Electromagnetic Waves
© 2006 by Taylor & Francis Group, LLC
Electromagnetic Waves Carlo G. Someda
Boca Raton London ...

Author:
Carlo G. Someda

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Electromagnetic Waves

© 2006 by Taylor & Francis Group, LLC

Electromagnetic Waves Carlo G. Someda

Boca Raton London New York

A CRC title, part of the Taylor & Francis imprint, a member of the Taylor & Francis Group, the academic division of T&F Informa plc.

© 2006 by Taylor & Francis Group, LLC

9589_Discl.fm Page 1 Friday, December 2, 2005 11:06 AM

Published in 2006 by CRC Press Taylor & Francis Group 6000 Broken Sound Parkway NW, Suite 300 Boca Raton, FL 33487-2742 © 2006 by Taylor & Francis Group, LLC CRC Press is an imprint of Taylor & Francis Group No claim to original U.S. Government works Printed in the United States of America on acid-free paper 10 9 8 7 6 5 4 3 2 1 International Standard Book Number-10: 0-8493-9589-5 (Hardcover) International Standard Book Number-13: 978-0-8493-9589-5 (Hardcover) Library of Congress Card Number 2005043549 This book contains information obtained from authentic and highly regarded sources. Reprinted material is quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable efforts have been made to publish reliable data and information, but the author and the publisher cannot assume responsibility for the validity of all materials or for the consequences of their use. No part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers. For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC) 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged. Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe.

Library of Congress Cataloging-in-Publication Data Someda, Carlo G. Electromagnetic waves / Carlo G. Someda.--2nd ed. p. cm. Includes bibliographical references and index. ISBN 0-8493-9589-5 1. Electromagnetic waves. I. Title. QC661.S66 2006 539.2--dc22

2005043549

Visit the Taylor & Francis Web site at http://www.taylorandfrancis.com Taylor & Francis Group is the Academic Division of Informa plc.

© 2006 by Taylor & Francis Group, LLC

and the CRC Press Web site at http://www.crcpress.com

To Gian Carlo Corazza and Enrique A.J. Marcatili who taught me how to mix Maxwell’s equations and practical life

To my grandchildren, Matilde and Edoardo

© 2006 by Taylor & Francis Group, LLC

Contents

Preface

xiii

Acknowledgments

xv

Preface to the Second Edition

xvii

1 Basic equations for electromagnetic ﬁelds 1.1 Introduction: Experimental laws 1.2 Maxwell’s equations and the charge continuity equation 1.3 Constitutive relations 1.4 Imposed currents 1.5 Divergence equations 1.6 Continuity conditions 1.7 The wave equation. The Helmholtz equation 1.8 Magnetic vector potential 1.9 Fitzgerald electric vector potential 1.10 Hertz vector potential 1.11 Further applications and suggested reading

1 1 4 5 9 11 13 15 18 23 25 26

References

33

2 Polarization 2.1 Introduction 2.2 Steinmetz representation of time-harmonic vectors 2.3 Parallel and orthogonal complex vectors 2.4 Properties of time-harmonic vectors 2.5 Properties of the complex vectors 2.6 Linear polarization ratio 2.7 Circular polarization ratio 2.8 Stokes parameters 2.9 The Poincar´e sphere 2.10 Evolution of polarization in a linear medium: Jones matrix 2.11 Further applications and suggested reading

35 35 36 37 38 39 40 41 42 45 46 48

References

51 vii

© 2006 by Taylor & Francis Group, LLC

viii

CONTENTS

3 General theorems 3.1 Introduction 3.2 Poynting’s theorem. Wave impedance 3.3 Uniqueness theorem 3.4 Reciprocity theorem 3.5 Equivalence theorem 3.6 Induction theorem 3.7 Duality theorem 3.8 TE-TM ﬁeld decomposition theorem 3.9 Spatial symmetries. Reﬂection operators 3.10 Further applications and suggested reading

53 53 53 58 61 63 69 71 72 75 79

References

87

4 Plane waves in isotropic media 4.1 Separability of variables in the homogeneous Helmholtz equation 4.2 Solution of the homogeneous Helmholtz equation in Cartesian coordinates 4.3 Plane waves: Terminology and classiﬁcation 4.4 Traveling waves. Phase velocity 4.5 Standing waves 4.6 Poynting vector and wave impedance 4.7 Completeness of plane waves 4.8 Reﬂection and refraction of plane waves 4.9 Fresnel formulas 4.10 Reﬂection in multilayer structures 4.11 Total reﬂection 4.12 Reﬂection on the surface of a good conductor 4.13 Further applications and suggested reading

89

90 93 96 98 101 105 108 113 115 118 122 124

References

131

5 Plane wave packets and beams 5.1 Modulated waves. Group velocity 5.2 Dispersion 5.3 The scalar approximation 5.4 The equations of geometrical optics 5.5 Geometrical optics: Electromagnetic implications 5.6 Examples of ray tracing in radio propagation and in optics 5.7 The WKBJ method 5.8 Further comments on the WKBJ method 5.9 Gaussian beams 5.10 Hermite-Gauss and Laguerre-Gauss modes 5.11 Reﬂection and refraction of Gaussian beams

133 133 137 140 142 147 149 152 155 157 162 168

© 2006 by Taylor & Francis Group, LLC

89

CONTENTS 5.12 On the completeness of a series 5.13 Further comments on rays and beams 5.14 Further applications and suggested reading

ix 171 173 174

References

183

6 Plane waves in anisotropic media 6.1 General properties of anisotropic media 6.2 Wave equations and potentials in anisotropic media 6.3 Birefringent media 6.4 Fresnel’s equation of wave normals 6.5 An application: Phase matching of two waves 6.6 Gyrotropic media 6.7 The Appleton-Hartree formula 6.8 An example of permittivity dyadic 6.9 Second example of permeability dyadic 6.10 Faraday rotation 6.11 Further applications and suggested reading

185 185 188 189 193 197 201 203 206 210 211 216

References

221

7 Waveguides with conducting walls 7.1 Introduction 7.2 Homogeneously ﬁlled cylindrical structures: Simpliﬁed proof of the TE-TM decomposition theorem 7.3 Waveguides with ideal conducting walls 7.4 Transmission modes of lossless cylindrical structures 7.5 Mode orthogonality 7.6 Some remarks on completeness 7.7 Rectangular waveguides 7.8 Circular waveguides and coaxial cables 7.9 Waveguides with nonideal walls 7.10 On wall impedances 7.11 Hybrid modes 7.12 Further applications and suggested reading

223 223

References

267

8 Waves on transmission lines 8.1 Introduction 8.2 Uniform transmission lines 8.3 Impedance transformation along a transmission line 8.4 Lossless transmission lines 8.5 Low-loss transmission lines 8.6 Partially standing waves

269 269 270 272 273 275 276

© 2006 by Taylor & Francis Group, LLC

224 228 229 235 238 239 244 250 254 258 260

x

CONTENTS 8.7 8.8 8.9 8.10 8.11 8.12 8.13

The Smith chart Remote measurement of the load impedance Impedance matching Transmission-line equations: An alternative derivation TEM and quasi-TEM propagation in planar lines The coupled-mode equations Further applications and suggested reading

278 282 284 291 297 301 305

References

311

9 Resonant cavities 9.1 Introduction 9.2 Separable coordinate systems in three dimensions 9.3 Completeness of resonator modes 9.4 Mode orthogonality in a perfect resonator 9.5 Lossless cylindrical cavities 9.6 Simple examples 9.7 Lossy resonators: Perturbation analysis. Intrinsic Q-factor 9.8 Resonators coupled to external loads. Loaded Q-factor 9.9 Open resonators 9.10 Stability of open resonators 9.11 Q-factor of an open resonator 9.12 Further applications and suggested reading

313 313 314 315 317 318 321 325 327 328 330 333 335

References

341

10 Dielectric waveguides 343 10.1 Introduction 343 10.2 Waves guided by a surface of discontinuity. The characteristic equation 344 10.3 Guided modes of a slab waveguide 349 10.4 Radiation modes of a slab waveguide 354 10.5 The cylindrical rod: Exact modes 356 10.6 Modal cut-oﬀ in the cylindrical rod 360 10.7 Weakly guiding rods: The LP modes 363 10.8 Dispersion in dielectric waveguides 369 10.9 Graded-index waveguides 375 10.10 The alpha proﬁles: An important class of multimode gradedindex ﬁbers 380 10.11 Attenuation in optical ﬁbers 385 10.12 Further applications and suggested reading 389 References

395

11 Retarded potentials

397

© 2006 by Taylor & Francis Group, LLC

CONTENTS 11.1 11.2 11.3 11.4 11.5 11.6 11.7 11.8 11.9 11.10 11.11 11.12

Introduction Green’s functions for the scalar Helmholtz equation Lorentz-gauge vector potentials in a homogeneous medium Field vectors in terms of dyadic Green’s functions Inhomogeneous media: Polarization currents Time-domain interpretation of Green’s functions Green’s function expansion into orthogonal eigenfunctions An example: Field in a rectangular box Spherical harmonics Multipole expansion An introduction to cylindrical harmonics Further applications and suggested reading

xi 397 398 401 404 406 407 410 412 414 420 424 426

References

431

12 Fundamentals of antenna theory 12.1 Introduction 12.2 Equivalent dipole moment of an extended source 12.3 Far-ﬁeld approximations 12.4 First example: Short electric-current element 12.5 Characterization of antennas 12.6 Behavior of receiving antennas. Reciprocity 12.7 Examples 12.8 Antenna arrays 12.9 Broad-side and end-ﬁre arrays 12.10 Further applications and suggested reading

433 433 435 437 439 444 448 454 464 469 471

References

477

13 Diﬀraction 13.1 Introduction 13.2 The diﬀraction integral: The vector formulation 13.3 Illumination conditions. Babinet’s principle 13.4 The scalar theory of diﬀraction 13.5 Diﬀraction formulas and Rayleigh-Sommerfeld 13.6 The Fresnel diﬀraction region 13.7 The Fraunhofer diﬀraction region 13.8 Examples 13.9 The ﬁeld near a focus: First example of Fresnel diﬀraction 13.10 Diﬀraction from a straight edge: Second example of Fresnel diﬀraction 13.11 A short note on the geometrical theory of diﬀraction 13.12 Further applications and suggested reading

479 479 479 484 488 493 495 497 500 508

References

521

© 2006 by Taylor & Francis Group, LLC

510 515 516

xii

CONTENTS

14 An introduction to the theory of coherence 14.1 Background and purpose of the chapter 14.2 The analytical signal 14.3 Complex degree of coherence 14.4 Temporal coherence of a source 14.5 Spatial coherence of a source 14.6 Higher-order coherence: An introduction 14.7 An introduction to photocount distributions 14.8 Modal noise in optical-ﬁber transmission systems: A short outline 14.9 Further applications and suggested reading

523 523 523 526 527 529 531 535

References

545

Appendices

547

A Vector calculus: Deﬁnitions and fundamental theorems

549

539 540

B Vector diﬀerential operators in frequently used reference systems 553 C Vector identities

555

D Fundamentals on Bessel functions D.1 Bessel, Neumann and Hankel functions D.2 Modiﬁed Bessel functions D.3 Bessel function formulas

557 557 560 561

References

565

Further Suggested Reading

567

© 2006 by Taylor & Francis Group, LLC

Preface Is there a real need to publish, in 1997, a new book on electromagnetic waves ? Clearly, the answer to this question will come from the market, not from the author. What the author can do is just to try to explain why he accepted to take this challenge. In the late 1970’s, UTET, one of the most important and prestigious Italian publishers, decided to prepare a series of volumes in electrical and electronic engineering, written by Italian authors. The overall purpose was to provide a complete cross-section of how these disciplines were taught in Italian universities. Francesco Barozzi, a very distinguished professor, then head of the Electrical and Electronic Engineering Department at the University of Trieste, was appointed as the editor in chief. He was a charming gentleman, with an immense, ubiquitous background in humanities and arts, which he never showed oﬀ. With a friendly smile, he always took an extremely calm approach to any kind of problem, making everybody feel at ease. To be selected by him as one of the authors was so rewarding, that my decision to prepare a volume for the UTET series was taken in less than one second. Writing the book was not equally fast: it was published in 1986. After that, for the ﬁrst few years its presence on the market was timid, but later sales increased to a very satisfactory level, which is now steadily maintained. Of course, this statement must be calibrated with respect to the limited size of the potential market: unfortunately, there is just one country where electromagnetic waves are taught in Italian! There are, on the other hand, several scientists around the world, who can read Italian. One of them, professor Michael C. Sexton, from University College Cork, Ireland, saw my book during one of his frequent visits to Padua in relationship to his research in the ﬁeld of plasma physics and fusion engineering. He was the ﬁrst who suggested to write an English version. Shortly after, Dr. Richard J. Black, who was then teaching at the Ecole Polytechnique in Montr´eal, Canada, and therefore was familiarizing himself with technical literature in the Romance languages, expressed the same opinion. They said “there is good stuﬀ in there.” I do not know precisely what they meant. On my part, I can say, comparing the contents of my book with those of many others, that I was fortunate to work, in a relatively short time interval, in three ﬁelds: microwaves, lasers, and ﬁber optics. What I tried to do in teaching, and consequently in writing my book, was to strengthen the links, and unify the background. The reader will judge whether I achieved this goal. The role Professor Sexton played was not only to trigger my interest in xiii © 2006 by Taylor & Francis Group, LLC

xiv

PREFACE

an English version of my book. He also contacted Chapman & Hall. The project was approved by the publisher under the condition that some changes were made, with respect to the Italian edition. The suggestions made to this purpose by anonymous referees matched exactly my own opinions. Indeed, my experiences as a student indicated that in the Latin countries there is a long tradition — maybe even a philosophical attitude — to teach “by concepts.” Conversely, in English-speaking countries one teaches — and learns — “by examples.” As a result, the English book could not be a mere translation, but had to be enriched with examples, and with more problems. As a net result, this book is very large. Maybe it is too thick, but I would not know what to cut. For this reason, I am very grateful to the publisher, who never put on me any pressure in order to reduce the market risks that this unusual size entails. C.G. Someda

© 2006 by Taylor & Francis Group, LLC

Acknowledgments The list of those towards whom I became indebted, during the preparation of this book, has no upper limits. Let me begin by repeating two names that I already mentioned in the Preface: those of Professor Michael C. Sexton and Dr. Richard J. Black. Their contributions range from suggesting the material to be added to the Italian version, and choosing examples and problems, to polishing my English, and adding to the references. In one word, their roles are quite similar to those of co-authors. It was very generous on their part to leave only to me the honor to sign this book. Next, I wish to thank an entire team of graduate students, post-doctoral fellows, and research assistants, who spent so much of their time and energy doing the initial translations, and then reading the manuscript over and over again, looking for no other reward but to see me happy and thankful. They were so many, that the list of their names would be too long, and the probability of forgetting someone would be too high. I am sure they understand what I mean, and recognize themselves in these lines. I want to mention explicitly, and to thank enthusiastically, Ms. Barbara Sicoli, who polished with professionality and patience the ﬁnal version of the manuscript, and Ms. Flavia Bergamin, who did the computer processing of almost all the ﬁgures. Finally, let me thank UTET, the Publisher of the Italian text, for the generous permission to publish the English version as a totally independent book, and Chapman & Hall, for being so patient during the long preparation of the manuscript, and for encouraging me when the goal looked out of my reach.

xv © 2006 by Taylor & Francis Group, LLC

Preface to the Second Edition

The ﬁrst edition of my “Electromagnetic Waves” is now seven years old not so old, in a ﬁeld where basic knowledge has a long record of stability, and technological breakthroughs occur on a time scale of several years, if not decades. Why a second edition, then? In fact, the main motivation does not come from the evolution of the contents, but from facts that took place in the book company arena. The publisher of the ﬁrst edition, Chapman & Hall, was acquired by another group - a prominent one - long ago, shortly after publishing my book. What I learnt during the following years was that, when similar events occur, regardless of the size and of the prestige of the acquiring group, redistributing the titles of the acquired company is a very complicated issue, which does not necessarily entail happy endings for all authors. It turned out that, without anybody being speciﬁcally guilty for that, the highly positive, I would almost say enthusiastic, reviews that my book had enjoyed when it appeared, were not followed by a marketing campaign of comparable intensity. Loss of enthusiasm can be contagious. Just to mention one of the negative fallouts, I never completed the exercise manual. Sales declined. The reader can imagine, then, how grateful I am to the Taylor & Francis Group, and to its CRC Press branch in particular, for proposing a second edition at the right moment. The ﬁrst one was almost sold out, and I was facing a bifurcation: either restart with new motivations, or abandon the whole project. I believe that the whole team wants now to show that we made the right decision. For what refers to the contents, the list of changes is very short. I have added, here and there, a few deﬁnitions, which are not new, but whose importance I had underestimated in the past; just to quote an example, antenna eﬀective height, in Chapter 12. The reference list has been updated, as a consequence of the wide set of ﬁrst-class books that were published in these seven years. The text of several problems has been edited, to remove some ambiguities or to complete, here and there, insuﬃcient data. The discovery of little imperfections in the problems occurred while their solutions were being prepared; this entails that the manual should be on the market soon after the book. Last but not least, several tens of misprints and minor mistakes that, in spite of automatic correctors etc., appeared in the ﬁrst edition (an extremely frustrating experience) have been taken care of. This is a signiﬁcant improvement, although, I am afraid, typographical perfection remains a dream. I am xvii © 2006 by Taylor & Francis Group, LLC

xviii

PREFACE TO THE SECOND EDITION

very grateful to an innumerable list of colleagues, friends, students, who had an active role in identifying these errors. To conclude, let me thank collectively all those who contributed to the second edition. Two persons, however, deserve more than anonymous thanks, as they have been instrumental in preparing the manuscript. One is Mr. Emanuele Zattin, who, while completing his education as a computer engineer, has developed a terriﬁc experience as a professional electronic editor. It took all his skills to recover the previous ﬁles, and make them compatible with state-of-the-art software. The second one is a junior colleague of mine, Antonio Daniele Capobianco. Without Tony’s continuous dedication and unlimited patience, preparation of this manuscript would have been an endless, painful process, not only for me, but for many other people.

C.G. Someda

© 2006 by Taylor & Francis Group, LLC

Electromagnetic Waves

© 2006 by Taylor & Francis Group, LLC

Electromagnetic Waves Carlo G. Someda

Boca Raton London New York

A CRC title, part of the Taylor & Francis imprint, a member of the Taylor & Francis Group, the academic division of T&F Informa plc.

© 2006 by Taylor & Francis Group, LLC

9589_Discl.fm Page 1 Friday, December 2, 2005 11:06 AM

Published in 2006 by CRC Press Taylor & Francis Group 6000 Broken Sound Parkway NW, Suite 300 Boca Raton, FL 33487-2742 © 2006 by Taylor & Francis Group, LLC CRC Press is an imprint of Taylor & Francis Group No claim to original U.S. Government works Printed in the United States of America on acid-free paper 10 9 8 7 6 5 4 3 2 1 International Standard Book Number-10: 0-8493-9589-5 (Hardcover) International Standard Book Number-13: 978-0-8493-9589-5 (Hardcover) Library of Congress Card Number 2005043549 This book contains information obtained from authentic and highly regarded sources. Reprinted material is quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable efforts have been made to publish reliable data and information, but the author and the publisher cannot assume responsibility for the validity of all materials or for the consequences of their use. No part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers. For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC) 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged. Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe.

Library of Congress Cataloging-in-Publication Data Someda, Carlo G. Electromagnetic waves / Carlo G. Someda.--2nd ed. p. cm. Includes bibliographical references and index. ISBN 0-8493-9589-5 1. Electromagnetic waves. I. Title. QC661.S66 2006 539.2--dc22

2005043549

Visit the Taylor & Francis Web site at http://www.taylorandfrancis.com Taylor & Francis Group is the Academic Division of Informa plc.

© 2006 by Taylor & Francis Group, LLC

and the CRC Press Web site at http://www.crcpress.com

To Gian Carlo Corazza and Enrique A.J. Marcatili who taught me how to mix Maxwell’s equations and practical life

To my grandchildren, Matilde and Edoardo

© 2006 by Taylor & Francis Group, LLC

Contents

Preface

xiii

Acknowledgments

xv

Preface to the Second Edition

xvii

1 Basic equations for electromagnetic ﬁelds 1.1 Introduction: Experimental laws 1.2 Maxwell’s equations and the charge continuity equation 1.3 Constitutive relations 1.4 Imposed currents 1.5 Divergence equations 1.6 Continuity conditions 1.7 The wave equation. The Helmholtz equation 1.8 Magnetic vector potential 1.9 Fitzgerald electric vector potential 1.10 Hertz vector potential 1.11 Further applications and suggested reading

1 1 4 5 9 11 13 15 18 23 25 26

References

33

2 Polarization 2.1 Introduction 2.2 Steinmetz representation of time-harmonic vectors 2.3 Parallel and orthogonal complex vectors 2.4 Properties of time-harmonic vectors 2.5 Properties of the complex vectors 2.6 Linear polarization ratio 2.7 Circular polarization ratio 2.8 Stokes parameters 2.9 The Poincar´e sphere 2.10 Evolution of polarization in a linear medium: Jones matrix 2.11 Further applications and suggested reading

35 35 36 37 38 39 40 41 42 45 46 48

References

51 vii

© 2006 by Taylor & Francis Group, LLC

viii

CONTENTS

3 General theorems 3.1 Introduction 3.2 Poynting’s theorem. Wave impedance 3.3 Uniqueness theorem 3.4 Reciprocity theorem 3.5 Equivalence theorem 3.6 Induction theorem 3.7 Duality theorem 3.8 TE-TM ﬁeld decomposition theorem 3.9 Spatial symmetries. Reﬂection operators 3.10 Further applications and suggested reading

53 53 53 58 61 63 69 71 72 75 79

References

87

4 Plane waves in isotropic media 4.1 Separability of variables in the homogeneous Helmholtz equation 4.2 Solution of the homogeneous Helmholtz equation in Cartesian coordinates 4.3 Plane waves: Terminology and classiﬁcation 4.4 Traveling waves. Phase velocity 4.5 Standing waves 4.6 Poynting vector and wave impedance 4.7 Completeness of plane waves 4.8 Reﬂection and refraction of plane waves 4.9 Fresnel formulas 4.10 Reﬂection in multilayer structures 4.11 Total reﬂection 4.12 Reﬂection on the surface of a good conductor 4.13 Further applications and suggested reading

89

90 93 96 98 101 105 108 113 115 118 122 124

References

131

5 Plane wave packets and beams 5.1 Modulated waves. Group velocity 5.2 Dispersion 5.3 The scalar approximation 5.4 The equations of geometrical optics 5.5 Geometrical optics: Electromagnetic implications 5.6 Examples of ray tracing in radio propagation and in optics 5.7 The WKBJ method 5.8 Further comments on the WKBJ method 5.9 Gaussian beams 5.10 Hermite-Gauss and Laguerre-Gauss modes 5.11 Reﬂection and refraction of Gaussian beams

133 133 137 140 142 147 149 152 155 157 162 168

© 2006 by Taylor & Francis Group, LLC

89

CONTENTS 5.12 On the completeness of a series 5.13 Further comments on rays and beams 5.14 Further applications and suggested reading

ix 171 173 174

References

183

6 Plane waves in anisotropic media 6.1 General properties of anisotropic media 6.2 Wave equations and potentials in anisotropic media 6.3 Birefringent media 6.4 Fresnel’s equation of wave normals 6.5 An application: Phase matching of two waves 6.6 Gyrotropic media 6.7 The Appleton-Hartree formula 6.8 An example of permittivity dyadic 6.9 Second example of permeability dyadic 6.10 Faraday rotation 6.11 Further applications and suggested reading

185 185 188 189 193 197 201 203 206 210 211 216

References

221

7 Waveguides with conducting walls 7.1 Introduction 7.2 Homogeneously ﬁlled cylindrical structures: Simpliﬁed proof of the TE-TM decomposition theorem 7.3 Waveguides with ideal conducting walls 7.4 Transmission modes of lossless cylindrical structures 7.5 Mode orthogonality 7.6 Some remarks on completeness 7.7 Rectangular waveguides 7.8 Circular waveguides and coaxial cables 7.9 Waveguides with nonideal walls 7.10 On wall impedances 7.11 Hybrid modes 7.12 Further applications and suggested reading

223 223

References

267

8 Waves on transmission lines 8.1 Introduction 8.2 Uniform transmission lines 8.3 Impedance transformation along a transmission line 8.4 Lossless transmission lines 8.5 Low-loss transmission lines 8.6 Partially standing waves

269 269 270 272 273 275 276

© 2006 by Taylor & Francis Group, LLC

224 228 229 235 238 239 244 250 254 258 260

x

CONTENTS 8.7 8.8 8.9 8.10 8.11 8.12 8.13

The Smith chart Remote measurement of the load impedance Impedance matching Transmission-line equations: An alternative derivation TEM and quasi-TEM propagation in planar lines The coupled-mode equations Further applications and suggested reading

278 282 284 291 297 301 305

References

311

9 Resonant cavities 9.1 Introduction 9.2 Separable coordinate systems in three dimensions 9.3 Completeness of resonator modes 9.4 Mode orthogonality in a perfect resonator 9.5 Lossless cylindrical cavities 9.6 Simple examples 9.7 Lossy resonators: Perturbation analysis. Intrinsic Q-factor 9.8 Resonators coupled to external loads. Loaded Q-factor 9.9 Open resonators 9.10 Stability of open resonators 9.11 Q-factor of an open resonator 9.12 Further applications and suggested reading

313 313 314 315 317 318 321 325 327 328 330 333 335

References

341

10 Dielectric waveguides 343 10.1 Introduction 343 10.2 Waves guided by a surface of discontinuity. The characteristic equation 344 10.3 Guided modes of a slab waveguide 349 10.4 Radiation modes of a slab waveguide 354 10.5 The cylindrical rod: Exact modes 356 10.6 Modal cut-oﬀ in the cylindrical rod 360 10.7 Weakly guiding rods: The LP modes 363 10.8 Dispersion in dielectric waveguides 369 10.9 Graded-index waveguides 375 10.10 The alpha proﬁles: An important class of multimode gradedindex ﬁbers 380 10.11 Attenuation in optical ﬁbers 385 10.12 Further applications and suggested reading 389 References

395

11 Retarded potentials

397

© 2006 by Taylor & Francis Group, LLC

CONTENTS 11.1 11.2 11.3 11.4 11.5 11.6 11.7 11.8 11.9 11.10 11.11 11.12

Introduction Green’s functions for the scalar Helmholtz equation Lorentz-gauge vector potentials in a homogeneous medium Field vectors in terms of dyadic Green’s functions Inhomogeneous media: Polarization currents Time-domain interpretation of Green’s functions Green’s function expansion into orthogonal eigenfunctions An example: Field in a rectangular box Spherical harmonics Multipole expansion An introduction to cylindrical harmonics Further applications and suggested reading

xi 397 398 401 404 406 407 410 412 414 420 424 426

References

431

12 Fundamentals of antenna theory 12.1 Introduction 12.2 Equivalent dipole moment of an extended source 12.3 Far-ﬁeld approximations 12.4 First example: Short electric-current element 12.5 Characterization of antennas 12.6 Behavior of receiving antennas. Reciprocity 12.7 Examples 12.8 Antenna arrays 12.9 Broad-side and end-ﬁre arrays 12.10 Further applications and suggested reading

433 433 435 437 439 444 448 454 464 469 471

References

477

13 Diﬀraction 13.1 Introduction 13.2 The diﬀraction integral: The vector formulation 13.3 Illumination conditions. Babinet’s principle 13.4 The scalar theory of diﬀraction 13.5 Diﬀraction formulas and Rayleigh-Sommerfeld 13.6 The Fresnel diﬀraction region 13.7 The Fraunhofer diﬀraction region 13.8 Examples 13.9 The ﬁeld near a focus: First example of Fresnel diﬀraction 13.10 Diﬀraction from a straight edge: Second example of Fresnel diﬀraction 13.11 A short note on the geometrical theory of diﬀraction 13.12 Further applications and suggested reading

479 479 479 484 488 493 495 497 500 508

References

521

© 2006 by Taylor & Francis Group, LLC

510 515 516

xii

CONTENTS

14 An introduction to the theory of coherence 14.1 Background and purpose of the chapter 14.2 The analytical signal 14.3 Complex degree of coherence 14.4 Temporal coherence of a source 14.5 Spatial coherence of a source 14.6 Higher-order coherence: An introduction 14.7 An introduction to photocount distributions 14.8 Modal noise in optical-ﬁber transmission systems: A short outline 14.9 Further applications and suggested reading

523 523 523 526 527 529 531 535

References

545

Appendices

547

A Vector calculus: Deﬁnitions and fundamental theorems

549

539 540

B Vector diﬀerential operators in frequently used reference systems 553 C Vector identities

555

D Fundamentals on Bessel functions D.1 Bessel, Neumann and Hankel functions D.2 Modiﬁed Bessel functions D.3 Bessel function formulas

557 557 560 561

References

565

Further Suggested Reading

567

© 2006 by Taylor & Francis Group, LLC

Preface Is there a real need to publish, in 1997, a new book on electromagnetic waves ? Clearly, the answer to this question will come from the market, not from the author. What the author can do is just to try to explain why he accepted to take this challenge. In the late 1970’s, UTET, one of the most important and prestigious Italian publishers, decided to prepare a series of volumes in electrical and electronic engineering, written by Italian authors. The overall purpose was to provide a complete cross-section of how these disciplines were taught in Italian universities. Francesco Barozzi, a very distinguished professor, then head of the Electrical and Electronic Engineering Department at the University of Trieste, was appointed as the editor in chief. He was a charming gentleman, with an immense, ubiquitous background in humanities and arts, which he never showed oﬀ. With a friendly smile, he always took an extremely calm approach to any kind of problem, making everybody feel at ease. To be selected by him as one of the authors was so rewarding, that my decision to prepare a volume for the UTET series was taken in less than one second. Writing the book was not equally fast: it was published in 1986. After that, for the ﬁrst few years its presence on the market was timid, but later sales increased to a very satisfactory level, which is now steadily maintained. Of course, this statement must be calibrated with respect to the limited size of the potential market: unfortunately, there is just one country where electromagnetic waves are taught in Italian! There are, on the other hand, several scientists around the world, who can read Italian. One of them, professor Michael C. Sexton, from University College Cork, Ireland, saw my book during one of his frequent visits to Padua in relationship to his research in the ﬁeld of plasma physics and fusion engineering. He was the ﬁrst who suggested to write an English version. Shortly after, Dr. Richard J. Black, who was then teaching at the Ecole Polytechnique in Montr´eal, Canada, and therefore was familiarizing himself with technical literature in the Romance languages, expressed the same opinion. They said “there is good stuﬀ in there.” I do not know precisely what they meant. On my part, I can say, comparing the contents of my book with those of many others, that I was fortunate to work, in a relatively short time interval, in three ﬁelds: microwaves, lasers, and ﬁber optics. What I tried to do in teaching, and consequently in writing my book, was to strengthen the links, and unify the background. The reader will judge whether I achieved this goal. The role Professor Sexton played was not only to trigger my interest in xiii © 2006 by Taylor & Francis Group, LLC

xiv

PREFACE

an English version of my book. He also contacted Chapman & Hall. The project was approved by the publisher under the condition that some changes were made, with respect to the Italian edition. The suggestions made to this purpose by anonymous referees matched exactly my own opinions. Indeed, my experiences as a student indicated that in the Latin countries there is a long tradition — maybe even a philosophical attitude — to teach “by concepts.” Conversely, in English-speaking countries one teaches — and learns — “by examples.” As a result, the English book could not be a mere translation, but had to be enriched with examples, and with more problems. As a net result, this book is very large. Maybe it is too thick, but I would not know what to cut. For this reason, I am very grateful to the publisher, who never put on me any pressure in order to reduce the market risks that this unusual size entails. C.G. Someda

© 2006 by Taylor & Francis Group, LLC

Acknowledgments The list of those towards whom I became indebted, during the preparation of this book, has no upper limits. Let me begin by repeating two names that I already mentioned in the Preface: those of Professor Michael C. Sexton and Dr. Richard J. Black. Their contributions range from suggesting the material to be added to the Italian version, and choosing examples and problems, to polishing my English, and adding to the references. In one word, their roles are quite similar to those of co-authors. It was very generous on their part to leave only to me the honor to sign this book. Next, I wish to thank an entire team of graduate students, post-doctoral fellows, and research assistants, who spent so much of their time and energy doing the initial translations, and then reading the manuscript over and over again, looking for no other reward but to see me happy and thankful. They were so many, that the list of their names would be too long, and the probability of forgetting someone would be too high. I am sure they understand what I mean, and recognize themselves in these lines. I want to mention explicitly, and to thank enthusiastically, Ms. Barbara Sicoli, who polished with professionality and patience the ﬁnal version of the manuscript, and Ms. Flavia Bergamin, who did the computer processing of almost all the ﬁgures. Finally, let me thank UTET, the Publisher of the Italian text, for the generous permission to publish the English version as a totally independent book, and Chapman & Hall, for being so patient during the long preparation of the manuscript, and for encouraging me when the goal looked out of my reach.

xv © 2006 by Taylor & Francis Group, LLC

Preface to the Second Edition

The ﬁrst edition of my “Electromagnetic Waves” is now seven years old not so old, in a ﬁeld where basic knowledge has a long record of stability, and technological breakthroughs occur on a time scale of several years, if not decades. Why a second edition, then? In fact, the main motivation does not come from the evolution of the contents, but from facts that took place in the book company arena. The publisher of the ﬁrst edition, Chapman & Hall, was acquired by another group - a prominent one - long ago, shortly after publishing my book. What I learnt during the following years was that, when similar events occur, regardless of the size and of the prestige of the acquiring group, redistributing the titles of the acquired company is a very complicated issue, which does not necessarily entail happy endings for all authors. It turned out that, without anybody being speciﬁcally guilty for that, the highly positive, I would almost say enthusiastic, reviews that my book had enjoyed when it appeared, were not followed by a marketing campaign of comparable intensity. Loss of enthusiasm can be contagious. Just to mention one of the negative fallouts, I never completed the exercise manual. Sales declined. The reader can imagine, then, how grateful I am to the Taylor & Francis Group, and to its CRC Press branch in particular, for proposing a second edition at the right moment. The ﬁrst one was almost sold out, and I was facing a bifurcation: either restart with new motivations, or abandon the whole project. I believe that the whole team wants now to show that we made the right decision. For what refers to the contents, the list of changes is very short. I have added, here and there, a few deﬁnitions, which are not new, but whose importance I had underestimated in the past; just to quote an example, antenna eﬀective height, in Chapter 12. The reference list has been updated, as a consequence of the wide set of ﬁrst-class books that were published in these seven years. The text of several problems has been edited, to remove some ambiguities or to complete, here and there, insuﬃcient data. The discovery of little imperfections in the problems occurred while their solutions were being prepared; this entails that the manual should be on the market soon after the book. Last but not least, several tens of misprints and minor mistakes that, in spite of automatic correctors etc., appeared in the ﬁrst edition (an extremely frustrating experience) have been taken care of. This is a signiﬁcant improvement, although, I am afraid, typographical perfection remains a dream. I am xvii © 2006 by Taylor & Francis Group, LLC

xviii

PREFACE TO THE SECOND EDITION

very grateful to an innumerable list of colleagues, friends, students, who had an active role in identifying these errors. To conclude, let me thank collectively all those who contributed to the second edition. Two persons, however, deserve more than anonymous thanks, as they have been instrumental in preparing the manuscript. One is Mr. Emanuele Zattin, who, while completing his education as a computer engineer, has developed a terriﬁc experience as a professional electronic editor. It took all his skills to recover the previous ﬁles, and make them compatible with state-of-the-art software. The second one is a junior colleague of mine, Antonio Daniele Capobianco. Without Tony’s continuous dedication and unlimited patience, preparation of this manuscript would have been an endless, painful process, not only for me, but for many other people.

C.G. Someda

© 2006 by Taylor & Francis Group, LLC

CHAPTER 2

Polarization

2.1

Introduction

As we said at the beginning of the previous chapter, we take it for granted that the reader has been previously exposed to the fundamentals of circuit and network theory. In particular, we assume that the reader is familiar with the representation of time-harmonic (i.e., varying sinusoidally in time) scalar quantities by means of complex numbers, i.e., with the so-called Steinmetz method (or procedure). This assumption is based on typical electrical engineering and/or electronics curricula, and also applies to curricula in physics in many universities. Those readers for whom this assumption does not apply can grasp the fundamentals of Steinmetz procedure in the next section. However, it is highly recommended that they read books that treat this subject in depth, and especially that they familiarize themselves with the Steinmetz method by solving many problems on circuits. For those readers who, on the contrary, feel conﬁdent that they know the procedure, the ﬁrst part of this chapter may look rather tedious. Nevertheless, we recommend that they read carefully at least the second part, beginning with Section 2.5, as otherwise some subtle details in the following chapters might be diﬃcult to grasp. Throughout this book, substantial beneﬁt can be gained if we assume, whenever possible, a purely monochromatic sinusoidal time dependence of the electromagnetic quantities that we are interested in, such as ﬁelds, potentials, etc. However, in contrast to circuit theory, here most of our variables are vectors, in three-dimensional (3-D) real space. We must ﬁrst ask ourselves how the Steinmetz method can be generalized to time-harmonic vectors. As we will see in the next section, this entails some additional diﬃculty. The rest of this chapter is then devoted to demonstrating why these complications are indeed worth the time and eﬀort they require. The net result will indeed be, that the complex representation of a time-harmonic vector can convey, in compact forms that lend themselves to very elegant analytical and geometrical interpretations, information on an important physical feature of the timeharmonic vector itself, namely, its polarization. Applications are found in all the following chapters, ranging from plane waves to waveguides, from beams to antennas, from diﬀraction to coherence. 35 © 2006 by Taylor & Francis Group, LLC

36 2.2

POLARIZATION Steinmetz representation of time-harmonic vectors

In real 3-D space, let us introduce a completely arbitrary cartesian orthogonal reference frame, {x1 , x2 , x3 }. Let {ˆ a1 , a ˆ2 , a ˆ3 } be the unit-vectors of this frame. We call time-harmonic, with angular frequency ω, a vector whose components in that reference frame vary sinusoidally in time, with angular frequency ω. Let then Sv (ω) be the set of all time-harmonic vectors of angular frequency ω: 3 a ˆn an cos(ωt + ϕn ) ; an ∈ R ; ϕn ∈ R ; an ≥ 0 ; 0 ≤ ϕn < 2π . Sv (ω) ≡ n=1

(2.1) It is easy to show that Sv (ω) satisﬁes the axioms that deﬁne a vector space over the real ﬁeld. We then deﬁne a complex vector as a linear combination of time-independent vectors, each of which belongs to the real 3-D space, where the scalar coefﬁcients are arbitrary complex numbers. Let C3 be the set of all the complex vectors deﬁned in this way: 3 jϕn a ˆn an e ; an ∈ R ; ϕn ∈ R ; an ≥ 0 ; 0 ≤ ϕn < 2π . (2.2) C3 ≡ n=1

Then, it is easy to show that C3 is also a vector space over the real ﬁeld. The relation: 3 3

jϕn jωt a a ˆn an cos(ωt + ϕn ) = Re a ˆn an e e = Re a ejωt ∼ (t) = n=1

n=1

(2.3) is a one-to-one and onto mapping, τ , between Sv (ω) and C3 . This relationship forms the complex representation of time-harmonic vectors, and is also referred to as the vectorial Steinmetz method. Let us deﬁne, in C3 , a scalar product as a straightforward extension of two well-known operations, the inner product of two real vectors, and the product of complex numbers, i.e.: 3 3 3 jϕn jψn a ˆn an e a ˆ n bn e an bn ej(ϕn +ψn ) . (2.4) · = a·b = n=1

n=1

n=1

With this deﬁnition of scalar product, C3 is not a unitary space, according to the deﬁnition given in classical textbooks on linear algebra (e.g., Halmos, 1958), because some axioms are violated. Some consequences, in particular those that involve the deﬁnitions of parallel and orthogonal complex vectors, will be outlined in the next section. Another consequence that we wish to point out is that the mapping in Eq. (2.3) is not an isomorphism, as there is no one-to-one matching between scalar products in the two vector spaces.

© 2006 by Taylor & Francis Group, LLC

PARALLEL AND ORTHOGONAL COMPLEX VECTORS

37

The scalar products in the two spaces are related to each other as follows: 1 Re{a · b∗ } = a ∼·b ∼ , 2

(2.5)

where the bar above the factors on the right-hand side indicates a time average over a period T = 2π/ω. We may then call the mapping in Eq. (2.3) a pseudoisomorphism.

2.3

Parallel and orthogonal complex vectors

Two complex vectors, A and B, are said to be parallel when there exists a complex number c, such that the relationship: A = cB

(2.6)

is satisﬁed. We want to preserve the validity, for complex vectors, of the following relation, which is well known to hold for real vectors and for real numbers: |A| = |c| |B| .

(2.7)

Note that, in order to do that, it is necessary to deﬁne the modulus of a complex vector in the following way: √ (2.8) |A| = A · A∗ . Another notion that we want to preserve from the space of real vectors is that the projection of a complex vector onto its own direction must coincide with its modulus. Then, to be consistent, the projection of a complex vector, A, onto the direction of another, completely arbitrary complex vector, C, must be deﬁned as follows: Ac = A · C∗ /|C| = (C · A∗ )∗ /|C|

.

(2.9)

Consequently, two complex vectors A and C are said to be orthogonal to each other when the following relationship: A · C∗ = A∗ · C = 0

(2.10)

is satisﬁed. Note that in general Eq. (2.10) does not imply A · C = 0. These results are probably surprising, and the reader might believe that they lead, sooner or later, to inconsistencies. On the contrary, we will see that these new rules are instrumental in some steps. For example, in Chapter 4 they will help us to correctly calculate the per-unit-area power carried by an arbitrarily polarized plane wave.

© 2006 by Taylor & Francis Group, LLC

38 2.4

POLARIZATION Properties of time-harmonic vectors

Any element in the set deﬁned by Eq. (2.1) can be written in the following way: 3 3 A a ˆn an cos ϕn cos ωt − a ˆn an sin ϕn sin ωt ∼ = n=1

n=1

= A cos ωt − A sin ωt ,

(2.11)

where the two vectors: A =

3

a ˆn an cos ϕn

,

A =

n=1

3

a ˆn an sin ϕn

(2.12)

n=1

are real and time-independent. Eq. (2.11) implies that, at any time t, the vector A ∼ belongs to the plane deﬁned by the two vectors in Eq. (2.12). Hence, for any time-harmonic vector originating at a given point O, the trajectory described vs. time by its terminal point is a plane curve. Furthermore, as the two vectors in Eq. (2.12) are linearly combined with time-periodic coeﬃcients, cos ωt and sin ωt, that trajectory has to be a closed curve. Let us show that in fact Eq. (2.11) is the parametric equation of an ellipse. Indeed, looking for the equation of that trajectory in any orthogonal reference frame {x , y } on the ˆx + y a ˆy , where a ˆx {A , A } plane (see Figure 2.1), we may write A= ∼ x a and a ˆy are obviously the unit vectors along the axes. Then, as a consequence of Eq. (2.11), we have: x y

= ax cos ωt − ax sin ωt , =

ay

cos ωt −

ay

(2.13)

sin ωt .

(2.14)

We now multiply Eq. (2.13) by ay , and Eq. (2.14) by ax , and subtract. Then, we multiply Eq. (2.13) by ay , and Eq. (2.14) by ax , and subtract. Finally, we square the two expressions, and sum, making use of the identity sin2 ωt + cos2 ωt = 1. The ﬁnal result is: (x ay − y ax )2 + (x ay − y ax )2 = (ax ay − ay ax )2

,

(2.15)

which is the standard form for the equation of an ellipse on the {x , y } plane. This result is usually expressed saying that a generic time-harmonic vector is elliptically polarized. It is usually convenient to express the ellipse in terms of the coordinate system corresponding to what are known as its principal axes. The principal axes are labelled x and y in Figure 2.1, and are parallel to the major and minor axes of the ellipse, respectively. In this reference frame, the instantaneous vector (2.11) can be expressed in the following form: (2.16) A ∼ = Fx cos(ωt + ϕ) − Fy sin(ωt + ϕ) , where Fx is a real vector parallel to the x axis, and Fy is a real vector parallel to the y axis.

© 2006 by Taylor & Francis Group, LLC

PROPERTIES OF THE COMPLEX VECTORS

39

Figure 2.1 The polarization ellipse, its principal axes, and a generic cartesian reference frame.

To ﬁnd the principal axes, when a vector is given in the form Eq. (2.11), is perfectly equivalent to ﬁnding the phase angle ϕ in Eq. (2.16). The problem is much simpler to tackle if one makes use of the corresponding complex vector, and therefore its solution is left to the next section. 2.5

Properties of the complex vectors

From Eq. (2.11) it follows that:

jωt . + j A ) e A = Re (A ∼

(2.17)

Comparing this with Eq. (2.3), we see that the complex vector A, which corresponds, according to the Steinmetz procedure, to the time-harmonic vector A ∼ , is connected to the two time-independent real vectors that appear in Eq. (2.11) by the simple relationship: A = A + j A

.

(2.18)

If we need to express the vector (2.18) in a generic orthogonal cartesian frame {x , y }, we may write: A = Ax a ˆx + Ay a ˆy

,

(2.19)

and the complex numbers Ax , Ay are the components of A in this frame, which is real, so that we may write: Ax

= A·a ˆ∗x = A · a ˆx

,

Ay

= A·a ˆ∗y = A · a ˆy

.

(2.20)

If we want to introduce the reference frame of the principal axes, then, combining Eq. (2.16) with Eq. (2.3), we get: A = Ax a ˆ x + Ay a ˆy = |Fx | ejϕ a ˆx ± j |Fy | ejϕ a ˆy

© 2006 by Taylor & Francis Group, LLC

.

(2.21)

40

POLARIZATION

This result shows that there is another way to deﬁne the principal axes of the ellipse represented by Eq. (2.15). It is the basis in which the ratio between the cartesian components of the complex vector A is purely imaginary, that is, the ratio: Ay , (2.22) p=j Ax is real. On the other hand, in a generic orthogonal basis, the ratio: p = j

Ay Ax

,

(2.23)

is a complex number. As is well known, a rotation of a real reference frame by an angle δ (see again Figure 2.1) entails the following transformation for the vector components: cos δ sin δ Ax Ax = . (2.24) − sin δ cos δ Ay Ay Then, comparing Eq. (2.22) and Eq. (2.23), we see that a simple way to ﬁnd the principal axes, starting from the vector expressed in a generic reference frame, is to impose that the rotation yield a real value for Eq. (2.22), i.e., to impose p = p∗ . We leave it as an exercise for the reader to prove that the ﬁnal result is the following rule: tan 2δ = j

p − p∗ 1 − p p∗

.

(2.25)

This point will be exploited in the next two sections. 2.6

Linear polarization ratio

The complex number in Eq. (2.23) is referred to as the linear polarization ratio of either the time-harmonic vector A ∼ , or the corresponding complex vector A, in the generic orthogonal reference frame {x , y }. When the reference frame is rotated by a generic angle δ , then the components are transformed according to Eq. (2.24), with δ replaced by δ . It is easy to check that the corresponding transformation rule for the linear polarization ratio is: p + j tan δ . (2.26) p = 1 + j p tan δ Eq. (2.26) is a bilinear relationship between p and p . The reader may prove, as an exercise, that an invariant under this transformation is the sign of the real part of the linear polarization ratio. This suggests that this sign has a meaning which is intrinsic to the vector A, independent of the reference frame. Indeed, we may observe that: • if Re(p ) < 0, i.e., when the sign in front of the last term in Eq. (2.21) is a plus, then the time-harmonic vector A ∼ rotates clockwise for an observer watching at the positive face of the {x, y} plane — we refer to this as a left-handed elliptical polarization;

© 2006 by Taylor & Francis Group, LLC

CIRCULAR POLARIZATION RATIO

41

• if Re(p ) > 0, i.e., when the sign in front of the last term in Eq. (2.21) is a minus, then the time-harmonic vector A ∼ rotates counterclockwise for an observer watching at the positive face of the {x, y} plane — we refer to this as a right-handed elliptical polarization; • ﬁnally, if Re(p ) = 0, i.e., if Fy = 0 in Eqs. (2.21) and (2.16), then the ellipse described by the point P = O+ A ∼ (t) degenerates into a segment — we refer to this as a linear polarization. In particular, when p = ±1, then |Fx | = |Fy | in Eqs. (2.21) and (2.16). The ellipse becomes a circle, and we refer to these cases as circular polarizations, either right- or left-handed, obviously depending on the sign of p . 2.7

Circular polarization ratio

Consider the following pair of complex vectors: √ √ ˆ = (ˆ ˆ = (ˆ R ax − j a ˆy )/ 2 , L ax + j a ˆy )/ 2 .

(2.27)

From what we saw in the previous section, they represent two circularly polarized time-harmonic vectors, a right-handed and a left-handed one, respectively. These two complex vectors: • belong to the (x, y) plane; • have moduli equal to unity: ˆ·R ˆ∗ = 1 , R

ˆ·L ˆ∗ = 1 , L

(2.28)

so they can be called unit vectors; • are orthogonal to each other, according to the deﬁnition in Eq. (2.10): ˆ·L ˆ∗ = 0 . R

(2.29)

Therefore, they form an orthonormal basis, which we refer to as the rotating frame. Any complex vector A belonging to the plane of Figure 2.1 can be expanded into this basis, as: ˆ + AL L ˆ . A = AR R

(2.30)

The components of A in this basis, the rotating frame, are related to those in the principal axis frame {x, y} by: AR

ˆ ∗ = √1 (Ax + j Ay ) , = A·R 2

AL

∗ 1 = A·L ˆ = √ (Ax − j Ay ) 2

.

(2.31)

The complex number: AL (2.32) AR is called the circular polarization ratio of the time-harmonic vector A ∼ , or of the corresponding complex vector A. q=

© 2006 by Taylor & Francis Group, LLC

42

POLARIZATION

All the deﬁnitions that we gave in this section originated from quantities that were expressed in the principal axis reference frame, {x, y}. If one starts from a completely arbitrary real orthogonal frame, {x , y }, rotated by an angle δ with respect to the principal axes, one gets, analogously: q =

AL = q ej2δ AR

.

(2.33)

Hence, the quantity |q| is independent of the reference frame, and therefore has an intrinsic meaning. This is reconﬁrmed by the relationship which links the circular polarization ratio to the linear polarization ratio deﬁned in the previous section, namely: 1 − p . (2.34) q = 1 + p This is again a bilinear transformation, which maps the loci |q | = constant on the q complex plane into curves on the p complex plane along which the sign of Re(p ) remains constant. In more detail, we have: • the line Re(p ) = 0 is mapped onto the circle |q | = 1 (locus of linear polarizations); • the half-plane Re(p ) < 0 is mapped onto the region |q | > 1, where lefthanded elliptical polarizations are located. In particular, q = ∞ corresponds to left-handed circular polarization; • the half-plane Re(p ) > 0 is mapped onto the region |q | < 1, where righthanded elliptical polarizations are located. In particular, q = 0 corresponds to right-handed circular polarization. 2.8

Stokes parameters

In the last two sections, we stressed that some properties of the linear and circular polarization ratios are independent of the reference frame one adopts on the plane of the ellipse, Eq. (2.15). Those remarks show how to proceed further, searching for quantities that characterize polarization, and are invariant with respect to a change in the reference frame, or at least, if they can not be constant, change in simple ways. Among many possibilities that were suggested in the past decades, the so-called Stokes parameters became quite popular, being both elegant and useful at the same time. In order to link their deﬁnition to what we have seen so far, let us start from Eq. (2.15), which describes the elliptical trajectory of the point P = O+ A ∼, in the generic frame {x , y } (see Figure 2.1), where the components of the complex vector A are expressed by Eq. (2.20). From now on, the notation that we have been using so far may be simpliﬁed, just because we are looking for quantities that are insensitive to a change in the reference frame. Therefore, let us write the components of the complex vector A as follows, omitting primes, although in general we are not using the

© 2006 by Taylor & Francis Group, LLC

STOKES PARAMETERS

43

Figure 2.2 Geometrical illustration of the symbols X, Y and δ (orientation angle).

principal axes as the reference system: Ax = bx ejϕx ,

Ay = by ejϕy ,

ϕ = ϕ x − ϕy ,

jωt A ∼ as Re(A e ) , (2.35) where the parameters bx , by , ϕx and ϕy are positive real. The equation of the ellipse, Eq. (2.15), can then be rewritten, after a little manipulation, as follows: 2 2 y x x y + −2 cos ϕ = sin2 ϕ . (2.36) bx by bx by

From this, one sees that the angle δ between the major axis of the ellipse and the x axis, referred to as the orientation angle (see Figure 2.2), is given by: 2bx by tan 2δ = 2 cos ϕ , (2.37) bx − b2y while the ratio between the lengths Y and X of, respectively, the minor and major axes of the ellipse is related to the parameters of Eq. (2.36) as follows: 2bx by 1 π Y = ± tan γ , γ = arcsin 2 , (2.38) sin ϕ , |γ| ≤ X 2 bx + b2y 4 where: (i) in line with the previous deﬁnitions, the plus and minus signs apply for the right-handed and left-handed polarizations respectively, and (ii) the quantity γ is referred to as the eccentricity angle. The four Stokes parameters are deﬁned as follows: s1

= Ax A∗x − Ay A∗y = b2x − b2y

s2

= Ax A∗y + A∗x Ay = 2bx by cos ϕ ,

s3

= −j(Ax A∗y − A∗x Ay ) = 2bx by sin ϕ ,

s0

= Ax A∗x + Ay A∗y = b2x + b2y

,

.

(2.39)

All the information concerning the polarization ellipse of Figure 2.1 is concentrated in the two real numbers γ and δ, plus the sign in Eq. (2.38). One

© 2006 by Taylor & Francis Group, LLC

44

POLARIZATION

may then wonder why we introduced a set of four real parameters. The answer may be given as follows. First of all, squaring the expressions in Eq. (2.39), and summing, we ﬁnd: (2.40) s21 + s22 + s23 = s20 , which shows that in any case the four Stokes parameters are not independent. A usual way to express this link in words is to refer to S ≡ (s1 , s2 , s3 ) as the Stokes vector, and to state that |S| = s0 . Furthermore, if we want to keep information on polarization of a time-harmonic vector separate from that on vector amplitude, we may use normalized amplitudes, i.e., impose: s0 = 1 .

(2.41)

In this way, the number of independent Stokes parameters goes down to two, with a further ambiguity in the sign of the third one. This result matches precisely our statement after Eq. (2.39). However, let us emphasize that we were entitled to make the normalization (2.41) only because we were dealing with a purely monochromatic, time-harmonic ﬁeld. The deﬁnition of the Stokes parameters has a broader scope of validity. For example, it may be used for narrow-band signals, i.e., for time-harmonic carriers modulated by slowlyvarying envelopes, a subject with which we will deal in Chapter 5. Another very important example of application of Stokes parameters beyond the scope of the previous sections in this chapter is the representation of incoherent (or partially coherent) ﬁelds, that is, ﬁelds whose slowly varying envelopes are random functions of time. In this case, the products of the type Aj A∗ j that appear in Eq. (2.39) have to be replaced by time averages, and therefore the results are strongly dependent on the assumptions one makes about the correlation properties of the ﬁeld. In all these cases, the diﬀerence between s0 and unity becomes a measure of the “depolarization” of the ﬁeld A ∼ . We advise the reader to return to these comments and reconsider them after studying the last chapter in this book which deals with statistical properties of e.m. ﬁelds. Once the Stokes parameters have been determined in a completely arbitrary reference frame, then it is straightforward to identify, starting from them, the polarization ellipse deﬁned with respect to the same frame. In fact, from Eq. (2.39), one ﬁnds: b2x =

s0 + s1 2

s0 − s1 2

,

b2y =

,

2γ = arcsin

,

ϕ = arctan

s3 s2

,

(2.42)

or, alternatively: 2δ = arctan

s2 s1

s3 s0

,

X 2 + Y 2 = s0

.

(2.43)

If the reference frame is rotated by an angle δ on the plane of the ellipse, then, as the reader may check as an exercise, two parameters (namely, s3 and s0 ) are invariant, while the other two obey the simple transformation rule: s1

© 2006 by Taylor & Francis Group, LLC

= s1 cos 2δ + s2 sin 2δ

,

´ SPHERE THE POINCARE

45

Figure 2.3 Geometrical representation of the Stokes parameters: the Poincar´ e sphere.

s2

= s1 sin 2δ − s2 cos 2δ

,

(2.44)

which is typical of a rotation of the {s1 , s2 } plane by an angle 2δ. The invariance of s3 with respect to this kind of basis transformation is connected to the fact that s3 measures the content of circular polarization in the vector A ∼ , as the reader can infer from the deﬁnition of s3 itself, and as will be reconﬁrmed in the next section. Many of the most interesting properties of Stokes parameters become simple and evident if we adopt a geometrical construction, which will in fact be dealt with in the next section. 2.9

The Poincar´ e sphere

The relationship in Eq. (2.40) between the Stokes parameters lends itself to an immediate geometrical interpretation. If we introduce, in real 3-D space, three mutually orthogonal axes labeled {s1 , s2 , s3 }, then Eq. (2.40) can be read as the equation of a spherical surface of radius s0 centered at the origin. This is usually referred to as the Poincar´e sphere (see Figure 2.3). For a purely monochromatic time-harmonic ﬁeld, as noted in the previous section, one may separate information on polarization from that on amplitude, normalizing the ﬁeld so that s0 = 1. In general, as the reader will be able to appreciate more easily after reading the ﬁnal chapter of this book, the radius of the Poincar´e sphere will convey information regarding the degree of polarization of the ﬁeld one is dealing with. Unit-radius will imply completely polarized ﬁeld, i.e., the state of polarization is purely deterministic and known at any time without any uncertainty. At the opposite extreme, the radius will

© 2006 by Taylor & Francis Group, LLC

46

POLARIZATION

go to zero for a completely unpolarized ﬁeld, i.e., for a ﬁeld whose state of polarization varies randomly in time, with a correlation time shorter than the response time of the instrumentation that measures it, i.e., shorter than the time scale over which the quantities in Eq. (2.39) are averaged. The reader can now easily grasp the simple relationship that exists between the position of a point on the Poincar´e sphere and the parameters of the ellipse traced out by the point P = O+ A ∼ . Indeed, the Eqs. in (2.43) indicate that latitude and longitude on the Poincar´e sphere equal 2γ and 2δ respectively, where γ and δ are respectively the ellipse eccentricity and orientation angles. Consequently: • the upper hemisphere (s3 > 0) is the locus of right-handed polarizations, and the lower hemisphere (s3 < 0) is the locus of left-handed polarizations; • the equator (s3 = 0) is the locus of linear polarizations; • the two poles (s1 = s2 = 0) represent circular polarizations, right-handed at the north pole, and left-handed at the south pole; • mirror symmetry with respect to the plane of the equator reﬂects that, in real 3-D space where the vector A ∼ is deﬁned, a clockwise rotation becomes counterclockwise, and vice-versa, if one reverses the sign of the coordinate axis orthogonal to the plane of the ellipse. We think it might now be simpler for the reader to understand why the Stokes parameters transform according to Eq. (2.44) when the reference frame on the plane of the ellipse is rotated by an angle δ. Purely circular polarizations are completely insensitive to such a change in the reference frame. Linear polarizations remain linear, but their orientation angle, i.e., the angle between the reference frame and the direction of polarization, changes when the axes are rotated. The periodicity in such a change must be π, and that is the source of the factor of two in front of δ in Eq. (2.44). Similarly, implications for elliptical polarizations can be deduced in a straightforward manner. 2.10

Evolution of polarization in a linear medium: Jones matrix

All the deﬁnitions that we gave so far, and all the operations that were described in this chapter, dealt with polarization as a local property. They referred to a vector which was deﬁned at a given point in real 3-D space. On the other hand, it is self-explanatory why, in a book on electromagnetic waves, it is of interest to study how the polarization of a vector wave evolves as it propagates. This subject can be dealt with in depth, grasping the physical meaning of the equations one is writing, only after one familiarizes oneself with the contents of the next three chapters. Nevertheless, it appears to be useful to present here, in a rather abstract form, a general formalism, which is consistent with what we saw in this chapter, and will be applicable to the subjects to be dealt with later on. Consider two points in real 3-D space, Pin and Pout , both situated on the same path along which a wave is propagating. Pin is “upstream,” Pout is

© 2006 by Taylor & Francis Group, LLC

EVOLUTION OF POLARIZATION

47

“downstream”. As we saw in the previous sections, polarization at each point is completely characterized by a complex vector with two nonvanishing components. For simplicity, let us suppose that the two vectors at Pin and Pout , Ain and Aout , respectively, lay in parallel z = constant planes. If the medium between Pin and Pout is linear, then the two vectors Ain and Aout must be linearly related to each other. If we adopt the formalism of Section 2.5, and use the same reference frame {x, y} (not necessarily the principal axes) at the two points, then the relationship between the vectors can be expressed in matrix form, as: J11 J12 Ainx Aoutx = . (2.45) Aouty J21 J22 Ainy The matrix denoted as (J) in Eq. (2.45) is referred to as the Jones matrix. If one changes the reference frame, rotating it by an angle δ (see again Figure 2.1), then the Jones matrix in the new system, (J ), is related to that in the old system by: (J ) = (R) (J) (R)−1 , (2.46) where (R) is the rotation matrix that appears in Eq. (2.24). The physical reasons why polarization may change as a wave propagates along a linear system are not simple to understand at this stage. As previously mentioned, they will be clariﬁed in the following chapters. However, it is rather intuitive that, independent of the causes, the two components of the vector A in may experience either a diﬀerential loss (or gain, in an active medium), or a diﬀerential phase shift. A simple example of the ﬁrst kind is an ideal polarizer, whose Jones matrix, in a reference frame where the y axis is along the blocking direction, is: 1 0 . (2.47) (J0 ) = 0 0 A simple example of the second kind is a so-called quarter-wave retardation plate, i.e., a device which introduces a delay of π/2 radians on one component (say the one along y) with respect to the other. Then (except for an undeﬁned factor of unit modulus in front of the matrix, which represents an irrelevant phase shift, equal for all components), the Jones matrix reads: 1 0 . (2.48) (Jπ/2 ) = 0 j When two or more linear systems are cascaded, along the path of a given wave, then it is evident from the deﬁnition that the Jones matrix of the whole system is the product, in the appropriate order, of the Jones matrices of the individual components. Some of the problems at the end of this chapter deal with these subjects. Another point, which is probably much less obvious, and therefore must be emphasized here, is that the Jones matrix does not always lend itself to the inverse operation, namely, to reconstruct the input vector polarization from that of the output vector. An example in this sense is provided by the ideal polarizer: the matrix in Eq. (2.47) cannot be inverted,

© 2006 by Taylor & Francis Group, LLC

48

POLARIZATION

as it is singular. Indeed, this corresponds to the physical fact that the output of an ideal polarizer is, by deﬁnition, linearly polarized in a given direction, irrespective of the input ﬁeld. Similarly, in general one can not infer that the Jones matrix remains invariant when a wave travels through the same device in the opposite direction, i.e., when the roles of the points Pin and Pout are interchanged. Such a statement will have to be checked against the so-called reciprocity theorem, to be proved in the next chapter. 2.11

Further applications and suggested reading

For further details regarding the complex (phasor) representation of time harmonic vectors, we refer to any text on mathematical physics and Section 1.3 of Yariv and Yeh (1984) for some pertinent comments as well as Section 1.4 of Azzam and Bashara (1987). For an in-depth treatment of polarized light including the Poincar´e sphere and Jones matrices, we refer to Azzam and Bashara (1987) — see also the introductory level optics treatment on pp. 48– 49 and 544–545 of Guenther (1990). We also refer to Kliger et al. (1990) and Chapter 5 of Yariv and Yeh (1984) for further applications in optics as well as Someda and Stegeman (1992) and references therein for advanced waveguide theory applications, and Galtarossa and Menyuk (2005) for a thorough coverage of the now popular subject of polarization mode dispersion (PMD) in optical ﬁbers. Problems 2-1

Derive the rule expressed by Eq. (2.25).

2-2 Show that the sign of the real part of the linear polarization ratio is an invariant under the transformation deﬁned by Eq. (2.26). 2-3 Show that, if the reference frame is rotated by an angle δ on the plane of the polarization ellipse, two Stokes parameters (s3 and s0 ) are invariant, while the other two obey the transformation rule expressed by Eq. (2.44). 2-4 Consider the following vector ﬁeld which corresponds to the magnetic ﬁeld of the so-called TE10 mode of a rectangular waveguide (see Chapter 7) and is deﬁned with respect to rectangular coordinates (x, y, z) for 0 < x < a, 0 < y < b a s H x =Asin(πx/a)exp(−γz),H y =0,H z =A(γπ/a)cos(πx/a)exp(−γz). a) First assume that γ is real. Show that in this case the vector is linearly polarized at any point, and ﬁnd its direction of polarization as a function of the x coordinate. b) Next assume that γ = jβ is imaginary. Calculate the linear polarization ratio and the circular polarization ratio of the vector as a function of x. Find the values of x, if any, for which the vector is linearly polarized, and

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

49

those for which it is circularly polarized. Distinguish between right-hand and left-hand circular polarizations, for a given sign of β. 2-5 Refer to the vector deﬁned in the previous problem, and let (see Chapter 7, Eq. (7.24)): √ γ = jβ = jω μ0 0 1 − (ωc /ω)2 . Discuss the shape and the orientation of the polarization ellipse as a function of ω, over the range ωc < ω < 2ωc , at the points x = a/4 and x = 3a/4. 2-6 Calculate the Stokes parameters, vs. ω, for the vector of the previous problem, assuming that the constant A is real. Study the corresponding curve on the Poincar´e sphere. 2-7 Consider the following vector ﬁeld, E, which corresponds to the electric ﬁeld of the so-called TE11 mode of a circular waveguide (see Chapter 7) and is deﬁned within the circle of radius a centered at the origin, on the z = 0 plane of a cylindrical coordinate frame, (r, ϕ, z) as Er = E0 J1 (1.84 r/a) (1/r) sin ϕ, Eϕ = (1.84/a) E0 J1 (1.84 r/a) cos ϕ, Ez = 0, where E0 is a constant which is in general complex. (For the deﬁnition of the Bessel function J and of its derivative J , see Appendix D). a) Show that this vector is linearly polarized at any point where it is deﬁned. b) Find how its direction of polarization changes along a circle r = constant, for example the circle r = a/2. 2-8 Write in component form, using cylindrical and cartesian coordinates, a vector, E1 , such that the vector E + E1 is circularly polarized at any point, where E is the vector deﬁned in the previous problem. 2-9 Consider the following vector ﬁeld, E, which is deﬁned in all 3-D space with respect to a spherical coordinate system (r, ϑ, ϕ) and corresponds to the electric ﬁeld of the short electric dipole (see Chapter 12): 1 M 1 e−κ r E = cos ϑ rˆ κ + jω0 2π r r2 1 e−κ r κ 1 M + 2 sin ϑ ϑˆ , + κ2 + jω0 4π r r r √ where κ = ω 0 μ0 , and M is a constant. For simplicity, retain only the term of order 1/r 2 in the r-component, and the term of order 1/r in the ϑcomponent. Find the linear polarization ratio and the axes of the polarization ellipse, as a function of ϑ, for ω = 2π 106 rad/s, r = 300 m. 2-10

Write the equation of the ellipse described, in the time domain, by the

© 2006 by Taylor & Francis Group, LLC

50

POLARIZATION

instantaneous vector of unit amplitude and angular frequency ω = 2π 108 rad/s √ which corresponds to the following Stokes parameters: S1 = 0, S2 = 1/ 2, √ S3 = −1/ 2. Find the polar coordinates of this point on the Poincar´e sphere. 2-11 Write the Jones matrix for a half-wave retardation plate, i.e., an element which introduces a phase delay of π radians between its principal axes. First, write the matrix in a reference system which coincides with its principal axes, as was shown in the text for the quarter-wave plate. Then, rewrite it in a reference system rotated by an angle ϑ with respect to the previous one. 2-12 Write the Jones matrix for the following combinations of cascaded elements. a) A quarter-wave plate followed by an ideal polarizer (with an arbitrary rotation between the axes of the two elements). b) An ideal polarizer followed by a quarter-wave plate (again with an arbitrary rotation). Show that, for a generic input state of polarization, the output states of the two combinations are not equal. Discuss the reason for this situation to occur. 2-13 A birefringent plate (see Chapter 6) is characterized, as a function of frequency, by the following Jones matrix: J11 = 1, J22 = exp(jωΔn/c), J12 = J21 = 0, where c = 3 108 m/s (speed of light in free space), Δn = 10−3 (index diﬀerence, or birefringence), = 1 mm (plate thickness). Find all the frequencies (or wavelengths, measured in free space) at which this device acts as a quarter-wave plate. 2-14 Suppose that the device studied in the previous problem can, in practice, be considered as a quarter-wave plate when the phase shift between J 11 and J22 deviates by less than 1% from its nominal value of ±π/2. Calculate the relative bandwidths, Δω/ωi , (i = 1, 2, ...), over which this tolerance is satisﬁed, around the nominal values ωi found in the previous problem. Discuss how these bandwidths behave, as ωi increases.

© 2006 by Taylor & Francis Group, LLC

References Azzam, R.M.A. and Bashara, N.M. (1987) Ellipsometry and Polarized Light. Paperback edition, Elsevier, Amsterdam, Chapters 1–2. Galtarossa, A. and Menyuk, C.R. (Eds.) (2005) Polarization Mode Dispersion. Springer, New York. Guenther, R.D. (1990) Modern Optics. Wiley, New York. Halmos, P.R. (1958) Finite-Dimensional Vector Spaces. Van Nostrand Reinhold, New York. Kliger, D.S., Lewis, J.W. and Randall, C.E. (1990) Polarized Light in Optics and Spectroscopy. San Diego, Academic. Someda, C.G. and Stegeman, G.I. (Eds.) (1992) Anisotropic and Nonlinear Optical Waveguides. Elsevier, Amsterdam. Yariv, A. and Yeh, P. (1984) Optical Waves in Crystals: Propagation and Control of Laser Radiation. Wiley, New York, Sections 1.3 and 5.1.

51 © 2006 by Taylor & Francis Group, LLC

CHAPTER 3

General theorems 3.1

Introduction

Chapter 1 was devoted to the diﬀerential equations which the e.m. ﬁeld satisﬁes, and to some manipulations that allowed us to pass to other equations, either in the same unknowns or in other variables. The small number and the apparent simplicity of those equations could lead one to believe that they could be solved with one procedure, or at most with a few procedures. Still, the number of problems where Maxwell’s equations have exact closed-form solutions is very small. Furthermore, the parameters in Maxwell’s equations can vary by many orders of magnitude. For example, frequencies of technical interest range from zero to over 1014 Hertz; conductivity may vary between zero and 106 Ω−1 ·m−1 , and so forth. Therefore, approximate methods, which are necessary in many instances, are extremely diversiﬁed. As a result, ad hoc discussions are required in a multitude of cases. Indeed, that is what we will do starting with Chapter 4 and proceeding through until the end of the book. However before we begin this, it is advisable to prove some theorems which follow directly from the equations that were either postulated or derived in Chapter 1. Obviously, these theorems must be satisﬁed by all the solutions to be found later on. Their purposes are diﬀerent. Some of them clearly aim at “economy”. For example, the uniqueness theorem of Section 3.3 illustrates the conditions which are necessary in order to obtain one solution to Maxwell’s equations. Others (for example, the equivalence theorem of Section 3.5) are mainly pointers to approximate methods. In no case, however, does the matter discussed in this chapter yield an algorithm for solving Maxwell’s equations. For this, the reader should wait until the following chapters. In some cases, either the statements, or the proofs, depend on whether we consider a generic behavior in the time domain, or time-harmonic complex vectors. This will be emphasized by splitting those theorems into two subsections. Whenever that is not done, it is understood that the proofs in the two domains are substantially the same. 3.2

Poynting’s theorem. Wave impedance

Under this heading we will present, separately for the time and frequency domains, a relation between quantities with the physical dimensions of power. It has an immediate physical interpretation as an energy budget. Furthermore, 53 © 2006 by Taylor & Francis Group, LLC

54

GENERAL THEOREMS

Figure 3.1 The region of interest in the proof of Poynting’s theorem.

the next section will show that the same relation plays another important role, as the lemma of a subsequent theorem. The proofs presented here require a linear anisotropic medium, whose dyadics , μ, σ are constant with respect to time t and angular frequency ω. It is trivial to reduce them to an isotropic medium, whereas it can be a diﬃcult exercise to extend them to cases in which and μ depend on ω. This point will be reconsidered in Chapter 5. 3.2.1

Time domain

In the time domain, the theorem can be stated as follows. Consider a region V , in ordinary 3-D space, bounded by a closed, regular, time-invariant surface, S. Then, the total instantaneous power that the sources in V generate equals the sum of three terms: (i) the rate of change vs. time of the electromagnetic energy stored in V , (ii) the instantaneous dissipated power (i.e., the Joule eﬀect) in V , and (iii) the instantaneous power transmitted through the surface S. All those terms should be understood in the algebraic sense (i.e., they may be positive or negative), except for dissipation, which is non-negative by deﬁnition. In particular, the so-called sources can be either generators, i.e., imposed currents such that, as a time average, they supply a positive power to the ﬁeld, or utilizers, i.e., imposed currents which correspond to a time-averaged negative power, like for instance electrical motors. To prove the time domain version of the theorem, let us write, at any point in the region V (Figure 3.1), Maxwell’s equations; in the case of a linear anisotropic medium, they are straightforward generalizations of Eqs. (1.47) and (1.48): ∂H ∼ , ∇× E = −μ· ∼ ∂t ∂E ∼ + σ· E + J · ∇× H = ∼ ∼0 ∼ ∂t

© 2006 by Taylor & Francis Group, LLC

(3.1) .

(3.2)

POYNTING’S THEOREM. WAVE IMPEDANCE

55

We then take the scalar product (from the left) of the ﬁrst equation by H ∼, and of the second by E . Next we subtract term by term, and use the vector ∼ identity (C.4), which reads: a·∇×b − b·∇×a = −∇·(a×b) . We obtain the following relation, holding at any point in V : ∂ 1 1 ∇·(E · E . (3.3) ∼×H ∼ ) = − ∂t 2 H ∼ ·μ· H ∼ ·· E ∼ +2 E ∼ −E ∼ ·σ· E ∼−J ∼0 ∼ Let us then integrate it over the whole region V . Taking account of the time-invariance of V , applying Gauss’ theorem (A.8) to the integral of the divergence of E ∼×H ∼ , and ﬁnally reordering the terms, we obtain: 1 ∂ − H ·μ· dV = J0 · E H + E ·· E ∼ ∼ ∼ dV ∼ ∼ ∂t V 2 ∼ V + E E n dS , (3.4) ∼ ·σ· E ∼ dV + ∼×H ∼ ·ˆ V

S

where n ˆ denotes the unit vector normal to S, directed outwards from V . Eq. (3.4) proves the above statement as can be rapidly veriﬁed by considering the physical meaning of each term on which we now comment. We ﬁrst consider the term on the left-hand side. The imposed current density J0 can be ∼ regarded as the motion of a charge distribution of density ρ0 and velocity v0, ∼ · J0 is the i.e., J0 = ρ0 v0. Then, if we recall Eq. (1.2), we establish that E ∼ ∼ ∼ ∼ per-unit-time and per-unit-volume work of the Lorentz force, when a current with density J0 circulates. Therefore, when we change its sign, i.e., we con∼ sider − E 0, this represents the per-unit-volume instantaneous power that ∼·J ∼ is transferred (in the algebraic sense) by the source to the e.m. ﬁeld {E ∼, H ∼}, at the expense of an energy source that supports the imposed current density J 0, ∼ but that does not appear in Maxwell’s equations, and that for this reason is often referred to as being of a “non-electrical nature”. Passing to the right-hand side, the reader should already know from the time-independent (or slowly varying) case, that the quadratic forms (H ∼ ∼ ·μ· H E )/2 are, respectively, the instantaneous magnetic and electric )/2 and (E ·· ∼ ∼ energy densities stored in V , measured in joules per cubic meter. The quadratic form E ∼ is the instantaneous dissipated power density measured in watts ∼ ·σ· E per cubic meter. The remaining term, consisting of the surface integral of the vector product E n, is the power transmitted through the surface S, as ∼×H ∼ ·ˆ one can deduce from two observations. The ﬁrst is that inside the region V there are no other possible uses for electromagnetic power, in addition to those already discussed. The second is that the sign of this surface integral changes its sign (while its absolute value remains the same) if we deal with the region that is complementary to V .

© 2006 by Taylor & Francis Group, LLC

56

GENERAL THEOREMS

Figure 3.2 Schematic diagram of a reﬂector antenna.

The vector function: 2

P (W/m ) , (3.5) ∼=E ∼×H ∼ is referred to as the Poynting vector. Its physical dimensions are those of a power density, i.e., watts per square meter. However, it is not legitimate to state that it has a physical meaning point by point. Only its ﬂux through a closed surface is physically identiﬁed as a power. Nonetheless, in the following chapters we will encounter many cases in which we consider the ﬂux of P ∼ through open surfaces, S , and attribute to it the physical meaning of power transmitted through S . Whenever we do so, this will be justiﬁed by the fact that there exists another surface, S , such that (i) S + S is a closed surface, and (ii) the ﬂux of P ∼ through S is zero. A typical example in this sense is a so-called aperture antenna, similar to a parabolic-reﬂector antenna (see Figure 3.2). The total power received by the antenna is usually computed as the Poynting vector ﬂux through its “aperture,” because the contribution through the paraboloidal surface, consisting of a good conductor, is supposed to be negligible. 3.2.2

Frequency domain

In the complex-vector representation of time-harmonic vectors, Eq. (3.4) is replaced by a relation involving complex quantities. We assume that the reader is familiar with the concept of complex power∗ for a lumped-element linear one-port network (Figure 3.3). Poynting’s theorem can then be stated as follows. Consider a region V in ordinary 3-D space, bounded by a regular closed time-invariant surface, S. Then, the total complex power delivered by the sources present in V equals ∗ For a linear one-port electrical device, this quantity (having physical dimensions of power) can be deﬁned as follows: (i) its real part is the active power, i.e., the time-average over a period of the instantaneous power; (ii) its imaginary coeﬃcient is the so-called reactive power, i.e., the product of twice the angular frequency, 2ω, times the amplitude of the sinusoidal component (at an angular frequency 2ω) of the energy stored in the device, with sign + or − depending on whether magnetic or electric energy dominates in the storage.

© 2006 by Taylor & Francis Group, LLC

POYNTING’S THEOREM. WAVE IMPEDANCE

57

Figure 3.3 The concept of complex power in linear circuit theory.

the sum of three terms: (i) the reactive power corresponding to the periodic variation in the energy stored in V , (ii) the active power dissipated in V , and (iii) the complex power transmitted through the surface S. The proof is quite similar to that of the previous case, but the results should not be mixed; let us stress the main diﬀerences. We start from the complex form of Maxwell’s equations, and multiply scalarly from the left the ﬁrst equation by H∗ , and the complex conjugate of the second equation by E. After subtracting, we use the identity (C.4). Then, we integrate over the region V . Finally, we divide all terms by 2, in order to correctly identify their physical meaning. We obtain: ∗ H ·μ·H E··E∗ J∗0 ·E E·σ·E∗ = dV + 2jω − − dV 2 2 4 4 V V V E×H∗ ·ˆ n dS . (3.6) + 2 S It is straightforward to check that Eq. (3.6) corresponds to what we stated. Notice that the ﬁrst term on the right-hand side is real, while the second is imaginary; the left-hand side and the ﬁnal term are, in general, complex. Also note that, as we pointed out previously in Chapter 2, the complex Poynting vector: E×H∗ , (3.7) P= 2 is not related to the time-domain Poynting vector, Eq. (3.5), by the Steinmetz method. Indeed, the one-to-one and onto mapping of complex vectors into time-harmonic vectors commutes only with linear operations in either of the two spaces, while Poynting vectors, being deﬁned via a vector product, are not linear with respect to the e.m. ﬁelds. We conclude this section giving a broad-scope deﬁnition, whose meaning probably looks obscure at this stage; hopefully, however, its application will clarify it in the following chapters. Let u ˆ be a unit-vector in ordinary 3-D real space; for any ﬁeld {E, H}, we deﬁne as the wave impedance in the direction of u ˆ, at any point Q, the dyadic operator η uˆ (Q) deﬁned by the following relationship: u) . (3.8) u ˆ×E×ˆ u = η uˆ (Q)·(H×ˆ We will see in the following chapters that the dyadic η uˆ (P ) reduces to a

© 2006 by Taylor & Francis Group, LLC

58

GENERAL THEOREMS

scalar in most cases of practical interest, where the components of E and H that are transverse with respect to u ˆ are orthogonal to each other. Hence, we may think of it as a generalized “ratio” of the components of E and H that are transverse with respect to u ˆ. Its usefulness consists in allowing us to express the component of the Poynting vector parallel to uˆ (and consequently, the ﬂux of P through a surface) in terms of only one of the two vectors E and H. The role is then the same as that of an impedance (or an admittance) in network theory, quantities which allow us to express the complex power ﬂowing into a port in terms of either only the voltage, or only the current at that port (Figure 3.3). In particular, if E and u ˆ are orthogonal to each other, we have: 1 1 (H∗ ×ˆ u) · η uˆ · (H×ˆ u) = E · (η u∗ˆ )−1 · (ˆ u×E∗ ×ˆ u) , (3.9) 2 2 where, to simplify the notation, the dependence on the point Q is implicitly understood. We leave it as an exercise for the reader to ﬁnd how this expression should be modiﬁed when H and u ˆ are orthogonal. P·u ˆ=

3.3

Uniqueness theorem

Whenever one has to solve diﬀerential equations, it is advisable, before trying to determine the unknowns, to identify conditions that guarantee, when they hold, the uniqueness of the solution. In this section we shall do that with reference to Maxwell’s equations. After that, we will be sure that, if we can solve the problem, there are no risks of other solutions, which might contradict experimentally our analytical (or numerical) results. At the same time, we will identify the minimum set of conditions that guarantee the uniqueness of the solution. However, we will also see later that there are situations, not at all rare, in which Maxwell’s equations can be solved only if one starts with a set of data larger than this minimum. What the uniqueness theorem tells us, is that the “extra” data can not be independent of the essential data. 3.3.1

Time domain

In the time domain, the theorem can be stated as follows. Consider a region V , in ordinary 3-D space, bounded by a closed, regular, time-invariant surface, S. Maxwell’s equations therein have a unique solution {E ∼, H ∼}, if the following quantities are known: • the ﬁeld sources, i.e., the imposed current density, J0, at each point of V , ∼ and at each time t after an initial time t0 , • the initial conditions, i.e., the values of E ∼ and H ∼ at each point of V at the initial time t0 , and • the boundary conditions, i.e., the components of either E ∼ or H ∼ tangential to the surface S, at each point on S, for all times t > t0 . To prove this, we proceed ab absurdo, i.e., we start by assuming that there are two solutions, {E ∼ ,H ∼ } and {E ∼, H ∼ }, and then we show that they must be

© 2006 by Taylor & Francis Group, LLC

UNIQUENESS THEOREM

59

the same in order to satisfy all the assumptions. Maxwell’s equations being linear, the diﬀerence between the two solutions:

e , h , (3.10) ∼=E ∼ −E ∼ ∼=H ∼ −H ∼ is in turn an e.m. ﬁeld. The term J0 being identical for the two solutions im∼ plies that the ﬁeld (3.10) satisﬁes homogeneous Maxwell’s equations, without sources, in V . Then, if we apply Poynting’s theorem in Eq. (3.4) to {e∼, h ∼}, we ﬁnd: 1 ∂ (h ·μ· h 0 = ∼+e ∼ ·· e ∼) dV ∂t V 2 ∼ + e e n dS . (3.11) ∼ ·σ· e ∼ dV + ∼×h ∼ ·ˆ V

S

It was assumed that we know unambiguously either Etan or Htan at each ∼ ∼ point on S; this allows us to determine that the last term of Eq. (3.11) is zero. Integrating the remaining terms with respect to time, from initial time t0 , and exploiting the initial-condition assumption for t = t0 at each point of V , we ﬁnally write: t 1 (h (3.12) ∼ ·μ· h ∼+e ∼ ·· e ∼) dV + t dt V e ∼ ·σ· e ∼ dV = 0 . V 2 0 The ﬁrst term on the left-hand side of Eq. (3.12) is an integral of deﬁnite positive quadratic forms; the second term is an integral of a semideﬁnite positive quadratic form. Therefore, this equation can only be satisﬁed in the case: e h (3.13) ∼= 0 , ∼= 0 , as we intended to show. 3.3.2

Frequency domain

In the complex-vector representation of time-harmonic vectors, the uniqueness theorem has to be modiﬁed, slightly but in a non-trivial manner. The assumptions, under which we will show that Maxwell’s equations have a unique solution in a region V , are the following: • the ﬁeld sources, i.e., the imposed current densities, J0 , are known at each point in V , • the boundary conditions, i.e., the components of either E or H tangential to the surface S, are known at each point on S, and • the medium contained in the region V has everywhere a nonzero conductivity, i.e., it is a lossy medium. The proof of this theorem starts oﬀ as does the previous one, but it diverges from it at the moment of applying Poynting’s theorem to the diﬀerence

© 2006 by Taylor & Francis Group, LLC

60

GENERAL THEOREMS

between the two solutions. Indeed, for complex vectors, Poynting’s theorem reads: h·μ·h∗ e··e∗ e·σ·e∗ dV + 2jω − 0= dV . (3.14) 2 4 4 V V We omitted the surface integral, equal to zero, for the same reason as in the previous case. The real and the imaginary parts of the right-hand side of Eq. (3.14) have to be both equal to zero, so we have: e·σ·e∗ dV = 0 , (3.15) 2 V h·μ·h∗ e··e∗ − dV = 0 . (3.16) 4 4 V It is straightforward to establish that if and only if we have σ = 0, these two relations imply e = h = 0, as required. This explains why it was necessary to add the third assumption. If σ = 0, i.e., in a lossless medium, we see that ﬁelds not sustained by any imposed current can indeed exist in V , simply provided that, as required by Eq. (3.16), the time-averaged energies stored in V in magnetic and electric forms are equal. Such an e.m. ﬁeld is referred to as a resonant ﬁeld, because of its clear analogy with a √ lumped-element LC circuit at resonance, i.e., at an angular frequency (ω = 1/ LC). It should be pointed out that assuming a lossless medium often simpliﬁes Maxwell’s equations remarkably. For this reason, we will use it quite often in the following chapters, but we must reconcile it with the uniqueness theorem. There is no contradiction, in those cases where the solution for σ = 0 is indeed the limit of the solution for the lossy case, as σ → 0. In fact, the third assumption of the uniqueness theorem requires only a non-vanishing loss, without requiring a minimum value for this loss, which can thus be arbitrarily small. Finally, let us mention that the uniqueness theorem can also be proved for a ﬁeld deﬁned throughout 3-D space, provided that the assumption on either Etan or Htan on a closed surface S is replaced by the following postulate (which is known as Sommerfeld’s radiation condition): lim r|E| = lim r|H| = 0 ,

r→∞

r→∞

(3.17)

where r is the distance from an arbitrarily chosen origin 0 (not at inﬁnity) to the point where the ﬁeld is calculated. The proof can be undertaken by the reader by following the guidelines of the previous proofs, applying Poynting’s theorem to a sphere of radius r centered at 0, and then letting r tend to inﬁnity. Comments on how the last proof should be adapted to problems where there are structures of cylindrical shape and inﬁnite length, such as waveguides, are postponed until Chapter 7.

© 2006 by Taylor & Francis Group, LLC

RECIPROCITY THEOREM 3.4

61

Reciprocity theorem

This section is devoted to a theorem that we will only prove for complex vectors representing time-harmonic ﬁelds: we leave as an exercise for the reader the version in the time domain. Before the proof, we need to introduce a deﬁnition, whose scope of applicability includes also circuit theory. Let J 0 be an imposed current density, deﬁned in a region V of ordinary 3-D space. Let {E, H} be an electromagnetic ﬁeld, deﬁned in the same region V , but not necessarily the ﬁeld sustained by the source J0 . We deﬁne the reaction of the ﬁeld {E, H} on the source J0 , in the region V , as the quantity: E·J0 dV . (3.18) R(E, J0 ) = 2 V Eq. (3.18) has the physical dimensions of power. However, even when {E, H} is the ﬁeld sustained by J0 , it is not the complex power supplied by the source J0 ; the latter, in fact, does not involve J0 , but −J∗0 (see Section 3.2). The main use of this new deﬁnition results in the theorem that we will prove shortly, which establishes under what conditions two diﬀerent current densities, J0a and J0b , deﬁned in the same region V of 3-D space, play interchangeable roles. Subscripts a and b will be used also to distinguish the ﬁelds sustained by these two diﬀerent sets of sources. For brevity, let us use terms like “source a, source b, ﬁeld a, ﬁeld b,” without fear of ambiguity. The Lorentz reciprocity theorem states that the reaction of the ﬁeld a on the source b is equal to that of the ﬁeld b on the source a, provided that: • the region V contains either a linear isotropic medium, or a linear anisotropic medium whose constitutive dyadics are all represented by diagonal matrices in a real orthogonal reference frame; • either V is the whole 3-D space, and then the ﬁelds a and b satisfy the Sommerfeld’s radiation conditions in Eq. (3.17), or, alternatively, V is bounded by a closed surface S, on which any e.m. ﬁeld deﬁned in V satisﬁes a boundary condition of the following type, whose physical meaning will become clear throughout the following chapters: n , Etan = Z Htan ×ˆ

(3.19)

where n ˆ is the unit vector normal to S ; Z [ohms] may be a function of spatial coordinates, but is independent of the ﬁeld {E, H}, and is referred to as the wall impedance. For simplicity, we will write in full detail the proof only for an isotropic medium in a bounded region, leaving as exercises the extensions to the other cases just stated. Maxwell’s equations for the source a and the associated ﬁeld read: ∇×Ea

= −jωμ Ha

∇×Ha

= jωc Ea + J0a

,

(3.20) .

(3.21)

We take the scalar product of the ﬁrst equation with Hb , and of the second with Eb ; then, we add term by term the two thus obtained relations. After

© 2006 by Taylor & Francis Group, LLC

62

GENERAL THEOREMS

that, we swap the roles of a and b, and repeat the same operations. Finally, we subtract term by term the two equations obtained in the two cases. The ﬁnal result is: Hb ·∇×Ea −Ea ·∇×Hb +Eb ·∇×Ha −Ha ·∇×Eb = J0a ·Eb −J0b ·Ea , (3.22) while all the other terms cancel out. We now divide Eq. (3.22) by 2 and integrate it over the region V . Then we apply to the two pairs of terms on the left-hand side the vector identity a · ∇×b − b · ∇×a = ∇·(b×a). We obtain: Ea ×Hb Eb ×Ha J0a ·Eb J0b ·Ea − dV − dV . (3.23) ·n ˆ dS = 2 2 2 2 S V V We can now apply the cross-product cyclic permutation rule to both terms in the integral on the left-hand side of Eq. (3.23). After that, we apply Eq. (3.19) to the ﬁeld a in one of the two terms, and to the ﬁeld b in the other term. We conclude that the left-hand side of Eq. (3.23) is zero. This ﬁnal result is exactly what we wanted to prove. As mentioned at the beginning of the section, this theorem means that, when its assumptions hold, one can interchange roles between two distinct sets of sources. A source a, taken alone, gives rise to a certain electric ﬁeld Ea at some points, where, in a diﬀerent situation, there is another source b; then the electric ﬁeld Eb generated by source b at the points that are sites of source a can not be independent of the ﬁeld Ea . It must in fact satisfy the equality between the two integrals on the right-hand side of Eq. (3.23). This theorem expresses, in more physical terms, the same concept as reciprocity in an electrical network consisting of linear, passive and reciprocal two-terminal components (resistors, inductors, mutually coupled inductors, and capacitors). In elementary circuit theory, this concept is often expressed in rather simplistic terms as follows: if we interchange the positions of a generator and of a current meter in a network, the reading of the latter remains unchanged. This familiarity with reciprocity may induce the reader to believe that all the consequences of the theorem match easily with intuition. This is not always true. To show this, let us consider the example of a so-called Y -junction (Figure 3.4). In simple intuitive terms, it is the conﬂuence of three inﬁnitely long waveguide structures, each of which, we will suppose, can propagate only one solution of Maxwell’s equations (apart from an arbitrary complex multiplicative constant). What is easy to reconcile with intuition, is that one can design such a structure so that, when a wave is coming in from arm 1, it is split in equal parts between arms 2 and 3. What, on the other hand, is not obvious at ﬁrst sight, is that, for the same structure, when a wave is coming in from arm 2, then, because of reciprocity, only one half of its power, at most, can be coupled to the output of arm 1. The reader may come back to this example at the end of Section 3.9, and rediscuss it in terms of symmetries. By far and large, the majority of the subjects that are dealt with in this

© 2006 by Taylor & Francis Group, LLC

EQUIVALENCE THEOREM

63

Figure 3.4 A microwave Y -junction.

book satisfy the assumptions of the reciprocity theorem. However, this should not induce the reader to draw the erroneous conclusion that all electromagnetic phenomena of practical interest are reciprocal. Nonreciprocal behavior is present in nature as well as in man-made devices. The most classical example of the ﬁrst class is ionospheric propagation. As for the second one, we may quote two-port and three-port devices that are widely used at microwave and optical frequencies. Isolators are two-port devices which, in their ideal version, transmit a signal without any loss in one sense, and introduce an inﬁnite attenuation in the opposite sense. Circulators are three-port devices which, again in their ideal version, pass all the power entering one port to the next one, and nothing to the other one. The usefulness of such devices in communications or radar systems is quite evident. From the proof that we gave in this section, it should be easy to grasp that, in order to get nonreciprocal behavior, it is strictly necessary to have a suitable anisotropic medium, and also to exploit it in a convenient way. In the ionosphere, anisotropy is caused by the Earth’s static magnetic ﬁeld. Isolators and circulators exploit either magnetized ferrites, or some types of crystals. We postpone discussion of more details regarding these subjects until Chapter 6. 3.5 3.5.1

Equivalence theorem Preliminaries

The contents of this section are more complicated than the previous ones. The path that we will take might initially look unnecessarily cautious and slow, but eventually it will turn out to be beneﬁcial to the reader. In Section 1.4, a known term was introduced into Maxwell’s equations, in order to model the physical sources of a ﬁeld. This procedure could lead one to believe that, in every problem, one must ﬁrst identify a unique, well-deﬁned function of space coordinates, J0 , that corresponds to the actual physical sources. The aim of this section is to illustrate that indeed the contrary is true: in order to determine the ﬁeld in a given region of 3-D space, multiple choices exist for replacing assigned sets of sources with other equivalent sets. Equivalent means that, if the boundary conditions are left unchanged,

© 2006 by Taylor & Francis Group, LLC

64

GENERAL THEOREMS

Figure 3.5 The region of interest in the proof of the equivalence theorem, and its subdivision as an internal and an external region.

the ﬁeld remains the same as that given by the original sources. The reader may correlate these results with others in circuit theory, in particular, with Th´evenin’s and Norton’s theorems on the equivalent generators (Kuo, 1966). All analogies between ﬁelds and circuits have deep logical roots, as circuits result always from an abstraction process that starts from the ﬁelds. The contents of this section apply only to media that are strictly linear, as all proofs involve superposition, and with losses, arbitrarily small but not strictly zero, as required by the uniqueness theorem of Section 3.3. The phrase “turn oﬀ the generators in a region of space R” will be used to indicate that, starting from a current density not identically zero in R, we pass to another one, which diﬀers from the previous one only because we impose J0 ≡ 0 in R. We will denote with R that region of 3-D space where the e.m. ﬁeld on which we elaborate is deﬁned, and with C the boundary of R. On C, we assume that the boundary conditions are such that, once we specify all the sources J 0 within R, then the ﬁeld is unique everywhere in R. Furthermore, we will assume that all the sources are localized within an internal subset of R, so that (see Figure 3.5) there are an inﬁnite number of regular closed surfaces completely contained inside R such that if S is any of these taken at random, then we have J0 = 0 in the region contained between S and C. We use the symbol Re to denote the portion of R external to S, and Ri to denote the portion internal to S. Of course, R can also be the whole space, in which case C is at inﬁnity and the ﬁeld satisﬁes the Sommerfeld’s radiation conditions in Eq. (3.17). Suppose that we aim only at determining the ﬁeld in the region Re , which does not contain sources. This case is quite frequent in practice. For example, we might be dealing with the far ﬁeld of a transmitting antenna, or with a diﬀraction problem. Then, according to the uniqueness theorem of Section 3.3,

© 2006 by Taylor & Francis Group, LLC

EQUIVALENCE THEOREM

65

Figure 3.6 The equivalence theorem, in the version which makes use of a perfect magnetic conductor.

any set of sources, situated either in Ri or on S, and capable of giving rise to the correct tangential component of H on the external face of S, would generate in Re a ﬁeld {E, H}, equal to that sustained by the true sources J0 . Recalling what we said in Section 1.6, we note that the problem may be reduced to that of being able to guarantee Htan = 0 on the internal face of S. In fact, once that is done, the correct Htan on the external face is supplied by a source consisting of a surface electric current, on S, having a (linear) density: JS = n ˆ×H (A/m) , (3.24) where n ˆ is the unit vector normal to S, oriented towards Re . The internal face of S does not belong to the region Re where we are interested in the ﬁeld. Hence, in order to ensure Htan = 0 on this face, we can take the liberty of modifying the medium in Ri . In nature, a material capable of guaranteeing Htan = 0 on its surface does not exist, not even as a rough approximation. Nevertheless, it is a useful mathematical device to deﬁne such a medium, and call it a perfect magnetic conductor. What we have obtained in this way is schematically illustrated in Figure 3.6. The generators in Ri are turned oﬀ, the current density of Eq. (3.24) is imposed at any point on S, and a “perfect magnetic conductor” has been placed all along the internal face of S. The ﬁeld {E, H} coincides in Re with that of Figure 3.5. In Ri , as the medium has been modiﬁed, the ﬁeld there is no longer deﬁned. Our goal has been, to a large extent, reached, because the physical sources have been replaced with equivalent ones. But the price we paid for this, inventing the ideal magnetic conductor, looks high. Another approach may start

© 2006 by Taylor & Francis Group, LLC

66

GENERAL THEOREMS

Figure 3.7 The equivalence theorem, in the version which makes use of a perfect electric conductor.

from the fact that the tangential magnetic and electric ﬁelds play perfectly interchangeable roles in the uniqueness theorem (see Section 3.3). Therefore, there is another recipe in order to get an e.m. ﬁeld {E, H} in Re which coincides with that of Figure 3.5. Its ingredients are sources which ensure the correct value of Etan on the external face of S. To obtain this, one can use another mathematical device, also without experimental correspondence: the deﬁnition of a magnetic surface current, whose (linear) density at any point of S is: MS = E×ˆ n (V/m) . (3.25) It ensures the correct value of Etan on the external face of S, provided we can impose Etan = 0 on the internal face. This can be done by postulating that in that place there is a perfect electric conductor, i.e., a medium having conductivity σ = ∞ (a subject that will be discussed again in Section 4.11). What we have obtained in this way is schematically illustrated in Figure 3.7. Once again, the ﬁeld {E, H} in Re coincides with that of Figure 3.5, whereas the ﬁeld in Ri is not deﬁned. Advantages and disadvantages are not the same as in the previous case, but comparable in level. 3.5.2

Theorem

None of the above-mentioned procedures is really attractive, mainly because they require modiﬁcation of the medium in the region Ri , so that the ﬁelds therein are no longer deﬁned. Hence the importance of the following statement, that for the time being will be proved for a source-free region. If we turn oﬀ the generators J0 in Ri , let both electric and magnetic surface currents in

© 2006 by Taylor & Francis Group, LLC

EQUIVALENCE THEOREM

67

Figure 3.8 The equivalence theorem, in the version which does not require any perfect conductor.

Eqs. (3.24) and (3.25) circulate on S, and leave the medium unchanged over the entire region R, then: • in the region Re , we have the same ﬁeld {E, H}, as that of the situation we started from, which is summarized in Figure 3.5, and • in the region Ri , the ﬁeld is identically zero (E = H = 0). The proof consists just of observing that the vector ﬁelds Eeq and Heq , deﬁned as: Eeq = E , Heq = H in Re , (3.26) Eeq = 0 , Heq = 0 in Ri , satisfy Maxwell’s equations in the whole region Ri , provided their sources are Eqs. (3.24) and (3.25), and also satisfy the boundary conditions on C that were satisﬁed by {E, H} in the case shown in Figure 3.5. Applying the uniqueness theorem, we conclude that the Eqs. (3.26) are indeed the solution of the problem schematically illustrated in Figure 3.8, as we intended to prove. It should be self-explanatory why one usually refers to the two surface densities, in Eqs. (3.24) and (3.25), taken together, as sources equivalent to J0 for the ﬁeld in region Re . 3.5.3

Absorbing currents

Let us study now what happens if, on the surface S, there are electric and magnetic surface currents with densities Ja and Ma opposite to the current densities JS and MS of Eqs. (3.24) and (3.25), i.e.: n×H , Ja = −ˆ

Ma = −E×ˆ n .

(3.27)

If the boundary conditions on C are invariant with respect to a change in sign of the ﬁeld (i.e., if the boundary conditions are satisﬁed by {−E, −H} when

© 2006 by Taylor & Francis Group, LLC

68

GENERAL THEOREMS

Figure 3.9 The absorbing currents.

they are by {E, H}, as is the case, for example, if they are the Sommerfeld’s radiation conditions in Eq. (3.17)), then the currents in Eq. (3.27), if alone, generate a ﬁeld which is everywhere the opposite of Eq. (3.26). Suppose now that the original sources J0 in Ri and the surface currents in Eq. (3.27) on S are present together. Then, by superposition, we get that the ﬁeld in such circumstances is: Ea = E − Eeq = E

,

Ha = H − Heq = H

in

Ri

,

Ea = E − Eeq = 0

,

Ha = H − Heq = 0

in

Re

.

(3.28)

So, this ﬁeld {Ea , Ha } coincides then with the original one, {E, H}, in Ri , where the sources are located. It is identically zero in the complementary region Re . This entails the name of absorbing currents (when they are looked at from region Re ) for Eqs. (3.27). We can also imagine having gone from the original situation (Figure 3.5) to that described by Eqs. (3.28) (Figure 3.9) turning oﬀ the generators in Re (which were zero), and replacing them by the surface current densities in Eq. (3.27). As we saw just now, in doing this, the ﬁeld in Ri remained unchanged, while that in Re became zero. Note that this result proves precisely that the equivalence theorem is also true for a region (Ri in our case) where there are sources. Another correct way to read these results is in the following terms. The equivalent sources are always given by Eqs. (3.24) and (3.25), provided that one takes the unit vector n ˆ oriented always away from the region where the equivalence theorem gives an identically zero ﬁeld. Consequently, Eqs. (3.24) and (3.25) can be interpreted also as absorbing currents, for the region Ri . This allows us to conclude by underlining the perfect symmetry in the roles that the two regions Ri and Re play in the equivalence theorem. Again making use of superposition, it is straightforward to treat an even

© 2006 by Taylor & Francis Group, LLC

INDUCTION THEOREM

69

more general case, in which one starts from a situation where there are sources both in Re and Ri . We leave this as an exercise for the reader. 3.5.4

Final remark

It has been shown that equivalent electric and magnetic surface currents can be treated on a par with known terms, if Maxwell’s equations are to be solved in a part of the region R where the ﬁeld is deﬁned. Unfortunately, however, these equivalent currents are not true known terms. Their expressions contain E and H on the surface S, quantities which are, in their turn, unknown. However, to use equivalent currents requires knowledge of E and H not in the entire region R, but only on a broadly arbitrary closed surface S. Consequently, the practical usage of the theorem resides in those numerous problems (see in particular Chapter 13), where the ﬁelds E and H on a suitable surface S can be either measured experimentally, or estimated via a satisfactory approximation. One can then start from the corresponding surface currents as working assumption, and elaborate on them. We will reconsider this at the beginning of Chapter 11. 3.6

Induction theorem

In this section we will prove a corollary of the equivalence theorem. It leads to relations that are per se exact. However, their practical use is in general limited to approximate approaches, where the theorem helps one to state a suitable “ansatz”. Let us start from the same assumptions as in Section 3.5 for the medium (linear, with arbitrarily small losses but never lossless). Let R denote the region of real 3-D space wherein the ﬁeld {E, H} is deﬁned. Suppose that on its contour surface, C, boundary conditions are such that the uniqueness theorem of Section 3.3 holds in R. Consider then a regular closed surface, S; it subdivides R into two domains, R1 and R2 (Figure 3.10). There are no imposed currents on it. The subscripts 1 and 2 underline that, as we saw in the previous section, the roles of the regions that were initially called “internal” and “external,” are interchangeable. The induction theorem states that the ﬁeld {E, H} in R can be expressed as the sum of two ﬁelds: E = Ei + E

,

H = H i + H

,

(3.29)

which are characterized by the following properties: • the ﬁeld {Ei , Hi }, known as the incident ﬁeld (or inducing, or imposed ), is diﬀerent from zero only in one of two regions — to illustrate the idea, in R1 — and identically zero in the other (which implies that all its sources are in R1 ); • the other ﬁeld {E , H }, referred to as the induced ﬁeld, is non-zero in all of R. It is sustained by those sources of {E, H} that are not sources of

© 2006 by Taylor & Francis Group, LLC

70

GENERAL THEOREMS

Figure 3.10 The induction theorem: schematic illustration.

{Ei , Hi }, and moreover by electric and magnetic surface currents, localized on S, whose densities are, respectively: ˆ , Ji = H i × n

Mi = n ˆ × Ei

,

(3.30)

where the unit vector normal to S, n ˆ , is directed towards the region where {Ei , Hi } is non-zero. To prove all this, consider, in addition to the actual sources of the total ﬁeld {E, H}, four surface current densities, localized on S. Two of them are given by Eq. (3.30), while the other two are the opposite of Eq. (3.30). Incidentally, note how arbitrary the ﬁeld {Ei , Hi } in Eq. (3.30) is, in doing this operation. The superposition of these surface currents is obviously equal to zero and therefore the e.m. ﬁeld, in this new situation, is identical to the ﬁeld {E, H}. We now subdivide all the currents present into two sets, S0 and S . The linearity of the medium allows us to consider the total ﬁeld {E, H} as the sum of the ﬁelds due separately to S0 and S . In the set S0 , we include the sources (all located in R1 ) that are able to generate the ﬁeld {Ei , Hi } on the surface S, and furthermore the surface currents opposite to those of Eq. (3.30). In the set S , we put Eq. (3.30), as well as all the sources of {E, H} that were not included in the set S0 . To reach our conclusion, it is enough to observe that the opposites to Eq. (3.30) are the absorbing currents of the ﬁeld {Ei , Hi } (i.e., they assure Ei = Hi = 0 in R2 ). Thus, S0 and S respectively sustain ﬁelds {Ei , Hi } and {E , H } that meet all the properties that were listed when we stated the theorem. A diﬀerent point of view regarding the current densities in Eq. (3.30) can help clarify the practical impact of this theorem. Let us use diﬀerent names

© 2006 by Taylor & Francis Group, LLC

DUALITY THEOREM

71

and symbols for the ﬁelds induced in the two regions R1 and R2 , for example: {E , H } = {Er , Hr } ,

“reﬂected ﬁeld” in R1

{E , H } = {Et , Ht } ,

“transmitted ﬁeld” in R2

; .

(3.31)

One of the assumptions was that the initial ﬁeld {E, H} does not have sources located on S. Therefore, its tangential components are continuous across that surface. With our new symbols, this continuity can be expressed as: n ˆ × Ei

= n ˆ × (Et − Er ) ,

Hi × n ˆ = (Ht − Hr ) × n ˆ .

(3.32)

Comparing Eq. (3.32) with Eq. (3.30), we see that the surface currents deﬁned by Eq. (3.30) are those that are required in order to have, in the passage across S, discontinuities in the induced ﬁeld {E , H } equal to the right-hand sides of Eqs. (3.32). As we said at the beginning, this theorem turns out to be really useful in practice only when one is clever in estimating the inducing ﬁeld. This must be a good approximation to the exact solution, and at the same time be simple to handle, in order to make the subsequent calculations simple and rapid. 3.7

Duality theorem

The proof of the equivalence theorem of Section 3.5 required the new concept of magnetic currents, which, as already stated, is a useful mathematical tool, with no experimental signiﬁcance. So far, we have only made use of surface magnetic currents, but nothing prevents us from also considering magnetic currents in a 3-D domain. Their use will be thoroughly justiﬁed much later, in Section 11.5. With this, Maxwell’s equation for complex vectors in a linear, time-invariant, anisotropic (or, as a special case, isotropic) medium, become symmetrical (apart from the signs) not only with respect to the unknowns, but also with respect to the imposed terms: ∇×E = −jω μ·H − M0 ∇×H = jω c ·E + J0

,

(3.33)

,

(3.34)

where the imposed magnetic current density is denoted as M0 . The so-called duality theorem is simply the following observation. If {E, H} is a solution of Eqs. (3.33) and (3.34) in a region R of real 3-D space, then another e.m. ﬁeld, also deﬁned in R, but in a diﬀerent medium and with different sources, is obtained via the following set of formal substitutions (where implicit factors of unitary magnitude are understood, to provide suitable conversion of physical dimensions): E = −H

,

H = E c

© 2006 by Taylor & Francis Group, LLC

=μ

J0 = −M0

, ,

μ = c

, .

M0 = J0

,

(3.35)

72

GENERAL THEOREMS

The ﬁeld {E , H } obtained in this way is referred to as the dual (or the conjugate) of the ﬁeld {E, H}. The same word is used also for sources and media. Obviously, the dual of the dual of a ﬁeld corresponds to the ﬁeld itself. Readers who have always seen Maxwell’s equations in the MKS SI system of units may be very surprised by this theorem, and wonder whether it bears any adherence to physical reality. If they consult a textbook (e.g., Jackson, 1975) where the so-called Gaussian CGS system of units is adopted, and pay attention to the fact that in this system, both and μ in free space are equal to unity, then it becomes rather intuitive that two dual ﬁelds may, at least in some cases, be very similar to each other. For example, there are cases where duality corresponds just to a change from one polarization to the orthogonal polarization. 3.8

TE-TM ﬁeld decomposition theorem

Electromagnetic potentials, in Chapter 1, gave us the feeling that there is a large degree of arbitrariness in the ways to express an e.m. ﬁeld in a linear medium. In particular, we saw that it may be expressed in many ways as the sum of two or more ﬁelds, via “superposition,” and that it is possible to impose some pre-selected requirements on each of these ﬁelds. The equivalence and induction theorems, of Sections 3.6 and 3.7, are two examples of how these degrees of freedom may be exploited. Another example will be supplied by this section, whose results will be used very often in Chapters 7, 9 and 10. We will show that in a simply connected region — for the relevance of this restriction, see again the footnote in Section 1.8 — without sources, containing a homogeneous linear isotropic medium, any e.m. ﬁeld {E, H} can be expressed as the sum of two e.m. ﬁelds (that will be labeled, respectively, with a subscript TE and a subscript TM), which, with respect to a unit vector a ˆ chosen in a completely arbitrary direction, are Transverse Electric (TE) and Transverse Magnetic (TM), namely: {E, H} = {ETE , HTE } + {ETM , HTM } , ETE ·ˆ a=0 ,

HTM ·ˆ a=0 .

(3.36)

Furthermore, we will prove that each of these two ﬁelds can be obtained, via standard vector-calculus operations, from a scalar function of spatial coordinates that satisﬁes the scalar homogeneous Helmholtz equation. Before we start showing this, let us warn the reader that the proof of this theorem is enormously simpliﬁed if one is entitled to make two further assumptions. The ﬁrst assumption is that we are entitled to disregard transverse electromagnetic (TEM) ﬁelds, i.e., solutions of Maxwell’s equations having E·ˆ a = 0 and H·ˆ a = 0. The second assumption is that one can guess the functional dependence of the ﬁeld on the coordinate along the direction of a ˆ, leaving only a complex parameter to be determined later on in this function. However, at this stage, we believe, the reader is totally unprepared to grasp the physical meaning and the relevance of these simplifying assumptions. For

© 2006 by Taylor & Francis Group, LLC

TE-TM FIELD DECOMPOSITION THEOREM

73

this reason, we will present here a very general, but also very demanding proof. If the use that is going to be made of this theorem, by a reader or by an instructor, is just to introduce guided-wave propagation (Chapters 7 and 10), then the rest of this section may be replaced by the simpliﬁed proof given in Section 7.2. The general proof begins observing that, given the assumptions on the sources and on the medium, a scalar electric potential ϕ = 0 is a Lorentz gauge. This implies, via Eq. (1.79), that ∇ · A = 0, and consequently, the magnetic vector potential satisﬁes an equation of the type: A = ∇×W

.

(3.37)

So we introduce another vector W (with arbitrary divergence), which in turn must satisfy the following equation, as one can show by inserting Eq. (3.37) into Eq. (1.80): (3.38) ∇×(∇×∇×W + κ2 W) = 0 . Incidentally, from the following vector identity (which can be proved by the reader as an exercise): ∇×(∇×∇×q) = −∇×(∇2 q) ,

(3.39)

(where q is any vector), it follows that Eq. (3.38) can also be replaced by: ∇×(∇2 W − κ2 W) = 0 .

(3.40)

Now, separating the longitudinal and transverse parts with respect to the direction (say z) of the unit vector a ˆ, we get: ˆ W=a ˆ×W×a ˆ + (W · a ˆ) a ˆ = WTM + L a

.

(3.41)

The divergence of W being arbitrary (while its curl is ﬁxed) implies that the scalar L is only speciﬁed to within the addition of an arbitrary function g, which must satisfy only the restriction ∇×(gˆ a) ≡ ∇g × a ˆ = 0. At the same time, the vector WTM is only speciﬁed to within the addition of a irrotational ˆ. vector W , perpendicular to a Now let us express the total e.m. ﬁeld as the sum of a contribution due ˆ. As for the latter, it may be immedito WTM , and another one due to L a ately shown that it is transverse electric (TE). In fact, from Eqs. (1.66) and (3.37), and from the vector identity (B.6), applied in the present case in which a ˆ = constant, we have: ˆ) = −jω∇L × a ˆ , ETE = −jωATE = −iω∇×(L a

(3.42)

ˆ = 0, as we wanted to show. and this relation requires ETE · a It is rather simple to prove now that L satisﬁes the scalar homogeneous Helmholtz equation: (3.43) ∇2 L − κ2 L = 0 . Actually, Eq. (3.43) does not follow necessarily from Eq. (3.40), but, as we said before, L contains an arbitrary additive function g, and this is enough for us to impose that the function L satisfy Eq. (3.43).

© 2006 by Taylor & Francis Group, LLC

74

GENERAL THEOREMS

At this point, we are able to prove that the ﬁeld contributed by WTM is transverse magnetic (TM). Namely, if we take into account Eq. (3.43), we see that WTM must also obey Eq. (3.40) and thus Eq. (3.38). Exploiting then the above mentioned existence of an arbitrary irrotational vector W , we can impose that WTM obeys the following equation: ∇×∇×WTM + κ2 WTM = 0 .

(3.44)

Incidentally, the steps from Eq. (3.38) to Eq. (3.44) are formally identical to those from Eq. (1.64) to Eq. (1.68), and thus equivalent to imposing ∇·WTM = 0. From Eqs. (1.60), (3.37) and (3.44), we get: HTM = (1/μ) ∇×ATM = (1/μ) ∇×∇×WTM = −(κ2 /μ) WTM

, (3.45)

a = 0, because WTM is perpendicular to a ˆ by construction. which yields HTM ·ˆ This is what we intended to prove. The last thing that remains to be shown is that the TM ﬁeld can also be obtained from a scalar function T satisfying the homogeneous Helmholtz equation: (3.46) ∇2 T − κ2 T = 0 . To do this, it is suﬃcient to invoke the duality theorem of Section 3.7, and start again from the very beginning of the present theorem, using the electric vector potential F instead of the magnetic vector potential A. The TE and TM ﬁelds interchange their roles, whence the conclusion. Summarizing, with the help of a few steps that follow simply from some vector identities of Appendix B, we obtain the following expressions for the TE ﬁeld: ETE HTE

= −jω∇L × a ˆ , 1 1 ∇×ETE ∂ ∂ = ∇L = ∇L , = a+ a+ −∇2 Lˆ −κ2 Lˆ −jωμ μ ∂z μ ∂z (3.47)

where L is a solution of Eq. (3.43). Dually, for the TM ﬁeld, we have: HTM

= −jω∇T × a ˆ ,

ETM

=

∇×HTM 1 = jωc c

ˆ− ∇2 T a

∂ ∇T ∂z

=

1 c

a− κ2 Lˆ

∂ ∇T , ∂z (3.48)

where T is a solution of Eq. (3.46). Comparison between the ﬁrst of Eqs. (3.48) and Eq. (3.45) illustrates how the scalar function T is connected with the vector WTM that was used in the ﬁrst part of the present proof, namely: WTM =

© 2006 by Taylor & Francis Group, LLC

1 ∇T × a ˆ . jωc

(3.49)

SPATIAL SYMMETRIES. REFLECTION OPERATORS

75

This relation also reveals where the residual liberty was hidden that was exploited when we imposed that T obeys Eq. (3.46). Indeed, Eq. (3.49) shows that the z-dependence of the function T is completely irrelevant for calculating the ﬁeld of Eq. (3.45), because the longitudinal component of ∇T disappears, when we take the cross product with the unit vector a ˆ. Hence, there was an arbitrary component of ∇T , up to when T was forced to satisfy Eq. (3.46). 3.9

Spatial symmetries. Reﬂection operators

Geometrical symmetries of the region where an electromagnetic ﬁeld is deﬁned, if properly exploited, can make its determination signiﬁcantly simpler. This statement has been known for a long time, but it became particularly attractive in recent times, when interactive numerical methods became widely used. Indeed, symmetries may permit great reductions in memory allocation and in processing time. The simplest case, as well as the most exploited one, is a symmetry plane. Other cases of considerable interest are symmetries with respect to an axis, and with respect to a point. To approach all these cases in a uniﬁed way, let us take as a starting point Maxwell’s equations written by components in a rectangular coordinate frame (x1 , x2 , x3 ). Our aim is to ﬁnd transformation rules which, when applied to one of their solutions, provide another solution, whenever a symmetry of one of those types is present. We will deal with media that are isotropic, but not necessarily homogeneous. Extensions to some classes of anisotropic media can be found in the literature, but the reader is advised to look for them only after studying Chapter 6. ˆi , Maxwell’s equations, in the generalExpanding the vectors as E = Σi Ei a ized form which also includes magnetic sources, are written by components, as: ∂E3 ∂E2 − ∂x2 ∂x3

= −jωμ H1 − M01

,

∂E1 ∂E3 − ∂x3 ∂x1

= −jωμ H2 − M02

,

∂E2 ∂E1 − ∂x1 ∂x2

= −jωμ H3 − M03

,

∂H3 ∂H2 − ∂x2 ∂x3

= jωc E1 + J01

,

∂H1 ∂H3 − ∂x3 ∂x1

= jωc E2 + J02

,

∂H2 ∂H1 − ∂x1 ∂x2

= jωc E3 + J03

.

(3.50)

Let P denote the generic point in the region R in which Eqs. (3.50) are

© 2006 by Taylor & Francis Group, LLC

76

GENERAL THEOREMS

satisﬁed. Consider now the following set of substitutions: P ≡ (x1 , x2 , x3 )

−→

P ≡ (x1 , x2 , −x3 )

,

E ≡ (E1 , E2 , E3 )

−→

E ≡ (E1 , E2 , −E3 ) ,

H ≡ (H1 , H2 , H3 )

−→

H ≡ (−H1 , −H2 , H3 ) ,

J0 ≡ (J01 , J02 , J03 )

−→

J0

M0 ≡ (M01 , M02 , M03 )

−→

M0 ≡ (−M01 , −M02 , M03 ) .

(3.51)

≡ (J01 , J02 , −J03 ) ,

One can verify immediately that if Eqs. (3.51) are applied all at once, then Eqs. (3.50) are unaltered. Therefore, if the point P belongs to the region R where Eqs. (3.50) are satisﬁed, then the vectors E (P ), H (P ) satisfy at the point P Maxwell’s equations, with sources {J0 (P ), M0 (P )} also located at the point P . Therefore, if the region R is symmetrical with respect to the plane x3 = 0, and if the boundary conditions are themselves symmetrical with respect to the same plane, Eqs. (3.51) deﬁne a new solution of Maxwell’s equations. We say that this new solution is obtained starting from the initial solution by reﬂection in the plane x3 = 0. It may be convenient to write Eqs. (3.51) in the following compact notation: {E , H } = R3 ∗ {E, H}

,

{J0 , M0 } = R3 ∗ {J0 , M0 } .

(3.52)

The operator represented by the symbol R3 ∗ is referred to as the reﬂection operator in the plane x3 = 0. Provided that the region R and the boundary conditions satisfy the appropriate symmetry conditions, also the quantities: {E, H } = −R3 ∗ {E, H}

,

{J0 , M0 } = −R3 ∗ {J0 , M0 } ,

(3.53)

which diﬀer from Eq. (3.51) only by a change in sign of all components, form a solution of Maxwell’s equations. Moreover, it can be veriﬁed immediately that R3 ∗ R3 ∗ = (−R3 ∗) (−R3 ∗) = I∗, where the last symbol stands for the identity operator. An e.m. ﬁeld which is invariant with respect to the application of the operator R3 ∗, namely, is such that in its entire domain of deﬁnition satisﬁes the following equalities: E1 (x1 , x2 , x3 )

= E1 (x1 , x2 , −x3 ) ,

E2 (x1 , x2 , x3 )

= E2 (x1 , x2 , −x3 )

E3 (x1 , x2 , x3 )

= −E3 (x1 , x2 , −x3 ) ,

H1 (x1 , x2 , x3 )

= −H1 (x1 , x2 , −x3 )

H2 (x1 , x2 , x3 )

= −H2 (x1 , x2 , −x3 ) ,

H3 (x1 , x2 , x3 )

= H3 (x1 , x2 , −x3 )

,

,

(3.54)

is called a ﬁeld of even symmetry (or, simply, an even ﬁeld) with respect to the plane x3 = 0. Analogously, a ﬁeld which obeys the relation {E, H} =

© 2006 by Taylor & Francis Group, LLC

SPATIAL SYMMETRIES. REFLECTION OPERATORS

77

−R3 ∗ {E, H}, i.e., satisﬁes a set of equations that diﬀer from Eqs. (3.54) only by the sign in each right-hand side, is referred to as a ﬁeld of odd symmetry (or, simply, an odd ﬁeld) with respect to the plane x3 = 0. If we let x3 = 0 in Eqs. (3.54), we see immediately that they imply E3 = 0 and H1 = H2 = 0 over the entire symmetry plane. Compare with what was said at the beginning of Section 3.5: the conclusion is that an even ﬁeld remains unaltered if an ideal magnetic conductor is introduced on the symmetry plane. Alternatively, if we let x3 = 0 in the relations that deﬁne an odd ﬁeld, then we establish in the same way that they imply, on the plane x3 = 0, zero tangential components of E, and zero normal component of H (E1 = E2 = 0, H3 = 0). Again comparing with Section 3.5, we conclude that an odd ﬁeld remains unaltered if an ideal electric conductor is introduced on the symmetry plane. A ﬁeld of odd symmetry, i.e., compatible with an ideal electric conductor wall, requires sources satisfying the relation: {J0 , M0 } = −R3 ∗ {J0 , M0 } .

(3.55)

This equation expresses in analytical terms the so-called method of images, which is probably known to the reader in the electrostatic case (Plonsey and Collin, 1961). Note that Eq. (3.55) involves a change in sign for some components (J01 , J02 , M03 ), passing from the “actual” sources to the “image” sources. It will be well-known to the reader that any scalar function of a single variable, deﬁned over an interval symmetrical with respect to the origin, can be expressed as the sum of its even and odd parts. Similarly, any e.m. ﬁeld, provided it is deﬁned in a region symmetrical with respect to the plane x3 = 0, may be split into its (i) even part:

{Ee , He } = (1/2) {E, H} + R3 ∗ {E, H} , (3.56) and its (ii) odd part:

{Eo , Ho } = (1/2) {E, H} − R3 ∗ {E, H}

.

(3.57)

Obviously: {E, H} = {Ee , He } + {Eo , Ho } . This property, if suitably exploited, is the key towards saving memory and computation time in numerical methods (see comments at the beginning of this section). However, since diﬀerent components in Eqs. (3.51) are treated diﬀerently in respect of their signs, Eqs. (3.56) and (3.57) are non-trivial extensions of the scalar case, and must be handled with care by a programmer. Let us go back to the beginning of this subject: it is evident that the coordinate x3 does not play a privileged role in Eqs. (3.50) with respect to the other two coordinates. Therefore, it is straightforward to deﬁne reﬂection operators with respect to the coordinate planes x1 and x2 . They will be denoted by R1 ∗

© 2006 by Taylor & Francis Group, LLC

78

GENERAL THEOREMS Table 3.1 Reﬂection operator multiplication table. I

R1

R2

R3

R1

R2

R3

R0

I R1 R2 R3

I R1 R2 R3

R1 I R3 R2

R2 R3 I R1

R3 R2 R1 I

R1 R0 R3 R2

R2 R3 R0 R1

R3 R2 R1 R0

R0 R1 R2 R3

R1 R2 R3 R0

R1 R2 R3 R0

R0 R3 R2 R1

R3 R0 R1 R2

R2 R1 R0 R3

I R3 R2 R1

R3 I R1 R2

R2 R1 I R3

R1 R2 R3 I

and R2 ∗, respectively. We leave it as an exercise for the reader to derive their explicit forms. All the previous considerations regarding even and odd parts, electric and magnetic walls, image sources, etc., can be repeated for these operators without any substantial change. Next we can deﬁne reﬂection operators with respect to coordinate axes. For example, the reﬂection operator with respect to the axis x3 is deﬁned by the following substitutions: P ≡ (x1 , x2 , x3 )

−→

P ≡ (−x1 , −x2 , x3 ) ,

E ≡ (E1 , E2 , E3 )

−→

E ≡ (−E1 , −E2 , E3 ) ,

H ≡ (H1 , H2 , H3 )

−→

H ≡ (−H1 , −H2 , H3 ) ,

J0 ≡ (J01 , J02 , J03 )

−→

J0 ≡ (−J01 , −J02 , J03 ) ,

M0 ≡ (M01 , M02 , M03 )

−→

M0 ≡ (−M01 , −M02 , M03 ) .

(3.58)

Finally, a reﬂection operator with respect to the origin can be deﬁned, as the following set of substitutions: P ≡ (x1 , x2 , x3 )

−→

P ≡ (−x1 , −x2 , −x3 ) ,

E ≡ (E1 , E2 , E3 )

−→

E ≡ (E1 , E2 , E3 ) ,

H ≡ (H1 , H2 , H3 )

−→

H ≡ (−H1 , −H2 , −H3 ) ,

(3.59)

J0 ≡ (J01 , J02 , J03 )

−→

J ≡ (J01 , J02 , J03 ) ,

M0 ≡ (M01 , M02 , M03 )

−→

M ≡ (−M01 , −M02 , −M03 ) .

Using Ri ∗ and R0 ∗ to denote the reﬂection operators with respect to the iaxis and to the origin, respectively, it can be shown that the seven so-deﬁned operators, plus the identity operator, form a set which is an Abelian group (e.g., Cracknell, 1968). Its elements are linked by the relationships summarized in Table 3.1. A simple and illustrative example of how useful symmetries can be is the fol-

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

79

lowing approach to a so-called directional coupler in waveguides, or striplines, etc., whose operation can be explained intuitively, without calculations. Although more formal deﬁnitions exist, let us consider here as a directional coupler a pair of inﬁnitely long transmission lines or waveguides, with a longitudinal symmetry plane, say x1 = 0 (see Figure 3.11). Because of symmetry, any ﬁeld in this structure can be decomposed into its even and odd part. Suppose now that the two lines or waveguides are “single-mode,” i.e., are designed in such a way, that each of them, if left alone, may support just one solution of Maxwell’s equations, uniquely characterized apart from a complex multiplicative constant. Then, at least as a ﬁrst approximation, the total ﬁeld in the coupler will be a superposition of the even (or in-phase) and odd (or out-of-phase) combinations of these “modes” of the individual waveguides. The even and odd combinations, corresponding to a magnetic and an electric wall in the x1 = 0 plane, respectively, are ﬁelds which satisfy Maxwell’s equations in two diﬀerent structures. Therefore, one should not be surprised if we state that they travel at diﬀerent velocities along the inﬁnitely long direction. Consequently, the total ﬁeld amplitude in each waveguide may evolve periodically, although the total power is conserved. The example shown in Figure 3.11 refers in particular to the case where all the power in one section was in the upper waveguide (for example, this was the “launching condition” at the input of the coupler). It shows that, perhaps in contrast with intuition, all the power may be transferred to the lower waveguide at suitable distances. The reader may try, as a demanding exercise, to explain in terms of symmetries (even and odd ﬁeld combinations) the reciprocity results for a Y -junction that we saw in Section 3.4. 3.10

Further applications and suggested reading

As all solutions of Maxwell’s equations must satisfy the theorems that form the core of this chapter, all the contents of the following chapters may be regarded as a set of further applications of the material in this chapter. In particular, Poynting’s theorem and the uniqueness theorem are quite pervasive, and therefore they may show up in any of the following chapters. The reciprocity theorem, as we already mentioned, will ﬁnd one of its main applications in Chapter 12, when we discuss the relationship between transmitting and receiving properties of a given antenna. Nonreciprocal electromagnetic phenomena, which occur when the assumptions of the reciprocity theorem are violated — namely, in some anisotropic media — will be introduced in Chapter 6. However, a much deeper knowledge of nonreciprocities, in terms of basic physics as well as of applications, is instrumental for those who intend to work in such areas as microwave devices and networks, optical telecommunications and sensors, ionospheric propagation, plasma physics. Many textbooks cover these subjects very well. The basic references can be found at the end of Chapter 6. The equivalence theorem is a fundamental starting point for retarded po-

© 2006 by Taylor & Francis Group, LLC

80

GENERAL THEOREMS

Figure 3.11 The principle of operation of a directional coupler.

tentials (Chapter 11) and for diﬀraction (Chapter 13). The induction theorem can prove very useful in the same context, but unfortunately space constraints oblige us to make its use minimal throughout this book. Interested readers can ﬁnd its applications in advanced books on diﬀraction, scattering, aperture antennas (e.g., Chapter 8 of Stutzman and Thiele, 1981; see also the list of references at the end of Chapter 13). Also, it may be useful to have a thorough understanding of the use of this theorem before getting involved in numerical methods (e.g., Ida, 1995; Press et al., 1992; Stutzman and Thiele, 1981, and references therein). As for the duality theorem, we said at the end of Section 3.7 that it may appear awkward, when presented, as herein, in the MKS system of units, where the orders of magnitudes of dual variables are, in most practical cases, quite diﬀerent. To appreciate why this theorem can be useful in practice, we insist on the point that the reader should deal, at least brieﬂy, with Maxwell’s equations written in the so-called Gaussian CGS system of units, where electric permittivity and magnetic permeability of free space are both equal to unity. These equations can be found, for example, in Jackson (1975). One important consequence was already pointed out in Section 3.7: there are many problems, in optics and in diﬀraction theory, where two dual ﬁelds diﬀer only in polarization. Another way to appreciate the potential uses of this theorem is to think, once again, of numerical methods for solving Maxwell’s equations. All

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

81

quantities handled by a numerical computer are adimensional, and must be normalized so that their numerical values fall within a suitable range. The TE–TM decomposition of an e.m. ﬁeld is the strategic starting point for waveguide theory, both at microwave and optical frequencies. The basic notions regarding guided-wave propagation, which are given in Chapters 7, 9 and 10, provide a very extensive set of examples of such applications. All of them, however, may be derived from the simpliﬁed proof to be given in Chapter 7. As we said in Section 3.8, the demanding proof that was given there broadens the theorem’s scope of validity. This entitles one to tackle more diﬃcult problems, compared to those of our chapters on waveguides. Advanced problems are dealt with in more specialized books, such as Collin (1991), Lewin et al. (1977), and Vassallo (1991). On these occasions, one needs ﬁelds that are TE and TM not with respect to a waveguide axis, but rather with respect to other directions. As for symmetries, several results which will be found in the following chapters are easy to reconcile with physical intuition if they are examined in terms of spatial symmetries of structures and, consequently, of the ﬁelds. The very well known example of the directional coupler, has already been mentioned, and will be brieﬂy reexamined in Chapter 8. Symmetries, as we pointed out in the previous section, are related to basic group theory notions. The advantages of a group-theory-based approach are beautifully outlined, for example, in Chapter 12 of Montgomery, Dicke and Purcell (1948), where many fundamental properties of microwave junctions are derived without getting into any detail of ﬁeld calculations. Similarly, fundamental properties of optical waveguides can be found in a very elegant way via group theory, as shown by Black and Gagnon (1995). For these reasons, the reader is strongly advised to come back and read again Section 3.9 after reading Chapter 10, particularly Section 10.7. Finally, we wish to stress that, from a tutorial viewpoint, it is very helpful for the reader to study the basic theorems of electric circuit theory, in parallel to these theorems on e.m. ﬁelds. Theorems in these two areas must overlap exactly, because circuit theory is another axiomatic theory whose postulates were derived from the same basic physical principles as those of our theory. Just to quote a few examples, uniqueness of voltages and/or currents in an electrical network may run into problems for perfectly lossless resonant circuits: compare with Section 3.3. Reciprocity is found in circuit theory to be intimately related with time invariance. Equivalent generator theorems (also called generalized Th´evenin’s theorems) are conceptually the same as the equivalence theorems of Section 3.5 (Corazza et al., 1969). The classical textbook by Newcomb (1966) remains a highly recommended starting point for those who look for advanced readings in circuit theory. Problems 3-1

For the ﬁeld of the TE10 mode of a rectangular waveguide (see Prob-

© 2006 by Taylor & Francis Group, LLC

82

GENERAL THEOREMS πx a

lems 1-23 and 2-4), E = Ey a ˆ y = E0 a ˆy sin exp(−jβz), H = Hx a ˆ x + Hz a ˆz , βE0 jE0 π πx πx with Hx = − ωμ sin a exp(−jβz), Hz = ωμ a cos a exp(−jβz), deﬁned for 0 < x < a, 0 < y < b, without sources, show that the ﬂux of the Poynting vector a) through the planes x = 0, x = a, y = 0, y = b, is equal to zero; b) through any plane z = constant, is real, and independent of z. Draw your conclusions. 3-2 Show that the results of part a) of the previous problem remain the same when jβ is replaced by γ, with real γ, in the argument of the exponentials and as a factor in the expression of Hx . On the other hand, this change modiﬁes deeply the conclusions of part b). Recalculate and draw suitable conclusions. 3-3 Calculate the wave impedance in the z direction, for the ﬁelds of the previous two problems. What happens to these wave impedances if the sign of β (or γ) is changed ? 3-4 The following ﬁeld represents a purely standing wave in free space, ˆy = E0 sin(2βz) a ˆ y , H = Hx a ˆx = to be studied in Chapter 4: E = Ey a E0 j η0 cos(2βz) a ˆx , where η0 = μ0 /0 . a) Write the Poynting vector. b) Find for which values of L the following equation is satisﬁed: L L 0 μ0 2 |E| dz = |H|2 dz . 4 4 0 0 c) Does this ﬁeld obey the uniqueness theorem ?

3-5 In the time domain, two ﬁelds, {E ∼ ,H ∼ } and {E ∼ ,H ∼ }, have identical sources and satisfy identical boundary conditions, but diﬀerent initial conditions. Show that, if the medium is lossy (i.e., for σ = 0), it must be limt→∞ (E ∼ −E ∼ ) = limt→∞ (H ∼ −H ∼ ) = 0, where t denotes time. 3-6 Consider a region, limited by two parallel conducting planes and extending to inﬁnity in all directions parallel to these planes. How should the Sommerfeld radiation conditions be modiﬁed, to ensure uniqueness of a ﬁeld deﬁned in this region ? (Hint. Introduce polar cylindrical coordinates.) 3-7 The ﬁeld of Problem 3-1 is incident on a source which consists of the following surface electric current density, imposed on the z = 0 plane: ˆy sin kx x [A/m] J0 = J a a

for 0 ≤ x ≤ a , 0 ≤ y ≤ b .

a) Calculate the reaction of the ﬁeld on this source, as a function of kx .

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

83

b) Discuss, in particular, what occurs when kx = nπ/a, with integer n, and suggest a physical interpretation of these results. 3-8 Two sources, equal to that of Problem 3-7, are separated by a distance L such that exp(−jβL) = −1. What is the total reaction of the ﬁeld of Problem 2-1 on this compound source ? Comment on the result. 3-9 In the deﬁnition of reaction, replace the imposed current densities with their complex conjugates. Retain all the other assumptions of the reciprocity theorem. Prove that a “modiﬁed” reciprocity theorem, involving “reactions” modiﬁed in this way, holds only if the medium is strictly lossless. Is this a meaningful, physically acceptable result ? (Hint. Reconsider what was said, about lossless media, after the uniqueness theorem.) 3-10 Consider the ﬁeld of Problem 3-1, deﬁned for −∞ < z < +∞. a) Calculate its equivalent electric and magnetic currents on the plane z = 0, i.e., the current densities on that plane such that the ﬁeld remains unchanged for z > 0, and is identically zero for z < 0. b) Calculate its absorbing currents on the plane z = 0, i.e., the current densities on that plane such that the ﬁeld remains unchanged for z < 0, and is identically zero for z > 0. 3-11 Repeat Problem 3-10, for a ﬁeld which diﬀers from the previous one in that β is replaced by −β. Comment on the results. 3-12 Consider two e.m. ﬁelds. The ﬁrst one is identical to that of Problem 31 for 0 < x < a, 0 < y < b, and identically zero outside that region. The second one has the same analytical expressions, which apply now for −∞ < x < +∞, −∞ < y < +∞. Calculate absorbing currents at x = 0, x = a, y = 0, y = b, which transform the second ﬁeld into the ﬁrst one. Compare the results with those of Problem 123. 3-13 Superimposing the ﬁeld of Problem 3-1 and a ﬁeld, of equal amplitude, which diﬀers from the former one in that β is replaced by −β, we get (assuming ˆy = 2E0 a ˆy sin πx ˆ x + Hz a ˆz , with that E0 is real): E = Ey a a cos(βz), H = Hx a βE0 jE0 π πx πx Hx = 2j ωμ sin a sin(βz), Hz = 2 ωμ a cos a cos(βz), for 0 < x < a, 0 < y < b. Calculate the equivalent currents in a generic plane z = constant, and discuss their behavior as functions of z. 3-14 A rectangular waveguide is deﬁned as an inﬁnitely long region of rectangular cross-section, ﬁlled by a homogeneous lossless dielectric and surrounded by a perfectly conducting wall (see Figure 7.4 in Chapter 7). Imagine

© 2006 by Taylor & Francis Group, LLC

84

GENERAL THEOREMS

a structure which consists of two identical waveguides, one above the other (i.e., 0 ≤ x ≤ a, −∞ < z < +∞ for both; 0 ≤ y ≤ b for waveguide 1, −b ≤ y ≤ 0 for waveguide 2). In the conducting wall between the two waveguides, there is a rectangular aperture S, at 0 ≤ x ≤ a, 0 ≤ z ≤ t. Assume that the “incident” ﬁeld, {E0 , H0 }, is deﬁned in waveguide 1, and coincides with that of Problem 3-1. Find the surface current densities of the induction theorem, Eqs. (3.30), on the aperture S. Next, suppose that between the two waveguides there are two rectangular apertures, each of the same size as the previous one, but with a distance d = π/(2β) between their centers. Calculate the current densities of the induction theorem on both apertures. What are the consequences which one should expect from this result, for the induced ﬁeld in region 2 ? 3-15 Write the dual of the ﬁeld of Problem 3-1, assuming, for simplicity, = 0 , μ = μ0 . 3-16 Consider the following e.m. ﬁeld, deﬁned for r < a in cylindrical coordinates: Er = Ez = 0, Eϕ = E0 J1 (3.83 r/a) exp(−jβz), Hr = E0 (β/ωμ) J1 (3.83 r/a) exp(−jβz), Hϕ = 0, Hz = j 3.83/(aωμ) E0 J0 (3.83 r/a) exp(−jβz), where E0 is a constant, and β is real (the so-called TE01 mode of a circular waveguide). (For the deﬁnition of Bessel functions, see Appendix D). Write explicit expressions for the components of the dual ﬁeld. 3-17

Consider again the ﬁeld of the previous problem.

a) Verify that it is TE with respect to the direction of a ˆz . b) Verify that its dual (see again Problem 3-16) is TM with respect to a ˆz . c) Find the corresponding pre-potential functions L and T , and ﬁnd the relationships which must be satisﬁed by β. (Hint. Exploit proportionality between longitudinal ﬁeld components and pre-potentials.) 3-18

Consider again the ﬁeld of Problem 3-1.

ˆy . a) Decompose it into its TE and TM parts with respect to a ˆ x and to a b) Carry out the same procedure for the ﬁeld given in Problem 3-16, and for its dual. 3-19 Show that the ﬁeld of Problem 3-13 (a mode of a so-called rectangular resonator, see Chapter 9) is: a) odd with respect to the planes x = 0, y = 0, z = 0; b) even with respect to reﬂection in the planes x = a/2, y = b/2, z = d/2, where d = π/β.

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

85

(Hint. Change reference frame, x = x − a/2, etc.) 3-20 Consider again a structure like that shown in Problem 3-14, but without any opening between the two waveguides. Consider a ﬁeld which is identical to that of Problem 3-1 in region 1, and identical to zero in region 2. Find its even and odd parts, with respect to the reﬂection operator in the plane y = 0. 3-21 Introduce a rectangular aperture between regions 1 and 2, as in Problem 3-14. Calculate the current densities of the induction theorem, assuming that the incident ﬁeld {E0 , H0 }, deﬁned now in the whole structure, is: a) the even ﬁeld as found in Problem 3-20; b) the odd ﬁeld as found in Problem 3-20. Comment on these results.

© 2006 by Taylor & Francis Group, LLC

References Black, R.J. and Gagnon, L. (1995) Optical Waveguides Modes: Symmetry, Coupling and Polarization. Elsevier, Amsterdam. Collin, R.E. (1991) Field Theory of Guided Waves. 2nd ed., IEEE Press, Piscataway, New Jersey. Corazza, G.C., Longo, G. and Someda, C.G. (1969) “Generalized Th´evenin’s Theorem for Linear n-Port Networks”. IEEE Trans. on Circuit Theory, 16 (4), 564–566. Cracknell, A.P. (1968) Applied Group Theory. Pergamon, Oxford. Ida, N. (1995) Numerical Modeling for Electromagnetic Non-Destructive Evaluation. Chapman & Hall, London. Jackson, J.D. (1975) Classical Electrodynamics. 2nd ed., Wiley, New York. Kuo, F.F. (1966) Network Analysis and Synthesis. 2nd ed., Wiley, New York, Section 7.2. Lewin, L., Chang, D.C. and Kuester, E.F. (1977) Electromagnetic Waves and Curved Structures. P. Peregrinus, Stevenage, UK. Montgomery, C.G., Dicke, R.H. and Purcell, E.M. (1948) Principles of Microwave Circuits. McGraw-Hill, New York. Newcomb, R.W. (1966) Linear Multiport Synthesis. McGraw-Hill, New York. Plonsey, R. and Collin, R.E. (1961) Principles and Applications of Electromagnetic Fields. McGraw-Hill, New York. Press, W.H., Flannery, B.P., Teukolsky, S.A. and Vetterling, W.T. (1992) Numerical Recipes in C: The Art of Scientiﬁc Computing. 2nd ed., Cambridge University Press, Cambridge, UK. Stutzman, W.L. and Thiele, G.A. (1981) Antenna Theory and Design. Wiley, New York. Vassallo, C. (1991) Optical Waveguide Concepts. Amsterdam, Elsevier.

87 © 2006 by Taylor & Francis Group, LLC

CHAPTER 4

Plane waves in isotropic media 4.1

Separability of variables in the homogeneous Helmholtz equation

The ﬁrst three chapters have provided a general framework. The next six chapters will be devoted to studying electromagnetic waves in regions which contain no sources, i.e., to the fundamentals of propagation. In this view, this section is devoted to some general remarks regarding the solution of the homogeneous Helmholtz equation. The preferential role that we attribute to this equation is related to the attempt made in this book to give a uniﬁed approach, based on electromagnetic potentials, to propagation (Chapters 4–10) and to radiation (Chapters 11–14). Note, though, that propagation phenomena could be studied by solving Maxwell’s equations directly, without using potentials. However, this loss of a uniﬁed approach would not entail any beneﬁt, in terms of less mathematical complexity. The content of this section applies to many of the partial diﬀerential equations of mathematical physics. For this reason, some of the following statements will look very generic, and possibly too vague, to the reader. To grasp their full meaning, we suggest to the reader to come back to this point after reading the rest of the present chapter. One of the most commonly used, and most deeply studied, techniques to solve partial diﬀerential equations, is the so-called method of separation of variables. Essentially, it consists of breaking a partial diﬀerential equation (PDE) into a set of ordinary diﬀerential equations (ODE’s), which can be solved separately from one another, by isolating each independent variable in a separate equation. Examples, presented in this chapter as well as in some of the following ones (Chapters 5, 7, 9, 10), will clarify many details on how this procedure works. However, before we proceed, it is worth brieﬂy discussing the scope of applicability of the technique. When an equation, such as the Helmholtz equation, contains partial derivatives with respect to ordinary coordinates in real space, then the main limitation stems from the fact that the equation is separable only in a few, welldeﬁned systems of orthogonal curvilinear coordinates. The phrase “the equation is separable” means that there is at least one set, S, of solutions obtained by separation of variables, which is dense enough, to allow one to express any possible solution of the PDE as a linear combination of the solutions belonging to S. Another way to express the same concept is to say that S is a complete set of solutions. 89 © 2006 by Taylor & Francis Group, LLC

90

PLANE WAVES IN ISOTROPIC MEDIA

The type of diﬃculties that may arise in practice from this limitation, can be understood as soon as one considers that solutions of any diﬀerential equation must always satisfy suitable boundary conditions. Indeed, it is easy to grasp that formal complications will arise as soon as the boundary is not a coordinate surface (or a set of coordinate surfaces), in one of the reference frames where the equation is separable. Nevertheless, separation of variables is a very widely used method. To a large extent, this is due to the drawbacks of the only method which oﬀers a broad-scope alternative, the so-called Green’s function method. This technique consists essentially of writing the solution of a PDE as a volume (or surface) integral of the ﬁeld sources multiplied by suitable functions, whose meaning will be identiﬁed in Chapter 11. It will be exploited in Chapters 11, 12 and 13, which are mainly devoted to the inhomogeneous Helmholtz equation, i.e., to ﬁelds with sources. We wish to point out that the Green’s function method is applicable also in the homogeneous case, but usually, in this case, it leads to more complication, compared with separation of variables, and this justiﬁes our preference. The relationship between the two methods will be discussed brieﬂy in Chapter 11. 4.2

Solution of the homogeneous Helmholtz equation in Cartesian coordinates

Let us suppose that the whole three-dimensional (3-D) real space is ﬁlled by a linear, isotropic and homogeneous medium, with parameters , μ, σ, and that imposed currents are zero everywhere. This, which will be assumed throughout this chapter, is, strictly speaking, a physically meaningless system. Nevertheless, it may build a model which applies with an adequate accuracy to systems which contain suﬃciently wide source-free homogeneous regions. A quantitative estimate of the expression “suﬃciently wide” will be given soon, through the deﬁnition of a length which characterizes the spatial evolution of e.m. phenomena, the wavelength. We aim at determining the electromagnetic ﬁelds which can exist in these conditions, using the potentials (A, ϕ) that were deﬁned in Section 1.8. Note that, under the present assumptions, ϕ = 0 is a Lorentz gauge. Therefore, the whole problem we want to tackle reduces to solving the homogeneous version of Eq. (1.80), i.e.: ∇2 A − κ2 A = 0 , where:

σ κ2 = −ω 2 μc = −ω 2 μ − j ω

(4.1)

.

(4.2)

Let us introduce a cartesian coordinate system (x1 , x2 , x3 ). In such a reference frame, the components of the Laplacian of a vector (see Appendix B) are the scalar Laplacians of the components of the vector. This enables us to

© 2006 by Taylor & Francis Group, LLC

SOLUTION OF THE HOMOGENEOUS HELMHOLTZ EQUATION

91

write Eq. (4.1) simply by components as: ∇2 Ai ≡

∂ 2 Ai ∂ 2 Ai ∂ 2 Ai + + = κ2 Ai 2 2 ∂x1 ∂x2 ∂x23

(i = 1, 2, 3)

.

(4.3)

As the three scalar equations (4.3), for the three components of A, are independent of each other, and the medium is linear, without loss of generality we can take A parallel to one of the axes, for example: A = Ax ˆ1

.

(4.4)

Separation of variables, anticipated in the previous section, consists of making the ansatz that A(x1 , x2 , x3 ) be written as a product of three functions, each of them depending on just one coordinate: A(x1 , x2 , x3 ) = f1 (x1 ) f2 (x2 ) f3 (x3 )

.

(4.5)

Substituting Eq. (4.5) in Eq. (4.3), and dividing by A (= 0), we get: f1 f f + 2 + 3 = κ2 f1 f2 f3

,

(4.6)

where fj denotes the second derivative of fj . The right-hand side of Eq. (4.6) is a constant. The left-hand side consists of three terms, each of them depending on a diﬀerent coordinate. Therefore, Eq. (4.6) can be satisﬁed if and only if the three terms are separately constant: fi = Si2 fi

,

(4.7)

with the condition:

(4.8) S12 + S22 + S32 = κ2 . 2 The quantities Si are the separation constants for Eq. (4.6). Eq. (4.8) is referred to as the separation condition. As we anticipated in Section 4.1, Eqs. (4.7) are three ordinary, mutually independent diﬀerential equations. In particular, in this case we found three harmonic equations, whose general integral, as well known, can be written in the form: (4.9) fi = Fai e−Si xi + Fbi eSi xi , where Fai , Fbi are arbitrary (and in general, complex) constants. Hence, any solution of the type in Eq. (4.5) could be written as a product of three functions of the type in Eq. (4.9). However we may notice that Eq. (4.8) deals with S i2 , not with Si = ± Si2 . Therefore, there is no loss of generality if we write Eq. (4.5) in the following form: A = A0 e−(S1 x1 +S2 x2 +S3 x3 ) ≡ A0 e−S·r

,

(4.10)

where A0 is an arbitrary constant. We have introduced the position vector ˆi , and deﬁned: r = Σi xi x ˆi (4.11) S = Σ i Si x as the propagation vector, whose components in our cartesian reference frame

© 2006 by Taylor & Francis Group, LLC

92

PLANE WAVES IN ISOTROPIC MEDIA

are Si (i = 1, 2, 3). Due to the constraint in Eq. (4.8) on the separation constants, this vector in Eq. (4.11) has to satisfy: S · S = κ2 = −ω 2 μc

.

(4.12)

Let us stress that, since S is a complex vector, S · S = |S|2 = S · S∗ , i.e., Eq. (4.12) is not the norm of the vector S. Let us introduce, for future use, explicit symbols for the real and the imaginary part of the propagation vector: S=a+jk

,

(4.13)

where a is referred to as the attenuation vector, and k as the phase vector. This notation permits us to rewrite Eq. (4.12) as follows: a2 − k2 + 2j a · k = κ2

.

(4.14)

Implications arising from Eq. (4.14) will be the subject of the next section. We will conclude this section by giving explicit expressions for the ﬁelds E and H that follow from the vector potential in Eq. (4.10). We emphasize that these ﬁelds may be derived directly, solving the Helmholtz equation for E or H by the method of separation variables. We leave this to the reader as an exercise. From Eq. (1.66), with ϕ = 0, it follows that: E = −jω A = E0 e−S·r

,

(4.15)

with E0 = −jω A0 x ˆ 1 = E0 x ˆ1 , which means that E0 is essentially an arbitrary constant. From Eq. (1.60), or from Maxwell’s equation for ∇ × E, using the vectorial identity (C.6), it follows that: H =

A0 −S·r 1 ∇ × A0 e−S·r x ∇ e ×x ˆ1 ˆ1 = μ μ

= −

S×E S × E0 −S·r S×A = = e μ jωμ jωμ

.

(4.16)

Note that the penultimate term in Eq. (4.16) gives a relation between E and H which is insensitive to the direction of the vector A. This means that this relationship between the electric and the magnetic ﬁeld vectors is always valid, and is not restricted to the case of a linear polarization, which was the restrictive meaning of Eq. (4.4). The last comment is a good starting point for a short discussion on how one should superimpose three solutions of the kind we just found, each of which gives one component of the vector A in the cartesian reference frame that we introduced before. From now on, unless the opposite is explicitly stated, we will restrict ourselves to superpositions where all three components of the vector A have the same propagation vector, S, i.e., to solutions of Eq. (4.1) of the kind: (4.17) A = A0 e−S·r ,

© 2006 by Taylor & Francis Group, LLC

PLANE WAVES: TERMINOLOGY AND CLASSIFICATION

93

where now A0 is a constant complex vector. The corresponding electromagnetic ﬁeld {E, H} is then: E = E0 e−S·r

,

H=

S × E0 −S·r e jωμ

,

(4.18)

where E0 = −jω A0 . Note that the previous steps provide a nontrivial generalization, compared with Eq. (4.4). Indeed, now we are in a position to change the reference frame in a completely arbitrary way. Should one be obliged to deal with a solution of Eq. (4.1) where the components Ai have diﬀerent propagation vectors (but always satisfying Eq. (4.12)), this ﬁeld could be thought of as a superposition of solutions of the type (4.17). A ﬁeld represented by Eq. (4.18) does not satisfy Sommerfeld’s radiation conditions. Indeed, for some directions of the position vector r (for example, when it is antiparallel to the attenuation vector), Eq. (3.17) is not satisﬁed. Therefore, the waves we just found (to be called plane waves) do not satisfy the uniqueness theorem, not even in the case of a lossy medium. On the other hand, this was an expected result, as we were looking for nontrivial solutions (i.e., solutions which do not coincide with E = H = 0) of Maxwell’s equations. However, in Section 4.7, where we use plane waves as a basis to expand more general ﬁelds, we will show that the solutions we just derived, although individually violating the uniqueness theorem, nevertheless can, as a set, be used in the construction of ﬁelds that satisfy rigorous physical constraints, such as Sommerfeld’s radiation conditions. Finally, inserting the second of Eq. (4.18) in Maxwell’s equation for ∇ × H, we get: S×E×S S×H =− , (4.19) E= −jωc ω 2 μc and, in the same way: S×H×S H=− . (4.20) ω 2 μc By similarities with real vectors, Eqs. (4.19) and (4.20) could make us think that the vectors E, H and S are three mutually orthogonal vectors. However the three vectors are complex, so that more attention has to be paid to orthogonality. The subject is postponed to the end of the next section for a more accurate discussion. 4.3

Plane waves: Terminology and classiﬁcation

Inserting Eq. (4.13), where a and k are real vectors, into Eq. (4.18), it follows that the loci where |E| = constant (and therefore |H| = constant, |A| = constant, also) are given by: a · r = constant

© 2006 by Taylor & Francis Group, LLC

,

(4.21)

94

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.1 Intrinsic propagation, phase and attenuation constants of a homogeneous medium, represented on the complex plane.

i.e., are the planes perpendicular to the attenuation vector a. They are called constant-amplitude or equiamplitude planes of the wave. We will comment on the term “wave” in the next section. In the same manner, it can be proved that loci where Ei = constant (the symbol f indicates the phase of the complex number f ), Ei being one of the components of the E vector in an arbitrary coordinate frame, are given by: k · r = constant

,

(4.22)

i.e., are the planes, called constant-phase or equiphase planes, perpendicular to the phase vector k. The equiphase surfaces of a wave are also called wavefronts. As its wavefronts are planes, the ﬁelds we found in the previous section are called plane waves. Properties of plane waves change drastically, depending on whether a and k are parallel (i.e., constant-phase and constant-amplitude planes coincide), or not. This subject will now be dealt with in detail. As in Section 1.7, let: κ = α + jβ κ2

,

(4.23)

2

be the root of = −ω μc which belongs to the ﬁrst quadrant of the complex plane (see Figure 4.1). Then, certainly we have β > 0, α ≥ 0, with the equality holding only if the medium is lossless. The quantities we denoted with α and β are called, respectively, the intrinsic attenuation constant and the intrinsic phase constant of the homogeneous medium we are dealing with. We can then rewrite the real and the imaginary part of Eq. (4.14) as follows: |a|2 − |k|2 2a·k

= Re{κ2 } = −ω 2 μ , = Im{κ2 } = ωμσ

(4.24)

.

(4.25) 2

Comparing signs on the two sides of Eq. (4.24), it follows that |k| > |a|2 , and therefore |k| > 0. Thus, solutions of Maxwell’s equations of the type in Eq. (4.18) can never have a constant phase throughout the region where they are deﬁned.

© 2006 by Taylor & Francis Group, LLC

PLANE WAVES: TERMINOLOGY AND CLASSIFICATION

95

From Eq. (4.25), we will draw diﬀerent information, depending on whether the medium is lossless (σ = 0) or lossy (σ = 0). Let us discuss separately the two cases. A) Lossless medium. From Eq. (4.25), it follows: a·k=0 ,

(4.26)

a=0 ,

(4.27)

which is satisﬁed either for: or for: a = 0 ,

a⊥k

,

(4.28)

where ⊥ means that a is perpendicular to k. When Eq. (4.27) holds, then we have |E| = constant and |H| = constant over the whole 3-D space, i.e., any plane is a constant-amplitude plane, for this kind of wave. Conventionally, we may decide (mainly by analogy with the following case of a lossy medium) to call constant-amplitude planes those which coincide with the constant-phase planes in Eq. (4.22). Such a wave is called a uniform plane wave. When, on the contrary, Eq. (4.28) holds — we notice, incidentally, that in this case, because of Eq. (4.24), |a| < |k| — then the constant-phase planes are orthogonal to the constant-amplitude planes. Such a wave is called evanescent. Many among the following sections and chapters will underline the physical relevance of this kind of waves, which, at ﬁrst sight, the reader might consider just as mathematical curiosities. B) Lossy medium. From Eq. (4.25) it follows that: a·k>0 .

(4.29)

This tells us ﬁrst, that |a| = 0, and second, that the angle between a and k is less than π/2. In the special case where a and k are parallel, i.e., the constant-phase planes coincide with the constant-amplitude planes, the wave is still called a uniform plane wave. When the angle between a and k is greater than zero, the wave is called dissociated. It is advisable to return to Eqs. (4.19) and (4.20), and discuss them in the light of the results we just found. In order to avoid a substantial amount of needless formal complication, let us restrict ourselves, for the rest of this section, to linearly polarized waves, i.e., waves whose ﬁeld vectors are proportional to real vectors. As a preliminary, note that, as a consequence of what was shown in Section 2.3, the identity B = (B · vˆ) vˆ + vˆ × B × vˆ, which, as well known, holds when B and vˆ belong to the real 3-D space, must be modiﬁed in the following way for the case of complex vectors: v × B) × vˆ∗ B = (B · vˆ∗ ) vˆ + (ˆ

.

(4.30)

Comparing now Eq. (4.30) with Eqs. (4.19) and (4.20), one sees that E, H and S are mutually orthogonal if and only if the wave is linearly polarized and, furthermore, S is parallel to S∗ . It is straightforward to check that this

© 2006 by Taylor & Francis Group, LLC

96

PLANE WAVES IN ISOTROPIC MEDIA

happens if and only if a and k are parallel (including also the particular case where a = 0). Therefore, both in a lossless and in a lossy medium, E, H and S are mutually orthogonal only for linearly polarized uniform plane waves. ∗ On the contrary, this is never possible for evanescent and for dissociated waves. This agrees with geometrical intuition: when a and k are not parallel, the vector propagation “covers” by itself two dimensions in space, so that it is impossible for E and H to be both mutually orthogonal, and perpendicular to the propagation vector as well. Finally it is worth noticing that Eqs. (4.19) and (4.20) always imply the following relations: E·S=0 , H·S=0 , (4.31) and similarly for their complex conjugates: E∗ · S∗ = 0 ,

H∗ · S∗ = 0 .

(4.32)

However, note that, because of what was proved in Section 2.3, these relations we just found are not orthogonality relations for E, H and S. 4.4

Traveling waves. Phase velocity

In order to clarify the basic physical features of plane waves, it is useful to transform, according to the usual Steinmetz procedure, complex ﬁeld vectors into the corresponding time-domain vectors. It is easy to check that, under this viewpoint, the quantities A, E and H given by Eqs. (4.17) and (4.18) have identical properties; therefore, we may concentrate on one of them. For instance, let us consider the electric ﬁeld E and, for the sake of simplicity, let us assume E0 to be real. The corresponding vectorial function of the time-space coordinates is then: −S·r jωt = E0 e−a·r cos(k · r − ωt) . e (4.33) E ∼ = Re E0 e This shows that E varies sinusoidally in time, and, if we leave aside the factor exp(−a · r), in space also. There is a well-deﬁned relationship between time and space dependencies, as the argument of the cosine function mixes the two coordinates. To underline this character, a wave of the kind in Eq. (4.33) is called a traveling wave. The factor exp(−a · r) is, as we said before, an attenuation term, which can be sensed in any direction, except for those perpendicular to a, i.e., those that lie in the constant-amplitude planes. The rate of change of this attenuation is maximum in the same direction as the vector a. This factor is not essential, when discussing the physical features of a traveling wave. Therefore, for the sake of simplicity, from now on it will be omitted. The factor cos(k · r − ωt) describes an undulatory motion. This can be veriﬁed by tracking, as a function of time, those points where the cosine factor ∗ Do not confuse orthogonality between instantaneous time-harmonic vectors, and orthogonality between the corresponding complex vectors. To clarify this point, solve Problem 4.2, at the end of this chapter.

© 2006 by Taylor & Francis Group, LLC

TRAVELING WAVES. PHASE VELOCITY

97

Figure 4.2 Understanding the behavior of the phase velocity.

is constant, i.e., points where the argument of the cosine is constant, that correspond to: d(k · r − ωt) = 0 , (4.34) where d( ) denotes the total diﬀerential of the expression in brackets. Since ω and k are constant, Eq. (4.34) is equivalent to: k · dr − ω dt = 0 .

(4.35)

If we let: dr = u ˆ dr

,

(4.36)

where the unit vector u ˆ has an arbitrary direction, and originates at the generic point P = O + r, then Eq. (4.35) can be written in terms of the following quantity: ω dr = , (4.37) u) ≡ vf (ˆ dt k·u ˆ whose dimensions are those of a velocity. This quantity is called phase velocity in the direction of u ˆ. It is easy to prove that vf (ˆ u) is not the same as a velocity of a point motion in kinematics. Indeed, its dependence on the angle ϑ between the direction of u ˆ, which we choose arbitrarily, and that of k, which is ﬁxed, is: ω . (4.38) u) = vf (ˆ |k| cos ϑ Accordingly, vf (ˆ u) is inversely proportional to cos ϑ, whereas in kinematics the component of a velocity vector v along the direction deﬁned by u ˆ would be directly proportional to cos ϑ. It can be seen with the help of Figure 4.2 that Eq. (4.38) is the velocity at which, to an observer, located in P and looking in the direction of u ˆ, would appear to move the intersection between the generic u) the speed of constant-phase plane and the direction of u ˆ. In no case is vf (ˆ a point mass. It will be shown in the following section that it is not even the velocity at which electromagnetic energy moves. So, it should not come as a surprise if it can be larger than the speed of light, eventually arbitrarily large, as is the case for ϑ → π/2. Normally, if u ˆ is not explicitly speciﬁed, then it is understood that the expression “phase velocity” refers to the value of Eq. (4.37) in the direction

© 2006 by Taylor & Francis Group, LLC

98

PLANE WAVES IN ISOTROPIC MEDIA

of k, i.e., the minimum value of Eq. (4.38), ω vf = . |k|

(4.39)

Now, following the scheme of the previous section, let us ﬁnd what are the values that Eq. (4.39) can take. A) Lossless medium. For a uniform plane wave, since a = 0, Eq. (4.24) gives: √ |k| = ω μ = β = 2π/λ , (4.40) where λ is the wavelength in the given medium, at the angular frequency ω. The modulus of the phase vector equals the intrinsic phase constant β, so that Eq. (4.39) yields: 1 (4.41) vf = √ = c . μ Therefore, for uniform plane waves in lossless media, the phase velocity (which is, in general, a feature of the wave) equals that quantity which in Section 1.7 was called the speed of light, which is, in turn, a feature of the medium. √ For an evanescent plane wave, since a = 0, Eq. (4.24) yields |k| > ω μ = β, where β denotes, as usual, the intrinsic phase constant of the medium. Hence, the equality (4.41) is replaced by the following inequality: vf < c .

(4.42)

Due to Eq. (4.42), an evanescent plane wave in a lossless medium is said to be a slow wave. B) Lossy medium. As in this case a = 0, the inequality in Eq. (4.42) holds, and therefore only slow waves can exist. Anyway, as σ → 0, we get vf → c for uniform plane waves, so that in media with small losses (σ ω), a ﬁrst-order approximation yields vf c. A higher-order approximation can be evaluated by the reader as an exercise. 4.5

Standing waves

The previous sections were devoted to the fundamental physical aspects of a solution of the Helmholtz equation of the kind in Eq. (4.10), assuming always that it was the only ﬁeld that existed in the region of interest. From previous experience with diﬀerential equations, and from elementary experimental observations on motion of waves (such as, for example, vibrating strings, or waves on the surface of a liquid), the reader can expect that as soon as boundary conditions are imposed on an e.m. ﬁeld, they can induce very substantial diﬀerences, with respect to what we saw up to now. At least one of these cases deserves to be investigated immediately. Consider two uniform plane waves which overlap in a lossless unbounded medium. Each wave is of the kind in Eq. (4.18). Their propagation vectors are

© 2006 by Taylor & Francis Group, LLC

STANDING WAVES

99

S = jk and −S = −jk = S∗ , respectively, so: E = E1 e−jk·r + E2 ejk·r

.

(4.43)

For simplicity, assume that E1 and E2 are real. Then, the corresponding time-dependent vector is: jωt E ∼ = Re(E e ) = 2E2 cos(k · r) cos ωt + (E1 − E2 ) cos(k · r − ωt) .

(4.44)

Whereas the second term in the ﬁnal expression of Eq. (4.44) is of the same type as Eq. (4.33), which has already been investigated in Section 4.4, the ﬁrst term is completely diﬀerent, as space and time coordinates appear in two distinct functions, which multiply each other, instead of being combined linearly in the argument of the same function. In particular, the time evolution is in synchronism throughout the whole space: the phase delays from point to point, that characterize Eq. (4.33), have disappeared in Eq. (4.44). What we see now is an oscillation, whose amplitude varies in space, according to a cos(k · r) rule. The points where amplitude reaches its maximum (called crests) are given by: k · r = mπ

(m = 0, ±1, ±2, . . .) ,

(4.45)

whereas the points where the oscillation amplitude is zero (called nodes) are given by: k · r = (2m + 1) π/2 (m = 0, ±1, . . .) . (4.46) Eqs. (4.45) and (4.46) deﬁne two families of planes, perpendicular to k. Comparison with Eq. (4.22) shows that, for each of the two traveling waves that overlap, these planes would have been constant-phase planes. Both families of planes in Eqs. (4.45) and (4.46) are periodic in space, with a period: Δr =

π ˆ λ ˆ k= k |k| 2

,

(4.47)

where kˆ is the unit vector of k and λ is the wavelength, deﬁned in Eq. (4.40). Planes belonging to the two families alternate with a constant spacing Δr = kˆ λ/4. The ﬁrst term in Eq. (4.44) is referred to as a pure standing wave, while the whole Eq. (4.44) is called a partially standing wave, given by superimposing a traveling wave with a pure standing wave. In Chapter 8 we will elaborate further on this point. In particular, we will see that the ratio between amplitudes of the traveling wave and of the pure standing one conveys very signiﬁcant information on boundary conditions where partial reﬂection of the traveling wave occurs. A partially standing wave, Eq. (4.44), is the most general case of superposition of two uniform plane waves propagating in opposite directions. To relax this last constraint, and to consider a superposition of two plane waves (of equal frequency) traveling in any possible direction, the factors exp(∓jk · r)

© 2006 by Taylor & Francis Group, LLC

100

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.3 Two uniform plane waves interfering at an angle: decomposition into a standing-wave factor and a traveling-wave factor.

have to be replaced by exp(−jk1 · r) and exp(−jk2 · r), where |k1 | = |k2 |. We leave it to the reader to develop the remaining steps, and reach the following conclusions: • the ﬁeld, in general, is still a superposition of a standing wave and a traveling wave; • the traveling wave is essentially the same as the one in Eq. (4.44); • the term representing the standing wave consists of a factor which is periodic in space, like: k1 − k 2 ·r , (4.48) cos 2 and has to be multiplied by another factor, like: e−j(k1 +k2 )·r/2

,

(4.49)

so that the wave is a standing one if we look at it in the direction of the vector k1 − k2 , as shown by Eq. (4.48); but at the same time, as shown by Eq. (4.49), it travels along the direction of the vector k1 +k2 , i.e., along the line that bisects the angle between the phase vectors of the two traveling waves (see Figure 4.3). Let us now return to the case described by Eq. (4.44), and, taking only the pure standing wave term, ﬁnd its instantaneous magnetic ﬁeld. It can be calculated by applying Eq. (4.18) to each of the terms that form the electric ﬁeld. The ﬁnal result reads: jωt k × E2 −jk·r jk·r e e − e H = Re j ∼ jωμ = 2

k × E2 sin(k · r) sin ωt ωμ

.

(4.50)

It shows that the nodes of the H ∼ ﬁeld coincide with the crests of the E ∼ ﬁeld, and vice-versa. In the time domain, E is in quadrature with respect to H ∼ ∼. The reader can prove that, for a pure standing wave, the time-averaged electric and magnetic energies, stored in a cylindrical volume of arbitrary

© 2006 by Taylor & Francis Group, LLC

POYNTING VECTOR AND WAVE IMPEDANCE

101

cross-section, and of length λ/2 along the direction of the phase vectors k of the superimposed plane waves, are equal to each other. 4.6

Poynting vector and wave impedance

For a traveling wave of the kind in Eq. (4.18), which satisﬁes Eq. (4.19), the Poynting vector P = E × H∗ /2 can be evaluated by means of the vector identity (C.11). As Eq. (4.13) yields S + S∗ = 2a, one gets: E0 × (S∗ × E∗0 ) −(S+S∗ )·r e 2jωμ |E0 |2 ∗ E0 · S∗ ∗ −2a·r S + E0 e = − 2jωμ 2jωμ |H0 |2 H∗ · S S− 0 H0 e−2a·r . = 2jω 2jω

P = −

(4.51)

So, P is a function of space coordinates only through the factor exp(−2a·r). It is then constant on any constant-amplitude plane in Eq. (4.21). This would imply that the ﬂux of the P vector through any plane in space is inﬁnite. This is physically untenable. Indeed, Eq. (4.18) is perfectly legitimate as a mathematical solution of Maxwell’s equations, but, as it requires an indeﬁnitely extended homogeneous medium, there are limitations to its physical validity. When these constraints are violated, one gets meaningless results, as the previous example shows. This point will be further clariﬁed in Section 4.7 and in Chapter 5. As for the direction of P, it is not easy to recognize it from Eq. (4.51), because, as we said in Section 4.3, E and S, or H and S, are not always orthogonal. It is advisable to analyze separately the peculiar aspects that Eq. (4.51) can assume for the various types of plane waves, following the scheme given in Section 4.3. A) Lossless medium, uniform plane wave. The propagation vector S = jk is orthogonal to both electric and magnetic ﬁelds. Moreover, since Eq. (4.40) holds, it is: |E0 |2 ˆ k . (4.52) P= μ 2 Poynting vector is hence real, so the complex power carried by a uniform plane wave consists only of active power. It is simple to verify that under the same circumstances, at any point in space it is: 1 1 |E|2 = μ |H|2 , (4.53) 4 4 as required by Poynting’s theorem (see Section 3.2) whenever P is real. Let us now evaluate the wave impedance, as deﬁned in Section 3.2. Along ˆ which is also called the direction of the the direction of the phase vector, k,

© 2006 by Taylor & Francis Group, LLC

102

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.4 The geometry of a TM plane wave.

wave normal, and can be seen from Eq. (4.52) to coincide with the direction of P, the wave impedance can be evaluated using Eq. (B.11). It reads: μ ≡η . (4.54) Zkˆ = Thus, the wave impedance in this case coincides with the quantity η = μ/ which is referred to as the intrinsic impedance of the medium, this name coming from the fact that it depends only on the medium parameters, not on the features of a speciﬁc wave propagating in it. It is very useful to memorize that the free-space intrinsic impedance η0 = μ0 /0 equals 120π 377 Ω. As for wave impedances in directions which diﬀer from that of k, it is of primary interest to derive and emphasize two results, to be used in Section 4.8. Both deal with linearly polarized waves, for which the electric and magnetic vectors can be assumed to be real. The ﬁrst one refers to a unit vector u ˆ lying in the plane deﬁned by E and k, and is referred to as the transverse magnetic or TM case. Application of Eq. (3.8) gives then (see Figure 4.4): ZTM = η cos ϑ ,

(4.55)

ˆ is the angle between u where ϑ = (ˆ u, k) ˆ and the wave normal. The second result refers to the case where u ˆ lies in the plane deﬁned by H and k, and is referred to as the transverse electric or TE case. It yields: ZTE = η/ cos ϑ ,

(4.56)

ˆ (see Figure 4.5). where, again, ϑ = (u, ˆ k) B) For an evanescent plane wave in a lossless medium, Eq. (4.51) becomes more complicated. Also for evanescent waves, it will become clear in the following sections that there is a special interest in the case where E is orthogonal to S (a case that, by analogy with the previous discussion, is called TE), and in the case where H is orthogonal to S (TM). Under these circumstances, we ﬁnd, respectively: P =

© 2006 by Taylor & Francis Group, LLC

|E0 |2 −2a·r e (k + ja) 2ωμ

(TE) ,

(4.57)

POYNTING VECTOR AND WAVE IMPEDANCE

103

Figure 4.5 The geometry of a TE plane wave.

|H0 |2 −2a·r e (k − ja) (TM) . (4.58) 2ω In both these cases, the Poynting vector is complex. Its real part is parallel to k, and hence perpendicular to the constant-phase planes. Its imaginary part is parallel to a, and hence lies in the constant-phase planes. There is a change in the sign of reactive power, passing from the TE to the TM case. P =

C) Lossy medium. In both the TE and TM cases, Eq. (4.51) still becomes Eqs. (4.57) and (4.58), respectively. However, now k is no longer orthogonal to a. This implies, on one hand, that along every direction lying in the constantamplitude planes there is a real component of P, but, on the other hand, that the wave normal direction can never lie on the constant-amplitude planes. Therefore, in the k direction there is always a ﬂow of both active and reactive power. D) Standing waves. Let us stress that the Poynting vector for these waves cannot be obtained by superposition of results calculated separately for the traveling waves which build up the standing wave, since the Poynting vector is derived through a nonlinear operation, with respect to the ﬁelds (the outer product of two vectors). We will deal only with the case of a pure standing wave in a lossless medium. We leave it to the reader to consider the other cases as problems, to be solved more conveniently after reading Chapter 8. Let us start from: E = E0 e−jk·r + E0 ejk·r

,

(4.59)

where E0 is orthogonal to S = jk, and H can be derived by means of Eq. (4.50). One then gets: P=

|E0 |2 |E0 |2 j2k·r e sin(2k · r) kˆ − e−j2k·r k = j 2ωμ η

.

(4.60)

Therefore the Poynting vector is parallel, as in the case A), to the k vector of a traveling wave, but it is purely imaginary, and it is varying periodically in space. It goes through zero on the planes: mπ (m = 0, ±1, ±2, . . .) , (4.61) k·r= 2

© 2006 by Taylor & Francis Group, LLC

104

PLANE WAVES IN ISOTROPIC MEDIA

i.e. (see Section 4.5), both at the nodes of E ∼ and at those of H. ∼ Its modulus reaches its maximum on the planes at half way between two successive nodes. It is suggested that the reader complete these results with a careful analysis of reactive power and of the distribution in space of stored electric and magnetic energy, relating this subject with the discussion at the end of Section 4.5. The wave impedance in the wave normal direction is purely reactive, and periodic in space: Zkˆ = jη cotg (k · r) .

(4.62)

This is strictly analogous to situations involving transmission lines, to be dealt with in Chapter 8. The reader should reconsider this section after dealing, in particular, with the input impedance of lossless lines terminated on purely reactive loads. E) Waves interfering at an angle. The last case we wish to discuss brieﬂy is that which was schematically shown in Figure 4.3, and discussed in the previous section for what refers to amplitudes and phases of the two ﬁelds. Once again, we insist on the fact that the Poynting vector for the overall ﬁeld cannot be inferred in an elementary way from those of the two interfering waves. We will brieﬂy summarize the main ﬁndings for the case in which the two interfering waves have equal amplitudes, leaving their derivation to the reader as an exercise. In such a case: • neither the Poynting vector, nor the wave impedance (the latter taken along any direction in space) change, as one moves along the direction of the vector k1 + k2 , because exponential factors like Eq. (4.49) and its complex conjugate disappear in the calculations; • the Poynting vector and the wave impedance vary periodically along the direction of the vector k1 − k2 . The laws for their variations in space are easily derived from Eq. (4.60) and from Eq. (4.62), respectively, by replacing k with (k1 − k2 )/2; • the component of the Poynting vector along the direction of k1 + k2 , and the wave impedance in the same direction, are real at any point in space, indicating that active power is ﬂowing in that direction; • the component of the Poynting vector along the direction of k1 − k2 , and the wave impedance in the same direction, are imaginary at any point in space, indicating that reactive power is ﬂowing in that direction; • in general, the Poynting vector and the wave impedance are polarization dependent. A more demanding exercise is to repeat all these calculations for the case where the two interfering waves are of diﬀerent amplitudes. At this stage, the reader should know how to do it. However, we suggest to wait until after reading Chapter 8, because analogies with transmission lines will be a source of self-conﬁdence, for this type of exercise.

© 2006 by Taylor & Francis Group, LLC

COMPLETENESS OF PLANE WAVES 4.7

105

Completeness of plane waves

In this section, which is linked to the introductory remarks of Section 4.1, we shall investigate the possibility of representing a generic electromagnetic ﬁeld as a linear superposition of plane waves. In order to proceed towards this aim, it is necessary ﬁrst to state explicit requirements that this ﬁeld must satisfy. This sentence is meant to underline that the question of what ﬁelds can be actually expanded into plane waves, does not have a unique answer. The more restrictive the requirements one places on the ﬁelds, the easier to prove the completeness of plane waves. Here, our main task is to show merely that it is possible to make such an expansion. Thus, we will start from fairly relaxed requirements, and consequently give a rather simple proof. More demanding ones can be found in some of the books that are listed among our references. Let us consider ﬁelds deﬁned in a region R which contains no sources, and is ﬁlled by an isotropic and linear medium. Moreover, let us suppose that this medium is stepwise homogeneous, i.e., the parameters c and μ, in R, vary in steps, at most, a ﬁnite number of times, as functions of the space coordinates. As we saw in Chapter 1, this implies that any ﬁeld {E, H} that may exist in R satisﬁes two requirements: a) both E and H satisfy almost everywhere (i.e., excluding at most a set of measure zero) the homogeneous Helmholtz equation, ∇2 a = κ2 a, where κ2 is stepwise constant; b) all the components of E and H in a ﬁxed reference frame are piecewise constant, i.e., have at most a ﬁnite number of discontinuities of the ﬁrst kind. We will now add a third assumption, whose purpose is to restrict our proof to physically meaningful ﬁelds, actually to those ﬁelds which store a ﬁnite energy, and carry a ﬁnite power. We will suppose that there is at least one direction in ordinary 3-D space, labeled z, such that the surface integral: 1 μ |H|2 + |E|2 dn , (4.63) N 4 is ﬁnite when N is any plane orthogonal to z. This is equivalent to saying that the ﬁeld {E, H} is square integrable, over any plane orthogonal to z. This assumption is well matched to situations (such as those treated in Chapters 7 and 9) where propagation takes place essentially in one direction. However, it is not diﬃcult to grasp that even for multidimensional propagation phenomena (such as, for example, waves of spherical symmetry), the physical need that the total energy must be ﬁnite can be translated into a mathematical requirement of the kind (4.63) on an arbitrary plane. We will show now that, under these assumptions, the ﬁeld {E, H} can be expanded as a linear superposition of an inﬁnite number of plane waves. Let us introduce a rectangular reference frame, (x1 , x2 , x3 ), whose x3 axis is parallel to the direction that we called z. Consider the generic Ei component in this system. Due to the assumptions, it is possible to Fourier transform E i with

© 2006 by Taylor & Francis Group, LLC

106

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.6 The loci of plane waves on the complex plane of the propagation constant.

respect to the “transverse” coordinates x1 , x2 : +∞ +∞ 1 dx1 dx2 Ei ej(k1 x1 +k2 x2 ) Ei (k1 , k2 , x3 ) = 2π −∞ −∞

.

(4.64)

Moreover, since Ei satisﬁes the scalar Helmholtz equation: ∇2 Ei ≡ Σl

∂ 2 Ei = κ2 Ei ∂x2l

,

(4.65)

exploiting the correspondence, in the Fourier transformation, between the ∂/∂xl operator and multiplication times jkl , the transformed component Ei satisﬁes: ∂ 2 Ei = −κ2 Ei . (4.66) (k12 + k22 ) Ei − ∂x23 Rearranging the terms, and letting: h2 = κ2 + k12 + k22

,

(4.67)

we see that Eq. (4.66) can be written as an ordinary diﬀerential equation: d2 Ei = h2 Ei dx23

,

(4.68)

whose general integral can be expressed as the linear superposition of exp(∓hx 3 ), if h = constant, and as follows, if h varies (independently of x3 ): Ei = fi (h) e−hx3

,

(4.69)

fi being an arbitrary function of h. From Eq. (4.67), noting that k1 and k2 are real and range from −∞ to +∞, it follows that, if κ2 = −ω 2 μ is real and negative (as is the case in lossless media), then h2 ranges from −ω 2 μ to +∞. Consequently, h spans the domain that consists of the real axis of the complex plane, plus the segment of the √ imaginary axis that lays between ∓jω μ (see Figure 4.6).

© 2006 by Taylor & Francis Group, LLC

COMPLETENESS OF PLANE WAVES

107

The inverse Fourier transformation then gives: +∞ +∞ 1 2 2 2 Ei = dk1 dk2 fi κ + k1 + k2 e−jk1 x1 −jk2 x2 −hx3 , 2π −∞ −∞ (4.70) which expresses Ei as a linear superposition of plane waves, each of which has ˆ1 + jk2 x ˆ2 + hˆ x3 . a propagation vector S = jk1 x Making use of Fourier transforms, we have proved plane wave completeness in the sense of the so-called averaged linear approximation: the right-hand side of Eq. (4.70) converges towards Ei , when the integration interval goes to inﬁnity, almost everywhere, i.e., at most with the exception of a set of discrete points where Ei has step discontinuities. This ends the proof that it is possible to expand E as a linear superposition of plane waves. It remains to be proved that the same expansion also yields the magnetic ﬁeld. It would not be enough to repeat the same procedure, starting from the Helmholtz equation for H. What has to be proved is that H can be expanded into plane waves with the same coeﬃcients as E. This step is necessary in order to prove that the whole ﬁeld {E, H} can be expressed as a linear combination of plane waves. Let us substitute E, by components expanded like Eq. (4.70), into Maxwell’s equation ∇×E = −jωμH. To prove that H is the linear superposition of same plane waves as E, is equivalent to showing that we can interchange the order between integration over k1 and k2 , and partial derivatives with respect to the coordinates. This is typically a question of uniform convergence of the integrals such as Eq. (4.70). It certainly applies where the components of E are continuous, i.e., within each one of the homogeneous media, which build up the piecewise homogeneous medium. So, this proves that the magnetic ﬁeld H can be expanded into plane waves with the same expansion coeﬃcients as E, except, at most, for what happens at the discrete points where the medium is discontinuous. The problem of what happens at these discontinuities is beyond the scope of this book, and the interested reader is referred to more advanced texts (e.g., Morse and Feshbach, 1953). The results of this section deserve some further comments. 1) Eq. (4.70), together with Eq. (4.67), shows that, in order to expand a generic ﬁeld {E, H}, uniform plane waves are not suﬃcient, as they cannot span the whole range of values which are taken by h. In particular, in lossless media (κ2 = −ω 2 μ < 0, and real) we see that it is necessary to take into account the whole set of the evanescent waves whose attenuation vector is along z, in addition to the set of plane waves traveling in all directions in 3-D real space. This point must be stressed, as it diﬀers signiﬁcantly from what the reader may know about Fourier transforms in the time domain, where the conjugate variable (frequency) spans only the real axis. 2) We have demonstrated that a square-integrable ﬁeld (which is a physically meaningful one) can be expanded as a superposition of non-squareintegrable waves. In this way, we have reconciled with physical intuition

© 2006 by Taylor & Francis Group, LLC

108

PLANE WAVES IN ISOTROPIC MEDIA

plane waves which, although physically absurd if taken literally, nevertheless are very powerful and simple mathematical tools. 3) A procedure which makes use of the Fourier transform automatically provides a quantitative criterion, to decide when a single plane wave is acceptable, at least as an approximation. Indeed, the “conjugate variables” relation between x1 , x2 and k1 , k2 allows us to write a well-known inequality, which links the minimum ranges where Ei and Ei have a nonnegligible amplitude: Δki Δxi ≥ 2π

(i = 1, 2)

.

(4.71)

Therefore, in order to express a ﬁeld as a single plane wave (i.e., in order to consider Δki ki ), it is necessary for this ﬁeld to have its amplitude nonnegligible over a region wide enough to satisfy the following condition: Δxi

2π ki

.

(4.72)

The components ki of the k vector depend, of course, on the direction of k itself. Nevertheless, a good term for comparison, which does not depend on the direction of k, can be easily found, as: λ=

2π |k|

.

(4.73)

This quantity was deﬁned in Section 4.4, and is called the wavelength of the plane wave of angular frequency ω. 4.8

Reﬂection and refraction of plane waves

Plane waves lend themselves to the description of how e.m. ﬁelds behave at a sharpe discontinuity in the parameters of the medium. The validity of the results one gets there in terms of single plane waves has to be checked with the criterion that we outlined in the previous section. For the sake of extreme simplicity, let us suppose that the whole 3-D space is divided by a plane surface, which separates two homogeneous media, whose parameters are labeled with a subscript m = 1, 2. Moreover, suppose that the medium 1 is lossless (σ1 = 0). Note, in particular, that the uniqueness theorem and the induction theorem do not hold under these assumptions. However, physical intuition, based on experimental evidence, suggests testing whether continuity of the tangential ﬁeld components on the discontinuity plane can be satisﬁed by superimposing an incident and a reﬂected wave in region 1, and assuming a transmitted wave in region 2. Let us suppose that the incident ﬁeld is a uniform plane wave. Let us also introduce a cartesian reference frame (Figure 4.7), whose plane x1 = 0 is placed at the discontinuity between the two media, and whose x3 axis is chosen so that the phase vector of the incident wave, ki , lies on the x3 = 0 plane. All these assumptions translate into the following formulas: for the incoming

© 2006 by Taylor & Francis Group, LLC

REFLECTION AND REFRACTION OF PLANE WAVES

109

Figure 4.7 Schematic representation of incident, reﬂected and transmitted waves, and the reference frame.

wave: Ei

= E0i e−jki ·r

Hi

=

Er

= E0r e−Sr ·r

Hr

=

,

kˆi × E0i −jki ·r e η1

.

(4.74)

,

(4.75)

.

(4.76)

For the reﬂected wave: ,

Sr × E0r −Sr ·r e jωμ1

and, for the transmitted wave: Et

= E0t e−St ·r

Ht

=

,

St × E0t −St ·r e jωμ2

Let us check whether these waves can satisfy, over the entire x1 = 0 plane, the continuity conditions in Eqs. (1.43) and (1.44), which can be written in the following form, usually referred to as Fresnel’s equations: x ˆ1 × (Ei + Er ) = x ˆ 1 × Et

,

ˆ 1 × Ht x ˆ1 × (Hi + Hr ) = x

.

(4.77)

In particular, these equations must hold at r = 0, i.e., we must have: x ˆ1 × (E0i + E0r )

= x ˆ1 × E0t

,

x ˆ1 × (H0i + H0r )

= x ˆ1 × H0t

.

(4.78)

In Eq. (4.77), each term can be written as the product of a factor like E0i , E0r , E0t , and of an exponential function, according to Eqs. (4.74), (4.75) and (4.76). Then, Eqs. (4.77) and (4.78) look like a system of four linear

© 2006 by Taylor & Francis Group, LLC

110

PLANE WAVES IN ISOTROPIC MEDIA

homogeneous equations in the three unknowns E0i , E0r , E0t . For this system to have nontrivial solutions, it is necessary that at most two of these equations are linearly independent. This happens if the three exponential functions have ˆ2 +x3 x ˆ3 , as in this case Eqs. (4.77) the same argument, for any value of r = x2 x coincide with Eqs. (4.78). This provides the following formulas, which describe the geometrical relationships of the three waves: jki · r = Sr · r = St · r .

(4.79)

For the reﬂected wave, writing Sr = ar + jkr , Eq. (4.79) yields: ar · r = 0 ,

k r · r = ki · r .

(4.80)

This pair of equations cannot be satisﬁed by an evanescent wave. Indeed, ar being orthogonal to kr , from the ﬁrst equation it follows that kr should lie on the plane x1 = 0; but then (remembering also that ki does not belong to √ this plane) the second equation would imply |kr | < |ki | = β1 = ω μ1 1 , thus violating Eq. (4.24). Therefore, we can conclude that the reﬂected wave is a uniform plane wave, with |kr | = |ki | = β1 . The second of Eqs. (4.80), for r = x ˆ3 , gives kr · x ˆ3 = 0, showing that kr lies on the plane x3 = 0, deﬁned by the vector ki and by the normal (x1 axis) to the discontinuity between the two media, and called the incidence plane. Moreover, as |kr | = |ki |, we must have: ϑr = ϑ i

,

(4.81)

where ϑr (ϑi ) is the angle between kr (ki ) and the x1 axis. They are called the reﬂection angle and the incidence angle, respectively. As for the transmitted wave, Eq. (4.79) reads: at · r = 0 ,

k t · r = ki · r .

(4.82)

A priori, there is no reason why this wave cannot be a dissociated one, or an evanescent one if medium 2 is lossless. We will see soon that, in some circumstances, this is indeed the case. In any case, the second of Eqs. (4.82) shows that also kt lies always in the incidence plane. Deﬁning then the transmission angle (or refraction angle) ϑt as the angle between kt and the x1 axis, the second of Eqs. (4.82) can be rewritten as: sin ϑt =

β1 sin ϑi |kt |

.

(4.83)

In order to evaluate |kt |, which is still unknown, it is necessary to distinguish whether medium 2 is lossless or lossy. A) Lossless medium. In Eqs. (4.82), we can have either at = 0, or at perpendicular to the x1 = 0 plane. For at = 0, the transmitted ﬁeld is a uniform plane √ wave, and is called the refracted wave. It has |kt | = β2 = ω μ2 2 . Eq. (4.83) can then be rewritten in the following form, which is known as Snell’s law, or

© 2006 by Taylor & Francis Group, LLC

REFLECTION AND REFRACTION OF PLANE WAVES sometimes as Descartes’ law : μ1 1 β1 sin ϑt = sin ϑi = sin ϑi β2 μ2 2

,

111

(4.84)

where:

μm m (m = 1, 2) , μ0 0 is the refractive index in the m-th medium. Eq. (4.84) then reads as: nm =

n2 sin ϑt = n1 sin ϑi

.

(4.85)

(4.86)

B) Again in a lossless medium, but considering now the possibility that: at · r = 0

(at = 0)

,

(4.87)

we see that this is an acceptable solution only if kt (which, because of what ˆ2 (i.e., was shown in Section 4.3, must be orthogonal to at ) is parallel to x if sin ϑt = 1). Moreover, we saw in Section 4.3 that |kt | > β2 , and therefore Eq. (4.83) tells us that an evanescent wave can exist in medium 2 only if the inequality: β2 n2 sin ϑi > = , (4.88) β1 n1 can be satisﬁed by a real value of the angle ϑi , as required by the initial assumption that the incident ﬁeld is a uniform plane wave. It is obvious that this condition can never be satisﬁed if n2 > n1 . What happens for n2 < n1 is extremely important, both conceptually and practically. It will be the subject of Section 4.11. C) Lossy medium. The transmitted wave has still to satisfy Eqs. (4.82). Hence, it is a dissociated wave, with constant-amplitude planes parallel to x 1 = 0 and constant-phase planes orthogonal to the direction deﬁned by Eq. (4.83). Note that in this case |kt | becomes a function of the angle ϑi . It can be proved (see Problems) that, instead of Eq. (4.83), one gets: 1/2 −1/2 σ22 n1 1 1+ 1+ 2 2 sin ϑt = sin ϑi . (4.89) n2 2 ω 2 cos2 ϑt This formula says that, for σ2 = 0, the ratio sin ϑt / sin ϑi does not depend only on the ratio between the refractive indices, but also on σ2 . Snell’s law has then to be looked at as a limit, which is approached when σ2 → 0. We will come back to this point in Section 4.12, when dealing with the case σ ω 2 . Before we close this section, let us brieﬂy mention what happens in the case (a very frequent one in practice) where there are more than two media, separated by parallel planes (see Figure 4.8). In all media but the last one, one wave is no longer suﬃcient: the number of continuity conditions to be satisﬁed by the ﬁeld in that medium doubled: in the previous case we had one plane of discontinuity in the parameters, in the

© 2006 by Taylor & Francis Group, LLC

112

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.8 A typical “multilayer” structure.

present case we have two such planes. Therefore, two waves are necessary in each medium, except the last one; they must diﬀer in the sign of the component along +x1 of their propagation vectors. Obviously, in general they will diﬀer in amplitude and phase too. When all continuity conditions are written explicitly, then one sees that the only way for this system to have nontrivial solutions is that conditions like Eqs. (4.81) and (4.83) be satisﬁed at each separation plane (the parameters in there must be those of the two media before and after that plane). However, at least as long as all the waves are uniform, to ﬁnd the transmission angle in the n-th medium it is not at all necessary to pass through explicit calculations which involve the other media. Indeed, exploiting the transitive law of equality, one may apply directly Snell’s law, Eq. (4.84), to the ﬁrst and to the n-th medium:

sin ϑn =

β1 sin ϑi = βn

μ1 1 sin ϑi μn n

.

(4.90)

The transitive law of equality makes the case where evanescent waves show up in one or more layers even more interesting than the present one. We will brieﬂy deal with it in Section 4.11.

© 2006 by Taylor & Francis Group, LLC

FRESNEL FORMULAS 4.9

113

Fresnel formulas

The previous section, devoted to the geometry of the reﬂection problem, has led us to conclude that nontrivial solutions of the system in Eq. (4.78) exist. We may then proceed to express the reﬂected and transmitted ﬁelds as function of the incident ﬁeld. These relationships depend on polarization of the incident ﬁeld. To determine them, the simplest way is to exploit some of the results of Chapter 2: we expand the incident wave, whose polarization is completely arbitrary, as the sum of two waves which have two preselected, linearly independent polarizations. Reﬂection and transmission of these two components can be studied separately, and the results can then be superimposed, to reconstruct what happens to the generic incident wave. Since Ei and Hi are, in any case, both transverse to ki , a convenient basis for the decomposition consists of two orthogonal linear polarizations, lying in the plane orthogonal to ki . The simplest results are obtained if we take the following pair: Ei perpendicular to the incidence plane, and Hi perpendicular to the incidence plane. They are referred to, respectively, as TE and TM polarizations.† We will discuss in full detail the case of lossless medium 2, and uniform plane transmitted wave. Suggestions will be given later on to the reader in order to extend the results to the other cases. For TM polarization (see Figure 4.9), Eqs. (4.78) read: E0i cos ϑi + E0r cos ϑr E0i E0r − η1 η1

= E0t cos ϑt E0t η2

=

,

.

(4.91)

Eliminating once E0t , and once E0r , we can ﬁnd the following expressions for the ratios E0r /E0i = ρ and E0t /E0i = τ , which are called reﬂection and transmission coeﬃcients, respectively: η2 cos ϑt − η1 cos ϑi η2 cos ϑt + η1 cos ϑi

ρTM

=

τTM

= (1 − ρTM ) η2 /η1 =

,

2η2 cos ϑt η2 cos ϑt + η1 cos ϑi

.

(4.92)

Similarly, for TE polarization, replacing the electric ﬁeld vectors in Figure 4.9 with the magnetic ones (and keeping track correctly of all the direction † There are other terminologies which are widely used for these two polarizations, like “parallel” and “perpendicular” polarizations, or “horizontal” and “vertical” polarizations. We prefer to avoid these terminologies because there is not a general consensus on them. Radio engineers tend to refer, when using these names, to the direction of the electric ﬁeld, while physicists, in particular in optics, tended, in the past, to refer to that of the magnetic ﬁeld. Recently (see, e.g., Born and Wolf, 1980), some very authoritative textbooks in optics adopted the convention related to the electric ﬁeld.

© 2006 by Taylor & Francis Group, LLC

114

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.9 The ﬁeld vectors of the incident, reﬂected and transmitted waves, in the TM case.

of the vectors that are orthogonal to the plane of the ﬁgure), we ﬁnd: η2 cos ϑi − η1 cos ϑt η2 sec ϑt − η1 sec ϑi = η2 cos ϑi + η1 cos ϑt η2 sec ϑt + η1 sec ϑi

ρTE

=

τTE

= 1 + ρTE =

2η2 sec ϑt η2 sec ϑt + η1 sec ϑi

.

, (4.93)

Eqs. (4.92) and (4.93) are known as Fresnel formulas. If we recall the TM and TE wave impedance formulas, Eqs. (4.55) and (4.56), the reﬂection coeﬃcients can be given a uniﬁed expression: ρ=

Z2 (ˆ x1 ) − Z1 (ˆ x1 ) Z2 (ˆ x1 ) + Z1 (ˆ x1 )

,

(4.94)

where Zm (ˆ x1 ) (m = 1, 2) is the TM or TE wave impedance in the m-th medium, in the direction normal to the plane that separates the two media. It can be proved that Eq. (4.94) holds also in the case of lossy medium 2. The reader who is familiar either with common transmission lines, or with the scattering matrix formalism in circuit theory, will recognize Eq. (4.94) as a very familiar formula. Those who see it for the ﬁrst time will understand the meaning of this comment as soon as they get to Chapter 8. Finally, in next section we will show that, when the transmitted wave in ˆ1 ) with medium 2 is evanescent, it is possible to replace the angle ϑt = (kˆt , x a generalized transmission angle, in general complex, such that Eqs. (4.92) and (4.93) remain formally valid even in this case, although the physics of the transmitted wave changes completely. An interesting question is whether there is any value for the angle of incidence, such that the reﬂection coeﬃcients vanish. To answer it, one has to remember that, under the assumptions which hold in this section, the transmission angle is related to the angle of incidence by Snell’s law, Eq. (4.84). Therefore, the larger angle is in the medium where the index is smaller. Before we proceed further, we must specify whether we deal with dielectric or with

© 2006 by Taylor & Francis Group, LLC

REFLECTION IN MULTILAYER STRUCTURES

115

magnetic media, because permittivity and permeability play equal roles in the refractive index but diﬀerent roles in the wave impedance. Let us deal with the ﬁrst case (i.e., μ1 = μ2 = μ0 ). Then, we see that the equation ρTE = 0 does not have any real solution, being impossible to make η2 cos ϑi = η1 cos ϑt . On the contrary, the equation ρTM = 0 has a solution, which, as the reader may verify by imposing this equation plus Snell’s law, Eq. (4.84), reads: sin ϑi = n2 /(n21 + n22 )1/2

.

(4.95)

This angle is referred to as the Brewster angle or the polarizing angle. The second name points out that when a plane wave in a completely arbitrary state of polarization impinges on a discontinuity at the Brewster angle, the reﬂected wave is linearly polarized in the TE state. This fact is exploited in practice, for example to build a laser which emits linearly polarized light (see Chapter 9). 4.10

Reﬂection in multilayer structures

At the end of Section 4.8, we stressed that if there are more than two media, separated by parallel planes (Figure 4.8), then two waves are necessary in each medium except the last one, to satisfy all the continuity conditions. We saw then that Eqs. (4.81) and (4.83) must be satisﬁed at each separation plane. This means that the geometrical part of the problem, i.e., the laws of reﬂection and refraction in a multilayer structure, are almost trivial extensions of the two semi-inﬁnite media. It is not so concerning the reﬂection and transmission coeﬃcients that we found in the previous section. In this section, we will just give a very schematic description of how one deals with this new problem, and then provide two simple examples. The reason why we postpone most of the applications, is because the same calculations, which would be very heavy now in terms of computational eﬀort, will be enormously simpliﬁed by the analogies that one can establish after reading Chapter 8. The procedure goes as follows. The Fresnel reﬂection coeﬃcients, ρTM and ρTE given by Eqs. (4.92) and (4.93), respectively, can still be used at the interface between the (n − 1)-th and the n-th medium, located (Figure 4.8) at x1 = 0. Then, one tracks the two waves in the (n − 1)-th medium, back to the plane x1 = −n−1 . Note that the up-going and the down-going waves undergo opposite phase shifts, in this operation. Therefore, the total ﬁeld in the (n − 1)-th medium at x1 = −n−1 will not be the same as that at x1 = 0. Still, at this stage we may think of it as a quantity which contains one undetermined complex multiplicative constant, because the relationship between the two waves is known. This means that we are facing again the same situation as at the beginning, namely, a plane of discontinuity where the ﬁeld has two degrees of freedom on the upstream side, and one degree of freedom on the downstream side. If we apply again the continuity conditions, we may now calculate the reﬂection coeﬃcient at the interface between the (n−2)-th and the (n−1)-th medium. We iterate this procedure as many times

© 2006 by Taylor & Francis Group, LLC

116

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.10 Schematic illustration of a Fabry-P´ erot interferometer.

as there are layers, and ﬁnally get the reﬂection coeﬃcients at the input of the structure. As we said before, the procedure that we outlined does often require very cumbersome calculations, especially at oblique incidence, i.e., for an incidence angle ϑi = 0, because Snell’s law has to used at the same time to ﬁnd the various ϑj ’s, j = 2, . . . , n, which in turn aﬀect the reﬂection and transmission coeﬃcients. The following examples are quite simple, as they refer to normal incidence, ϑi = 0, and to three media, i.e., two half-spaces and just one layer in between. Let us study ﬁrst a layer between two semi-inﬁnite media having equal refractive indices, n1 = n3 = n2 . The media are supposed to be non-magnetic dielectrics. The reﬂection coeﬃcient at x1 = 0, obviously independent of polarization, is: n2 − n1 η1 − η2 = . (4.96) ρ= η1 + η2 n2 + n1 We now track back the two waves in medium 2, from x1 = 0 to x1 = −T (Figure 4.10). At those frequencies at which the thickness T of the layer is an integer multiple of the half-wavelength in medium 2, i.e., T = mπ/β 2 , m = 1, 2, . . ., the relative phase of the two waves in medium 2 at x1 = −T is the same as at x1 = 0, exactly as if there were no gap between medium 1 and medium 3. As these are identical, the reﬂection coeﬃcient at x1 = −T , as seen from medium 1, is equal to zero. At these frequencies, all the power

© 2006 by Taylor & Francis Group, LLC

REFLECTION IN MULTILAYER STRUCTURES

117

Figure 4.11 Design of an anti-reﬂection layer.

incident on the ﬁrst discontinuity is transmitted through layer 2, and emerges in medium 3. At all other frequencies, the reﬂection coeﬃcient at x1 = −T is not zero. Its modulus reaches its maximum value at those frequencies at which T equals an odd number of quarters of the wavelength in medium 2, a result that the reader may interpret in rather intuitive terms. The details of how the reﬂection coeﬃcient varies as a function of frequency depend on other details that will be easy to understand after reading Chapters 8 and 9. In any case, however, it is a periodic function of frequency, and its minima are equal to zero. The device schematically shown in Figure 4.10 is known as a Fabry-P´erot interferometer or as an ´etalon. In our second example, Figure 4.11, the two semi-inﬁnite media have different refractive indices, n1 = n3 , and the purpose of the layer between them is to cancel reﬂection at the front of this layer, i.e., at x1 = −T . The media are again supposed to be non-magnetic dielectrics. Physical intuition tells us that one way to reach our goal is to design the thickness and the indices in such a way, that two backward waves, due to reﬂections at x1 = 0 and at x1 = −T , cancel each other. Aiming at this, we assume that the thickness of layer 2 is an odd integer multiple of a quarter of the wavelength in medium 2, i.e., T = (2m + 1)π/(2β2 ), m = 0, 1, . . .. This entails that the phase diﬀerence between the forward and backward traveling waves shifts by an odd integer multiple of π radians, as we move from the upper side of layer 2 to the lower side. The reﬂection coeﬃcient at x1 = 0, again independent of polarization, is: ρ0 =

η3 − η2 n2 − n3 = η3 + η2 n2 + n3

,

(4.97)

and then, from what we just said, it follows that the ratio between the backward and the forward traveling waves in medium 2 at x1 = −T is ρ = −ρ0 = (n3 − n2 )/(n3 + n2 ). What this implies is that, in the ﬁeld continuity conditions, Eqs. (4.78), if written at x1 = −T , the transmitted

© 2006 by Taylor & Francis Group, LLC

118

PLANE WAVES IN ISOTROPIC MEDIA

electric ﬁeld E0t must be replaced by E0t (1 + ρ ), and the transmitted magnetic ﬁeld H0t must be replaced by H0t (1 − ρ ). The ratio between these two ﬁelds is then not equal any more to the intrinsic impedance of medium 2, η 2 , but to: 1 + ρ n3 = η2 . (4.98) ηequiv = η2 1 − ρ n2 If we insert Eq. (4.98) into Eq. (4.94) as the impedance after the discontinuity plane, we ﬁnd a reﬂection coeﬃcient: ρ1 =

ηequiv − η1 n1 − n22 /n3 = ηequiv + η1 n1 + n22 /n3

,

(4.99)

which clearly indicates that there is no reﬂection there (ρ1 = 0) for: n2 = (n1 n3 )1/2

.

(4.100)

For reasons that will become clearer after reading Chapter 8, the anti-reﬂection device which has been illustrated in this example is referred to as a quarterwave impedance-matching transformer. 4.11

Total reﬂection

Let us now study what happens if, for medium 2 being lossless, we have, n2 n2 < n1 , ϑi > ϑc = arcsin , (4.101) n1 where the symbols are the same as in the previous section. The angle ϑ c deﬁned by Eq. (4.101) is referred to as the critical (or limiting) angle. In this case, Snell’s law Eq. (4.83) has no real solutions for ϑt , since it should require sin ϑt > 1. We may conclude that the transmitted wave cannot be a uniform plane wave. On the other hand, there is a solution which, as we mentioned in Section 4.8, consists of an evanescent wave. In this entire chapter, we are working in such conditions that the uniqueness theorem can never be invoked. Yet, the evanescent wave is the solution, as demonstrated experimentally. Its study provides the best possible example of why evanescent waves are important in practice. In the reference frame of Figure 4.7, as we saw in Section 4.8, the evanescent wave must have its attenuation vector at parallel to the x1 axis, and its phase vector kt parallel to the x2 axis. The modulus of kt is then given by Eq. (4.83) with ϑt = π/2, i.e.: |kt | = β1 sin ϑi (> β2 ) , (4.102) while from Eq. (4.24) we get: 1/2 2 1/2 |at | = |kt |2 − β22 = β1 sin2 ϑi − β22

.

(4.103)

Physically, it is evident that at must be oriented in the positive x1 sense; otherwise, the amplitude would grow exponentially as the ﬁeld penetrates into medium 2.

© 2006 by Taylor & Francis Group, LLC

TOTAL REFLECTION

119

Now let us show that, under these circumstances, the wave impedances, both TE and TM, in medium 2, are purely imaginary, i.e., pure reactances. Let us deal ﬁrst with the TE case. Because of the continuity condition in Eq. (1.46) on the normal component of the vector D across the plane x1 = 0, we see that Et has to be parallel to x3 , like Ei and Er . Moreover, since the propagation vector St = at + jkt has a purely real component along x1 , from the second of Eqs. (4.18) it follows that Htan = (at × Et )/jωμ is in quadrature with Et , so x1 ) is imaginary, q.e.d.. The TM case can be analyzed in the same that Z2TE (ˆ way, starting from the continuity condition for the normal component of B, and exploiting now Eq. (4.19). The two procedures yield: Z2TE (ˆ x1 ) =

jωμ |at |

,

Z2TM (ˆ x1 ) =

|at | jω

,

(4.104)

where |at | is given by Eq. (4.103). Given the physical meaning of the wave impedance, these results show that the Poynting vector component along x1 is imaginary, which is equivalent to saying that the evanescent wave does not carry any active power in the x1 direction. This result should be linked with what was said in Section 4.6. We will show now that, in these cases, both ρTE and ρTM , on the plane x1 = 0, are of unit modulus. Indeed, from Eq. (4.78), recalling that E0t and H0t are related by Eq. (4.104), we ﬁnd that, ρTM

=

(|at |/jω2 ) − η1 cos ϑi (|at |/jω2 ) + η1 cos ϑi

,

ρTE

=

(jωμ2 /|at |) − η1 sec ϑi (jωμ2 /|at |) + η1 sec ϑi

,

(4.105)

and hence: |ρTM | ≡ 1

,

|ρTE | ≡ 1

.

(4.106)

This implies that we have, for both the TE and the TM case (we shall therefore omit the subscript): Ei × H∗i Er × H∗r Ei × H∗i ·x ˆ1 ≡ −ρρ∗ ·x ˆ1 = − ·x ˆ1 2 2 2

.

(4.107)

This relation shows that the reﬂected wave carries, in the direction orthogonal to the separation plane, an active power density equal to that of the incoming wave. This yields the name total reﬂection for this phenomenon, which — let us emphasize it again — can occur only for n2 < n1 , i.e., when passing from a medium of “higher optical density” to a medium of “lower optical density” (such as, for example, from glass to air, but not vice-versa). The reader can prove that, when total reﬂection occurs, superimposing the incoming and the reﬂected wave one gets a ﬁeld like that which was described at the end of Section 4.5 and in Problem 3.7. This ﬁeld behaves as a purely standing wave along x1 (its nodes and crests are x1 = constant planes, spaced by λ/(4 cos ϑi )), and as a traveling wave along x2 , with phase velocity

© 2006 by Taylor & Francis Group, LLC

120

PLANE WAVES IN ISOTROPIC MEDIA

vf (ˆ x2 ) = c1 / sin ϑi = ω/(βi sin ϑi ), i.e., synchronous with the phase velocity in medium 2 where the wave normal is parallel to the x2 axis. In the rest of this section, we will show how total reﬂection can be described with a formalism which is very similar to that of Section 4.8, provided one deﬁnes a suitable complex transmission angle, ϑ¯t . Let us start from the postulate that Snell’s law holds also beyond the critical angle, i.e., for ϑi > ϑc . This implies: n1 sin ϑi > 1 . (4.108) sin ϑ¯t = n2 Then, let ϑ¯t = ϕ + jψ, with real ϕ and ψ. Exploiting the identity sin ϑ¯t = −j sinh(j ϑ¯t ), and expanding the sinh function into its real and imaginary parts, as: sin ϑ¯t = cosh ψ sin ϕ − j sinh ψ cos ϕ , (4.109) we see that Eq. (4.108) is satisﬁed when: π sin ϑ¯t = cosh ψ . (4.110) ϑ¯t = + jψ , 2 Moreover, also for ϑ¯t complex, we have cos2 ϑ¯t + sin2 ϑ¯t = 1; thus, sin ϑ¯t larger than 1 implies that cos ϑ¯t is imaginary if, as reconﬁrmed by an expansion of cos ϑ¯t analogous to Eq. (4.109), which, using Eq. (4.110), is: cos ϑ¯t = ±j sinh ψ

.

(4.111)

Now comparing Eq. (4.102) with Eq. (4.107), and using Eq. (4.110), it follows that: |kt | = β2 sin ϑ¯t = β2 cosh ψ . (4.112) From Eq. (4.103), we get: |at | = jβ2 cos ϑ¯t = β2 sinh ψ

.

(4.113)

Eqs. (4.112) and (4.113) allow us to say that ϑ¯t is the angle that a complex vector, of modulus β2 , must form with the x1 axis, in order that its components along x2 and x1 equal kt and jat , respectively. Note that ϑ¯t is not the angle between x1 and the propagation vector St = at + jkt , as, for an evanescent wave, |St | = β2 . Finally, thanks to Eq. (4.113) we see that there is a perfect formal matching between Eq. (4.105) on one side, and Eqs. (4.92), (4.93) on the other side. This means that the complex angle in Eq. (4.110) also provides correct results for the reﬂection coeﬃcients. All the contents of this section, up to now, hold provided that medium 2 extends to inﬁnity along the positive x1 half-axis. Let us now consider a case where the medium with refractive index n2 , smaller than n1 , has a ﬁnite thickness. For the sake of simplicity, let us conﬁne ourselves to the case, shown in Figure 4.12, where n3 = n1 ; more complicated examples can be found among the problems. Continuity conditions like in Eq. (4.77) must be satisﬁed, now, on two parallel planes, x1 = 0 and x1 = d. It is straightforward to check that the minimum

© 2006 by Taylor & Francis Group, LLC

TOTAL REFLECTION

121

Figure 4.12 Illustration of partially frustrated total reﬂection.

number of waves which are needed to solve this problem for an arbitrary value of the incidence angle is ﬁve: an incoming and a reﬂected wave in medium 1, two evanescent waves in medium 2 (with opposite directions, along +x 1 and −x1 , for their attenuation vectors), and a transmitted wave in medium 3. For n3 = n1 , it is easy to show, using Snell’s law and the transitive law of equality, that: (4.114) ϑ3 = ϑ i , so the transmitted wave goes in the same direction as the incident wave in medium 1. In medium 2, the ﬁeld is the sum of the two evanescent waves: ∗

E = E2+ e−S2 ·r + E2− eS2 ·r H =

,

S2 × E2+ −S2 ·r S∗2 × E2− S∗2 ·r e − e jωμ2 jωμ2

.

(4.115)

The reader can prove that, for this ﬁeld, the wave impedance in the x1 direction has a nonzero real part and depends on x1 . This is in deep contrast with a single evanescent waves, whereas it has analogies and diﬀerence with stationary waves arising as superposition of two uniform plane waves, discussed in Section 4.5. Indeed, one uniform traveling wave carries zero reactive power, while superposition of two counter-propagating ones results in a stationary wave with nonzero reactive power. For evanescent waves, each carries zero active power, while superposition of two counter-attenuating ones gives rise to a wave that carries an active power in the direction of the attenuation vector. This provides a physical explanation for the presence of a uniform plane wave, carrying active power, in medium 3, which otherwise would appear as an inconsistency under the viewpoint of conservation of energy. The physical phenomenon of Figure 4.12 is called partially frustrated total reﬂection or, by analogy with quantum mechanics, electromagnetic tunneling. The analogy with quantum mechanics is reinforced by the impossibility of explaining this phenomenon in terms of ray optics, which would always predict

© 2006 by Taylor & Francis Group, LLC

122

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.13 Plane-wave reﬂection at the surface of a good conductor.

zero power in medium 2 as soon as the critical angle is exceeded, independently of its thickness. In Chapter 10, it will be shown that electromagnetic tunneling plays a fundamental role in guided-wave propagation along a dielectric rod, such as an optical ﬁber. When the geometry is not planar (e.g., when it is cylindrical), there are interesting variations to be added on the same theme, as we will see later. 4.12

Reﬂection on the surface of a good conductor

In technical application of electromagnetic waves, very commonly one is required to ﬁnd a ﬁeld {E, H} deﬁned in a region ﬁlled by a low-loss (or lossless) dielectric medium, bordered by a high-conductivity medium. Usually, knowing the ﬁeld inside the conducting medium is not of primary interest. Thus, it is unnecessary to solve Maxwell’s equations in the conductor, and impose continuity conditions at its borders. Rather, one looks for suitable boundary conditions, to be imposed on the ﬁeld deﬁned in the dielectric medium, at the dielectric-conductor interface. It is our aim, in this section, to ﬁnd such conditions starting from Maxwell’s equations inside the conductor. Let us consider a plane surface between two semi-inﬁnite media, as in Figure 4.13. As we saw in Section 4.8, in medium 1 a uniform plane wave is reﬂected with ϑr = ϑi , independently of the nature of medium 2. What happens in medium 2 will determine the value of the reﬂection coeﬃcient. Suppose that medium 2 is a good conductor, i.e.: σ2 ω2

.

(4.116)

Before we proceed, let us point out that, in some cases, approximations following from Eq. (4.116) may cause errors: we are comparing real and imaginary quantities, and this can result in unacceptable consequences for the phases of complex numbers, although the results are correct for their moduli. As we saw in Section 4.8, the wave in medium 2 is in general dissociated.

© 2006 by Taylor & Francis Group, LLC

REFLECTION ON THE SURFACE OF A GOOD CONDUCTOR

123

Its attenuation vector at is parallel to x1 . Moreover, from Eq. (4.89) we see that: lim ϑt = 0 . (4.117) (σ2 /ω2 )→∞

Thus, when Eq. (4.116) holds, we can take ϑt 0, i.e., kt is almost parallel to at . Moreover, from Eqs. (4.116), (4.24) and (4.25), we get: |kt |2 − |at |2 2 at · kt

.

(4.118)

Eq. (4.118) means that |at | |kt |, so that Eq. (4.25) becomes: |at | |kt | (ωμ2 σ2 /2)1/2 = (πf μ2 σ2 )1/2

,

(4.119)

where f = ω/2π is the frequency of the incident wave. Therefore, the transmitted wave is attenuated along x1 as exp[−(πf μ2 σ2 )1/2 x1 ] as it penetrates into the conductor. The quantity: δ = 1/(πf μ2 σ2 )1/2

,

(4.120)

whose dimensions are those of a length, gives an estimate of the depth up to which the transmitted wave has a nonnegligible amplitude. It is referred to as the penetration depth (or, especially in the past, as the skin depth). In practice, the results that we found for the indeﬁnitely extended medium 2 are reliable if the conductor thickness is equal to a few times the penetration depth. The wave impedance along x1 in the conducting medium can be obtained from Eq. (4.18), taking into account that, because of what we just saw: ˆ1 St = at + jkt (1 + j) (πf μ2 σ2 )1/2 x

,

(4.121)

and noting that, since S is parallel to x ˆ1 , both Et and Ht are perpendicular to x ˆ1 , irrespective of the incidence angle ϑi . In conclusion, it reads: 1/2 πf μ2 jωμ2 Zxˆ1 = = (1 + j) . (4.122) St · x ˆ1 σ2 This impedance depends only on the frequency and on the medium parameters. In contrast to previous results, it does not depend on the direction of incidence of the incoming wave. Therefore, it is a characteristic of the conducting medium, not of the wave propagating in it. This justiﬁes its name, wall impedance, and entitles us to use Eq. (4.122) as a boundary condition on the plane which limits the dielectric medium. The scope of validity to this approach may be limited by curvature of the separation surface; this point will be clariﬁed in Chapter 7. From Eq. (4.122), it follows that: lim Zxˆ1 = 0 ,

σ2 →∞

(4.123)

where no quantity characterizing the incident wave appears. This reminds us of the concept of an ideal electric conductor, found in Chapter 3, which is a useful mathematical tool in many problems, and tells

© 2006 by Taylor & Francis Group, LLC

124

PLANE WAVES IN ISOTROPIC MEDIA

us that it can be regarded as a medium with inﬁnite conductivity, imposing the boundary condition: Etan = 0 (4.124) on its surface. Note that Eq. (4.120) gives δ → 0 as σ2 → ∞, so that the thickness of an ideal electric conductor can be arbitrarily small. 4.13

Further applications and suggested reading

In Chapter 1, we presented Maxwell’s equations, and other equations which follow from them. In Chapter 3, we derived theorems which are based on Maxwell’s equations. In Chapter 4, we began to solve Maxwell’s equations, and this process will now occupy the remainder of the book. Solutions, subject to boundary conditions and often to other limiting assumptions, never have a scope of validity as general as the fundamental equations and theorems of the ﬁrst two chapters. Nevertheless, we may say that, in this chapter, we started oﬀ with solutions whose validity and whose usefulness are very broad. Plane waves appear to be much more ubiquitous than any other time-varying electromagnetic ﬁeld. Rather than writing a list of applications, which, irrespective of its length, could never aim at being complete, this section will limit itself to a few basic recommendations to the reader, and to the explanation of why some subjects have been omitted in this chapter, although they may be very important. First, let us remind the reader that electromagnetic plane waves in a homogeneous medium, being solutions of one of the most widely used equations in mathematical physics — the Helmholtz scalar equation — share an extremely large number of properties with waves of completely diﬀerent physical natures (e.g., Elmore and Heald, 1969; Pain, 1993; Young, 1976), such as acoustic waves in ﬂuids (e.g., Ford, 1970; Frisk, 1994; Brekhovskikh, 1960; Brekhovskikh and Godin, 1990/1991; Meyer and Neumann, 1972; Pierce, 1981), elastic waves in solids (e.g., Ashcroft and Mermin, 1976; Kittel, 1986), wave functions associated with probability distributions of particles in Quantum Mechanics. An impressive example in this sense is provided by the Fresnel reﬂection and transmission formulas (Section 4.9). They were regarded, historically, as the ﬁrst irrefutable proof of the wave nature of light, and were published in 1816, i.e., more than 50 years before Maxwell’s equations. The wave nature of light is by no means a synonym for the electromagnetic nature, a concept which was ﬁrst proposed by Maxwell many decades later. Incidentally, the history of undulatorial theories in physics is dealt with in an ad hoc introductory chapter of Born and Wolf (1980). It may be very useful for the reader to compare the content of this chapter with that of suitable textbooks on acoustics (e.g., Ford, 1970), classical analytical mechanics (e.g., Goldstein, 1980), and quantum mechanics (e.g., Cohen-Tannoudji et al., 1977; Dicke and Wittke, 1960; Merzbacher, 1970;

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

125

Schiﬀ, 1968). As we will see later, very strong similarity between electromagnetic waves and other waves exists also when dealing with guided propagation phenomena. We recommend that the reader return to these references, after reading Chapters 7, 9 and 10. In particular, consulting the more-than-onecentury old Theory of Sound (1877) by Lord Rayleigh would, at the same time, satisfy cultural appetites from the viewpoint of physics, and make the reader more curious about history of science. Plane waves in unbounded isotropic media suﬀer from a number of very severe limitations, which make them inadequate to model some basic phenomena. Just to quote one, at the end of this chapter, the reader does not yet know at which velocity the energy associated with an electromagnetic wave travels. The weak points of plane waves have not been emphasized in this chapter just because it would have been pedagogically wrong and pointless not to insist ﬁrst on their strong points. Their disadvantages, and how to correct them, e.g., using “local plane waves” (Snyder and Love, 1983, Chapter 35), will be contents of the next chapter. We postpone more suggestions for further readings until the end of Chapter 5. Some experienced readers will probably be surprised not to ﬁnd in this chapter an analysis (or design) of multilayer dielectrics. Indeed, this is one of the most important problems that can be tackled using plane waves. The reason why it was not dealt with in detail now is, again, a pedagogical choice. We believe that problems of this kind are dealt with in a much simpler fashion, especially for what concerns calculations, by analogy with electrical transmission lines. These will be covered in depth in Chapter 8. So, it will be advisable to come back to multilayers, and solve again some of the problems presented at the end of this chapter, after studying Chapter 8. Of course, talking about multilayers, under the viewpoint of theory and under that of a broad choice of examples of application, a classical reference that it is highly recommended for consultation is Born and Wolf (1980). We also refer to the work of Brekhovskikh (1960) for waves in multilayer media. Finally, with regard to plane waves in good conductors, in telecommunications they are of primary importance for microwave transmission, to evaluate waveguide loss (see Chapter 5), and in radio propagation, to study penetration into natural conductors, such as salt water at low frequencies, the ionosphere in the MHz range (Budden, 1961), etc. Outside the telecom world, keeping in mind what was said in the ﬁnal section of Chapter 1, interaction of electromagnetic ﬁelds with living tissues, and applications to material diagnostics, are just two among the most eminent examples. The relevant bibliographical suggestions are those that were already quoted in Chapter 1.

Problems 4-1 Show that the Helmholtz equation for the ﬁeld vectors E and H can be solved directly by separation of variables. Show that a generic solution of

© 2006 by Taylor & Francis Group, LLC

126

PLANE WAVES IN ISOTROPIC MEDIA

the equation for E and one of the equation for H do not form a solution of Maxwell’s equations. 4-2 A plane wave traveling in air has an electric ﬁeld amplitude of 10 mV/m. Derive its magnetic ﬁeld in the following cases. a) The wave is uniform, travels in the x3 direction, and the electric ﬁeld is linearly polarized along the x1 axis. b) The wave is uniform, travels in the x3 direction, and the electric ﬁeld is circularly polarized in the (x1 , x2 ) plane. c) The wave is evanescent, with its phase vector in the x3 direction, its attenuation vector in the x2 direction, and the electric ﬁeld is linearly polarized along the x1 axis. Show that in the ﬁrst case only the electric ﬁeld and the magnetic ﬁeld are orthogonal complex vectors. Show that also in the second case the corresponding vectors in the time domain are orthogonal at any instant. 4-3 Find a better approximation than vf c for a uniform plane wave in a low-loss medium where σ ω. Show that the ﬁrst correction is proportional to the square of the loss tangent. 4-4 Derive explicitly all the features of the ﬁeld resulting from two waves of equal amplitude interfering at an angle, as described in Section 4.6. Repeat the same exercise for two waves of diﬀerent amplitudes. 4-5 Show that the following expressions for power reﬂectivity (i.e., ratio of reﬂected and incident power densities), for plane waves at the separation between two semi-inﬁnite media, are equivalent to those which can be derived from the Fresnel formulas: RTE =

sin2 (ϑi − ϑt ) sin2 (ϑi + ϑt )

RTM =

tan2 (ϑi − ϑt ) tan2 (ϑi + ϑt )

for TE waves, and:

for TM waves, respectively. Reconsider the Brewster angle problem in the framework of these new results.

4-6 A layer of a lossless medium, 1 m in thickness, with a refractive index n = 4, is placed between an air-ﬁlled half-space and an ideally conducting plane (located at x1 = 0). A uniform plane wave, at a frequency f = 6 GHz, impinges from air, travels through the layer, and is reﬂected back. Find the period of the standing wave in the dielectric layer, along the x1 direction, for the following cases:

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

127

a) incidence along the x1 axis; b) incidence along a direction such that the refracted wave in the layer travels at 45 degrees with respect to the x1 axis; c) incidence at an angle (in air) of 45 degrees with respect to the x1 axis. 4-7 A uniform plane wave, impinging at normal incidence on the plane (x1 = 0) of separation between two lossless dielectric media, conveys a power density of 1 mW/mm2 . The power density of the transmitted wave is 3 dB below that of the incident wave. Find the modulus of the electric ﬁeld of the reﬂected wave in the following two cases: a) the medium in which the incident and the reﬂected waves travel is air; b) the medium in which the transmitted wave travels is air. Repeat the problem assuming that the incidence angle is π/4 and that the power densities on the x1 = 0 plane remain the same as in the previous cases. 4-8 A TE-polarized uniform plane wave impinges with an incidence angle of π/4 radians on the plane x1 = 0, separating air from a lossless dielectric medium. The reﬂection coeﬃcient equals −0.5. Find the refractive index of the second medium, the reﬂection coeﬃcient for a TM-polarized uniform plane wave coming along the same direction, and the reﬂection coeﬃcients for a TE- and a TM-polarized plane wave coming from the second medium, and giving rise to refracted waves with a transmission angle (in air) equal to π/4 radians. 4-9 Two semi-inﬁnite lossless dielectric media have refractive indices equal to 3 and to 1, respectively. At a frequency of 3×1014 Hz, experimental observation of an evanescent wave in the second medium shows that it decays in the ratio 1/e over a distance of 1 μm, measured along the normal to the plane of separation. a) Find the corresponding angle of incidence, in medium 1. b) Evaluate the phases of the corresponding TE and TM reﬂection coeﬃcients in medium 1, on the separation plane. 4-10 Show that when a wave is incident at the Brewster angle, Eq. (4.95), then the transmitted wave travels at an angle ϑt = π/2 − ϑi . Hence, show that it is impossible to ﬁnd two media such that the Brewster angle at their interface equals π/4. √ 4-11 A lossless layer, with a refractive index equal to 3 and a thickness of 1/8 μm, has been sandwiched between the semi-inﬁnite media of Problem 4-9. a) At what frequencies does it operate as an impedance-matching layer (i.e., reﬂection is canceled at normal incidence) ?

© 2006 by Taylor & Francis Group, LLC

128

PLANE WAVES IN ISOTROPIC MEDIA

b) Which is, in this structure with an intermediate layer, the critical angle for total internal reﬂection of a plane wave coming from the medium whose index equals 3 ? 4-12 Calculate the Brewster angle for the two semi-inﬁnite media of Problem 4-9, and the reﬂection coeﬃcient for a TE polarized wave incident at that angle. Suppose now that a uniform plane wave of unspeciﬁed polarization is incident at the same angle as in the previous question of this problem, but subsequent to the introduction of the intermediate layer discussed in the previous problem. Is the reﬂected wave still TE-polarized ? 4-13 A lossless dielectric layer is located between two air-ﬁlled semi-inﬁnite half-spaces. A uniform plane wave impinges normally on it. The resulting partially-standing wave pattern has a period of 4 mm in air, of 1 mm in the dielectric layer. a) Find the frequency of the wave. b) Find the thicknesses of the layer for which the peak-to-valley ratio of the standing wave, in the half-space where the incident wave comes from, is maximum, and those for which it is minimum. c) Does the layer thickness aﬀect the peak-to-valley ratio of the partially standing wave in the layer ? 4-14 A uniform plane wave, coming from a lossless medium, impinges at oblique incidence onto the plane surface which delimits another semi-inﬁnite medium. Compare the following two cases: a) medium 2 is lossless, and the incident wave undergoes total reﬂection; b) medium 2 is a good conductor. 1. 2. 3. 4. 5. 6.

Are Are Are Are Are Are

the reﬂection coeﬃcients comparable in modulus ? they dependent on the polarization of the incident wave ? their phases identical, or in general not ? their phases polarization-dependent or not ? the attenuation vectors in medium 2 in the same direction ? the phase vectors in medium 2 in the same direction ?

4-15 A typical value for the conductivity of copper is 5.8 × 107 Ω−1 m−1 . Permittivity and permeability may be taken, as an approximation, equal to those of free space. a) Evaluate the order of magnitude of the maximum frequencies at which copper can still be considered to be a good conductor. Calculate the frequency at which the penetration depth in copper equals 1 mm. Find the value of the surface impedance of a ﬂat copper wall at that frequency.

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

129

b) Repeat these calculations for the case of graphite, considered as a good conductor at industrial frequencies. Its conductivity, strongly temperaturedependent, is of the order of 105 Ω−1 m−1 . Are the numbers obtained in this case consistent with each other ? If not, explain why. 4-16 To establish a radio communication link with an immersed submarine, one requires a wave penetration depth of at least 10 meters. Find the maximum usable frequency. For the conductivity of sea water, a typical value is 5 Ω−1 m−1 . Magnetic permeability is practically the same as that of free space. 4-17 Evaluate the modulus of the phase vector for a plane wave in sea water, at a frequency of 500 Hz. Compare it with the modulus that the wave vector has, at the same frequency, for a plane wave in a lossless medium whose dielectric permittivity is equal to that of sea water (whose relative permittivity is of the order of 80). Explain the origin of this very large diﬀerence. (Hint. Use a graphical representation of −ω 2 μc and of its roots on the complex plane.)

© 2006 by Taylor & Francis Group, LLC

References Ashcroft, N.W. and Mermin, N.D. (1976) Solid State Physics. Holt, Rinehart and Winston, New York. Born, M. and Wolf, E. (1980) Principles of Optics. 6th ed., Pergamon, Oxford. Brekhovskikh, L.M. (1960) Waves in Layered Media. Academic, New York. Brekhovskikh, L.M. and Godin, O.A. (1990) Acoustics of Layered Media I: Plane and Quasi-Plane Waves. Springer-Verlag, Berlin. Brekhovskikh, L.M. and Godin O.A. (1991) Acoustics of Layered Media II: Point Source and Bounded Beams. Springer-Verlag, Berlin. Budden, K.G. (1961) Radio Waves in the Ionosphere. Cambridge University Press, Cambridge, UK. Budden, K.G. (1988) The Propagation of Radio Waves: The Theory of Radio Waves of Low Power in the Ionosphere and Magnetosphere. 1st paperback ed. with corrections, Cambridge University Press, Cambridge, UK. Cohen-Tannoudji, C., Diu, B. and Lalo¨e, F. (1977) M´ecanique Quantique. Hermann, Paris. Dicke, R.H. and Wittke, J.P. (1960) Introduction to Quantum Mechanics. AddisonWesley, Reading, MA. Elmore, W.C. and Heald, M.A. (1969) Physics of Waves. McGraw-Hill, New York (Dover, New York, 1985). Ford, R.D. (1970) Introduction to Acoustics. Elsevier, Amsterdam. Frisk, G.V. (1994) Ocean and Seabed Acoustics: A Theory of Wave Propagation. Prentice-Hall, Englewood Cliﬀs, NJ. Goldstein, H. (1980) Classical Mechanics. 2nd ed., Addison-Wesley, Reading, MA. Kittel, C. (1986) Introduction to Solid State Physics. Wiley, New York. Merzbacher, E. (1970) Quantum Mechanics. Wiley, New York. Meyer, E. and Neumann, E.-G. (1972) Physical and Applied Acoustics: An Introduction. Academic, New York. Morse, P.M. and Feshbach, H. (1953) Methods of Theoretical Physics. McGraw-Hill, New York. Pain, H.J. (1993) The Physics of Vibrations and Waves. 4th ed., Wiley, New York. Pierce, A.D. (1981) Acoustics: An Introduction to its Physical Principles and Applications. McGraw-Hill, New York. Schiﬀ, L.I. (1968) Quantum Mechanics. McGraw-Hill, International. Snyder, A.W. and Love, J.D. (1983) Optical Waveguide Theory. Chapman & Hall, London. Young, H.D. (1976) Fundamentals of Waves Optics and Modern Physics. 2nd ed., McGraw-Hill, New York.

131 © 2006 by Taylor & Francis Group, LLC

CHAPTER 5

Plane wave packets and beams 5.1

Modulated waves. Group velocity

Throughout the previous chapter, the reader had to bear in mind two fundamental warnings. One was that a single plane wave of ﬁnite amplitude does not make physical sense, as it would ﬁll the whole space with an inﬁnite energy. The second was that, on the other hand, plane waves form a complete set. Consequently, combinations of plane waves may build up physically meaningful and, more signiﬁcantly, technically important results. Some of them are still mathematically simple, and fairly intuitive, and this chapter is devoted to them. At ﬁrst sight, the subjects which form the chapter might look rather heterogeneous, but this is more apparent than real, as we shall see while we proceed. As we just said, a single plane wave, with an angular frequency ω, invades the whole 3-D space. At any point, its ﬁeld varies sinusoidally vs. time t, for t going from −∞ to +∞; consequently, it does not convey information from one point to another. It is because of this that we accepted that the phase velocity vf , deﬁned in Section 4.4, could be arbitrarily large. This is not a violation of the fundamental postulate of the special theory of relativity, since vf can never be the speed of propagation for a signal that synchronizes two “clocks”. Those readers who have been exposed to the fundamentals of communication theory know that, in order to transmit information, one needs a signal with a ﬁnite spectral width, such as a modulated wave. In order to determine the speed at which the information content carried by an e.m. wave travels, we must then start from a non-monochromatic ﬁeld {E ∼, H ∼}. Actually, in Section 4.7, completeness was proved only for strictly monochromatic plane waves. However, when regularity of functions and derivatives assures that operators in the time domain and in the space domain commute, a ﬁeld having a ﬁnite spectral width in the frequency domain can also be expanded into plane waves, provided it satisﬁes, as a function of space coordinates, conditions similar to those given in Section 4.7. A very simple model, which enables us to determine the speed we are looking for, at least in a lossless medium, consists of two uniform plane waves, traveling in the same direction (i.e., having parallel phase vectors k), but diﬀerent in angular frequencies, ω0 ± Δω, respectively. This notation is reminiscent of an amplitude modulation, with a modulation angular frequency Δω. In the following Section 5.2, however, we will show that the results obtained in this 133 © 2006 by Taylor & Francis Group, LLC

134

PLANE WAVE PACKETS AND BEAMS

way are quite general. To simplify the calculations, without any loss in the scope of validity of the results, we will suppose that these waves have equal polarizations and equal amplitudes, represented by a vector E which, again for simplicity, we assume to be real. Let us introduce rectangular coordinates such that the waves propagate along the x3 axis. In the time domain, the total electric ﬁeld is then expressed by

−j(β+ x3 −ω0 t−Δωt) , (5.1) + E e−j(β− x3 −ω0 t+Δωt) E ∼ (r, t) = Re E e where β+ and β− are the phase constants β = ω μ(ω) (ω) at the angular frequencies ω0 + Δω and ω0 − Δω, respectively. One may always write β+ = βm + Δβ

,

β− = βm − Δβ

,

(5.2)

where βm = (β+ +β− )/2. Quite often, β(ω) is a smooth function of ω, between ω0 − Δω and ω0 + Δω. In that case, there is an angular frequency ωm , in that interval, such that βm = β(ωm ). In general, ωm diﬀers from ω0 , but if Δω is small enough, then we can let βm β(ω0 ). Then, inserting Eqs. (5.2) into Eq. (5.1), we ﬁnd (5.3) E ∼ = E cos(ω0 t − βm x3 ) 2 cos(Δβx3 − Δωt) . This expression looks like a traveling wave (with angular frequency ω0 and phase constant βm ), wrapped in an envelope cos(Δβx3 − Δωt). The loci of constant envelope amplitude are the solutions of d(Δβx3 − Δωt) = 0. They are planes perpendicular to the x3 axis, moving along that axis with a velocity v = Δω/Δβ

.

(5.4)

If the function β(ω) can be expanded as a Taylor series dβ 1 d2 β (ω − ω0 ) + (ω − ω0 )2 + . . . , (5.5) dω 2 dω 2 then, if we let the spectral width of the signal tend to zero, we see that the velocity (5.4) tends to 1 dω = . (5.6) vg (ω0 ) = dβ dβ ω=ω0 dω ω0 β(ω) = β(ω0 ) +

The quantity deﬁned by Eq. (5.6) is referred to as the group velocity of a uniform plane wave with angular frequency ω0 . The procedure that led us to this result indicates that, for a narrow-band signal (Δω ω0 ), traveling in a medium where β(ω) is a smooth function satisfying all the requirements that we introduced, the group velocity (5.6) equals indeed the velocity at which the signal travels. But let us stress that this is not always true. For example, in a case where dβ/dω = 0, Eq. (5.6) would yield an inﬁnite speed, whereas Eq. (5.4) shows that, for Δβ = 0, the envelope propagates at a

© 2006 by Taylor & Francis Group, LLC

MODULATED WAVES. GROUP VELOCITY

135

ﬁnite speed. Furthermore, this value must be less than the speed of light, c, to be compatible with the special theory of relativity. In all cases where these requirements are not satisﬁed by the quantity in Eq. (5.6), we will keep calling it “group velocity,” but we will not state that, in these cases, the signal envelope propagates at the group velocity. While we wait until the next section to show what happens to signals whose spectrum is more complicated than just two frequencies, let us show that Eq. (5.6) has also another physical meaning, namely that of the velocity of energy transport along the direction of propagation. For a uniform plane wave in a lossless medium, in Maxwell’s equations we may write ∇× = (−jk)× = ˆ (−jβ k)×. Then, diﬀerentiating all terms in Maxwell’s equations with respect to ω, we get the following relations: −j

∂(ωμ) ∂H ∂ (k × E) = −j H − jωμ ∂ω ∂ω ∂ω

−j

∂ ∂(ω) ∂E (k × H) = j E + jω ∂ω ∂ω ∂ω

,

.

(5.7)

We take then the scalar product of the ﬁrst one by H∗ , and that of the second one by E∗ . Then, we subtract the ﬁrst one from the second one, and we take into account also the following relationships, which stem directly from Maxwell’s equations and allow us to get rid of several terms: −j k ×

∂E · H∗ ∂ω

= −jω E∗ ·

−j k ×

∂H · E∗ ∂ω

= jωμ H∗ ·

∂E ∂ω

∂H ∂ω

, .

Finally, we obtain: ˆ ∂β = ∂(ωμ) H · H∗ + ∂(ω) E · E∗ (E × H∗ · kˆ + E∗ × H · k) ∂ω ∂ω ∂ω This equation can be rewritten as: 1 ∂(ωμ) ∂(ω) ∗ ∗ ˆ H·H + E·E Re{P · k} = vg 4 ∂ω ∂ω

.

.

(5.8)

(5.9)

It shows that the real part of the Poynting vector ﬂux through a unit-area surface, orthogonal to the direction of propagation, equals the group velocity times a term whose dimensions are those of an energy density. As in ﬂuid dynamics, this tells us that the group velocity is the speed at which energy is carried by this e.m. ﬁeld. We note, though, that the expression for the energy density in general is not the same as the one which appears in Poynting’s theorem, Section 3.2, which was: 1 (μ H · H∗ + E · E∗ ) . 4

© 2006 by Taylor & Francis Group, LLC

(5.10)

136

PLANE WAVE PACKETS AND BEAMS

The two expressions coincide if and only if: ∂ ∂μ =0 , =0 , (5.11) ∂ω ∂ω that is, in what is called a nondispersive medium. The apparent discrepancy between the results can be explained if we remember that in Section 3.2 we dealt with a purely sinusoidal time-harmonic ﬁeld, while now we are studying what happens when we superimpose two plane at diﬀerent frequencies. As powers and energies result from nonlinear operations, even a very simple superposition can deeply modify previous results. In Chapter 4, we saw that, for a monochromatic uniform plane wave in a lossless medium, energy is uniformly distributed throughout the whole space. We could not track an energy “carried” by the wave, and indeed, for such a wave, the divergence of Poynting vector, P = E × H∗ /2, is zero everywhere. On the contrary, the modulated wave that we saw in this section does carry an energy. Its Poynting vector has a nonvanishing divergence, as we can prove in the following way. Consider, as an extension of the previous argument, the following e.m. ﬁeld: 1 E = E(ω0 + Δω) ej(ω0 +Δω)t + c.c. ∼ 2 1 E(ω0 − Δω) ej(ω0 −Δω)t + c.c. + 2 1 ∂E j(ω0 +Δω)t Δω e + c.c. E(ω0 ) + 2 ∂ω ∂E 1 Δω ej(ω0 −Δω)t + c.c. (5.12) + E(ω0 ) − 2 ∂ω ∼ H ∼ = +

∂H Δω ej(ω0 +Δω)t + c.c. H(ω0 ) + ∂ω ∂H 1 Δω ej(ω0 −Δω)t + c.c. H(ω0 ) − 2 ∂ω 1 2

,

(5.13)

where c.c. means the complex conjugate of the preceding term. Suppose that {E(ω0 ), H(ω0 )} is a uniform plane wave, then, ∂E/∂ω and ∂H/∂ω are related to each other by two equations that can be obtained by diﬀerentiating Maxwell’s equations. Now, let us calculate the instantaneous Poynting vector which corresponds to the ﬁeld in Eqs. (5.12), (5.13), including also the products between ﬁelds at diﬀerent frequencies, because, as one lets Δω → 0, the time-averaged values of these products tend to become quantities that are not zero. Let us drop the terms of order (Δω)2 . The assumption that {E(ω0 ), H(ω0 )} is a uniform plane wave implies that the contribution E(ω0 ) × H∗ (ω0 ) and alike have zero divergence. Then, from Maxwell’s equations and from their derivatives with respect to ω, taking into account the relation (C.4) ∇ · (a × b) = (∇ × a) · b − a · (∇ × b), it follows

© 2006 by Taylor & Francis Group, LLC

DISPERSION that:

137

∂E∗ ∂H∗ + ×H ∇· E× ∂ω ∂ω

=j

∂(ωμ) ∂(ω) H · H∗ + E · E∗ ∂ω ∂ω

. (5.14)

The Poynting vector computed from Eqs. (5.12) and (5.13) contains the terms on the left-hand side of Eq. (5.14), minus their complex conjugates. If we subtract Eq. (5.14) from its complex conjugate, we ﬁnd twice the quantity on the right-hand side. So, we see that the energy density which appears on the right-hand side of Eq. (5.9) is indeed related, in the sense of Poynting’s theorem, to that part of the Poynting vector which originates from products of ﬁelds at two diﬀerent frequencies. As we said, only this part of the e.m. energy travels, so these ﬁndings reconﬁrm that the interpretation we gave, for the physical meaning of group velocity, is correct. 5.2

Dispersion

In the previous section, group velocity, on one hand, was calculated at a single frequency, according to Eq. (5.6). On the other hand, its physical meaning can be grasped only if we consider non-monochromatic waves. The whole picture is clear and unambiguous if Eq. (5.6) is independent of ω, over a ﬁnite frequency interval around ω0 . But if vg varies within the signal spectral range, then further developments become necessary. The phenomena which occur when dvg /dω = 0 go under the generic name of dispersion. The historical origin of this term comes from chromatic dispersion of prisms, which, as well known, separate spectral components of white light because glass has a wavelength dependent refractive index. Nowadays, the word dispersion, if used in a transmission context, is essentially a synonym to delay distortion, and means that diﬀerent spectral components travel at diﬀerent speed. Other phrases that are also of general use, regardless of their historical origins, are: • normal dispersion: dvg /dω < 0 (dvg /dλ > 0); • anomalous dispersion: dvg /dω > 0 (dvg /dλ < 0). In the following chapters we will see that these concepts and these names apply not only to plane waves, but to electromagnetic propagation in general. This section is devoted to an example which is based on plane waves, but the concepts we develop here have a much broader scope of validity. Let us consider a packet of uniform plane waves that travel along one direction, say z. Let their spectrum be centered on a “carrier” angular frequency ω0 , and their time-domain envelope, at an initial abscissa z = 0, be a Gaussian function, which simpliﬁes calculations and so allows us to stress physical results. We choose the time reference system such that at t = 0 the center of the Gaussian pulse is at z = 0. Any component of the e.m. ﬁeld can be written as: ψ(z = 0, t) = A exp(−t2 /τ 2 ) cos ω0 t

© 2006 by Taylor & Francis Group, LLC

.

(5.15)

138

PLANE WAVE PACKETS AND BEAMS

The symbol ψ is intended to underline that the following results are independent of the physical dimensions of the quantity one explicitly deals with. To follow the temporal and spatial evolution of the wave packet, it is convenient to use the (time-domain) Fourier transform of Eq. (5.15), Ψ(ω), which is easily calculated, as the Fourier transform of a Gaussian function is a Gaussian function also: √ +∞ 2 2 π −δ2 /(2α)2 e e−α x ejδx dx = . (5.16) α −∞ We get therefore: +∞ 1 Ψ(ω) = √ ψ e−jωt dt 2π −∞ 2 2 Aτ −(ω−ω0 )2 τ 2 /4 √ e . (5.17) = + e−(ω+ω0 ) τ /4 2 2 At any time t ≥ 0 and at any abscissa z ≥ 0, ψ (z, t) can be calculated as ∼ an inverse Fourier transform: +∞ 1 √ ψ (z, t) = Ψ(ω) e−j[k(ω) z−ωt] dω . (5.18) ∼ 2π −∞ Let us assume that the phase constant k, as a function of ω, can be expanded as a Taylor series∗ around ω0 : dk d2 k (ω − ω0 )2 (ω − ω0 ) + + ... . (5.19) dω dω 2 2 Suppose now that, throughout the spectral range where Ψ(ω) has a nonnegligible amplitude compared with Ψ(ω0 ), k varies slowly enough to neglect, in Eq. (5.19), all the terms of order larger than 2 in (ω − ω0 ). Moreover, take into account that, by deﬁnition, it is dk/dω = 1/vg , and introduce the symbol k = d2 k/dω 2 = −(1/vg2 ) dvg /dω. Substituting Eq. (5.19), within the approximation that we just made, into Eq. (5.18), yields: Aτ j[k(ω0 ) z−ω0 t] +∞ −ξ2 τ 2 /4 −j vzg −t ξ jk zξ2 /2 √ e e e e dξ ψ (z, t) = ∼ 4 π −∞ Aτ −j[k(ω0 ) z−ω0 t] +∞ −ξ2 τ 2 /4 −j vzg −t ξ −jk zξ2 /2 √ e + e e e dξ . 4 π −∞ (5.20) k(ω) = k(ω0 ) +

For k = 0, i.e., in a dispersionless medium, Eq. (5.20) becomes much simpler, and, using again Eq. (5.16), it gives: 2 2 . (5.21) ψ (z, t) = A cos ω0 t − k(ω0 ) z e−(t−z/vg ) /τ ∼ Hence, if there is no dispersion, at any z the wave packet is still modulated ∗ As for the part of the spectrum which is centered on −ω0 , it is enough to recall that, for uniform plane waves, k(−ω) = −k(ω).

© 2006 by Taylor & Francis Group, LLC

DISPERSION

139

Figure 5.1 An amplitude-modulated Gaussian pulse.

by a Gaussian envelope whose temporal width, measured at 1/e amplitude, is still 2τ (see Figure 5.1). Accordingly, with no dispersion the envelope travels undistorted, and its speed of propagation coincides with the group velocity, vg . In the presence of dispersion (i.e., for k = 0), the pulse envelope can no longer travel unchanged, as t grows. To show this, let us collect the ﬁrst and the third factor, in each integral in Eq. (5.20), into one complex exponential, and simplify the notation letting: t = t −

z vg

,

D=

2k z τ2

,

T 4 = τ 4 + 4k 2 z 2 = τ 4 (1 + D2 )

. (5.22)

Then, use again Eq. (5.16), which also holds for complex α2 . We get: 2 4 A √ 1 + jD ej[k(ω0 ) z−ω0 t−2k zt /T ] ψ (z, t) = 2 ∼ 2 1+D +

1 − jD e−j[k(ω0 ) z−ω0 t+2k

zt2 /T 4 ]

2

e−t

τ 2 /T 4

. (5.23)

This new expression diﬀers signiﬁcantly from Eq. (5.21), in the traveling wave factor at angular frequency ω0 as well as in the envelope. Changes in the traveling wave (terms in parenthesis) aﬀect mainly its phase, as the factor √ 1 ± jD is trivial, being constant in time. On the contrary, the change represented by the last term in the argument of the exponentials is signiﬁcant, because for z = constant it introduces a t2 time dependence, i.e., a distortion that can be detected with a phase-sensitive receiver. As for the envelope, changes due to dispersion do not aﬀect the motion of the pulse center, which travels again at the velocity vg , nor, in a qualitative sense, the pulse shape, which remains Gaussian. They aﬀect the pulse width, at 1/e amplitude, which is now given by: T2 = 2 τ 2 + 4k 2 z 2 /τ 2 = 2 τ 2 + 4k 2 vg2 t2M /τ 2 , 2Δt = 2 (5.24) τ where tM = z/vg is the time at which, at a given distance z, the pulse am-

© 2006 by Taylor & Francis Group, LLC

140

PLANE WAVE PACKETS AND BEAMS

plitude reaches its maximum. So, as it propagates, the wave packet spreads, and its width tends asymptotically to be proportional to the distance it covered: 2Δt → 4 |k | vg tM /τ . This occurs independently of the sign of k , i.e., both in the normal and in the anomalous dispersion regime. It is important to stress that this last point is a consequence of our choice for the wave packet model. There are other cases (typical ones are frequency-modulated pulses, or “chirped” ones, i.e., pulses that are both frequency and amplitude modulated), where a suitable choice in the dispersion sign may lead to pulse compression as propagation distance grows, until a minimum width is reached. After that, the pulse width starts growing again. Readers who are interested in this subject can ﬁnd further hints at the end of this chapter. The initial width τ appears in the denominator of the parameter which determines the rate at which the Gaussian pulse spreads vs. z. This is because, as one sees from Eq. (5.17), τ −1 is a measure of the spectral width of the pulse. The broader the pulse spectrum in the frequency domain, the higher its sensitivity to a given dispersion in the medium. The last comment should also be read as a warning, when one is dealing with pulses whose τ is very small. In such cases, the spectrum may be so broad, that one cannot neglect the higher-order terms in Eq. (5.19), and the whole content of this section may become inaccurate. Another warning is for the cases where k is not a smooth function of frequency, as happens often around resonant molecular or atomic transitions in material media. Also in these cases the results of this section may become invalid. 5.3

The scalar approximation

In the two previous sections, as well as in most of Chapter 4, we elaborated on one or more components of the e.m. ﬁeld, in a cartesian reference frame, in ways that were not signiﬁcantly aﬀected by the vector nature of these ﬁelds. It is self-explanatory why changing a vector problem into a scalar one is always a very attractive target, but in the case of e.m. waves, it can be reached in an exact way only in rare cases, like those of Chapter 4. Much more often, it can be reached through approximations, as we shall see, for instance, in Chapter 13. Most of the problems that we will deal with in the rest of this chapter belong to the latter class. Therefore, it may be useful, before considering individual cases, to discuss brieﬂy the reasons why a scalar treatment involves approximations, as well as the reasons why, notwithstanding the approximate nature of the approach, we use it; and ﬁnally, to ﬁnd criteria to assess the order of magnitude of the errors that this approach entails. The ﬁrst reason why, in most sections of this chapter, a scalar treatment cannot be rigorous, is because we deal often with waves in non-homogeneous media. In Sections 4.8 and 4.9, it was shown that, at a sharp discontinuity in either or μ, polarization of the incident ﬁeld plays a fundamental role. In a smoothly varying medium (i.e., where and μ are continuous and diﬀerentiable

© 2006 by Taylor & Francis Group, LLC

THE SCALAR APPROXIMATION

141

functions of space coordinates, as we shall suppose in the following sections), we must rediscuss the steps which in Chapter 1 led us from Maxwell’s equations to the Helmholtz equation. In general, ∇ · E = 0 and ∇ · H = 0, so we must take into account the terms ∇(∇ · E) and ∇(∇ · H) when we want to pass from ∇ × ∇× to ∇2 . On the other hand, we know that ∇ · (E) = 0 and ∇ · (μH) = 0, always. These relations can be rewritten in the following form, thanks to the vector identity (C.3): ∇ · E + ∇ · E = 0 , ∇μ · H + μ ∇ · H = 0 . From this we ﬁnd:

(5.25)

∇ · E ∇ × ∇ × E = −∇ E + ∇(∇ · E) = −∇ E − ∇ , ∇μ · H , (5.26) ∇ × ∇ × H = −∇2 H + ∇(∇ · H) = −∇2 H − ∇ μ 2

2

so in an inhomogeneous medium the Helmholtz equation must encompass one more term, which is sensitive to the direction of the vector E (or H) through the scalar product times ∇ (or ∇μ). This is an equation which is not easy to break down into scalar problems in an exact way. The reasons why, nonetheless, even in inhomogeneous media the scalar approximation is used very often, come on one hand from mathematical diﬃculties due to the additional terms in Eqs. (5.26); on the other hand, from the richness of the scientiﬁc literature on scalar waves (acoustic, elastic, quantummechanical waves, etc.), which obey the scalar Helmholtz equation also in inhomogeneous media. Results from these neighboring ﬁelds are often acceptable approximations for e.m. wave propagation also, especially when they lead to numerical methods. One consequence of the scalar approximation is that physical dimensions of the wave lose importance: all the cartesian components of ﬁelds and of the potentials satisfy the same equation, namely: ∇2 ψ − κ2 ψ = 0 .

(5.27)

So, the wave function ψ can have arbitrary physical dimensions. It is often convenient to assume that it is adimensional. Finally, let us look for conditions that must be satisﬁed in order to insure that errors due to the scalar approximation are acceptable. From Eqs. (5.25) and (5.26), we can extract a necessary condition: the medium has to be slowly varying, this phrase meaning that the moduli of the following vectors (whose dimensions are m−1 ): ∇ ∇μ , , (5.28) μ must be small, compared with the scale length over which E and H evolve in space. In some cases (as we will see in Section 5.6), one may ﬁnd additional,

© 2006 by Taylor & Francis Group, LLC

142

PLANE WAVE PACKETS AND BEAMS

more restrictive constraints, but whenever this is not the case, then the typical ﬁeld evolution scale is the wavelength, λ. Let us stress once more that this condition is only a necessary one. There are no suﬃcient conditions with a broad scope of validity; they must be assessed individually for each problem. 5.4

The equations of geometrical optics

Propagation of light, meaning electromagnetic radiation whose wavelength is small compared to the size of all the objects which are present along its path, is often approached, at the most elementary level, in a purely geometrical way. Let us see how we may connect rigorous e.m. theory, i.e., Maxwell’s equations, with simpliﬁed models in terms of “rays,” starting from the axiomatic theory of Chapter 1 and introducing suitable approximations. Taking completeness of plane waves as a generic hint, let us suppose that, in a lossless inhomogeneous medium, any solution of Eq. (5.27) can be written in the following form: (5.29) ψ(r) = C(r) e−jk0 S(r) , √ where k0 = ω μ0 0 is the intrinsic phase constant of free space, while C(r) and S(r) are two real functions of spatial coordinates, which we will call amplitude and phase functions, respectively. At a generic (but ﬁxed) point P = O + r0 , Eq. (5.29) looks like a uniform plane wave. For this reason, the assumption we have introduced now is usually referred to as a “ﬁeld expansion into local plane waves”.† The ansatz Eq. (5.29) enables us to introduce easily some approximations. Indeed, if we substitute Eq. (5.29) into Eq. (5.27), make use of some vector identities (see Appendix C), and divide by exp(−jk0 S(r)) = 0, we get:

(5.30) ∇2 C − jk0 (2∇C · ∇S + C∇2 S) − (∇S)2 k02 + κ2 C = 0 . Having assumed that C and S are real, we may equate to zero separately the second term, which is imaginary, and the sum of the ﬁrst and the third ones, which are real. Considering these, we have: (∇S)2 = −

κ2 1 ∇2 C + 2 2 k0 k0 C

.

(5.31)

The ﬁrst term on the right-hand side in this equation, equal to ω 2 μ/ω 2 μ0 0 , is always ≥ 1. The second term would be equal to 1, if C were a solution of the Helmholtz equation in free space. In such a case, C would vary in space with a period equal to λ0 = 2π/k0 . But in our case, remember that C is just the amplitude of the wave Eq. (5.29). If it varies appreciably on a length scale comparable to λ0 , then the applicability of the idea of local plane waves is not † If we allow the functions C(r) and S(r) to take complex values, then we encompass evanescent local plane waves in the expansion. This entails a remarkable broadening in the scope of validity, but the subject is now premature. We will come back to it in Section 5.13.

© 2006 by Taylor & Francis Group, LLC

THE EQUATIONS OF GEOMETRICAL OPTICS

143

justiﬁed. Hence, we may assume: |∇2 C| k02 |C|

.

(5.32)

Some constraints on the results that follow from this assumption will be outlined in Section 5.13. Eq. (5.31) is then approximated as follows: (∇S)2 = n2

,

(5.33)

where n = (μ/0 μ0 )1/2 is the refractive index, which was already deﬁned in Section 4.9, but now becomes, in general, a function of spatial coordinates. If one is able to solve Eq. (5.33), then one can determine the surfaces over which S(r) = constant, i.e., the constant-phase surfaces of the wave Eq. (5.29). As the reader may know from elementary optics, the shape of such surfaces is strictly related to the capability of a wave of creating “images” of its sources. For this reason, Eq. (5.33) is often referred to as the eikonal equation (or the equation of images, from Greek ικ˜ ω ν = image). Within the approximations that we have accepted when we wrote Eq. (5.32), geometry of the wave propagation is fully contained in Eq. (5.33). Indeed, if the ﬁeld can be expanded, at any point R = O + r, into local plane waves, then each of these waves, utilizing what we saw in Section 4.3, is characterized by a phase vector perpendicular to the constant-phase surface which passes through R. The lines which envelope the local phase vectors, i.e., the lines orthogonal to the surfaces S(r) = constant, are called ray paths. If what one aims at is to ﬁnd the ray paths, the procedure can be simpliﬁed, with respect to solving Eq. (5.33) and then calculating the orthogonal trajectories. We can ﬁnd from Eq. (5.33) another diﬀerential equation where the unknown is the position along the generic ray. To this purpose, we introduce (see Figure 5.2) a curvilinear coordinate u which varies along the ray with a metric coeﬃcient equal to unity (i.e., the ratio between any path length along the ray, and the corresponding change in the coordinate u, equals one). Then, u ˆ = dr/du equals the unit vector tangent to the ray, and d/du = u ˆ·∇ is the direction-derivative operator along the direction of the ray, orthogonal everywhere to a constant-phase surface. Thus we may write: ∇S = n

dr = nˆ u . du

(5.34)

To reach our goal, we must now get from Eq. (5.34) an equation where the function S does not appear explicitly. Let us consider the quantity ∇(n2 ) = 2n ∇n, which is also equal to ∇(nˆ u · nˆ u) and therefore can be expanded (see Eq. (C.2)) as follows: ∇(A · B) = (A · ∇) B + (B · ∇) A + A × (∇ × B) + B × (∇ × A) . Remembering that: ∇ × (nˆ u) = ∇ × ∇S = 0 ,

© 2006 by Taylor & Francis Group, LLC

(5.35)

144

PLANE WAVE PACKETS AND BEAMS

Figure 5.2 The curvilinear coordinate u, along the trajectory of a generic ray.

we get: 2n ∇n = 2n(ˆ u · ∇) (nˆ u) .

(5.36)

Dividing by 2n (diﬀerent from zero everywhere) and rewriting the direction derivative u ˆ · ∇ as d/du, we have: d dr n = ∇n . (5.37) du du Eq. (5.37) is called the ray equation. The procedures for its direct integration, frequently used for propagation in non-homogeneous media (like the atmosphere and the ionosphere at radio frequencies, graded-index lenses and optical ﬁbers at optical frequencies), are called ray tracing. Nowadays, in most cases they are implemented numerically, and many packages for computer-aided design based on ray tracing are commercially available. In several cases of practical interest, Eq. (5.37) is still too complicated to be integrated directly. Then, one makes use of a further simplifying approximation, based on the assumption that all the rays that matter in practice are nearly parallel to a direction, z, called the optical axis of the system. Examples in this sense can be optical instruments with several lenses, whose spacing is not small compared to their transverse dimensions, but they extend also to multiple-reﬂector antennas, together with their feeders. In all these cases we may introduce the so-called paraxial approximation: du dz

,

(5.38)

where z indicates now a cartesian coordinate, measured along the optical axis. So, Eq. (5.37) becomes: dr d n = ∇n . (5.39) dz dz This equation is much simpler than Eq. (5.37), as the derivative along the unknown direction of a ray has been replaced by an ordinary derivative. However, one has to memorize that Eq. (5.39) is a further approximation, compared with the eikonal equation. This can be proved, for instance, by computing the curl

© 2006 by Taylor & Francis Group, LLC

THE EQUATIONS OF GEOMETRICAL OPTICS

145

Figure 5.3 The ray equation as generalized Snell’s law: geometrical scheme. M = center of curvature.

of both sides of Eq. (5.39). Provided that all the continuity requirements be satisﬁed, so that the operators ∇× and d/dz commute, one obtains: d ∇ × (nˆ u) = 0 , (5.40) dz i.e., ∇ × (nˆ u) = constant along the z-axis; on the contrary, as we have shown shortly ago, the eikonal equation yields ∇ × (nˆ u) = 0 along a ray. These two results coincide exactly only for a ray strictly parallel to the z axis, whence the conclusion. Let us show now that Eq. (5.37) can be thought of as a generalized Snell’s law (Eq. (4.84)). Take the scalar product of Eq. (5.37) by any unit-vector vˆ lying in the plane orthogonal to ∇n, i.e., tangent to the n = constant surface that passes through the point R = O + r, which belongs to the ray we are dealing with (see Figure 5.3). We get: vˆ ·

dn dˆ u d (nˆ u) = vˆ · u ˆ +n · vˆ = ∇n · vˆ = 0 . du du du

(5.41)

In particular, if we let vˆ = ˆb (the unit vector orthogonal to the plane deﬁned by the vectors ∇n and u ˆ), then the scalar product u ˆ · ˆb vanishes. So (being n = 0), we obtain: dˆ u ˆ ·b=0 . (5.42) du This equation shows that, in the proximity of the point R, variations of the vector u ˆ along the ray must also be orthogonal to the unit vector ˆb, and consequently, co-planar with u ˆ and with ∇n. Hence, in the proximity of any point R, a ray is well ﬁtted by a planar curve. The restrictions which are expressed by the phrases “in the proximity of R” and “ﬁtted” can be relaxed, if, in the whole medium we are dealing with, the index distribution is such that the vector ∇n remains everywhere parallel to a given plane (see an example in Section 5.6). From what we said, we may conclude that dˆ u/du is the curvature vector of

© 2006 by Taylor & Francis Group, LLC

146

PLANE WAVE PACKETS AND BEAMS

the ray trajectory: dˆ u cˆ =c= du ρ

,

(5.43)

where ρ is the radius of curvature at the point R (see again Figure 5.3). Let ϕ now be the angle between u ˆ and ∇n, whose physical signiﬁcance is the same as that of the incidence angle of Section 4.8. Set now vˆ = a ˆ, the unit vector orthogonal to ∇n and belonging to the plane (ˆ u, ∇n). Then, from Eq. (5.43) we ﬁnally get: −

1 dn sin ϕ + n cos ϕ = 0 . du ρ

(5.44)

Note that, if the direction of ∇n remains constant in the neighborhood of R, then it is du = −ρ dϕ. Therefore, in such a case Eq. (5.44) can be rewritten in the form: d (n sin ϕ) = 0 , (5.45) du which is, as we anticipated, a generalized Snell’s law which holds, as we just said, when the direction of ∇n remains constant. This happens, for example, in the media where the refractive index depends only on one cartesian coordinate. In cylindrically symmetrical media, where ∇n is everywhere in the radial direction (see an example in the next section), this statement applies only to rays whose trajectories belong to planes passing through the symmetry axis, usually referred to as meridional rays. For skew rays, the reader can show as an exercise that what follows from the ray equation, Eq. (5.37), is n cos ϑz = constant, where ϑz is the angle between the symmetry axis of the medium and the tangent to the ray trajectory. This also can be though of as a generalized Snell’s law. The case of spherical symmetry is also discussed in some of the problems at the end of this chapter. In the paraxial approximation, direction derivatives are replaced by the ordinary derivative with respect to z. Restrictions on the consequent results have already been pointed out a shortwhile ago. Let us now go back to the equation that we get if we set the imaginary part of Eq. (5.30) equal to zero, namely: 2∇C · ∇S + C ∇2 S = 0 .

(5.46)

Multiply it by C, and make use of the identity (C.3) ∇·(ϕa) = ∇ϕ·a+ϕ∇·a, as well as of Eq. (5.34). The result is: u) = 0 . ∇ · (C 2 ∇S) = ∇ · (C 2 nˆ

(5.47)

In order to grasp the meaning of this relation, let us imagine a surface, T , which consists entirely of ray trajectories (see Figure 5.4). By deﬁnition, u ˆ is parallel to a ray everywhere. Hence, Eq. (5.47) indicates that this surface T is a “ﬂux tube” of the quantity C 2 n. In other words, if AS denotes that portion of a generic constant-phase surface (S = constant) which is inside T , then

© 2006 by Taylor & Francis Group, LLC

GEOMETRICAL OPTICS: ELECTROMAGNETIC IMPLICATIONS

147

Figure 5.4 A family of rays seen as a “ﬂow tube”.

Eq. (5.47) is equivalent to writing: C 2 n dAS ≡ |ψ|2 n dAS = constant AS

,

(5.48)

AS

(constant meaning independent of the curvilinear coordinate u measured along one of the rays lying on T ). Eq. (5.47), or equivalently Eq. (5.48), express the intensity conservation law of geometrical optics. Its electromagnetic implications will be discussed in the next section. These equations provide a “ﬂuid-dynamics” model of wave propagation. They are also a good starting point if one wants to point out the limitations of geometrical optics. Indeed, let us consider a focus, that is a point where, according to Eqs. (5.37), an inﬁnite number of rays converge into an inﬁnitesimal cross-section As . Then, Eq. (5.48) implies C → ∞. This result is not only physically unrealistic, it is also in contrast with the approximations that we introduced at the beginning of this section, based on Eq. (5.32). This point will be re-examined in the following sections. Going back to Eq. (5.46), we can deduce from it another relationship: by simply taking into account Eq. (5.34), we get: 1 2 dC 1 = − ∇2 S = − ∇ · (nˆ u) . C du n n

(5.49)

This equation is easy to integrate along a generic ray path, yielding: ! u ∇2 S ! u ∇·(nu) ˆ − du − du = C 2 (0) e 0 n . (5.50) |ψ(u)|2 ≡ C 2 (u) = C 2 (0) e 0 n The last formula can be used while tracing the rays, in order to track, from one point to the next, along a ray, how the intensity C 2 (u) evolves. 5.5

Geometrical optics: Electromagnetic implications

The basic equations of geometrical optics, Eqs. (5.33), (5.37) and (5.48), were obtained after the scalar approximation. In fact, they can be obtained also from the e.m. ﬁeld vectors, provided one makes an ansatz similar to Eq. (5.29), namely: E(r) = e(r) e−jk0 S(r)

© 2006 by Taylor & Francis Group, LLC

,

H(r) = h(r) e−jk0 S(r)

,

(5.51)

148

PLANE WAVE PACKETS AND BEAMS

where the vectors e and h have to be supposed as real. If we insert Eqs. (5.51) into Maxwell’s equations (without losses and without sources), the conclusions we reach are the same as in the previous section. The main reason why we repeat this procedure is simply because we want to avoid a misunderstanding. We do not want the reader to believe that a “vectorial” theory of geometrical optics could be more accurate that the “scalar” one. Indeed, let us show that, when one arrives at the equations of geometrical optics, then the problem has already become a scalar one, perhaps without us noticing it, even if one starts from the ansatz in Eq. (5.51) and makes use of Maxwell’s equations. To show that, we introduce Eqs. (5.51) into Maxwell’s equations. We get: ∇ × E ≡ (∇ × e) e−jk0 S + (−jk0 ∇S) × e e−jk0 S = −jωμ h e−jk0 S , (5.52) and likewise for ∇ × H. Divide by k0 exp(−jk0 S) and neglect, for the same reasons that were outlined at the beginning of the previous section, the terms in k0−1 . We obtain: ∇S × e nη h , ∇S × h −n

e η

,

(5.53)

where the refractive index n = (μ/μ0 0 )1/2 and the wave impedance η = (μ/)1/2 are now functions of spatial coordinates, since in general the medium is not homogeneous. Taking into account Eq. (5.34), we get: e . (5.54) u ˆ×e ηh , u ˆ×h− η From Eqs. (5.54) we may deduce some interesting results, that were implicit in the scalar theory of the previous section. In particular, note that these results depend critically on S(r) being a real-valued function. A) The vectors e, h and ∇S (or u ˆ) are mutually orthogonal. Therefore, E and H are transverse, at any point R, with respect to the ray path passing through R. B) As η is real, e and h, and consequently E and H, are in phase, at any point R. C) As a consequence, the Poynting vector P = E × H∗ /2 is everywhere real and parallel to the ray direction. This is the reason why it became common, even in other contexts, to call “direction of the electromagnetic ray” the direction of the real part of the Poynting vector. We will speak again about this terminology in Chapter 6. D) The wave impedance in the ray direction is, at any point R, equal to η(R), the intrinsic impedance of the medium (or rather, its local value at the point R). E) From the previous points A) and D) we can infer that, at any point R, the time-averaged values of the electric and magnetic energy density equal

© 2006 by Taylor & Francis Group, LLC

EXAMPLES OF RAY TRACING

149

each other: 1 1 (R) E(R) · E∗ (R) = μ(R) H(R) · H∗ (R) . 4 4

(5.55)

F) From Eq. (5.55) and from Poynting’s theorem, as there are neither losses nor sources, it follows that the divergence of the Poynting vector equals zero everywhere. Based on this and on the previous point C), Eq. (5.48) indicates that the conserved quantity, referred to as the intensity, in the scalar approximation |ψ|2 n, is proportional to the modulus of the Poynting vector : |H|2 |E|2 ≡η . (5.56) |ψ|2 n ∝ |P| ≡ 2η 2 The physical dimensions of the proportionality constant in the previous relationship are not yet speciﬁed, since the physical dimensions of the scalar wave function ψ are unspeciﬁed as well. However, we may notice that Eq. (5.56) makes it necessary that the refractive index n shows up explicitly in the relationship between ψ and the e.m. vectors E, H. Furthermore, n is involved implicitly, through the local wave impedance η. This point will be raised again and clariﬁed in Section 5.7. In conclusion, we have shown in this section that the approximations of geometrical optics imply that, locally at any point, the e.m. ﬁeld has all the attributes of a uniform plane wave. Therefore, referring to such a ﬁeld as a set of local plane waves, as we did before (after Eq. (5.29)), appears to be a legitimate procedure. Clearly, this approach is intrinsically an approximation. Indeed, we have shown in Chapter 4 that a wave, when impinging on a steep change in parameters of the medium, must be reﬂected in part. The theory we developed here loses any track of reﬂection, as shown by the invariance of the ﬂux of Poynting vector along a ﬂux tube. This is one of its basic drawbacks, or at least limitations. 5.6

Examples of ray tracing in radio propagation and in optics

An isotropic spherically symmetric medium, whose refractive index decreases monotonically as the radial coordinate increases, is a reasonable model for studying radio wave propagation in the earth atmosphere, at heights below the ionosphere. Assume that above the ground (i.e., for r > r0 ) the refractive index varies as:

n(r) = n0 1 − a(r − r0 ) = n0 [1 − ah] , (5.57) where n0 = n(r0 ) is the refractive index at the ground level, and h = r − r0 is the height above the ground. For r < r0 , n(r) is not deﬁned. If we restrict ourselves to heights such that |a| h 1, then Eq. (5.57) can be looked at as a truncated Taylor series, for a widely arbitrary function n(r), being a =

© 2006 by Taylor & Francis Group, LLC

150

PLANE WAVE PACKETS AND BEAMS

Figure 5.5 A spherical coordinate reference frame, to solve the ray equation in a spherically symmetrical medium.

−(1/n0 ) (dn/dr)r=r0 . Therefore, the model is more general than one may think at ﬁrst sight. In spherical coordinates‡ (Figure 5.5), it is convenient to re-write the ray equation (5.37) recalling Eq. (5.43) and using the following identity: d ∂ 1 ∂ =u ˆ · ∇ = cos ϕ + sin ϕ du ∂r r ∂ϑ

.

We obtain then: ∂n 1 ∂n 1 d (nˆ u) ≡ cos ϕ u ˆ + sin ϕ u ˆ + n cˆ = ∇n ≡ −n0 aˆ r. du ∂r r ∂ϑ ρ

(5.58)

(5.59)

The term ∂n/∂ϑ is equal to zero, because of the spherical symmetry. Since cˆ and u ˆ are orthogonal, we see that u ˆ, cˆ and rˆ lay on the same plane and, consequently, the trajectory of any ray is on a plane passing through the origin. Integration of Eq. (5.59) becomes very simple in the cases (quite relevant for terrestrial radio links) where the ray trajectories have little inclination with respect to the horizon. Then, we may assume ϕ π/2, | cos ϕ| 1 everywhere, while ∂n/∂r and n/ρ are of the same order of magnitude. Therefore we get the following equation: n

1 cˆ = −n0 aˆ r , ρ

(5.60)

which shows that, for a > 0, the unit vector cˆ is oriented towards the center of the sphere, indicating that rays bend towards the ground as they propagate. Furthermore, if we assume n n0 on the left-hand side, Eq. (5.60) gives: ρ=

1 = constant . a

(5.61)

‡ Note that, to be consistent with symbols used in the previous section, ϕ is not a coordinate angle in the spherical reference frame, but the angle between the ray and the index gradient. Anyhow, thanks to symmetries in the problem, two spherical coordinates, r and ϑ, will be enough in this section.

© 2006 by Taylor & Francis Group, LLC

EXAMPLES OF RAY TRACING

151

Figure 5.6 Radio transmission beyond the geometrical horizon.

This means that, within this approximation, all the ray trajectories are circles of radius (1/a). If (1/a) > r0 (which is the normal situation in the atmosphere), then it is possible to draw circles of this family, which pass through two points above the ground, without intersecting the earth surface (see Figure 5.6). This result implies that a radio link can be established between two points each of which is beyond the geometrical horizon with respect to the other. As a second example, let us deal with cylindrical symmetry, i.e., a medium whose refractive index depends only on the distance r from an axis, which we take as the z axis of a cylindrical reference frame (Figure 5.7). This can be a model for some types of graded-index optical waveguides (or lenses), to be studied in more detail in Section 10.9 and in the following ones. In particular, we concentrate on the so-called parabolic index proﬁle:

, (5.62) n2 = n20 1 − (r/a)2 where n0 is the maximum index, on axis, and a is a suitable normalizing radius, which sets the scale on which the index decreases as we get away from the axis. The region of interest for ray propagation is where one can assume: r/a 1 . 2

(5.63) ) (n20 /n) rˆ,

in the radial direction The index gradient is found to be −(r/a and always inwards: the cylindrical medium is a focusing one. In the paraxial approximation du dz, no derivative of n appears on the left-hand side of the ray equation, Eq. (5.37), which then reads: dˆ u = −ˆ r(r/a2 ) (n0 /n)2 −ˆ r(r/a2 ) . (5.64) dz This shows that the ray is subject to a radial “pull-back” force, proportional to the elongation away from the z axis. Therefore, by analogy with mechanics of a simple pendulum, the reader may conclude that those solutions of Eq. (5.64) that are contained in a plane passing through the z axis are sinusoidal trajec-

© 2006 by Taylor & Francis Group, LLC

152

PLANE WAVE PACKETS AND BEAMS

r r

a

4

u

∇n

0

n20

n2

z ∇n

-4

Figure 5.7 Ray trajectories in a parabolic-index slab.

tories (see again Figure 5.7). They are called meridional rays. Combinations of sinusoidal motions with equal periods in two orthogonal planes give rise to skew rays, which are, in the most general case, helical trajectories. The reader should, as an exercise, explain why the assumption that the two periods are equal is justiﬁed by Eq. (5.64). For a signiﬁcant number of additional examples of ray tracing in structures which are of interest to optical communications, the reader may consult, for instance, van Etten and van der Plaats (1991). 5.7

The WKBJ method

The subject of this section, although closely related to the previous one, is not easy to link with intuition. It looks advisable to postpone the discussion of its physical meaning and of its relationship with the previous topics, and to outline its mathematical treatment ﬁrst. The technique which is currently referred to as the WKBJ method (from the initials of the four mathematicians, Wentzel, Kramers, Brillouin, and Jeﬀreys, who initiated it, independently of each other) will be outlined in this section in its version for uniform plane waves. The same method can be applied to many other circumstances; examples in this sense will be discussed in the next section and again in Chapter 10. Similarly to Section 5.4, let us suppose that the equation to be solved is (5.27). So, we assume that the scalar approximation holds. Consider a lossless medium, and suppose it is inhomogeneous just because and μ vary only in one direction which we take as the z axis of an orthogonal reference frame: = (z)

,

μ = μ(z) ,

κ2 = −ω 2 μ(z) (z) = −k 2 (z) .

(5.65)

Another assumption, which links the WKBJ method to plane waves, but makes an important diﬀerence with respect to the previous section, is that we

© 2006 by Taylor & Francis Group, LLC

THE WKBJ METHOD

153

want the wave function be separable, namely: ψ(x, y, z) = φ(x, y) f (z) .

(5.66)

Consequently, to solve the scalar Helmholtz equation: ∇2 ψ − κ2 ψ = 0 ,

(5.67)

we have merely to solve the following ordinary diﬀerential equation:

d2 f 2 2 2 = κ (z) + k , t f ≡ −β (z) f dz 2

(5.68)

where kt2 is a separation constant. We will restrict ourselves to the case kt2 < ω 2 μ, i.e., β 2 (z) > 0. The case β 2 (z) < 0 is thoroughly treated in the literature (e.g., Budden, 1961). For β 2 an arbitrary function of z, Eq. (5.68) does not admit an exact solution. The WKBJ approximate solution requires that β 2 (z) be slowly varying on the scale of a local characteristic length, Λ(z) = 2π/|β(z)|: 2 dβ |β| = β . (5.69) dz Λ 2π To clarify the meaning of this characteristic length, let us refer to the case where the unknown function is a plane wave; then, we have kt2 = 0, β 2 = −κ2 , and the condition in Eq. (5.69) is equivalent to assuming that the parameters of the medium vary slowly with respect to the local wavelength λ = 2π/k(z): d(μ) μ (5.70) dz λ . Note that Eq. (5.69) is more general than Eq. (5.70). This will be very useful in the following, mainly in Chapter 10. Let us now set: f (z) = eg(z) . (5.71) Diﬀerentiating with respect to z, we obtain: f = (g 2 + g ) f

.

(5.72)

g 2 + g = −β 2 (z)

.

(5.73)

Then, Eq. (5.68) becomes:

Now it is time to exploit the assumption in Eq. (5.69). Remember that, in a homogeneous medium (β = constant), Eq. (5.27) has uniform plane waves as solutions, for which g(z) = −jβz, and then g = −jβ. Therefore, let us presume now, and verify later, that, when the condition in Eq. (5.69) holds, we are entitled to say that: |g 2 | |g |

.

(5.74)

Then, a ﬁrst approximation to Eq. (5.73) reads: g 2 −β 2 (z) .

© 2006 by Taylor & Francis Group, LLC

(5.75)

154

PLANE WAVE PACKETS AND BEAMS

Anyway, it is easy to check that this approximation is not satisfactory. Eq. (5.75) would give an imaginary g , whence also an imaginary g(z), and so, through Eq. (5.71), |f (z)| = constant. The corresponding wave function ψ = φf would not even satisfy energy conservation, which is respected by geometrical optics. In conclusion, this approximation is actually worse than geometrical optics. In order to improve on Eq. (5.75), we can use a result which is a straightforward consequence of Eq. (5.75) itself, namely: dβ , (5.76) dz and substitute it into the exact equation (5.73). The binomial expansion of the square root, cut oﬀ at the second term (as allowed by Eq. (5.74)), gives then: dβ 1 d ln(β 1/2 ) . (5.77) g ±j β(z) 1 ± j = ±j β(z) − 2 2β (z) dz dz g ±j

Integrating Eq. (5.77), and substituting the result into Eq. (5.71), we get: ! A ±j z β(z ) dz 0 f (z) = 1/2 e , (5.78) β (A = arbitrary constant), which is the ﬁrst-order WKBJ solution, one of the key results of the method. Eq. (5.78) expresses the phase delay of the wave as the result of an inﬁnite number of inﬁnitesimal phase shifts. This is the reason why the integral in the previous exponential function is sometimes referred to as phase memory. The factor β −1/2 in front of the exponential means intensity conservation: in the one-dimensional case (i.e., for φ(x, y) = constant), this factor plays the same role as n−1/2 in the intensity conservation law of geometrical optics |ψ|2 n × area = constant. Hence, we have many indications that geometrical optics and the WKBJ method are strictly interconnected. Indeed, it can be shown that the two methods give results of equal accuracy, measured as the number of powers of Λ that are accounted for in a series expansion of the unknown wave functions and of the equations they must satisfy. Nevertheless, the two theories diﬀer in their scope of application. In principle, the assumption of separation of variables makes the WKBJ method less general than geometrical optics. On the other hand, in practice that assumption makes the WKBJ method more versatile than geometrical optics. For example, one can apply WKBJ to study the dependence of the wave function on a coordinate which, in the paraxial approximation (see Section 5.4), is orthogonal to the z direction. This point will be dealt with in detail in Section 10.10. It is easy to show that also the WKBJ method collapses under some critical conditions, which make at least its ﬁrst-order version, which we saw so far, not applicable. These conditions coincide with those that are critical for geometrical optics, that is to say, they occur near the focal points. Indeed, if we insert

© 2006 by Taylor & Francis Group, LLC

FURTHER COMMENTS ON THE WKBJ METHOD

155

the approximate solution in Eq. (5.78) into the diﬀerential equation (5.73), we ﬁnd that the validity of that solution is subject to the following necessary condition: 3 1 dβ 2 1 1 d2 β − (5.79) β2 . 4 β dz 2 β dz 2 This condition is not satisﬁed where β(z) ∼ = 0, i.e., where kt2 = −κ2 . The surfaces where β(z) = 0 are called caustics. It can be shown, with a rather lengthy procedure, that the same surfaces are loci of focal points when the same problem is examined in terms of ray tracing. In Section 5.13 we will quote brieﬂy some procedures to overcome these restrictions. Eq. (5.79) is more restrictive than Eq. (5.69), which was our starting point. Indeed, Eq. (5.69) is veriﬁed also in the proximity of the extrema of β(z), but in general near such points there are no restrictions on the second derivative of β(z), so Eq. (5.79) is not necessarily satisﬁed. One must then adopt additional care when applying the WKBJ solution near a maximum or minimum of β(z). Finally, let us stress that the WKBJ method is intrinsically a scalar method, like geometrical optics. It is not possible to gain in accuracy by going to a vectorial formulation of the same method. In fact, if one looks for a WKBJ solution for an e.m. ﬁeld, i.e., for a behavior like Eq. (5.78) for all components of the vectors E and H, then one must suppose that every component separately satisﬁes an equation like Eq. (5.73). This entails necessarily a scalar approximation, i.e., an implicit assumption that the double curl operator is replaced by a Laplacian operator even in non-homogeneous media, where this step is not rigorous. 5.8

Further comments on the WKBJ method

Several propagation experiments (for example, radio wave sensing of the ionosphere, or testing of graded-index optical waveguides) deal with slowly varying media of unknown characteristic. One tries to exploit the results of the WKBJ method to evaluate the parameters of the medium from measurements on waves that propagate through it (“inverse WKBJ”). If what is measured is a wave transmitted through the unknown medium, then the equations to be used are those of the previous section. In particular we may use Eq. (5.78), where the value of f (z) at the receiver (z = z0 ) is now a known quantity, and the quantity to be determined is β(z ) for 0 ≤ z ≤ z0 . It is evident that the known quantities are not enough to evaluate β(z ). One needs additional information. Usually, one starts from a model of β(z ) which contains a number of unknown parameters equal to the number that can be evaluated from Eq. (5.78). These unknown parameters are then ﬁtted to the experimental data. Often, however, transmitted waves are much more diﬃcult to measure than reﬂected waves. For example, ionospheric soundings become much simpler to perform if the receiver is placed near the transmitter. This requires the test waves to propagate vertically. However, the WKBJ method is not the most

© 2006 by Taylor & Francis Group, LLC

156

PLANE WAVE PACKETS AND BEAMS

suitable one in order to evaluate distributed reﬂections, due to slow variations in β(z). Indeed, the two solutions of Eq. (5.78), corresponding to waves traveling in opposite directions, look as if they were completely decoupled; but physical intuition says that a strong localized reﬂection should be expected to occur where the incident wave (e.g., Eq. (5.78) with a − sign) reaches a caustic, i.e., points where β = 0. This intuitive idea relies on the fact that beyond the caustic β(z) becomes imaginary, meaning that there our wave becomes evanescent, a situation which is reminiscent of total reﬂection (Section 4.11). To solve this problem with the inverse WKBJ method, one is obliged to work in a region where the approximations of the previous section fail. Anyway, it can be shown (although the proof is too time-consuming to be reported here) that the corrections required because Eq. (5.69) does not hold are small, and physically intuitive. We may then begin to discuss reﬂections starting from the results of the previous section, in particular from Eq. (5.78). Let us set z = 0 at the point of observation of the reﬂected wave, and call z0 the height of the caustic, i.e., β(z0 ) = 0. Because of what we said above, we suppose that Eq. (5.78) applies over the whole range 0, z0 . If fi and fr are the incident and the reﬂected wave, respectively, then we have: " ! β(0) −j z β(z ) dz 0 e fi (z) = fi (0) , (5.80) β(z) " fr (0)

β(z0 ) −j e β(0)

= fr (z0 ) " = fr (z0 )

β(z0 ) j e β(0)

!0 z0

! z0 0

β(z ) dz

β(z ) dz

.

(5.81)

Our goal is a relationship between the reﬂection coeﬃcient at z = 0, ρ = fr (0)/fi (0), and that at z0 . We need ﬁrst to ﬁnd a relationship between fi (z0 ) and fr (z0 ). This cannot be derived rigorously from Eq. (5.78), because this equation gives fi → ∞ and fr → ∞ as z → z0 . One way to overcome the problem could be to suppose that these two inﬁnites are equal: fi (z0 ) = fr (z0 ). Then we would obtain: ! z0 −j2 β(z ) dz 0 . (5.82) ρ=e In reality, this is an arbitrary statement. Eq. (5.78) says only that both wave amplitudes diverge when β(z) → 0. To be more accurate, one can show (e.g., Budden, 1961), through a more careful approximation, that the solutions of Eq. (5.73) near a caustic can be expressed as combinations of Airy functions (Abramowitz and Stegun, Eds., 1965). The only change that this entails with respect to Eq. (5.82) is simply a phase shift of π/2 radians. The ﬁnal result is then: ! z0 π ! z0 −j2 β(z ) dz −j 2 β(z ) dz + 2 0 0 =e . (5.83) ρ=je As we anticipated, this phase correction is easy to link to physical intuition.

© 2006 by Taylor & Francis Group, LLC

GAUSSIAN BEAMS

157

Indeed, it expresses, within the limits of the scalar approximation, the phase shift that we found in Section 4.11 for total reﬂection, which we called then the Goos-H¨ anchen shift. Here as well as there, the shift is a consequence of passing from a real wave impedance to an imaginary one, at z = z0 . 5.9

Gaussian beams

In the previous sections, we gained some experience on how to ﬁnd approximate solutions of Maxwell’s equations in slowly varying media. We can now exploit that experience, to ﬁnd some other approximate solutions, ﬁrst in a homogeneous medium, although they can be extended to some classes of nonhomogeneous media as well. They are physically more realistic than a single plane wave, particularly because they carry a ﬁnite power. At the same time they are still reasonably simple. In a homogeneous lossless medium, the vector Helmholtz equation is strictly valid. It can be written by components in any cartesian reference frame, like for plane waves. So, it is easy to tackle in scalar terms. Therefore, let us consider just one component of the electric ﬁeld, with an angular frequency ω. Let us call (z) the direction, for the time being chosen with total freedom, of a √ phase vector k (|k| = ω μ). We may always write, without any physical implication: E(r) = φ(r) e−jk·r

.

(5.84)

This is simply a change of variables, but soon it will allow us to give a physical meaning to some approximations that we shall make. For a uniform plane wave traveling in the direction of k, we know that φ(r) is a constant. Within the approximation of geometrical optics, the amplitude function C is slowly varying. Therefore, it appears legitimate to see what happens to Eq. (5.84) if the variations of φ vs. z are slow, on the length scale of λ = 2π/k. If we recall the ﬁeld expansion as an integral of plane waves (Section 4.7), then we understand why a ﬁeld like Eq. (5.84) with slowly varying φ is called a beam of plane waves propagating in the z direction. Sometimes it is also called a ray beam. Quantitatively, the approximation we need to make reads: 2 ∂ φ 2k ∂φ . (5.85) ∂z ∂z 2 If we insert Eq. (5.84) into the scalar Helmholtz equation, and neglect the term ∂ 2 φ/∂z 2 because of Eq. (5.85), we get: ∂2φ ∂2φ ∂φ + 2 = 2jk 2 ∂x ∂y ∂z

.

(5.86)

This equation is rather unusual in classical electrodynamics, but, on the contrary, is very well known in Quantum Mechanics, where it is referred to as the (two-dimensional) Schr¨ odinger equation for a free particle, and is usually

© 2006 by Taylor & Francis Group, LLC

158

PLANE WAVE PACKETS AND BEAMS

written as: 2m ∂φ , h ¯ ∂t where φ is the wave function associated with the particle, m is the particle mass, t is time, h ¯ = h/2π, h being Planck’s constant. This equation has been studied so widely, that we can adapt to our needs solutions that can be found in the literature. In particular, we will now study in depth the simplest element in a complete set of discrete, linearly independent solutions. The other elements of this set will be examined in Section 5.10, and shown to share many basic features of the ﬁrst solution. Let then φ be a complex function, whose amplitude, in the generic transverse plane z = constant, varies like a circularly symmetrical Gaussian function: k 2 φ = A exp −j P (z) + r , (5.87) 2q(z) ∇2 φ = −j

where r 2 = x2 + y 2 . The complex quantities P (z) and q(z) are still unknown. Inserting the ansatz (5.87) into Eq. (5.86), we ﬁnd: k2 j (5.88) 2k P + + r2 2 (1 − q ) = 0 , q q where the primes denote derivatives with respect to z. The quantities in brackets are functions only of z, the ﬁrst bracket is multiplied by a constant, while the second one is multiplied by a function of x and y. Therefore, Eq. (5.88) can be satisﬁed everywhere only if the two quantities in brackets are separately equal to zero: q

= 1 ,

P

= −

j q

(5.89) .

(5.90)

The general integrals of these two equations are simply found to be respectively: q P

= z + q0

, z = −j ln 1 + q0

(5.91) .

(5.92)

In the second integral, an arbitrary integration constant B has been omitted, since in Eq. (5.87) P appears in the argument of an exponential function, so the factor exp(B) can be incorporated into the arbitrary constant A. The function q(z), being complex, plays two roles, through Eq. (5.87). It aﬀects the length scale of the Gaussian dependence on x and y in the transverse plane z = constant, and also the shape of the constant-phase surfaces of the wave. As for P (z), it characterizes the slow variations in amplitude and phase of φ as a function of z. However, Eq. (5.87) becomes simpler to analyze if we replace P (z) and q(z), which were useful in making the diﬀerential equations

© 2006 by Taylor & Francis Group, LLC

GAUSSIAN BEAMS

159

Figure 5.8 Behavior of the complex function q vs. the longitudinal coordinate z.

easy, with new variables, whose physical meaning is more direct. First, let us separate real and imaginary parts of 1/q: 1 1 1 = − 2j q R kw2

,

where R and w are real function of z. Eq. (5.87) then becomes: kr2 r2 φ = A exp −j P (z) + . − 2 2R w

(5.93)

(5.94)

From this expression we can easily grasp the signiﬁcance of R(z) and w(z). Recalling Eq. (5.84), we see that the ﬁeld is given by φ multiplied by exp(−jkz); variations in the phase of φ due to P (z) are slow, with respect to λ (see Eq. (5.92)). Consequently, the constant-phase surface passing through the generic point of coordinates (x, y, z) satisﬁes the equation k(z +r 2 /2R) = constant. This says that the surface is an ellipsoid, but near that point it is well approximated by a spherical surface, centered on the z axis, and of radius R. So, physically R(z) is the radius of curvature of the wavefront at the point (0, 0, z). As for w(z), from Eq. (5.94) we see that it equals the radius of a circle, on the transverse plane z = constant, where the modulus of φ equals 1/e times its value on the z axis. It is clear why 2w is referred to as the Gaussian beam diameter, or the spot size. From Eq. (5.91) it is clear that there is one (and only one) value of z, for which the real part of q vanishes (while the imaginary part is independent of z). This is in fact a peculiar z = constant plane, because Eq. (5.93) shows that, when Re(q) = 0, then 1/R = 0. Furthermore (see Figure 5.8), w takes its minimum value on the same plane. Hence, this plane is referred to as the beam waist plane. From now on, it is convenient to choose it as the z = 0 plane. Let w0 denote the minimum of w. In the new reference frame, we have

© 2006 by Taylor & Francis Group, LLC

160

PLANE WAVE PACKETS AND BEAMS 5

R(z)/ z

w(z)/ w

0

4

3

2

1

0

0

1

2

3

4

λ | z|/(πw 20 ) Figure 5.9 Beam width and radius of curvature, vs. normalized distance from the beam waist.

q0 = q(z = 0), so that:

w02 . (5.95) 2 This simpliﬁes very much the z-dependencies of w and R. Indeed, from Eqs. (5.91), (5.93) and (5.95), letting k = 2π/λ, we get: 2 zλ w2 (z) = w02 1 + , (5.96) πw02 q0 = jk

R(z) = z 1 +

πw02 λz

2 .

(5.97)

We see that the wavefront curvature and the relative width of the beam depend on only one variable, namely the ratio between the area of the beam waist, πw02 , and the product λz. They are plotted in Figure 5.9. As we see, the narrower the spot on the beam waist, with respect to λ, the faster the beam diverges as we move away from the waist, and the larger the curvature of the wavefront at a given distance |z|. This is in perfect agreement with diﬀraction theory (see Chapter 13), but in sharp contrast with geometrical optics. This is the ﬁrst important indication that Gaussian beams are indeed an improved approximation, compared to ray optics, especially in the vicinity of focal planes. Eq. (5.96) is, on the plane (z, w), the equation of a hyperbola (Figure 5.10) whose asymptotes are given by: w=±

© 2006 by Taylor & Francis Group, LLC

λ z πw0

.

(5.98)

GAUSSIAN BEAMS

161

Figure 5.10 The 1/e hyperbola and its asymptotes.

Hence, far from the beam waist, i.e., for z πw02 /λ, the beam diverges linearly vs. z, with an angular aperture: 2ϑ = 2 arctan

λ 2 = 2 arctan πw0 kw0

.

(5.99)

This relation conﬁrms that a tightly focused Gaussian beam (small spot size) diverges rapidly, away from the waist plane, while if one wants a well collimated beam, i.e., a beam consisting of almost parallel rays, it is necessary that the beam cross-section be not too small. This comment is fundamental not only in optics, but also in the case of aperture antennas. It expresses the socalled “uncertainty principle” between “conjugate variables,” i.e., between two quantities which span the domains of two functions related to each other by a Fourier transform. It is, in this sense, the same as the relationship between widths of a signal in the time domain and in the frequency domain, which we recalled in Section 4.7. In the present case, the Fourier transform is the plane-wave expansion of the ﬁeld on the beam waist plane. Let us add a short comment, which stems from terminology, but whose real purpose is to stress the underlying links with other subjects, to be dealt with in some of the following chapters. The quantity: πw02 w02 = (5.100) 2 λ is called the Rayleigh range. This name is deliberately reminiscent of a result obtained by Lord Rayleigh, at the end of the 19th Century: in a focused pencil of rays, when the distance z from the focal plane is less than z0 , results based on geometrical optics are severely contradicted by experiment. The physical phenomenon that governs proagation in that range is diﬀraction. This is a further conﬁrmation that Gaussian beams are indeed the big improvement we were looking for, with respect to ray optics. To give Eq. (5.94) a ﬁnal shape, let us substitute Eq. (5.95) into Eq. (5.92). We get: 2z 2z 2z + j arctan − 2 j P (z) = ln 1 − j = ln 1 − j . (5.101) kw02 kw02 kw0 z0 = −jq0 = k

© 2006 by Taylor & Francis Group, LLC

162

PLANE WAVE PACKETS AND BEAMS

Note that:

2 1/2 2z λz w(z) 1 − j = 1+ − = kw02 πw02 w0

,

(5.102)

λz χ = − arctan − 2 . (5.103) πw0 Incidentally, we point out that for any z this quantity χ is between ±π/2. This reconﬁrms our previous statements, that contributions to the phase coming from the factor encompassing P (z) were much smaller than the others. We then get: w0 −jχ e−jP (z) = e , (5.104) w(z) and the whole solution of Eq. (5.86) ﬁnally reads:

kr2 w0 exp −j(kz + χ) − j E=A . (5.105) exp −r2 /w2 (z) w(z) 2R(z) and let:

Note that the factor w0 /w(z) is necessary in order to maintain the power ﬂux through the z = constant planes independent of z, as required by the medium being lossless and homogeneous. In order to strengthen the ties between the contents of this section and their technical fallouts, Table 5.1 shows a set of basic data regarding some of the most popular state-of-the-art laser sources. 5.10

Hermite-Gauss and Laguerre-Gauss modes

Before making the ansatz (5.87), we justiﬁed why the previous section was devoted entirely to one solution of Eq. (5.86). A countable inﬁnity of solutions, forming a complete set, share all the basic features of Eq. (5.105). The completeness of the set will be shortly discussed in Section 5.12. In this section, we will look for the other solutions of Eq. (5.86), starting with a less restrictive ansatz than Eq. (5.87), with more freedom for the dependence of φ(x, y, z) on the transverse coordinates x, y. As we aim at separating variables, we are tempted to set: k 2 r + ξ(z) , (5.106) φ = f1 (x) f2 (y) exp −j P + 2q where f1 (x), f2 (y) and ξ(z) are unknown functions of one coordinate each. Suppose that P and q remain the same functions of z as in the previous section. Then, inserting Eq. (5.106) into Eq. (5.86), exploiting Eqs. (5.89) and (5.90), and dividing by f1 f2 , we ﬁnd: f2 f1 2kx f1 2ky f2 −j −j (5.107) + = 2kξ , f1 q f1 f2 q f2 where the primes denote ordinary derivatives, without any ambiguity. Contrary to what may appear at ﬁrst sight, Eq. (5.107) does not allow

© 2006 by Taylor & Francis Group, LLC

HERMITE-GAUSS AND LAGUERRE-GAUSS MODES

163

Table 5.1 State-of-the-art lasers. Name

Active medium

Wavelength (μm)

Ruby

Crystal (Cr3+ in Al2 O3 )

0.6943

Neodymium: Yag

(Nd

3+

Crystal in Y3 Al5 O12 )

Typical angular divergence (radians)

Typical power range 2.5–25 kW (pulsed)

1.0641

0.5–2×10−3

1–100 W (cw)

Helium-Neon

Gas mixture

0.6328 (1.15, 3.39)

10−3

0.5–2.5 mW (cw)

Argon Ion

Ionized gas (Ar)

0.35–0.52 (typ. 0.488)

< 10−3

0.2–20 W

Carbon Dioxide

Gas mixture (CO2 )

10.4

1.5–2.5×10−3

1 W–1.5 kW (cw)

Gallium Arsenide

Semiconductor (GaAs)

∼ 0.85

∼ 0.3 (in the juction plane) ∼ 1 (normal to the juction)

< 100 mW ∼ (cw)

All-ﬁber

Er-doped glass ﬁber

1.55

10−1

< 100 mW ∼ (cw)

separation of variables. In fact, the two terms on the left-hand side contain q, which is a function of z. To reach our goal, rather than writing Eq. (5.106) we have to let: k 2 r + ξ(z) , (5.108) φ = g1 (t1 ) g2 (t2 ) exp −j P + 2q where:

√ x √ y , t2 = 2 , (5.109) 2 w w are transverse coordinates normalized to the width (5.96) of the Gaussian function, which depends on z. If we substitute Eq. (5.108) into Eq. (5.86), take correctly into account all the terms like (dgi /dti ) (dti /dz), and make use of the following relationship, which comes from Eqs. (5.96) and (5.97): t1 =

2 then we obtain:

g1 g − 2t1 1 g1 g1

d(w2 ) w2 = R dz

+

,

g2 g − 2t2 2 g2 g2

(5.110)

= kw2 ξ

,

(5.111)

where now primes denote derivatives with respect to ti on the left-hand side,

© 2006 by Taylor & Francis Group, LLC

164

PLANE WAVE PACKETS AND BEAMS Table 5.2 Hermite polynomials. Deﬁnition

(−1)n (2t)m−2n

[m/2]

Hm (t) = m!

n=0

n! (m − 2n)!

= (−1)m et

2

dm −t2 e dtm

where [m/2] is the integer part of m/2. Examples H0 (t) = 1 H1 (t) = 2t H2 (t) = 4t2 − 2 H3 (t) = 8t3 − 12t

and z on the right-hand side. The three independent variables are now t1 , t2 , z, and we can separate them. The equations for g1 and g2 are identical, namely: d2 gi dgi − 2ti + 2mi gi = 0 dt2i dti

(i = 1, 2)

.

(5.112)

When the separation constant mi equals a non-negative integer, then the previous equation has a polynomial solution, known as the Hermite polynomial of degree mi . Deﬁnition and examples of Hermite polynomials are shown in Table 5.2. When mi is not a non-negative integer, then Eq. (5.112) still has power-series solutions, but they diverge at inﬁnity faster than exp(t2i ). Hence, when multiplied by the Gaussian function, they would yield ﬁelds which are not modulus-square integrable. So they must be disregarded, as non-physical solutions for our problem. The corresponding equation for ξ(z) reads: 1 . kw2 Inserting Eq. (5.96) into it, its general integral is found to be: zλ ξ = −(m1 + m2 ) arctan + constant . πw02 ξ = −2(m1 + m2 )

(5.113)

(5.114)

Therefore, Eq. (5.86) has the following discrete set of solutions: √ x √ y

w0 Hm 1 Hm 2 exp −r2 /w2 E = A 2 2 w w w 2 λz πr − (m1 + m2 + 1) arctan · exp −j kz + . λR πw02 (5.115) One of their most interesting properties is that they all share the same behavior as functions of z, for what concerns the beam diameter, 2w(z), and the wavefront radius of curvature, R(z). This reconﬁrms that eventually it

© 2006 by Taylor & Francis Group, LLC

HERMITE-GAUSS AND LAGUERRE-GAUSS MODES

165

has been sensible to devote all the previous section to one particular solution, which belongs of course to this set and corresponds to m1 = m2 = 0 in the equations of this section. The solutions in Eq. (5.115) are called Hermite-Gauss modes. An interesting property that can be proven for the Hermite polynomial of any degree m is that its m zeros are all real. Therefore, the orders m1 and m2 in Eq. (5.115) have a simple physical meaning: they are the numbers of zeroes of the function E, Eq. (5.84), along the x and y axes, respectively. They are called the transverse wavenumbers of the Hermite-Gauss modes, which in turn (for reasons similar to those outlined in Section 5.5) are usually labelled TEMm1 ,m2 modes. The Hermite-Gauss modes are linearly independent solutions of Eq. (5.86). This stems from the following orthogonality relation between Hermite polynomials (Abramowitz and Stegun, Eds., 1965): +∞ 2 e−t Hm (t) H (t) dt = 0 for = m . (5.116) −∞

At this stage, let us recall that a Gaussian beam is an approximate solution of the Helmholtz equation that must satisfy the condition in Eq. (5.85). Indeed, not all the solutions of the paraxial equation (5.86) that we have found now, satisfy the restriction in Eq. (5.85). As m1 and m2 grow, the z-derivative of the exponent in Eq. (5.115) grows indeﬁnitely as well. It is left to the reader as an exercise to show that, taking into account Eqs. (5.113) and (5.114) (and assuming, for simplicity, zλ πw02 ), the inequality in Eq. (5.85) is satisﬁed only by those modes for which: w02 . (5.117) λ2 Fortunately, in all cases of practical interest (like laser beams, open resonators — see Chapter 9 — and aperture antennas), most of the ﬁeld power is in the lowest-order Hermite-Gauss modes. Thus, an expansion of the total ﬁeld as a sum of Hermite-Gauss modes does not contradict the initial assumption of slow variations in φ vs. z. For conceptual reasons, to be discussed in the next section, as well as for practical ones, we are driven to ﬁnd solutions of Eq. (5.86) in cylindrical coordinates (r, ϕ, z) as well. Passing from a cartesian system to a cylindrical one, the Gaussian function, circularly symmetrical around the z axis, remains unchanged. The other amplitude factors can be found following essentially the same way as in the previous case. The ﬁnal results are: r

w0 h exp −r2 /w2 E = A w(z) w λz πr2 − (2p + + 1) arctan · exp −j kz ± ϕ + , λR πw02 m1 + m2 4π 2

(5.118)

© 2006 by Taylor & Francis Group, LLC

166

PLANE WAVE PACKETS AND BEAMS Table 5.3 Associated Laguerre polynomials. Deﬁnitions Lp (t) =

p (−1)m p + m=0

where Ls (t) =

p−m

m!

s (−1)m s m=0

m!

m

tm = (−1)

tm = e t

d Lp+ (t) dt

ds (e−t ts ) dts

is the ordinary Laguerre polynomial of degree s. Examples L0 (t) = 1 L1 (t) = + 1 − t 1 1 L2 (t) = ( + 1) ( + 2) − ( + 2) t + t2 . 2 2

where and p are two non-negative integers. We set: 2 √ r 2r 2 Lp h = , w w2

(5.119)

where Lp (t) is the so-called associated Laguerre polynomial of orders , p (see Table 5.3), which are polynomial solutions of the following equation: dL d2 L + pL = 0 . (5.120) + ( + 1 − t) dt2 dt From Eq. (5.118) we see that, when ϕ varies from 0 to 2π, then E, Eq. (5.84), repeats periodically times. For this reason is called the azimuthal wavenumber, and must be an integer, otherwise we would get a multivalued function, which is physically unacceptable. From general properties of Laguerre polynomials (see Abramowitz and Stegun, Eds., 1965), it follows that p equals the number of zeroes of the polynomial Lp , which are all real and positive. For this reason p is called the radial wavenumber. Linear independence among the Laguerre-Gauss modes (also labeled TEMp ) is guaranteed by the following orthogonality relation between generalized Laguerre polynomials of equal azimuthal order: ∞ e−t t Lp (t) Lq (t) dt = 0 for p = q . (5.121) t

0

Well known orthogonality relations apply among the functions e−jϕ over the interval (0, 2π), and so entail modes of diﬀerent azimuthal order to be always orthogonal. Figure 5.11 shows some very famous experimental results of intensities (proportional to |E|2 ) on a z = constant plane, for some Hermite-Gauss modes. Figures 5.12 and 5.13 show the behavior of a few Hermite and Laguerre polynomials. They also indicate that there is no one-to-one matching between these

© 2006 by Taylor & Francis Group, LLC

HERMITE-GAUSS AND LAGUERRE-GAUSS MODES

167

Figure 5.11 Very famous early experimental results: a laser oscillating in various Hermite-Gauss modes (from Boyd, G.D. and Gordon, J.P. (1961) Bell System Technical Journal, 40, 489–508).

© 2006 by Taylor & Francis Group, LLC

168

PLANE WAVE PACKETS AND BEAMS

Hn n3 H5 125

H3 27

H4 64

H2 8

x

Figure 5.12 Examples of Hermite polynomials.

Ln ( x )

L2

L3 x

L5

L4

Figure 5.13 Examples of Laguerre polynomials.

two families of polynomials. Their only relationship is that any Hermite-Gauss mode can be expanded as a series of Laguerre-Gauss modes, and vice-versa. 5.11

Reﬂection and refraction of Gaussian beams

At the beginning of Section 5.9, we introduced an important assumption, namely, that the medium in all the region of 3-D space where a Gaussian

© 2006 by Taylor & Francis Group, LLC

REFLECTION AND REFRACTION OF GAUSSIAN BEAMS

169

Figure 5.14 Reﬂection of a wavefront on a spherical mirror.

beam is deﬁned, be homogeneous. It occurs often that one wants to know how a Gaussian beam behaves at a sharp discontinuity in the medium. This is the case when one analyzes the reﬂection of the beam on a mirror, for example in an optical resonator (see Chapter 9), or the transmission of a beam through a lens. To be rigorous, one could expand the incident beam into plane waves, and apply the results of Chapter 4 to each individual wave. This procedure is quite cumbersome. Frequently, the problem can be simpliﬁed through approximations based on ray optics. We shall discuss them now. Let us start from a perfectly reﬂecting spherical mirror (Figure 5.14) of focal length F , meaning that its curvature radius equals 2F . We take the focus of the mirror as the origin of coordinates. Let us deal explicitly only with the TEM00 mode. The results can be easily extended to higher-order modes, as well as to a thin lens. Suppose (this is the source of approximation) that the distance Ri between the waist of the incident Gaussian beam and the point where the mirror intersects the z axis is so large, that we may approximate the ﬁeld impinging on the mirror with a non-uniform (i.e., amplitude not constant on a constantphase surface) spherical wave, of radius Ri . In this case, the elementary laws of ray optics say that the reﬂected ﬁeld is also a spherical wave, whose curvature radius is related to Ri and to F as: 1 1 1 . (5.122) =− + Rr Ri F The reﬂected spherical wave is also non-uniform. We assumed a perfect reﬂector, so the reﬂected amplitude distribution coincides with that of the incident wave, whence is still Gaussian and its width equals that of the incident wave: wr (z = −F ) = wi (z = −F ). If Rr , given by Eq. (5.122), is also large enough, so that the reﬂected wave is in turn a good approximation to a Gaussian beam, then the complex parameter qr (z) of the reﬂected beam deﬁned by Eq. (5.93): 1 1 1 = − 2j , (5.123) qr Rr kw2

© 2006 by Taylor & Francis Group, LLC

170

PLANE WAVE PACKETS AND BEAMS

Figure 5.15 Reﬂection on a spherical mirror: the ray picture.

is related at the mirror (z = −F ) to that of the incident beam, qi , by the following equation: 1 1 1 = + . (5.124) qr (−F ) qi (−F ) F The diﬀerence in sign between Eq. (5.122) and Eq. (5.124) is simply a consequence of Eq. (5.97), which gives R < 0 for z < 0. Once the value (5.124) of the complex parameter qr (−F ) is known, then q can be calculated at any plane z = constant by means of Eq. (5.91), which can be rewritten as follows: q(z) = q(−F ) + (z + F ) .

(5.125)

It is of interest to see whether we can write how qr at the generic plane z = z2 is related to qi at another generic plane, z = z1 , skipping Eqs. (5.124) and (5.125). Looking at Figure 5.15, we see that R = (x/x ), where x = dx/dz (for paraxial rays). While traveling in the region between mirrors (or lenses), the transverse coordinate x(z) of the generic ray varies linearly vs. z, in proportion to Eq. (5.125), while x = dx/dz remains constant. At each reﬂection, x(z) remains constant while x/x follows Eq. (5.124). Hence, during the whole path from z = z1 to z = z2 , the parameter q undergoes exactly the same transformations as the ratio x/x . Now, it is well known that a spherical mirror maps, in a “one-to-one and onto” way, a family of rays leaving a source point, into another family of rays, passing through an image point. This operation performed by the mirror is fully characterized, in mathematical terms, by a 2 × 2 matrix: A B xi xr = . (5.126) C D xr xi From what we said, it follows that Gaussian beam reﬂections obey the following relationship: qr (z = z2 ) =

© 2006 by Taylor & Francis Group, LLC

A qi (z = z1 ) + B C qi (z = z1 ) + D

,

(5.127)

ON THE COMPLETENESS OF A SERIES

171

where the coeﬃcients A, B, C and D can be determined by elementary ray optics. Eq. (5.127) is usually referred to as the ABCD law. Also a thin lens maps a family of straight lines into another family of straight lines. So, refraction of Gaussian beams at thin lenses can also be described by Eq. (5.127). When using Eq. (5.124), to distinguish between a lens and a mirror it is important to pay attention to the sign of the focal length F . Many contents of this section will be used again in Section 9.10, which deals with stability of open resonators.

5.12

On the completeness of a series

In Section 4.7 we discussed completeness of plane waves and we showed that, in general, a monochromatic ﬁeld, which carries a ﬁnite power through a plane, can be expressed as a plane wave integral. In other words, the basis for expanding such a ﬁeld is a continuous inﬁnity of elements. On the other hand, in Section 5.10 we mentioned on several occasions that we were dealing with a countable set of functions, and yet we stated that it was a complete set. This means that the most general ﬁeld we can have under these circumstances is not an integral, but a series. There is an apparent decrease in generality, so we must discuss, at least qualitatively, the conditions the ﬁeld must satisfy in order to allow such an expansion. Completeness problems, related to a partial diﬀerential equation, are well deﬁned only when boundary conditions are speciﬁed. In general, the weaker the boundary conditions, the wider the set, in order to be complete. So, the proof given in Section 4.7 referred to rather weak boundary conditions, as the only requirement was that the ﬁeld be modulus-square integrable. In such a case, a set could be complete only if it was a continuum. One may ask an interesting question: how stringent must the boundary conditions become, in order to reach completeness with a countable set of basis elements ? If we recall how a Fourier integral is related to a Fourier series, it is easy to conclude that a suﬃcient condition for the spatial spectrum of a ﬁeld to become discrete, is that the ﬁeld should satisfy periodic boundary conditions; or, equivalently, that the ﬁeld be non-zero only over a ﬁnite spatial interval, I. In the latter case, as well known from signal analysis, the ﬁeld can be repeated periodically, all the way to inﬁnity.§ Sometimes, the series convergence could be nonuniform at the extremes of the interval I. When this happens, it is not possible to diﬀerentiate the series term by term. In this framework, we will speak of completeness of a set only § Maybe the terminology used in this section biases the reader towards rectangular coordinates. However, the same concepts, with minor changes, apply in other types of reference frames. For example, in cylindrical coordinates, we shall consider expansions in series of trigonometric functions of the azimuthal coordinate, and Bessel functions of the radial coordinate.

© 2006 by Taylor & Francis Group, LLC

172 in the sense that:

PLANE WAVE PACKETS AND BEAMS 2 n lim cm fm dI = 0 , ψ − n→∞ I

(5.128)

m=1

where ψ can be any solution of the partial diﬀerential equation, {fm } is the complete set, the {cm }’s are the expansion coeﬃcients, and I, as we said, is the interval where the ﬁeld is deﬁned. All the subjects of Chapters 7 and 9 (waveguides with conducting walls and resonant cavities) will enable us to use periodic extension of ﬁelds deﬁned over ﬁnite intervals. It is easy, in those cases, to expect complete sets to be discrete; but Gaussian beams do not belong to the same class. We infer from this that periodicity is a suﬃcient condition for passing from a continuum to a discrete basis, but not a necessary one. There must be another explanation which applies to Gaussian beams. The answer can be identiﬁed as a boundary condition which, although set at inﬁnity, is much more restrictive than simply to require that the ﬁeld be square-integrable. To simplify the proof of this statement, let us use a onedimensional notation, which applies to both Laguerre-Gauss and HermiteGauss modes. Assume that F is a function that can be expanded as a series of Gaussian beams: ∞ ∞ 2 2 Pi (u) e−u = e−u Pi (u) , (5.129) F (u) = i=0

i=0

where Pi (u) is the i-th polynomial of the suitable type. From the last equation it follows that F (u) can be written as the product of a Gaussian function times a power series in u. F (u) is deﬁned in a region which extends to inﬁnity, and Eq. (5.129) must hold in all that region. Then, as u grows, F (u) has to 2 decrease fast enough, so that the Taylor series for F (u) eu converges all the way to inﬁnity. In other words, it must be: n u2 (5.130) lim F (u) e − Pi (u) = 0 , n→∞ i=1

for any u ∈ (0, +∞). As we said above, this constraint is stronger than squareintegrability. In practical problems where Gaussian beams turn out to be convenient (typically, laser beams and aperture antennas), Eq. (5.130) is easily satisﬁed for very simple physical reasons. For example, in an open resonator (Chapter 9), i.e., in a system consisting of two mirrors facing each other, the ﬁnite size of the mirrors in the transverse directions ensures that the ﬁeld is conﬁned, in practice, within a ﬁnite region. For u greater than a certain ﬁnite value, it is F (u) 0. Similarly, for a laser beam, or for the ﬁeld emitted by an aperture antenna, physical intuition leads to the same conclusion. With these limitations and margins of caution, one can speak of completeness of Gaussian beams in all cases where we need to use them.

© 2006 by Taylor & Francis Group, LLC

FURTHER COMMENTS ON RAYS AND BEAMS 5.13

173

Further comments on rays and beams

There are many links between geometrical optics and Gaussian beams. The ﬁrst one is in the terminology. Indeed, at the beginning of Section 5.9 we said that a tentative solution of the type φ(x, y, z) exp(−jkz), where φ is a slowly varying function of z, could be called a beam of rays. In contrast to a single ray, however, such a beam satisﬁes a basic requirement, the “minimum uncertainty relationship,” ΔxΔkx ≥ 2π, between its location in real space and its location in the spatial frequency space. As the reader may know from previous studies, and which will be reconﬁrmed in Chapter 13, this relationship is also fundamental for diﬀraction. Geometrical optics misses completely this point because it deals with each ray as an entity that is totally independent of what surrounds it. It comes as a consequence that energy ﬂows in ﬂux tubes, according to a ﬂuid-dynamics pattern. A critical analysis would show that WKBJ results are also aﬀected by the same limitations since they neglect diﬀraction. This shows that Gaussian beams are more accurate solutions of Maxwell’s equations than WKBJ solutions. Consequently, while geometrical optics and the WKBJ method fail near foci and caustics, where ray densities grow indeﬁnitely and ﬂux-tube cross-sections collapse, Gaussian beams do not have critical points, and are acceptable — although approximate — solutions in the whole 3-D space. It takes a more accurate analysis (e.g., Marcatili and Someda, 1987) to point out some inconsistencies in tightly focused Gaussian beams, in the neighborhood of the waist. As for geometrical optics and the WKBJ method, we wish to point out that suitable “correction procedures” have been developed, which allow one to overcome part of their drawbacks. The results are interesting, but too specialized to be discussed here in detail. Let us just say that there are rules that enable one to track rays beyond a focus. What happens near the focus can never be described in terms of rays. Similarly, there are connection formulas which link to each other two WKBJ solutions deﬁned on each side of the return point, or caustic. They go under the generic name “enlarged WKBJ method,” and yield, among other things, the result on phase shift that was mentioned in Section 5.8. However, these connection formulas are still rather poor approximations. In particular, they fail when there two or more caustics are close to each other. However, it would be naive to think that beams are superior to rays, as solutions of Maxwell’s equations, only because a larger number of powers of λ are kept in the approximation. If this were the case, then one might improve very signiﬁcantly on geometrical optics just by keeping more powers of λ. So, there must be other reasons. In fact, the true cause is that, in geometrical optics, the amplitude and phase functions are assumed to be real. Following this assumption, we divided the Helmholtz equation into two equations, where no term of order k0−1 appears. At this stage, we have diminished the initial level of accuracy. Removing the

© 2006 by Taylor & Francis Group, LLC

174

PLANE WAVE PACKETS AND BEAMS

assumption of real amplitude and phase functions, one ﬁnds Gaussian beams. In this sense, beams can be looked at as complex rays. Physically, what makes the diﬀerence are evanescent waves, which are left out of geometrical optics on one side, but on the other hand play a crucial, although partly hidden, role in Gaussian beams, near the beam waist, where the Poynting vector changes gradually its direction. In the literature, the reader can ﬁnd very important examples (e.g., Felsen and Marcuvitz, 1973) that show how other methods can be pushed to any level of accuracy in terms of powers of 1/k0 , at the price of a rapid growth in computation complexity. 5.14

Further applications and suggested reading

The logical thread of this chapter can be summarized as follows. A monochromatic plane wave is a simple, exact solution (in an unbounded homogeneous medium) of Maxwell’s equations, but it can not be taken literally, because it is unphysical under several viewpoints. In order to ﬁnd similar but physically acceptable solutions, one has to introduce approximations. Diﬀerent approximations are required under diﬀerent circumstances. It is rather unusual for a textbook to present several approximations side by side in one chapter. Our purpose was to help the reader to make comparison, stressing the relationships that exist often among diﬀerent approximations. It happens much more frequently to ﬁnd books which deal in depth with just one, or with a few, of these approximate solutions. Therefore, it is quite easy, for the interested reader, to proceed, on each individual subject, well beyond the introductory level of this chapter. Readers who have been exposed to fundamentals of telecommunications know well that the contents of the ﬁrst two sections (group velocity and dispersion) are extremely important in that context. What happens to signals which travel through non-ideal transmission media must be studied in depth by telecommunication engineers, regardless of the physical mechanism which is responsible for non-ideal propagation. Just to quote one among many classical reference books in this area, in Sunde (1969) the subject of dispersion-related signal distortion is covered in detail, for both analog and digital signals, and well linked to the other crucial issues in transmission and detection. In Section 5.2 we brieﬂy mentioned that pulses may be compressed in the time domain by making a proper choice of the dispersion sign in the medium where they propagate. This property has been exploited for decades at microwave frequencies, where shortening radar pulses is intended to increase spatial resolution (Skolnik, 1980; Baden Fuller, 1990). In the last decades, it has been transferred, very successfully, to the optical frequency range (Rudolph and Wilhelmi, 1989). Grating-assisted compression of laser pulses has been the way to generate the shortest man-made events that are known today, i.e., optical pulses in the range of a few femtoseconds (1 fs = 10−15 s). In optical telecommunications, ingenious techniques have been demonstrated for pulse

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

175

compression such as reversing signal spectrum midway between the transmitter and the receiver with a chirped grating (Hill et al., 1994; Kashyap et al., 1996) or reversing the sign of the dispersion using a DISpersion COmpensating (“DISCO”) optical ﬁber. Section 5.3 was mainly intended to set the stage for the following ones, but it could also be looked at as a hint for those who are interested in understanding more in depth the relation between scalar waves (encountered in many other branches of Physics) and vector waves. Some chapters in the classical treatise by Morse and Feshbach (1953) are quite relevant in this respect. They should be at least sampled by our readers. We also refer to Black and Ankiewicz (1985) for a discussion of analogies between optics and other areas of physics. The following three sections formed a package devoted to a subject — geometrical optics — which has many applications, in basic physics, in optical engineering, in communication engineering, etc. Chapter III of Born and Wolf (1980) is an in-depth view of basic theoretical arguments, those in favor of this approximation as well as those that show its limits. Chapters IV, V and VI in the same book describe exhaustively applications to imaging systems, and to the theory of aberrations (Buchdahl, 1968/1970), a subject which is also of potential interest to engineers involved in aperture antenna design. Readers who want to become more familiar with recent developments in imaging systems should invest some of their time studying graded-index (GRIN) components, and their computer-aided design, based on numerical solutions of the ray equation. They may start from the basic concepts outlined in Part I of Snyder and Love (1983), and then proceed to consult more specialized books, like Marchand (1978), for detailed, sophisticated theories, or Iga et al. (1984) for descriptions of GRIN engineering applications. We also said that geometrical optics has been for a long time, and still is, an important background item also for communication engineers. Radio wave propagation in the slowly-varying (and also random-varying) troposphere is instrumental in designing radio-relay and radar systems. Just to mention a few basic references on this subject, Chapters 33 and 34 of the handbook edited by Jordan (1986) focus on radio transmission problems. On the other hand, Sunde (1969) deals mainly with consequences on transmitted signals. Passing from the troposphere to the ionosphere, in addition to being slowly varying this medium is anisotropic as well. Should one wait for Chapter 6 ? In part, the answer is yes, but we can say now that even in the ionosphere ray tracing remains important for radio engineers. Highly recommended reading is Budden (1961). Ray tracing application to telecom problems enjoyed a second youth in the 1970’s, when optical ﬁbers became practical, and most of the interest was concentrated, for about one decade, on multimode ﬁbers. Snyder and Love (1983) is one of the most complete — and most popular — among the books which distilled that knowledge. Even at times, like the present ones, when single-mode ﬁbers are much more widely used than multimode ones for long-haul systems, it is still mandatory for a telecommunication engineer to

© 2006 by Taylor & Francis Group, LLC

176

PLANE WAVE PACKETS AND BEAMS

be exposed to the geometrical-optics treatment of graded-index multimode ﬁbers. This subject will be dealt with, to some extent, in Chapter 10. Not surprisingly, ionospheric propagation and graded-index optical guides are also among the main examples of application of the WKBJ method to e.m. wave propagation. References given in the previous paragraph, including our own Chapter 10, are still valid for the contents of Sections 5.7 and 5.8. We should also point out that the WKBJ method is applied, very widely, in other branches of pure and applied physics, ranging from underwater propagation (Frisk, 1994) to solid-state electronics (e.g., Kittel, 1996). Gaussian beams, the subject of Sections 5.9 through 5.12, were invented in the early ’60s, to model laser beams and resonators accurately enough to account for their basic diﬀraction limits. It is strongly recommended to come back to this subject after reading Chapter 13. Gaussian beams were successfully used, later on, also for designing aperture antennas (see Chapter 12), especially those with multiple reﬂectors, each of which is in the near ﬁeld of at least another one. Still, their main applications remain in the laser area. Classical textbooks on optical oscillators (see, e.g., Svelto, 1989) devote several chapters to Gaussian beams. All the students who include Quantum Electronics in their curricula must master this subject, and be able to solve problems on Gaussian beams. A special problem, obviously related to the previous ones, is propagation of Gaussian beams in lens-like media, i.e., in media where a slight radial gradient in refractive index may counteract, at least in part, beam divergence caused by diﬀraction. Gaussian beams in media with gain or loss are also required in some very practical problems. Straightforward introductions to these subjects can be found in Yariv (1991) and in Adams (1981). Other important practical problems that have been modeled successfully in terms of Gaussian beams are those related to source-to-ﬁber and ﬁber-to-ﬁber coupling in optical communication systems (Senior, 1992; Tosco, Ed., 1990). In particular, in the case of multimode ﬁbers the latter problem was an interesting example where the application of the whole family of Hermite-Gauss modes (Rizzoli and Someda, 1981) yields fundamental improvements with respect to ray-based approaches. Finally, the content of Section 5.13 should be taken just as a hint. It suggests that the links between Gaussian beams and geometrical optics are deep and fundamental. Clarifying them was a task that was successfully accomplished by theoreticians who investigated e.m. ﬁelds in very abstract terms, without direct interest in designing devices. A very suitable reference, in order to appreciate this, although very demanding reading, is the previously quoted book by Felsen and Marcuvitz (1973). Problems 5-1 Evaluate the group velocity, vg , and the dispersion, k , as a function of the angular frequency ω, in a medium where the phase constant for a plane wave (see Section 6.9) is deﬁned as β+ = (ω/c) 1 + ωM /(ω0 + ω).

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

177

5-2 Repeat the previous exercise with the following expression for the phase constant (see again Section 6.9): β− = (ω/c) 1 + ωM /(ω0 − ω). Note that in this case there is a frequency band where the phase constant is imaginary, i.e., the plane wave is evanescent. The group velocity is not deﬁned over that band. 5-3 Evaluate the group velocity, vy , and the dispersion, k , as a function of frequency, in a dielectric medium, near an atomic or molecular resonant absorption peak at ω = ω0 , where the permittivity exhibits a “reactive tail” of the Lorentzian absorption line, i.e.: = 0 + Δ

2Δω ω02 + 4Q2 Δω 2

,

where Δω = ω − ω0 , and Δ, ω0 and Q are real and positive, while μ = μ0 . Find the frequency intervals where dispersion is normal and those where it is anomalous. (Hint. d/dω = d/(dΔω).) 5-4 A wave is characterized by a phase constant β = (ω/c) [1 − (ωc /ω)2 ]1/2 (deﬁned for ω > ωc > 0, with c = constant). Find its group velocity vs. ω. Show that vf vg = c2 . Show that β → 0 for ω → ∞. Discuss the physical interpretation of these results. √ 5-5 A wave is characterized by a phase constant β = ω μ [1 − (ωc /ω)2 ]1/2 , but now, in contrast to the previous problem, depends on frequency as in Problem 5-3. Assume ω0 ωc . Show that the contributions to dispersion due to d/dω = 0, and those due to the factor [1 − (ωc /ω)2 ]1/2 , have equal signs in a frequency range close to ω0 , where both yield anomalous dispersion; opposite signs far from ω0 , where the “material dispersion” (Problem 5-3) is normal while the “waveguide dispersion” (Problem 5-4) remains anomalous. 5-6 A transmission medium is characterized by a dispersion coeﬃcient |k | = 0.05 ns2 /km, and is 20 km in length. What is the required initial half-width of a Gaussian pulse, for its ﬁnal width to be 3 ns ? Why are there two solutions to this problem ? Which one is more convenient in practice ? 5-7 A simple model for a “ﬂat” atmosphere is a dielectric medium (μ = μ0 ) where the permittivity varies in the vertical direction according to = 1 (1 − az), where 1 = 1.05 0 , a = 10−5 m−1 . Estimate the minimum frequency at which this can be looked at as a slowly varying medium. Discuss whether polarization (horizontal or vertical electric ﬁeld) has an eﬀect on the validity of the slowly-varying approximation. 5-8

Explain why Eq. (5.64) enables us to assume that the two periods, in

© 2006 by Taylor & Francis Group, LLC

178

PLANE WAVE PACKETS AND BEAMS

two diﬀerent meridional planes over which a meridional ray can be projected, are equal. 5-9 Show that, in a spherically symmetrical medium, i.e., for a refractive index n = n(r), the generalized Snell’s law reads: d n(r) r sin ϕ = 0 , du where ϕ is the angle between the tangent to the ray and the position vector r, which, due to the spherical symmetry, is parallel to ∇n. Show that any ray propagating through such a medium is meridional, i.e., belongs entirely to a plane through the origin. (Hint. Show ﬁrst that d(r × nˆ u)/du = 0 and make use of the ray equation.) 5-10 The Luneburg lens is a sphere where the refractive index varies as n(r) = [2 − (r/a)2 ]1/2 for r ≤ a, while n = 1 outside the sphere. Using the result of the previous problem, show that a beam of parallel rays, incident on this lens, is focused “on the back” of the sphere, i.e., on the point where the surface r = a intercepts the diameter parallel to the beam. 5-11 The so-called ﬁsh-eye lens, ﬁrst investigated by Maxwell, is a spherically symmetrical medium where the refractive index varies, for any r, as n(r) = n0 [1 + (r/a)2 ]−1 (n0 is a real constant). a) Show that all the rays emitted by a point source located on the r = a spherical surface pass through the symmetrical point with respect to the center of the lens. b) Show that all the rays emitted by a source located at a point where r > a pass through a point, located on the same diameter, on the opposite side with respect to the origin, and having a radial coordinate r = a2 /r < a (and vice-versa). 5-12 The numerical aperture of an optical system, exhibiting rotational symmetry around an “optic axis” z, is deﬁned as N A = sin ϑM , where ϑM is the maximum angle (measured in air) which a ray propagating through the system can form with the z axis. a) Show that if the refractive index varies along z, either in steps or gradually, then N A = n(z) sin ϑM (z). b) For a rotationally symmetrical medium where n decreases parabolically vs. r (see Eq. (5.62)), show that the numerical aperture (deﬁned with reference to meridional rays) is maximum on the axis and decreases gradually for growing r. 5-13 Consider a parabolic graded-index slab, i.e., a medium where the refractive index is independent of one of the transverse coordinates, y, and varies

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

179

with the other one according to n2 = n20 [1 − 2Δ(x/a)2 ] for |x| < a, and then remains constant for x > a. Rays may reach the borderline r = a, but are not supposed to penetrate further. a) Discuss the relationship between the numerical aperture (deﬁned as in the previous problem, with reference to rays in the (x, z) plane) and the relative index diﬀerence, Δ (which is always supposed to be much less than unity). b) For n0 = 1.5, N A = 0.1, and a maximum index gradient 104 m−1 , ﬁnd the corresponding slab width, 2a. 5-14 Consider again the parabolic graded-index slab of the previous problem. Suppose that the scalar wave equation, Eq. (5.67), can be approached by separation of the variables x and z, and furthermore, that ∂ 2 /∂z 2 = kz2 = constant. Write explicitly the restriction which plays the role of Eq. (5.69), in order to use the WKBJ method, for the equation which governs the x-dependence. Discuss in detail whether it is satisﬁed in the region where |x| a. 5-15 A lossless slab whose refractive index n(z) varies slowly along z (normal to the slab faces), and is independent of x and y, is sandwiched between two semi-inﬁnite lossless media of refractive indices n1 and n2 , at z = 0 and z = L, respectively. Show that the WKBJ method yields the following expression for the reﬂection coeﬃcient at z = 0: n2 n(0) − n1 n(L) . ρ= n2 n(0) + n1 n(L) 5-16 To use a graded-index layer as an antireﬂection coating, one sets n(L) = n2 . Find how the expression of the previous problem changes in this case. Compare this device to the quarter-wavelength antireﬂection coating discussed in Chapter 4, in terms of practical parameters, such as thickness, fabrication tolerances, bandwidth. 5-17 Taking into account Eqs. (5.113) and (5.114) — and assuming, for simplicity, |z|λ πw02 — show that only those TEM modes which satisfy the restriction (5.117) do not violate the inequality (5.85). 5-18 Draw a comparison between |f (z)|2 for the ﬁrst-order WKBJ solution, and |φ(z)|2 , where φ is a TEM0,0 Gaussian beam expressed by Eq. (5.87). Discuss the relationship between these results and the conservation law of geometrical optics, Eq. (5.48). 5-19 The emitting surface of a heterojunction semiconductor laser is, roughly, a rectangular strip. Its width in the junction plane is Δx = 100 μm, and its thickness in the orthogonal direction is Δy = 10 μm. The operating wavelength is 1.3 μm. Suppose that the emitted beam can be modeled as a TEM 0,0

© 2006 by Taylor & Francis Group, LLC

180

PLANE WAVE PACKETS AND BEAMS

astigmatic (i.e., non-circularly-symmetrical) Gaussian beam. Find the beam angular widths in the (x, z) and (y, z) planes, assuming that the beam waist is located on the output facet of the device. After reading Chapter 13, compare these results with those based on diﬀraction theory. 5-20 In cylindrical coordinates, a CO2 laser emits a TEM0,2 beam at a wavelength λ = 10 μm. Far from the waist, the beam full angular divergence is 2 × 10−2 rad. The electric ﬁeld, measured on the beam axis at z = 1 m (distance from the beam waist) equals 100 V/m. Find: a) the spot size on the plane of the beam waist; b) the maximum electric ﬁeld on the beam waist; c) the total power of the beam; d) the coordinates of the points, on the waist plane, where the ﬁeld vanishes. 5-21 A Gaussian beam, emitted by a Nd:YAG laser source (λ = 1.06 μm) whose waist half-width equals 2 mm and whose angular divergence equals 10−4 radians, has to be focused through a circular aperture, 30 mm in diameter, at a distance of 1.4 m from the waist plane. Design a suitable lens (i.e., ﬁnd its distance from the beam waist and its focal length). As a design criterion, note that about 99% of the power of a Gaussian beam is transmitted through an aperture whose radius is 3 times that of the beam at 1/e its maximum value. 5-22 A circular aperture of radius a is illuminated by a uniform ﬁeld, E = E0 = constant (while E = 0 for r > a). Expand this ﬁeld into a set of Laguerre-Gauss modes. (Hint. Use the orthogonality relation Eq. (5.121), and the following identity (Abramowitz and Stegun, 1965, p. 786): ∞

(x) .) e−t Lp (t) dt = e−x L(x) − Lp−1 p x

5-23 Repeat the previous problem, assuming now that the illuminated surface is annular (i.e., b < r < a, b > 0) and that the ﬁeld on it is E = E0 r−1 cos ϕ (again E = 0 out of this region). 5-24 A successful way to model an oﬀset joint between two multimode optical ﬁbers (Rizzoli and Someda, 1981) was to express the ﬁeld in each ﬁber as a series of Hermite-Gauss modes. Show that the so-called overlap integral (i.e., the coeﬃcient in the expansion) between the TEMp,q and the TEMm,n Hermite-Gauss modes, of equal widths w, for an oﬀset d between the centers of the two Gaussian functions along the x axis, and perfect alignment along

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING the y axis, is expressed by d2 n δqn exp − 2w 2 d d tμv = Hm xi − wi Hp xi + (π2p+m p!m!)1/2 i=1 (2)w (2)w

181

,

where δqn = Kronecker’s symbol, a = core radius, s = p + q, [p, q] ≡ v = [s(s + 1)/2] + p + 1 is the so-called principal mode number, and [m, n] ≡ μ. For wi , xi see Abramowitz and Stegun (1965, pp. 890–924).

© 2006 by Taylor & Francis Group, LLC

References Abramowitz, M. and Stegun, I.A. (Eds.) (1965) Handbook of Mathematical Functions. Dover, New York. Adams, M.J. (1981) An Introduction to Optical Waveguides. Wiley, Chichester. Baden Fuller, A.J. (1990) Microwaves: An Introduction to Microwave Theory and Techniques. 3rd ed., Pergamon Press, Oxford. Black, R.J. and Ankiewicz, A. (1985) “Fiber-optic analogies with mechanics”. American Journal of Physics, 53, 554–563. Born, M. and Wolf, E. (1980) Principles of Optics: Electromagnetic Theory of Propagation, Interference and Diﬀraction of Light. 6th (corrected) ed., reprinted 1986, Pergamon Press, Oxford. Buchdahl, H.A. (1968) Optical Aberration Coeﬃcients. Dover, New York. Buchdahl, H.A. (1970) An Introduction to Hamiltonian Optics. Cambridge University Press, UK. Budden, K.G. (1961) Radio Waves in the Ionosphere. Cambridge University Press, Cambridge, UK. Etten, W., van der and Plaats, J., van der (1991) Fundamentals of Optical Fibers Communications. Prentice-Hall, New York. Felsen, L.B. and Marcuvitz, N. (1973) Radiation and Scattering of Waves. PrenticeHall, Englewood Cliﬀs, NJ. Frisk, G.V. (1994) Ocean and Seabed Acoustics: A Theory of Wave Propagation. Prentice-Hall, Englewood Cliﬀs, NJ. Hill, K.O. et al. (1994) “Chirped in-ﬁber Bragg gratings for compensation of opticalﬁber dispersion”. Optics Letters, 19 (17), 1314–1316. Iga, K., Kokubun, Y. and Oikawa, M. (1984) Fundamentals of Microoptics: Distributed-Index, Microlens, and Stacked Planar Optics. Academic, Tokyo and Orlando, FL. Jordan, E.C., (Ed.) (1986) Reference Data for Engineers: Radio, Electronics, Computer and Communications. 7th ed., Howard W. Sams & Co., Indianapolis. Kashyap, R. et al. (1996) “1.3 m long super-step-chirped ﬁbre Bragg grating with a continuous delay of 13.5 ns and bandwidth 10 nm for broadband dispersion compensation”. Electronics Letters, 32 (19), 1807–1809. Kittel, C. (1996) Introduction to Solid State Physics. 7th ed., Wiley, New York. Marcatili, E.A.J. and Someda, C.G. (1987) “Gaussian beam are fundamentally different from free-space modes”. IEEE J. of Quantum Electronics, 23 (2), 164–167. Marchand, E.W. (1978) Gradient Index Optics. Academic, New York. Morse, P.M. and Feshbach, H. (1953) Methods of Theoretical Physics. Vols. I and II, McGraw-Hill, New York. Rizzoli, V. and Someda, C.G. (1981) “Mutual inﬂuence among imperfect joints in multimode-ﬁbre link”. Electronics Letters, 24, 906–907. Rudolph, W. and Wilhelmi, B. (1989) Light Pulse Compression. Harwood, New 183 © 2006 by Taylor & Francis Group, LLC

184

REFERENCES

York. Saleh, B.E.A. and Teich, M.C. (1991) Fundamentals of Photonics. Wiley, New York. Senior, J.M. (1992) Optical Fiber Communications: Principles and Practice. 2nd ed., Prentice-Hall, New York. Skolnik, M.I. (1980) Introduction to Radar Systems. 2nd ed., McGraw-Hill, New York. Snyder, A.W. and Love, J.D. (1983) Optical Waveguide Theory. Chapman & Hall, London. Sunde, E.D. (1969) Communication System Engineering Theory. Wiley, New York. Svelto, O. (1989) Principles of Lasers. 3rd ed., translated from Italian and edited by D.C. Hanna, Plenum, New York. Tosco, F. (Ed.) (1990) Fiber Optic Communications Handbook. TAB Professional and Reference Books, Blue Ridge Summit, PA. Yariv, A. (1991) Optical Electronics. 4th ed., Saunders College Publishing, Philadelphia.

© 2006 by Taylor & Francis Group, LLC

CHAPTER 6

Plane waves in anisotropic media 6.1

General properties of anisotropic media

We brieﬂy mentioned in Section 1.3 that a medium is said to be anisotropic, from an electromagnetic point of view, when it has the following property: at any point, and considering generic directions for the vectors E, H, at least one of the following statements is true: • the total current density vector Jt = J + jωD is not parallel to the electric ﬁeld vector E; • the magnetic induction vector B is not parallel to the magnetic ﬁeld intensity vector H. Tackling anisotropic media analytically is not too complicated when the above mentioned vectors are linked to one another by linear relationships. As is well known, it is absolutely general to express a linear relationship between two vectors in terms of a dyadic operator, i.e., in terms of a second-rank tensor accompanied by the rule of taking its inner product with the vector on its right. If B and Jt are linked by a linear relationship to both E and H, i.e., if we can write: Jt

= g·E+λ·H ,

B

= ν ·E+μ·H ,

(6.1)

then one refers to this as a linear bianisotropic medium. The special case λ = ν = 0 is referred to as a linear anisotropic medium. In the latter case, it is a generalized habit to use the notation of Eqs. (1.23), (1.24) and (1.25), which we rewrite here for future convenience: = σ·E ,

(6.2)

D = ·E ,

(6.3)

J

B

= μ·H .

(6.4)

In a linear isotropic medium, one may always think that Eqs. (6.2), (6.3), and (6.4) still hold, with σ = σ¯i, = ¯i, μ = μ¯i, where ¯i is the identity dyadic. In the following, whenever we speak about anisotropic media, we will implicitly exclude this trivial case. An anisotropic conductor will be, in our terminology, a medium with σ = σ¯i, but with = ¯i and μ = μ¯i. The 185 © 2006 by Taylor & Francis Group, LLC

186

PLANE WAVES IN ANISOTROPIC MEDIA

deﬁnitions of anisotropic dielectric and anisotropic magnetic media are now obvious. Before studying how plane waves propagate in anisotropic media, let us see how some fundamental results, which we obtained in the previous chapters for isotropic media, can either be extended, or modiﬁed. A) Lossless media. Like in the isotropic case, in Maxwell’s equations: ∇ × H = jω · E + σ · E + J0 ∇ × E = −jωμ · H

,

,

(6.5) (6.6)

the right-hand side of Eq. (6.5) can be written in a more compact form if we deﬁne the following dyadic: 1 c = + σ . (6.7) jω By similarity with the isotropic case, we are inclined to refer to this quantity as the complex permittivity dyadic. However, there is a fundamental diﬀerence, with respect to the isotropic case. Here, in fact, for the medium to be lossless it is not required that all the elements of the dyadic c should be real. This conclusion can be drawn directly from Poynting’s theorem. If we introduce the dyadic in Eq. (6.7) into Maxwell’s equations, then the proof of the theorem (see Section 3.2) yields an expression which no longer contains an explicit dissipation term (i.e., a term proportional to the conductivity σ), but instead we ﬁnd that the other volume integral becomes now: E · ∗c · E∗ H∗ · μ · H −2jω − dV . (6.8) 4 4 V For this quantity to be imaginary (i.e., to represent a reactive power), it is not necessary that the dyadics c and μ be real. In fact, the real part of Eq. (6.8) is zero, regardless of the particular integration volume V (i.e., the medium is lossless), if and only if: E · ∗c · E∗ − E∗ · c · E + H · μ∗ · H∗ − H∗ · μ · H = 0 ,

(6.9)

where E, H are arbitrary. This occurs if and only if both dyadics c and μ are Hermitian, i.e.: c = †c , μ = μ† , (6.10) where the symbol † means transposed and conjugate. A medium which can be lossless even with σ = 0, i.e., a “lossless conductor,” sounds puzzling. In fact, we will see later in this chapter that, under some circumstances, charged particles can move, under the inﬂuence of an e.m. wave, in such a way that their convection current is in quadrature with the electric ﬁeld of the wave. The ﬁeld-current product is imaginary, giving rise to a purely reactive power, although the current is not a displacement one. It is evident, however, that this conduction without dissipation results from a model of particle motion where any friction is completely neglected, hence necessarily an approximate model.

© 2006 by Taylor & Francis Group, LLC

GENERAL PROPERTIES OF ANISOTROPIC MEDIA

187

After learning that complex elements in c do not necessarily imply that the medium is lossy, we could drop the distinction between the symbols and c . However, using two diﬀerent symbols may help to prevent the mistake of thinking that E∗ · c · E/4 is always the (time-averaged) electric energy density. Indeed, it can be shown that, in some lossless media, and for some values of the frequency, this quantity can even become negative, which is clearly unacceptable for an energy density. This apparent nonsense can be explained, taking into account that, in such a medium, there is also a storage of distributed kinetic energy, because of the motion of the charged particles. Without getting now into details that are diﬃcult to grasp at this stage, let us repeat that it is because of these problems that, in the following, we will preserve the distinction between , which refers to a zero-conductivity anisotropic dielectric medium, and c , for an anisotropic medium with a ﬁnite conductivity. Another surprise for the reader may be the fact that, similarly, some media can be characterized by a complex permeability dyadic μ. This will also be clariﬁed by means of an example, later on in this chapter. It is well known that, in a complex space, any Hermitian operator is described by a diagonal matrix in a suitable orthogonal reference frame (Smirnov, 1961). Therefore, we can state that for any lossless anisotropic dielectric or magnetic medium, there is a basis — in general, a complex one — where c or μ is represented by a diagonal matrix. However, in a medium characterized by c and μ that are both Hermitian, we cannot state for sure that there exists an orthogonal basis where both dyadics are represented by diagonal matrices. B) Reciprocity. Let us repeat the proof of the reciprocity theorem (Section 3.4), for a linear anisotropic medium, using a real rectangular reference frame, and writing all the vectors by components. Then, we see that the theorem holds if and only if c and μ are both represented by symmetric matrices: ˜ c , (6.11) (c ) = # , (μ) = μ where the symbol ˜ means transposed. As known from linear algebra (Smirnov, 1961), symmetry of a matrix is a property which is invariant under an orthogonal transformation, i.e., a change of basis in real 3-D space. In general, it is not invariant under a unitary transformation, i.e., the most general change of basis in a complex space. Incidentally, this is in contrast with the Hermitian property, which is preserved by unitary transformations. This implies that, while we could speak of lossless media without any need to specify what kind of reference frame we were taking, on the contrary, when the subject is reciprocity in an anisotropic medium, statements are unambiguous only as long as they are restricted to linearly polarized ﬁelds. In engineering contexts, this point is often taken for granted. However, let us stress that it may entail important consequences, that will be outlined in detail in Section 6.6. For the time being, in order to grasp the ﬁrst meaning of this discussion, the reader may verify that the reaction, deﬁned in Section 3.4 and intended to be a mea-

© 2006 by Taylor & Francis Group, LLC

188

PLANE WAVES IN ANISOTROPIC MEDIA

sure of reciprocity, is not an invariant with respect to an arbitrary change of basis in the complex ﬁeld. Indeed, even a very trivial change — multiply one of the basis vectors by a phase factor — modiﬁes the reaction. Mathematically, this surprising behavior is related to the fact that Steinmetz’s representation of time-harmonic quantities is not an isomorphism between the space of real time-harmonic vectors and the 3-D complex vector space, and does not preserve the inner product. 6.2

Wave equations and potentials in anisotropic media

Dyadic operators do not always commute with the curl operator, neither with the divergence one. Accordingly, Maxwell’s equations in anisotropic media cannot be handled in the same way as in the ﬁrst chapter. In general, it is not possible to obtain from Eqs. (6.5) and (6.6) a simple equation, like the Helmholtz equation, containing either E or H, but not both. For a dielectric anisotropic medium (scalar μ), it is possible to obtain an equation where only E appears. To do that, take the curl of Eq. (6.6), replace ∇ × H with Eq. (6.5), and make use of the deﬁnition of the Laplacian of a vector. This yields: ∇2 E − ∇(∇ · E) + ω 2 μ c · E = jωμ J0

.

(6.12)

On the other hand, in such a medium it is impossible to derive an equation where the only unknown is H, at least as long as further information on the dyadic c is not provided. Eq. (6.12) in general is not the same as the Helmholtz equation, because the identity (C.5) ∇ · ∇ × a = 0, applied to Eq. (6.5), gives: ∇ · (c · E) = −

1 ∇ · J0 jω

.

(6.13)

If there are no sources, or if the sources have zero divergence, this yields: ∇ · (c · E) = 0 .

(6.14)

In general the last equation does not entail ∇ · E = 0. Therefore, the second term on the left-hand side in Eq. (6.12) does not disappear. In the following sections, we will see that only in a few special cases Eq. (6.12) can undergo signiﬁcant simpliﬁcation. Duality tells us that in an anisotropic magnetic medium the governing equation is: (6.15) ∇2 H − ∇(∇ · H) + ω 2 c μ · H = 0 , where the second term cannot be eliminated, since Eq. (6.6) and the identity ∇ · ∇ × a = 0 entail: ∇ · (μ · H) = 0 , (6.16) which does not necessarily lead to ∇ · H = 0. In this type of medium, we are unable to write a simple and general equation for the vector E. These problems in writing a diﬀerential equation, either for E or for H, are

© 2006 by Taylor & Francis Group, LLC

BIREFRINGENT MEDIA

189

accompanied by parallel problems if we try to develop a theory of e.m. potentials, similar to the one we built in Chapter 1 for isotropic media. Indeed, let us omit the very complicated case where c and μ are both dyadics (fortunately, not a case of primary importance), and take the case of a homogeneous anisotropic dielectric. We try to repeat the initial steps of Section 1.8. So, we introduce a magnetic vector potential A, whose scope of deﬁnition can be very broad. The ﬁeld vectors can be derived from it, as follows: 1 1 E= c −1 · (∇ × ∇ × A − μ J0 ) , (6.17) H= ∇×A , μ jωμ where c −1 is the inverse of c . The vector A is easily shown to obey the equation: (6.18) ∇ × c −1 · ∇ × ∇ × A − ω 2 μA − μ c −1 · J0 = 0 . In a simply connected domain, we may try to solve Eq. (6.18) introducing an arbitrary scalar function ϕ, as in isotropic media. This yields an equation which diﬀers formally from Eq. (1.65) only because the dyadic c replaces the scalar c : ∇ × ∇ × A − ω 2 μ c · A − μJ0 = −jωμ c · ∇ϕ .

(6.19)

In reality, we run into problems as soon as we try to simplify Eq. (6.19) through the choice of ϕ. Indeed, Eq. (6.19) and the usual identity ∇·∇×a = 0 give us an equation which contains ∇ · (c · A), not ∇ · A. As a consequence, even in the simplest case where ∇ · J0 = 0 (let us recall that we showed that under this assumption, in an isotropic medium, ϕ = 0 is a Lorentz gauge), now we are unable to get the Helmholtz equation from Eq. (6.19), because the term ∇(∇ · A) in general is not zero. The only result that we can reach is that, in those special cases where Eq. (6.12) becomes the Helmholtz equation, then the same occurs to Eq. (6.19), provided the ϕ = 0 gauge is adopted. The duality theorem of Section 3.7 indicates how to cover the case of a magnetic anisotropic medium, in terms of an electric vector potential F of the same type as the one we saw in Section 1.9. The details are left to the reader as an exercise. In conclusion, we saw in this section that anisotropic media are impossible to deal with at the same level of generality as we did in previous chapters. Consequently, we will be often forced to introduce some “ansatz” on the ﬁelds. The most widely used assumption is to deal only with plane waves of a given polarization. Another consequence is that it appears convenient to classify anisotropic media into classes, which diﬀer from each other in the manner of simplifying the equations that the ﬁelds must obey. 6.3

Birefringent media

In this section we will deal with one of the most important classes of anisotropic media. We will assume that they are lossless dielectrics. Including loss would make the mathematics much heavier, without any major gain in understanding

© 2006 by Taylor & Francis Group, LLC

190

PLANE WAVES IN ANISOTROPIC MEDIA

the physics. The dual case of magnetic media with identical formal properties is left to the reader as an exercise. A lossless anisotropic dielectric medium is referred to as birefringent if its dyadic is real, i.e., if all the elements of the matrix () which represent the dyadic in a real reference frame, are real. As we said, we maintain a distinction between and c ; otherwise, some consequences that we are going to draw from Eq. (6.22) would no longer be valid. This deﬁnition, combined with the assumption that the medium is lossless, Eq. (6.10), implies that, in any real rectangular coordinate system (x1 , x2 , x3 ), the matrix () is symmetric: ij = ji

(i, j = 1, 2, 3)

.

(6.20)

Any (3 × 3) real symmetric matrix has three, mutually orthogonal, real eigenvectors (Smirnov, 1961). Therefore, in any birefringent dielectric, there is a particular rectangular reference frame — from now on, referred to as the (x, y, z) system — where its matrix () is diagonal: ⎤ ⎡ ⎡ ⎤⎡ ⎤ x 0 0 Dx Ex ⎣ Dy ⎦ = 0 ⎣ 0 y 0 ⎦ ⎣ Ey ⎦ . (6.21) Dz 0 0 z Ez If the medium is homogeneous, this reference frame is constant in space throughout the region occupied by the medium. In this case, it is called the system of principal axes of the dielectric. The matrix elements x , y , z in Eq. (6.21) are referred to as the principal relative permittivities. The steps one has to go through, to pass from permittivity in a generic frame to Eq. (6.21), lend themselves to a geometrical interpretation, which might help us later on in understanding e.m. wave propagation in such a medium. Let us start from a generic frame (x1 , x2 , x3 ), where is described by the matrix (ij ), and deﬁne the following quadratic form: 3 i,j=1

ij xi xj =

3

ij xi x∗j

,

(6.22)

i,j=1

where x∗j = xj , since the coordinates are real. This form is easily seen to be deﬁnite positive. In fact, if we replace xi with the component Ei of the electric ﬁeld, Eq. (6.22) becomes four times the time-averaged electric energy density, which must be positive by deﬁnition for any nonvanishing ﬁeld. Writing that a deﬁnite positive form like Eq. (6.22) is equal to a constant, one gets an equation which, in analytical geometry, is the equation of an ellipsoid. Passing from a generic reference frame to the coordinate system where Eq. (6.21) holds, the ellipsoid equation reads: x x2 + y y 2 + z z 2 = constant .

(6.23)

Using the language of geometry, we call this the ellipsoid equation in the frame of its principal axes. This explains the names that were introduced earlier. Eq. (6.21) reconﬁrms that, in general, in an anisotropic dielectric medium

© 2006 by Taylor & Francis Group, LLC

BIREFRINGENT MEDIA

191

the vectors D and E are not parallel. Further information on how these two vectors lay can be obtained if we make the assumption (to be conﬁrmed later on), that a linearly polarized plane wave is allowed to propagate in this medium. Assume then, on the basis of what we saw in Chapter 4, that all the ﬁeld vectors vary in space according to exp(−j k · r), where k is a constant real vector. Then, the curl operator may be replaced by (−j k)×. Maxwell’s Eqs. (6.5) and (6.6), in a region without sources, read: −j k × H = jω D ,

(6.24)

−j k × E

(6.25)

= −jωμ H .

From these we see that H is perpendicular to D, to E, and to k. Consequently, these three vectors must all lay on one plane. Eq. (6.24) also shows that D is perpendicular to k. On the other hand, Eqs. (6.24) and (6.25) do not yield any information on the direction of E, except for the previous point that it belongs to the plane deﬁned by k and D. This entails that, in general, the vector E has a nonvanishing component along the direction of propagation. Consistently with the previous assumption of linear polarization, let us assume H = H0 exp(−jk · r), with a real H0 . Then, the vectors E0 = E/ exp(−jk · r) and D0 = D/ exp(−jk · r) are also real, and can be depicted as in Figure 6.1. The Poynting vector P = E × H∗ /2 is also real, and furthermore, being perpendicular to H0 , it belongs to the same plane as D0 , E0 and k. By construction, it is also orthogonal to E0 , and therefore in general it is not parallel to the phase vector k. The angle that the Poynting vector makes with the phase vector equals the angle between E0 and D0 . Accordingly, for a plane wave in a birefringent medium power ﬂows, generally speaking, in a direction (called the ray direction) which is not the same as the normal to the constant-phase planes (the wave normal direction). In Section 4.11, we saw that superpositions of evanescent waves in isotropic media do convey power in a direction perpendicular to their phase vector, but let us stress, to avoid any ambiguity, that the phenomenon that we are discussing here is completely diﬀerent, from a physical point of view. Let us aim now at ﬁnding a link between the vectors D and E, for any direction of the phase vector. Let us deﬁne a refractive index n as a function of the direction of k, as follows: k = n β0 kˆ

,

(6.26)

where β0 = ω(μ0 0 )1/2 , while kˆ is the unit vector in the direction of k. Such a deﬁnition stems from one of the physical meanings of the refractive index in an isotropic medium, namely that of ratio between phase velocity in free space, and phase velocity in the medium. On the other hand, it does not entail anything about the relationship between D and E. Insert then Eq. (6.26) into Eqs. (6.24) and (6.25), assume for simplicity μ = μ0 , and eliminate H. The result is: (6.27) D = n2 0 kˆ × E × kˆ = n2 0 Et ,

© 2006 by Taylor & Francis Group, LLC

192

PLANE WAVES IN ANISOTROPIC MEDIA

Figure 6.1 A plane wave in an anisotropic medium: relationships between its electromagnetic vectors.

where Et is the transverse component of E, with respect to the direction of propagation. To determine n as a function of the parameters of the medium, it is convenient to exploit the inverse permittivity (also called impermittivity) dyadic b, deﬁned as: b = 0 −1

,

(6.28)

0 E = b · D .

(6.29)

such that: ˆ Let us project Eq. (6.29) on a constant-phase plane (perpendicular to k), and take into account that D belongs to this plane. Thus, we obtain: 0 Et = bt · D ,

(6.30)

ˆ where the dyadic bt is the projection of b onto a plane orthogonal to k, therefore it is a dyadic in a 2-D subspace and can be described by a 2 × 2 matrix (Smirnov, 1961). The system consisting of Eqs. (6.30) and (6.27) yields an eigenvalue equation: 1 bt · D = 2 D . (6.31) n Therefore, n2 can be identiﬁed as the inverse of one of the eigenvalues of the operator bt . This enables us, ﬁrst of all, to count how many values the refractive index can take for a given direction of propagation. Let us take a rectangular coordinate frame with an axis parallel to the direction of propˆ Then, thanks to our initial assumptions on , the 2 × 2 agation (i.e., to k). matrix (bij ) representing bt is real and symmetric. Consequently, it has two real eigenvalues, in general diﬀerent from each other. In conclusion, for waves traveling along a generic direction, the refractive index may take two diﬀerent values. It is self-explanatory, now, why these media are called birefringent. Eq. (6.22), as already said, is a positive deﬁnite quadratic form (in lossless

© 2006 by Taylor & Francis Group, LLC

FRESNEL’S EQUATION OF WAVE NORMALS

193

media). Hence, both eigenvalues are positive.∗ Each eigenvalue gives rise to two opposite values of n, which correspond, obviously, to two waves traveling ˆ in opposite directions, ±k. Under the assumptions that were made at the beginning of this section, the operator bt has two real orthogonal eigenvectors. Each of them belongs to one of the above mentioned eigenvalues. This means that there are two possible, mutually orthogonal linear states of polarization for the D vector of plane waves traveling in a birefringent medium. Any other state of polarization is a ﬁeld whose dependence on spatial coordinates cannot be written in the simple form exp(−jk · r). So, to see how a ﬁeld, known at a given point, in an unspeciﬁed state of polarization, evolves in space, ﬁrst we must decompose it as the sum of the two linear polarizations corresponding to the eigenstates that we have just encountered. These are then two waves which travel independently of each other. The total ﬁeld at any distance is to be written as the sum of those two linearly polarized waves, after they propagate over that distance. As these two waves have two diﬀerent phase constants, the net result is that the state of polarization of the total ﬁeld, in general, is a function of distance, along the direction of propagation. The reader should be able to explain, as an exercise, why this function is periodic in space. 6.4

Fresnel’s equation of wave normals

In the previous section we studied the fundamentals of wave propagation in birefringent media, but we did not arrive at operative rules. We will show now how to calculate the main features of plane waves which travel in these media. Solving the eigenvalue equation (6.31) requires two cumbersome operations, to invert the matrix which represents , and to change the coordinate system. There is a faster way to compute the refractive indices, for a given direction of propagation, starting from Eqs. (6.3) and (6.27), which (i being the identity dyadic) entail: ˆ kˆ . (6.32) ( − n2 0 i) · E = −n2 0 (E · k) In the principal-axes reference frame (where is diagonal), it becomes very simple to write Eq. (6.32) by components. Let kˆ ≡ (vx , vy , vz ). We get: (n2 − x ) Ex

= n2 vx (kˆ · E) ,

(n2 − y ) Ey

= n2 vy (kˆ · E)

(n2 − z ) Ez

= n2 vz (kˆ · E) .

, (6.33)

∗ Many results of this chapter can be extended to the case where c = + (1/jω) σ is real. In that case, however, as we saw in Section 6.2, the quadratic form E ∗ · c · E does not have a physical meaning which assures that n2 is always positive. When a negative value is found for n2 , thus an imaginary value is found for ±n, the corresponding plane wave becomes extremely similar, physically, to the evanescent waves of Chapter 4. Such a formalism can be very useful in modeling ionospheric reﬂections (see, e.g., Budden, 1961).

© 2006 by Taylor & Francis Group, LLC

194

PLANE WAVES IN ANISOTROPIC MEDIA

Multiply the ﬁrst of these equations by vx , the second by vy , the third by vz , sum the three equations, and divide both sides by kˆ · E (leaving out isotropic media, where kˆ · E vanishes). We get the following equation: vy2 vx2 v2 1 + + 2 z = 2 2 2 n − x n − y n − z n

,

(6.34)

currently referred to as Fresnel’s equation of the wave normals. It allows an easy computation of n2 , once the principal permittivities and the propagation direction (vx , vy , vz ) are given. Eq. (6.34) is biquadratic in n, because of the identity vx2 + vy2 + vz2 = 1. Hence, it always gives two solutions for n2 , in agreement with what we saw in Section 6.3. When written in a rational form, it holds also in isotropic media (x = y = z = r ), and in this case it gives, √ as it must give, solutions n = ± r that are independent of the direction of propagation. The directions of the two linear polarizations allowed for D are now simple to determine by means of geometry. In the principal-axes reference frame (x, y, z), consider the ellipsoid, centered in the origin, deﬁned by the following equation (note that it is not the same as Eq. (6.23)): x2 y2 z2 + + =1 . x y z

(6.35)

This surface is called index ellipsoid, or the optical indicatrix. We will show now that, for a wave traveling parallel to a unit vector kˆ = (vx , vy , vz ), the two directions that are allowed for linearly polarized D vectors are the major and minor axes of the ellipse which is the line where the ellipsoid Eq. (6.35) intersects the plane orthogonal to kˆ passing through the origin whose equation reads: (6.36) vx x + v y y + v z z = 0 . We will also prove that, with a suitable normalization, the lengths of these two axes are equal to the indices n given by Eq. (6.34). To do this, let us ﬁrst point out that the axes we are looking for are the minimum and maximum diameter lengths for the ellipse. Therefore, they may be found as conditional extrema of the distance from the origin, r, or of a convenient power of r. The simplest choice is to deal with: r 2 = x2 + y 2 + z 2

.

(6.37)

2

The constraints we must impose on r are Eqs. (6.35) and (6.36), meaning that the ellipse axes terminate at points that belong to the ellipsoid and to the ˆ The problem can be tackled with the standard method plane orthogonal to k. of Lagrange multipliers. Namely, we write a linear combination of Eqs. (6.37), (6.36) and (6.35), with auxiliary parameters 2λ1 , λ2 , i.e.: 2 x y2 z2 + + F (x, y, z) = x2 +y 2 +z 2 +2λ1 (vx x+vy y +vz z)+λ2 ; (6.38) x y z next, we take the partial derivatives of Eq. (6.38) with respect to x, y, z, and

© 2006 by Taylor & Francis Group, LLC

FRESNEL’S EQUATION OF WAVE NORMALS

195

Figure 6.2 Geometrical construction of the refractive indices and the allowed states of polarization, using the optical indicatrix.

set them equal to zero. We get the following system of equations: m =0 (m = x, y, z) . m + λ 1 vm + λ 2 m

(6.39)

The parameters λ1 and λ2 can be eliminated combining Eqs. (6.35), (6.36) and (6.39). So, we get three equations, the ﬁrst of which reads: 2 vx x vy y vz z r 2 − 1 x = r vx + + . (6.40) x x y z The other two can be obtained simply by permuting x with y and with z. These equations deﬁne, by construction, the extrema of the ellipse axes. Now, we observe that the ﬁrst of Eqs. (6.33) becomes Eq. (6.40) if we set n2 = r2 , x = Dx = x Ex . The same occurs with the other two components and the other two coordinates. This proves that n is proportional to the axis length, and, at the same time, it proves that the directions of the axes coincide with those allowed for D (see Figure 6.2). In other words, this completes our proof. We run into a case of special interest when two out of the three principal permittivities are identical, i.e., for x = y = z . The ellipsoid described by Eq. (6.35) has then a rotational symmetry around the z axis. In this case, for ˆ one of the axes of the ellipse where the ellipsoid intersects any direction of k, the plane normal to kˆ is, at the same time, a diameter of the circle where the ellipsoid intersects the plane z = 0 (which, in turn, is either a minimum or a maximum in length among all the intersection ellipses, depending on whether x is smaller or larger than z ). Consequently, in these media one of the two values for n (usually referred to as the ordinary index, n0 ) is independent of the direction of propagation, while the other one (referred to as the extraordinary ˆ The names given to the indices index, ne ) does depend on the direction of k. reﬂect the habit of calling ordinary and extraordinary waves, respectively,

© 2006 by Taylor & Francis Group, LLC

196

PLANE WAVES IN ANISOTROPIC MEDIA

the plane waves associated with n0 and with ne . The term birefringence, in a given direction, is in current use to denote the absolute value of the diﬀerence, |n0 − ne |, for a pair of plane waves traveling in that direction. Note, however, that sometimes it is used for the relative diﬀerence, |n0 − ne |/n0 , and on other occasions it means the absolute diﬀerence between the two phase constants, |n0 − ne |β0 . The quantity LB = 2π/|n0 − ne |β0 is called the beat length. It is simple to show, as an exercise, that, if an ordinary wave and an extraordinary one travel in the same direction, then the state of polarization of the total ﬁeld, resulting from the overlap of the two waves, evolves periodically as a funtion of distance, and LB is its spatial period. The values of n are both independent of kˆ only in a medium where x = y = z , i.e., in an isotropic medium. In this case, the ellipsoid given by Eq. (6.35) becomes a sphere, and all directions orthogonal to kˆ are legitimate for D. If, on the contrary, z = x = y , then there is only one direction for kˆ — the z-direction — for which we ﬁnd ne = n0 . For waves traveling along this axis, the directions allowed for D are all those in the plane orthogonal to z. A medium of this kind is called a uniaxial medium. Trigonal, tetragonal and hexagonal crystals fall within this category (Nye, 1985). When the three principal permittivities diﬀer from each other, it is simple to prove that there are two (and only two) planes through the origin that intersect the ellipsoid in circles. Hence, there are two directions of propagation (normal to those planes), for each of which the two values of n2 are equal, and any direction normal to propagation is allowed for D. These media are referred to as biaxial. Refraction at the separation between an isotropic and an anisotropic medium can be now studied along the guidelines of Section 4.8. For the sake of brevity, let us omit evanescent waves, which can be dealt with by the reader as an exercise. Then, in general (i.e., except for the special cases where the refracted waves travel along one of the optic axes) there are two transmitted waves, traveling in two diﬀerent directions, and orthogonally polarized. This phenomenon is referred to as double refraction. Each refracted wave obeys Snell’s law, Eq. (4.84), with the value of n which is relevant to that wave (either the ordinary, or the extraordinary index). Reﬂection and transmission coeﬃcients can be evaluated by decomposing any incident wave into the two polarizations that correspond to the ordinary and extraordinary waves in the birefringent medium, respectively, and then applying Fresnel’s formulas of Section 4.9 to each of them. Another refraction phenomenon, diﬀerent from those we saw so far, is peculiar of biaxial media. It is referred to as conical refraction. To outline it, let us consider a biaxial slab, deﬁned by two planes normal to one of the optic axes, and illuminated by a uniform plane wave at normal incidence. By definition of optic axis, any direction in the normal plane is permissible for D, and the refractive index n is independent of the direction of D. Snell’s law in Eq. (4.84) shows that, for normal incidence, the direction of the phase vector does not change as the wave passes from one medium to the other. However,

© 2006 by Taylor & Francis Group, LLC

AN APPLICATION: PHASE MATCHING OF TWO WAVES

197

Figure 6.3 Conical refraction.

as we saw in Section 6.2, the vector E in general is not in the same direction as the vector D, having a longitudinal component (see Figure 6.1). Consequently, the Poynting vector is not parallel to the optic axis, and furthermore, its direction, belonging always to the plane orthogonal to the magnetic ﬁeld, depends on that of the vector D. Suppose now that the incident wave is depolarized, i.e., is a random superposition of waves whose D vector is linearly polarized in all directions orthogonal to the propagation direction, with a ﬂat probability distribution. The corresponding directions for the vector E form a surface, which can be shown (see Born and Wolf, 1980, Chapter 14) to be ˆ by a cone. Note, however, that the axis of this cone cannot be parallel to k: deﬁnition of biaxial medium (the three principal permittivities are diﬀerent), an optic axis can never coincide with one of the principal axes, and therefore the length of the longitudinal component of E depends on the direction of D. Consequently, the directions of the Poynting vector also form a cone, whose ˆ Hence, its intersection with the slab output plane is axis is not parallel to k. an ellipse. So, if the slab is illuminated over a ﬁnite round spot, and if the birefringent slab is thick enough, compared to the spot radius, then light (i.e., active power) emerges at the output face through an elliptical-ring-shaped area (see Figure 6.3). It is self-explanatory why this phenomenon is called, as we said before, conical refraction. Note that, as the wave normals remain always orthogonal to the input and output planes, there is no further refraction when the light leaves the slab. A few numerical data on typical uniaxial and biaxial crystals will be given at the end of the next section. 6.5

An application: Phase matching of two waves

Quite often, at least at optical frequencies, the nonlinear phenomena of highest practical interest (e.g., harmonic generation, frequency mixing, parametric ampliﬁcation, etc.) take place in birefringent media. Physics of nonlinear

© 2006 by Taylor & Francis Group, LLC

198

PLANE WAVES IN ANISOTROPIC MEDIA

phenomena in crystals, and nonlinear e.m. wave propagation, are outside the scope of this book. Those readers who are interested in such topics are strongly encouraged to consult several specialized books (Bloembergen, 1996; Zernike and Midwinter, 1973; Agrawal, 1995; Someda and Stegeman, Eds., 1992; etc.). However, most nonlinear wave interactions share an important problem that lends itself to be tackled now, as an application of what we developed in the previous section. For two waves to reach a signiﬁcant eﬃciency in a nonlinear interaction, they must travel with the same phase velocity. Otherwise, local contributions might still be strong, but would not add up coherently, and the average eﬀect over long distances would be essentially zero. Similar considerations apply to nonlinear processes involving three or more waves. Satisfying such constraints among interacting waves is referred to as phase matching. We will provide an example, in a birefringent medium, showing that phase matching is achieved only if the waves travel along selected directions. For simplicity, let us deal with a uniaxial crystal, and leave to the reader, as an exercise, biaxial ones. So, we let† n20 = x = y ; n2z = z . Suppose that, over the spectral range of interest, n0 and nz are both monotonic functions of frequency. Then, if f1 and f2 are the frequencies of two waves involved in a given nonlinear process, we have, inevitably, n0 (f1 ) = n0 (f2 ), and nz (f1 ) = nz (f2 ). Phase matching between two ordinary waves is not possible, regardless of their direction of propagation, neither it is possible between two extraordinary waves. One must try to phase match an ordinary wave and an extraordinary one. Let ϑ be the angle between the direction along which these two waves propagate, z , and the optic axis, z. We are looking for a value of ϑ such that the following equation: n0 (f1 ) = ne (f2 )

,

(6.41)

is satisﬁed, ne being the extraordinary index along z . To that purpose, let us express ne as a function of n0 , nz and ϑ, starting from Fresnel’s equation of the wave normal. To avoid a 0/0 form, let us ﬁrst rationalize Eq. (6.34), and then make use of the following symbols (see Figure 6.4): vx = sin ϑ ,

vy = 0 ,

vz = cos ϑ .

(6.42)

After simple algebra, we get: 1 = ne

cos2 ϑ sin2 ϑ + n20 n2z

1/2 .

(6.43)

† While n0 is called ordinary index without any ambiguity, the term “extraordinary index” is used by some authors to denote what we labelled here as nz , by other authors to denote the quantity (6.43). In this book, we will adhere consistently to the latter rule.

© 2006 by Taylor & Francis Group, LLC

AN APPLICATION: PHASE MATCHING OF TWO WAVES

199

Figure 6.4 Relationship between the propagation direction, z , and the optic axis, z.

So, the condition in Eq. (6.41) becomes an equation in ϑ which reads: 1 1 1 1 − 2 − sin2 ϑ 2 = . (6.44) n0 (f2 ) nz (f2 ) n20 (f2 ) n20 (f1 ) A real solution for ϑ, i.e., a direction of propagation satisfying Eq. (6.41), exists if the two quantities in brackets have the same sign, and if the one on the left-hand side is larger, in absolute value, then the one on the right-hand side. Hence, there are solutions to the problem either for: n0 (f2 ) > n0 (f1 ) > nz (f2 )

,

(6.45)

n0 (f2 ) < n0 (f1 ) < nz (f2 ) .

(6.46)

or for: In fact, in our formalism the two symbols f1 and f2 are completely interchangeable. (They are no longer interchangeable only if at a certain frequency, between f1 and f2 , the inequality between n0 and nz is reversed). So, the actual meaning of Eqs. (6.45) and (6.46) is that, given n0 and nz at one frequency (say f1 ), phase matching is possible at all those frequencies f2 at which n0 (f2 ) is within the range between n0 (f1 ) and nz (f1 ). The polarizations of the two waves, on the other hand, cannot be interchanged. It is easy to prove, as an exercise, that the extraordinary wave must be the higher frequency, if n0 is an increasing function of frequency f , and Eq. (6.45) applies; the lower frequency, if n0 is still an increasing function of frequency f , but Eq. (6.46) applies. The reverse is true if the ordinary index is a decreasing function of frequency. A numerical example may help clarifying how one proceeds. Let us deal with second-harmonic generation in KDP (Deuterated Potassium Phosphate), one of the most widely used materials in bulk integrated optics. It is a uniaxial crystal. Suppose that the light source is a ruby laser, emitting in the red, at λ1 = 694 nm. The second-harmonic wavelength is then λ2 = λ1 /2 = 347 nm, in the green. The measured values for the refractive indices can be found in

© 2006 by Taylor & Francis Group, LLC

200

PLANE WAVES IN ANISOTROPIC MEDIA Table 6.1 Typical values of KDP refractive indices at various wavelengths (T = 25◦ C). λ [nm] 266 355 532 694 1064

n0

nz

1.5593 1.5311 1.5123 1.5050 1.4938

1.5099 1.4858 1.4705 1.4653 1.4599

Typical values of KDP refractive indices at T = −100◦ C. λ [nm] 266 355 532 694 1064

n0

nz

1.5651 1.5363 1.5172 1.5097 1.4985

1.5133 1.4890 1.4736 1.4685 1.4630

the literature, and are: n0 (f1 ) = 1.506

,

nz (f1 ) = 1.466

,

n0 (f2 ) = 1.534

,

nz (f2 ) = 1.487 .

We see that n0 (f1 ) falls in the range between n0 (f2 ) and nz (f2 ). Hence, an ordinary wave at the fundamental frequency can be phase matched to an extraordinary second-harmonic wave. To ﬁnd the direction in which the two waves have to travel, we take Eq. (6.43) and rewrite it as: −1/2 cos2 ϑ 1 − cos2 ϑ + . 1.506 = (1.534)2 (1.487)2 The solution is easily found to be ϑ 50◦ . One of the features of birefringent media that the previous example is intended to point out is that, usually, the refractive indices and the birefringence (i.e, the diﬀerence between indices) depend signiﬁcantly on wavelength. They also depend strongly on temperature. Table 6.1 stresses this message, showing, as an example, ordinary and extraordinary indices in KDP at a few selected wavelengths and at two temperatures. On the other hand, Table 6.2 shows a list of birefringent crystals of practical interest. Because of what was pointed out just now, the ﬁgures in the table are indicative, and are meant to give the reader just a feeling of the orders of magnitude of the indices and of the birefringence. To solve actual technical problems, the reader should look for more detailed data vs. wavelength and vs. temperature, which can be found in many books and handbooks (e.g., Yariv and Yeh, 1984; Dmitriev et al., 1991; Tosco, Ed., 1990).

© 2006 by Taylor & Francis Group, LLC

GYROTROPIC MEDIA

201

Table 6.2 Some typical birefringent crystals. Uniaxial: positive birefringence Quartz Zircon Rutile ZnS

n0 1.54 1.92 2.62 2.35

nz 1.55 1.97 2.90 2.36

1.52 1.51 1.66 2.19 2.29 2.42

1.48 1.47 1.49 2.18 2.20 2.37

ny 1.526 1.582 1.620 1.938

nz 1.530 1.588 1.627 1.947

negative birefringence (ADP) (NH4 )H2 PO4 (KDP) KH2 PO4 Calcite LiTaO3 LiNbO3 BaTiO3 Biaxial Feldspar Mica Topaz YAlO3

6.6

nx 1.522 1.552 1.619 1.923

Gyrotropic media

An anisotropic dielectric is referred to as gyrotropic, if there is a real rectangular reference frame, (x1 , x2 , x3 ), where the dyadic c is represented by a matrix of the form: ⎞ ⎛ 11 12 0 (6.47) (c ) = ⎝ −12 11 0 ⎠ . 0 0 33 This deﬁnition makes use of a particular reference frame, and this may appear as a weak point. Indeed, we might replace it with other deﬁnitions based on intrinsic properties. The reason why this one was preferred, is because, in the cases of practical interest, physical considerations make it quite evident that one direction is “special” for that medium. Then, we choose it as the x3 axis of the reference frame. On the other hand, all the directions orthogonal to the special one are interchangeable. In fact, let us show that the matrix in Eq. (6.47) is invariant under any rotation of the reference frame around the x3 axis. Such a rotation does not aﬀect, obviously, the x3 components, while its eﬀects in the (x1 , x2 ) plane are described by the sub-matrices: cos ϑ sin ϑ cos ϑ − sin ϑ −1 (M ) = , (M ) = , (6.48) − sin ϑ cos ϑ sin ϑ cos ϑ

© 2006 by Taylor & Francis Group, LLC

202

PLANE WAVES IN ANISOTROPIC MEDIA

where ϑ is the angle of rotation around x3 , in the sense that in the new frame the upper-left (2 × 2) submatrix of Eq. (6.47) is replaced by: 11 12 11 12 −1 (M ) (M ) = , (6.49) −12 11 −12 11 whence we draw the conclusion that Eq. (6.47) does not change. In short, the choice of x1 and x2 , on the plane normal to x3 , is completely arbitrary. The examples given in the following sections will enlighten all these statements, and provide them with a stronger physical meaning. An anisotropic magnetic medium such that, in a real rectangular reference frame, its dyadic μ is represented by a matrix with the same features as Eq. (6.47), is also called gyrotropic, or rather gyromagnetic. Most of what we say in this section on dielectric gyrotropic media applies in a straightforward way to gyromagnetic media as well. In Section 6.9 we will discuss an example of such a medium. Combining Eq. (6.47) with Eq. (6.10), i.e., the condition (c ) = (c )† for an anisotropic medium to be lossless, we see that in a lossless gyrotropic medium all the diagonal elements of the matrix in Eq. (6.47) are real, whereas all the oﬀ-diagonal ones are imaginary: 11 = ∗11

,

33 = ∗33

,

12 = −∗12

.

(6.50)

In this section, similarly to what we did in the previous ones, we will focus on lossless media, gaining in simplicity without losing too much in generality. We will study plane waves which travel along one of the axes of the reference frame where Eq. (6.47) applies. We will postpone propagation in an arbitrary direction to the next section. First, let us see whether uniform plane waves can propagate in a direction orthogonal to x3 , for example along x1 . These results will apply, of course, to any direction in the (x1 , x2 ) plane, as a consequence of Eq. (6.49). By deﬁnition a1 ) a ˆ1 , of uniform plane waves, it must be ∇2 E = −k 2 E and ∇(∇·E) = −k 2 (E·ˆ where a ˆ1 is the unit vector of the x1 axis. Thus, Eq. (6.12) reads now:

ˆ1 ) a . (6.51) ˆ1 ω 2 μ · E = k 2 E − (E · a Note that this equation is reminiscent of Eq. (6.30); this is not a pure coincidence. Writing now Eq. (6.51) by components, we reach easily the conclusion that it has two solutions, namely: E = E3 a ˆ3 12 E = E2 − a ˆ1 + a ˆ2 11

, ,

k 2 = ω 2 μ 33 , 2 k 2 = ω 2 μ 11 + 12 11

.

(6.52)

This shows that there are two possible states of polarization, and that their phase velocities are diﬀerent. At ﬁrst sight, this is similar to birefringent media (see Section 6.3), but there is a subtle diﬀerence, as one of the allowed states of polarization (the one in the (x1 , x2 ) plane) is elliptical, since 12 is imaginary.

© 2006 by Taylor & Francis Group, LLC

THE APPLETON-HARTREE FORMULA

203

Furthermore, note that this wave can be evanescent, and actually is evanescent for |11 | < |12 |. When the phase vector k is parallel to x3 axis, the picture is completely diﬀerent. The matrix representation of Eq. (6.51) reads then: 11 12 E1 E1 ω2 μ = k2 . (6.53) −12 11 E2 E2 The characteristic equation of the matrix in Eq. (6.53) is: (11 − κ)2 = −212

,

(6.54)

whose solutions are: (6.55) κL,R = 11 ± j 12 . Because of the conditions in Eq. (6.50), or, equivalently, since the matrix () is Hermitian, both eigenvalues of (t ), in Eq. (6.55), are real. The corresponding normalized eigenvectors: 1 1 1 1 (L ) = √ , (R ) = √ (6.56) 2 j 2 −j represent the clockwise and counterclockwise rotating √ unit-vectors, respecˆ = 1/ 2 (ˆ ˆ = tively, well known from Chapter 2, namely L a1 + jˆ a2 ) and R √ a1 − jˆ a2 ). The net result is that uniform plane waves may travel along 1/ 2 (ˆ the x3 axis, provided they are circularly polarized. Their propagation constants are given by: 2 kL

= ω 2 μ(11 + j12 )

,

(6.57)

2 kR

= ω 2 μ(11 − j12 )

.

(6.58)

These results entail very important consequences, to be discussed in Section 6.10. 6.7

The Appleton-Hartree formula

Continuing along the guidelines of the previous section, let us deal now with waves traveling in an arbitrary direction, neither x3 nor orthogonal to it. It may be helpful to change reference frame. We take new rectangular coordinates, y1 , y2 , y3 , with the y3 axis parallel to the wave vector, and with the medium preferential direction x3 lying in the (y1 , y3 ) plane. Let 1 , 2 (= 0), and 3 = (1 − 21 )1/2 be the direction cosines of the “old” x3 axis with respect to the new axes yi . Then, the reader may check, as an exercise, that in the new frame the dyadic is represented by the following matrix: ⎞ ⎛ 2 3 12 1 3 (33 − 11 ) 3 11 + 21 33 ⎠ . ⎝ (6.59) 11 1 12 −3 12 2 2 1 3 (33 − 11 ) −1 12 1 11 + 3 33 We saw in Section 6.3 that for a uniform plane wave in a birefringent medium the vector D is orthogonal to y3 , and belongs to the plane deﬁned

© 2006 by Taylor & Francis Group, LLC

204

PLANE WAVES IN ANISOTROPIC MEDIA

Figure 6.5 The electric ﬁeld vector and the electric displacement vector in the (y1 , y2 , y3 ) reference system.

by y3 and by the vector E. This result still holds in the case at hand, as one can prove as an exercise. In the reference frame {yi }, we can express these statements through the following formulas (see Figure 6.5): D3 = 0 , D2 E2 = = −jp D1 E1

,

(6.60)

where p is the linear polarization ratio of the D ﬁeld, as deﬁned in Chapter 2. Combining Eq. (6.60) with the constitutive relation D = · E (written by components using the matrix in Eq. (6.59)), we get the following seconddegree equation in p: 3 12 33 p2 + j21 (11 33 − 211 − 212 ) p − 3 12 33 = 0 .

(6.61)

Solving Eq. (6.61) we get the polarizations of the two uniform plane waves which can propagate in the y3 direction. Consider a lossless gyrotropic medium (i.e., 12 imaginary and 11 , 33 real). Then, when both Eqs. (6.57) and (6.58) turn out to be positive (i.e., if |12 | < 11 ), Eq. (6.61) divided by the imaginary unit has real coeﬃcients, and a real and positive discriminant, which automatically entails two real roots. From the properties of the polarization ratio p discussed in Chapter 2, we conclude then that the axes of the polarization ellipse coincide with y1 and y2 . When the medium is lossy, the last points are no longer true. However, an interesting property can be derived in any case, as follows: the ﬁrst coeﬃcient and the last one in Eq. (6.61) are equal and opposite; hence, the product of the two roots of the equation has to be −1: p1 · p2 = −1 .

(6.62)

Again because of the properties of p seen in Chapter 2, Eq. (6.62) entails that the two corresponding time-dependent vectors D ∼ describe ellipses which

© 2006 by Taylor & Francis Group, LLC

THE APPLETON-HARTREE FORMULA

205

Figure 6.6 Schematic illustration of the two allowed states of elliptical polarization.

are mirror images of each other, with respect to the line y1 = y2 , i.e., the straight line at 45 degrees between the y1 and y2 axes (see Figure 6.6). We can derive now an expression for the refractive index, deﬁned always through Eq. (6.26). To this purpose, rewrite Eq. (6.27) as E1 D1 (6.63) = n2 0 , D3 = 0 . D2 E2 Next, write D1 , D2 and D3 (= 0) as functions of E1 , E2 and E3 , using the matrix in Eq. (6.59). The third equation obtained in this way allows us to express E3 as a function of E2 and E1 , namely:

1 ( . (6.64) − ) E + E E3 = 2 11 33 1 3 1 1 12 2 1 11 + 23 33 If we insert this into the expressions for D1 , and eliminate E2 by means of Eq. (6.60), we end up with the following general expression for n2 : n2 =

33 11 + j3 12 p 0 21 11 + 23 33

.

(6.65)

Replacing p with −1/p thanks to Eq. (6.62), we get n2 =

33 3 12 + j11 p 0 jp(21 11 + 23 33 )

.

(6.66)

If p is real (lossless medium, 12 imaginary), then n2 is real. If we suppose j3 12 p < 0, then, Eq. (6.66) is always positive, while Eq. (6.65) is either positive or negative, depending on whether 11 > 3 |12 p|, or 11 < 3 |12 p|. However, the complete opposite occurs if we suppose j3 12 p > 0. When n2 is negative, we have an evanescent wave, whose physical meaning will be outlined in Section 6.10. In a lossy medium, n2 is complex. Hence, any wave traveling in such a medium undergoes an attenuation.

© 2006 by Taylor & Francis Group, LLC

206

PLANE WAVES IN ANISOTROPIC MEDIA

What we said concerning waves traveling in an arbitrary direction in a gyrotropic medium is of practical interest to radio engineers, as it applies to ionospheric propagation, where the preferential direction x3 is that of the terrestrial magnetic ﬁeld. The direction of propagation, y3 , and the direction x3 deﬁne a plane (in our notation, it is the y2 = 0 plane) which is usually referred to as the “magnetic meridian plane”. The form taken by Eq. (6.65) or by Eq. (6.66), when one inserts values for ij which are typical of the ionosphere, is known in the literature as the Appleton-Hartree formula (Budden, 1961). 6.8

An example of permittivity dyadic

An e.m. wave traveling in an ionized gas is always accompanied by a convection current, because of the motion of electrons and, to a smaller extent, of positive ions, although they are much heavier than electrons. In this section we will see how, under suitable assumptions, this convection current can be accounted for, in a complex permittivity dyadic. The equation of motion for an individual electron in an e.m. ﬁeld, driven by the Lorentz force in Eq. (1.2), neglecting collisions, is: dv ∼

= −e(E (6.67) ∼+v ∼×B ∼) , dt where m = electron rest mass (= 9.1091 10−31 kg), e = electron charge magnitude (= 1.6021 10−19 C), v ∼ = instantaneous velocity. For a uniform plane wave in a medium whose intrinsic impedance does not diﬀer too much from that of free space, the ratio |B|/|E| is so small that, for non-relativistic particles (i.e., for | v ∼ | c), the second term on the right-hand side of Eq. (6.67) is negligible, compared to the ﬁrst one. From now on, this term will be omitted. However, it happens often, in plasma physics laboratories as well as in nature in the ionosphere, that an ionized gas is placed in a strong static magnetic ﬁeld, whose induction vector will be denoted as B0 . In the presence of this B0 and in the absence of a time-harmonic ﬁeld (i.e., for E ∼ = 0), Eq. (6.67) becomes the equation of motion along a helix, whose axis is parallel to the ﬂow lines of B0 . It is rather intuitive that propagation of a wave in such a medium is strongly anisotropic, with signiﬁcant diﬀerences between the direction of B 0 and all the orthogonal ones. Let us assume, for simplicity, that B0 is constant throughout the region of interest, and introduce rectangular coordinates with the x3 axis parallel to B0 . From Eq. (6.67) one sees that for B0 → ∞ the radius of curvature of the helical electron trajectories tends to zero. So, let us ﬁrst make a simpliﬁed analysis under the assumption B0 = ∞, which forbids the electrons to move in the plane transverse to x3 so that all quantities of interest depend only on x3 and time. In the time domain, the convection current density can then be written as: m

J =ρ v , ∼3 ∼ ∼3

© 2006 by Taylor & Francis Group, LLC

(6.68)

AN EXAMPLE OF PERMITTIVITY DYADIC

207

where ρ is the instantaneous electron charge density.‡ The electrons may move ∼ with a nonvanishing average velocity, v0 , referred to as the drift velocity. In the rest of this section, we will use a subscript 0 for the dc components, and a subscript a for the ac components. Then, Eq. (6.68) reads: J0 + Ja = (ρ0 + ρa) (v0 + va) . ∼ ∼ ∼

(6.69)

Clearly, J0 = ρ0 v0 . Assuming that ρa and va are both time-harmonic, with an ∼ ∼ angular frequency ω, then the time-dependent part contains terms at angular frequencies ω and 2ω. Assuming that the latter are negligibly small, we may write: (6.70) J ∼ = ρ0 va + ρa v0 . ∼a ∼ ∼ Our next, time-consuming task is to derive from Eq. (6.70) how the ac electric ﬁeld is related to the ac current density Ja. In fact, to link va with the ∼ ∼ electric ﬁeld, we note ﬁrst that, if we adopt a ﬂuid-dynamics viewpoint, we may write, in general: ∂ v3 ∂ v3 d v3 ∂ v3 ∂ v3 ∂x3 ∼= ∼+ ∼ = ∼ + v3 ∼ ∼ ∂x3 dt ∂t ∂x3 ∂t ∂t

.

(6.71)

The last term, a product of two quantities at ω, can be neglected as we did a moment ago with ρava. Introducing then new symbols (capital letters) such ∼∼ that va= Re(Va ejωt ), etc., the left-hand side of Eq. (6.67) can be replaced by: ∼ ∂Va m jω Va + v0 . (6.72) ∂x3 Assume now that all ac quantities depend on x3 as exp(−κx3 ), where κ is still unknown. Using calligraphic capital letters to indicate that Va = Va e−κx3 , etc., Eq. (6.67) becomes: e (6.73) (jω − κv0 ) Va = − E3 . m To reach our ﬁnal goal, we need now a suitable expression for ρa, to be used ∼ in Eq. (6.70). We get it from the continuity Eq. (1.29), namely: ∂ρ ∼ ∇· J ∼ = − ∂t which, under our assumptions, yields: κJa = −jω Pa

,

(6.74)

,

(6.75)

where ρa= Re{Pa ejωt e−κx3 }. Re-writing now Eq. (6.70) for the complex ∼ ‡ This analysis neglects electrostatic forces. This implies an underlying assumption, namely that the ionized gas, on average, is neutral, i.e., the average ion charge density is equal, in absolute value, to the electron charge density. When this assumption is not veriﬁed, as in a particle beam, then the results that we will ﬁnd here hold only as a ﬁrst approximation.

© 2006 by Taylor & Francis Group, LLC

208

PLANE WAVES IN ANISOTROPIC MEDIA

quantities represented by capital letters, and using for Va and Pa the expressions which follow from Eqs. (6.73) and (6.75), we get: ρa e Ja = −jω E3 . (6.76) m(jω − κv0 )2 Deﬁning the quantity: ωp =

N e2 0 m

1/2 ,

(6.77)

which is referred to as the plasma angular frequency, Eq. (6.76) becomes: Ja = −jω 0

ωp2 E3 (jω − κv0 )2

.

(6.78)

From our initial assumption that any electron motion perpendicular to x3 is forbidden, it follows that in the planes orthogonal to x3 the medium behaves like free space. So, we may write: ⎞ ⎛ 0 0 0 (c ) = ⎝ 0 0 0 ⎠ , (6.79) 0 0 33 with:

33 = 0

ωp2 1− (ω + jκv0 )2

.

(6.80)

In these expressions, ρava has been neglected with respect to (ρ0 va + ρa ∼∼ ∼ ∼ v0 ), and Va ∂Va /∂z has been neglected with respect to (jωVa + v0 ∂Va /∂z). Consequently, when v0 = 0, i.e., for a plasma at rest, these formulas are correct only if ρ0 ρa and |ω/κ| Va , i.e., if the wave phase velocity is large, ∼ compared with the electron speed. Now let us remove the assumption B0 = ∞, and take into account, in the electron motion, the components orthogonal to B0 . The starting point is again Eq. (6.67), where we still neglect the contribution v ∼×B ∼, due to the ac component of B . Eq. (6.68) is replaced by the complete vector equation: ∼ J (6.81) ∼=ρ ∼v ∼ . Assume, for simplicity, that the average velocity is equal to zero (v0 = 0). Then, the ac component of Eq. (6.81) can be written (neglecting, as usual, products of ac components on the right-hand side), as: (6.82) J = ρ 0 va . ∼a ∼ Suppose now that a uniform plane wave propagates in this medium, in an arbitrary direction. Then, all complex quantities representing time-harmonic functions are products of constants times a function of spatial coordinates of the type: (6.83) exp(−κ · r) = exp(−κ1 x1 − κ2 x2 − κ3 x3 ) .

© 2006 by Taylor & Francis Group, LLC

AN EXAMPLE OF PERMITTIVITY DYADIC

209

Under these circumstances, the left-hand side of Eq. (6.67) can be written as: m (jω − κ1 Va1 − κ2 Va2 − κ3 Va3 ) V a . (6.84) If the particle velocities are small compared to the wave phase velocity, then Eq. (6.84) can be approximated as: jω m V a

.

(6.85)

In the same reference frame as before (x3 parallel to B0 ), Eq. (6.67) may be written by components in a matrix form as: ⎛ ⎞ eB0 ⎛ ⎞ ⎛ ⎞ jω 0 E1 ⎜ ⎟ Va1 m ⎜ ⎟ e ⎜ ⎟ ⎜ −eB0 ⎟⎜ ⎟ (6.86) ⎜ ⎝ Va2 ⎠ = − ⎝ E2 ⎠ . jω 0 ⎟ ⎜ m ⎟ m ⎝ ⎠ V E3 a3 0 0 jω Inverting the matrix in Eq. (6.86), and setting: e B0 , (6.87) ωc = m (this quantity is referred to as the cyclotron angular frequency), we get: ⎛ ⎞ ω jωc 0 ⎜ ωc2 − ω 2 ωc2 − ω 2 ⎟⎛ ⎞ ⎛ ⎞ ⎜ ⎟ E1 Va1 ⎜ ⎟ ⎜ ⎟⎜ ⎟ ω e ⎜ −jωc ⎜ ⎟ ⎟ ⎝ E2 ⎠ . 0 V = −j (6.88) ⎝ a2 ⎠ 2 2 2 2 ⎜ ⎟ ωc − ω m ⎜ ωc − ω ⎟ ⎜ ⎟ Va3 1 ⎠ E3 ⎝ 0 0 − ω Going back to Eq. (6.82), and from this to Maxwell’s equations, we ﬁnally ﬁnd the dyadic c we were looking for. In our reference frame, it is represented by a matrix of the type in Eq. (6.47) with the following elements: ωp2 11 = 0 1 + 2 , ωc − ω 2 12

= −21 = j

33

= 0 1 −

ωp2 0 ωc 2 ω ωc − ω 2

ωp2 ω2

.

,

(6.89)

Note that the denominators in 11 , 12 and 21 , all vanish if ω = ωc . The medium has a resonant behavior when the angular frequency of the e.m. wave equals the angular velocity of the helical electron motion around the ﬂow lines of the vector B0 .

© 2006 by Taylor & Francis Group, LLC

210

PLANE WAVES IN ANISOTROPIC MEDIA

It is left to the reader as an exercise to show that, under suitable conditions, the medium exhibits an evanescence, as brieﬂy mentioned in Section 6.7. The reader should also elaborate on the physical relationship between this evanescence and the frictionless motion of charged particles. 6.9

Second example of permeability dyadic

In this section, we will be obliged to use concepts, terminology and notation taken from magnetostatics, a discipline which is not dealt with in this book. We take it for granted that the reader had been exposed to these notions in basic physics courses. An interest in refreshing them can easily be satisﬁed by consulting classical textbooks on static ﬁelds (e.g., Plonsey and Collin, 1961). In the so-called semiclassical treatment of ferromagnetic media (i.e., if a minimal amount of Quantum Mechanics is invoked), the magnetic dipole moment due to the individual electron spin, m ∼ , is looked at as the result of a rotation of the electron around its axis, with an angular momentum, Q ∼, proportional to m . In formulas: ∼ m ∼=gQ ∼ ,

(6.90)

where g is called the gyromagnetic constant of the electron (g = −1.76 × 1011 T/s). When immersed in a magnetic induction ﬁeld B ∼, the dipole moment m experiences a torque: ∼ τ =m ×B ∼ ∼ ∼

.

(6.91)

Hence, the dynamics of the vector m ∼ obeys the equation: dm ∼ dt

=gm ∼×B ∼

.

(6.92)

In a saturated magnetic material, if we neglect, as a ﬁrst approximation, demagnetizing factors related to the shape of the material itself, we may write: B ∼ = μ0 (H ∼×M ∼ ) = μ0 (H ∼ +N0 m ∼)

,

(6.93)

where N0 is the spin moment density (number per unit volume). Multiplying both sides of Eq. (6.92) times N0 , and using Eq. (6.93), we get: dM ∼ dt

= −g μ0 H ∼×M ∼ .

(6.94)

This is easily recognized as the equation of a precession around the vector H ∼. If the H ﬁeld has a dc component, called orienting ﬁeld, and if this com∼ ˆ3 , is parallel to the x3 axis, then, in the saturation regime, the ponent, H0 a ˆ3 , and use static component of M ˆ3 . We will call it M0 a ∼ is also parallel to a a subscript a to denote complex quantities which represent ac components.

© 2006 by Taylor & Francis Group, LLC

FARADAY ROTATION

211

From Eq. (6.94), we then get: ⎛ ⎞ ⎛ ⎞⎛ ⎞ jω −g μ0 H0 0 −Ha2 Ma1 ⎜ ⎟ ⎜ ⎟⎜ ⎟ jω 0 ⎠ ⎝ Ma2 ⎠ = g μ0 H0 ⎝ Ha1 ⎠ ⎝ g μ 0 H0 0

0

jω

Ma3

.

(6.95)

0

Solving this for the vector Ma , and replacing it into: Ba = μ0 (Ha + Ma ) ≡ μ · Ha

,

(6.96)

we get, in our reference frame, a matrix representation for the dyadic μ, whose elements are: ω0 ωM , μ11 = μ22 = μ0 1 + 2 ω0 − ω 2 μ12 μ33 = μ0

= −μ21 = j ,

μ0 ωωM ω02 − ω 2

,

μ13 = μ23 = μ31 = μ32 = 0 ,

(6.97)

where we set: ω0

= −g μ0 H0

,

ωM

= −g μ0 M0

.

(6.98)

The minus signs are due to the fact that g < 0 for an electron. Eqs. (6.97) show that, according to the deﬁnition given in Section 6.6, a magnetized ferrite is indeed a gyrotropic medium. The angular frequency ω 0 plays a role which is very similar to that of the cyclotron angular frequency ω c in a gyrotropic plasma. Physically, in fact, both these quantities are related to angular velocities (of rotation around B0 for ωc , of precession around H0 for ω0 ). Eqs. (6.97) exhibit a resonant behavior centered at ω = ω0 , like Eqs. (6.89). Examples of actual microwave devices which exploit gyromagnetic media, as well as of very similar optical devices, will be given at the end of the next section. 6.10

Faraday rotation

Some features of plane waves traveling along the preferential direction, x3 , in a gyrotropic medium, were derived in Section 6.6. However, the subject was left open for further discussion. We will see in this section that those waves provide a very interesting example — probably, the example which is most widely used in practice — of nonreciprocal electromagnetic behavior. Let us start oﬀ with two preliminary remarks, restricting ourselves, once again, to lossless media. A) As a matter of argument, assume that 12 is imaginary and positive. Then,

© 2006 by Taylor & Francis Group, LLC

212

PLANE WAVES IN ANISOTROPIC MEDIA

the right-hand side of Eq. (6.58) is always positive. The right-hand side of Eq. (6.57) is positive or negative, depending on whether 11 > |12 |, or 11 < |12 |. In the latter case, the root kL is imaginary: the corresponding wave undergoes an attenuation along x3 , while its phase does not change in this direction. The medium is assumed to be lossless, so an attenuation cannot be attributed to power dissipation. It must originate from an evanescence, similarly to what we saw in Chapter 4. In general, 11 and 12 are frequency dependent. So, if ω is changed, one can pass from a situation where both circularly polarized waves are uniform plane waves, to a situation where one of them is evanescent, while the other one is still uniform. B) The terminology “clockwise” or “counterclockwise,” that we adopted in Section 6.6 for the rotation of the instantaneous vectors represented either by ˆ or by L, ˆ Eqs. (6.56), refers to an observer looking at the positive face of the R x3 = 0 plane, regardless of whether the wave is traveling in the positive or negative sense of the x3 axis. This entails that: • as long as we deal with waves traveling in the sense of increasing x3 , the wave which rotates like a left-handed screw has a propagation constant equal to the positive square root of Eq. (6.57), the wave which rotates like a right-handed screw has a propagation constant equal to the positive square root of Eq. (6.58); • when we deal with waves traveling in the sense of decreasing x3 , the two propagation constants are interchanged, i.e., the negative square root of Eq. (6.58) for the left-handed screw, the negative square root of Eq. (6.57) for the right-handed screw. It is precisely this exchange between propagation constants when the sense in which the waves travel is reversed, that is responsible for nonreciprocity, as we will see in a shortwhile. Suppose now that, at the frequency of interest, kL and kR are both real. Let a uniform plane wave impinge on a plane orthogonal to x3 , for example on the x3 = 0 plane, traveling in the direction of increasing x3 , and assume that its electric ﬁeld is linearly polarized along a direction lying in the x3 = 0 plane, say x1 (Figure 6.7). This vector E can be written by components in the basis Eq. (6.56), which consists of the eigenvectors of the matrix () and therefore is easy to track as we move through the gyrotropic medium. In fact, the electric ﬁeld at any value of x3 can be simply written as: E0 ˆ −jkR x3 E0 ˆ −jkL x3 e e +√ R E= √ L 2 2

.

(6.99)

The vector in Eq. (6.99), sum of two circularly polarized vectors which are equal in modulus (for any x3 ), remains (see Chapter 2) linearly polarized everywhere. However, at a generic x3 , its direction no longer coincides with the input one. It is at an angle ϑ, with respect to the x1 axis, that can be

© 2006 by Taylor & Francis Group, LLC

FARADAY ROTATION

213

Figure 6.7 Schematic illustration of Faraday rotation.

found if we rewrite Eq. (6.99) in the form: kL +kR kL − kR kL − kR x3 + a x3 ˆ1 cos ˆ2 sin e−j 2 E = E0 a 2 2

x3

.

(6.100) From this, we ﬁnd: ϑ = arctan

E2 kL − kR x3 = E1 2

.

(6.101)

So, we see that the direction of polarization is rotated by an angle proportional to x3 , the proportionality factor (sometimes referred to as circular birefringence) being (kL −kR )/2 = 0 for 12 = 0, i.e., for a gyrotropic medium. This phenomenon is known as the Faraday rotation. As kL and kR are interchanged when propagation is reversed (see the previous comment B) this rotation is peculiar in that the sign of the rotation angle (with respect to a ﬁxed reference frame) does not change when the direction of propagation is reversed. Consequently, let us write, rather than Eq. (6.101): ϑ=

kL − kR |x3 | 2

.

(6.102)

This implies that the reciprocity theorem (Section 3.4) does not apply to the medium at hand. To prove this statement, let us deal with two sets of sources, satisfying the following requirements. The ﬁrst set generates a wave which travels in the positive x3 direction, and whose electric ﬁeld equals Eq. (6.99) on the x3 = 0 plane. The second set generates a wave which travels in the negative x3 direction, and whose electric ﬁeld on the plane x3 = equals Eq. (6.99) in amplitude, but is polarized at an angle ϑ0 = (kL − kR )/2 with respect to the x1 axis. Let us compute the reactions of these two ﬁelds on the sources, according to the deﬁnition given in Section 3.4. It is easily seen that, except for the very special cases where ϑ0 is an integer multiple of 2π, the two reactions are not equal. This proves, as we intended to do, that we are facing a nonreciprocal phenomenon.

© 2006 by Taylor & Francis Group, LLC

214

PLANE WAVES IN ANISOTROPIC MEDIA

Figure 6.8 Illustration of a microwave isolator.

Faraday rotation is widely exploited in practice, especially at microwave and optical frequencies. One of the simplest nonreciprocal devices, and at the same time one of the most useful ones, is the so-called isolator. Its function is to introduce a strongly nonreciprocal attenuation — ideally, no attenuation in one direction, an inﬁnite attenuation in the opposite one — aiming at making an oscillator insensitive to any change in its load. The most common types of isolators (although not the only ones) are based on Faraday rotation. Figure 6.8 provides a schematic illustration of its waveguide version at microwave frequencies, Figure 6.9 shows an optical implementation. The two are much more similar in their essence than at ﬁrst sight. Although the ﬁrst one was developed several decades before the second, we will brieﬂy comment only on the latter one, because understanding some features of the ﬁrst requires concepts to be developed in Chapter 7. The reader should come back to it after the chapter on conducting-wall waveguides. The optical isolator is a Faraday rotator (whose gyromagnetic medium is, usually, a garnet crystal), designed in such a way that the rotation angle ϑ equals π/4, and placed between two linear polarizers that are rotated with respect to each other also by π/4. As illustrated schematically in the ﬁgure, a backward wave is strongly attenuated by polarizer A, while a forward wave passes with very low attenuation through polarizer B. Another Faraday-rotation-based nonreciprocal device which is widely used at microwave frequencies is the so-called circulator. It is an n-port junction, with n ≥ 3, deﬁned, in its ideal version, as follows. When a wave impinges on the i-th port, all the incident power comes out from the i + 1-th port (understanding that port n + 1 is the same as port 1). Nothing is reﬂected, or transmitted to the remaining “decoupled” n − 2 ports. Figure 6.10 provides a schematic illustration of a 3-port circulator in a waveguide conﬁguration (part a), and the symbol used for it in circuit diagrams (part b). A qualitative explanation of how a Y -junction 3-port circulator operates may proceed as follows. Because of the conducting walls in the planes orthogonal to the static induction ﬁeld, the clockwise and counterclockwise waves of the previous analysis are no longer traveling, but standing waves. Their propagation constants along the vertical axis remain diﬀerent. As they are not in an unbounded medium, but in a closed environment, this diﬀerence in longitudinal propagation constant entails, through Maxwell’s equations, that their azimuthal phase velocities (i.e., the velocities at which they rotate around the static ﬁeld) are diﬀerent too. Then, a suitable design can make the two

© 2006 by Taylor & Francis Group, LLC

FARADAY ROTATION

215

Figure 6.9 Illustration of an optical isolator.

Figure 6.10 A 3-port microwave circulator. (a) View of a waveguide conﬁguration. (b) Symbol in use for circuit diagrams.

waves excited by a ﬁeld entering through port 1 meet in phase at port 2, in counterphase at port 3, whence the conclusion. A typical use of the 3-port circulator is in radars, where it allows the transmitter (connected to port 1) and the receiver (connected to port 3) to share the same antenna (connected to port 2) without “seeing” each other, which would be extremely detrimental since the transmitted power is many orders of magnitude larger than the received one. In addition to its technical relevance, which is illustrated also by some of

© 2006 by Taylor & Francis Group, LLC

216

PLANE WAVES IN ANISOTROPIC MEDIA

the problems at the end of the chapter, Faraday rotation is very important as a concept. It stresses that, if we want to avoid any ambiguity in discussing reciprocity in an anisotropic medium, it is strictly mandatory to use real reference frames, and to deal only with linearly polarized ﬁelds. To clarify the meaning of this statement, let us recall that, as pointed out in Section 6.6, the matrix () of a gyrotropic medium becomes diagonal (and therefore, symmetˆ ric) in a complex reference frame where two of the basis unit vectors are R ˆ Consequently, circularly polarized plane waves traveling in a gyrotropic and L. medium obey equations that are formally identical to those governing uniform plane waves in isotropic media. This might induce one to claim that circularly polarized waves in a gyrotropic medium behave reciprocally, but such a statement does not agree with the deﬁnition of reciprocity, based on the reaction, that we gave in Section 3.4. The origin of this apparent ambiguity is that the reaction, deﬁned through a scalar product between two complex vectors, without taking the complex conjugate of one of its factors, is not an invariant with respect to any change of basis in a complex space. In particular, the reaction does change, in general, when we pass from linearly polarized basis vectors to rotating basis vectors. 6.11

Further applications and suggested reading

This chapter contains only those fundamental notions about anisotropic media, that may be considered essential ingredients, nowadays, in the education of an advanced engineer who intends to work in the area of information technology. The level of knowledge which is necessary in order to become a specialist in propagation in anisotropic media is orders of magnitude larger than the contents of this short chapter. The list of further suggested readings is potentially unlimited. Similarly to what we said at the end of Chapter 4, we ﬁrst stress that linear anisotropic media are very pervasive in physics and in engineering. This statement applies, of course, also to the mathematical formalism which is normally used for their analysis, i.e., tensors and their matrix representations, that is shared by many diﬀerent ﬁelds of application (e.g., Babuska and Cara, 1991; Hijden, 1987; Nayfeh, 1995 and references that treat the propagation of mechanical, e.g., seismic waves in anisotropic media). To broaden one’s own general background in physical sciences, one should at least examine a broad-scope, comprehensive book, like, for instance, Nye (1985), where the mathematical properties of anisotropic media are presented in detail, and then followed by applications, ranging from magnetic and electric properties, to theory of elasticity, from transport phenomena to crystal optics. An extremely important concept that the reader can grasp from such a book is the physical relevance, in some problems, of tensors whose rank is higher than 2 (2 is, of course, the rank of dyadics, that we used in this chapter). Other treatments of crystal optics include Juretschke (1974). For what refers, more speciﬁcally, to further reading on electromagnetic

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

217

waves in anisotropic media, let us mention the following examples. Optics of crystals is the title of a famous chapter in Born and Wolf (1980), where the beauty of the presentation matches the richness of the scientiﬁc contents. Our emphasis is on electrical anisotropy. Materials with magnetic anisotropy are treated in, e.g., Tarling and Hrouda (1993) — see also Negi and Saraf (1989). For a treatment of electromagnetic wave propagation in bianisotropic media, we refer to Kong (1990) and Staelin et al. (1993). Propagation in magnetoplasmas, with emphasis on ionospheric phenomena, is thoroughly dealt with in Budden (1961). Those who are interested in the fundamentals of plasma physics and magnetohydrodynamics can ﬁnd them, e.g., in Jackson (1975), Chandrasekhar (1989), Sitenko and Malnev (1995), and Dendy, Ed., (1993), while Miyamoto (1989) focuses on applications of plasmas in thermonuclear fusion for energy production, and Sturrock (1994) includes material regarding astrophysical and geophysical as well as laboratory plasmas. As we already mentioned in Section 6.5, many among the experiments that opened, in the ’60s, the new ﬁeld of nonlinear optics — in bulk media, at that time — were performed in anisotropic media, birefringence being a crucial ingredient for obtaining phase matching. The fundamentals of nonlinear optics, as well as their links with propagation in anisotropic media, are well covered, e.g., in Bloembergen (1996) — see also Boyd (1992), Butcher and Cotter (1990), Newell and Moloney (1992), Shen (1984), and Yariv and Yeh (1984). Much more recently, thanks to the availability of very-low-loss optical ﬁbers, where interaction lengths could become as large as several kilometers, guidedwave nonlinear optics became much more attractive than its bulk precursor. Fundamentals on nonlinear phenomena in waveguides can be found, e.g., in Agrawal (1995). More speciﬁcally, Someda and Stegeman, Eds., (1992) covers the area where guided-wave propagation overlaps with anisotropy (for example, random birefringence in optical ﬁbers) and with nonlinearities. Propagation problems in anisotropic waveguides are an area where we may state — anticipating something to be seen in Chapter 10 — very few problems can be solved analytically in closed form. Numerical methods become then extremely important; a relevant example is provided by Zoboli and Bassi in Someda and Stegeman, Eds., (1992). The reader should remember what we said in Section 3.9 about symmetries: careful exploitation of spatial symmetries in numerical methods can yield enormous savings in processing time and in memory occupation. Notice, however, that in anisotropic media symmetry operations are legitimate only if the dyadic operators that characterize materials commute with the symmetry operators that characterize the geometry of the region where the ﬁeld is deﬁned. Problems 6-1 When the dyadics c and μ are non-singular, i.e., when both have an inverse, then Eqs. (6.3) and (6.4) can be inverted, expressing the electric and

© 2006 by Taylor & Francis Group, LLC

218

PLANE WAVES IN ANISOTROPIC MEDIA

magnetic ﬁeld vectors as functions of the displacement and induction vectors. Write the corresponding expression of the Poynting vector. Derive conditions under which the factors contained in this expression commute. 6-2 Show that if at least one of the dyadics c and μ is not Hermitian, then the quantity in Eq. (6.8) may have a nonvanishing real part. Find a way to distinguish media where this real part is positive (lossy media) and those where it is negative (media with gain). 6-3 Write Eq. (6.12) by components, in a rectangular reference frame, in a region without imposed currents. 6-4 Show that, in those cases where Eq. (6.12) becomes a Helmholtz equation, the same occurs to Eq. (6.19), provided the ϕ = 0 gauge is adopted. 6-5 Starting from the contents of Section 6.2, and applying the duality theorem of Section 3.7, discuss the case of a magnetic anisotropic medium in terms of an electric vector potential F, with the same physical dimensions as in Section 1.9. 6-6 Using duality, analyze a magnetic medium whose constitutive relations are the dual of those of the birefringent dielectric medium studied in Section 6.3. 6-7 Show that, if a plane wave is launched into a lossless birefringent medium in a generic direction and with a generic state of polarization, then its state of polarization evolves periodically as a function of the distance traveled by the wave. Hence, express the period as a function of the ordinary and extraordinary indices for that direction of propagation. Does the state of polarization still evolve periodically in the presence of loss ? 6-8 In a rutile crystal (a uniaxial dielectric medium whose refractive indices are on the list in Table 6.2), plane waves propagate along the straight line at 45 degrees between the x and z axes, z being the optic axis. Evaluate the values of the ordinary and extraordinary refractive indices. Use both approaches explained in the text: ﬁrst, ﬁnd the eigenvalues of the reduced inverse permittivity dyadic bt . Next, solve Fresnel’s equation of the wave normals as shown in Section 6.5. 6-9 A rutile slab is cut with planar parallel faces, spaced by 10 mm, orthogonal to the y axis. The surrounding medium is air. The incident wave impinges with an incidence angle ϑi = π/4. The plane of incidence is the (x, y)-plane. Applying Snell’s law and the Fresnel formulas of Chapter 4, ﬁnd the directions of propagation of the TE and TM refracted waves (note the socalled double refraction), and the transmission coeﬃcients. What is the angle

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

219

between the directions of propagation of ordinary and extraordinary waves when they emerge in air after the slab ? Is this result valid in general ? 6-10 Find the angles between the phase vectors and the Poynting vectors (ray directions) for the refracted waves within the slab of the previous problem.

6-11 Prove that, in a uniaxial medium where n0 is an increasing function of frequency f , phase matching between fundamental frequency and second harmonic can be achieved only if the extraordinary wave is the second harmonic when Eq. (6.45) applies; only if it is the fundamental frequency when Eq. (6.46) applies. Prove that the reverse is true, if the ordinary index is a decreasing function of frequency. 6-12 One method of fabricating an optical polarizer is to cut a prism of uniaxial medium in such a way that a plane wave, emerging from it towards air, undergoes total internal reﬂection for one polarization (either ordinary, or extraordinary), but not for the other one. Find the range of values for the angle of incidence for which this occurs, both for calcite and for zircon (see Table 6.2). Specify the totally reﬂected wave, in both cases. 6-13 As mentioned brieﬂy in Chapter 2, a quarter-wave plate is a slab of birefringent material which converts a linearly polarized input wave into a circularly polarized output wave. Consider three slabs of mica (see Table 6.2), each of them cut normally to one of the axes x, y, z. The incident wave is linearly polarized at 45 degrees with respect to the axes laying in the input plane of the slab. a) Calculate the (minimum) λ/4-slab thicknesses in the three conﬁgurations. b) In reality, if the incident wave (in air) is linearly polarized, the output polarization will be slightly elliptical because the reﬂection coeﬃcients along the two principal axes will be slightly diﬀerent. Evaluate the deviation from circular polarization, in terms of percentage. 6-14 Show that, if a plane wave is launched into a lossless gyrotropic medium in the x3 direction, with a generic state of polarization, then its state of polarization evolves periodically as a function of the distance traveled by the wave (provided both circularly polarized “eigenwaves” propagate at the frequency of interest). Express the period as a function of the propagation constants of the two circularly polarized eigenwaves. Does the state of polarization still evolve periodically with distance in the presence of loss ? 6-15 Prove that a lossless anisotropic medium, where the only plane waves that propagate in one direction (x3 ) preserving their state of polarization, are circularly polarized, is gyrotropic.

© 2006 by Taylor & Francis Group, LLC

220

PLANE WAVES IN ANISOTROPIC MEDIA

6-16 Prove that the change in reference frame described at the beginning of Section 6.7 yields Eq. (6.59). 6-17 Show that, analogous to what was shown in Section 6.3, for a uniform plane wave in a gyrotropic medium where the reference frame of Section 6.7 is used, the vector D is orthogonal to y3 and, if the wave is linearly polarized, it belongs to the plane deﬁned by y3 and by the electric ﬁeld vector E. 6-18 Study under which conditions one of the plane waves in the medium studied in Section 6.8 is evanescent. Can both waves be evanescent ? 6-19 In the highly ionized layers of the Earth’s ionosphere, electron densities are of the order of 1012 m−3 . A typical value for the terrestrial static magnetic induction ﬁeld is 3 × 10−5 T. a) Calculate the corresponding plasma frequency and cyclotron frequency. b) Repeat the calculation for a laboratory plasma with 1016 electrons per cubic meter and a 10 T induction ﬁeld. 6-20 Sketch the plots of the dispersion diagrams, k(ω) vs. ω, for the two circularly polarized waves whose phase constants are expressed by Eqs. (6.57) and (6.58), when the elements of the permittivity matrix are expressed by Eqs. (6.89). Consider separately the cases ωp > ωc , ωp < ωc . 6-21 Evaluate the intensity of the static magnetic ﬁeld H0 which is required in order to have a resonant frequency f0 = 1 GHz, in a saturated ferrite. 6-22 To build a microwave isolator (see Section 3.4) based on Faraday rotation, a saturated ferrite bar is used, with a static magnetic ﬁeld, H0 , applied in the direction along which the wave propagates. a) Approaching the problem in the dual way with respect to a gyrotropic dielectric medium, express the phase constants of the circularly polarized waves, and then the per-unit-length Faraday rotation, for a gyromagnetic medium with = 0 , and the permeability dyadic found in Section 6.8. Note: assume that the bar cross-section is wide enough to approximate the propagation constants in it with those of plane waves in an unbounded medium. b) Evaluate the intensity of the static magnetic ﬁeld required in order to have a Faraday rotation of π/4 radians over a length = 30 mm at a frequency f = 5 GHz. The saturation magnetization M0 equals 106 A/m.

© 2006 by Taylor & Francis Group, LLC

References Agrawal, G.P. (1995) Nonlinear Fiber Optics. 2nd ed., Academic, San Diego. Babuska, V. and Cara, M. (1991) Seismic Anisotropy in the Earth. Kluwer Academic, Boston. Bloembergen, N. (1996) Nonlinear optics. 4th ed., World Scientiﬁc, River Edge, NJ. Born, M. and Wolf, E. (1980) Principles of Optics: Electromagnetic Theory of Propagation, Interference and Diﬀraction of Light. 6th (corrected) ed. (reprinted 1986), Pergamon, Oxford. Boyd, R.W. (1992) Nonlinear Optics. Academic Press, Boston. Budden, K.G. (1961) Radio Waves in the Ionosphere. Cambridge University Press, Cambridge, UK. Butcher, P.N. and Cotter, D. (1990) The Elements of Nonlinear Optics. Cambridge University Press, Cambridge, UK. Chandrasekhar, S. (1989) Plasma Physics, Hydrodynamic and Hydromagnetic Stability, and Applications of the Tensor-Virial Theorem. University of Chicago Press, Chicago. Dendy, R. (Ed.) (1993) Plasma Physics: An Introductory Course. Cambridge University Press, UK. Dmitriev, V.G., Gurzadyan, G.G. and Nikogosian, D.N. (1991) Handbook of Nonlinear Optical Crystals. Springer-Verlag, Berlin. Hijden, J.H.M.T. van der (1987) Propagation of Transient Elastic Waves in Stratiﬁed Anisotropic Media. North Holland, Amsterdam. Jackson, J.D. (1975) Classical Electrodynamics. Wiley, New York. Juretschke, H.J. (1974) Crystal Physics: Macroscopic Physics of Anisotropic Solids. Benjamin, Advanced Book Program, Reading, MA. Kong, J.A. (1990) Electromagnetic Wave Theory. 2nd ed., Wiley, New York. Miyamoto, K. (1989) Plasma Physics for Nuclear Fusion. Rev. ed., MIT Press, Cambridge, MA. Nayfeh, A.H. (1995) Wave Propagation in Layered Anisotropic Media, with Applications to Composites. Elsevier, Amsterdam. Negi, J.G. and Saraf, P.D. (1989) Anisotropy in Geoelectromagnetism. Elsevier, Amsterdam. Newell, A.C. and Moloney, J.V. (1992) Nonlinear Optics. Advanced topics in the interdisciplinary mathematical science, Addison-Wesley, Redwood City, CA. Nye, J.F. (1985) Physical Properties of Crystals: Their Representation by Tensors and Matrices. 1st paperback edition with corrections and new material, Clarendon Press, Oxford, UK. Plonsey, R. and Collin, R.E. (1961) Principles and Applications of Electromagnetic Fields. McGraw-Hill, New York. Shen, Y.R. (1984) The Principles of Nonlinear Optics. Wiley, New York. Sitenko, A.G. and Malnev, V. (1995) Plasma Physics Theory. Chapman & Hall, 221 © 2006 by Taylor & Francis Group, LLC

222

REFERENCES

London. Smirnov, V.I. (1961) Linear Algebra and Group Theory. McGraw-Hill, New York. Someda, C.G. and Stegeman, G.I. (Eds.) (1992) Anisotropic and Nonlinear Waveguides. Elsevier, Amsterdam. Staelin, D.H., Morgenthaler, A.W. and Kong, J.A. (1993) Electromagnetic Waves. Prentice-Hall, Englewood Cliﬀs, NJ. Sturrock, P.A. (1994) Plasma Physics: An Introduction to the Theory of Astrophysical, Geophysical, and Laboratory Plasmas. Cambridge University Press, UK. Tarling, D.H. and Hrouda, F. (1993) The Magnetic Anisotropy of Rocks. Chapman & Hall, London. Tosco, F. (Ed.) (1990) Fiber Optic Communications Handbook. TAB Professional and Reference Books, Blue Ridge Summit, PA. Yariv, A. and Yeh, P. (1984) Optical Waves in Crystals: Propagation and Control of Laser Radiation. Wiley, New York. Zernike, F. and Midwinter, J.E. (1973) Applied Nonlinear Optics. Wiley, New York.

© 2006 by Taylor & Francis Group, LLC

CHAPTER 7

Waveguides with conducting walls

7.1

Introduction

What we saw in Chapters 4, 5 and 6 indicates that there are no physical reasons why in a homogeneous medium the energy carried by an electromagnetic wave should travel along a given direction. Indeed, Chapters 12 and 13, devoted to the fundamentals of antenna theory, will reconﬁrm that sources in a homogeneous medium cannot radiate just along one linear path, but necessarily in all directions, at least within a given cone. Guided propagation, which means sending almost all the energy conveyed by an electromagnetic wave along a chosen path, inevitably requires an inhomogeneous medium. However, depending on the applications and especially on the frequency in use, nonhomogeneities can either consist of combining dielectrics and conductors, or merely using diﬀerent dielectric media. In principle, electromagnetic ﬁelds penetrate in all media whose conductivity is ﬁnite, and satisfy continuity conditions on the surfaces between diﬀerent media. In practice, we may encounter two drastically diﬀerent situations. In a structure which consists only of dielectrics, in none of them is the ﬁeld negligible, even as a ﬁrst approximation. On the other hand, the step from a good conductor (deﬁned as in Section 4.12) to an ideal conductor, in which the ﬁeld is identically zero, is, at least in the ﬁrst approach to the problem, an excellent trade-oﬀ between being realistic and simplifying the mathematics. In fact, it allows replacement of the continuity conditions at the interface between dielectric and metal with boundary conditions for the ﬁeld, deﬁned only in the dielectric. For historical reasons, but above all for their mathematical simplicity which follows from what we said so far, conducting-wall waveguides will be dealt with ﬁrst, in this chapter. Dielectric waveguides will be studied later, in Chapter 10. The subject will not be tackled in terms of plane waves, although their completeness, proved in Section 4.7, would enable us to do that. In fact, a diﬀerent approach will yield more compact results, practically useful and formally elegant at the same time. We will invoke a plane wave approach only at some points where it provides an intuitive explanation for some of the results obtained in the other way. 223 © 2006 by Taylor & Francis Group, LLC

224

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.1 A waveguide of unspeciﬁed geometry, and inﬁnite length.

7.2

Homogeneously ﬁlled cylindrical structures: Simpliﬁed proof of the TE-TM decomposition theorem

Let us consider, in real 3-D space, a region ﬁlled with a homogeneous medium and surrounded by walls, whose physical nature does not have to be speciﬁed now, which extend to inﬁnity along one direction, z, and whose cross-section is constant vs. z. We will summarize these assumptions calling this region a “cylindrical structure in the z direction”. Let a ˆ z be the unit vector in the z direction, u1 , u2 be two generic orthogonal curvilinear coordinates in the plane orthogonal to z (see Figure 7.1). We will show that, under these circumstances, the TE-TM decomposition theorem can be proved in a greatly simpliﬁed way, compared with Section 3.8, while at the same time a physical identiﬁcation is possible for the rather mysterious pre-potentials that were introduced at that stage. As we anticipated in Section 3.8, there is some price to pay for this simpliﬁcation. One is the need to make an ansatz (easy to justify) for the z-dependence of the ﬁelds, the other one is a loss in generality, to be discussed and by-passed later in this section. As for the ansatz, it consists of the following assumptions, which can ﬁnd their justiﬁcation either in what we saw in Section 4.7 or, a posteriori, in the results that will follow from them, in particular for what concerns completeness of the solutions: E(u1 , u2 , z)

=

¯ 1 , u2 ) exp(−γz) E(u

,

H(u1 , u2 , z)

=

¯ 1 , u2 ) exp(−γz) H(u

.

(7.1)

In Section 3.8 it was shown that any ﬁeld in such a region can be expressed as the sum of two ﬁelds which are transverse electric (TE) and transverse magnetic (TM) with respect to an arbitrary real unit vector a ˆ. From now on, unless the opposite is explicitly stated, TE and TM will be understood with respect to a ˆz . To simplify the proof as much as possible, let us deal, for the time being, with rectangular coordinates (u1 = x, u2 = y), and write Maxwell’s equations by components using the symbolic determinant notation (see Appendix B) for the curl operator. We get: ∂Ez + γ Ey ∂y

© 2006 by Taylor & Francis Group, LLC

= −jωμ Hx

,

HOMOGENEOUSLY FILLED CYLINDRICAL STRUCTURES ∂Ez = −jωμ Hy , −γ Ex − ∂x ∂Hz + γ Hy ∂y

= jωc Ex

225

,

∂Hz = jωc Ey , (7.2) ∂x plus the two components along z, which we omit here as they are not relevant to the following. These equations can be looked at as an algebraic system of four linear equations in the four unknowns Ex , Ey , Hx , Hy , whose solution is simply found to be: ∂Hz 1 ∂Ez + jωμ Ex = − 2 γ , γ − κ2 ∂x ∂y ∂Hz 1 ∂Ez − jωμ γ Ey = − 2 , γ − κ2 ∂y ∂x ∂Hz ∂Ez 1 +γ −jωc Hx = − 2 , γ − κ2 ∂y ∂x ∂Hz ∂Ez 1 + γ jω , (7.3) Hy = − 2 c γ − κ2 ∂x ∂y −γ Hx −

where, as usual, κ2 = −ω 2 μc . These equations prove the ﬁrst part of the decomposition theorem, namely that any ﬁeld in the region of interest can be expressed as the sum of a TE and a TM ﬁeld. The transverse electric ﬁeld consists of Hz and of four transverse components expressed by the terms in the right column, on the right-hand side of Eqs. (7.3). The transverse magnetic ﬁeld consists of Ez and of four transverse components expressed by the terms in the left column, on the right-hand side of Eqs. (7.3). However, it is extremely important to point out that this proof holds only for γ 2 = κ2 . Otherwise, the quantity in the denominators on the right-hand side goes to zero. Implications of this restriction will be discussed shortly. Before we proceed to the second part of the decomposition theorem, we note, incidentally, that Eqs. (7.3) can be rewritten in an intrinsic form, independent of the (x, y) reference frame, as follows: E¯t

=

γ2

1 (−γ ∇t Ez + jωμ a ˆz × ∇t Hz ) , − κ2

¯t H

=

γ2

1 (−γ ∇t Hz − jωc a ˆz × ∇t Ez ) , − κ2

(7.4)

where the transverse gradient ∇t is deﬁned through the position: ∇ = ∇t + a ˆz

© 2006 by Taylor & Francis Group, LLC

∂ = ∇t − a ˆz γ ∂z

,

(7.5)

226

WAVEGUIDES WITH CONDUCTING WALLS

and can be expressed in a generic system of orthogonal curvilinear coordinates u1 , u2 as: tˆ1 ∂ tˆ2 ∂ ∇t = · + · , (7.6) h1 ∂u1 h2 ∂u2 where h1 , h2 are the metric coeﬃcients of the (u1 , u2 ) reference frame. The second part of the decomposition theorem says that the TE part of the ﬁeld can be derived from a scalar function which satisﬁes the scalar Helmholtz equation, and the same is true for the TM part. In the case at hand, it is straightforward to show that these two scalar functions are (apart from proportionality constants, which are needed to reconcile them with Section 3.8) the longitudinal ﬁeld components Hz and Ez , respectively. In fact, Eqs. (7.3) show, on one hand, that when these two components are known, the entire ﬁeld can be calculated by means of straightforward operations. On the other hand, the ﬁelds H and E, in a homogeneous medium without sources, satisfy the vector Helmholtz equation, as we saw in Chapter 1, and therefore their components along the straight axis z satisfy the scalar Helmholtz equation, as we had to prove. In more detail, if we deﬁne a transverse Laplacian ∇2t through the position: ∂2 = ∇2t + γ 2 , (7.7) ∂z 2 we see that the two functions of the two transverse coordinates u1 , u2 satisfy an equation of the type: ∇2t F = −χ2 F , (7.8) ∇2 = ∇2t +

where F is either Hz or Ez , and: χ2 = −(κ2 − γ 2 ) = 0 .

(7.9)

This is an eigenvalue equation, similar to the Helmholtz equation but different from it because in the 3-D case the eigenvalue χ2 = −κ2 is unique, and is known when the frequency and the medium are speciﬁed, whereas in Eq. (7.8) the eigenvalue is unknown. We will see shortly that indeed Eq. (7.8), at a given frequency, has an inﬁnity of discrete solutions, for suitable boundary conditions. Each eigenvalue corresponds to a value of γ, whose physical meaning is, according to Eq. (7.1), that of a propagation constant in the z direction. We understand then that diﬀerent solutions of Maxwell’s equations of the form in Eq. (7.1), corresponding to diﬀerent eigenvalues in Eq. (7.8), will have diﬀerent propagation properties. The “spatial spectrum” of the total ﬁeld which can exist in a cylindrical structure will be radically diﬀerent from a plane-wave spectrum. The link between the quantities and the symbols used in this section and those of Section 3.8 are simply found, by comparison, to be: L = μ Hz /χ2

,

T = −c Ez /χ2

.

(7.10)

Hence, if one follows the approach of Section 3.8, the expressions in Eq. (3.47) and (3.48) for the TE and TM ﬁelds, respectively, become the following ones,

© 2006 by Taylor & Francis Group, LLC

HOMOGENEOUSLY FILLED CYLINDRICAL STRUCTURES

227

where symbols like L and T indicate, as in the rest of this section, that the exponential dependence on z has been factored out: E¯TE

= −jω ∇t L × a ˆz

¯ TE H

= −

E¯TM

=

¯ TM H

,

1 2 γ (κ − γ 2 ) L a ˆz − ∇t L , μ μ

1 γ (κ2 − γ 2 ) T a ˆz + ∇t T c c

= −jω ∇t T × a ˆz

.

(7.11)

, (7.12)

Comparing Eqs. (7.12) (which, being based on the general proof of Section 3.8, are valid also for χ2 = 0) and Eqs. (7.4), we see that a ﬁeld corresponding to the eigenvalue χ2 = 0 can be non-identically zero only if: Ez ≡ 0

,

Hz ≡ 0

,

(7.13)

i.e., for a transverse electro-magnetic (TEM) ﬁeld. We will see shortly that the loss in generality caused by the restriction χ2 = 0 is not harmful, because the guided propagation of a TEM ﬁeld is quite peculiar. Indeed, its ﬁrst peculiarity, which can be derived starting either from Eqs.(7.11) or from Eqs. (7.12), is that: ¯ ×a E¯ = ±η H ˆz , (7.14) 1/2 is, as usual, the intrinsic impedance of the homogeneous where η = (μ/c ) medium in the cylindrical structure. The sign is either + or − depending on whether γ = +κ or γ = −κ . Hence, for a TEM ﬁeld the propagation constant and the wave impedance coincide with those of a uniform plane wave traveling in the z direction, in an unbounded medium with the same characteristics as those of the medium which ﬁlls the cylindrical structure. We will see in Section 7.4 that none of the other solutions of Maxwell’s equations in a waveguide can have these properties. In the same section we will also see the conditions under which a cylindrical structure can propagate a TEM ﬁeld. For the time being, let us point out that for a TEM wave, the ﬁelds, the potentials and the pre-potentials, as functions of the transverse coordinates, all obey an equation of the following type: ∇2t F = 0 ,

(7.15)

which is the 2-D Laplace equation. Then, we will be entitled to invoke all the uniqueness theorems which hold for this equation, with which the reader should be familiar from electrostatics. To close this section, let us emphasize that we proved that any ﬁeld in a homogeneous cylindrical structure can be expressed as the sum of a TE ﬁeld, a TM ﬁeld and, in some cases to be clariﬁed, a TEM ﬁeld. Whether one of these ﬁelds may exist in a given structure independently of the others, this can be said only when one knows what boundary conditions have to be satisﬁed in that structure. This will be the subject of Section 7.3 and again of Section 7.9.

© 2006 by Taylor & Francis Group, LLC

228 7.3

WAVEGUIDES WITH CONDUCTING WALLS Waveguides with ideal conducting walls

To get involved in the details of guided propagation, let us start from structures where metallic walls shield completely the region where the guided ﬁeld is conﬁned. This approach is convenient, for the reasons that were mentioned in the introduction, both in terms of concepts and of technicalities. The problem is further simpliﬁed in a drastic manner if we assume that the walls consist of an ideal electric conductor (see Section 4.12). Although, obviously, this is not realistic, nevertheless all the results of practical interest can be obtained from this model, at most by introducing further reﬁnements which are usually small and simple. Examples will be given in Section 7.9 and later on. It is well known that an ideal electric conductor imposes Etan = 0 on its surface. Then, through Maxwell’s equations, also the normal component of H, Hn , vanishes on that surface. Given the ansatz Eq. (7.1), these boundary conditions (not independent of each other) on the contour C of the cylindrical region where the guided ﬁeld is deﬁned (Figure 7.1) read: E¯tan = 0 ,

Hn = 0 .

(7.16)

We will show now that, if we leave aside the TEM case assuming χ2 = 0, then Eqs. (7.16) are satisﬁed on C if the following conditions are satisﬁed on the same surface: ⇐⇒ T = 0 , (7.17) Ez = 0 d Hz dn

= 0

⇐⇒

dL =0 dn

,

(7.18)

where d/dn is, as usual, the normal derivative on C. To prove this statement, we note ﬁrst that Eq. (7.17) implies Ez = constant on the surface C. Then ∇t Ez is orthogonal to C at any point. Conversely, Eq. (7.18) implies that the normal component of ∇t Hz vanishes on C, so that this vector is tangent to C at any point. Consequently, the general expressions in Eqs. (7.4) for the transverse components show that Et is normal to C and Ht is tangent to C everywhere. If, furthermore, we recall that Ez = 0 on C by assumption, then we conclude that Eqs. (7.16) are satisﬁed everywhere on C. Note that Eqs. (7.17) and (7.18) are independent of each other, unlike the two Eqs. (7.16). In particular, Eq. (7.17) is compatible with taking Hz ≡ 0 (L ≡ 0), which means it is applicable to a TM ﬁeld. Eq. (7.18) is compatible with taking Ez ≡ 0 (T ≡ 0), meaning it is applicable to a TE ﬁeld. A very important consequence that we may draw is that TE transmission modes and TM transmission modes exist in a cylindrical structure with ideal conducting walls. In Section 7.5 we will see that they form a complete set, in some cases provided that the TEM mode is included also. As we brieﬂy said at the end of Section 7.2, existence and completeness of TE and TM modes are properties which are not shared by all cylindrical structures. They follow from the speciﬁc assumptions in Eq. (7.16). Finally, note that Eqs. (7.17) and (7.18) are two scalar boundary conditions.

© 2006 by Taylor & Francis Group, LLC

TRANSMISSION MODES

229

We saw in Section 7.2 that the complete expressions for TE and TM ﬁelds may be derived from the solutions of two scalar partial diﬀerential equations. We may then conclude that guided propagation in cylindrical structures with ideal conducting walls can be thought of, under a mathematical point of view, as a scalar problem, in spite of the vector physical nature of the guided ﬁeld. 7.4

Transmission modes of lossless cylindrical structures

From now on, unless the opposite is explicitly stated, we will assume that the dielectric in the cylindrical structure is lossless. The walls were supposed in the previous section to be ideal conductors, so the whole structure is strictly lossless. The quantity κ2 = −ω 2 μ (constant, because the dielectric was supposed to be homogeneous at the beginning of Section 7.2) is real and negative. This property will be exploited several times in this section. 7.4.1

Properties of all transmission modes

A1. Real, positive eigenvalues All the eigenvalues in Eq. (7.9) of Eq. (7.8) are real and positive (except for the zero eigenvalue of the TEM mode). This follows from the 2-D form of the ﬁrst Green identity (A.10), namely: ∂ψ d , (7.19) ϕ ∇2t ψ + ∇t ϕ · ∇t ψ dS = ϕ ∂n S which we apply to the cross-section of the structure, S, and to its contour, , setting ϕ = F, ψ = F ∗ , F being any solution of Eq. (7.8). For all modes but the TEM, the boundary conditions in Eqs. (7.17) and (7.18) make the right-hand side of Eq. (7.19) equal to zero. Then, through Eq. (7.8) we get: 2 ∗ ∗ FF dS = ∇t F · ∇t F ∗ dS . (7.20) χ S

S

Both surface integrals in Eq. (7.20) are real and positive, whence the conclusion on χ2 . A2. Critical frequency For each transmission mode (labeled with an index n, shown later in Section 7.5 to be an integer) there is one positive value (zero for the TEM mode) of the angular frequency ω, at which the longitudinal propagation constant γ vanishes. Indeed, if we rewrite Eq. (7.9) as: γn2 = χ2n + κ2 = χ2n − ω 2 μ , and set it equal to zero, then it is satisﬁed for: χn ω = √ ≡ ωcn , μ

© 2006 by Taylor & Francis Group, LLC

(7.21)

(7.22)

230

WAVEGUIDES WITH CONDUCTING WALLS

which is a real and positive solution because of the previous property A1. The symbol ωcn underlines that this is called the critical angular frequency of the n-th mode. It is also called the cut-oﬀ angular frequency, for reasons to be explained in the next paragraph. A3. Cut-oﬀ of a transmission mode For the n-th transmission mode, the longitudinal propagation constant γ n is real at all frequencies smaller than its critical frequency, imaginary at all frequencies higher than its critical frequency. To prove it, we start from Eq. (7.21), which indicates that γn2 > 0 for ω < ωcn , γn2 < 0 for ω > ωcn . Hence, we may let: √ for ω < ωcn , (7.23) γn = αn = ω μ (ωcn /ω)2 − 1 ∈ Re √ for ω > ωcn . (7.24) γn = jβn = jω μ 1 − (ωcn /ω)2 ∈ Im This tells us that any transmission mode in a waveguide does not propagate (i.e., its phase does not change vs. z) at frequencies below its critical frequency (whence the name “cut-oﬀ frequency”). As the structure is strictly lossless, the attenuation described by Eq. (7.23) cannot be caused by energy dissipation. Its physical nature will be clariﬁed by looking also at the following properties. A4. Wave impedance The wave impedance in the z direction is imaginary for ω < ωcn , and real for ω > ωcn . To prove this, note that for a TE mode (Ez ≡ 0), the electric and magnetic nonvanishing terms in Eqs. (7.4) are in quadrature when Eq. (7.23) holds, are in phase when Eq. (7.24) holds. The same is true for the nonvanishing terms in the case of a TM mode (Hz ≡ 0). Further details will be added in the following subsections. Comparing with what we saw in Chapter 4 concerning propagation constants and wave impedances of plane waves, we may conclude that the physical phenomenon why a mode is attenuated vs. z below its cut-oﬀ frequency, is an evanescence. Then, the mode cannot convey any active power in the longitudinal direction of the cylindrical structure. However, in strict correspondence to what we saw in Chapter 4, this statement is correct only in the presence of one wave which attenuates in one sense along the z axis, and loses any validity for two superimposed waves which decay in the two opposite senses. A5. Phase and group velocities For any mode that can propagate in a cylindrical structure (i.e., at ω > ωcn ), the longitudinal phase velocity as a function of frequency can be obtained from Eq. (7.24) and reads: vpn =

© 2006 by Taylor & Francis Group, LLC

ω =c βn

1−

1 ω

cn

ω

2

,

(7.25)

TRANSMISSION MODES 231 √ where c = 1/ μ is the speed of light in the dielectric of the cylindrical structure. Eq. (7.25) indicates that vpn tends to inﬁnity for ω → ωcn , and decreases monotonically as ω increases, approaching c (from above) as ω → ∞. From what we saw in Chapter 4 on phase velocity of plane waves in any direction, this result suggests that at its cut-oﬀ frequency a guided mode consists of plane waves which travel along directions in the plane orthogonal to a ˆz . As ω is increased, i.e., as the wavelength becomes smaller with respect to the transverse geometrical dimensions of the structure, the plane waves the mode consists of, travel along directions which make smaller and smaller angles with that of a ˆz direction, gradually tending to a situation where they ignore the presence of walls. The group velocity: ω 2 1 cn =c 1− vgn = , (7.26) ∂βn ω ∂ω behaves, vs. frequency, in a way which agrees perfectly with the previous comment. Indeed, it equals zero at cut-oﬀ (ω = ωcn ), and increases as the frequency increases, tending asymptotically to c (from below) for ω → ∞. From Eqs. (7.25) and (7.26) it follows that: vgn vpn = c2

.

(7.27)

This relationship holds only in the structures that we are studying here. It cannot be extrapolated, unfortunately, to the dielectric waveguides of Chapter 10, where the behavior of group velocity vs. frequency is of paramount importance, and turns out to be much more complicated. The functions in Eqs. (7.23), (7.24), (7.25), (7.26) are sketched in Figure 7.2. According to the deﬁnitions given in Section 5.2, Eq. (7.26) shows that propagation of a guided mode in a metallic waveguide is aﬀected by anomalous dispersion (dvgn /dω > 0). Eq. (7.26) also shows that two guided modes having diﬀerent critical frequencies (i.e., diﬀerent eigenvalues) have, at the same frequency, diﬀerent group velocities. Eq. (7.25) shows that they have diﬀerent phase velocities also. Consequently, if several modes propagate together, a modulating signal is distorted because the group delay from the waveguide input to its output has as many values as the propagating modes. This cause of distortion, often referred to as “multipath,” in general will be superimposed onto the classical causes of signal distortion, which, as well known, for a two-port linear system read |H(jω)| = constant, H(jω) = aω + b, where H(jω) is the system transfer function. By aﬃnity with what we described in Section 5.2, phenomena related to multiple values of the group delay are also referred to as multimode (or inter-modal) dispersion. Eq. (7.27) shows that in a waveguide where the dielectric is homogeneous it is never possible, not even at a single value of ω, to have equal phase or group velocities for modes whose cut-oﬀ frequencies are diﬀerent. In Chapter 10 we will see that, on the contrary, curves of phase or group velocities vs. ω of

© 2006 by Taylor & Francis Group, LLC

232

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.2 Attenuation constant, phase constant, phase velocity, group velocity vs. frequency, for a generic mode of a lossless metallic waveguide.

diﬀerent modes may intersect each other in waveguide where the dielectric is not the same in the entire cross-section. On the other hand, the previous formulas show that two modes having equal eigenvalues have the same phase and group velocities at any frequency. Modes which are physically distinct but have equal eigenvalues are called degenerate. Their role is quite particular, and becomes extremely important in problems such as power exchange among modes of the same structure or of coupled structures, as we shall clarify later in this chapter and in the two following ones.

A6. The fundamental mode

The mode corresponding to the lowest eigenvalue compatible with the boundary conditions imposed by a given structure, χ20 , is called the fundamental mode of the structure. If it is not degenerate, it is the only mode which can propagate alone in that structure. This occurs over the frequency range between f0 = χ0 c/(2π) and the smallest among the cut-oﬀ frequencies of all the other modes. This is the only frequency range where the problems due to intermodal dispersion may be avoided.

© 2006 by Taylor & Francis Group, LLC

TRANSMISSION MODES

233

Figure 7.3 Behavior of the wave impedance of TE and TM modes as functions of frequency, plotted on the Smith chart.

7.4.2

Properties of TE modes

B1. Wave impedance The wave impedance in the direction of the unit vector a ˆ z , which can be calculated letting Ez = 0 in Eqs. (7.4), is: ηTE =

jωμ γn

.

(7.28)

It is constant over the cross-section of the structure. As for its frequency dependence, Figure 7.3 illustrates it on the Smith chart, to be deﬁned in the next chapter, but it can be easily described even without this graphical tool. One must distinguish, as we said in the previous subsection, whether the frequency is smaller or larger than fcn = ωcn /(2π). For f < fcn , Eq. (7.28) is an inductive reactance, equal to zero for f = 0, then increasing as f increases, becoming inﬁnite for f = fcn . Passing through this value, which corresponds to an ideal open circuit, Eq. (7.28) fcn , and becomes a resistance for f > √ then tends asymptotically to η = μ/ for f → ∞, while γn → jω μ, i.e., while the generic TE mode tends to become a plane wave traveling in the z direction. We ﬁnd an inﬁnite ηTE for f = fcn because at this frequency the ¯ vanishes, so that nothing changes if a cross-section transverse component of H of the structure is “metallized” by means of an ideal magnetic conductor (see Chapter 3). B2. Poynting vector The Poynting vector of the generic TE mode is given by: 1 1 1 ∗ −2αn z ∗ 2 ∗ jωμγn |∇t Hz | a ˆz − 2 jωμHz ∇t Hz , Pn = E × H n = e 2 2χ4n 2χn (7.29) where αn = 0 for f ≥ fcn . Its longitudinal part (the ﬁrst term) is imaginary ˆ z . Its for f < fcn , real for f > fcn , and is oriented in the same sense as a

© 2006 by Taylor & Francis Group, LLC

234

WAVEGUIDES WITH CONDUCTING WALLS

transverse component, parallel to ∇t Hz , can change in phase from one point to another in the cross-section only if the phase of Hz depends on the transverse coordinates. If this is not the case, then the phases of Hz∗ and ∇t Hz cancel each other, and we may state that the transverse component of the Poynting vector is imaginary at any frequency. In fact, we will see in the next section that Hz = constant is always a legitimate assumption. Therefore, we may indeed state that the transverse Poynting vector is indeed purely imaginary. 7.4.3

Properties of TM modes

These properties are “dual” of those of the TE modes, so they will be stated without proofs. C1. Wave impedance The wave impedance in the direction of a ˆ z is: γn ηTM = . (7.30) jω Its behavior vs. ω is also plotted schematically in Figure 7.3, and can be described as follows. For f = fcn we have ηTM = 0: at this frequency, E¯t = 0, so that nothing changes if the cross-section is short-circuited by means of an ideal electric conductor. C2. Poynting vector The Poynting vector of the generic TM mode is given by: 1 1 −2αn z 2 ∗ Pn = e ˆz − 2 jωEz ∇t Ez , − 4 jωγn |∇t Ez | a 2χn 2χn

(7.31)

where again αn = 0 for f ≥ fcn . For its longitudinal and transverse components, refer to Subsection B2. 7.4.4

Properties of the TEM mode

D1. Uniqueness of the TEM mode We saw in Section 7.2 that, when χ2 = 0, the ﬁeld must be TEM in order not to give inconsistent relationships. Now let us show also that the inverse is true, namely that a ﬁeld can be TEM only for χ2 = 0. In fact, for χ2 = 0, Eqs. (7.4) show that if we set Ez = 0, Hz = 0, then we necessarily have E¯t = 0, ¯ t = 0, i.e., the trivial solution of Maxwell’s equations in a region without H sources. Therefore the eigenvalue corresponding to a TEM ﬁeld is unique. For this reason we speak of the TEM mode. D2. Existence of the TEM mode The TEM mode may exist only in cylindrical structures where the crosssection of the dielectric is not a simply connected domain. Indeed, we saw

© 2006 by Taylor & Francis Group, LLC

MODE ORTHOGONALITY

235

that the TEM ﬁeld satisﬁes Eq. (7.15): ∇2t F = 0

(F = L, T ) ,

(7.32)

which is the Laplace equation. A well-known theorem ensures the uniqueness of its solution, when the values of either F or ∂F/∂n are known over the whole contour. Suppose that the border, , consists of only one ideal conductor, where Etan = 0. Then, elaborating on Eqs. (7.12) one may show that this assumption implies ∂L/∂n = 0 on and T = constant on . These conditions are compatible with a ﬁeld identically zero everywhere. Because of the uniqueness theorem that we have just quoted, the vanishing ﬁeld is the solution of the problem. In conclusion, the TEM ﬁeld cannot exist if the cross-section is a simply connected domain. Examples of non-simply connected regions of practical interest are two-wire lines and coaxial cables. The latter will be studied in this chapter. Most of the contents of the following chapter apply to both. One section in the next chapter will be devoted to striplines, which are also very important examples of structures with non-simply connected cross-sections. However, in striplines the dielectric is inevitably not homogeneous throughout the cross-section, and consequently they do not ﬁt rigorously into the patterns of this chapter. Their fundamental mode, as we shall see, is called “quasi-TEM,” and can be dealt with only by means of approximations. D3. TEM cut-oﬀ frequency In non-simply connected structures, the TEM mode is the fundamental mode. This is a corollary of the property D1 (the eigenvalue of the TEM mode is zero): through Eq. (7.22) we see that the cut-oﬀ frequency of the TEM mode is also equal to zero: (7.33) fcTEM = 0 . By deﬁnition of the fundamental mode, the conclusion is obvious. Eq. (7.33) tells us that the frequency range where only one mode propagates begins at zero frequency for those structures whose cross-sections are not simply connected. From direct current, it extends to the smallest among the critical frequencies of the higher-order modes. A typical example of higherorder modes in coaxial cable will be dealt with in Section 7.8. 7.5

Mode orthogonality

Propagation in waveguides without sources is described by homogeneous equations. In any mathematical problem of this type, it is important to discuss linear independence of solutions. In this section, we will prove some relationships which assess linear independence of diﬀerent solutions in its strongest possible form, namely orthogonality relationships. They can be looked at as extensions to an inﬁnite-dimensional vector space (the space of the solutions of Eq. (7.8)) of a property which is well known for eigenvalue problems of ﬁnite dimension, namely that eigenvectors belonging to diﬀerent eigenvalues are orthogonal.

© 2006 by Taylor & Francis Group, LLC

236 7.5.1

WAVEGUIDES WITH CONDUCTING WALLS Orthogonality between two TE modes or two TM modes

As usual, let S be the cross-section, normal to the unit vector a ˆ z , of a cylindrical structure surrounded by ideal conductors. Let be the border of S ( = S ∩ C, where C is the structure wall). Let us write on S the 2-D form of the Green theorem (A.11), where ϕ and ψ are two functions of the two transverse coordinates, regular over the domain S: ∂ϕ ∂ψ 2 2 −ψ ϕ ∇t ψ − ψ∇t ϕ dS = ϕ d . (7.34) ∂n ∂n S Let now ϕ and ψ be two functions satisfying Eq. (7.8). To be speciﬁc, let ϕ = Hzj , ψ = Hzk (i.e., we are dealing with the TEj and TEk modes). Both functions satisfy the boundary condition of Eq. (7.18), so the right-hand side of Eq. (7.34) is zero. The left-hand side can be modiﬁed exploiting Eq. (7.8), so we write: 2 2 χk − χj Hzj Hzk dS = 0 . (7.35) S

We get immediately the following orthogonality relationship: Hzj Hzk dS = 0 for χ2j = χ2k .

(7.36)

S

If we let ϕ = Ezj , ψ = Ezk (i.e., if we deal with two TM modes), satisfying the boundary condition of Eq. (7.17), then the right-hand side of Eq. (7.34) is again zero, and we reach the following result: Ezj Ezk dS = 0 for χ2j = χ2k . (7.37) S

Two additional orthogonality relationships can be proved starting from Eqs. (7.36) and (7.37), and using the ﬁrst Green identity written in the form in Eq. (7.19). In fact, a proportionality relationship similar to Eq. (7.20) yields then: E¯tj · E¯tk dS = 0 for χ2j = χ2k , (7.38)

S

¯ tj · H ¯ tk dS = 0 H

for

χ2j = χ2k

.

(7.39)

S

These last relationships hold both for TE and for TM modes. In the case of two (or more) degenerate modes (as we said, modes which are physically diﬀerent but belong to equal eigenvalues, χ2j = χ2k , j = k), what we may say for certainty is that there exist always two (or more) linear combinations of such modes, that satisfy the same orthogonality relationships that we proved now. These linear combinations are still propagation modes, as we will see in Section 7.7. The algorithm by means of which orthogonal modes can be obtained from non-orthogonal modes is, in abstract terms, the same as that which is known in linear algebra as the Schmidt orthogonalization procedure (Dicke and Wittke, 1960). It can be sketched as follows: one chooses,

© 2006 by Taylor & Francis Group, LLC

MODE ORTHOGONALITY

237

arbitrarily, one among the degenerate modes, and leaves it as it is. Then, one takes the second mode (the order is still arbitrary if the modes are more than two) and subtracts from it its projection onto the ﬁrst mode. From the third mode, one subtracts the projection onto the ﬁrst one and the projection onto the second one; one proceeds in the same manner with the following modes. 7.5.2

Orthogonality between a TE mode and a TM mode

To prove that the transverse ﬁelds of a TE and a TM mode also satisfy Eqs. (7.38) and (7.39), even when their eigenvalues are equal, let us apply Stokes’ theorem in Eq. (A.9) to the vector ψ∇ϕ, where ψ and ϕ are two regular scalar functions of space coordinates. Exploiting the identity (C.6), so that ∇ × (ψ∇ϕ) = ∇ψ × ∇ϕ, we have: (∇t ψ × ∇t ϕ) · a ˆz dS = ψ∇ϕ · d . (7.40) S

Let ψ = Ezk (a TM mode) and ϕ = Hzj (a TE mode). Then, it is ψ ≡ 0 on , so the right-hand side of Eq. (7.40) vanishes. The result is: ∇t Ezk × ∇t Hzj · a ˆz dS = 0 . (7.41) S

Thanks to the cyclic permutation rule for the mixed product of three vectors and to Eq. (7.5), from Eq. (7.41) it follows that Eqs. (7.38) and (7.39) still hold, without any restriction on the eigenvalues. 7.5.3

Consequences of mode orthogonality

The total power ﬂowing through the cross-section of a waveguide with ideal conducting walls is the sum of the powers carried by the individual modes, since all “cross terms”: 1 Etj × H∗tk · a ˆz dS (j = k) , (7.42) 2 S vanish. Strictly speaking, this comment follows not from Eqs. (7.38) and (7.39), but from two similar equations, where Etk and Htk are replaced by E∗tk and H∗tk . These relations are easily proved following the previous lines, and can be veriﬁed as an exercise. If a ﬁeld {E, H} in a waveguide is expressed as a series of propagation modes, then the expansion coeﬃcients can be simply calculated by an “orthogonal projection,” i.e., evaluating integrals of the type: E · E∗k dS , H · H∗k dS . (7.43) S

S

A third fundamental consequence of mode orthogonality will be outlined in the next section.

© 2006 by Taylor & Francis Group, LLC

238 7.6

WAVEGUIDES WITH CONDUCTING WALLS Some remarks on completeness

In Section 7.2, we made use (and gave a simpliﬁed proof) of the theorem shown in Section 3.8. It says that in a homogeneous cylindrical structure any e.m. ﬁeld can be split as the sum of a TE and a TM ﬁeld (leaving aside, for the time being, the peculiar TEM case). In Section 7.3, we proved that, in a structure where the boundary conditions of Eq. (7.16) hold, there are TE and TM ﬁelds such that each of them, individually, satisﬁes these boundary conditions. The scope of validity of this second result is much narrower than that of the ﬁrst one, which is insensitive to boundary conditions: in fact, Section 7.11 will provide an important counter-example, where the ﬁrst result holds but the second does not. There are many problems in which an important prerequisite is to be able to “count” the modes of a structure, i.e., to be able to assess whether a given set of solutions is complete or not. In the theory of partial diﬀerential equations, boundary conditions and completeness are always linked to each other. This statement matches, in our case, with what we did in Section 4.7, where the assumption of a square-integrable ﬁeld was instrumental to prove completeness of plane waves, and in Section 5.12, when we discussed completeness of Gaussian beams. Let us now take the case of closed cylindrical structures, where boundary conditions hold on a closed surface which is all at a ﬁnite distance from any point where the ﬁeld is deﬁned. Then, we may proceed by analogy with what happens to signals in the time domain, in particular with the well-known links between the Fourier integral (for boundary conditions at inﬁnity) and the Fourier series (for boundary conditions at the two ends of an interval). This analogy suggests that a complete set may consist of an inﬁnity of discrete elements. However, the waveguide problem is 2-D, and in two dimensions a discrete inﬁnity can be built up in an inﬁnite number of possible ways. There is no reason to be sure, a priori, that any of these procedures will eventually lead to a complete set. To give an example of a set which is obviously not complete, let us take an inﬁnity of solutions which diﬀer from each other only in their dependence on one coordinate. In summary, cardinality of a set (i.e., the number of its elements) is not enough to assure completeness. In any case, completeness of a countable set of solutions of a partial differential equation (PDE) is a subject that can be kept within tolerable limits of size and complexity only if the PDE is separable, that is to say, it can be reduced to ordinary diﬀerential equations (ODE’s), each of these in one unknown quantity, a function of only one coordinate. In a 2-D case, like the one at hand, a complete set of solutions of the PDE consists then of all the products of a countable inﬁnity of linearly independent solutions of the ODE with respect to the ﬁrst coordinate, multiplied by a countable inﬁnity of linearly independent solutions of the ODE with respect to the second coordinate. As for Eq. (7.8), it can be shown that, for positive real eigenvalues, it is separable only in four types of orthogonal coordinates: rectangular cartesian;

© 2006 by Taylor & Francis Group, LLC

RECTANGULAR WAVEGUIDES

239

circular polar (where the coordinate lines are straight half-lines in radial directions, and circles); parabolic (whose coordinate lines are two families of confocal, mutually orthogonal parabolas); elliptical (whose coordinate lines are confocal ellipses and confocal hyperbolas, mutually orthogonal).∗ When dealing with a cylindrical structure of any shape, one can always adopt a coordinate frame which belongs to this set, but clearly, such a reference frame is beneﬁcial only when the boundary conditions are imposed on coordinate lines, or at least on lines that can be broken into pieces of coordinate lines. When this is not the case, then Eq. (7.8) can still be separated into two ODE’s, but their solutions cannot satisfy the boundary conditions on an individual basis, and the ﬁnal result is that each mode is not expressed as the product of two functions, but as the product of two series of functions. Nowadays, many numerical methods have been developed on these premises, and can in fact study waveguides of any shape within reasonable processing times and memory occupation, but it remains extremely hard to assess the completeness of a set of solutions obtained in this way. All the structures dealt with in this book have cross-sections which not only allow the separation of variables, but also enable us to satisfy boundary conditions on an individual basis, i.e., with any single element in a countable inﬁnity of solutions of two ODE’s. Modes will be expressed as products of individual functions, not of series. No formal mathematical proof will be given for the completeness of these sets. However, the reader’s background on periodic functions will be enough to make him/her conﬁdent that we do not neglect any physically meaningful solution. However, one should not extrapolate from these examples, erroneously, that, given a separable 2-D PDE, the products of any two countable inﬁnities of linearly independent solutions of the two ODE’s form a complete set. This is not true in general even if all these functions satisfy the boundary conditions. A very simple counter-example goes as follows: given a complete set of linearly independent solutions, take away one of its elements. The set which is left is still a countable inﬁnity, and its elements are still linearly independent, but clearly it cannot be complete, because none of the linear combinations of the elements which are left can express the term which was taken away. 7.7

Rectangular waveguides

Waveguides of rectangular cross-section provide an excellent example of application for the previous results, being at the same time technically signiﬁcant and mathematically simple. Let us introduce rectangular coordinates (x, y, z), and place the origin in a corner of the cross-section (Figure 7.4), the x-axis along the longer side, of length a, b being the length of the shorter side. In this reference frame, the dif∗ The proof of this statement is too long to be developed here. Let us say just that it follows from the theory of analytical functions of a complex variable. It can be found rather easily in the literature (e.g., Morse and Feshbach, 1953).

© 2006 by Taylor & Francis Group, LLC

240

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.4 A rectangular waveguide, and its reference frame.

ferential operator which appears in Eq. (7.8) reads ∇2t = (∂ 2 /∂x2 )+(∂ 2 /∂y 2 ), and separation of variables is simply the 2-D version of the procedure we saw in Section 4.2. Let: F(x, y) = X(x) Y (y) . (7.44) Then, the same steps as in Section 4.2 yield a pair of harmonic equations: X = −kx2 X

,

Y = −ky2 Y

,

(7.45)

whose separation constants are linked to the eigenvalue −χ2n = (κ2 − γn2 ) as follows: (7.46) kx2 + ky2 = χ2n . We know from Section 7.4 that χ2n is real and positive, and from Section 7.3 that F must satisfy boundary conditions at ﬁnite distance. This suggests that it may be convenient to write the general integrals of Eq. (7.45) in terms of sine and cosine functions, i.e.: X

= A sin kx x + B cos kx x ,

Y

= C sin ky y + D cos ky y

,

(7.47)

where A, B, C and D are arbitrary constants.† The boundary conditions on the walls are diﬀerent, as we saw before, for TE and for TM modes. For TE modes, Eq. (7.18) holds, and once the variables are separated it reads: dX =0 dx

for

x=0, x=a ,

dY =0 dy

for

y=0, y=b .

(7.48)

Diﬀerentiating Eqs. (7.47) and imposing the conditions of Eqs. (7.48), we † So far, we cannot exclude that either kx or ky could be imaginary, so that Eqs. (7.47) contain one pair of hyperbolic functions with real argument. However, we will see soon that this is incompatible with the boundary conditions on the ideal conducting wall. Hyperbolic functions, on the other hand, cannot to be disregarded in open structures, like a pair of parallel conducting planes, in which case they represent ﬁelds which radiate in the direction along which the structure is open.

© 2006 by Taylor & Francis Group, LLC

RECTANGULAR WAVEGUIDES

241

Table 7.1 Field components of TE and TM modes in a rectangular waveguide. TE MODES nπ mπx nπy Ex = E o cos sin b a b mπ mπx nπy Ey = −Eo sin cos a a b Ez = 0 TM MODES Ex = ηTM Hy

nπ mπx nπy sin cos b a b mπ mπx nπy Hy = H o cos sin a a b Hx = −Ho

Ey = −ηTM Hx Ez = −

Ey ηTE Ex Hy = ηTE χ2 mπx nπy Hz = Eo cos cos jωμ a b Hx = −

χ2 mπx nπy Ho sin sin jω a b

Hz = 0

get immediately A = C = 0, and furthermore: mπ , a nπ sin ky b = 0 , (7.49) =⇒ ky = b where m and n are two integers. Accordingly, the longitudinal magnetic ﬁeld component of the generic transverse electric mode (TEm,n ) reads: nπy mπx cos , (7.50) Hz = H cos a b where H is an arbitrary constant. The other ﬁeld components, obtained from Eq. (7.50) through Eq. (7.3), are shown in Table 7.1, where, to simplify the notation for the transverse components, which are the most commonly used ones, we set H = Eo χ2 /(jωμ). The eigenvalue corresponding to Eq. (7.49) is given by: mπ 2 nπ 2 χ2m,n = + , (7.51) a b which lends itself to the simple geometrical interpretation of Figure 7.5. The corresponding cut-oﬀ frequency, according to Eq. (7.22), is: 1/2 m 2 n 2 1 + . (7.52) fcm,n = √ 2 μ a b sin kx a = 0

=⇒

kx =

Negative values for m and n should not be taken into account, since a change in sign does not aﬀect Eq. (7.52), nor Eq. (7.50). In the ﬁeld components shown in Table 7.1, changes in sign would not be signiﬁcant, as they can always be considered as part of the arbitrary constant Eo . Zero is a legitimate value

© 2006 by Taylor & Francis Group, LLC

242

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.5 Geometrical construction of the modal eigenvalues in a rectangular waveguide.

for either m or n, but not simultaneously for both, because in this case in fact the longitudinal component in Eq. (7.50) would become a constant, and consequently all the transverse components of E and H would vanish. This purely longitudinal magnetic ﬁeld, constant with respect to all coordinates, would be the magnetostatic ﬁeld of an inﬁnitely long rectangular solenoid, not a traveling guided wave. As for the TM modes, what we said for the TE modes remains valid until Eq. (7.47). The boundary condition on the wall is now Eq. (7.17), which once the variables are separated reads: X=0

for

x=0, x=a ,

Y =0

for

y=0, y=b .

(7.53)

Imposing these conditions on Eqs. (7.47) yields B = D = 0, and then Eqs. (7.49) hold again. So, the longitudinal electric ﬁeld component of the generic transverse mode TMm,n is: nπy mπx sin , (7.54) Ez = E sin a b where E is an arbitrary constant. The corresponding transverse ﬁeld components, which can be derived through Eq. (7.3), are also shown in Table 7.1, where, again to simplify the notation for the transverse components, we set E = −Ho χ2 /(jω). The eigenvalue corresponding to Eq. (7.54) is still given by Eq. (7.51), but now neither m nor n can take the zero value, because this would imply E z ≡ 0 in Eq. (7.54) and thus result in an identically zero ﬁeld. Negative values for m and n are also to be left out, for the same reasons as for TE modes. Eq. (7.51) holding for TM and for TE modes implies that, for m · n = 0, the TEm,n and TMm,n modes are degenerate. Any linear combination of two such modes is then a ﬁeld which not only satisﬁes the boundary conditions on the waveguide walls, but also depends exponentially on the longitudinal

© 2006 by Taylor & Francis Group, LLC

RECTANGULAR WAVEGUIDES

243

coordinate z and is characterized by the eigenvalue (7.51). Therefore, this ﬁeld in turn can be looked at as a mode of the waveguide. In Section 7.9 we will provide an example of a problem — losses in waveguides with nonideal metallic walls — where special attention has to be paid in the case of degenerate modes and of their linear combinations. As we said before, for TE modes Eq. (7.51) holds also for m = 0 or for n = 0. These modes are not degenerate with TM modes. Consequently, the smallest among all the values that are allowed for χ2m,n corresponds to m = 1 and n = 0, i.e., it belongs to the TE1,0 mode. This is, in conclusion, the fundamental mode of a rectangular waveguide. Letting m = 1 and n = 0 in Eq. (7.52), we get its cut-oﬀ frequency: fc =

c 2a

,

(7.55)

where c = (μ)−1/2 is the speed of light in the dielectric inside the waveguide. Eq. (7.55) shows that the distance a between the short walls of the waveguide equals a half wavelength (measured in the dielectric that ﬁlls the guide) at the cut-oﬀ frequency of the fundamental mode. This is a hint towards a physical interpretation of the “cut-oﬀ” phenomenon in terms of plane waves which impinge on the walls and are reﬂected on them, giving rise to standing waves in the waveguide. If the distance between two parallel metallic walls is smaller than one half of the plane-wave wavelength, then no standing wave resulting as a superposition of uniform plane waves can satisfy the condition of zero tangential ﬁeld on both walls, whose distance is shorter than the distance between two nodes of the electric ﬁeld, irrespectively of the direction of propagation of the two interfering traveling waves. The conditions on the two walls can be satisﬁed by the superposition of two evanescent waves whose attenuation vectors are parallel to the z axis and whose phase vectors (sat√ isfying |k| > ω μ) are orthogonal to the two walls. For these slow waves (see Chapter 4), two consecutive nodes in their interference pattern are less then λ/2 apart from one other. This interpretation reconﬁrms that the ﬁeld attenuation in the z direction, which cannot be caused by power dissipation since the structure is lossless by assumption, is indeed an evanescence. When, on the contrary, a half wavelength is shorter than the distance between the walls, say a, then two nodes of the electric ﬁeld at a distance a may indeed result from the superposition of two uniform plane waves, provided their phase vectors have a nonvanishing component in the z direction. The net result is a ﬁeld which propagates in the axial direction of the guide. Higher-order modes behave according to the same mechanism, diﬀering from each other in position and in number of the ﬁeld nodes in the interference pattern, which, for m = 0 and n = 0, results from reﬂections on the four walls. Examples of ﬁeld lines, for some of the simplest and yet most important modes of rectangular waveguides, are shown in Figure 7.6.

© 2006 by Taylor & Francis Group, LLC

244

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.6 Field lines of a set of signiﬁcant modes, in a rectangular waveguide.

7.8

Circular waveguides and coaxial cables

The subjects of this section are of technical interest and, at the same time, rather simple as far as calculations are concerned. The mathematical background, although not complicated, is much less dealt with, in standard curricula in electrical engineering and in physics, than that of the previous section. For this reason, Appendix D is devoted to outline it in some detail. Several contents of this section will be useful background points for very important subjects to be dealt with in Chapter 10. Consider an inﬁnitely long structure with a lossless dielectric either inside a circular-cylinder conductor (cylindrical waveguide, Figure 7.7), or between

© 2006 by Taylor & Francis Group, LLC

CIRCULAR WAVEGUIDES AND COAXIAL CABLES

245

Figure 7.7 A circular waveguide, and its reference frame.

Figure 7.8 A coaxial cable.

two circular-cylinder conductors with the same axis (coaxial cable, Figure 7.8). In either case, let us introduce a cylindrical coordinate system (r, ϕ, z), whose z axis is obviously the axis of the cylinders. The diﬀerential operator of Eq. (7.8) then reads: 1 ∂2 ∂2 1 ∂ 1 ∂ ∂ 1 ∂2 + 2 = . (7.56) ∇2t = 2 + r + 2 2 ∂r r ∂r r ∂ϕ r ∂r ∂r r ∂ϕ2 To separate the variables, let: F = R(r) φ(ϕ) ,

(7.57)

insert Eqs. (7.57) and (7.56) into Eq. (7.8), and then multiply both sides times r2 /Rφ. We obtain the following ODE’s, where ν 2 is the separation constant: φ = −ν 2 φ , 1 ν2 R + R + χ2m − 2 R = 0 . r r

(7.58) (7.59)

Eq. (7.58) is, once more, the harmonic equation, whose general integral can be written in either of the following forms: φ(ϕ)

© 2006 by Taylor & Francis Group, LLC

= G sin νϕ + G cos νϕ ,

(7.60)

246

WAVEGUIDES WITH CONDUCTING WALLS φ(ϕ)

= A ejνϕ + B e−jνϕ

,

(7.61)

where G , G (or A, B) are arbitrary constants. The ﬁrst form is more suitable if one wants to stress that the generic solution can be thought of as a superposition of two states of polarization, which correspond to G = 0 and to G = 0, respectively. The second form, on the other hand, represents φ(ϕ) as a superposition of two traveling waves which describe a helical path around the z axis, one in the clockwise sense (B = 0), the other one in the counter-clockwise sense (A = 0). This viewpoint becomes particularly attractive in the limit of geometrical optics, i.e., when a ray picture of propagation is associated with these two waves, as we shall see in Chapter 10. The two polarizations in Eq. (7.60) can then be looked at as standing waves which result from interference of two helical waves (Eq. (7.61)) of equal amplitudes. To be physically meaningful, Eq. (7.57) must be a single-valued function. Therefore, it must be: φ(ϕ + 2lπ) = φ(ϕ) , (7.62) where l can be any integer. Consequently, ν must be an integer. To stress this point, we will change the symbol, in agreement with standard conventions in mathematical handbooks and textbooks, and replace ν with n (= 0, 1, . . .). Negative values for n are not required, for the same reasons as in Section 7.7. Under the assumption χ2m = 0, i.e., if we leave aside the TEM mode, Eq. (7.59) is the Bessel equation of integer order ν = n. Its general integral can be expressed in several forms (see Appendix D). For the problems discussed in this section, the most convenient one is: R(r) = C Jn (χm r) + D Yn (χm r) ,

(7.63)

where Jn (x) and Yn (x) are, respectively, the Bessel function of the ﬁrst kind, of order n, and the Bessel function of the second kind (also called the Neumann function), of order n. Some examples are plotted in Figure D.1 and D.2 (see again Appendix D). Note in particular that, for any order n, Yn tends to −∞ for ν → 0. An inﬁnite value of the longitudinal ﬁeld component F is unphysical, so the term Yn (χm r) must be disregarded, i.e., we must set D = 0 in Eq. (7.63), whenever the z axis (where r = 0) belongs to the region where the e.m. ﬁeld is deﬁned. This occurs in cylindrical waveguides (Figure 7.7), but not in coaxial cables (Figure 7.8). From now on, it is convenient to deal separately with TE and TM modes. For the ﬁrst ones, the longitudinal ﬁeld component F = Hz must satisfy the boundary condition (7.18), which in the present case, keeping track of Eq. (7.57), reads: dR ≡ χm C Jn (χm a) = 0 , (7.64) dr r=a

where a is the radius of the cylindrical wall, and Jn (x) indicates the ﬁrst derivative of Jn (x) with respect to its argument. Let pn,m be the m-th zero of Jn (x). Tables of such zeros can be found in several handbooks. Usually, they

© 2006 by Taylor & Francis Group, LLC

CIRCULAR WAVEGUIDES AND COAXIAL CABLES

247

Table 7.2 Field components of TE and TM modes in a circular waveguide. ∇t = a ˆr

∂ 1 ∂ +a ˆϕ ∂r r ∂ϕ

TE MODES n sin nϕ Er = Eo Jn (χm r) − cos nϕ r cos nϕ Eϕ = Eo χm Jn (χm r) sin nϕ Ez = 0

Eϕ ηTE Er Hϕ = ηTE cos nϕ χ2 Hz = m Eo Jn (χm r) sin nϕ jωμ Hr = −

TM MODES n Jn (χm r) r

Er = ηTM Hϕ

Hr = H o

Eϕ = −ηTM Hr

Hϕ = Ho χm

Ez = −

χ2m Ho Jn (χm r) jω

cos nϕ sin nϕ

sin nϕ − cos nϕ cos nϕ Jn (χm r) sin nϕ

Hz = 0

are labeled also by a second integer, m = 1, 2, . . . (see, e.g., Abramowitz and Stegun, Eds., 1965). Then, the eigenvalue for the TEn,m mode is given by: pn,m , (7.65) a √ and the corresponding critical frequency is fcn,m = pn,m /(2πa μ). The longitudinal ﬁeld component can be written in the form: r cos nϕ Hz = H Jn pn,m (7.66) sin nϕ , a where H is an arbitrary constant. Eq. (7.66) stresses that, for n = 0, the two terms in Eq. (7.60) correspond, as we said before, to two distinct polarizations, giving two modes which are orthogonal (in the sense of Section 7.5) over the range 0 − 2π and must then be looked at as a pair of degenerate modes. The other ﬁeld components, obtained from Eq. (7.66) through Eq. (7.4), are shown in Table 7.2, which also shows the ∇t operator expressed in cylindrical coordinates. As in the previous section, to simplify the notation for the transverse components, which are the most commonly used ones, we set H = Eo χ2 /(jωμ). For the TM modes, the boundary condition of Eq. (7.17) reads, in the case at hand: (7.67) R(a) ≡ C Jn (χm a) = 0 . χn,m =

Let pn,m be the m-th zero of Jn (x). The eigenvalue of the TMn,m mode is then given by: pn,m , (7.68) χn,m = a

© 2006 by Taylor & Francis Group, LLC

248

WAVEGUIDES WITH CONDUCTING WALLS

where again a is the wall radius. The corresponding critical frequency is √ fcn,m = pn,m /(2πa μ), and the longitudinal ﬁeld component is: r cos nϕ Ez = E Jn pn,m , (7.69) sin nϕ a where E is an arbitrary constant. This expression, like Eq. (7.66), stresses that there are two orthogonal and degenerate polarizations for n = 0. The transverse ﬁeld components are also shown in Table 7.2, where again, for the usual reasons, we set E = −Ho χ2 /(jω). Among all the zeros of the functions Jn and of their derivatives Jn — incidentally, all of them are irrational numbers — the smallest one is p1,1 1.84. Hence, the fundamental mode of the circular guides is the TE1,1 mode, whose critical frequency is: fc =

1.84 √ 2πa μ

.

(7.70)

This is of the same order of magnitude as the cut-oﬀ frequency of the fundamental mode in a rectangular waveguide whose broad side equals in length the diameter of the circular waveguide, given by Eq. (7.55). This indicates that the physical nature of the cut-oﬀ phenomenon is not aﬀected by the waveguide geometry in a very signiﬁcant way. As for coaxial cables, we said before that in the presence of the inner conductor the r = 0 axis is out of the region where the ﬁeld is deﬁned, and this implies that the second term in Eq. (7.63) cannot be disregarded. As a matter of fact, it plays a fundamental role in the higher-order modes, but not in the fundamental mode, which, as the structure does not have a simply connected cross-section, is deﬁnitely the TEM mode, according to what we saw in Section 7.4. The TEM mode originates from the Laplace equation, whose solutions, as we said in Section 7.4, obey a well-known uniqueness theorem. We may exploit this fact, and derive the solution of Eq. (7.15) which corresponds to the TEM mode in a heuristic way, without any risk of loss in generality. Let us suppose ν = 0 in Eq. (7.58). This in turn gives φ = constant, and Eq. (7.59) becomes: 1 (7.71) R + R = 0 , r whose general integral is: R = C1 ln r + C2

,

(7.72)

where C1 , C2 are arbitrary constants. Let us set F = T in Eqs. (7.12). Then, we get the following expressions for the ﬁeld of the TEM mode: √ 1 ˆr e−jω μz , E = jω μ/ C1 a (7.73) r H = jω C1

© 2006 by Taylor & Francis Group, LLC

√ 1 a ˆϕ e−jω μz r

.

(7.74)

CIRCULAR WAVEGUIDES AND COAXIAL CABLES 249 √ Apart from the propagation factor exp(−jω μ z), Eq. (7.73) coincides with the electrostatic ﬁeld in a cylindrical capacitor where the per-unit-length √ charge is ±jω (2π) C1 μ on the outer and inner conductors, respectively. Eq. (7.74) coincides — again apart from the propagation factor — with the stationary magnetic ﬁeld in a cylindrical inductor where the longitudinal current intensity is ±2πω C1 in the inner and outer conductors, respectively. The wave impedance in the z direction, for the e.m. ﬁeld given by Eqs. (7.73) and (7.74), is η = μ/. This point may generate many comments, most of which will be made in Section 8.11. For what concerns higher-order modes in coaxial cables, the boundary conditions must be satisﬁed on two cylindrical surfaces, r = a and r = b, a and b being the radii of the outer and inner conductors, respectively. Consequently, one of the two arbitrary constants in the general integral in Eq. (7.63) can be expressed in terms of the other before one proceeds to ﬁnd the eigenvalues. For TE modes, the condition in Eq. (7.18) becomes dR/dr = 0 at r = a and r = b. Dividing side by side the two equations one gets in this way, we ﬁnd the following equation in the unknown χm : Jn (χm a) Y (χm a) = n Jn (χm b) Yn (χm b)

,

(7.75)

which is referred to as the characteristic equation of the TE modes. One can prove that it has a countable inﬁnity of discrete solutions. The corresponding longitudinal ﬁeld component can be expressed as:

cos nϕ Hz (r, ϕ) = H Yn (χm a) Jn (χm r) − Jn (χm a) Yn (χm r) , (7.76) sin nϕ where H is an arbitrary constant and, once again, two orthogonal polarizations for n = 0 are written explicitly. For TM modes, the boundary condition of Eq. (7.17), namely R(a) = R(b) = 0, leads to the following characteristic equation: Yn (χm a) Jn (χm a) = Jn (χm b) Yn (χm b)

,

and then to the following longitudinal ﬁeld component:

cos nϕ Ez (r, ϕ) = E Yn (χm a) Jn (χm r) − Jn (χm a) Yn (χm r) sin nϕ

(7.77)

,

(7.78)

which lends itself to the same comments as Eq. (7.76). It can be shown that, amongst all the solutions χm = 0 of Eqs. (7.75) and (7.77), the smallest one is the ﬁrst solution of Eq. (7.75) for n = 0. The corresponding TE0,1 mode is then the ﬁrst higher-order mode of the coaxial cable. Its eigenvalue lends itself to an approximation which leads to a physically enlightening expression for its critical frequency, namely: fc c/π(a + b) .

(7.79)

This corresponds to saying that at cut-oﬀ the wavelength is equal to the

© 2006 by Taylor & Francis Group, LLC

250

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.9 Field lines of a set of signiﬁcant modes, in a circular waveguide.

length of the mean circle between the conductors. Comments are then similar to those that were given on Eq. (7.55) in Section 7.7. Examples of ﬁeld lines, for some of the simplest and yet most important modes of circular waveguides, are shown in Figure 7.9. 7.9

Waveguides with nonideal walls

An ideal conducting wall is an abstraction, but nevertheless the theory of lossless waveguides is useful in practice. Most of its results apply well to real waveguides whose walls have a large but ﬁnite conductivity, σ ω. The only important quantity of practical interest which is absolutely impossible to evaluate by means of a lossless model is the power loss that a mode propagating in a real waveguide suﬀers because of dissipation, either in the nonideal conducting walls or in the imperfect dielectric. This inconvenience can be overcome using a procedure, to be described in this and in the following section, which is part of a broad-scope chapter of Mathematical Physics, known as “perturbation theory” (see, e.g., Morse and Feshbach, 1953). Let us make ﬁrst some general comment, which may turn out to be useful also in some of the following chapters. To ﬁnd an e.m. ﬁeld in a homogeneous dielectric surrounded by a good conductor in a rigorous way would require the following procedure: ﬁrst, to solve Maxwell’s equations in the dielectric and in the conductor; then, to apply continuity conditions to the tangential components at the interface between the two media. This is always cumbersome, and often impossible, as we may show with a surprisingly simple example. In a rectangular waveguide, continuity conditions should be imposed on four segments belonging to four diﬀerent coordinate lines. This problem has no analytical solution, as we will see again in the case of dielectric waveguides (Chapter 10). What we saw in Section 4.12 on plane waves in good conductors provides a

© 2006 by Taylor & Francis Group, LLC

WAVEGUIDES WITH NONIDEAL WALLS

251

good hint towards an approximate solution. Namely, if what we are actually interested in is mainly the ﬁeld in the dielectric, then we may characterize the conductor in terms of its wall impedance, that is to say in terms of the boundary condition it imposes on the ﬁeld at its surface. Indeed, one fundamental feature of a wall impedance is that it does not depend on the ﬁeld we are looking for in the dielectric. In a waveguide, it is not always simple to assess which value should be given to the surface impedance of walls which consist of a given material, as the geometrical shape can play an important role also. This statement will be clariﬁed by some means of examples in the next section. However, once a suitable wall impedance has been identiﬁed, then to ﬁnd the transmission modes of the waveguide means to solve an eigenvalue equation, similar to Eq. (7.8) but with some major changes. In fact, the boundary conditions are no longer Eqs. (7.17) and (7.18). As a consequence, the eigenvalues are not necessarily real and positive. Still, in most cases this new eigenvalue problem is very hard to solve rigorously. What is much simpler is to use perturbation theory, a procedure which yields a sequence of approximate solutions that converge towards the exact one. To do that, the basic requirement is to know, and to use as the starting point, the exact solution of another eigenvalue problem which is similar enough to the one that one cannot solve exactly. To ﬁnd a general criterion to assess what “similar enough” does actually mean, would require a discussion whose size and depth are well beyond the scope of this book. For the problem at hand, we will restrict ourselves to the following statement. In Section 4.12 we saw that the wall impedance of a good conductor tends to zero as σ → ∞. Accordingly, the exact solutions of a “similar problem” to start from will be, for us, the modes of the lossless waveguides that we studied in the previous sections. Furthermore, let us state that, for all practical purposes, attenuation constants of the modes above cut-oﬀ are obtained with suﬃcient accuracy as ﬁrst-order corrections to the eigenvalues of the lossless waveguide modes. This phrase means that calculation of attenuation constants requires only the knowledge of “zero order” ﬁelds, i.e., of the modes of the lossless guide. However, let us stress that the procedures to calculate ﬁrst-order corrections diﬀer drastically, depending on whether one is dealing with non-degenerate or degenerate modes. There are fundamental reasons behind this diﬀerence, which we will try to clarify by means of an example, later in this section. For a non-degenerate mode, we proceed as follows. We assume that the ﬁeld of the (m, n)-th mode of an ideal waveguide (Zw = 0, and lossless dielectric) is present in a waveguide with the same geometry, but with a ﬁnite wall impedance, Zw (later we will discuss also the case of a lossy dielectric). We calculate the per-unit-length active power which, under these conditions, would ﬂow into the walls. By deﬁnition of wall impedance, it equals |Htan m,n |2 d , (7.80) WP (m,n) = Re[Zw ] 2

© 2006 by Taylor & Francis Group, LLC

252

WAVEGUIDES WITH CONDUCTING WALLS

where is the contour of the waveguide cross-section, while the suﬃx “tan” denotes, as usual, the tangential component on the wall. The power lost in a piece of waveguide of inﬁnitesimal length dz, WP (m,n) dz, is proportional to the active power WT (m,n) ﬂowing through the waveguide cross-section (the “transmitted” power): ¯ tm,n |2 |H dS , (7.81) WT (m,n) = Zm,n 2 S where Zm,n is the wave impedance in the z direction for the (m, n) mode, ¯ tm,n is the transverse component of the magnetic ﬁeld. As the ﬁeld is a and H ¯ tan . ¯ t is proportional to H solution of homogeneous equations, H Proportionality between Eq. (7.80) and Eq. (7.81) entails that, as z increases, the transmitted power decreases exponentially. Hence the e.m. ﬁeld also decreases exponentially vs. z. Power being proportional to the squared modulus of the ﬁeld, the ﬁeld attenuation constant, αm,n , is one half of the power decay constant, so we ﬁnally ﬁnd: αm,n =

WP (m,n) 2 WT (m,n)

.

(7.82)

This result is accurate enough only if higher-order corrections are much smaller than Eq. (7.82). It can be shown that one necessary condition for this to occur is that Eq. (7.82) be much smaller than unity. Note that this requirement is violated as the (m, n) mode approaches its cut-oﬀ frequency, since then WT (m,n) goes to zero and Eq. (7.82) tends to inﬁnity. We may draw the conclusion that near cut-oﬀ it becomes demanding to calculate the attenuation constant accurately enough. Second-order corrections are, in general (see, e.g., Morse and Feshbach, 1953), much more cumbersome to calculate than ﬁrst-order ones, as they require summations which extend, at least in principle, over the whole inﬁnity of modes of the lossless waveguide. The reason why the ﬁrst-order procedure, as we described it, can be inadequate for degenerate modes is not too diﬃcult to explain at this stage. In fact, in each term of the summations involved in the second-order correction, the denominator contains the diﬀerence between the eigenvalue corresponding to the (m, n)-th mode which we are dealing with, and the eigenvalue of another mode. In the degenerate case, at least one among these diﬀerences vanishes, the corresponding term in the summation becomes inﬁnitely large, and it does not make sense any longer to call it a second-order correction. To avoid this, one has to make sure that the numerator of that term in the summation is also zero. In the case of the problem at hand, it can be shown that the numerator is of the type: ¯ tan m,n · H ¯∗ H tan q,r d . (7.83) WP (m,n;q,r) = Zw 2 Eq. (7.83) can be looked at as a measure of how much the two modes, (m, n) and (q, r), interact, i.e., loose their orthogonality, in the lossy structure. In

© 2006 by Taylor & Francis Group, LLC

WAVEGUIDES WITH NONIDEAL WALLS

253

precise terms, Eq. (7.83) is the per-unit-length complex power which ﬂows into the imperfect conductor because of the interplay between the two modes, i.e., is the diﬀerence between the total per-unit-length power loss and the sum of those which can be attributed to the two modes individually. If Eq. (7.83) is non-zero, then the two modes are not orthogonal. If, furthermore, they are degenerate, then one of them cannot propagate without being accompanied by the other. It becomes erroneous to calculate its loss as if it were traveling alone. The correct procedure for two degenerate modes, (m, n) and (q, r), is then the following one. First, one must ﬁnd two linear combinations of these modes, such that for them the integral in Eq. (7.83) is zero. They (see Section 7.5) are also modes of the ideal lossless guide, but, in contrast to the modes (m, n) and (q, r), they remain orthogonal in the lossy waveguide. One can then apply the procedure in Eq. (7.82) to each of these new modes. An example may help to clarify these points. Take the modes TE11 and TM11 in a rectangular waveguide, and suppose that Zw (to be discussed in more detail in the next section) is constant over the waveguide contour and ¯ , respectively, be ¯ and H can be taken out of the integral in Eq. (7.83). Let H the TE11 and the TM11 modal ﬁelds. From Table 7.1, setting all the arbitrary constants equal to one, as they are irrelevant in these calculations, we have: πx πy πy 1 πx ¯ = jβ11 π 1 sin πx cos πy a ˆ cos sin a ˆ cos a ˆz , + H x y + cos aχ211 b a b b a b a b (7.84) 1 πx πy πx πy jπk 1 ¯ = sin cos a ˆx − cos sin a ˆy H , (7.85) χ211 b a b a a b where: χ211

=

π 2 a

+

π 2

,

b

Calculating Eq. (7.83) as: WP = Z w

0

a

η=

μ

Hx Hx∗ dx + Zw

we ﬁnd: WP =

,

k=ω

0

b

√

μ .

Hy Hy∗ dy

Zw β11 kπ 2 (a − b) . 2 χ411 ab

,

(7.86)

(7.87)

(7.88)

Therefore, except for a square waveguide (a = b), the TE11 and TM11 modes are not orthogonal, but coupled to each other by the losses in the wall. To ﬁnd linear combinations of these modes for which the interaction term in Eq. (7.83) vanishes can be phrased as an eigenvalue problem, as it corresponds to diagonalizing the loss matrix: W WP , (7.89) W= WP W

© 2006 by Taylor & Francis Group, LLC

254

WAVEGUIDES WITH CONDUCTING WALLS

where W , W are Eq. (7.80) calculated for the TE11 and the TM11 modes, respectively, while WP is Eq. (7.88). The eigenvalues of the matrix in Eq. (7.89) are, clearly: 1 W + W ± (W − W )2 + WP2 . (7.90) λ1,2 = 2 2 These quantities have to be inserted in the numerator of Eq. (7.82) to obtain the correct values for the attenuation constants of the two orthogonalized modes. The magnitude of the error one makes by using Eq. (7.82) without accounting for the coupling between the degenerate modes can be calculated, as an exercise, using for Zw the value which was found in Section 4.11 for plane waves impinging on a ﬂat good conductor. The subject of wall impedance will be dealt with in more detail in the next section. To conclude this section, let us deal brieﬂy with loss due to an imperfect dielectric inside a waveguide. The concepts remain the same: one assumes that the modal ﬁelds of the lossless structure propagate in the lossy one, and evaluates the corresponding per-unit-length power dissipation. Dividing it by the transmitted power, and then by 2 to keep track of the relationship between ﬁeld amplitude and power, one ﬁnally gets the rate at which the ﬁeld decreases exponentially vs. z. Formulas, on the other hand, are much simpler than those of loss due to imperfect conductors. Indeed, the ﬁnal result turns out to be the same for TE and for TM modes, and also independent of whether the mode is degenerate or not. It reads: αd =

2

ω η 1 − (fc /f )2

,

(7.91)

where the notation c = −j has been adopted for the complex permittivity, η = μ/ is the intrinsic impedance of the dielectric in the lossless waveguide, and fc is the cut-oﬀ frequency of the mode we are dealing with. Necessary conditions for the validity of Eq. (7.91) are , and frequency not too close to the cut-oﬀ frequency.

7.10

On wall impedances

The right value for the surface impedance for good conducting walls of a waveguide can always be found using in the right way the same approach as in Section 4.11 for the similar problem of a plane wave impinging on the surface of a half-space good conductor. Namely, we will ﬁrst introduce in the Helmholtz equation those approximations which are derived from the deﬁnition of a good conductor (σ ω). We will see that then, as a consequence, the wave impedance along the normal to the conductor surface becomes independent of the conditions of incidence on the dielectric-conductor interface. Hence, its value may be taken as the wall impedance. We will see that often it is the same as that of Section 4.11; often, but, surprisingly perhaps, not always. Let us deal ﬁrst with a rectangular waveguide (Figure 7.10), and begin from

© 2006 by Taylor & Francis Group, LLC

ON WALL IMPEDANCES

255

Figure 7.10 Rectangular waveguide.

the walls orthogonal to the x axis. Separating variables as in Section 7.7, we get the following equation: d2 X = −(ω 2 μ − jωμσ + γ 2 − ky2 ) X ≡ −(kx2 − jωμσ) X dx2

.

(7.92)

In a lossless structure, kx = mπ/a can become large only if ω is large. Hence, good conductor means always ωμσ kx2 = |ω 2 μ + γ 2 − ky2 |. Consequently, the quantity in parenthesis on the right-hand side of Eq. (7.92) can be approximated as −jωμσ. (The same warnings as in Section 4.12 hold, for approximations where real numbers are compared with imaginary ones). From now on, everything ﬂows as in Section 4.12, and obviously the same occurs for the walls orthogonal to the y axis. We conclude that the wall impedance to be used in rectangular waveguide is Eq. (4.107), namely: 1/2 πμf . (7.93) Zw = (1 + j) σ For circular waveguides, the appropriate wall impedance is again given by Eq. (7.93). The proof, which requires separation of variables in cylindrical coordinates, is left to the reader as an exercise. One might then jump to the conclusion that Eq. (7.93) is valid in general, for any shape of the waveguide. This is erroneous. In most cases the error is numerically small, but not insigniﬁcant, as it may aﬀect, e.g., how the attenuation depends on frequency. As a counter-example, we will deal brieﬂy with waveguides with elliptical cross-section. We will show that in such a case the wall impedance Zw must be taken as a function of position on the waveguide contour, and furthermore, it takes diﬀerent values for TE and TM modes. The rather unexpected conclusion is that the wall impedance is anisotropic, although the wall is made of an isotropic conductor. To prove all these statements, we ﬁrst introduce (Figure 7.11) a system of elliptical-cylinder coordinates (ζ, η, z), related to cartesian coordinates (x, y, z) as follows: x = q cosh ζ cos η , y = q sinh ζ sin η , (7.94) where 2q is the distance between foci (which are the same for all the elliptical coordinate lines ζ = constant). Obviously, these foci are also those of the

© 2006 by Taylor & Francis Group, LLC

256

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.11 An elliptical coordinate system.

waveguide contour, which is then a coordinate line, ζ = ζ0 . Also, q is related to the ellipticity e, and to the lengths of the major and minor axes of the contour ellipse, 2a and 2b, respectively, as: 2 1/2 b q . (7.95) = e= 1− a a To ﬁnd the correct wall impedance, one must go from the beginning through the procedure we outlined before. Accordingly, the ﬁrst step is to solve Eq. (7.8) under the good-conductor approximation, σ ω. In elliptical coordinates, Eq. (7.8) reads (see Morse and Feshbach, 1953): ∂2F ∂2F + + χ2n q 2 (sinh2 ζ + sin2 η) F = 0 . ∂ζ 2 ∂η 2

(7.96)

We can ﬁnd its general integral (see, e.g., Abramowitz and Stegun, Eds., 1965), and then, similarly to what we did in Section 7.8, we can impose on it the restriction that it remains ﬁnite for ζ → 0. The net result can be written as:

(4) (7.97) F = M c(n) (ζ ; ρ) A ce(n) (η ; ρ) + B se(n) (η ; ρ) , where the symbols have the following meanings: • A, B = arbitrary constants; n = order of the functions; (4)

• M c(n) = radial Mathieu function of the fourth kind;

© 2006 by Taylor & Francis Group, LLC

ON WALL IMPEDANCES

257

• ce(n) = even angular Mathieu function of the ﬁrst kind; • se(n) = odd angular Mathieu function of the ﬁrst kind; • ρ = χ2n q 2 /4. The assumption of a good conductor implies χ2n −jωμσ in Eq. (7.96), and also, in most cases, |χn | a 1. Then, we can make use of an asymptotic ex(4) pression for M c(n) , which is valid for |χn q cosh ζ| → ∞. This is a necessary step in order to obtain wave impedances in the radial direction that are independent of the order n, i.e., independent of the mode under examination. Otherwise, we could not call it a wall impedance. The expressions one gets in this way are diﬀerent depending on whether we ˆz or Etan = Eη a ˆη . In detail, it can be shown (Falciasecca et take Etan = Ez a al., 1971; Lewin et al., 1977) that, for ζ = ζ0 , i.e., on the waveguide wall, it is: 1/2 πμf 1 Eη = (1 + j) , (7.98) ZPTE ≡ Hz σ (1 − e2 cos2 η)1/2 ZPTM

≡

Ez = (1 + j) Hη

πμf σ

1/2

(1 − e2 cos2 η)1/2

.

(7.99)

It is evident that these expressions are not the same as Eq. (7.93). Changes are due, from a mathematical viewpoint, to the fact that the metric coeﬃcient of the coordinate η changes along the coordinate line (ζ = ζ0 ) where we calculate partial derivatives in order to express the ﬁeld components. In a rectangular or circular waveguide, on the contrary, the metric coeﬃcient remains constant along the boundary. From a physical viewpoint, changes with respect to Eq. (7.93) can be explained by looking at the current ﬂow lines inside the conducting wall. We saw in Section 4.12 that, when a uniform plane wave impinges on a conducting half-space, the propagation vector of the transmitted wave is essentially perpendicular to planar interface between the two media. Consequently, the current density J = σE is parallel to the planar interface, at any depth inside the conductor. An equivalent surface current — which is a concept intimately related to that of surface impedance — is easy to deﬁne, given that all ﬂow lines are parallel. In a rectangular waveguide, if we neglect the corner eﬀects, the situation remains essentially the same. Current ﬂow lines in the walls are everywhere parallel to the wall surfaces. The same occurs again in a circular waveguide, where the current ﬂow lines have only an axial and an azimuthal component, always parallel to the surface. What we ﬁnd in an elliptical waveguide is completely diﬀerent. The current ﬂow lines, at the generic P point, have nonvanishing components along the coordinate elliptical line which passes through P , and consequently (see Figure 7.11) they are not parallel to the waveguide contour. At the same time, the distance between two given coordinate ellipses (which delimit a current ﬂow tube) changes as we move in the azimuthal direction. Also the Poynting vector inside the wall can be shown not to be perpendicular everywhere to the surface of the wall. All these comments justify why Eqs. (7.98) and (7.99)

© 2006 by Taylor & Francis Group, LLC

258

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.12 Schematic view of a helical waveguide.

are surprisingly diﬀerent from the wall impedances we have found so far. In particular, we note that: • none of the wall impedances in an elliptical waveguide is the same as Eq. (7.93); • the wall impedance is anisotropic, since Eqs. (7.98) and (7.99) are not the same; • all the impedances on the elliptical wall change as we move on the waveguide contour. The diﬀerence between Eqs. (7.98) and (7.99), on one hand, and Eq. (7.93), on the other hand, does not induce any relevant diﬀerence if the purpose is to ﬁnd the reﬂection coeﬃcient of an elliptical object (such as an antenna reﬂector, for example). On the contrary, with a “small” diﬀerence repeated many times, as in the case of a propagating mode which rebounds again and again on a waveguide wall, then the total diﬀerence on the attenuation constant can become quite signiﬁcant (Lewin et al., 1977). Experimental results on the frequency dependence of loss reconﬁrm that the correct values are those which one obtains with the procedure we outlined in this section. 7.11

Hybrid modes

In the two previous sections, the purpose of the wall impedance was to model more realistically structures which had been analyzed before with an idealized lossless model. The lossless model was a good starting point for the perturbation method. Consequently, we kept on using the terms TE and TM for the modes of waveguides with ohmic losses, although most of them were, strictly speaking, neither TE nor TM. Wall impedances are used in a slightly diﬀerent sense in modeling structures which do not have an ideal conducting wall, not even as a ﬁrst approximation. A typical example in this sense are the helical waveguides, structures (see Figure 7.12) whose walls are made of coiled wires or ribbons, the spacing between two coils being usually much smaller than the thickness of the wire or ribbon. These structures strongly discriminate the low attenuation of one family of modes whose currents ﬂow essentially along the wires or ribbons (the TE0,n family) from the high losses for all other modes. Their behavior can be studied using as a model an anisotropic wall, whose impedance, as a ﬁrst approximation, can be taken equal to zero in the azimuthal direction,

© 2006 by Taylor & Francis Group, LLC

HYBRID MODES

259

diﬀerent from zero in the longitudinal direction. Apart from the interest in their practical applications, which is nowadays much less than a few decades ago, what we wish to point out in this section is a concept, namely that, when wall impedances are ﬁnite, a complete set of solutions can no longer consist of only TE and TM modes. On the contrary, it is necessary to take into account other ﬁelds, which are called hybrid modes, are more complicated than TE and TM modes, and will play a dominant role in dielectric waveguides, to be studied in Chapter 10. To show this, let us start from the general expressions in Eqs. (7.11) and (7.12), and suppose that the anisotropic wall, with a circular cross-section of radius a, sets the following boundary conditions: Eϕ

= 0 ,

Ez

= ZHϕ

(7.100) ,

Z = 0 .

(7.101)

First of all, let us verify whether these conditions can both be satisﬁed by a TM ﬁeld, as given by Eq. (7.12). In cylindrical coordinates, after separation of variables, Eq. (7.100) becomes: ∂T ∝ R(a) φ (ϕ) = 0 , (7.102) ∂ϕ r=a

where, as usual, the prime indicates the derivative with respect to the argument, while Eq. (7.101) becomes: ∂T 2 2 . (7.103) (κ − γ ) T (a) = jω Z ∂r r=a

From what we saw in Section 7.8 it follows that, for an azimuthal order n = 0, φ (ϕ) is not identically zero. Therefore, Eq. (7.102) can be satisﬁed only for R(a) = 0. This condition implies T (a) = 0, and if we insert this into Eq. (7.103) we get R (a) = 0. Nontrivial solutions of the Bessel equation (7.59) do not have any value for their argument such that the function and its ﬁrst derivative, both vanish. Thus, Eqs. (7.102) and (7.103) are mutually incompatible, except for n = 0 (φ = constant). This means that TM0,n modes do exist in this type of waveguide, but, on the contrary, there are no TM modes that depend on the azimuthal coordinate. As for TE modes, we can show in a similar way that they cannot exist, except again for n = 0. In fact, assuming Ez ≡ 0 implies, via Eq. (7.101), Hϕ = 0, and because of Eq. (7.11) this entails: ∂L ∝ R(a) φ (ϕ) = 0 . (7.104) ∂ϕ r=a At the same time, the condition Eq. (7.100), again through Eq. (7.11), becomes: ∂L ∝ R (a) φ(ϕ) = 0 , (7.105) ∂r r=a and from now one can proceed as in the previous case.

© 2006 by Taylor & Francis Group, LLC

260

WAVEGUIDES WITH CONDUCTING WALLS

Evidently, TE0,n and TM0,n modes do not form a complete set, as the waveguide can certainly propagate ﬁelds that depend on ϕ. There must be further solutions of Maxwell’s equations. The decomposition theorem of Section 7.2 is quite general, and therefore one should be able to express any solution as a sum of a TE ﬁeld, of the type in Eq. (7.11), plus a TM ﬁeld, of the type in Eq. (7.12). Then, let L = A1 R1 φ1 and T = A2 R2 φ2 , with A1 and A2 arbitrary constants. The boundary conditions (7.100) and (7.101) become: nγ A2 R2 φ2 = 0 , (7.106) −A1 χ R1 φ1 + a jγ A1 n R1 φ1 , (7.107) −χ2 A2 R2 φ2 = Z jω A2 χ R2 φ2 + ωμa where primes indicate, as usual, derivatives with respect to arguments (note that the argument of Ri is χr, while that of φi is nϕ); furthermore, −χ2 = κ2 − γ 2 . From the system of Eqs. (7.106) and (7.107), we can: • eliminate one arbitrary constant Ai , showing that there is only one degree of freedom for amplitude and phase of each mode in this family; • derive an equation in implicit form which links the propagation constant γ with the angular frequency ω. It is usually referred to as the characteristic equation or the dispersion relationship of the structure. A simple way to ﬁnd it is to set the determinant of the system Eqs. (7.106) and (7.107), in the two unknowns Ai , equal to zero: jγZ R1 F −χ2 R2 − jωχ ZR2 F2 n 1 = a ωμaχ R1 F1 nγR2 F2

.

(7.108)

For this equation in the unknown χ to admit solutions, it is necessary that the dependence on ϕ be the same on both sides, i.e.: F1 F2 =± F1 F2

.

(7.109)

The two choices for the sign in Eq. (7.109) lead to two distinct sets of solutions for Eq. (7.108), resembling what we will see in Chapter 10. There is a need for a new terminology, as we cannot call TE and TM these new modes of helical guides. However, no general agreement on new names has been found throughout the technical literature. In Chapter 10, on the other hand, we will see that there is a general consensus on one nomenclature for the hybrid modes of cylindrical dielectric waveguides. 7.12

Further applications and suggested reading

We said at the beginning of this chapter that one of the reasons why we deal here with waveguides with conducting walls before tackling dielectric waveguides is because they came ﬁrst in time. In fact, waveguides surrounded

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

261

by metal walls were among the key ingredients in the development of radar, which was the ﬁrst example of practical use of microwaves, which were also referred to, at that time, as hyperfrequencies. A popular misconception is that those early developments were essentially empirical, based on a cut-and-try strategy. Those who want to see how advanced and sophisticated were, on the contrary, the ﬁrst major theoretical models in the area, should take at least a quick tour through the volumes known as the MIT Radiation Laboratory Series, published by McGraw-Hill in the late 1940’s. Readers who are familiar with the names of the most prominent theoretical physicists of that time will recognize some of them among the authors of these books. This was because during World War II most physicists in the United States became engaged in applied research for military aims — not only the very famous “Manhattan Project,” but also other developments, including of course radar. Accordingly, mathematical tools which were advanced, for that time, and were in use in ﬁelds such as quantum mechanics, were promptly and successfully transferred to modeling microwave devices. Because of this tidy and neat mathematical background, waveguides with ideal conducting walls are an ideal tutorial introduction to guided-wave propagation, much simpler than through any other similar structure. Furthermore, waveguides induce a very beneﬁcial step in the learning process of telecommunication engineers and other students with similar curricula. This step consists of the fact that students who have familiarized themselves with waveguides begin to look at ﬁelds, transmission lines, and electrical circuits, in a uniﬁed way, considering them as three ways to model the same physical reality. This cross-fertilization process can still be appreciated, as a major contribution, in at least one of the books in the previously quoted MIT Series, namely in Montgomery, Dicke and Purcell (1948). Readers who, at this stage, do not yet grasp the meaning and the importance of this comment, should come back to it after studying the next chapter. For decades, the word microwaves has had two meanings. It stood for a frequency range and, at the same time, for a set of technologies, which were then based mainly on waveguides. At present, this is no longer true. At those frequencies which are still called microwaves, up to an upper limit which is not sharply deﬁned, but is deﬁnitely higher than 10 GHz, and still growing, striplines dominate by far and large over waveguides. As we said when we introduced the TEM mode, striplines, although similar to those structures where the TEM mode propagates, are not easy to model analytically, as they require some approximation. For this reason, they have been deferred to the next chapter. While striplines were displacing waveguides from many applications, most of the electronics gradually became solid-state. Historically, microwave electronics consisted of special tubes which encompassed waveguiding structures, like sequences of cavities, helices, periodic combs, etc. The reader who wants to become familiar with modern microwave technology can be referred to textbooks like Baden Fuller (1990), Gardiol (1984) and Cronin (1995), and to more advanced books like Collin (1966/1991). Those who want to learn

© 2006 by Taylor & Francis Group, LLC

262

WAVEGUIDES WITH CONDUCTING WALLS

about classical microwave tubes, still in use for some applications, may refer to Pierce (1950). Conducting-wall waveguides have still some practical application, in addition to being pedagogically useful. Compared to striplines, their main advantage is that all the power they convey travels in a well deﬁned, closed region. This can be crucial in non-telecommunication applications, like industrial heating, domestic cooking, medical diagnosis and therapy, etc., where shielding humans from the ﬁeld is essential, to avoid unwanted interaction with biological tissues. Industrial, scientiﬁc and medical (ISM) applications of microwaves are widely described in the literature (e.g., see Roussy and Pearce, 1995, and references therein). Another case where waveguides are used, because they can convey very high powers with relatively low loss (especially if operated in the TE0n modes of either circular or helical waveguides), is plasma heating. Plasmas, as well known, have been and still are intensively explored, as the way towards generating energy by controlled nuclear fusion, and microwave heating is looked at as one of the possible ways to reach the threshold temperatures that are needed. The ﬁeld of plasma diagnostics also relies heavily on microwaves up to about 100 GHz. For these subjects, the reader is referred to Heald and Wharton (1978), Hutchinson (1987), Lochte-Holtgreven (1995) and references therein. Finally, some kinds of particle accelerators, for high-energy physics experiments, contain parts — typically, pipes where particles gain kinetic energy at the expenses of a radio-frequency or microwave ﬁeld — which can be looked at, and designed, as waveguiding structures. An excellent reading for those who are interested in these problems is Panofsky and Phillips (1962). Problems 7-1 A rectangular waveguide is interrupted on the plane z = L by a ﬂat ideal conductor, i.e., is a short-circuited waveguide. There are no sources between z = 0 and z = L. Assume that the waveguide wall and the dielectric inside it are all lossless. At a frequency such that only the TE10 mode is above its cut-oﬀ frequency, state whether the real and the imaginary parts of the ﬂux of the Poynting vector through the plane z = 0 vanish or not, specifying whether the answers depend on the length L. Repeat the problem assuming that the TE10 mode is below its cut-oﬀ frequency. 7-2 Derive two relationships which diﬀer from Eqs. (7.38) and (7.39) only because Etk and Htk are replaced by E∗tk and H∗tk , respectively. Comment on their physical meanings. 7-3 Estimate the error if Eq. (7.82) is used for a degenerate mode without accounting for its coupling with the other degenerate mode. Use for the wall impedance Zw the value found in Section 4.12 as the wave impedance for plane waves in a ﬂat conductor.

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

263

7-4 Return to Problem 1-17, and show that, in general, the ﬁeld of any TM mode can be derived from a purely longitudinal magnetic vector potential ˆz . A = Az a Derive the constant of proportionality between Az and Ez . Why is it diﬀerent from −jω ? 7-5 Return to Problem 1-18, and show that, in general, the ﬁeld of any TE mode can be derived from a purely longitudinal electric vector potential F = Fz a ˆz . Derive the constant of proportionality between Fz and Hz . Why is it diﬀerent from −jω ? 7-6 An important theorem of circuit theory, usually referred to as Foster’s theorem, states the following. Let X(ω) be the input reactance of a perfectly lossless two-terminal lumped-element network. Then, ∂X/∂ω > 0 for any network, where ω is, as usual, the angular frequency. a) Verify that the wave impedances of TE and TM modes below cut-oﬀ obey this rule, for waveguides of any shape and any size. b) For any mode of a rectangular waveguide, calculate the magnetic and electric energies stored in a semi-inﬁnite waveguide, z > 0, and, using the reactive power term in Poynting’s theorem, verify again that these results agree with Foster’s theorem. 7-7 Show that in the presence of loss or gain, i.e., when the real parts of the propagation constants of modes above cut-oﬀ are nonvanishing, two generic modes are not power-orthogonal, i.e., they do not satisfy Eq. (7.42). 7-8 Design an air-ﬁlled rectangular waveguide (i.e., calculate the lengths of its sides, a and b) so that the frequency f = 3 GHz is 1.3 times greater than the cut-oﬀ frequency of its fundamental mode, and 0.7 times the cut-oﬀ frequency of the ﬁrst higher-order mode. 7-9 A rectangular waveguide of width a = 23 mm and height b = 11.5 mm can propagate the TE10 mode at a frequency of 6 GHz, only if ﬁlled with a suitable dielectric. a) Find the range of values for the relative permittivity, r , for which the waveguide propagates only the fundamental mode. b) Study how the phase constant of the fundamental mode, as a function of frequency, varies when the permittivity of the medium varies within this range of values. 7-10 A rectangular waveguide whose dimensions are the same as in the previous problem, ﬁlled with air (r = 1), goes from a receiving antenna to a receiver for a total distance L = 5 m and conveys a signal whose spectrum is

© 2006 by Taylor & Francis Group, LLC

264

WAVEGUIDES WITH CONDUCTING WALLS

centered on f = 1.5 fc , where fc is the cut-oﬀ frequency of the fundamental mode. Evaluate the diﬀerential phase and group delays (i.e., the diﬀerences in velocities times the distance) over this length, for two frequencies which diﬀer from the center of the spectrum by ±5%. 7-11 A standard WR-15 rectangular waveguide (3.76 mm×1.88 mm) ends abruptly in an aperture. Assume that the wave impedance beyond the aperture equals the free-space intrinsic impedance. Find the reﬂection coeﬃcient for the TE10 mode at a frequency of 60 GHz, and the percentage of active power which is reﬂected back into the waveguide. (Comment. This problem is much simpler to solve after reading the following chapter.) 7-12 Two rectangular waveguides with diﬀerent major sides (a1 < a2 ) and equal minor sides (b1 = b2 ) are butt-jointed in the z = 0 plane. Waveguide 1 propagates the TE10 mode only, which travels in the positive z direction and impinges on the plane of discontinuity. Find which modes of waveguide 2 are excited, and which are their amplitudes, if a) the two waveguides are aligned in a corner (x = 0); b) they are aligned at their centers (the point x = a1 /2 of waveguide 1 is the same as the point x = a2 /2 of waveguide 2). (Hint. Expand an arc of cosine function into a set of harmonic functions with a diﬀerent period. Make use of orthogonality.) 7-13 Repeat the previous problem for two rectangular waveguides with equal major sides (a1 = a2 ) and diﬀerent minor sides (b1 < b2 ). Waveguide 1 propagates the TE10 mode only. a) The two waveguides are aligned in a corner (y = 0). b) They are aligned at their centers (the point y = b1 /2 of waveguide 1 is the same as the point y = b2 /2 of waveguide 2). 7-14

Consider a lossless circular waveguide.

a) Making use of Bessel function identities and integrals given in Appendix D, prove explicitly that its modes are orthogonal. b) Repeat the same problem for the modes of a coaxial cable. 7-15 Find the modes of a semicircular-waveguide, and those of a waveguide whose cross-section is a quarter of a circle, always assuming that the walls are ideal electric conductors. 7-16 Starting from the wave equation in a good conductor, and using separation of variables in cylindrical coordinates, show that the wall impedance

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

265

given by Eq. (7.93) is applicable when dealing with circular waveguides and coaxial cables. 7-17 Consider a circular waveguide with a low-loss wall, and use Eq. (7.93) for the wall impedance. a) Show that the perturbation formula for the attenuation, Eq. (7.80), indicates that the theoretical losses of the TE0,m modes are monotonically decreasing functions of the frequency, although the wall impedance increases in proportion to the square root of the frequency. b) Provide a physical explanation for this result. (Hint. Study the current density of these modes in the walls, as a function of frequency, for a constant transmitted power.) c) Why none of the modes of a rectangular waveguide behaves in the same way ? 7-18 A circular waveguide whose modes are all below their cut-oﬀ frequencies may be used as an attenuator in a coaxial cable: the internal conductor is removed over a certain length, and short-circuited to the external one at the two extremes of that length. a) Neglecting backward waves, i.e., assuming a purely exponential z-dependence for the TE11 mode below cut-oﬀ, write a “calibration chart” of an attenuator whose diameter is 20 mm and whose length is 50 mm, i.e., its attenuation (in decibels) vs. frequency, from dc to the cut-oﬀ frequency of the TE11 mode. b) In these calculations, all the structures have been assumed to be lossless. Consequently, attenuation in the device cannot be caused by power dissipation. Where does the lost power go ? Is it rigorous to accept, under these circumstances, the previous assumption that backward waves are negligible ? 7-19 In coaxial cables, the diameters of the inner and outer conductors which are in current use, as standards, in the design of the so-called Type N connectors, are 0.120 inches and 0.276 inches, respectively. Assuming that the dielectric between them has the same permittivity as air, evaluate (approximately) the uppermost frequency at which only the TEM mode propagates in the cable. Discuss how this cut-oﬀ frequency scales vs. relative permittivity of the dielectric. 7-20 A so-called microwave Y -junction is made of three identical rectangular waveguides, as sketched in Figure 3.4. Assume that they are lossless, and that only the TE10 mode is above cut-oﬀ. When waveguide 1 is connected to a generator, then the two waves at the outputs of waveguides 2 and 3 are equal in amplitude and in phase. By reciprocity, when two waves of equal amplitudes and phases are sent into waveguides 2 and 3, a wave in the TE10 mode emerges

© 2006 by Taylor & Francis Group, LLC

266

WAVEGUIDES WITH CONDUCTING WALLS

from waveguide 1. Using symmetry arguments (see Section 3.9), discuss what happens when: a) two waves of equal amplitude but phase-shifted by π radians are sent into waveguides 2 and 3; b) one wave is sent into waveguide 2 whereas no wave is sent into waveguide 3.

© 2006 by Taylor & Francis Group, LLC

References Abramowitz, M. and Stegun, I.A. (Eds.) (1965) Handbook of Mathematical Functions. Dover, New York. Baden Fuller, A.J. (1990) Microwaves: An Introduction to Microwave Theory and Techniques. 3rd ed., Pergamon, Oxford, New York. Collin, R.E. (1966) Foundations for Microwave Engineering. McGraw-Hill Physical and Quantum Electronics Series, McGraw-Hill, New York. Collin, R.E. (1991) Field Theory of Guided Waves. IEEE Press, New York. Cronin, N.J. (1995) Microwave and Optical Waveguides. Institute of Physics, Bristol, Philadelphia. Dicke, R.H. and Wittke, J.P. (1960) Introduction to Quantum Mechanics. AddisonWesley, Reading, MA. Falciasecca, G., Someda, C.G. and Valdoni, F. (1971) “Wall impedances and application to long-distance waveguides”. Alta Frequenza, 40, 426–434. Gardiol, F.E. (1984) Introduction to Microwaves. Artech House, Dedham, MA. Heald, M.A. and Wharton, C.B. (1978) Plasma Diagnostics with Microwaves. R.E. Krieger Pub. Co., Huntington, New York. Hutchinson, I.H. (1987) Principles of Plasma Diagnostics. Cambridge University Press, Cambridge, UK. Lewin, L., Chang, D.C. and Kuester, E.F. (1977) Electromagnetic Waves and Curved Structures. P. Peregrinus, Stevenage, UK. Lochte-Holtgreven, W. (1995) Plasma Diagnostics. American Vacuum Society Classics, AIP Press, New York. Montgomery, C.G., Dicke, R.H. and Purcell, E.M. (Eds.) (1948) Principles of Microwave Circuits. McGraw-Hill, New York. Reprinted by Peter Peregrinus, London, for IEE (1987). Morse, P.M. and Feshbach, H. (1953) Methods of Theoretical Physics. McGraw-Hill, New York. Panofsky, W.K.H. and Phillips, M. (1962) Classical Electricity and Magnetism. 2nd ed., Addison-Wesley, Reading, MA. Pierce, J.R. (1950) Traveling-Wave Tubes. Bell Telephone Laboratories Series, Van Nostrand, New York. Roussy, M.G. and Pearce, J.A. (1995) Foundations and Industrial Applications of Microwave and Radio Frequency Fields: Physical and Chemical Processes. Wiley, New York.

267 © 2006 by Taylor & Francis Group, LLC

CHAPTER 8

Waves on transmission lines 8.1

Introduction

We will use the phrase “transmission line” for any electrical system which, in one geometrical dimension, i.e., along the axis labeled z, extends signiﬁcantly more than along the others, and more than the wavelength. When we talk of “common” or “ordinary” transmission lines, they consist of two or more conductors, whose cross-sections, orthogonal to the preferential z direction, are separated by dielectrics. Simple examples are parallel wire pairs, and coaxial cables. More sophisticated examples are provided by microwave stripline circuits (see Section 8.11), the main diﬀerence coming from the fact that often their dielectric is not homogeneous. When the conductors are connected through lumped∗ passive components (resistors, capacitors, inductors, transformers), and fed by lumped voltage or current generators, then usually they can convey an electric current at arbitrarily low frequencies. This is the reason why the classical electricalengineering approach to transmission lines starts oﬀ as an extension of lumpedcircuit theory, i.e., begins by postulating Ohm’s law and Kirchhoﬀ’s laws. We will take it for granted that the readers know them from basic physics classes, and follow this standard approach in the ﬁrst part of the present chapter. We will shortly be able to show that waves may propagate along these lines. They are expressed in terms of voltages and currents, so they must be looked at as being electromagnetic waves. Our subsequent task will be to reconcile these waves on transmission lines with those e.m. waves that we have studied so far. This will be done also due to the fact that we have used very frequently, for our waves, one of the most typical concepts of circuit theory, that of impedance. Once the link is established between common transmission lines and what we saw in the previous chapters, then the transmission-line formalism, developed for actual lines, becomes suitable, and extremely useful, as an abstract model for transmission/reﬂection phenomena with any kind of e.m. wave, no matter whether free or guided. These subjects will be dealt with in the last sections of this chapter. ∗ “Lumped” means that the geometrical size of an element is negligible with respect to wavelength, at all frequencies of interest, so that none of the electrical quantities which can be deﬁned on that element, like voltage and current, experiences a signiﬁcant phase delay from one point to another within that element. 269 © 2006 by Taylor & Francis Group, LLC

270

WAVES ON TRANSMISSION LINES

Figure 8.1 Equivalent circuit for a section of inﬁnitesimal length of a uniform transmission line.

8.2

Uniform transmission lines

To retain simplicity, we will deal only with transmission lines whose characteristics are constant along the direction of propagation, z. We will refer to them as uniform transmission lines. In any transmission line, energy can be stored along its length under electric and under magnetic form. Some energy can also be dissipated along the line, under the form of ohmic losses in non-ideal conductors and in non-ideal dielectrics. Intuitively, then, it should be possible to model the electromagnetic behavior of any line in terms of distributed parameters. Magnetic energy stored in an inﬁnitesimal length dz can be thought of as being stored in an inﬁnitesimal inductance. Ohmic losses in the conductors, over an inﬁnitesimal length, can be attributed to an inﬁnitesimal resistance. The same can be said for the electric energy, stored in an inﬁnitesimal capacitance, and for losses in the dielectric, modeled as an inﬁnitesimal shunt conductance. In a uniform transmission line, these properties must, by deﬁnition, be constant in the z direction. Consequently, we can assume, as a starting point, that the suitable model for an inﬁnitesimal length dz is the “equivalent circuit” shown in Figure 8.1, where the parameters r, l, c, g are independent of z. We may then deﬁne a per-unit-length series impedance Z (expressed in ohms per meter), as the series of the per-unit-length inductance l (expressed in henrys per meter) and the per-unit-length resistance r (expressed in ohms per meter), Z = r + jωl. Similarly, we deﬁne a per-unit-length shunt admittance Y = g + jωc (expressed in inverse ohms per meter) as the parallel of the per-unit-length capacitance c (expressed in farads per meter) and the perunit-length conductance g (expressed in inverse ohms per meter). Note that this presentation of the “equivalent circuit” is rather phenomenological: at this stage, it should be regarded as a postulate. However, it will be justiﬁed in more physical terms in Section 8.10. We postpone this justiﬁcation because it is less urgent than the basic results which can be derived from the circuit itself. Applying Ohm’s law and Kirchhoﬀ’s laws (for voltages around a closed path and for currents in a node, respectively) for time-harmonic voltages and currents at steady state, from the circuit of Figure 8.1 we get the following

© 2006 by Taylor & Francis Group, LLC

UNIFORM TRANSMISSION LINES

271

diﬀerential equations, where the voltage V and the current I are unknown functions of the longitudinal coordinate z: dV = −ZI dz

,

dI = −Y V dz

,

(8.1)

where: Z = (r + jωl) , Y = (g + jωc) . (8.2) Eqs. (8.1) are frequently referred to as the telegraphist’s equations. Diﬀerentiating them again with respect to z and replacing in one of them the ﬁrst derivative by means of the other one, we get: d2 I d2 V = ZY V , = ZY I . (8.3) 2 dz dz 2 Eqs. (8.3) are sometimes referred to as the telephonist’s equations. They can also be looked at as one-dimensional forms of the Helmholtz equation. The general integral of either one can be expressed in terms of exponential functions: V (z) = V+ e−Γz + V− eΓz

I(z) = I+ e−Γz + I− eΓz

,

,

(8.4)

where:

Γ = + (r + jωl) (g + jωc) . (8.5) Since Γ is in general complex, each term in Eqs. (8.4) looks like a wave, traveling either in the forward or in the backward direction along the z axis. Consequently, Γ is referred to as the propagation constant of the transmission line. Current symbols are Γ = α + jβ, with α and β real, referred to as the attenuation constant and the phase constant of the transmission line. Eqs. (8.4) cannot be looked as the general integral of the system in Eqs. (8.1), as long as the four arbitrary constants are independent. Indeed, the order of the system in Eqs. (8.3) being higher, it has also spurious solutions, i.e., solutions which do not satisfy Eqs. (8.1). To restrict ourselves to the actual solutions, we must introduce Eqs. (8.4) into Eqs. (8.1). So, we ﬁnd that the constants must be related among themselves as follows: I+ =

V+ ZC

The quantity:

,

I− = −

V− ZC

.

(8.6)

" ZC =

(r + jωl) (g + jωc)

,

(8.7)

is referred to as the characteristic impedance of the transmission line. In general, it is complex, and this indicates that, in a uniform transmission line which is not further speciﬁed, a voltage and a current waves which travel in the same direction† are not in phase, at the same distance from the origin along the line, z. † To avoid an unnecessary cumbersome language, on many occasions throughout this chapter we will use the word “direction” with the meaning of “oriented direction,” or “sense”.

© 2006 by Taylor & Francis Group, LLC

272

WAVES ON TRANSMISSION LINES

The quantity: V (z) I ∗ (z) , (8.8) 2 where the star denotes as usual the complex conjugate, is the complex power across the generic cross-section of the line, ﬂowing in the positive direction of the z axis (i.e., active power is actually ﬂowing in the positive z direction if the real part of Eq. (8.8) is positive, in the negative direction if Re[Pc ] < 0). Inserting Eq. (8.6) into Eq. (8.8), we see that Pc consists of a term as∗ /2) exp(−2αz), plus a term associsociated with the forward wave, (V+ I+ ∗ ated with the backward wave, (V− I− /2) exp(2αz), plus two “cross” terms, ∗ ∗ /2) exp(−j2βz) and (V− I+ /2) exp(j2βz). (V+ I− The remaining two arbitrary constants in the expressions for voltage and current have to be determined by two boundary conditions. Normally, these are imposed by a generator at one end of the line, and a load at the other end. The rest of this chapter will be devoted mainly to how these boundary conditions aﬀect voltage and current along the entire transmission line. Pc (z) =

8.3

Impedance transformation along a transmission line

The ratio between voltage and current, at any value of z along the transmission line, deﬁnes an input impedance: Z(z) =

V+ e−Γz + V− eΓz V (z) = ZC I(z) V+ e−Γz − V− eΓz

.

(8.9)

The input admittance is, obviously, the inverse of Eq. (8.9). Suppose that the line is terminated, at z = 0, on a linear, passive twoterminal network, whose impedance ZL = V (0)/I(0) we refer to as the load impedance. Incidentally, note that normally we will take the positive z direction from the generator to the load, then, choosing the load terminal section as the origin for the z coordinate implies that the line is in the region z ≤ 0. The input impedance at any distance z can be obtained, as a function of z and of the load impedance, imposing Z(0) = ZL in Eq. (8.9). The result is easily derived as: Z(z) = ZC

ZL cosh Γz − ZC sinh Γz ZC cosh Γz − ZL sinh Γz

.

(8.10)

Another useful quantity to deﬁne at any distance z from the load is the (voltage) reﬂection coeﬃcient, equal to the ratio between the backward and the forward voltage waves at that z. Due, once again, to Eq. (8.9), it is expressed by: I− eΓz V− eΓz =− = ρL e2Γz , (8.11) ρ(z) = −Γz V+ e I+ e−Γz There will be no danger of ambiguity, as the waves considered in this chapter can only travel along the line.

© 2006 by Taylor & Francis Group, LLC

LOSSLESS TRANSMISSION LINES

273

where ρL is the (voltage) reﬂection coeﬃcient at z = 0, i.e., at the load terminals. To simplify the expression of ρ(z) as a function of the input impedance or admittance, and vice-versa, it is convenient to deﬁne the so-called normalized impedance and admittance, namely: ζ(z) =

Z(z) ZC

,

y(z) =

1 Y (z) ZC = = YC Z(z) ζ(z)

.

(8.12)

The relationships between these adimensional quantities and the reﬂection coeﬃcient (adimensional also) are easily found by combining the deﬁnitions in Eqs. (8.12) and (8.11), and read: ζ(z) =

1 + ρ(z) 1 − ρ(z)

,

y(z) =

1 − ρ(z) 1 + ρ(z)

,

whose inverse is: ζ(z) − 1 Z(z) − ZC YC − Y (z) 1 − y(z) ρ(z) = = = = ζ(z) + 1 Z(z) + ZC 1 + y(z) YC + Y (z)

(8.13)

.

(8.14)

Let us stress how similar these expressions are to Fresnel’s reﬂection formulas, Eqs. (4.92) and (4.93), which we derived for plane wave reﬂection in Section 4.9. This similarity will induce us to elaborate a broad-scope abstract interpretation for the formalism of transmission lines. This point will be developed in Section 8.10, where, following the same line of thought, we will deﬁne normalized voltages and currents, quantities which obey Ohm’s law with normalized (i.e., adimensional) impedances or admittances and therefore are broadly arbitrary in their physical dimensions. Coming back to common transmission lines, the previous relationships enable us to express the complex power, Eq. (8.8), in terms of the reﬂection coeﬃcient, as: ∗ V+ I+ Pc (z) = . (8.15) 1 − ρ(z) ρ∗ (z) + ρ − ρ∗ 2 An exercise which is left to the reader is to separate real and imaginary parts in the last equation, i.e., to learn how to express active and reactive power in terms of the reﬂection coeﬃcient. This should be done twice, initially assuming that the characteristic impedance is real, and then removing this assumption. The diﬀerence will help the reader to appreciate why the content of the following two sections is very valuable in practice. 8.4

Lossless transmission lines

All the relationships that were deﬁned or derived in the previous sections become much simpler in the ideal case of a transmission line whose conductors and dielectrics are lossless. The reader may easily obtain, setting the per-unit-length series resistance r and shunt conductance g equal to zero, the following results for propagation constant, characteristic impedance, nor-

© 2006 by Taylor & Francis Group, LLC

274

WAVES ON TRANSMISSION LINES

malized impedance and admittance, reﬂection coeﬃcient, voltage and current along a lossless transmission line: √ Γ = jβ = jω lc , (8.16) l , (8.17) ZC = c ζ(z)

ζL − j tan βz ζL cos βz − j sin βz = cos βz − jζL sin βz 1 − jζL tan βz

=

1 yL − j tan βz = ζ(z) 1 − jyL tan βz

y(z) =

V− j2βz e V+

ρ(z)

= ρL ej2βz =

V (z)

= V+ e−jβz + V− ejβz V+ −jβz V− jβz e − e ZC ZC

I(z) =

,

,

(8.18) (8.19)

,

(8.20)

,

(8.21) .

(8.22)

For a lossless transmission line the characteristic impedance Eq. (8.17) is real, like the intrinsic impedance in a lossless medium (and consequently the wave impedance in any direction in real space, for a uniform plane wave). To insist on this analogy, the quantity λ = 2π/β is called the wavelength on the transmission line. Voltage and current of either a forward or a backward traveling wave are in phase, at any z. Another result which, as the reader may easily check, is peculiar to lossless transmission lines, and holds irrespectively of their loads, is that the reﬂection coeﬃcient, the normalized input impedance and the normalized input admittance are periodic functions of z, with a period equal to λ/2. The modulus of the reﬂection coeﬃcient is constant along a lossless line. Voltage and current are periodic functions, of period λ. As for the complex power, Eq. (8.15) shows that along a lossless line its real part (i.e., the active power): Pa =

∗ V+ I+ 1 − ρ(z) ρ∗ (z) 2

,

(8.23)

is constant vs. z, while the coeﬃcient of its imaginary part (i.e., the reactive power): Q(z) = −j

∗ V+ I+ [ρ − ρ∗ ] , 2

(8.24)

varies sinusoidally, with a period equal to λ/2. These properties will be exploited to develop a graphical tool, the Smith chart, to be introduced in Section 8.7.

© 2006 by Taylor & Francis Group, LLC

LOW-LOSS TRANSMISSION LINES 8.5

275

Low-loss transmission lines

Strictly lossless transmission lines are not physical entities, in the same sense as the strictly lossless media of the previous chapters. In the real world, losses are inevitable. However, with very rare and special exceptions, like attenuators, transmission lines are of practical interest only if their losses are small enough, at least over a suitable frequency range. Consequently, the lossless model studied in the previous section may be looked at as a “zero-order” approximation for real lines, similarly to what we did for conducting-wall waveguides in Chapter 7. In fact, we will see soon that, for lossy lines, exact solutions are often unnecessarily cumbersome. We will show then how ﬁrst-order approximations for real lines can be derived by elaborating on the lossless model. The phrase “small loss,” still ambiguous up to this stage, will be used from now on to designate precisely the cases where these ﬁrst-order approximations are acceptable. As we said before, the propagation constant Eq. (8.5) is, in general, a complex quantity, Γ = α+jβ. One may always start from the following expression: Γ2 = (r + jωl) (g + jωc) = (rg − ω 2 lc) + jω(rc + lg) ,

(8.25)

and separate real and imaginary parts, obtaining: α2 − β 2

= rg − ω 2 lc ,

2αβ

= ω(rc + lg) .

(8.26)

Solving this as a system of two equations in two unknowns, we get: 1 α = √ (r2 + ω 2 l2 ) (g 2 + ω 2 c2 ) − (ω 2 lc − rg) , 2 1 β = √ (r2 + ω 2 l2 ) (g 2 + ω 2 c2 ) + (ω 2 lc − rg) . 2

(8.27)

As anticipated, these expressions are complicated, compared with Eq. (8.16), but become remarkably simpler for small losses, i.e., for r/ωl 1, g/ωc 1. Indeed, truncated binomial expansions for the square roots (neglecting terms of order rg/ω 2 lc and smaller) give: g r c l r g ∼ + = YC + ZC α = , 2 l 2 c 2 2 √ g 2 1r ∼ − β = ω lc 1 + . (8.28) 8 ωl ωc From Eq. (8.7) in a similar way we ﬁnd: g r 3g 2 rg r2 + j − − + ZC ∼ 1+ = ZC 8ω 2 l2 8ω 2 c2 4ω 2 lc 2ωc 2ωl

,

(8.29)

= 1/YC is the characteristic impedance of a lossless line having the where ZC same per-unit-length inductance and capacitance as the actual line.

© 2006 by Taylor & Francis Group, LLC

276

WAVES ON TRANSMISSION LINES

Note that, at ﬁrst order in r/ωl and g/ωc, the phase constant remains the same as for an ideal lossless line. The same occurs for the real part of the characteristic impedance. Accordingly, ﬁrst-order eﬀects are only a non-zero attenuation constant, and a reactive part in the characteristic impedance. As we said in Section 7.9, one of the most important results of perturbation theory in general (see, e.g., Morse and Feshbach, 1953, or Dicke and Wittke, 1960) is that n-th order corrections for an eigenvalue can be calculated by means of the n − 1-th order eigenvectors. What this means in our case is that (exactly like in Section 7.9) we can compute the ﬁrst-order attenuation constant from the zero-order voltages and currents, i.e., from what we found for lossless lines in the previous section. In our case, this statement is easy to prove. Insert the lossless-line voltage and current forward waves (the ﬁrst terms in Eqs. (8.21) and (8.22), respectively) into the following formulas, which express the per-unit-length power loss, Pl (in watts per meter) for a forward traveling wave, and the active power P (in watts) carried by that wave through the generic section of the line: 1 , (8.30) r |I+ (z)|2 + g |V+ (z)|2 Pl = 2 P

=

Y |V+ (z)|2 |I+ (z)|2 ZC = C 2 2

.

(8.31)

On a lossy line, where forward voltage and current decay like e−αz , active power has to vary like P (z) = P0 e−2αz , and power loss over an inﬁnitesimal length is then dP = −Pl dz = −2α P (z) dz. Then, from Eqs. (8.30) and (8.31), the attenuation constant turns out to be: gZC rY , (8.32) α= C + 2 2 which is the same as Eq. (8.28), as we intended to prove. 8.6

Partially standing waves

In this section we will refer to lossless lines, unless the opposite is explicitly stated. However, most of the results are not diﬃcult to extend to low-loss lines. The reader is encouraged to do that as an exercise. The general expression Eq. (8.4) for voltage along a lossless line can be rewritten as follows: V (z) = |V+ | ej(φ+ −βz) + |V− | ej(βz+φ− )

,

(8.33)

where φ+ and φ− are the phases of the complex numbers V+ and V− , respectively. Adding and subtracting |V− | e−jβz ejφ+ , Eq. (8.33) can be rewritten as:

. (8.34) V (z) = |V+ | − |V− | ej(φ+ −βz) + |V− | ej(βz+φ− ) + e−j(βz−φ+ ) The ﬁrst term on the right-hand side is a forward traveling wave, of ampli-

© 2006 by Taylor & Francis Group, LLC

PARTIALLY STANDING WAVES

277

tude |V+ | − |V− |. Comparison with Section 4.5 tells us that the second term is a pure standing wave, of maximum amplitude 2 |V− |. In practice, it occurs often that one can measure either voltage or current (or sometimes a ﬁeld related to these) on a transmission line by means of an instrument whose response is slow (i.e., it averages over a large number of periods, at the angular frequency ω) and quadratic (i.e., the output of the instrument is proportional to the modulus square of either voltage or current). In view of this, let us write explicitly:

(8.35) |V (z)|2 = |V+ |2 1 + |ρL |2 + 2 |V+ |2 |ρL | cos(2βz + φL ) , where |ρL | and φL are modulus and phase, respectively, of the load reﬂection coeﬃcient, ρL = ρ(0) = V− /V+ . The voltage modulus reaches its maximum value: (8.36) |V (zM )| = |V+ | + |V− | , in those sections where: 2β zM + φL = 2kπ

(k = 0, 1, 2, . . .) ,

(8.37)

and its minimum value: |V |m = |V+ | − |V− | ,

(8.38)

in those sections where: 2β zm + φL = (2k + 1) π

.

(8.39)

It is straightforward to check that the sections where the voltage modulus is maximum, Eq. (8.37), are also those where the current modulus is minimum, and vice-versa for the sections Eq. (8.39), since:

|I(z)|2 = |I+ |2 1 + |ρL |2 + 2 |I+ |2 |ρL | sin(2βz + φL ) . (8.40) Two adjacent minima (or maxima) are separated by a distance: (zM +1 − zM ) =

λ π = β 2

,

(8.41)

whereas the distance between any maximum and either one of the adjacent minima is λ/4. From now on, when using phrases like “voltage maximum” (or minimum) we will refer to the voltage modulus, without any danger of ambiguity. The same applies, of course, to the current. We might have derived the same results also from the following expressions: = |V+ | |1 + ρ(z)| ,

(8.42)

ZC |I(z)| = |V+ | |1 − ρ(z)| ,

(8.43)

|V (z)|

which lend themselves to the simple graphical interpretation shown in Figure 8.2, based on the fact that, as Eq. (8.20) shows, when z varies along a lossless line, ρ(z) describes a circle.

© 2006 by Taylor & Francis Group, LLC

278

WAVES ON TRANSMISSION LINES

Figure 8.2 Complex-plane illustration of the behavior of the quantities 1 ± ρ vs. distance, along a lossless transmission line.

The adimensional quantity: S=

1 + |ρ| |V |max |V+ | + |V− | = = |V |min |V+ | − |V− | 1 − |ρ|

,

(8.44)

is referred to as the voltage standing wave ratio, and frequently designated by its initials, VSWR. When a transmission line is closed on a passive load, i.e., on a two-terminal network such that, at any frequency of interest, active power ﬂowing into it is non-negative, then Eq. (8.21) shows that |ρ| ≤ 1, and consequently S ≥ 1. In the sections where voltage is maximum (and current is minimum), voltage and current are in phase, so the input impedance is real, and indeed it equals: Z(zmax ) =

|V (z)|max |V+ | + |V− | ZC = SZC = |I(z)|min |V+ | − |V− |

.

(8.45)

Similarly, in those sections where voltage is minimum (and current is maximum), voltage and current are again in phase, the input impedance is real, and equals: Z(zmin ) =

8.7

|V (z)|min |V+ | − |V− | ZC ZC = = |I(z)|max |V+ | + |V− | S

.

(8.46)

The Smith chart

In Section 8.3, we derived two formulas, Eq. (8.10) and its reciprocal, which enable us to calculate the input impedance and admittance in any section along a transmission line when the load impedance is speciﬁed. Those relationships are much more cumbersome than the corresponding transformation rule for the reﬂection coeﬃcient, Eq. (8.11). Reﬂection coeﬃcient and input impedance (or admittance) are related to each other by a one-to-one and onto mapping. Hence, it is self-explanatory why it is of interest to exploit this mapping and the simple rules which govern how ρ changes vs. z, to by-pass the cumbersome formulas for Z(z), Y (z).

© 2006 by Taylor & Francis Group, LLC

THE SMITH CHART

279

We derived in Section 8.3 the relationship between normalized input impedance (or admittance) and reﬂection coeﬃcient: ρ=

1−y ζ −1 = ζ +1 1+y

.

(8.47)

Relationships of the kind aτ η + bτ + cη + d = 0, where a, b, c, d are complex constants, and τ, η are two complex variables, are called bilinear transformations. It can be shown (Schwerdtfeger, 1961) that any bilinear transformation maps any circle of the τ complex plane into a circle on the η complex plane, and vice-versa. Furthermore, any bilinear transformation maps two mutually orthogonal pencils of circles of one of these complex planes into two mutually orthogonal pencils of circles on the other complex plane, and vice-versa. This statement can be extended to the case of two orthogonal sets of parallel straight lines, which can be looked at as degenerate pencils of circles, with base points at inﬁnity. In particular, this can be applied to the straight lines parallel to the real and to the imaginary axis of the ζ plane (or of the y plane), i.e., to the loci of constant real part of normalized impedance (or admittance), and of constant imaginary part of the normalized impedance (or admittance). In other words, we know a priori that these lines will be mapped by Eq. (8.47) into mutually orthogonal pencils of circles on the complex plane of ρ. To proceed to the calculations, let ζ = r + jx (y = g + jb), and ρ = u + jv, where r, x, g, b, u and v are all real.‡ Starting from: r + jx =

1 + (u + jv) 1+ρ = 1−ρ 1 − (u + jv)

,

(8.48)

and separating real and imaginary parts, we ﬁnd: r=

1 − (u2 + v 2 ) (1 − u)2 + v 2

,

x=

2v (1 − u)2 + v 2

These two equations can be rewritten as: 2 r 1 u− + v2 = 1+r (1 + r)2

,

.

(8.49)

(8.50)

and:

2 1 1 (u − 1)2 + v − = 2 . (8.51) x x Eq. (8.50) reconﬁrms that, on the ρ plane, the loci r = constant are circles. r , 0 and its radius is Each of them is centered at a point of coordinates 1+r given by 1/(1 + r) (see Figure 8.3). Similarly, Eq. (8.51) reconﬁrms that the ‡ Several textbooks on transmission lines use the symbols z and x for the normalized impedance and the normalized reactance, respectively. At the same time, they use one of these symbols to denote the longitudinal coordinate along the line. To avoid ambiguity, we call z the longitudinal coordinate, ζ the normalized impedance. Throughout this section, x means the normalized reactance. Elsewhere, it may indicate a transverse coordinate. The text will always make the symbol unambiguous.

© 2006 by Taylor & Francis Group, LLC

280

WAVES ON TRANSMISSION LINES

Figure 8.3 The constant-resistance loci, on the complex plane of the reﬂection coefﬁcient.

Figure 8.4 The constant-reactance loci, on the complex plane of the reﬂection coefﬁcient.

loci x = constant on the ρ plane are circles, each of which is centered at a point of coordinates (1, 1/x) and has a radius given by 1/|x| (see Figure 8.4). Their sets are indeed two orthogonal pencils. Each pencil is parabolic, i.e., instead of having two distinct base point as the more common elliptic pencils, it has one double base point (the same for both pencils). This base point is located at (1,0). “Double” means that all the circles belonging to the same pencil not only pass through that point, but also have the same tangent in that point. For the r = constant circles, their tangent in common is the line u = 1, parallel to the imaginary axis. For the x = constant circles, the tangent in common is the real axis, v = 0. As we saw before, for passive loads we have |ρ| ≤ 1, and correspondingly, r ≥ 0 in Eq. (8.50). Consequently, the region of the ρ plane where passive loads are mapped is the interior of the circle |ρ| = 1, including the circle itself, which is the locus of the lossless loads. The plot of the r = constant circles and of the portions of the x = constant which belong to this part of the ρ plane is usually referred to as the Smith impedance chart, and is shown in Figure 8.5. An equivalent chart in terms of admittance is easy to build, noting that

© 2006 by Taylor & Francis Group, LLC

THE SMITH CHART

281

70 (+ jX /Z

45

1.4

1.2

1.0

0.8

55

1.6

1.8

0.6 60

2.0

0.5

06 0. 44 0. 0 14

4

0.

0.0 5

25 0.4

0.2

20 3.0

0.6

9 0.2 30

4.0

15

5.0

0.2

IND UCT IVE

20 10 0.25 0.26 0.24 0.27 0.23 0.25 0.24 0.26 0.23 0.27 REFLECTION COEFFICIENT IN DEG REES LE OF ANG ISSION COEFFICIENT IN TRANSM DEGR LE OF EES ANG

8

0.

0.6

10

0.1

0.4

20

0.2

50

20

10

5.0

4.0

3.0

2.0

1.8

1.6

1.4

1.2

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

1.0

50

0.1

50

RESISTANCE COMPONENT (R/Zo), OR CONDUCTANCE COMPONENT (G/Yo) 0.2

20

0.4

10

-10 1.0

E IV CT DU IN

2.0

-4

5

0.14 -80 -4 0

0.15 0.35

0.13

TR S. RF S. A A W. L. W. TT N P L L EN SM EA O O . . C K SS [ SS C [dB O O (C dB O ] EF EF O ] EF F, F, NS F E P T. or P) I

2

0.12

0.37

40

-70

-1

06

0

0.0

1

0.4

0.4

0.39

0.38

RADIALLY SCALED PARAMETERS

∞ 40 30 0

10 20

1 0.9

5

0.8

3

15 2

0.7

4

0.6

2.5

10

3

4

0.5

0.4

5

2

8 6

0.3

7

1.8

6 8

0.2

9

1.6

5 10

0.1

1.4

4 12

3 14

0.05

1.2 1.1 1 2 20 0.01

1

15

TOWARD LOAD —> 10 7 5

1 1 30 ∞ 0

0.1

0 0

1.1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0 1

0.99

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

CENTER 1

1.1

1.2

1.3 1.4

0.4

0.6

1.2

1.3 0.95

1.4

0.8

1.5

0.9

WAVELE 0.49 NGTH S TOW ARD 0.0 0.49 AD WA 160 0 1. Let ρ = 1/ρ: in this way, the region outside the circle |ρ| = 1 is mapped onto the region inside that circle, and vice-versa. y(ρ) =

© 2006 by Taylor & Francis Group, LLC

282

WAVES ON TRANSMISSION LINES

Consequently, plots over Figure 8.5 can be used to represent active loads, provided one reads that plane as that of the inverse of the reﬂection coeﬃcient. Many commercially available modern instruments for testing active electronic devices, typically microwave transistors, encompass a function which allows passing from the ρ plane to the ρ−1 plane on their polar display, by simply pushing a button. 8.8

Remote measurement of the load impedance

An elegant and useful application of the theory that we developed in this chapter comes from a problem that is frequently encountered in actual telecommunication systems. It occurs often that an unknown load impedance, located at the end of a transmission line, is diﬃcult to access. For example, this is typical with antennas installed on the top of buildings, metallic towers, etc., where it is diﬃcult to carry the necessary measuring instrumentation and to operate it in a reliable way. Finding the unknown load impedance from data measured at a distance, along the transmission line, is a remarkable improvement. We saw in Section 8.3 that if the reﬂection coeﬃcient is known, in modulus and phase, at a known distance z from the load, then, in principle, it is straightforward to calculate the load impedance or admittance (normalized to the characteristic impedance of the line). We will show here that there is a rather simple experimental recipe to implement this theory. The required instrument is referred to as a voltage standing wave indicator (VSWI), and consists of: a voltage probe that can be moved in a controlled way along the transmission line (usually, it is inserted in a movable mount equipped with a calibrated vernier scale); an RF peak voltmeter (an instrument which measures the modulus of the RF voltage, when connected to the probe); ﬁnally, a known load impedance, normally a short circuit. The ﬁrst step is to ﬁnd the modulus of the reﬂection coeﬃcient, which we can calculate when we know the VSWR. Moving then the probe along the line, one looks for a maximum, and measures |V |max by means of the RF peak voltmeter. Next, one looks for an adjacent minimum, and measures |V |min . Assuming, at least during an initial phase, that the line is lossless, we ﬁnd: |ρ| = |ρL | =

|V |max − |V− |min S−1 = S+1 |V |max + |V− |min

.

(8.53)

If we measure the distance between two consecutive minima (or maxima) (equal to λ/2) by means of the sliding probe, and if the distance between the load and the n-th voltage maximum on the line, zL , is also known, then in principle it is easy to calculate the phase of the load reﬂection coeﬃcient ϕL = arg(ρL ) as follows: ϕL = 2β(zL − n λ/2) = 2π(2zL /λ − n)

.

(8.54)

In practice, however, this procedure is unsuitable, because the result would be aﬀected by a large error bar. The ﬁrst uncertainty is encountered in the

© 2006 by Taylor & Francis Group, LLC

REMOTE MEASUREMENT OF THE LOAD IMPEDANCE

283

|V|

a

b c

βz

Figure 8.6 Typical patterns of partially standing waves. a) VSWR = 1. b) VSWR = 2. c) VSWR = ∞.

location of a maximum. Indeed, in any partially standing wave, minima (of the modulus, see Figure 8.6) are sharper than maxima. Therefore, locating a minimum is more accurate than locating a maximum. This simply implies that ±π is added to Eq. (8.54). However, another more serious reason why this modiﬁed procedure is not yet adequate is because, as well known, an unknown quantity measured as the diﬀerence between two quantities of similar magnitude is inevitably aﬀected by an enlarged relative error bar. In the present case this inconvenience may be extremely signiﬁcant, because quite often it is extremely impractical to measure geometrically the distance z L . For example, this occurs whenever a line twists or bends, between the load and the measuring equipment. All these inconveniences are avoided if one measures ϕL by comparison with a reference standing wave, namely the wave which one ﬁnds on the same line when the unknown load is replaced by a known impedance. To be a good choice, this known load must yield a large VSWR, with sharply deﬁned voltage minima. An excellent combination of this requirement and simplicity is a short circuit. Incidentally, note that an open circuit would be equally good in principle, but in practice it is not, because at high frequency a transmission line whose remote terminals are open radiates, similar to an antenna. If a line is terminated on a short circuit, then the distance between the n-th voltage minimum and the load terminals is z¯SC = nλ/2. If we call z¯L the abscissa of the n-th voltage minimum when the line is terminated on the unknown load, then the unknown load phase is simply: zSC − z¯L ) ± π ϕL = 2β(¯

.

(8.55)

To perform the tests that were described in this section, there are commercially available VSWI’s which encompass a slotted line, i.e., a section of either coaxial cable or rectangular waveguide, whose outer wall is slotted in the longitudinal direction, parallel to the current ﬂow in the wall itself. The slot accommodates a movable voltage probe.

© 2006 by Taylor & Francis Group, LLC

284

WAVES ON TRANSMISSION LINES

Figure 8.7 A uniform transmission line, connecting a generator and a load.

8.9

Impedance matching

Normally, a transmission line of ﬁnite length is connected to a generator at one end (the left side in Figure 8.7), and to a load at the other hand. Let ZG be the internal impedance of the generator (either in series to an ideal voltage generator whose open-circuit voltage is VG , or in parallel to an ideal current generator whose short-circuit current is IG = VG /ZG ). In general, there are reﬂected waves on the line. As we saw in the previous sections, and also in some of the previous chapters, reﬂected waves are responsible for unwanted eﬀects. First, the power delivered to the load becomes smaller than that carried by the forward wave. Second, keeping constant the active power delivered to the load P , the maximum voltage along the line can be expressed as: (8.56) |VM | = 2ZC P S , where ZC is, as usual, the characteristic impedance of the line, and S is the VSWR. The larger the reﬂected waves, the larger S, and, as consequence, |VM |. This must be accounted for, to avoid electrical discharges along the line. Last but not least, in general the reﬂection is frequency dependent, causing signal distortion in the case of ﬁnite spectral width. This intuition is reconﬁrmed by evaluating the voltage transfer function of a lossless transmission line of length l, i.e., the ratio between output and input voltages: H(jω) =

ZL VL = Vi ZL cos βl + jZC sin βl

.

(8.57)

It is easy to verify that in general Eq. (8.57) does not satisfy the distortionless transmission conditions (|H(jω)| = 1, H(jω) ∝ ω). However, if ZL = ZC , i.e., when there are no reﬂections, then H(jω) = e−jβl and the signal does not undergo any distortion. An impedance matching device is a passive (whenever possible, lossless) two-port network, whose purpose is to eliminate reﬂections at one of its ports, when the other one is connected to a given load (Figure 8.8). The purpose of this section is to illustrate three of the most commonly used impedance matching networks, which are based on lossless transmission lines. From the elementary theory of electrical circuits, we know that the active

© 2006 by Taylor & Francis Group, LLC

IMPEDANCE MATCHING

285

Figure 8.8 Deﬁnition of an impedance-matching network.

jXG

-jXG

RG

ZL

RG

VG

Figure 8.9 Power matching between a given generator and its load.

power delivered by a generator of given internal impedance, ZG = RG + jXG , reaches its maximum value when the load connected to the generator terminals, Zi (see Figure 8.9) equals the complex conjugate of ZG , i.e., for Zi = RG − jXG . This is referred to as the power matching condition. The complex power delivered by the generator under these circumstances is easily calculated from the deﬁnition of complex power (e.g., Eq. (8.8)). For a powermatched load it is I = VG /(2RG ), so the active power delivered to the load is: |VG |2 , (8.58) Pa = 8RG and the reactive power simply equals Pa times the ratio (−XG /RG ). Eq. (8.58) is referred to as the generator’s available power. The power matching condition, which is of paramount importance to electrical engineers involved in power systems, in general is not the best answer if we refer to signal transmission problems. In fact, power matching implies that the imaginary parts of the generator and load impedances have opposite signs. Because of the diﬀerent frequency dependence of reactances of opposite signs, their absolute values can be equal just at one frequency, not over a ﬁnite bandwidth, except for the case where they are equal to zero. Therefore, in general power matching entails signal distortion. The right recipe in a telecommunication context is not to look for the absolute maximum of the power delivered to the load, but for a conditional maximum, namely, under the restriction that it must be frequency-independent (at least over a narrow bandwidth). Again from elementary circuit theory, it can be shown that this

© 2006 by Taylor & Francis Group, LLC

286

WAVES ON TRANSMISSION LINES

corresponds to: Zi = ZG , (8.59) which is usually referred to as the uniformity matching condition. Note, however, that in the following we will restrict ourselves to lossless transmission lines, or at most to low-loss lines. It will be taken for granted that their characteristic impedance ZC is real. As the loads, and often the generators also, will be matched to this real ZC , the distinction between power matching and uniformity matching disappears. The purpose of an impedance-matching network is then to provide an input impedance which satisﬁes two independent requirements, if expressed in real numbers: real part equal to ZC , imaginary part equal to zero. We may then expect that the design of an impedance matching network should involve (at least) two real variables. All the following examples reconﬁrm this expectation. However, we will see that the physical nature of these variables changes completely as we pass from one network to another. 8.9.1

The quarter-wavelength transformer

In Section 4.10 we showed that, for plane waves, a lossless dielectric layer, whose thickness equals a quarter of the wavelength, can in fact cancel the Fresnel reﬂection at the interface between two semi-inﬁnite lossless media, provided its intrinsic impedance is given a suitable value. The Smith chart of Section 8.7, while on one hand it simpliﬁes the understanding of what we did in Section 4.10, on the other hand it suggests that the same results apply to transmission lines, provided the load impedance is real. Indeed, when a lossless line whose length equals an odd integer multiple of λ/4 is connected to a generic load impedance ZL , its input impedance equals: Zi =

2 ZC ZL

.

(8.60)

As the characteristic impedance of the lossless quarter-wavelength line, Z C , is real, Eq. (8.60) cannot be made equal to a real number (the characteristic impedance of the line to be matched) unless ZL is itself real. Using a , would mean introducing lossy quarter-wavelength line, with a complex ZC an unwanted attenuation. A suitable solution to the problem can be found if we simply accept that the quarter-wavelength line be inserted at a suitable distance d from the load (Figure 8.10), rather than connected directly to its terminals. Indeed, we know from Section 8.3 and the following ones that, as we move along the line, we ﬁnd periodically sections where the input impedance is real. They are either maximum-voltage sections, where:

Z(z) = SZC

,

(8.61)

,

(8.62)

or minimum-voltage sections, where: Z(z) = ZC /S

© 2006 by Taylor & Francis Group, LLC

IMPEDANCE MATCHING

287

Figure 8.10 Load matching by means of a quarter-wave line.

where S is the VSWR. On the Smith chart, these sections correspond to the intersections of the circle |ρ(z)| = constant, i.e., the locus of the reﬂection coeﬃcient along the line, with the real axis of the ρ plane, i.e., the locus of real impedances. In the ﬁrst case, combining Eqs. (8.60) and (8.61), the recipe for the intrinsic impedance of the quarter-wave line is found to be: √ = Z(z) ZC = ZC S , (8.63) ZC while in the second case it is found to be: √ = Z(z) ZC = ZC / S ZC

.

(8.64)

These formulas (even better if re-written in terms of normalized impedances, /ZC ) are strictly reminiscent of how one calculates the transforming ratio, ZC n : 1, for an ideal transformer to convert a resistive impedance Ro into an other resistive impedance Ri , n = Ri /Ro . For this reason, in the current terminology of telecommunications, lossless lines whose lengths equal an odd integer multiple of λ/4 are often referred to as quarter-wavelength transformers. Note, however, that the analogy disappears when the loads are not purely resistive. Indeed, the sign of the imaginary part of the impedance does not change as one goes through an ideal transformer, while, on the other hand, Eq. (8.60) converts an inductive load into a capacitive input impedance, and vice-versa. We said in the ﬁrst part of this section that we expected impedance-matching network designs to be based on two real parameters. In the case at hand, the two parameters are the characteristic impedance of the quarter-wave section, and its distance from the load, which equals either: d(S) = (λ/2) (ϕL /2π) + m(λ/2)

,

(8.65)

in the case of Eq. (8.63), and: d(1/S) = d(S) ± λ/4

,

(8.66)

in the case of Eq. (8.64), where ϕL is the phase angle of the load reﬂection coeﬃcient, and m = 0, 1, 2, . . .. As for the choice between the solutions in Eqs. (8.63) and (8.64), a suitable criterion is to make the line between the

© 2006 by Taylor & Francis Group, LLC

288

WAVES ON TRANSMISSION LINES

Figure 8.11 Single-stub impedance-matching network.

load and the λ/4-section as short as possible. Indeed, there are reﬂected waves on this piece of line, and from Eq. (8.64) one may check that the deviation from distortionless transmission grows as the line length grows. Sometimes, however, other more practical criteria become dominant with respect to this. For example, in a coaxial cable to diminish the characteristic impedance is much easier than to increase it, as discussed in one of the problems at the end of this chapter. 8.9.2

Impedance matching with one or two stubs

From the Smith chart, it is easy to check that if a lossless line is either shortcircuited, or open-circuited at its end, then its input impedance (and admittance, of course) is purely reactive, irrespectively of its length. Therefore, it is of great interest to study whether any load can be impedance-matched to a lossless transmission line by inserting along the line a pure reactance, either in parallel or in series. In principle, the answer is positive for all these combinations. In practice, however, open-circuited lines are disregarded, because they inevitably radiate an active power from their terminals, thus violating the requirement of being lossless. Insertions in series are also disregarded, normally, being more cumbersome than those in parallel. So, whenever we mention impedance-matching stubs, one should refer automatically to shortcircuited line sections in parallel to the main line, as shown schematically in Figure 8.11. From the general expression in Eq. (8.18), with ζL = 0, we ﬁnd immediately that the normalized input impedance of a stub is: ζi = 1/yi = j tan βl

,

(8.67)

where l is the stub length. Thus, any reactance, from −∞ to +∞, can be realized in this way. Therefore, any normalized admittance lying on the g = 1 circle of the −ρ plane (i.e., of the Smith admittance chart, see Figure 8.12), can be matched to a lossless transmission line. It is understood that normalization is with respect to the characteristic admittance of this line, YC = 1/ZC . So, the only requirement to be satisﬁed by what is located between the load and

© 2006 by Taylor & Francis Group, LLC

IMPEDANCE MATCHING

289

Figure 8.12 Illustration of the operation principle of a single-stub impedance matching network, on the Smith admittance chart.

the stub is to transform the point which represents the load into a point on the circle g = 1. Again from Figure 8.12 we see that one simple way to do that, applicable to any load, is to leave a suitable length of the main transmission line between the terminals of the load and those of the stub. Indeed, the circle |ρL | = constant (where ρL is the reﬂection coeﬃcient of the load) intersects the circle g = 1 at two points (called A and B in Figure 8.12). One of them is in the halfplane where b > 0, b being the normalized susceptance; the other one is where b < 0. To yield impedance matching, the stub must provide a susceptance which is equal in modulus and opposite in sign to that seen along the main line. Therefore, one solution (point A) entails an inductive stub (with a length l < λ/4) and the other one (point B) entails a capacitive stub (with a length λ/2 > l > λ/4). Once again the design of this impedance-matching network relies on two real parameters, namely the length of the piece of line between the load and the stub (to be calculated as ϑ/(2πβ), where the angle ϑ is shown in Figure 8.12), and the length of the stub. Note that in the last lines it was taken somewhat for granted that the stub was made with a line with the same characteristic admittance as the main line. This is not necessary at all. When this is not the case, the only change to be made with respect to what we said is the way to calculate the length of the stub, as illustrated in detail in some of the following problems. Incidentally, the reader may verify as an exercise that designing this type of device (which is currently referred to as a single-stub impedance matching

© 2006 by Taylor & Francis Group, LLC

290

WAVES ON TRANSMISSION LINES

Figure 8.13 Double-stub impedance-matching network.

network) using analytical tools, instead of the Smith chart, is possible but much more cumbersome. Another device which is of common use in practice is the double-stub impedance-matching network, schematically illustrated in Figure 8.13. We will call stub 1 the one which is connected to the same terminals as the load, stub 2 the one upstream towards the generator. Contrary to the single stub, distances along the main line are ﬁxed. The most commonly used values for the distance between the stubs are either λ/4, or λ/8, or 3λ/8. The two real variables exploited in the design of this network are the two lengths of the stubs. To outline how a double-stub network operates, note ﬁrst that the stub 2 must play the same role as the stub in the previously discussed single-stub network. Therefore, matching is possible only if the point which represents the input admittance on the main line, as seen from the terminals of the stub 2, lies on the circle g = 1 of the admittance Smith chart (see Figure 8.14). Consequently, as nothing can be modiﬁed on request between the stubs, the point which represents the input admittance immediately upstream of the stub 1 must lay on the curve where the circle g = 1 is mapped by the line section between the stubs. Figure 8.14 shows these curves for the three common cases of a distance equal to λ/4, to λ/8 and to 3λ/8. To avoid confusion, the rest of the procedure is illustrated in detail only for the case λ/4. The other ones can be completed by the reader as an exercise. The purpose of the stub 1 is to modify the input admittance, from the given load value to a value lying on the transformed circle, designated as the A-circle. A stub in parallel does not aﬀect the real part of the admittance. Therefore, we pass from the point yL to the point on the A-circle along a constant-g line. This intersects the A-circles at two points, indicating that, for a load like the one shown in Figure 8.14, there are two solutions. For each of them, the susceptance of the stub 1 is found as the diﬀerence between that at the point on the A-circle, and that of the load. The susceptance of the stub 2 is then computed in the same way as for a single-stub matching network. The main advantage of the double stub is to avoid sliding contacts along the

© 2006 by Taylor & Francis Group, LLC

TRANSMISSION-LINE EQUATIONS

291

Figure 8.14 Illustration of the operation principle of a double-stub impedance matching network, on the Smith admittance chart.

main line. Its main drawback is that it cannot match all possible loads. Indeed, if the normalized load admittance is located in the region whose g = constant lines do not intersect the transformed circle (to be clear, the A-circle in the previous example), then the procedure does not yield any solution. The main reason why double stubs spaced by either λ/8 or 3λ/8 are often preferred to those spaced by λ/4 is because their “non-match regions” are smaller. The further choice between them is a trade-oﬀ: the ﬁrst one is shorter, thus giving less distortion, but two closely spaced stubs may tend to interact with each other in a rather unpredictable way. A powerful solution which makes it possible to match any load is to use a triple stub. In such a case, there are several possible solutions for a generic load to match. The consequent main drawback, usually, is that the device is designed (or made empirically) using a time-consuming cut-and-try procedure. Double stubs could also be explained and designed in terms of line equations, but the Smith chart turns out to be much simpler, as the reader can verify as an exercise. 8.10

Transmission-line equations: An alternative derivation

In this section we will show how the equations which govern voltage and current waves along transmission lines can be derived from Maxwell’s equations, provided some suitable assumptions are satisﬁed. The ﬁrst purpose is to show that what we have seen so far in this chapter, although derived in a rather phenomenological way from the circuit of Figure 8.1, is not in contradiction

© 2006 by Taylor & Francis Group, LLC

292

WAVES ON TRANSMISSION LINES

with the axiomatic approach followed in the rest of the book. Another purpose, may be obscure now, but hopefully not so at the end of this section, is to show that transmission lines provide useful and elegant equivalent models for propagation phenomena in other guiding structures, like the conducting-wall waveguides of the previous chapter, and dielectric waveguides of Chapter 10. Let us consider a lossless homogeneous cylindrical structure whose crosssection is not a simply connected domain, so that it can propagate the TEM mode. Introduce rectangular coordinates where z is the longitudinal axis, and write homogeneous Maxwell’s equations by components. Under the assumptions that the ﬁeld is transverse electromagnetic, i.e.: Ez ≡ 0 ,

Hz ≡ 0 ,

(8.68)

the x and y components read: ∂Ey = jωμ Hx ∂z

,

∂Ex = −jωμ Hy ∂z

,

(8.69)

∂Hy ∂Hx = −jω Ex , = jω Ey . (8.70) ∂z ∂z The z components are omitted as they are not required in the following steps. The ﬁrst of Eqs. (8.69) and the second of Eqs. (8.70) form a system with two unknowns, Ey and Hx . The other two equations form another system (in Ex and Hy ), independent of the previous one and formally identical with it, except for a sign. Therefore, it is enough for us to investigate in detail only one of these systems, say the second one. All conclusions could then be repeated for the other one. Being familiar with separation of variables, we suggest to start oﬀ with the following ansatz: Ex (x, y, z)

= V (z) Ex (x, y)

,

Hy (x, y, z)

= I(z) Hy (x, y)

.

(8.71)

These two equations must be read as if they encompass two arbitrary multiplicative constants, A and B, i.e., they can always be replaced by:

1 Ex (x, y) , Ex = V (z) Ex (x, y) = A V (z) A

1 Hy (x, y) . (8.72) Hy = I (z) Hy (x, y) = B I(z) B Note that even the physical dimensions of A and B are arbitrary. Therefore, the physical dimensions of the quantities called V , I, Ex and Hy in Eqs. (8.71) are not yet deﬁned at this stage. Introducing Eqs. (8.71) into (8.69) and (8.70), we get immediately: dV Ex (x, y) dz

© 2006 by Taylor & Francis Group, LLC

= −jωμ I Hy (x, y) ,

TRANSMISSION-LINE EQUATIONS

293

Figure 8.15 Kirchhoﬀ ’s laws applied to the equivalent circuit are identical to Maxwell’s equations for TEM propagation.

dI Hy (x, y) = −jω V Ex (x, y) . (8.73) dz We saw in Section 7.2 that the wave impedance of a TEM ﬁeld traveling in the ±z direction is ±η. The same cannot be said, in general, for a superposition of TEM ﬁelds that propagate in both directions along the z axis. Indeed, on several occasions (in Chapter 4 and later) we saw that superimposing two waves is a delicate operation, from the viewpoint of the wave impedance. However, there was one point which was invariably true in all the cases that we examined: namely, the wave impedance did not depend on the transverse coordinates x and y. Consequently, we will assume now that the two functions of the transverse coordinates Ex (x, y) and Hy (x, y) in Eqs. (8.71) are proportional, i.e.: (8.74) Ex (x, y) = N Hy (x, y) . Clearly, the proportionality constant N is related to the arbitrary constants A and B of the previous comment. In particular, there is nothing wrong in letting Ex = Hy in Eq. (8.72), thus obtaining N = A/B. Because of Eq. (8.74), Eqs. (8.73) become: dV dz

= −

dI dz

= −jω N V

jωμ I N

, .

(8.75)

If we give the arbitrary constants in Eqs. (8.72) physical dimensions such that V and I get, respectively, dimensions of a voltage and of a current, then Eqs. (8.75) become identical with the telegraphist equations, Eqs. (8.1). Other choices for A and B do not entail any substantial change. The degrees of freedom in A and B simply mean that the circuit of Figure 8.15, where Eqs. (8.75) are satisﬁed as Kirchhoﬀ’s laws, should be looked at as an equivalent circuit for any TEM mode propagation, not necessarily as a phenomenological model, as it might have appeared after the initial derivation of the equations from the circuit shown in Figure 8.1. The circuit shown in Figure 8.15 may look as a simpliﬁed (or idealized) version of that of Figure 8.1. A natural question to ask then, is whether passing

© 2006 by Taylor & Francis Group, LLC

294

WAVES ON TRANSMISSION LINES

from a lossless medium to a lossy one in Maxwell’s equations, Eqs. (8.69) and (8.70), could yield an equivalent circuit like that of Figure 8.1. It is simple to verify that, if we replace by c , only the second of Eqs. (8.75) changes, so that its equivalent circuit encompasses now an inﬁnitesimal parallel conductance: g dz = Re [jωc N ] dz = N σ dz

.

(8.76)

No series resistance emerges in this way. The reason is because a series resistance is, strictly speaking, incompatible with the TEM assumption. Indeed, a current ﬂowing through conductors with ﬁnite conductivity necessarily entails a nonvanishing longitudinal component for the electric ﬁeld. As we saw at the beginning of this chapter, we may elaborate on Eqs. (8.75) so that each unknown, V (z) and I(z), is isolated. The order of the diﬀerential equations increases. Incidentally, note that this procedure is essentially the same as the one used in Chapter 1 to pass from Maxwell’s equations to the Helmholtz equation. The equations we get in this way are: d2 I d2 V 2 = −ω μ V , = −ω 2 μ I . (8.77) dz 2 dz 2 Eqs. (8.77) are essentially the same as the telephonist equations (8.3). Note that the arbitrary constant N of Eqs. (8.75) disappeared in the last equations. Their general integral is of the type: V = V+ e−jβz + V− e+jβz

,

(8.78)

where V+ and V− are arbitrary constants, and: √ β = ω μ .

(8.79)

Hence, for waves propagating along a TEM transmission line, phase constants are always equal to the intrinsic phase constant of the homogeneous lossless dielectric medium between the (ideal) conductors, independently of the geometrical shape of the structure cross-section. From Eq. (8.78) and the ﬁrst of Eqs. (8.75) we ﬁnd: I=

Nβ Nβ V+ e−jβz − V− ejβz = I+ e−jβz + I− ejβz ωμ ωμ

where now we let:

I± = ±V± N

V± =± μ (η/N )

.

,

(8.80)

(8.81)

The physical dimensions of V and I are still undeﬁned. So are those of the arbitrary constant N , and consequently of η/N , which appears in Eq. (8.81). Independently of the choice of N , this quantity plays the physical role of an impedance, and is proportional to the intrinsic impedance of the dielectric, η = μ/. Let us illustrate, by means of some examples, how one may exploit the still arbitrary constants so as to get a set of enlightening results. 1) If we give N the physical dimensions of an impedance, i.e.: [N ] = Ohm ,

© 2006 by Taylor & Francis Group, LLC

(8.82)

TRANSMISSION-LINE EQUATIONS

295

then η/N is adimensional, while V and I have the same dimensions. In particular, V and I can also be taken adimensional. In such a case we have [Ex ] = V/m, [Hy ] = A/m. Everything is ready for applying the so-called scattering-matrix formalism (Corazza, 1974). An interesting sub-case is to choose N = η. If we start from it, and extend the formalism of this section to TE and TM modes, as far as possible, then we ﬁnd again the results shown in Figure 7.3. 2) If we let N be adimensional, then it is possible (although not necessary) to let V have the physical dimensions of voltage, and I those of a current. Then, we have [Ex ] = [Hy ] = m−1 . If we require that for ω → 0 our V and I tend to become the same as voltage and current in the time-independent case, then N cannot be chosen arbitrarily, but must be derived from two deﬁnitions of the following types: V =− E · d1 , I= H · d2 , (8.83) 1

2

where 1 is an open path which belongs to the generic z = constant plane, and whose extrema fall on the two conductors of the line, while 2 is a closed path (also belonging to the generic z = constant plane), concatenated with only one of these conductors. Given that Eqs. (8.71) hold, Eqs. (8.83) lead to: Ex a ˆx · d1 = −1 , Hy a ˆy · d2 = 1 . (8.84) 1

2

3) If, independently of the physical dimensions one gives to Ex , we choose: N= |Ex |2 dx dy , (8.85) S

where S is the cross-section of the waveguiding structure, then the complex power which ﬂows through this cross-section is given by: E × H∗ V I∗ · aˆz dS = . (8.86) Pc ≡ 2 2 S This makes the relationship between complex power on one side, and V and I on the other side, formally identical with that which holds at the terminals of a one-port electrical network (see Figure 8.16). Note, however, that the dimensions of V and I are still undeﬁned. To make Eq. (8.86) really the same as the usual circuit-theory relationship, it is necessary that, in addition to Eq. (8.85), one of Eqs. (8.84) be valid too. One can then introduce the scattering-matrix formalism starting from any possible choice of N . To show that, let us rewrite Eqs. (8.75) in the following form: ⎞ ⎛ jωμ 0 d V V N ⎠ . (8.87) = −⎝ I dz I jωN 0 To diagonalize the matrix on the right-hand side of Eq. (8.87) (i.e., to ﬁnd its

© 2006 by Taylor & Francis Group, LLC

296

WAVES ON TRANSMISSION LINES

Figure 8.16 Relationship between voltage, current, and complex power.

√ eigenvalues ±iβ = ±iω μ and its eigenvectors) is straightforward algebra, yielding the following equations: da dz

= −jω

db dz

= jω

where: a =

1 2

b

1 2

=

√

√

V η/N V η/N

μ a

,

μ b ,

+

−

(8.88)

η/N I

,

η/N I

Their inverse relationships clearly are: η V = (a + b) , N " N (a − b) . I = η

.

(8.89)

(8.90)

The new variables Eqs. (8.88) are often referred to as the normal modes of the structure. Their practical importance is probably not evident, at this stage, because this whole section suﬀers from limitations due to some of the simplifying initial assumptions. In fact, after writing Eqs. (8.83), we always mentioned two conductors. However, what we proved in Section 7.4 was that the requirement for the existence of the TEM mode were at least two conductors, without an upper limit to their number. In a structure with n > 2 conductors, it is evident that the TEM mode, in spite of its eigenvalue being unique, has more degrees of freedom than what we assumed when we wrote Eqs. (8.71). Namely, in a multiwire transmission line, one can deﬁne n − 1 independent voltages (each of them is the voltage between one common reference conductor and the remaining n − 1 independent conductors) and n − 1 independent currents (the n-th current in the common reference being the “return” one, equal and

© 2006 by Taylor & Francis Group, LLC

TEM AND QUASI-TEM PROPAGATION IN PLANAR LINES

297

ε rε 0 t

2d w

ε0

t d

w

εrε0

Figure 8.17 Schematic view of a stripline and a microstrip.

opposite to the sum of the others). Dealing with such a structure in terms of 2(n − 1) independent diﬀerential equations would make the formalism unbearably heavy, unless one passes through diagonalizing a matrix like in Eq. (8.87). Further developments on this subject are left to Section 8.12. 8.11

TEM and quasi-TEM propagation in planar lines

For at least three decades, the whole world of electronics evolved in the direction of exploiting planar technologies more and more. Microwaves were not at all an exception to this rule. Waveguides of the types discussed in the previous chapters, coaxial cables, or wire pairs, are not very well matched to this trend. The tendency to build passive transmission lines which consisted of two or more ﬂat conductors, lying on parallel planes and separated by suitable dielectrics (e.g., Itoh, Ed., 1987) was then a natural evolution. Figure 8.17 shows schematically the most commonly used types of planar lines. The conducting planes whose edges are not shown in the drawings are, usually, much wider than all the other transverse geometrical dimensions. In the theoretical analyses they are usually taken as inﬁnitely wide. Unless the opposite is explicitly speciﬁed, they are at a common “ground” potential, also when they are more than one. Voltages of interest in the propagation along these structures are those of the other conductors, with respect to ground. The current in the ground electrodes is the opposite of the (algebraic) sum of currents in all the other conductors. The technological progress, from waveguides to planar lines, is not accompanied by a similar progress in modeling. On the contrary, none of the structures shown in Figure 8.17 is simple to analyze theoretically. Approximations are necessary, almost invariably, often for more than one reason, as we shall explain now. The theory of guided e.m. waves, as developed in the previous chapter, re-

© 2006 by Taylor & Francis Group, LLC

298

WAVES ON TRANSMISSION LINES

E field H field

ε0

εr ε0 Figure 8.18 Electric and magnetic ﬁeld lines in the cross-section of a typical microstrip.

quires a homogeneous medium in the region where the guided ﬁeld travels. This is the case, for example, for the symmetrical stripline shown in the upper part of Figure 8.17. Then, as the cross-section is not a simply connected domain, the general theory tells us that the fundamental mode is TEM. All the contents given so far in this chapter apply. The only diﬃculty that one still has to overcome is calculating the per-unit-length parameters l and c (plus, if required, r and g). Often, though, these entail approximations, because the geometry of the structure does not lend itself to a rigorous analysis. An example will be provided in a short while. In other cases — typically, the microstrip shown in the lower part of Figure 8.17, which is probably the most practical among all those examples — part of the ﬁeld travels in the dielectric between the conductors, and part in the air above them (see Figure 8.18). The theory of the previous chapter no longer applies. Intuition tells us that a structure with more than one conductor can operate at arbitrarily low frequencies, so its fundamental mode must have something in common with the TEM mode. However, it is easy to verify that, if we impose the condition that the longitudinal components of the electric and magnetic ﬁelds are zero everywhere, then we do not have enough degrees of freedom left to satisfy all the continuity conditions at the air-dielectric interface, plus the boundary conditions on the conductor surfaces. As a consequence, the fundamental mode is not TEM, but quasi-TEM. The whole content of this chapter then applies to it only as an approximation, superimposed upon the diﬃculty of ﬁnding the per-unit-length parameters. In the following we will discuss brieﬂy an example, to show what this can mean in practice. 8.11.1

First example: the stripline

For a lossless structure which can propagate the TEM mode, we know from Section 8.4 that its characteristic impedance and its phase constant are given √ by Eqs. (8.16) and (8.17), respectively, namely ZC = l/c and β = ω lc. Furthermore, from the previous section we know (see Eq. (8.79)) also that √ β = ω μ. Therefore, when the parameters μ and of the dielectric medium

© 2006 by Taylor & Francis Group, LLC

TEM AND QUASI-TEM PROPAGATION IN PLANAR LINES

299

are known, the phase constant is known, and to determine the characteristic impedance it is enough to determine either c, or l. As the reader may know from previous classes of static ﬁelds, a powerful technique to calculate capacitances, even for apparently complicated geometries, is known as conformal mapping. It is explained in several advanced textbooks on static ﬁelds (e.g., Plonsey and Collin, 1961), and cannot be dealt with in this book because of space limits. When applied to the geometry of Figure 8.17, part a), under the assumption of negligible thickness t, it yields a per-unit-length capacitance such that the corresponding characteristic impedance is: K(k) η √ , (8.91) Zc ≈ 4 K( 1 − k 2 ) where η = (μ/)1/2 is, as usual, the characteristic impedance of the homogeneous dielectric medium, K(k) indicates the so-called complete elliptic integral of the ﬁrst kind (Abramowitz and Stegun, Eds., 1965), and:

πw −1 k = cosh , (8.92) 4d the symbols w and d being explained in Figure 8.17, part a). The practical use of Eq. (8.91) can be simpliﬁed by using two approximate expressions for the elliptic integral, namely:

πw η ln 2 coth , (8.93) Zc 2π 8d for w/d ≤ 1.12, and: πη Zc , (8.94) 8[ln 2 + πw/(4d)] for w/d > 1.12. The assumption of negligible thickness, on the other hand, makes it impossible to calculate the ohmic loss for a ﬁnite conductivity and the consequent attenuation constant. Indeed, a conductor of zero cross-section and ﬁnite conductivity would entail an inﬁnitely large loss, which is obviously unrealistic. For a ﬁnite thickness t, a perturbation approach like that of Section 7.9 yields an attenuation constant: πw/2d + ln(8d/πt) Rs αc = , (8.95) 2Zc d ln 2 + πw/4d where RS is, once again, the surface resistance of a conducting half-space, which we found ﬁrst in Section 4.12. 8.11.2

Second example: the microstrip

It is easy to show that, as we said before, for all the structures shown in Figure 8.17 except for part a), the fundamental mode cannot be rigorously TEM, as they encompass more than one dielectric. Indeed, from the decomposition theorem of Section 3.8 we see that a nontrivial TEM solution (i.e.,

© 2006 by Taylor & Francis Group, LLC

300

WAVES ON TRANSMISSION LINES

with nonzero transverse components) requires that the phase constant be √ β = ω μi in each medium (i = 1, 2), which is clearly impossible if 1 = 2 . Approximations are required to begin with. We will outline the way of thinking with reference to the microstrip (Figure 8.17, part b) which is by far and large the most widely used in practice, but the essential features apply also to the other structures. At relatively low frequency, we may still assume that the longitudinal components are negligible compared to the transverse ones. (The reader may justify this, as an exercise, in quantitative, terms starting from the decomposition theorem). Within this approximation, the only problem we are left with is to calculate a per-unit-length capacitance, to replace Eq. (8.91). However, conformal mapping no longer provides exact results when the dielectric is not homogeneous. Another approximation is needed; the most widely accepted way to proceed is to calculate ﬁrst Eq. (8.91) as if the dielectric were air everywhere, and then correct it by means of an “equivalent” relative permittivity, deﬁned as that of a dielectric medium ﬁlling all the half-space above the ground plane, yielding the same per-unit-length capacitance as that of Figure 8.17, part b). Several answers can be found in the literature. The most commonly used one is given in Schneider (1969): (εr − 1) 1 . (8.96) 1+ εr = 1 + 2 1 + 10d/w At higher frequencies, neglecting the longitudinal components yields unacceptable results, in particular for the phase constant, which is of paramount importance in designing components and circuits. An exact theory, in terms of potentials, can be found in the literature (Collin, 1991), but is too long and demanding to be reported here. A widely used approximation is that of Yamashita et al. (1979): √ εr − εr β = + εr , (8.97) k0 1 + 4F −1.5 where k0 = 2π/λ0 is the free-space phase constant, and: √

4d εr − 1 w 2 F = 0.5 + 1 + 2 ln 1 + λ0 d

.

(8.98)

One essential feature is that the dispersion in this structure becomes larger when the relative permittivity becomes higher, and when the ratio w/d becomes smaller, i.e., when the structure becomes more compact. Therefore, design of microstrip circuits is inevitably a trade-oﬀ between contradictory requirements. As for attenuation due to ﬁnite conductivity, its order of magnitude can be estimated with a rather rough approximation, if a microstrip is considered as a pair of stripes of identical width w. The result of the perturbation approach is then simply: (8.99) αc = Rs /(wZc ) ,

© 2006 by Taylor & Francis Group, LLC

THE COUPLED-MODE EQUATIONS

301

where Rs is the usual surface resistance of a half-space good conductor, and Zc is the characteristic impedance of the microstrip. In agreement with intuition, Eq. (8.99) overestimates the attenuation, since, in reality, the width of the conducting backplane is much larger than w. More precise calculations are easily found in modern textbooks on microwaves, and are omitted here because of space limits. 8.12

The coupled-mode equations

Completeness of a set of modes in a perfect cylindrical structure was discussed in Section 7.6. The conclusion was that a suitable countable inﬁnity of orthogonal modes could form a basis for expanding any ﬁeld in such a structure. One must reconsider completeness, from another point of view, when dealing with an imperfect structure, I, which diﬀers slightly from a perfect one, P , whose modes are known, whilst those of the imperfect structure cannot be found exactly. A typical example in this sense is a waveguide whose metallic walls have a ﬁnite conductivity (see Section 7.9). In such a case, a set of modes {Ei , Hi } is said to be complete when it satisﬁes the following convergence requirement: n n ∗ ∗ ∗ lim Vi (z) Eti × Ht (z) − Ii (z) Hti · a Et (z) − ˆz dS = 0 , n→∞

S

i=1

i=1

(8.100) where {Et , Ht } are the transverse components of any ﬁeld which may exist in I; S is the cross-section of I; {Eti , Hti } are the (z-independent) transverse components of the i-th mode of the perfect structure P ; {Vi , Ii ; i = 1, 2, . . .} are complex coeﬃcients, which, in general, depend on z. For the time being, we may assume that these coeﬃcients are adimensional. The condition in Eq. (8.100) means that the two series Σ(Vi Eti ) and Σ(Ii Hti ) converge towards Et and Ht , respectively, at any point on S , except, at most, the points of a zero-measure subset. To state and to prove what are the most general conditions for Eq. (8.100) to be satisﬁed, is long and complicated, but may be found in the literature (Collin, 1991). We prefer to invest some space and eﬀort in introducing a formalism which is convenient to use whenever Eq. (8.100) holds. Through it, from Maxwell’s equations one obtains a set of ordinary diﬀerential equations (ODE’s) for {Vi (z), Ii (z)}. The derivation is formally reminiscent of that of Section 8.10, and therefore may be phrased in the typical terminology of transmission lines. The only problems we are going to face in this derivation come from the fact that, for the two series in Eq. (8.100), uniform convergence is not assured at every point. Thus, we are not guaranteed that they can be diﬀerentiated term by term. In the structure I, where there are no sources, Maxwell’s equations, projected on the z = constant plane, read: ∂Et ∂z

© 2006 by Taylor & Francis Group, LLC

= jωμ a ˆ z × H t + ∇ t Ez

,

302

WAVES ON TRANSMISSION LINES

∂Ht = −jω a ˆ z × Et + ∇ t H z . (8.101) ∂z When Eq. (8.100) holds, in the previously speciﬁed sense of this statement, we may set: Et =

∞

Vi (z) Eti

,

Ht =

i=1

∞

Ii (z) Hti

.

(8.102)

i=1

Let us recall that {Ei , Hi } is the i-th mode of the structure P . For the sake of generality, we will suppose that it is a hybrid mode (i.e., Ez = 0, Hz = 0). Then, using Eqs. (7.11) and (7.12), we may write: Eti Hti

jγi ∇t Ti ω

,

jγi ∇t Li + ∇t Ti × a ˆz ωμ

.

= ∇t Li × a ˆz − =

(8.103)

The functions Li and Ti satisfy: ∇2t Fi = −χ2i Fi

(χ2i = γi2 − κ2 ) .

(8.104)

To obtain the ODE’s in Vi (z) and Ii (z) we are looking for, at ﬁrst sight one would think it is enough to express Ez and Hz in terms of Vi and Ii in Eqs. (8.101). Another step, based on mode orthogonality in the perfect structure P , would then allow us to isolate individual coeﬃcients (i = 1, 2, . . .) in Eqs. (8.101). In reality, this procedure would not be correct. As the series do not converge uniformly, ∇t Ez and ∇t Hz in Eqs. (8.101) cannot be calculated by term-by-term diﬀerentiation of Eqs. (8.102). It is necessary to pass ﬁrst from Eqs. (8.101) to equations where Ez and Hz do not go through diﬀerential operators. To this purpose, let us take the scalar product of the ﬁrst of Eqs. (8.101) ∗ ∗ /2, and of the second one with a ˆz × Etm /2. Let us integrate with a ˆz × Htm over S , take into account mode orthogonality, and impose on each of them a normalization condition (power through S equal to 1 W). All these operations result in the following equation: ∗ Eti × Htm ·a ˆz dS = δi,m , (8.105) 2 S where δi,m is Kronecker’s delta. Let us write also the following identities, which can be proved by the reader as an exercise using Appendix C and taking into account Eq. (8.104): ∇t Ez · ∇t Tm ∇t Ez × ∇t Lm

= ∇t · (Ez ∇t Tm ) + χ2m Ez Tm = ∇t × (Ez ∇t Lm ) .

,

(8.106) (8.107)

Let us also consider two identities which diﬀer from the previous ones only because Ez is replaced by Hz , and because Tm and Lm have been interchanged. We omit writing them explicitly for the sake of brevity.

© 2006 by Taylor & Francis Group, LLC

THE COUPLED-MODE EQUATIONS

303

In the equations which were derived from Eqs. (8.101), all the terms involved in the latest identities show up inside integrals over the surface S . Gauss’ theorem (A.8) can be applied to the integral containing the ﬁrst term on the right-hand side in Eq. (8.106), and to the twin term, containing Hz . Stokes’ theorem (A.9) can be applied to the integral containing the right-hand side of Eq. (8.107), and to the twin term containing Hz . The results are the following equations: dVm jγ ∗ = −jωμ Im + m χ2m Ez Tm∗ dS Nm dz ω S ∗ ∂Tm∗ jγm ˆ d , (8.108) d + + Ez Ez (∇t L∗m · ) ω ∂n 1 dIm Nm dz

∗ jγm 2 = −jω Vm − χ Hz L∗m dS ωμ m S ∗ ∂L∗m jγm ˆ d , (8.109) d + − Hz Hz (∇t Tm∗ · ) ωμ ∂n

where: • ∂/∂n is the normal derivative; • = S ∩ C is the boundary of the section S , oriented clockwise with reference to the unit vector a ˆz ; ˆ • is the unit vector tangent to ; • Nm = |Etm |/|Htm | is a generalization, for the m-th mode, of the constant N of Eq. (8.74), and is proportional to the wave impedance of the m-th mode. At this stage, Ez and Hz in Eqs. (8.108) and (8.109) are no longer arguments of diﬀerential operators. Thus, non-uniform convergence of their series expansions does not cause problems any more. These series expansions can be obtained through the longitudinal components of Maxwell’s equations, if we let the transverse ﬁelds be expressed by Eqs. (8.102), namely: ∞

Ez

1 ∇t × Hti ∇ × Ht = Ii (z) jω jω i=1

=

=

∞

Ii (z) Ezi = −

i=1

∞ 1 Ii (z) χ2i Ti jω i=1

,

(8.110)

.

(8.111)

∞

Hz

= −

=

1 ∇t × Eti ∇ × Et = Vi (z) jωμ −jωμ i=1

∞ i=1

© 2006 by Taylor & Francis Group, LLC

Vi (z) Hzi =

∞ 1 Vi (z) χ2i Li jωμ i=1

304

WAVES ON TRANSMISSION LINES

Let us now place these expressions into Eqs. (8.108) and (8.109), and use commutation between the integral over S and the sum with respect to i. Taking into account further orthogonality relations, of the type in Eq. (7.35), between pre-potentials of the perfect structure P , the only terms among the integrals over S which survive are those whose arguments contain products of the type Lm L∗m and Tm Tm∗ . Because of the normalization in Eq. (8.105), the integrals containing these products are constants: Lm L∗m dS = Am , (8.112) S

S

Tm Tm∗ dS

= Bm

,

(8.113)

although it is impossible to specify their explicit values as long as the boundary conditions on the walls of P are not speciﬁed. In conclusion, Eqs. (8.108) and (8.109) become: dVm dz

∞ jωμ = − Im + cm Im + Cmn In Nm n=1

dIm dz

= −jω Nm Vm + dm Vm +

∞

,

Dmn Vn

,

(8.114)

n=1

whose coeﬃcients are shown in Table 8.1. In the perfect structure P , orthogonality assures that each mode may exist by itself. This implies that in P it must be Cmn = 0 and Dmn = 0 for m = n. Then, Eqs. (8.114) reduce to the standard transmission line equations that we studied for most of this chapter. This makes it self-explanatory why in general Eqs. (8.114), for the imperfect structure I, are referred to as the coupled-line equations, or also as the generalized telegraphist equations. Analogy with the previous section entails us to let, for each mode: " η Nm (am − bm ) . (am + bm ) , Im = (8.115) Vm = Nm η Introducing these new variables into Eqs. (8.104), we get the following equations, usually referred to as the coupled-mode equations: dam = −Γm am + (Fmn an + Gmn bn ) , dz n=m

dbm dz

= −Γm bm −

(Gmn an + Fmn bn ) ,

(8.116)

n=m

where Γm is the propagation constant of the m-th mode, provided it is alone (i.e., an = bn = 0 for n = m) in the imperfect structure I. In general, it is Γm = γm (the propagation constant of the same mode in the perfect structure), as Γm accounts also for the other terms in Eqs. (8.114) that correspond to n = m.

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

305

Table 8.1 Expressions for the coeﬃcients of the coupled-mode equations. Cm = −

∗ χ4m γm Bm ω 2 2 Nm

dm = −

χ4m ∗ γm Nm Am ω 2 μ2

Cmn = −

∗ 2 1 ! χn 1 ! ∂Tm∗ γm χ2n ˆ d T + Tn (∇t L∗m · ) d n ω 2 2 Nm ∂n −jω Nm

Dmn = −

∗ 2 ! ! χn ∂L∗m γm χ2n ˆ d N L + d Nm Ln (∇t L∗m · ) m n ω 2 μ2 ∂n jωμ

The mode coupling coeﬃcients are related to the quantities shown in Table 8.1 via: Nn Nm Cmn + Dmn , 2Fmn = Nm Nn Nn Nm 2Gmn = Cmn − Dmn . (8.117) Nm Nn Many examples of solutions of Eqs. (8.116) are available in the literature. Their physical meaning changes from problem to problem. Perhaps the most typical example is that of the directional coupler, which was brieﬂy illustrated in Section 3.9 when we discussed symmetries and their implications. Further suggestions to the reader will be given in the next section. 8.13

Further applications and suggested reading

Solving Maxwell’s equations, or using circuit theory, are, as we said on several occasions, two diﬀerent approaches to one physical reality. The diﬀerence is in the axioms. In circuit theory, the axioms are Ohm’s law and Kirchhoﬀ’s laws. In our approach to e.m. waves, the axioms are Maxwell’s equations and the charge continuity equation. In this chapter we have shown that, with an additional assumption (namely, that of TEM propagation), the ﬁeld approach leads to the same transmission lines equations as the circuit approach. The link between the two approaches is broader, in its scope of validity, than TEM waves, as we outlined brieﬂy in Section 8.10. Hence, one of the most important messages that the reader can extract from this chapter is that, in order to understand more in depth high-frequency e.m. ﬁelds, one should have a strong background in circuit theory also. We recommend reading the classical treatise by Newcomb (1966) that we have quoted on several occasions in this book, and at least one other text, more focused on distributed-parameter circuits, such as, for instance, Ghausi and Kelly (1968). In particular, the Smith chart of Section 8.7 is a crucial intersection between the circuit culture and the e.m. wave culture. On this front, further interest may be stimulated in those readers who like abstraction and mathematical approaches, by saying that bilinear transforms help to solve many other problems related to impedance transformations through various types of linear

© 2006 by Taylor & Francis Group, LLC

306

WAVES ON TRANSMISSION LINES

networks. The short but condensed booklet by Schwerdtfeger (1961), that we quoted when we introduced the Smith chart, is also a good starting point in this direction — see also Liboﬀ and Dalman (1985). Another valuable resource is Chapter 7 (by R. Beringer) of Montgomery, Dicke and Purcell (1948). The reader is encouraged to consult it after studying our Chapter 9 on resonant cavities. A point that might look trivial at ﬁrst sight, but in our opinion is not so, is that transmission-line based problems show up in a very large number, and in a very large variety, of occasions in the professional life of anybody in electronic engineering. Solving many problems is a valuable investment for an EE student, in view, for example, of an employment interview. For this reason, the collection of problems at the end of this chapter is wider than the average over the rest of the book. All other textbooks covering this subject are good sources of problems. The last comment has brought us towards rather practical issues. Continuing in the same direction, let us stress that what has been said in this chapter (and in the whole book) on striplines and microstrips, has to be thought of just as an elementary introduction. All telecommunication engineers should learn much more than that. Many ﬁrst-class textbooks cover the area both under the viewpoint of theory and that of applications. Just to quote a few, Collin (1991) is highly recommended in the ﬁrst category; Wadell (1991), Gupta et al. (1996), and Bhat and Koul (1989) are representative for the second one. Other useful texts include Freeman (1996) and Elliott (1993) (general microwave texts), and Christopoulos (1995), Fache et al. (1993), Faria (1993) (texts on mathematical modeling of microwave and strip transmission lines). On the other hand, the reader must be encouraged not to underestimate the power of the rather abstract approach of Sections 8.10 and 8.12. Without these connections, it would be very hard to understand why the transmission line equations and the coupled-mode equations show up in problems where there are no conductors at all, such as integrated optics, or where it remains hard to visualize two coupled lines, such as microwave tubes based on electron beams, or particle accelerators, or traveling-wave microwave masers. Suggested reading in these areas include Unger (1977) on planar dielectric waveguides, Tamir, Ed., (1979) on integrated optics, Gilmour (1986), Chodorow and Susskind (1964), and Sims and Stephenson (1963) on microwave tubes, Pierce (1950/1954/1974) on traveling wave tubes, electron beams, and waves as well as some excellent popularizations, Louisell (1960) on coupled modes, Roussy and Pearce (1995) on industrial microwave device applications. However, especially in the case of integrated optics, it is preferable for the reader to consult these books only after studying our Chapter 10. Let us conclude with a comment on the coupled-mode equations (CME’s). They are still a popular subject of discussion and investigation, especially in integrated optics. One of the reasons why they are popular is because some important devices (typically, directional couplers and ﬁlters) cannot be studied analytically without approximation. However, there is a more subtle reason.

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

307

It is because the “perfect” structure P one starts from, in order to study an “imperfect” structure I (see Section 8.12), is not unique. For example, to study a directional coupler which encompasses a grating, one can start either from the modes of the two isolated waveguides, or from those (sometimes designated as supermodes) of two parallel waveguides. These also are available in exact closed form, for some geometries. Looking back at the general derivation of Section 8.12, the reader, if familiar with perturbation theory, should be able to understand why the accuracy of the perturbation solutions depends on the perfect structure one starts from. For very instructive details on this point, recent research papers by Huang et al. (1996) should be consulted, since, to the best of our knowledge, the subject has not yet been published as a book — see Huang (1994) for an overview, and, e.g., Sarangan and Huang (1994), and Huang and Hong (1992) for example applications. Problems 8-1 Express the active and the reactive power in the generic section of a transmission line in terms of the reﬂection coeﬃcient, separating real and imaginary parts of Eq. (8.15). Assume ﬁrst that the characteristic impedance of the transmission line is real. Then, repeat the exercise assuming a complex characteristic impedance. 8-2 Extend all the results of Section 8.6 to the case of low-loss transmission lines. 8-3 For the lossless transmission lines shown in Figure 8.19, we have R G = Zc1 = 50 Ω, Zc2 = 100 Ω, VG = 10 V. In the line with characteristic impedance Zc1 , the VSWR equals 4/3. The voltage modulus goes through a minimum where there is a discontinuity in the characteristic impedance. a) Evaluate the load impedance ZL . b) Calculate the complex power delivered to the load impedance for XG = 0, for XG = RG , and for XG = −RG , assuming that the length of the line with characteristic impedance Zc1 equals an odd integer multiple of λ/4. 8-4 In the lossless transmission line shown in Figure 8.20, the phase constant is the same as in free space. The frequency is f = 100 MHz. We have RG = Zc = 50 Ω, RL = 60 Ω, VG = 20 V, C = (1/π) pF, L = 0.1/(4π) mH. The line lengths, 1 and 2 , are both equal to 2 m. a) Set R1 = 40 Ω, and calculate the active power delivered to the load resistance RL . b) Find the range of values of R1 for which the total active power, delivered to R1 and to RL , is more than one half of the generator’s available power. 8-5

For the lossless transmission line shown in Figure 8.21, at the frequency

© 2006 by Taylor & Francis Group, LLC

308

WAVES ON TRANSMISSION LINES d

jXG ZL

RG VG

ZC2

ZC1

Figure 8.19

ZG

R

VG

ZC

L

C

RL

ZC z=a

z=b

z=0

Figure 8.20

f = 600 MHz, we have Zc = 200 Ω, d1 = 15 m, XL = 100 Ω, XC = −400 Ω, VG = 10 V. In the Zc1 part of the line, the voltage modulus goes through a minimum 0.5 m on the left of the terminals 22 , while the VSWR is 4. The active power delivered to the load R equals 40 mW. a) Find all the values of the distance d2 which are compatible with these data. b) Find the unknown values of the resistances R and RG . c) Find voltage and current at a generic distance from the generator. 8-6 Reconsidering the transmission line shown in the ﬁgure of Problem 84, and retaining all the other parameter values, ﬁnd the internal impedance 1

RG

2

d1

XL

VG

XC

ZC 2’

Figure 8.21

© 2006 by Taylor & Francis Group, LLC

3

R

ZC 1’

d2

3’

FURTHER APPLICATIONS AND SUGGESTED READING

309

of the voltage generator which satisﬁes the power matching condition at the generator terminals. Repeat the problem modeling the generator as an ideal current generator in parallel with an internal admittance. 8-7 Verify that one can design a single-stub impedance-matching network using an analytical approach, instead of using the Smith chart, but that the procedure becomes quite cumbersome. 8-8 Work out in full detail the procedures to design a double-stub impedancematching network for the two cases shown in Figure 8.14 but not discussed in the text, i.e., when the distance between the stubs is equal either to λ/8 or to 3λ/8. 8-9 Enumerate relative advantages and disadvantages of single- and doublestub impedance-matching networks. Distinguish between two situations. First, consider a laboratory environment, where the load impedance is changed often, and in general is not known a priori. Second, consider a mass production environment, where many identical pieces are fabricated. 8-10 A low-loss transmission line, whose nominal characteristic impedance is 50 Ω, exhibits a loss of 2 dB over a length of 10 m. The loss contributed by the parallel conductance is negligibly small. Calculate the per-unit-length series resistance r, and the corresponding correction to the phase constant with respect to the lossless case, at 900 MHz, assuming a per-unit-length capacity of 50 pF/m. 8-11 For the network discussed in Problem 8-3, design a single-stub impedance-matching network at the frequency f = 600 MHz, to be placed upstream with respect to the discontinuity in the characteristic impedance, but as close as possible to it. Discuss two possibilities, namely a stub with characteristic impedance Zc1 and with characteristic impedance Zc2 . Calculate the complex power at the load, at the discontinuity, and at the terminals of the stub. 8-12 For the network discussed in Problem 8-4, design a quarter-wavelength impedance-matching network to be placed upstream with respect to the resistor R = 40 Ω. Find both solutions, one with an increase and the other with a decrease in characteristic impedance. Then, increasing the frequency to 200 MHz, and leaving the whole network unchanged, calculate the VSWR at the generator terminals. 8-13 A lossless transmission line, with phase constant equal to that of free space and characteristic impedance Zc = 100 Ω, is closed on a load impedance ZL = 100 − j50 Ω, and fed by a generator with an open-circuit voltage VG = 10 V and an internal impedance ZG = 100 + j50 Ω. The operating

© 2006 by Taylor & Francis Group, LLC

310

WAVES ON TRANSMISSION LINES

frequency is 300 MHz. Design a double-stub impedance-matching network connected to the load terminals, and with a spacing d = 0.25 m between the stubs. Study how the current and voltage vary with z, from the generator to the load. Calculate the complex power at the terminals of the load and at those of the two stubs. 8-14 For a lossless coaxial cable (whose outer and inner radii are denoted as a and b, respectively), show that if voltage and current are deﬁned as in the electrostatic case (i.e., according to Eqs. (8.83)), then the characteristic impedance of the cable equals Zc = (μ/)1/2 ln(a/b)/(2π). 8-15 In an air-ﬁlled coaxial cable, the inner conductor is of radius b = 2.5 mm. Making use of the formula that was proved in the previous problem, adjust the radius of the outer conductor such that Zc = 120 Ω. 8-16 The cable of the previous problem is connected to a 50 Ω resistive load impedance, at the frequency of 300 MHz. A metallic sleeve can be used as a λ/4 impedance-matching device. Find its length, its outer radius, and its distance from the load terminals. Repeat the same problem for a resistive load impedance of 300 Ω. 8-17 Show, using the decomposition theorem of Section 3.8 or of Section 7.2, that, within each homogeneous dielectric medium present in a microstrip, the longitudinal ﬁeld components are negligibly small compared to the transverse ones. 8-18 Using one of the approximations, Eqs. (8.93) or (8.94), design a stripline, in a homogeneous dielectric medium having r = 2.2 and a spacing d = 10 mm, so that its characteristic impedance is 50 Ω. For copper conductor whose thickness is t = 0.2 mm, evaluate the attenuation constant vs. frequency between 1 and 2 GHz. (Copper conductivity: 5.8 × 107 Ω−1 m−1 .) 8-19 A microstrip for which w/d = 4 is fabricated with copper on alumina, the dielectric for which r = 9.8. Evaluate the equivalent permittivity, Eq. (8.96), and then the corresponding phase constant, dispersion (i.e., the second derivative of β with respect to ω), and attenuation constant, at 3.5 GHz. 8-20 Using Appendix C and taking into account Eq. (8.104), derive Eqs. (8.106) and Eq. (8.107).

© 2006 by Taylor & Francis Group, LLC

References Abramowitz, M. and Stegun, I.A. (Eds.) (1965) Handbook of Mathematical Functions. Dover, New York. Bhat, B. and Koul, S.K. (1989) Stripline-Like Transmission Lines for Microwave Integrated Circuits. Wiley, New York. Chodorow, M. and Susskind, C. (1964) Fundamentals of Microwave Electronics. McGraw-Hill, New York. Christopoulos, C. (1995) The Transmission-Line Modeling Method: TLM. IEEE/OUP Series on Electromagnetic Wave Theory, Institute of Electrical and Electronics Engineers, New York/Oxford University Press, Oxford. Collin, R.E. (1991) Field Theory of Guided Waves. IEEE Press, New York. Corazza, G.C. (1974) Fondamenti di Campi Elettromagnetici e Circuiti. Vol. II, Patron, Bologna. Dicke, R.H. and Wittke, J.P. (1960) Introduction to Quantum Mechanics. AddisonWesley, Reading, MA. Elliott, R.S. (1993) An Introduction to Guided Waves and Microwave Circuits. Prentice-Hall, Englewood Cliﬀs, NJ. Fache, N., Olyslager, F. and de Zutter, D. (1993) Electromagnetic and Circuit Modelling of Multiconductor Transmission Lines. Oxford Engineering Science Series, Clarendon Press, Oxford/Oxford University Press, New York. Faria, J.A.B. (1993) Multiconductor Transmission-Line Structures. Wiley Series in Microwave and Optical Engineering, Wiley, New York. Freeman, J.C. (1996) Fundamentals of Microwave Transmission Lines. Wiley Series in Microwave and Optical Engineering, Wiley, New York. Ghausi, M.S. and Kelly, J.J. (1968) Introduction to Distributed-Parameter Networks: with Application to Integrated Circuits, Holt, Rinehart and Winston, New York. Gilmour, A.S., Jr. (1986) Microwave Tubes. Artech House Microwave Library, Artech House, Dedham, MA. Gupta, K.C. et al. (1996) Microstrip Lines and Slotlines. 2nd ed., Artech House Microwave Library, Artech House, Boston. Huang, W.-P. and Hong, J. (1992) “A coupled-waveguide grating resonator ﬁlter”. IEEE Photonics Technology Letters, 4 (8), 884–886. Huang, W.-P. (1994) “Coupled-mode theory for optical waveguides: an overview”. J. Optical Society of America A, 11 (3), 963–983. Huang, W.-P., Guo, Q. and Wu, C. (1996) “A polarization-independent distributed Bragg reﬂector based on phase-shifted grating structures”. J. Lightwave Technology, 14 (3), 469–473. Itoh, T. (Ed.) (1987) Planar Transmission Line Structures. IEEE Press Selected Reprint Series, IEEE Press, New York. Liboﬀ, R.L. and Dalman, G.C. (1985) Transmission Lines, Waveguides, and Smith Charts. Macmillan, New York. 311 © 2006 by Taylor & Francis Group, LLC

312

REFERENCES

Louisell, W.H. (1960) Coupled Mode and Parametric Electronics. Wiley, New York. Montgomery, C.G., Dicke, R.H. and Purcell, E.M. (Eds.) (1948) Principles of Microwave Circuits. McGraw-Hill, New York. Reprinted by Peter Peregrinus, London, for IEE (1987). Morse, P.M. and Feshbach, H. (1953) Methods of Theoretical Physics. McGraw-Hill, New York. Newcomb, R.W. (1966) Linear Multiport Synthesis. McGraw-Hill, New York. Pierce, J.R. (1950) Traveling-Wave Tubes. Bell Telephone Laboratories Series, Van Nostrand, New York. Pierce, J.R. (1954) Theory and Design of Electron Beams. Bell Telephone Laboratories Series, 2nd ed., Van Nostrand, New York. Pierce, J.R. (1974) Almost All About Waves. MIT Press, Cambridge, MA. Plonsey, R. and Collin, R.E. (1961) Principles and Applications of Electromagnetic Fields. McGraw-Hill, New York. Roussy, M.G. and Pearce, J.A. (1995) Foundations and Industrial Applications of Microwave and Radio Frequency Fields: Physical and Chemical Processes. Wiley, New York. Sarangan, A.M. and Huang, W.-P. (1994) “A coupled mode theory for electron wave directional couplers”. IEEE J. Quantum Electronics, 30 (12), 2803–2810. Schneider, M.V. (1969) “Microstrip lines for microwave integrated circuits”. Bell System Technical Journal, 48 (5), 1421–1444. Schwerdtfeger, H. (1961) Introduction to Linear Algebra and the Theory of Matrices. 2nd ed., Noordhoﬀ, Groningen. Sims, G.D. and Stephenson, I.M. (1963) Microwave Tubes and Semiconductor Devices. Interscience Publishers, New York. Tamir, T. (Ed.) (1979) Integrated Optics. 2nd corrected and updated ed., SpringerVerlag, Berlin. Unger, H.-G. (1977) Planar Optical Waveguides and Fibres. Clarendon Press, Oxford. Wadell, B.C. (1991) Transmission Line Design Handbook. Artech House Microwave Library, Artech House, Boston. Yamashita, E., Atsuki, K. and Ueda, T. (1979) “An approximate dispersion formula of microstrip lines for computer-aided design of microwave integrated circuits”. IEEE Trans. Microwave Theory & Techniques, MTT-27 (12), 1036–1038.

© 2006 by Taylor & Francis Group, LLC

CHAPTER 9

Resonant cavities 9.1

Introduction

In the previous chapters, we showed what theoretical background is necessary, in order to comply with a self-explanatory technical requirement: how to convey electromagnetic energy along a given path. Similarly, this chapter is devoted to the theoretical basis required to answer another very important technical requirement: how to store electromagnetic energy in a given region in 3-D space. In elementary circuit theory, as well known, one usually starts oﬀ by introducing two ideal components, the capacitor and the inductor. Their purpose is to model, in the most simplistic way, two physical realities, that consist of storing energy, electric and magnetic, respectively, within given regions. The next step is to combine these two elements, and introduce the concept of a resonant circuit. This approach is strongly biased by previous knowledge of electrostatics and magnetostatics. Indeed, the two ideal components are rather easy to deﬁne as long as electric and magnetic ﬁelds are time-invariant. But in a rapidly time-varying regime, electric and magnetic ﬁelds are inseparable. Then, as we will see in this chapter, if we start from Maxwell’s equations, we may run directly into some objects whose behavior is essentially that of resonant LC circuits; but the main diﬀerence is that it is impossible to split these objects into two separate L and C components. As we saw in the previous chapter (and also in terms of plane waves, in Chapter 4), the input impedance of a short-circuited single-mode line is purely reactive. It is inductive, if the length of the line is less than λg /4, capacitive if λg /4 < < λg /2. Let us consider a transmission line of length λg /2, shortcircuited at both ends (Figure 9.1). Any plane orthogonal to the z axis (except for the one z = λg /4) splits the line in two parts, one of them having an inductive input impedance, the other having a capacitive input impedance. The choice of this plane is, once again, completely arbitrary; this indicates that the resonant behavior is an intrinsic property of the whole system in itself. The deﬁnition of a capacitor and an inductor is an artifact, and was used just as a link with known concepts. In the example shown in Figure 9.1, a complete round trip, from one of the short-circuit planes to the other and back, along the z axis, has a total length of a wavelength. Consider then a wave that passes through the generic point P with a given phase φ. When it passes again through P , in the same sense as before, it has the same phase φ. In the remainder of this chapter we will see 313 © 2006 by Taylor & Francis Group, LLC

314

RESONANT CAVITIES

Figure 9.1 A short-circuited coaxial cable, half wavelength in length, seen as a resonant circuit.

that this comment has a very broad scope of validity. A resonant behavior always entails that there are closed paths (in general, three-dimensional paths) whose total length is an integer multiple of a wavelength. In Chapter 7, we saw that Maxwell’s equations with 2-D boundary conditions admit solutions at any frequency. In this chapter, on the contrary, we will see that 3-D boundary conditions can be satisﬁed only at discrete frequencies; this is a consequence of the phase conservation requirement along closed paths, that we just illustrated. This frequency selection process reinforces the analogy with networks consisting of lumped inductors and capacitors, which also may resonate only at discrete frequencies. 9.2

Separable coordinate systems in three dimensions

In some of the previous chapters, we saw that separation of variables is a powerful method to solve the Helmholtz equation. On the other hand, we saw in Chapter 7 that it suﬀers from intrinsic limitations, due to the ﬁnite number of reference systems in which it can be applied. We will see now brieﬂy how those limitations extend from the 2-D to the 3-D case. An electromagnetic resonator is, once again, a region in 3-D space, surrounded by a surface S, which can either be completely at a ﬁnite distance from an origin (closed resonators), or extend in part to inﬁnity (open resonators). On this surface, we impose boundary conditions that the ﬁeld must satisfy. When the Helmholtz equation is solved by separation of variables, in a given reference system, boundary conditions are simple to express only if the surface S is a coordinate surface, or can be subdivided into coordinate surfaces, in that reference system. When this is not the case, then we may still use a complete set of solutions obtained by separation of variables, but the ﬁeld must be written as a series expansion. Such a procedure is always a source of nuisance; sometimes, it causes conceptual problems also, as consequences of whether it is legitimate or not, to diﬀerentiate a series term by term. For these reasons, it is very important to know from the beginning that the 3-D Helmholtz equation is separable only in a ﬁnite number of coordinate systems. We must omit the proof for the sake of brevity; it is very similar to

© 2006 by Taylor & Francis Group, LLC

COMPLETENESS OF RESONATOR MODES

315

that of the 2-D case, and can be found in the same references. In any case, we will provide now the list of those coordinate systems (the so-called “separable coordinate systems”), where separation of variables is applicable. The most general case (in the sense that all the others may be looked at as its degenerate cases) is that of orthogonal ellipsoidal coordinates (ξ1 , ξ2 , ξ3 ), which are implicitly deﬁned by the following relations, where (x1 , x2 , x3 ) is an orthogonal cartesian system: 1/2 (ξ12 − a2i ) (ξ22 − a2i ) (ξ32 − a2i ) i = 1, 2 , xi = j = 2, 1 (= i) a2i (a2i − a2j ) x3

=

ξ1 ξ2 ξ3 a1 a2

,

ξ1 > a1 > ξ2 > a2 > ξ3 > 0 .

(9.1)

It can be proved (Morse and Feshbach, 1953) that its coordinate surfaces are confocal, mutually orthogonal quadric surfaces, namely: • ellipsoids: ξ1 = constant; • one-fold hyperboloids: ξ2 = constant; • two-fold hyperboloids: ξ3 = constant. The degenerate cases, which complete the set of separable coordinates for the Helmholtz equation, are the following (in brackets, the corresponding coordinate surfaces): 1. orthogonal cartesian coordinates (planes); 2. circular cylindrical coordinates (circular cylinders; half-planes; planes); 3. elliptical cylindrical coordinates (elliptical cylinders; hyperbolic cylinders; planes); 4. parabolic cylindrical coordinates (two families of parabolic cylinders; planes); 5. rotation parabolic coordinates (two families of rotation paraboloids; planes); 6. paraboloidal coordinates (two families of elliptical paraboloids; one family of hyperbolic paraboloids); 7. spherical coordinates (spherical surfaces; cones; planes); 8. prolate spheroidal coordinates (prolate spheroids; two-fold revolution hyperboloids; planes); 9. oblate spheroidal coordinates (oblate spheroids; one-fold revolution hyperboloids; planes); 10. conical coordinates (spherical surfaces; one-fold hyperboloids; cones with elliptical cross-section). 9.3

Completeness of resonator modes

Solving the Helmholtz equation by separation of variables is legitimate, similarly to what we saw in Chapters 4 and 7, provided we can prove that the

© 2006 by Taylor & Francis Group, LLC

316

RESONANT CAVITIES

solutions found in this way form a complete set. Completeness of resonator modes can be proved similarly to what we saw in Chapter 4 for plane waves, and in Chapter 7 for waveguide modes. When we compared Chapter 4 and Chapter 7, we noticed that the dimension of a complete set decreased (from a continuum to a countable inﬁnity), as we passed from the very mild restriction that the ﬁeld be square integrable, to boundary conditions at ﬁnite distance along two directions. At a given frequency, the component of the propagation vector along a given direction can vary continuously in an unbounded medium, but is discretized in a waveguide. We may wonder now whether in a resonator (for simplicity, a closed one) there is a further reduction in the size of a complete set, as we impose boundary conditions along three directions. The answer can be given intuitively, going back to the case (see Section 9.1) of a cavity that consists of a ﬁnite length of a lossless cylindrical structure, whose transmission modes we assume to know. Two perfect conductors close the structure at its ends. The i-th mode of the cylindrical guide can travel either way; so, we must associate it with two arbitrary complex constants, Pi , Qi . They represent amplitudes and phases of the two counterpropagating waves, on the plane z = 0. The boundary conditions that have been imposed at the ends of the resonator can be expressed as two linear homogeneous equations in Pi , Qi . This system has non-trivial solutions if and only if its determinant vanishes. From this condition we obtain an equation that admits solutions only for some discrete values of the longitudinal propagation constant of the i-th mode βi , and then, as βi is a function of frequency, only for suitable discrete values of ω, ωi,j = 2πfi,j (j = 1, 2, . . .). In conclusion, a lossless resonator, at a ﬁxed frequency, has always a ﬁnite number of resonant modes. This number is zero, when the frequency we are referring to is not one of those particular fi,j ’s that were deﬁned a few lines above, which are obviously called resonant frequencies. The corresponding modes of an ideal resonator form a complete set, in the same sense as those discussed in Chapter 7. If we take a structure whose geometrical shape is identical to that of the ideal resonator, but whose boundary conditions are diﬀerent from those on a perfect electric conductor, then we know that a suitable series of modes of the ideal resonator converges, in the linear mean sense, towards any ﬁeld that may exist in the non-ideal structure. This applies not only to resonators with ﬁnite conductivity walls, but also to resonators coupled to other structures, by means of holes, irises, or similar devices; they are referred to as loaded resonators. Actually, a completely closed resonant cavity, without any access, is an easy model to analyze; but it would be useless, if we had not noticed that, as a consequence of mode completeness, the ﬁeld in a loaded resonator may be expressed as a series of modes of the isolated resonator. As we saw in Chapter 7, in a non-ideal waveguide, mode orthogonality collapses; therefore, the ﬁeld in general cannot be thought of as being a single mode in a rigorous sense. Modes under their cut-oﬀ frequencies are necessary

© 2006 by Taylor & Francis Group, LLC

MODE ORTHOGONALITY IN A PERFECT RESONATOR

317

to fulﬁl the boundary conditions, where the real waveguide departs from the ideal one. In the same way, in a loaded resonator, a mode of the corresponding ideal resonator may be present at frequencies which diﬀer from its resonant frequency fn,m . Its amplitude coeﬃcient, if all the other parameters are kept constant, decreases as |f − fn,m | increases, but in principle it may diﬀer from zero at any frequency. This point will be clariﬁed, in more quantitative terms, in Section 9.8; but we thought it might be useful to notice now that it is strictly related to the notion of completeness. 9.4

Mode orthogonality in a perfect resonator

Similarly to what was carried out in Section 7.5, we will prove an orthogonality theorem for the modes of a perfectly lossless, completely isolated cavity. The last assumption entails that there are no imposed currents, neither inside the cavity, nor on its walls. The proof holds for two modes with diﬀerent resonant frequencies. For two modes with the same resonant frequency (called degenerate modes), one can always orthogonalize them, with a procedure based on projections and subtractions (known as Schmidt’s orthogonalization); this is essentially the same as we saw for transmission modes in a perfect waveguide, and therefore it will not be repeated here. The proof of the orthogonality theorem is similar to that of Poynting’s theorem (Section 3.2); the main diﬀerence comes from the fact that we have to deal with two modes, having resonant frequencies ωh , ωk , respectively. Both satisfy homogeneous Maxwell’s equations: (a)

∇ × Eh = −jωh μ Hh

(b)

∇ × Hh = jωh Eh

(c)

∇ × Ek = −jωk μ Hk

(d)

∇ × Hk = jωk Ek

,

, ,

.

(9.2)

Dot-multiply both sides of equation (a) times H∗k , and the complex conjugate of equation (d) by Eh . Then subtract side by side, and use the vector identity (C.4), which we rewrite here for convenience: ∇ · (Eh × H∗k ) = H∗k · ∇ × Eh − Eh · ∇ × H∗k

.

(9.3)

The net result is a formula, similar to one which one ﬁnds on the way towards Poynting’s theorem, before making a volume integration. We integrate it over the cavity volume V , and then apply Gauss’ theorem (A.8) to the integral of the left-hand side of Eq. (9.3) to convert it into a surface integral. It is easy to show that this integral vanishes. In fact, the perfect conductor surrounding the cavity imposes Etan = 0 at any point. Following again the same path as in Poynting’s theorem, we notice that two more terms disappear: the one due to sources (there are none in the cavity, by assumption), and the one due to

© 2006 by Taylor & Francis Group, LLC

318

RESONANT CAVITIES

losses (again zero by assumption). So, we are left with the following relation: ωh μ Hh · H∗k dV = ωk Eh · E∗k dV . (9.4) V

V

If we repeat the same steps starting from Eqs. (9.2) (b) and (c), and take the complex conjugate for each term in this new relation, we get: ∗ μ Hh · Hk dV = ωh Eh · E∗k dV . (9.5) ωk V

V

Now let us multiply Eq. (9.4) times ωk , Eq. (9.5) times ωh , and then subtract; we get the ﬁnal result, namely: Eh · E∗k dV = 0 . (9.6) (ωh2 − ωk2 ) V

Multiplying Eq. (9.4) times ωh and Eq. (9.5) times ωk , we obtain a similar result for the magnetic ﬁelds. Thus if the two modes have diﬀerent resonant frequencies, the following orthogonality relations hold: Eh · E∗k dV = 0 (ωh = ωk ) , (9.7) V

μ Hh · H∗k dV = 0

(ωh = ωk )

.

(9.8)

V

As we saw in Chapter 7, orthogonality is a strong form of linear independence, with links with the property of completeness. It is also of great help in practical problems, like expanding a generic ﬁeld into a series of modes. Just to quote one example of application, it may be useful in calculating the ﬁelds excited by a source, either in the cavity, or localized on one of its walls. For more details, the reader should see some of the problems. 9.5

Lossless cylindrical cavities

Let us start, like in Section 7.2, from a cylindrical structure. We call z its axis, and a ˆz its unit vector. Inside the structure there is a homogeneous lossless medium; the lateral surface, C, is a perfect electric conductor. Let us then cut a ﬁnite length of this structure, and close it by means of two perfectly conducting planes, orthogonal to the z axis, at z = 0 and z = d, respectively (Figure 9.2). We saw in Section 7.2, and in the following ones, that each transmission mode of the inﬁnitely long cylindrical structure can be obtained from a scalar pre-potential (either L or T , depending on whether the mode is TE or TM). It is of the form F(u1 , u2 )·exp(−γz). But now, we introduced two discontinuities, orthogonal to the z axis. They cause reﬂections, which couple forward and backward propagating waves. A reasonable ansatz is then to assume that the pre-potentials are of the following type: F = F(u1 , u2 ) f (z)

© 2006 by Taylor & Francis Group, LLC

(F = L, T ; f = , t)

.

(9.9)

LOSSLESS CYLINDRICAL CAVITIES

319

Figure 9.2 A cylindrical cavity obtained from a waveguide of unspeciﬁed geometry.

Here, L(u1 , u2 ) and T (u1 , u2 ), which depend on the transverse coordinates, must satisfy the boundary conditions on C, Eqs. (7.17) and (7.18). Consequently, they remain the same as for TE and TM modes of the inﬁnitely long cylindrical structure. The only quantities which are still unknown are (z) and t(z). The conducting planes at z = 0 and z = d impose the following conditions, to both TE and TM modes: Et = 0

for z = 0 , z = d

.

(9.10)

Incidentally, notice that Eq. (9.10) is not the same as Ht = 0: the total ﬁeld in the cavity is the sum of forward and backward waves, whose wave impedances have opposite signs. The reader should compare this with what we saw in Chapter 8, and also in Chapter 4 on standing plane waves. Let us take now Eqs. (3.47) and (3.48), which apply to any TE or TM ﬁeld in a homogeneous medium, and introduce Eq. (9.9) into them. We get the following expressions for the ﬁeld transverse components: ˆz ) (z) ETEt = (∇t L × a

,

ETMt =

j dt(z) ∇t T ωc dz

.

(9.11)

for TE modes ;

(9.12)

for TM modes .

(9.13)

This shows that an alternative way to express Eq. (9.10) is: (0) = (d) = 0 dt dt = =0 dz z=0 dz z=d

The TE/TM terminology refers, as usual, to the z axis. But it is worth noticing that, within a given resonator, modes can be classiﬁed in diﬀerent ways, depending on which direction the term “transverse” refers to. In particular, this becomes evident in those cavities where the cross-section remains constant along two or three axes. We know from Section 7.2 that the functions F satisfy the modiﬁed Helmholtz equation (7.8), and from Chapter 3 that the functions F satisfy the complete

© 2006 by Taylor & Francis Group, LLC

320

RESONANT CAVITIES

Helmholtz equation. Consequently, the functions f (z) must satisfy: d2 f = γ2f (γ 2 = χ2 − ω 2 μ) , (9.14) dz 2 where χ2 is real and positive (see Section 7.4). Solutions to Eq. (9.14) are compatible with Eqs. (9.12) and (9.13) only for γ 2 < 0. In this case, letting γ 2 = −k 2 , the solutions we are looking for are: π (p = 1, 2, . . .) , (9.15) (z) = A sin kz , k=p d π t(z) = B cos kz , k=p (p = 1, 2, . . .) , (9.16) d where A, B are arbitrary constants.∗ From the pre-potentials L and T that have been determined in this way, we may now calculate the ﬁelds of the cavity resonance modes, using the generalpurpose formulas in Eqs. (3.47) and (3.48). But the reader can prove, as an exercise, that they can be found in a faster way by changing slightly Eqs. (7.8) and (7.9). These changes consist simply of replacing the exponential functions in Eqs. (7.1) with sine and cosine functions, Eqs. (9.15) and (9.16). Let us keep the same symbols as in Chapter 7 for the z-independent factors ¯ TEt and H ¯ TM have in the waveguide modes; in particular, let E¯TE , E¯TMt , H the same meanings as in Eqs. (7.11) and (7.12). Then, ﬁnally we derive the following expressions, where the subscript t stands, as usual, for transverse components with respect to the z axis: TEm,n,p mode : Ez ≡ 0 ,

Hz = Hz sin p

Et = E¯TE sin p

πz d

,

πz d

,

¯ TEt cos p Ht = H

πz d

.

(9.17)

TMm,n,p mode : πz , Hz ≡ 0 , d πz ¯ TM cos p πz . (9.18) , Ht = H Et = E¯TMt sin p d d Since γ 2 and χ2m,n are linked by Eq. (9.14), it must be: pπ 2 ω 2 μ = χ2m,n + . (9.19) d The results of this section deserve further comments. First of all, one should notice that, to build a resonance mode in a cavity obtained from a waveguide, Ez = Ez cos p

∗ The case p = 0, in Eq. (9.16), which entails E TMt ≡ 0, is compatible with Eq. (9.13), but it is acceptable only for χ2 = 0, i.e., if T is not constant vs. u1 and/or vs. u2 . If this is not the case, then p = 0 must be disregarded, because (like for p = 0 in Eq. (9.15)) it would correspond to a ﬁeld identical to zero.

© 2006 by Taylor & Francis Group, LLC

SIMPLE EXAMPLES

321

a waveguide mode above its cut-oﬀ frequency is required. Indeed, no nontrivial solution to Eq. (9.14) is compatible with Eqs. (9.12) and (9.13) if we suppose γ 2 < 0 in Eq. (9.14): a resonant mode can never be a superposition of two evanescent waves. As anticipated in Section 9.1, it is a superposition of two traveling waves, one forward, the other one backward, and it is straightforward to detect that indeed there are two expressions of the type exp(−jkz) ∓ exp(jkz) in Eqs. (9.15) and (9.16). Resonance occurs when the round-trip length, for the traveling wave that undergoes reﬂections at both ends of the cavity, is an integer multiple of the wavelength. Notice that, as one can easily infer from Chapter 8, this result depends on the phase shift at the reﬂections, at both ends, being equal to π radians. Should the cavity be terminated by “mirrors” with a diﬀerent phase shift, this should be accounted for. However, if we recall from Chapter 7 that, for the (m, n)-th waveguide mode, βm,n = 2π/λg , where λg is the guided wavelength, then the condition k = pπ/d can be rewritten as: 2d = pλg

,

(9.20)

where p is an integer. This reconﬁrms the previous statement. Any solution like Eq. (9.17) or (9.18) obeys Eq. (9.19), where χ2m,n is determined by the shape and the size of the waveguide. This reconﬁrms that, for a given cavity (i.e., once the transverse eigenvalues are given, and the length d is also given), Eq. (9.9) is satisﬁed only at discrete frequencies, f m,n,p . If f = fm,n,p , the ﬁeld is identically zero. This result applies only to a strictly lossless cavity, perfectly isolated from the rest of the universe. What happens in a real cavity has been described in Section 9.3. 9.6 9.6.1

Simple examples Rectangular resonators

Given a rectangular box, the z axis (with the meaning given to it in the previous sections) can be taken parallel to any edge. Diﬀerent choices for the z axis give rise to diﬀerent nomenclatures for the resonator modes, but such changes do not correspond to any physical diﬀerence. ¯ t and Hz , to be inserted into Eqs. (9.17), TE modes. The expressions for E¯t , H come from Eq. (7.50), through Eqs. (7.4) and (7.5). For the TEm,n,p mode, we get: Ez

≡ 0 ,

Hz

= H cos

Ex

=

nπy pπz mπx cos sin a b d

,

mπx nπy pπz jωμ nπ cos sin sin H 2 kc b a b d

© 2006 by Taylor & Francis Group, LLC

,

322

RESONANT CAVITIES

Figure 9.3 Rectangular resonator.

Ey

= −

mπx nπy pπz jωμ mπ sin cos sin H 2 kc a a b d

,

Hx

= −

mπx nπy pπz H pπ mπ sin cos cos 2 kc d a a b d

,

Hy

= −

mπx nπy pπz H pπ nπ cos sin cos 2 kc d b a b d

,

(9.21)

where a, b and d are the lengths of the three sides of the cavity (Figure 9.3); H is an arbitrary constant; kc2 = (mπ/a)2 + (nπ/b)2 is a symbol taken from Chapter 7. ¯ t and Ez , to be inserted into Eqs. (9.18), TM modes. The expressions for E¯t , H come from Eq. (7.54), through Eqs. (7.4) and (7.5). For the TMm,n,p mode, we get: nπy pπz mπx sin cos , Ez = E sin a b d Hz

≡ 0 ,

Ex

= −

mπx nπy pπz E pπ mπ cos sin sin kc2 d a a b d

,

Ey

= −

mπx nπy pπz E pπ nπ sin cos sin kc2 d b a b d

,

Hx

=

Hy

= −

mπx nπy pπz jω nπ sin cos cos E 2 kc b a b d mπx nπy pπz jω mπ cos sin cos E 2 kc a a b d

, ,

(9.22)

where the notation is the same as above; E is an arbitrary constant. The reader can ﬁnd, as an exercise, which, among the TE and TM modes, can have one or two integer subscripts equal to zero. No mode can have all three subscripts equal to zero.

© 2006 by Taylor & Francis Group, LLC

SIMPLE EXAMPLES

323

Figure 9.4 Cylindrical resonator of circular cross-section.

9.6.2

Cavities of circular cross-section

For a cavity shaped like a circular cylinder, of radius a and length d (Figure 9.4), we use of course cylindrical coordinates r, ϕ, z, the z axis being the cylinder axis. In such a case there is no ambiguity in the mode nomenclature; but this has no physical meaning. Indeed, one might think that the resonator shown in Figure 9.4 is a piece of radial transmission line, short-circuited at r = a, or a structure where waves go in loops in the azimuth direction, closing their path at Δϕ = 2π. ¯ t and Hz , to be inserted into Eqs. (9.17), TE modes. The expressions for E¯t , H come from Eq. (7.66), through Eqs. (7.4) and (7.5). For the TEn,m,p mode, we get: Ez Hz Er Eϕ Hr

Hϕ

≡ 0 ,

r cos nϕ pπz , = H Jn pn,m sin a sin nϕ d sin nϕ pπz r jωμ a2 Jn pn,m , n sin = H 2 − cos nϕ pn,m r a d r cos nϕ a pπz , = −jωμ H J p sin pn,m n n,m a sin nϕ d r cos nϕ H pπ pπz p J , = a cos n n,m p2 d a sin nϕ d n,m sin nϕ pπz r H a2 pπ Jn pn,m n cos = − 2 − cos nϕ pn,m r d a d

, (9.23)

where Jn is the ﬁrst derivative of the Bessel function of the ﬁrst kind, Jn (u), with respect to its argument u (see Appendix D); H is an arbitrary constant; pn,m , the m-th zero of Jn (u), is linked by Eq. (7.65) to the eigenvalue of the TEn,m mode in the circular waveguide the cavity was derived from. The functions in brackets stand for the two possible polarizations of each mode

© 2006 by Taylor & Francis Group, LLC

324

RESONANT CAVITIES

with n = 0. Upper or lower rows must be taken together, to get the correct ﬁeld components.

¯ t and Ez , to be inserted into Eqs. (9.18), TM modes. The expressions for E¯t , H come from Eq. (7.69), through Eqs. (7.4) and (7.5). For the TEn,m,p mode, we get: r cos nϕ pπz , Ez = E Jn pn,m cos a d sin nϕ Hz

≡ 0 ,

Er

=

Eϕ

=

Hr

=

Hϕ

=

r cos nϕ pπ pπz Jn pn,m , − a sin pn,m d a d sin nϕ sin nϕ r pπz E a2 pπ Jn pn,m n , sin − cos nϕ p2n,m rd a d sin nϕ r pπz jωE a2 Jn pn,m , n cos − 2 − cos nϕ pn,m r a d r cos nϕ jωE pπz , − a Jn pn,m cos pn,m a d sin nϕ E

(9.24)

where the symbols are the same as for Eqs. (9.23), except for the arbitrary constant E and for pn,m (m-th zero of Jn (u)), linked to the eigenvalue of the TMn,m mode of the circular waveguide by Eq. (7.68). Some examples of ﬁeld lines, for a few, simple modes of rectangular and circular resonators, are shown in Figure 9.5.

9.6.3

Spherical cavities

In Chapter 11 we will solve the Helmholtz equation using separation of variables in spherical coordinates (r, ϑ, ϕ). The main reason why this subject is postponed, is because its main applications deal with ﬁeld sources (like antennas). After studying Chapter 11, the reader should come back to this section, and impose on each spherical harmonic the boundary condition Etan = 0 on a spherical surface of radius r = a. In this way, the modes of a spherical resonator will be found. Another reasonably simple exercise is to impose the same condition also on coordinate planes ϕ = constant, or on coordinate cones ϑ = constant. What one obtains in this way are the modes of more complicated structures, which are of some interest in problems related to feeders for aperture antennas (see Chapter 12 and 13), or to tapers between waveguides of diﬀerent cross-section.

© 2006 by Taylor & Francis Group, LLC

LOSSY RESONATORS

325

Figure 9.5 Field lines of a set of signiﬁcant modes, in rectangular and circular resonators.

9.7

Lossy resonators: Perturbation analysis. Intrinsic Q-factor

From the idealized model of perfectly conducting walls, to a more realistic one where the walls are good conductors (σ ω, but ﬁnite), the step is very similar to that of Chapter 7 about waveguides. On the grounds of what we saw in Section 4.12, the surface which surrounds the resonator can be characterized in terms of its wall impedance: πμf [Ω] , (9.25) Z0 = (1 + j) σ

© 2006 by Taylor & Francis Group, LLC

326

RESONANT CAVITIES

which was originally derived for a ﬂat surface. As we saw in Chapter 7, one can prove that this result is correct also for cylindrical or spherical walls. On the other hand, the wall impedance becomes much more complicated (involving metric coeﬃcients of the coordinate system) if the curvature is not constant along the wall itself. On these premises, the ﬁrst order perturbation theory is very similar to that of Chapter 7. We start oﬀ supposing that a mode of the ideal cavity can exist in the cavity with ﬁnite wall impedance, although it cannot be rigorously so. We calculate the ohmic losses under this assumption. As in Chapter 7, special caution is needed in the case of degenerate modes: among all the linear combinations of two or more degenerate modes, which are still modes of the ideal cavity, one must identify those that remain orthogonal in the presence of losses. From lumped-element circuit theory we know that power dissipation in a resonant circuit, WJ , is of little signiﬁcance, unless correlated with the amount of energy which is stored in the circuit, E. The link between these two quantities is, also for a resonant cavity, the adimensional number: Q0 =

ω0 E WJ

,

(9.26)

which is referred to as the quality factor, or factor of merit, or ﬁgure of merit (or, for short, the Q-factor) of the resonator mode characterized by the resonance angular frequency ω0 . Let us recall (from Chapter 3) that, for a resonant ﬁeld, the time-averaged energies stored in electric and in magnetic form are equal and the total energy, constant in time, is equal to twice those averages. If one continues along the same examples as in Section 9.6, one sees that, as a general trend, the Q-factor in a given cavity tends to increase with the order of the resonance mode. This can be intuitively justiﬁed as follows. Consider ﬁrst a cavity with lossless walls. In it, the electric ﬁeld components of a given mode have nodes, i.e., surfaces on which the tangential components of E are identically zero. Some of these nodes are located on the resonator walls, but some may be inside. The number of internal nodes grows as the mode order grows. All the nodal surfaces can be metallized with perfect electrical conductors, without aﬀecting the ﬁeld; this operation is equivalent to subdividing the resonator into several cavities, each of which resonates in a lower order mode, compared to that of the original cavity. When we calculate the Q-factor, only the real (external) walls, being non-ideal, contribute to loss. So, the power dissipated in the “large” resonator is less than the sum of the powers that would be dissipated in all the “small” resonators if they would be really separated by the same imperfect conductor the real walls are made of. On the other hand, the total stored energy obviously equals the sum of the stored energies. Consequently, the Q-factor is larger in the large resonator, which is equivalent to saying that (at least within a given “family” of modes) it grows as the order of the mode grows. From circuit theory we know that in a lumped-element resonator the Q-

© 2006 by Taylor & Francis Group, LLC

RESONATORS COUPLED TO EXTERNAL LOADS

327

factor determines the full width at half maximum of the so-called resonance line, i.e., of the Lorentzian function

f (ω) = 1 + jQ

1 ω0 ω − ω0 ω

.

(9.27)

In other words, the Q-factor of a resonant cavity mode is a measure how diﬃcult it is to inject into the cavity a signal whose frequency is diﬀerent from the resonant one. 9.8

Resonators coupled to external loads. Loaded Q-factor

A completely isolated resonant cavity does not bear any practical interest. Any use of a resonator requires coupling it to other structures, that allow to inject energy into it. Some typical coupling tools at microwave frequencies are: • connections either with waveguides or with other cavities through holes called irises or diaphragms; • connections with transmission lines, or with lumped-element circuits, by means of conductive wires or probes, either straight or loop-shaped; • connections with other elements (transmission lines or other cavities) by means of electron beams which pass through the resonator. To evaluate the external coupling of a lossless resonator, suppose that the cavity is at steady state: it is fed by a source which compensates exactly for the power that the cavity loses to the outside world. We deﬁne as the external quality factor of the resonator (or better, of its mode characterized by the resonance frequency ω0 ) the adimensional quantity: Qex =

ω0 E Wex

,

(9.28)

where E, as in the previous section, is the energy stored in the resonator at steady state, while Wex is the active power leaving the resonator to the outside world. If we consider a passive resonator, as we have always done so far, then the quantity in Eq. (9.28), if inserted into Eq. (9.27), determines, as in the previous section, the spectral width of the Lorentzian function which characterizes the mode under test. It is useful to add that for a resonator containing sources, the Q-factor deﬁned by Eq. (9.28) determines, again through Eq. (9.27), a linewidth which is not that which one measures when the source is running in the large-signal regime. To stress this point, the quantity in Eq. (9.28) is often referred to as the cold cavity Q-factor. In a practical resonator, losses and coupling with the outside world must be accounted for at the same time. In all cases of interest, Q0 and Qex are both much greater than one, meaning that the energy stored in the cavity is much larger than the energy lost in the resonator over one period T = 2π/ω0 ,

© 2006 by Taylor & Francis Group, LLC

328

RESONANT CAVITIES

as well as than the energy delivered to the external load during the same time interval. One consequence is that internal losses and energy given to the external load may be thought of as additive, although in general this is not rigorous. Consequently, it makes sense to deﬁne a loaded quality factor as: QL =

ω0 E WJ + Wex

.

(9.29)

Its link with the other quantities deﬁned in this section and in the previous one is: 1 1 1 = + , (9.30) QL Q0 Qex the loaded Q-factor Eq. (9.39) determines, as usual through Eq. (9.27), the spectral width of the frequency response of a cavity. 9.9

Open resonators

The examples shown in Section 9.6 reconﬁrm that resonance in a cavity means satisfying a relationship between the geometrical dimensions of the cavity itself and the wavelength at the resonant frequency. When the geometrical shape of the cavity is simple, this relation is simple also. If the shape of the cavity is complicated, this relation becomes complicated but the concept remains the same. At optical frequencies (of the order of 1014 –1015 Hz), to make geometrical dimensions comparable with wavelengths becomes a technological challenge. It is necessary to think of resonators which operate in high-order modes, in contrast to what is normally the case for microwave cavities. If we take a cavity of a given shape and given size, and increase the frequency, then the growth in the resonance mode order is accompanied by an increase in the relative density of resonances, on the frequency axis (deﬁned simply as the number of modes, n, whose resonance frequencies fall within a unit length on the f axis, Δf = 1 Hz, divided by the frequency at which this interval is centered, f0 ). Note that the relative density is much more relevant than the absolute one, n, because n/f0 is the quantity to compare with the cavity Q-factors in order to establish whether in practice it is possible to operate a cavity in a single mode. Single-mode operation requires at the same time a low relative density of modes and a high Q. In a cavity with metallic walls, of the same type as those we dealt with so far, it is impossible to meet these requirements at optical frequencies. This is because at these frequencies there is no material which satisﬁes σ ω. To by-pass the problem, one may proceed as follows: • replace the metallic walls with dielectric mirrors, which are dielectric multilayers (see Section 4.10) designed to yield high reﬂection coeﬃcient at the wavelength of interest; • decrease the relative density of modes by eliminating some of the walls, making an open resonator, i.e., a 3-D region conﬁned between two highly

© 2006 by Taylor & Francis Group, LLC

OPEN RESONATORS

329

Figure 9.6 Schematic examples of open resonators.

Figure 9.7 A Gaussian beam in an open resonator.

reﬂecting surfaces along one direction, z, but extending to inﬁnity in the planes orthogonal to z. Some examples of open resonators, with plane mirrors (Fabry-P´erot cavities, see Section 4.10) and with spherical mirrors, are sketched in Figure 9.6, parts a) and b), respectively. Some rays are also sketched in the same ﬁgure, to illustrate how one can justify in intuitive terms high-Q-factor modes in an open resonator. Indeed, for suitable geometries one ﬁnds rays that are trapped in the cavity and undergo an inﬁnite number of reﬂections on the two mirrors in spite of the fact that these have ﬁnite transverse dimensions. As we shall see in the next section, some quantitative information may be grasped from the intuitive ray optics model. However, a much larger amount of information can be extracted from Gaussian beams. In fact, we saw in Chapter 5 that, far from the beam waist (|z| πw02 /λ, see Figure 9.7), the wavefronts of a Gaussian beam are essentially spherical, and their radius of curvature, as given by Eq. (5.96), is: 2 2 πw0 z . (9.31) R(z) = z 1 + λz If a beam impinges on a spherical mirror whose radius of curvature equals that of the wavefront, reﬂection gives rise to another Gaussian beam whose parameters are identical to those of the incident beam, and travels in the opposite direction. Intuition indicates that suitable combinations of two mirrors can make the beam go up and down indeﬁnitely in a stable way. Predictions based on this model can be made quantitative, for mirrors having any radius of

© 2006 by Taylor & Francis Group, LLC

330

RESONANT CAVITIES

curvature, by means of the “ABCD law” of Section 5.11. All the main parameters of a mode of an open resonator can be calculated in terms of Gaussian beams: resonance frequency, total stored energy and its spatial distribution, and also, with a few further notions that will be given in Section 9.11, the Q-factor. In particular, the resonance frequency can be calculated ensuring that the total phase shift along a closed path, taking into account phase shifts due to reﬂections on the two mirrors, has to be an integer multiple of 2π. Making use of the results of Section 5.7, this requirement may be written as: arg ρ1 + arg ρ2 + 2kL + 2[χ(z2 ) + ξ(z2 ) − χ(z1 ) − ξ(z1 )] = 2nπ

,

(9.32)

where z1 and z2 are the coordinates of the centers of the two mirrors; L = (z1 − z2 ) is the cavity length; n is an integer; χ and ξ have the same meanings as in Sections 5.9 and 5.10; ρ1 and ρ2 are the reﬂection coeﬃcients of the two mirrors, whose arguments are normally independent of the mode one deals with. The fourth term on the left-hand side of Eq. (9.32) is usually much smaller than the third, so it is neglected in ﬁrst approximations; however, it is very important in principle, as it contains the term ξ(z) which, as we saw in Section 5.10, depends on the transverse order of the Gaussian beam, labeled in Section 5.10 as (m1 , m2 ) for the Hermite-Gauss modes and as (, p) for the Laguerre-Gauss modes. It is just because of this term that the resonance frequency of the TEMm1 ,m2 ,n or TEM,p,n depends slightly on the transverse order, for a given longitudinal order. This aﬀects signiﬁcantly the emission spectrum of laser sources, whenever they may oscillate on more than one “transverse mode”. Let us stress that such a result would be impossible to obtain using ray optics. 9.10

Stability of open resonators

In the previous section we ﬁrst gave the motivations for using open resonators at optical frequencies, and then we saw how they can be studied in terms of Gaussian beams. These beams, as we saw in Chapter 5, have wavefronts that in principle extend to inﬁnity in all directions orthogonal to the z axis. On the contrary, mirrors of open resonators are obviously of ﬁnite size. This contrast entails two diﬀerent types of consequences, that will be the subjects of this section and of the next one. Normally, mirrors are far away from focal planes. In other words, they are located in regions where evanescent waves do not contribute in a signiﬁcant way to the Gaussian beam. Therefore, as a ﬁrst approximation we may think in terms of ray optics, in order to ﬁnd under what conditions a ray remains trapped in the cavity when it undergoes multiple reﬂections on the two mirrors. Let us consider then what is sketched in Figure 9.8. The cavity consists of two spherical mirrors, with focal lengths R1 /2 and R2 /2, located at a distance L from each other. Let us restrict ourselves to meridional rays, i.e., rays belonging to planes passing through the cavity axis z, like those shown

© 2006 by Taylor & Francis Group, LLC

STABILITY OF OPEN RESONATORS

331

Figure 9.8 Deﬁnitions of the geometrical parameters of an open resonator.

Figure 9.9 Ray reﬂection at a mirror at one end of an open resonator.

in Figure 9.8 which belong to the (x, z) plane. Just to ensure that there is no ambiguity, suppose that the n-th reﬂection of a given ray takes place on mirror 2, the (n + 1)-th on mirror 1, and so on. Let xn be the value of the transverse coordinate x at the site of the n-th reﬂection, and xi = (dx/dz)i (i = . . . , n − 1, n, n + 1, . . .) be the slope of the ray after the i-th reﬂection, with respect to the positive direction of the z axis.† The law of reﬂection, if applied on mirror 1 (Figure 9.9), gives the following equation (where the angles are small enough to be approximated by their tangents): xn+1 = −xn −

xn+1 f1

.

(9.33)

Furthermore, as the slope of a ray remains constant between two consecutive reﬂections, it is straightforward to write that: xn+2 − xn+1 = L xn+1

,

xn+1 − xn = −L xn

.

(9.34)

Replacing Eq. (9.33) into the ﬁrst of Eqs. (9.34) and eliminating xn by means of the second of Eqs. (9.34), rearranging the terms, we obtain the † Note that the signs of x depend on the sense in which the ray propagates: x n+1 > 0 i means that, traveling from mirror 1 towards mirror 2, x is increasing. On the contrary, xn > 0 means that, traveling from mirror 2 towards mirror 1, x is decreasing.

© 2006 by Taylor & Francis Group, LLC

332

RESONANT CAVITIES

following equation, without derivatives:

L xn+2 + xn = xn+1 2 − f1

.

(9.35)

It is self-explanatory why the same equation holds for the previous trips, those in one direction: L xn + xn−2 = xn−1 2 − , (9.36) f1 as well as those in the other direction:

L xn+1 + xn−1 = xn 2 − f2

.

(9.37)

Adding Eqs. (9.35) and (9.36) term by term, and eliminating from the righthand side the term xn+1 +xn−1 by means of Eq. (9.37), we obtain the following ﬁnite diﬀerence equation: xn+2 − 2b xn + xn−2 = 0 , where we set: b=1−

L L L2 L L − + =2 1− 1− −1 . f1 f2 2f1 f2 2f1 2f2

(9.38)

(9.39)

The ﬁrst thing of interest is to know whether Eq. (9.38) has stable solutions, that is whether there are ray trajectories for which the value of the transverse coordinate x does not diverge as the number of reﬂections increases indeﬁnitely. For this purpose we let: xn = x0 ejnϑ

,

(9.40)

where the unknown ϑ has to be thought of, for the sake of generality, as being complex. Substituting Eq. (9.40) into Eq. (9.38), we ﬁnd: ej2ϑ + e−j2ϑ = 2b ,

(9.41)

2ϑ = arccos b .

(9.42)

whose solution is: Hence, ϑ is real if |b| < 1. In this case, Eq. (9.40) is a periodic solution of Eq. (9.38). As long as a mode of the cavity can be reasonably approximated by plane waves obeying the ray-optics laws, the previous result indicates that these waves remain conﬁned inside the cavity for |b| < 1. For |b| > 1, on the contrary, ϑ becomes imaginary, and so the transverse coordinate Eq. (9.40) diverges exponentially as the number of round trips through the cavity increases. This indicates that, after a ﬁnite number of reﬂections, the generic ray will miss the mirrors and leave the cavity. The ﬁrst case is referred to as a stable resonator, the second one as an unstable resonator. It is useful to plot the regions where |b| > 1 or |b| < 1 on the cartesian

© 2006 by Taylor & Francis Group, LLC

Q-FACTOR OF AN OPEN RESONATOR

333

Figure 9.10 The stability diagram for open resonators.

plane {L/2f1 , L/2f2 }. In Figure 9.10, the blank regions correspond to geometries that yield stable resonators, while the dashed regions correspond to unstable resonators. Negative values of one of the coordinates correspond to a convex mirror; so, points whose coordinates are both negative (a pair of convex mirrors) are all in the region of instability, in agreement with intuition. They do not ﬁnd any practical application. On the other hand, cavities which are moderately unstable (|b| slightly larger than 1) are used in laser technology, when high-power output beams are required. Indeed, for a given beam power, the energy stored in an unstable cavity is signiﬁcantly smaller than in a stable one, and this can simplify problems related to materials. The price one pays when the stability is lowered (that is, when |b| is increased) is a reduced mode selectivity, that is, a lower quality (both vs. time and vs. spatial coordinates) of the output beam. 9.11

Q-factor of an open resonator

Also for open cavities it is of great interest to deﬁne an adimensional quantity, called merit or quality factor or, for short the Q-factor. It can still be deﬁned by multiplying the resonance frequency by the energy stored in the resonator, and dividing it by the time-averaged power loss. There are no diﬀerences with respect to what we saw in other sections from the viewpoint of principles, but now it is more diﬃcult to separate power lost in the resonator from that delivered outside (see Sections 9.7 and 9.8), a distinction which allowed us to deﬁne an unloaded Q-factor and a loaded Q-factor for closed resonators. In an open resonator, both stable and unstable, mirrors have ﬁnite diameters, while, as we said before, wavefronts of Gaussian beam extend indeﬁnitely. Therefore, every reﬂection is accompanied by loss of that amount of

© 2006 by Taylor & Francis Group, LLC

334

RESONANT CAVITIES

power which corresponds to the part of the beam which travels at distances from the cavity axis larger than the mirror size. This power goes obviously out of the cavity, but must be treated conceptually as lost power, because it is not delivered outside in any usable way. Consequently, it is included (together with absorption either within the cavity or on the mirrors) in the unloaded Q-factor. The case of one (or two) partially transparent mirror is completely diﬀerent. If a mirror reﬂects only a fraction of the incident power, the complementary part can be delivered to an external beam. This power passing through the partially transparent mirror has to be accounted for in the deﬁnition of an external Q-factor. When the losses due to ﬁnite size of the mirrors and those due to absorption are small with respect to the energy stored in the resonator, the two contributions to the unloaded factor Q0 can be added, following the rule used in Section 9.8. The low-loss assumption ﬁts always well stable cavities, provided the mirrors are made with low-loss materials. Then, we may write: 1 1 1 = + Q0 Qa Qd

,

(9.43)

where Qa = ω0 E/Wa is the Q-factor due to absorption (Wa = time-averaged absorbed power), while Qd is the contribution due to losses at mirrors edges. To ﬁnd a way to express these losses, one needs diﬀraction theory, to be dealt with in Chapter 13. Here, we will just quote the formulas we need now, without any proof. All the quantities that we need can be expressed as functions of an adimensional parameter, usually referred to as the Fresnel number (Born and Wolf, 1980) which, for two spherical mirrors, can be deﬁned as: a1 a2 , (9.44) N= λL where λ is the wavelength, L is the cavity length, a1 and a2 are the radii of the cross-sections of the mirrors (see Figure 9.8). The fraction of the power of a Gaussian beam, τ , which is lost in one reﬂection on the mirror, as a function of N , is well approximated by simple expressions for N 1 (diﬀraction losses small compared to the other losses of interest) and for N 1 (that means diﬀraction losses dominant on the others). Respectively, we have: τ 16π 2 N e−4πN and:

τ 1 − π2N 2

(N 1) (N 1) .

,

(9.45) (9.46)

A simple expression for Qd as a function of τ (Ramo et al., 1994) can be easily linked to intuitive properties of Gaussian beams that we saw in Chapter 5. If the radii of curvature of the mirrors are large enough, then the energy distribution in the cavity, on a transverse plane, is almost independent of the longitudinal coordinate, and the ratio between average and maximum values of the energy density depends only on the order of the mode. The

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

335

Poynting vector is equal to the product of the energy density times the group velocity, that we can assume equal to the speed of light in the medium which ﬁlls the cavity. Therefore, the energy stored in the cavity: E = πw02 L U

,

(9.47)

(w0 = waist width of the Gaussian beam) and the power incident on the mirror: (9.48) Wi = πw02 c U , can be easily expressed as functions of the average energy density, U . Taking now into account the fact that power loss at any reﬂection, Wd , is the product of τ times Eq. (9.48), we ﬁnally get: Qd =

2πL ω0 E ω0 L = = Wd τc τ λ0

,

(9.49)

where λ0 = 2πc/ω0 is the resonance wavelength. So, the quality factor due to diﬀraction losses is, if we keep all other conditions constant, proportional to the number of wavelengths L/λ0 contained in the cavity in the axial direction, i.e., to the integer number n of Eq. (9.32) (“longitudinal order” of the resonance mode). This agrees with the statements of Section 9.7, about the Q-factor growing when the order of the mode grows, and justiﬁes why very high values can be obtained without too much diﬃculty for Q-factors at optical frequencies. To ﬁnish this section, it may be useful to recall again that all the deﬁnitions and calculations in the sections on open resonators (Section 9.9 and following) refer to cavities ﬁlled with a passive medium. In the language of laser technology, these are called cold cavity modes. When cavities are ﬁlled with an active medium, capable of amplifying the resonant wave (the necessary energy comes from other sources), then there are important changes both in the Q-factor, with respect to that deﬁned in this section, and in the oscillation frequencies, which are “pulled” by the active medium to diﬀerent values with respect to those given by Eq. (9.32). These problems are largely discussed in books on lasers (Svelto, 1989; Yariv, 1976). 9.12

Further applications and suggested reading

We take it for granted that the background in basic physics of all the readers includes resonances in other types of passive systems, typically mechanical ones. The relationship between them and what we have seen in this chapter is not just a vague analogy. The links are very fundamental. Just to quote one of them, a mechanical system can be resonant only if it can store energy under two diﬀerent forms at diﬀerent times in its periodic motion (e.g., kinetic and potential in a pendulum, kinetic and elastic in a spring). Resonance occurs when these two energies balance each other, exactly like electric and magnetic energies in resonant cavities. It may be of some interest, at least for those readers who have an incli-

© 2006 by Taylor & Francis Group, LLC

336

RESONANT CAVITIES

nation for the History of Science, to know that the mathematical treatment of resonant cavities, which is rather sophisticated compared to that of simple mechanical systems, has been developed long ago, up to a very advanced level, in the framework of acoustics. The very famous treatise Theory of Sound by Lord Rayleigh, now more than one century old, remains a masterpiece of scientiﬁc literature as well as a very important marker in the advancement of science. It is still a highly recommended, very instructive reading. After these suggestions related to classical physics, we also invite the readers to strengthen their background on resonant systems in Quantum Mechanics, typically, harmonic oscillators (Louisell, 1960). An important analogy in this direction is that between discrete states in a quantized system, and discrete modes in a resonator. The two systems are eigenvalue problems, and it is immediately noted that energy and frequency are proportional, through the Planck constant. Concepts like mode degeneracy and perturbation theory (like the one we used for calculating Q-factors), both in this chapter and in the previous one, may become much more familiar through a cross-correlation with elementary Quantum Mechanics. Going back to e.m. resonators, it is rather surprising that microwave resonators and optical resonators were studied in completely separate textbooks for decades, in spite of the fact that both are easy to explain, at least qualitatively, in terms of transmission lines (Chapter 8). It is even more surprising since one of the main applications of resonators, ﬁltering, was shared by the two frequency ranges. Probably, this was because most of the experts in microwaves, being electrical engineers, had a strong background in circuit theory, and so they designed their devices using mathematical tools originating from this area, typically those based on topological matrices and on Laplace transforms. On the other side, most of the physicists who worked in optics preferred to use frequency-domain approaches. To see this contrast, one may consult, for example, the sections devoted to ﬁltering properties of Fabry-P´erot ´etalons in Born and Wolf (1980), on one side, and Chapters 4 and 5 of Montgomery, Dicke and Purcell, Eds., (1948) on the other side. Nowadays, thanks to the success of optics in telecommunications, it is much easier to ﬁnd the two subjects side by side. However, this comment was not inserted simply as a historical detour. It was intended to remind that, if one has to design resonators for ﬁltering purposes, then it may be advisable, irrespectively of the frequency band, to know at least something in the area of network theory. A sequence of readings in this sense may consist of Desoer and Kuh (1969), Collin (1966) and Howe (1974), to fulﬁl the basic needs and possibly trigger further interest. Expert readers may include dielectric resonators, a subject that was omitted here only because of space limits. One may consult for example Kajfez and Guillon, Eds., (1986). Other readers may wish to be introduced to computer-aided design of ﬁltering devices. At microwave frequencies, one may then consult Gupta et al. (1981). Another practical application of resonant cavities, comparable in importance to ﬁltering, is positive feedback, as required in order to build an oscil-

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

337

lator. This extends to microwave and optical frequencies the role played in conventional electronics by lumped-element LC circuits. However, it has been stressed many times in the previous sections that an oscillating device can no longer be modeled as a purely electromagnetic system. Indeed, the ﬁeld in it is coupled to something (e.g., an active medium in lasers, or a chargedparticle beam in microwave tubes) which provides gain to the ﬁeld. One must be warned that this coupling aﬀects some of the most important characteristics of the ﬁeld. Ramo et al. (1994) is a recommended source of introductory information on this point, for the microwave case. Textbooks on lasers (e.g., Svelto, 1989) should be consulted for the optical-frequency case. In this framework, it may be interesting to look again at the relationship between microwave and optical resonators. Section 10.15 of Ramo et al. (1994) outlines why it took the work of several outstanding scientists, to grasp the advantages of open resonators, compared to closed ones, when the ﬁrst lasers were built. Another important point is that the type and the quality of the resonator aﬀect both temporal and spatial coherence of the sources. As the subject of coherence will be introduced in Chapter 14, readers are advised to come back to the present chapter after reading that one, and link the two together. Problems 9-1 Show that for any resonant mode of a closed cavity with perfectly conducting walls, all the ﬁeld components can be obtained following the same approach as for the propagation modes in waveguides, provided one replaces the exponential functions in Eqs. (7.1) with either sine or cosine functions, following the indications of Eqs. (9.15) and (9.16). 9-2 Discuss which among the TE and the TM modes of a rectangular resonator can have one or two subscripts equal to zero. Repeat the problem for the modes of a resonator with circular cross-section. Establish a link between these two answers, describing intuitively how the lines of force of the electric and magnetic ﬁelds of a given mode in a rectangular resonator deform into those of the corresponding mode in a circular resonator, using a cavity of elliptical cross-section as an intermediate step. 9-3 If one or two ratios between side lengths, in a rectangular resonator, are rational numbers, then we may expect some modes to become degenerate. a) Find which modes are degenerate when a = 2b = d/2. b) Among these sets of degenerate modes, only some satisfy orthogonality conditions like Eqs. (9.7) and (9.8). Discuss which do and which do not. 9-4 The fundamental mode in a resonator is clearly deﬁned as the mode with the lowest resonance frequency. In a rectangular cavity, the ordering of the resonant frequencies depends on the inequalities between the side lengths.

© 2006 by Taylor & Francis Group, LLC

338

RESONANT CAVITIES

a) Keeping a > b, discuss which is the fundamental mode as a function of d. b) Find a design rule that maximizes the spacing between the resonance frequency of the fundamental mode and that of the next higher-order mode. 9-5 In a resonator of circular cross-section, resonant frequencies are ordered in a way which depends on the ratio between the length and the diameter. Repeat, for this case, the two questions of the previous problem. 9-6 In a spherical coordinate frame, impose the boundary condition Etan = 0 ﬁrst on the coordinate planes ϕ = constant, and then on the coordinate cones ϑ = constant. Find the corresponding solutions of Maxwell’s equations. Discuss the physical structures in which these solutions exist as resonant modes or as propagation modes. (Hint. Restrict the discussion to TE and TM waves, for which partial derivatives with respect to ϕ are identically zero). 9-7 Using the contents of Section 7.10, and with the help of a suitable handbook dealing with Mathieu’s functions (e.g., Abramowitz and Stegun, 1965), express the longitudinal ﬁeld components for the resonant modes of a cylindrical cavity with elliptical cross-section, with an ideal conducting wall. 9-8 Express the intrinsic Q-factors of the generic TE and TM mode of a cubic resonator (a = b = d) whose wall impedance is expressed by Eq. (9.25). In particular, discuss how the Q-factor of a given mode depends on the resonant frequency, which in turn depends on the cavity size. 9-9 As sketched in Section 9.1, a piece of coaxial cable whose ends are sortcircuited resonates at those frequencies where its length equals a half wavelength. a) Express the intrinsic Q-factor due to losses in the good conductor which is used to make the two cylinders and the two ﬂat terminal surfaces. b) What happens to the intrinsic Q-factor if the cavity resonates in a higherorder longitudinal mode ? (Hint. Do not forget the loss contributed by the two terminal ﬂat walls.) 9-10 Calculate the intrinsic Q-factor of a coaxial cavity, for an outer radius a = 5 mm, a characteristic impedance of the cable of 120 Ω, a dielectric permittivity inside the cable equal to 0 , copper walls (conductivity: 5.8 × 107 Ω−1 m−1 ), resonant frequency 600 MHz. 9-11 On some occasions, losses in the dielectric which ﬁlls a resonant cavity may contribute signiﬁcantly to the intrinsic Q-factor. a) Assuming that the conducting walls are ideal, express the intrinsic Q-factor due to a dielectric with a small loss tangent, δ, for the modes of the cubic resonator of Problem 9-8, and for the TEM modes of the coaxial cavity of Problem 9-9.

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

339

b) Which is the expression of the intrinsic Q-factor that accounts for both conductor and dielectric losses, assuming that they are small enough to be additive ? 9-12 Tunable resonators are often used to measure frequencies in current microwave practice. They are usually referred to as wavemeters. The higher the Q-factor, the sharper the resonance, and so, quite obviously, the more accurate the frequency measurement. This is one of the reasons why cavities of circular cross-section, resonating in a low-loss TE0,1,p mode (see Section 7.11), are usually preferred to rectangular cavities. To support this statement, calculate the intrinsic Q-factor of the fundamental mode of a cubic cavity (see the previous problem) resonating at 3 GHz, and that of a cylindrical cavity whose radius equals the side length of the previous cavity and resonates in the TE0,1,1 mode at the same frequency. Assume that the wall surfaces are covered with a galvanic silver layer (conductivity) thicker than the skin depth. 9-13 Another reason why cavities of circular cross-section are preferred as wavemeters is because there are no currents ﬂowing through the contact between the cylindrical wall and the sliding piston which makes the resonator tunable. To support this statement, ﬁnd the current densities in all the walls of a circular-cylinder cavity, in the TE0,1,1 mode. Are there other modes of the same cavity which enjoy the same property of not having currents ﬂowing through that contact ? 9-14 At microwave frequencies, one method of measuring the permittivity of a dielectric material is to observe the shift in the resonance frequency, Δω, when a small sample of the material is placed inside a given cavity. In fact, following a variational approach, Slater proved that Δω/ω0 = ΔU/U0 , where ΔU is the change in the energy stored in the cavity caused by the presence of the sample, while ω0 and U0 are the values which correspond to the cavity without the sample. a) To enhance the sensitivity of the measurement, is it convenient to place a given dielectric sample near a maximum of the electric ﬁeld or near one of the magnetic ﬁeld ? b) Assuming that the ﬁeld remains unperturbed when the small sample is introduced, evaluate Δω for a cubic sample having r = 3, 3 mm in side, placed in the center of a rectangular cavity with a = 2b = d = 30 mm, ﬁlled with a material having r = 1.5 and resonating in the TE101 mode. 9-15 In Chapter 4 we discussed a device, shown in Figure 4.10, which was referred to as the Fabry-P´erot interferometer. Actually, this name is usually given to any structure which consists of two ﬂat parallel mirrors, separated by a homogeneous medium. It is often used as a frequency (or rather, wavelength) meter at optical frequencies, by making it possible to vary continuously the distance between the two mirrors.

© 2006 by Taylor & Francis Group, LLC

340

RESONANT CAVITIES

a) To increase the resolution of the frequency measurement, the device is fabricated with dielectrics which exhibit very low loss. What determines then the spectral width of the resonance line, Eq. (9.27) ? b) Evaluate this external Q-factor of the open resonator. (Hint. Model the device as a piece of lossless transmission line, with a characteristic impedance Z1 , and a length equal to a multiple of λ/2, sandwiched between two inﬁnitely long transmission lines, with characteristic impedances Z2 .) c) How does the external Q-factor vary as a function of the diﬀerence between Z1 and Z2 ? 9-16 The mirrors which deﬁne the cavity in a typical semiconductor laser are simply the discontinuities between the high-index semiconductor chip, and air. a) Assuming a refractive index n = 3 (note that in Eq. (9.32) it is k = nk0 , where k0 is deﬁned in free space), calculate the resonant frequencies and the corresponding free-space wavelengths of the longitudinal modes (i.e., neglect the χ and ξ terms) for a chip length L = 300 μm. b) For the lowest-order answer to the previous question, estimate the frequency separation between two consecutive Hermite-Gauss modes, taking w0 = 5 μm at the beam waist. 9-17 Find one or more physical arguments to justify why (see Figure 9.10): a) an open resonator with two convex mirrors (f1 and f2 both negatives) is always unstable; b) a convex mirror and a concave one can make a stable resonator, provided their distance L is not too large. 9-18 Consider a Helium-Neon laser, emitting at λ = 0.633 μm. a) Show that, even when the mirrors are so small that diﬀraction losses are extremely high (i.e., for τ close to unity), the estimated Q-factor, based on Eq. (9.49), is quite high if the cavity length is in the range of 100 mm. b) Show that mirror radii in the millimeter range are suﬃcient to let Qd grow by many orders of magnitude.

© 2006 by Taylor & Francis Group, LLC

References Abramowitz, M. and Stegun, I.A. (Eds.) (1965) Handbook of Mathematical Functions. Dover, New York. Born, M. and Wolf, E. (1980) Principles of Optics: Electromagnetic Theory of Propagation, Interference and Diﬀraction of Light. 6th (corrected) ed., reprinted 1986, Pergamon, Oxford. Collin, R.E. (1966) Foundations for Microwave Engineering. McGraw-Hill, New York. Desoer, C.A. and Kuh, E.S. (1969) Basic Circuit Theory. McGraw-Hill, New York. Gupta, K.C., Garg, R. and Chadha, R. (1981) Computer-Aided Design of Microwave Circuits. Artech House, Dedham, MA. Howe, H. (1974) Stripline Circuit Design. Artech House, Dedham, MA. Kajfez, D. and Guillon, P. (Eds.) (1986) Dielectric Resonators. Artech House, Dedham, MA. Louisell, W.H. (1960) Coupled Mode and Parametric Electronics. Wiley, New York. Montgomery, C.G., Dicke, R.H. and Purcell, E.M. (Eds.) (1948) Principles of Microwave Circuits. McGraw-Hill, New York. Reprinted by Peter Peregrinus, London, for IEE (1987). Morse, P.M. and Feshbach, H. (1953) Methods of Theoretical Physics. McGraw-Hill, New York. Ramo, S., Whinnery, J.R. and Van Duzer, T. (1994) Fields and Waves in Communication Electronics. 3rd ed., Wiley, New York. Svelto, O. (1989) Principles of Lasers. 3rd ed., translated from Italian and edited by D.C. Hanna, Plenum, New York. Yariv, A. (1976) Introduction to Optical Electronics. Holt, Rinehart and Winston, New York.

341 © 2006 by Taylor & Francis Group, LLC

CHAPTER 10

Dielectric waveguides 10.1

Introduction

In all the waveguides dealt with in Chapter 7, the dielectric medium where the guided ﬁeld propagates was assumed to be homogeneous. In other structures, which may also be of practical interest, the medium is not homogeneous; for example, this is the case for microwave devices made with ferrite-loaded waveguides, of the same kind as those that were mentioned in Chapter 6. Still, these guides have a boundary deﬁned by a cylindrical conducting surface. Their analysis is similar, at least with reference to the fundamentals, to what we saw in Chapter 7, although some changes are not trivial. There is still another set of wave guiding structures, whose essential feature is no metallic wall to conﬁne the ﬁeld within a closed region. The dielectric medium does in principle extend to inﬁnity, in the directions orthogonal to the waveguide axis. What drives the ﬁeld energy along a given direction is a suitable set of inhomogeneities — either sharp discontinuities, or smooth variations — in the permittivity of the medium. These structures diﬀer radically from those of Chapter 7, and we must approach them in another way. The theoretical foundations of dielectric waveguides were established in the early decades of the 20th century. However, the space devoted to them in standard Electrical Engineering curricula was extremely marginal until 1970. What made them quite popular, subsequently, was the explosion of ﬁber optics in long-haul telecommunications, and the strictly correlated research eﬀort in integrated optics. Interest in exploiting optical frequencies for telecommunications had been stimulated, in the early ’60’s, by the invention of the laser. If one thinks of guiding waves at optical frequencies along a given path, one easily sees that there are at least three very important reasons why the metallicwall waveguides of Chapter 7 are absolutely unsuitable. Firstly, the size of the guide would be too small, both for manufacturing and handling it. Secondly, conductors would introduce too much loss; actually, the correct physical picture of the interaction between an e.m. ﬁeld and charge carriers at the surface of a conductor, at optical frequencies, is not simple, and requires quantummechanical methods. Thirdly, for a link length and a bandwidth of technical relevance, dispersion in a metallic waveguide would be prohibitively large. In this chapter, we will lay down the theoretical foundations of how e.m. waves are guided either by step discontinuities, or by graded inhomogeneities. When this can be done, we will make the analytical approach as similar as possible to that of Chapter 7. In some cases, however, we will be forced to 343 © 2006 by Taylor & Francis Group, LLC

344

DIELECTRIC WAVEGUIDES

adopt approximate techniques. In these instances, we will follow as closely as possible some ideas that were outlined in Chapter 5, e.g., the WKBJ technique and geometrical optics. However, we will point out that in general all these approximations become accurate only when the transverse dimensions of the waveguides are fairly large compared to the wavelength. Consequently, we will be reasonably satisﬁed with some analytical solutions only for highly multimode waveguides, not for single-mode or few-mode ones. In these cases, we will face a rather “black or white” situation, where either we are able to ﬁnd exact solutions, or we have to resort to numerical methods. Because of the wide technical interest in optical ﬁbers, we will deal explicitly with waveguides where the magnetic permeability is constant throughout the structure, and normally equal to that of free space. However, all the results of this chapter can easily be translated, by duality, into results which apply to structures where the dielectric permittivity is constant and the guiding discontinuities aﬀect the magnetic permeability μ. This chapter is organized as follows. We will commence with an unspeciﬁed geometry, where the dielectric permittivity exhibits just one step discontinuity. We will show that a structure of this kind supports guided modes (a ﬁnite number, at any ﬁnite frequency) and unbounded modes (a continuum). With the help of two examples — slab waveguides and cylindrical rods — we will show that only a subset of the guided modes are TE and TM, while the others modes are hybrid. Then, we will outline why the exact analytical modes of step-index circular waveguides are too cumbersome, and how they can be replaced by approximate solutions, whose most interesting feature is that their transverse ﬁelds are linearly polarized. Next, we will elaborate further on dispersion in dielectric rods, a very important issue in characterizing optical ﬁbers. We will subsequently proceed to graded-index waveguides, show how they can be studied with approximate analytical techniques, and discuss brieﬂy their performance limitations. Finally, we will just touch upon losses in dielectric waveguides. 10.2

Waves guided by a surface of discontinuity. The characteristic equation

The mathematical approach to an inhomogeneous dielectric waveguide varies, depending on whether the structure consists of a ﬁnite number of homogeneous media, or of a medium whose permittivity varies continuously as function of transverse coordinates. It is convenient to deal ﬁrst with waveguides made of several homogeneous materials, starting from the simplest case, that of two lossless dielectric media, referred to as core and cladding, separated by a cylindrical surface, C, parallel to the z axis. Suppose that the intersection between C and the plane z = constant is a regular closed curve, . At each point on we may deﬁne in an unambiguous way three mutually orthogonal unit vectors, tangent, normal and binormal to , denoted by tˆ, n ˆ, a ˆz , respectively. The last one is also the unit vector in the direction of the z axis. They are related as

© 2006 by Taylor & Francis Group, LLC

WAVES GUIDED BY A SURFACE OF DISCONTINUITY

345

Figure 10.1 A dielectric waveguide of unspeciﬁed geometry, consisting of two homogeneous media.

tˆ × n ˆ = a ˆz (see Figure 10.1). Let co , cl be the permittivities of the inner and outer medium, with co > cl , and μ be the permeability of both media (when not speciﬁed otherwise, suppose μ = μ0 ). We will also use the following symbols, whose meanings are simple extensions of other symbols used in the √ previous chapters: −κ2i = ω 2 μi , βi = ω μi (i = co, cl). Within each of the two regions, since the medium in it is homogeneous and there are no sources, we are entitled to apply the decomposition theorem of a generic ﬁeld into a TE and a TM ﬁeld, which was proved in Section 3.8 and, in a simpliﬁed way which is suitable for the case at hand, in Section 7.2. Indeed, we will retain the basic ansatz of Chapter 7, that all the unknown functions (pre-potentials and ﬁelds) depend on the longitudinal coordinate through an exponential function, in general of complex argument, exp(−γz). Italic symbols will be used, as in Chapter 7, for the functions of the transverse coordinates where the exponential dependence on z has been factored out. The decomposition theorem also tells us that the longitudinal ﬁeld components obey eigenvalue equations like Eq. (7.8), namely: ∇2t Fi = −χ2i Fi

(i = co, cl)

,

(10.1)

χ2i = (γ 2 + ω 2 μi ) = (γ 2 − κ2i )

.

(10.2)

where Fi is either Hiz or Eiz , and:

The logical steps to take from now on are: • to write the total ﬁeld, in the core and in the cladding, as the sum of the TE and the TM ﬁelds; • to express the continuity conditions of the tangential ﬁeld components, Eqs. (1.43) and (1.44), on the core-cladding interface surface, in terms of these ﬁelds; • to study whether this system has solutions, and whether they are discrete, or form a continuum. The ﬁrst step is elementary, regardless of the waveguide shape: E¯i = E¯TEi + E¯TMi

,

¯ i = E¯TEi + E¯TMi H

(i = co, cl)

.

(10.3)

The second step is simple as long as we do it using the general-purpose

© 2006 by Taylor & Francis Group, LLC

346

DIELECTRIC WAVEGUIDES

symbols that have been introduced so far: tˆ · (E¯TEco + E¯TMco )

=

tˆ · (E¯TEcl + E¯TMcl ) ,

ζˆ · (E¯TEco + E¯TMco )

=

ζˆ · (E¯TEcl + E¯TMcl ) ,

¯ TEco + H ¯ TMco ) tˆ · (H

=

¯ TEcl + H ¯ TMcl ) , tˆ · (H

¯ TEco + H ¯ TMco ) ζˆ · (H

=

¯ TEcl + H ¯ TMcl ) ζˆ · (H

.

(10.4)

Comparison between this set of four equations and the boundary conditions in a metallic waveguide, Eqs. (7.17) and (7.18), shows immediately that the mathematical complexity of this problem is much greater than that of Chapter 7. For all the ﬁeld components involved in Eqs. (10.4) we might write the general expressions, either Eqs. (7.4) or Eqs. (7.11) and (7.12). This would give us the correct perception that the system in Eqs. (10.4) contains four unknown functions, Hiz and Eiz , in addition to the parameter γ which is still undetermined. However, from a practical viewpoint this is an insigniﬁcant progress, unless the following condition is satisﬁed: the interface surface between core and cladding, C, is a coordinate surface in a reference frame where the Helmholtz equation is separable. To grasp the actual meaning of this point, the reader should return to Section 7.6. It tells us that the geometries which can be studied in closed form are very few: planar slabs, circular-core and elliptical-core waveguides (parabolic waveguides are never found in practice). Suppose then that we are in one of these fortunate cases, and let (u1 , u2 , z) be the suitable coordinates, (h1 , h2 , h3 = 1) be their metric coeﬃcients, and u1 = a = constant be the equation of the coordinate surface C. Then, each of Eqs. (10.1) can be split, as we saw in Chapter 7, into two second-order ordinary diﬀerential equations (ODE’s). So, in general every function Fi is a product of a linear combination of two functions of u1 , multiplied by a linear combination of two functions of u2 , and therefore it contains four arbitrary constants. At this stage, we must draw a sharp distinction between two possible cases, which will be clariﬁed by means of examples in the following sections. Let us proceed as follows. A) In the ﬁrst case, purely physical considerations enable us either to break a general integral into two parts (e.g., separating even and odd symmetry solutions), or to eliminate, as unacceptable, some of the functions of u 1 or u2 which show up in the general integrals (e.g., disregard functions which diverge to inﬁnity within their domain of deﬁnition). On some occasions, as we will see later, a procedure of this kind is capable of cutting the number of arbitrary constants in each function Fi from four to one, hence the total number of arbitrary constants in the problem from sixteen to four. In those cases, we may write: Hiz (u1 , u2 )

© 2006 by Taylor & Francis Group, LLC

= Ai Qi (u1 ) Mi (u2 )

(i = co, cl) ,

WAVES GUIDED BY A SURFACE OF DISCONTINUITY Eiz (u1 , u2 ) = Bi Ri (u1 ) Ni (u2 )

(i = co, cl) ,

347 (10.5)

where Ai , Bi (i = co, cl) are the remaining four arbitrary constants. Using either Eqs. (7.4) or Eqs. (7.11) and (7.12), Eqs. (10.4) can be rewritten in the following form, where it is understood that all the functions of u1 are evaluated at u1 = a: 1 dQco γ 1 dNco Aco Mco + Bco Rco h1 du1 jωco h2 du2 1 dQcl γ 1 dNcl = Acl Mcl + Bcl Rcl , h1 du1 jωcl h2 du2 Bco

2 κ2co − γ 2 κco − γ 2 Rco Nco = Bcl Rcl Ncl , jωco jωcl

1 dRco dMco γ 1 Aco − Qco Nco + Bco jωμ h2 du2 h1 du1 1 dRcl dMcl γ 1 = Acl − Qcl Ncl , + Bcl jωμ h2 du2 h1 du1

Aco (κ2co − γ 2 ) Qco Mco = Acl (κ2cl − γ 2 ) Qcl Mcl .

(10.6)

For this system of four homogeneous equations in the four variables Ai , Bi to have non-trivial solutions, the determinant of the coeﬃcients (in square brackets) in Eqs. (10.6) must be equal to zero. This yields a characteristic equation, where the unknown is the propagation constant γ. In general, it is a transcendental equation in the complex ﬁeld, since in the initial ansatz γ was supposed to be complex, and other parameters involved in the arguments of the functions M, N, Q, R can be complex too. However, it is strictly mandatory to check whether the solutions of the characteristic equation are consistent with the physical arguments that were used in order to arrive at Eqs. (10.6), eliminating some of the mathematically correct solutions of the ODE’s. The examples will show that this requirement is satisﬁed, for any ﬁnite value of the angular frequency ω, only by a ﬁnite number of solutions of the characteristic equation, all of which correspond to imaginary values of γ, and to e.m. ﬁelds whose power travels along the z direction. We may express these ﬁndings by saying that the structure has a ﬁnite number of guided modes. It can also be shown that, quite generally, the values of β = −jγ, which correspond to solutions of this kind, fall within the range deﬁned by the intrinsic phase constants of the cladding, βcl , and the core, βco , namely: βcl < β < βco

.

(10.7)

Clearly, from what we saw in the previous chapters, we may infer that this ﬁnite number of solutions cannot be a complete set of modes. To build a

© 2006 by Taylor & Francis Group, LLC

348

DIELECTRIC WAVEGUIDES

complete set, we may account also for solutions of a diﬀerent physical type, to be sketched as the next point. B) In the second case, on the contrary, one is not able to collect additional physical information, allowing us to write Eqs. (10.5) and so reducing the number of arbitrary constants. To quote just an example, this occurs for a cylindrical rod when one wants to describe ﬁelds traveling radially in the cladding. In general, they must be expressed as superpositions of Hankel functions of the ﬁrst and the second kind (see Appendix D). Under these circumstances, Eqs. (10.4) and the separation of variables yield, instead of Eqs. (10.6), a set of equations where the number of unknowns is larger than the number of equations. According to the Rouch´e-Capelli theorem of algebra, such a set has nontrivial solutions for any value of the propagation constant γ. Therefore, in addition to the ﬁnite number of guided modes, any dielectric waveguides has a continuum of modes. To grasp their physical nature, one has to check what values of γ are compatible with Maxwell’s equations in the core and in the cladding. It can be shown that, regardless of the waveguide shape, in a lossless structure the net result is as follows: √ 1. γ can take imaginary values, provided that |γ| < βcl = ω μcl . This continuum of solutions is referred to as the set of the radiation modes. Each of them consists of uniform plane waves, traveling in oblique directions with respect to the waveguide axis; 2. γ can take real values. This is also a continuum of solutions, referred to as the set of the evanescent modes. Each of them consists of evanescent plane waves whose attenuation vector is parallel to the waveguide axis, and consequently whose phase vector is orthogonal to the z axis. Such modes may reach an appreciable amplitude only in the immediate neighborhood of the ends of a waveguide of ﬁnite length. To summarize, at any ﬁnite value of the frequency a dielectric waveguide supports: • a ﬁnite number of guided modes; • a continuum of radiation modes; • a continuum of evanescent modes. The number of guided modes depends on the frequency, similarly to the number of propagating modes in a metallic waveguide. However, there is a major diﬀerence between metallic and dielectric waveguides from the viewpoint of cut-oﬀ. A mode below cut-oﬀ in a metallic waveguide cannot propagate, but does maintain its individual features. When a guided mode goes to its cut-oﬀ frequency in a dielectric waveguide, it loses its discrete nature, and merges into the continuum of radiation modes. Let us point out that in general the characteristic equation of the system (10.6) has other solutions, in addition to the ﬁnite number of guided modes, if γ is permitted to scan the whole complex plane, without restrictions. More precisely, these additional solutions form a countable inﬁnity, and

© 2006 by Taylor & Francis Group, LLC

GUIDED MODES OF A SLAB WAVEGUIDE

349

all of them are characterized by complex γ’s, with nonvanishing real and imaginary parts. At ﬁrst sight, each of these solutions might look like a generalized version of a guided mode, the attenuation accounting for some power radiated laterally. For this reason, ﬁelds of this kind, satisfying Eq. (10.1) and the following equations, have been called leaky modes of the dielectric waveguide. However, it is easy to show that they cannot be true rigorous modes of a lossless structure. Indeed, the characteristic equation can be derived through a crucial step, the selection based on physical requirements among the mathematically correct solutions of Eq. (10.1). Leaky modes violate these physical requirements, and because of this contradiction they cannot be looked at as physically meaningful solutions of the problem. Still, they can be quite useful in practice (Snyder and Love, 1983), to approximate the radiation ﬁeld (i.e., the total ﬁeld minus the guided modes) of a dielectric waveguide as a summation (in general, over a small number of terms) rather than as a line integral in the complex plane. In particular, in a structure aﬀected by loss, where the cut-oﬀ mechanism (i.e., a guided mode becoming radiating) is not simple to describe, leaky modes can be helpful. We will brieﬂy come back to this subject at the end of Section 10.4. Another important diﬀerence between dielectric and metallic waveguides is that, in general, the set of Eqs. (10.6) does not break into two sets of two equations, one including only Hiz (TE modes), and one including only Eiz (TM modes). Hence, in general dielectric waveguides have hybrid modes. TE and TM modes may exist, but are not enough to form a complete set. From this viewpoint, dielectric waveguides may look similar to waveguides surrounded by walls with a non-zero impedance (see Chapter 7). The main diﬀerence is that in a dielectric waveguide the continuity conditions may be expressed as a matching between mode-dependent wave impedances in the direction normal to C, in the core and in the cladding, but not in terms of a ﬁeld-independent wall-impedance. 10.3

Guided modes of a slab waveguide

We choose the ﬁrst example of application, for the broad-scope approach of the previous section, such that we may use rectangular coordinates, which simplify separation of variables and the following steps. However, we cannot apply the contents of the previous section to a rectangular waveguide, since its contour is not one coordinate surface, as it consists of four segments belonging to diﬀerent coordinates surfaces. In contrast to rectangular metallic waveguides, rectangular dielectric waveguides cannot be solved analytically in exact terms (see Snyder and Love, 1983). Therefore, our ﬁrst example will deal with a slab waveguide, which extends indeﬁnitely in one direction (say y) orthogonal to that of propagation (say z) (see Figure 10.2). Let 2a be the slab width along the x direction. We will consider only ﬁelds which are independent of the y coordinate. This assumption will allow us to clarify some basic ideas, but implies a loss in gen-

© 2006 by Taylor & Francis Group, LLC

350

DIELECTRIC WAVEGUIDES

Figure 10.2 Symmetrical step-index slab waveguide.

erality, whose consequences might appear to invalidate some of the statements of the previous section (in particular, the point referring to hybrid modes). The overall picture will be fully clariﬁed only after Section 10.5. Let us get into the details of the general approach outlined in Section 10.2, writing: Qi (u1 ) (i = co, cl) , (10.8) = Xi (x) Ri (u1 ) and following the same path as in Section 7.7, starting from the Helmholtz equation in the core and in the cladding. One gets that the ODE for Xi reads: Xi = −ki2 Xi

,

(10.9)

which is, once again, the harmonic equation, where: ki2 = −Γ2i = γ 2 − κ2i = γ 2 + ω 2 μi

(i = co, cl)

The general integral of Eq. (10.9) can be written as: sin kco x (even modes) Xco = C (odd modes) cos kco x in the core, and∗ :

⎧ x ⎨ e−Γcl |x| |x| Xcl = D ⎩ −Γ |x| e cl

.

(10.10)

(10.11)

(even modes) (10.12) (odd modes)

in the cladding, with C and D arbitrary constants. Solutions of the form exp(Γcl |x|) in the cladding have been disregarded as they diverge when x goes to inﬁnity, therefore are not physically meaningful. Assuming the ﬁeld is independent of the y-coordinate is equivalent to setting ∗ Calling even and odd the modes of a symmetrical slab waveguide is not a uniquely deﬁned process. In fact, when the longitudinal ﬁeld components in Eqs. (10.11) and (10.12) are even functions of x, then the transverse ﬁeld components, which descend from the longitudinal ones through a derivative with respect to x, are odd, and vice-versa. This fact is strictly correlated to what we saw about symmetries in Chapter 3. Denominations adopted here are not consistent with those given in Chapter 3, but are the most widely adopted in optical waveguide theory, and reﬂect the fact that transverse components are more important (and larger) than the longitudinal ones.

© 2006 by Taylor & Francis Group, LLC

GUIDED MODES OF A SLAB WAVEGUIDE

351

Mi ≡ Ni ≡ 0 in Eqs. (10.6). Then, this system breaks down into two sets, of two equations each in two unknowns. One of these sets identiﬁes TE modes, the other one, TM modes. To be more speciﬁc, continuity equations corresponding to even TE modes (where the word even is used according to the previous footnote) read: C kco cos kco a = −D Γcl e−Γcl a

,

2 C kco sin kco a = −D Γ2cl e−Γcl a

.

(10.13)

Setting equal to zero the determinant of the coeﬃcients of Eqs. (10.13) yields the characteristic equation (also called the dispersion relation): kco tan kco a = Γcl

,

(10.14)

which must be combined with the following equation, which stems from Eq. (10.10): 2 kco + Γ2cl = ω 2 μ(co − cl )

.

(10.15)

Any solution of Eq. (10.14) corresponds to a nontrivial solution of Eqs. (10.9), which is the longitudinal component, Hz , of an even TE mode. The corresponding transverse ﬁeld components can be obtained from it by means of Eqs. (7.4). In detail, if we let Hz = Xi (i = co, cl), we obtain the following ﬁeld in the core: jωμ −γ Ey , Hy = 0 , Ex = 0 , Ey = C cos kco x , Hx = kco jωμ (10.16) and in the cladding: Ex = 0 ,

Ey =

jωμ D e−Γcl |x| Γcl

,

Hx =

−γ Ey jωμ

,

Hy = 0 . (10.17)

As we anticipated in the previous footnote, the transverse ﬁeld components are even, for the reasons explained in Section 3.9. The characteristic equation and the ﬁeld components of the odd TE modes, and those of the even and odd TM modes, can be obtained in similar ways, and are summarized in Tables 10.1 and 10.2. All the characteristic equations shown in those tables are to be combined with Eq. (10.15). One of the main points in the theory of the previous section was that at any ﬁnite frequency there is a ﬁnite number of guided modes. So, we should prove that each of the four characteristic equations has, for a ﬁxed ﬁnite value of ω, a ﬁnite number of solutions. Let us prove it for Eq. (10.14), leaving to the reader as an exercise the very similar proofs for the other three equations. By means of Eq. (10.15), and imposing that Γcl is real in order not to lose the physical features of the guided modes, one derives: 2 − β2 |kco | ≤ ω μ(co − cl ) = βco , (10.18) cl where the equal sign holds only for ω = 0. Furthermore, by means of Eq. (10.15),

© 2006 by Taylor & Francis Group, LLC

352

DIELECTRIC WAVEGUIDES Table 10.1 TE modes of the symmetrical slab waveguide.

EVEN TE MODES Characteristic equation: kco tan kco a = Γcl Field components: Core Cladding Hz = C sin kco a

Hz = C sin kco x Ey =

jωμ C cos kco x kco

Hx = −

Ey =

β Ey ωμ

x −Γcl (|x|−a) e |x|

jωμ C sin kco a e−Γcl (|x|−a) Γcl

Hx = − Ex = H y = 0

β Ey ωμ

ODD TE MODES Characteristic equation: kco cot kco a = −Γcl Field components: Core Cladding Hz = C cos kco a e−Γcl (|x|−a)

Hz = C cos kco x Ey = −

jωμ C sin kco x kco

Hx = −

β Ey ωμ

Ey = −

jωμ x C cos kco a e−Γcl (|x|−a) Γcl |x|

Hx = − Ex = H y = 0

Eq. (10.14) can be rewritten as: tan kco a =

β Ey ωμ

2 − β 2 − k2 βco co cl kco

.

(10.19)

The left-hand side of Eq. (10.19) is a periodic function of kco , the right-hand side is a monotonically decreasing function of kco (see Figure 10.3). Within the range deﬁned by Eq. (10.18), a periodic curve and a monotonical one can intersect each other only at a ﬁnite number of points, hence the number of guided modes is ﬁnite. Eq. (10.10) and the inequality in Eq. (10.18) entail that the phase constant of any guided mode, β = −jγ, satisﬁes, for ω = 0, the following constraints: βcl < β < βco

,

(10.20)

which were also anticipated in the general discussion of Section 10.2. As ω tends to zero, Eq. (10.14) always has one solution, the corresponding phase constant also going to zero, together with kco and Γcl . This result indicates that one guided TE mode may propagate in the slab at an arbitrarily low frequency, which makes a fundamental diﬀerence with respect to metallic waveguides. The same is true for the ﬁrst even TM mode as well, but not for any odd mode, neither TE nor TM. Physically, these lowest-order guided

© 2006 by Taylor & Francis Group, LLC

GUIDED MODES OF A SLAB WAVEGUIDE

353

Table 10.2 TM modes of the symmetrical slab waveguide. EVEN TM MODES Characteristic equation: Field components:

1 1 kco tan kco a = Γcl co cl

Core

Cladding

Ez = C sin kco x

Ez = C sin kco a

Hy = − Ex =

jωco C cos kco x kco

β Hy ωco

Hy = − Ex = Ey = H x = 0

x −Γcl (|x|−a) e |x|

jωcl C sin kco a e−Γcl (|x|−a) Γcl

β Hy ωcl

ODD TM MODES Characteristic equation: Field components:

1 1 kco cot kco a = − Γcl co cl

Core

Cladding

Ez = C cos kco x

Ez = C cos kco a e−Γcl (|x|−a)

Hy = Ex =

jωco C sin kco x kco β Hy ωco

Hy = Ex = Ey = H x = 0

jωcl x C cos kco a e−Γcl (|x|−a) Γcl |x| β Hy ωcl

k co a

Figure 10.3 Graphical solution of the characteristic equation of the even TE modes.

© 2006 by Taylor & Francis Group, LLC

354

DIELECTRIC WAVEGUIDES x TE0

TE1

TE2

a Ey

0

-a

a

b

c

Figure 10.4 Examples of low-order TE modes in a slab waveguide. a) Lowest-order even mode. b) Lowest-order odd mode. c) First higher-order even mode.

modes tend to become uniform plane waves in the cladding when ω tends to zero, thus becoming less and less sensitive to the presence of a core. This can never be the case for an odd mode, where the ﬁelds in the two half-spaces which form the cladding, have opposite phase. To reconﬁrm that this is the correct physical explanation of the diﬀerence in cut-oﬀ behavior between the lowest-order even and odd modes, the reader may prove, as an exercise, that in an asymmetrical slab waveguides where the upper and lower half-spaces have diﬀerent permittivities, all the guided modes have a non-zero cut-oﬀ frequency. Amplitude distributions of the various ﬁeld components, and ﬁeld lines, for the lowest-order even and odd guided modes of a slab waveguide, are sketched in Figure 10.4. 10.4

Radiation modes of a slab waveguide

For the radiation modes of a slab waveguide, there are no physical grounds to eliminate, in the cladding, any of the oscillatory solutions (or of the two exponential solutions with imaginary argument) of Eq. (10.9). Therefore, we let: (10.21) Xcl = q ejkcl x + p e−jkcl x , where p and q are two arbitrary constants. Eq. (10.15) is replaced by: 2 2 − kcl = ω 2 μ(co − cl ) . kco

(10.22)

Continuity of tangential components at x = ±a yields, both for TE and for TM modes, two linear homogeneous equations in three unknowns, p, q (from Eq. (10.21)) and C (from Eq. (10.11)). The Rouch´e-Capelli theorem says that such a system has an inﬁnity of nontrivial solutions for any values of the coeﬃcients. Hence, we cannot ﬁnd additional conditions on kco and kcl . The radiation modes are a continuum, whose transverse ﬁeld parameters are

© 2006 by Taylor & Francis Group, LLC

RADIATION MODES OF A SLAB WAVEGUIDE

355

related only by Eq. (10.22). They must satisfy only one restriction, coming 2 from Eq. (10.10) as a consequence of kcl > 0, which is necessary in order to have, as required, an oscillatory behavior in the cladding. The restriction reads: β ≤ βcl , (10.23) where β is now the phase constant of the generic radiation mode. We leave it to the reader as an exercise to study the evanescent modes, namely the solutions with real γ, which must satisfy the restriction ki ≥ βi (i = co, cl), coming again from Eq. (10.10). Let us strengthen the links between the mathematical description of radiation modes and physical intuition, and try to clarify why Eq. (10.21) contains two terms, one corresponding to a wave traveling away from the slab, the other one to a wave traveling towards the slab. It would be erroneous to believe that radiation modes must be able to describe only ﬁelds excited by a source located at |x| < a, inferring from this starting point that waves traveling towards the slab are not physically meaningful. There are at least three reasons why this viewpoint would be wrong; a) To be complete, a set of modes must be able to describe a ﬁeld radiated by a source located on one side of the slab, at an arbitrary distance from it. This also requires waves which travel towards the slab; b) A cladding extending indeﬁnitely in the x direction is a physically acceptable model only if it represents the limit of a cladding with a ﬁnite thickness, d, when d goes to inﬁnity. Centripetal waves are needed in order to describe that part of the radiated ﬁeld which is reﬂected back by the external edge of the cladding. This also tells us that it would be erroneous to disregard centripetal waves; c) The whole centripetal ﬁeld, expressed as an integral over the continuum of radiation modes, can carry zero power, without implying zero amplitude for the individual centripetal waves. This is a well known property of the Fourier integral, and the reader is probably familiar with it from timedomain signal analysis. All these comments reconﬁrm that, as we remarked in Section 10.2, dielectric and metallic waveguides are totally diﬀerent concerning that part of the ﬁeld which does not propagate under the form of guided modes. In dielectric guides, each transmission mode loses its individual features as soon as it is cut-oﬀ. The “invention” of leaky modes can be thought of as an attempt to eliminate this diﬀerence, using another description for that part of the ﬁeld which complements the guided modes. A non-real solution of the characteristic equation, in the complex ﬁeld, could be looked at as an “extension” of a real solution, when a guided mode has gone through a cut-oﬀ condition, i.e., when the angular frequency is below a critical value ωcn , similarly to what occurs in Chapter 7. Every complex solution of the characteristic equation would entail Re γ = 0, and thus an attenuation which is an increasing function of ωcn − ω. These solutions form

© 2006 by Taylor & Francis Group, LLC

356

DIELECTRIC WAVEGUIDES

Figure 10.5 Step-index circular dielectric rod.

a countable inﬁnity and, as mentioned in Section 10.2, are often referred to as leaky modes. However, let us stress once again that the term “modes” might be misleading. As we said in general in Section 10.2, and as we saw in detail in Section 10.3, the characteristic equation was obtained on the grounds of a well deﬁned physical assumption — a guided mode — and there is no reason why it should hold when the physics changes. To appreciate the usefulness of leaky modes as an approximation, and to appreciate the mathematical subtleties of the procedure, the reader is referred to specialized books (Snyder and Love, 1983).

10.5

The cylindrical rod: Exact modes

This section is devoted to the guided modes of a cylindrical dielectric rod, of radius a (Figure 10.5). This example is very important under a practical viewpoint, because of the impact of optical ﬁbers in modern telecommunication technology and in other related ﬁelds, like sensors. We will show that these guided modes can be identiﬁed rigorously, but using their exact expressions makes it complicated to express some important quantities, like power density ﬂowing along the waveguide. For this reason, we will develop an approximate theory in the following section. Radiation modes of a cylindrical waveguide do not exhibit signiﬁcant diﬀerences with respect to those of a slab waveguide, except for mathematical details (Hankel functions replacing exponentials of imaginary argument) which can be left to the reader as an exercise. Let us follow the general procedure of Section 10.2, taking into account the contents of Section 7.8 about the Helmholtz equation in cylindrical coordinates. A physical selection criterion is to disregard those solutions of Bessel’s equations which diverge, either for r < a or for r > a. Accordingly, we can set: i = co, cl ; Mi (10.24) = ejνϕ ν = 0, ±1, ±2, . . . ; Ni

© 2006 by Taylor & Francis Group, LLC

THE CYLINDRICAL ROD: EXACT MODES ur Qco = Jν (kco r) = Jν Rco a

357 ,

(10.25)

where Jν is the Bessel function of the ﬁrst kind and of integer order ν, and: wr Qcl , (10.26) = Kν (Γcl r) = Kν Rcl a where Kν is the modiﬁed Bessel function of the second kind and of integer order ν. The following adimensional quantities have been deﬁned in these equations: u = kco a = a ω 2 μco − β 2 = a ω 2 μco + γ 2 = a γ 2 − κ2co ,

w = Γcl a = a

−ω 2 μcl + β 2 = a

−(γ 2 + ω 2 μcl ) = a

κ2cl − γ 2 , (10.27)

where β = −jγ is the phase constant of the generic guided mode. As for the harmonic equation which governs the dependence on the azimuthal coordinate, we chose to express its solutions in terms of exponential functions, Eq. (10.24), rather than in terms of cos(νϕ) and sin(νϕ), because the continuity conditions, Eqs. (10.6), involve the functions and also their derivatives, so Eqs. (10.24) end up making the calculations easier. The metric coeﬃcients are h1 = 1, h2 = r. Following a standard notation, let us use Jν and Kν for the derivatives of Bessel functions with respect to their arguments. Then, Eqs. (10.6) read as follows: Aco u Jν (u) − Bco

βν βν Jν (u) = Acl w Kν (w) − Bcl Kν (w) , jωco jωcl

Bco Aco

u2 Jν (u) jωco

= −Bcl

w2 Kν (w) jωcl

,

βν Jν (u) βν Kν (w) + Bco u Jν (u) = Acl + Bcl w Kν (w) , jωμ jωμ Aco u2 Jν (u) = −Acl w2 Kν (w)

.

(10.28)

For ν = 0, we can separate Eqs. (10.28) into two sets of two equations, one in the two unknowns (Aco , Acl ), the other in the two unknowns (Bco , Bcl ). In fact, it is not surprising that these modes, independent of the azimuthal coordinate ϕ, behave similarly to those independent of the y coordinate in the slab waveguide of Section 10.3. The set which consists of the ﬁrst and the fourth of Eqs. (10.28) corresponds to TE modes, and yields the following characteristic equation: K (w) J0 (u) =− 0 u J0 (u) w K0 (w)

(TE0,p ) .

(10.29)

The set which consists of the second and the third of Eqs. (10.28) corresponds

© 2006 by Taylor & Francis Group, LLC

358

DIELECTRIC WAVEGUIDES

to TM modes, and yields the following characteristic equation: co

J0 (u) K0 (w) = −cl u J0 (u) w K0 (w)

(TM0,p ) .

(10.30)

We see from these characteristic equations that the TE0,p and the TM0,p modes are not degenerate (co = cl ), but tend to become degenerate as co → cl . This comment will be expanded and used in the following section. Modes whose azimuthal order is not equal to zero are hybrid modes, as the set of Eqs. (10.28), for ν = 0, does not break down into two separate sets. Their characteristic equation can be written in a compact form if we set: jν (u) =

Jν (u) u Jν (u)

,

bν (w) =

Kν (w) w Kν (w)

,

(10.31)

and introduce an adimensional quantity, v, proportional to the angular frequency ω, deﬁned as follows: v 2 = u2 + w2 = a2 ω 2 μ(co − cl )

.

(10.32)

Dividing the ﬁrst and the third of Eqs. (10.28) by Jν Kν , eliminating Acl and Bcl by means of the second and the forth equation and, ﬁnally, setting the determinant of the remaining system of two equations in Aco , Bco equal to zero, the characteristic equation then reads: 2 1 1 2 2 2 2 (jν + bν ) (κco jν + κcl bν ) = −ν β + 2 , (10.33) u2 w where u, w and β are related through Eqs. (10.27). Another, perhaps more elegant way to write the characteristic equation is: 1 co 1 cl 2 , (10.34) + + (jν + bν ) (co jν + cl bν ) = ν u2 w2 u2 w2 which must be associated with Eq. (10.32). To reach this result, we divide Eq. (10.33) by v 2 , write 1/u2 +1/w2 = v 2 /(u2 w2 ), replace v 2 with Eq. (10.32), and ﬁnally, make use of the following expression, which can be obtained taking the squares of the two Eqs. (10.27), multiplying the ﬁrst by cl and the second by co , and summing them: β 2 a2 (co − cl ) = co w2 + cl u2

.

(10.35)

Note that Eq. (10.34) breaks into Eqs. (10.29) and (10.30) for ν = 0. For ν = 0, Eq. (10.34) is invariant with respect to the change ν → −ν, since (see , bν = b−ν . Therefore, each solution of Eq. (10.34) for Appendix D) jν = j−ν ν = 0 corresponds to a pair of degenerate modes. This agrees with physical intuition: waves propagating helically in an isotropic medium cannot distinguish between traveling clockwise or counterclockwise. This enables us to assume ν > 0 from now on, to simplify our calculations. Our next aim is to show that the solutions of the characteristic equation may be separated into two sets, which are physically diﬀerent from each other.

© 2006 by Taylor & Francis Group, LLC

THE CYLINDRICAL ROD: EXACT MODES

359

A suitable way to reach this result is simply to rewrite Eq. (10.34) as a secondorder algebraic equation in jν , and to solve it, while introducing, for future beneﬁts, two new variables, namely: co + cl , Δ = co − cl , = (10.36) 2 and using a concise notation, fν2 (u, w), for the right-hand side of Eq. (10.34). Make use of the following expressions for derivatives of Bessel functions (see Appendix D): ν Jν (u) − Jν+1 (u) , (10.37) Jν (u) = u when taking the positive square root of the discriminant, and: ν Jν (u) + Jν−1 (u) , u when taking the negative one. The ﬁnal results are, respectively: 1/2 Jν+1 (u) u Δ2 2 ν 2 = b + co fν (u, w) ub + − Jν (u) co ν u co 4 ν Jν (u) = −

for the plus sign in front of the discriminant, and: 1/2 Jν−1 (u) u Δ2 2 ν =− bν + co fν2 (u, w) ubν + − Jν (u) co u co 4

(10.38)

(10.39)

,

(10.40)

for the negative one. Let us emphasize once again that Eqs. (10.39) and (10.40) assume ν > 0, and each solution of either one corresponds to a pair of degenerate modes having identical behaviors vs. the radial coordinate, given by Eqs. (10.25) and (10.26), while they diﬀer in the sign in front of the azimuthal coordinate, exp(±jνϕ). The solutions of Eq. (10.39) are referred to as the EHν,p modes, those of Eq. (10.40) as the HEν,p modes. Radial and azimuthal components of the e.m. ﬁeld may be obtained from the longitudinal components, i.e., from Eqs. (10.24), (10.25) and (10.26), expressing Eqs. (7.4) in cylindrical coordinates. They are summarized in Table 10.3. The reader can verify, as an exercise, that radial components satisfy the following continuity conditions (which, as shown in Chapter 1, are not independent of those which hold for tangential components): co Erco (r = a) = cl Ercl (r = a) Hrco (r = a) = Hrcl (r = a) .

, (10.41)

Table 10.3 shows that the longitudinal components of these exact modal ﬁelds are fairly simple but, on the other hand, transverse components are complicated, mainly because Bessel functions of integer order ν can be written as superpositions of two Bessel functions of diﬀerent orders (to choose between ν, ν + 1, and ν − 1). Transverse components are fundamental in practice, e.g., to calculate the Poynting vector. Furthermore, it can be shown (see the rest of the chapter) that, in agreement with physical intuition, when the diﬀerence

© 2006 by Taylor & Francis Group, LLC

360

DIELECTRIC WAVEGUIDES Table 10.3 Exact modes of the circular step-index rod. Core (ρ ≤ 1)

Cladding (ρ ≥ 1)

Hz

A Jν (uρ) ejνϕ

A

Jν (u) Kν (wρ) ejνϕ Kν (w)

Ez

B Jν (uρ) ejνϕ

B

Jν (u) Kν (wρ) ejνϕ Kν (w)

Er

−

j jνϕ e · u2

· B Eϕ

−

βau Jν (uρ)

jωμ0 νa +A Jν (uρ) ρ

j jνϕ e · u2

Hr Hϕ

jβaν · B Jν (uρ) − A ωμ0 aw Jν (wρ) ρ ωco − Eϕ β ωco Er β

μ

HE modes:

B=j

EH modes:

B = −j

0

co

μ

j Jν (u) jνϕ e · w2 Kν (w)

· B

βaw Kν (wρ)

jωμ0 νa +A Kν (wρ) ρ

j Jν (u) jνϕ e · w2 Kν (w)

jβaν · B Kν (wρ) − A ωμ0 aw Kν (wρ) ρ ωcl − Eϕ β ωcl Er β

A 0

co

A

co − cl is not too large, the magnitude of the transverse components is much larger than that of longitudinal components. The practical need to simplify transverse ﬁeld components is a valid reason why the approximate theory of Section 10.7 is often preferred to the exact theory presented in this section. 10.6

Modal cut-oﬀ in the cylindrical rod

A suitable way to clarify the physical diﬀerence between EH and HE modes, is to analyze their cut-oﬀ conditions. As we found in general in Section 10.2, and saw reconﬁrmed by the example of Section 10.3, a guided mode is cut-oﬀ when its ﬁeld does not decay, in the radial direction, fast enough to prevent power from ﬂowing in the radial direction. This is translated into mathematical terms as w → 0, to be replaced in the previous equations. To proceed further, we make use of the following approximations (Abramowitz and Stegun, Eds., 1965) for small values of w: bν (w) −ν/w in the case of Eq. (10.39), and bν (w) −(ν/w) − 1/[2(ν − 1)] in the case of Eq. (10.40). Keeping also track that Eq. (10.32) implies that at cut-oﬀ (w = 0) it is u = v = vc (critical value of v parameter), after some algebra one ﬁnds that the cut-oﬀ condition arising

© 2006 by Taylor & Francis Group, LLC

MODAL CUT-OFF IN THE CYLINDRICAL ROD

361

Table 10.4 Cut-oﬀ normalized frequencies for some low-order modes. vc

0

2.405

3.832

5.136

5.520

Modes

HE11

TE01 , TM01

EH11 , HE12

EH21

TE02 , TM02

from Eq. (10.39) reads: Jν (vc ) = 0

(EHν,p modes ; ν = 1, 2, . . .) ,

(10.42)

while that arising from Eq. (10.40) reads: vc

2(ν − 1) ¯ Jν (vc ) = Jν−1 (vc ) cl

(HEν,p modes ; ν = 1, 2, . . .)

.

(10.43)

In particular, the latter equation, for ν = 1 and vc = 0, gives Jν (vc ) = 0, showing that at the cut-oﬀ (but only at cut-oﬀ) HE1,p modes and EH1,p−1 modes are degenerate. Eqs. (10.42) and (10.43) show that, for any ﬁnite value of the normalized frequency v deﬁned by Eq. (10.32), each family of guided modes is a ﬁnite set. This result could also be predicted similarly to what was carried out in Section 10.3 on the characteristic equation of the slab modes, i.e., observing that one side of these equations is monotonic, while the other one is a function which goes, sequentially, through zero and through inﬁnity. Eq. (10.43) also shows that the fundamental mode is the HE1,1 mode, which can propagate at an arbitrarily small frequency (vc = 0) like the fundamental mode of a slab. We will see in the following sections that fundamental modes of a slab and of a cylindrical rod are physically very similar, in spite of their labels appearing to be completely diﬀerent from each other. A very practical matter is to establish which are the ﬁrst higher-order modes. The question can be answered letting w = 0 in Eqs. (10.29) and (10.30), and this yields J0 (vc ) = 0. Therefore, the next modes are the TE0,1 and the TM0,1 . They are degenerate only at cut-oﬀ. The cut-oﬀ value for the normalized frequency is vc = 2.405. Cut-oﬀ values of the normalized frequency for other modes are shown in Table 10.4. For a better understanding of the role played by various parameters in the design of practical single-mode and multimode ﬁbers, let us rewrite the normalized frequency, deﬁned by Eq. (10.32), as follows, in terms of the refractive indices of core and cladding: √ √ 2πa ncl 2Δ , (10.44) v = ka ncl 2Δ = λ0 √ where k = ω μ0 0 = 2π/λ0 is the free-space wave number, and: 2Δ ≡

n2co − n2cl n2cl

(10.45)

is the relative index diﬀerence. In comparison with Chapter 7, we notice that, while in a conducting-wall waveguide the cut-oﬀ frequency of a given mode

© 2006 by Taylor & Francis Group, LLC

362

DIELECTRIC WAVEGUIDES Table 10.5 Typical values of refractive index for glasses in common use.

Pure silica (SiO2 ) GeO2 -doped silica F-doped silica Halide glasses Chalcogenide glasses

1.4585 1.4585 – 1.465 1.445 – 1.4585 1.41 – 1.72 1.7 – 2.5

depends on the guide size and on the refractive index of the dielectric medium which ﬁlls it, in a dielectric waveguide it depends on these parameters plus another one, namely the core-cladding index relative diﬀerence. For an optical ﬁber to be single-mode, v must be less than 2.405. Wavelengths of highest practical interest (measured in free space) are in the near infrared, more precisely in the range between 0.8 μm (the so-called “ﬁrst window”) and 1.55 μm (“third window”; these names are related to historical breakthroughs in the development of optical communications) where silica glass is very transparent (see Section 10.11). The refractive index of pure silica, at those wavelengths, is somewhat less than 1.5 (see Table 10.5). These √ facts indicate that the product a 2Δ must be smaller than a few units (depending on the exact wavelength) multiplied by 10−7 , if a is measured in meters. Actual choices for a and 2Δ are dictated by other, contradictory requirements (see Snyder and Love, 1983), typically bending loss and coupling loss. However, for all ﬁbers which are standard commercial products, a is a few micrometers, and 2Δ is less than 0.01. These are highly demanding requirements from a technological viewpoint. This point explains why highly multimode ﬁbers, where a is of the order of tens of micrometers and 2Δ reaches a few percent so that v 2.405, have attracted extremely serious technical attention for decades, in spite of some major limitation that will be outlined in Section 10.8. On the other hand, few-mode ﬁbers, where v > 2.405, but not v 2.405, are of interest only for some special applications, like sensors, which are beyond the scope of this book (Dakin and Culshaw, 1988). Incidentally, let us point out that, as 2Δ 1 in all cases of practical interest, we have nco + ncl 2ncl , so that Eq. (10.45) can be rewritten as Δ

nco − ncl ncl

( 1)

.

(10.46)

The practical importance of having Δ 1 will be thoroughly appreciated in the next three sections. Finally, let us try to explain why the cut-oﬀ conditions which were derived from the characteristic equations justify the strange terminology which is in current use for the dielectric waveguide modes. In fact, Eq. (10.42) appears to be the same as the equation which, in a circular waveguide with conducting walls, yields cut-oﬀ frequencies for those modes which are most frequently referred to as TM, but are called E modes by other authors. As we saw here, modes of a dielectric rod are hybrid; still, as a result of this purely formal

© 2006 by Taylor & Francis Group, LLC

WEAKLY GUIDING RODS: THE LP MODES

363

similarity, one set was named EH, and then, by analogy, the other one was named HE. 10.7

Weakly guiding rods: The LP modes

We saw in the previous section that, for most dielectric waveguides used in practice, and in particular for typical optical ﬁbres, if we return to the notation of Eq. (10.36), we ﬁnd: Δ ¯ . (10.47) A dielectric structure where Eq. (10.47) is satisﬁed is referred to as weakly guiding. Introducing this assumption into Eq. (10.34), we ﬁnd that both characteristic equations, for HE and for EH modes, result in the following approximate form: 1 1 ∼ + 2 jν + bν = ±ν . (10.48) u2 w This tells us that HE and EH modes tend to become degenerate as Δ → 0. Knowing from Section 10.5 that there is a strict two-fold degeneracy related to replacing ν with −ν, we see that for Δ → 0 we tend towards a four-fold degeneracy of the guided modes, in a dielectric rod. In mathematical terms, the idea of a “quasi-degeneracy” is not well founded. In any eigenvalue problem, degenerate solutions are critical, in the sense that some diﬃculties show up when two eigenvalues are equal, but disappear as soon as they are not equal, irrespectively of how close they are, provided they are distinct. If we adopt this viewpoint, then two modes which correspond to two diﬀerent eigenvalues can never be linearly combined to yield another mode, even if the two eigenvalues are very close to each other. However, in a weakly guiding dielectric waveguide of circular cross-section, the errors we make if we treat as degenerate EH and HE modes whose eigenvalues are not strictly identical can be shown (Snyder and Love, 1983) to be small, compared to errors that we make neglecting other perturbations, like slight ellipticity, absorption and scattering losses, bends, slight unintentional anisotropy, etc. As a result, the ﬁelds that we build up in this section are not exact solutions of Maxwell’s equations, and therefore should be called pseudomodes, rather than modes, but they match very well experimental results on actual ﬁbers aﬀected by imperfections, and in this sense can be referred to as modes. Let us proceed as if the two families of modes were degenerate. It is easy to show that, using the same notation as in the previous section, degeneracy does not occur between the HEν,m and the EHν,m mode, but between the HEν,m and the EHν−2,m mode. Indeed, using the recurrence formula: 2(ν − 1) Jν−1 (u) = Jν (u) + Jν−2 (u) . (10.49) u Eq. (10.43) shows that for ¯/cl → 1 the cut-oﬀ equation for the HEν,m mode becomes: (10.50) Jν−2 (vc ) = 0 ,

© 2006 by Taylor & Francis Group, LLC

364

DIELECTRIC WAVEGUIDES

which coincides with the cut-oﬀ condition for the EHν−2,m mode, Eq. (10.42), which is insensitive to the approximation cl ¯. This four-fold degeneracy can be exploited to derive drastically simpler transverse components. The price to pay is to have more complicated longitudinal components, but, as we said before, these are far less used in practical calculations. Furthermore, the reader can prove as an exercise (working, to simplify calculations, on the modes of a slab waveguide or on the TE and TM modes of a cylindrical rod) that the longitudinal components are much smaller than the transverse ones. In a weakly guiding structure, this ratio can be shown to be of order Δ1/2 , where Δ is still deﬁned by Eq. (10.45). The procedure which yields the simpliﬁed transverse ﬁelds can be summarized as follows, omitting calculations for the sake of brevity. To start with, let = 1, 2, . . .. (For = 0, there is nothing to add to what was said in Section 10.5). Then: • take the transverse components of the HE+1,m mode and those of the EH−1,m mode (see Table 10.3); • use Bessel and modiﬁed Bessel function recurrence relations to express all these components as linear combinations of Bessel and modiﬁed Bessel functions of two orders, to be chosen among , + 1, − 1; for example, let us take those of orders and + 1; • look for a linear combination of the HE+1,m mode and the EH−1,m mode ensuring that it does not contain Bessel functions of one of these remaining orders, say +1. Note that this step is not trivial at all, as the HE+1,m mode and the EH−1,m mode have diﬀerent dependences on the azimuthal coordinate, expressed by exp[j( ± 1) ϕ], respectively. The reason why one can ﬁnd a positive answer to this requirement is precisely the four-fold mode degeneracy, i.e., the possibility of combining linearly not only the above mentioned functions of ϕ, but also exp[−j( ± 1) ϕ]. The number of independent coeﬃcients one can play with then becomes suﬃcient to eliminate J+1 and K+1 from all the transverse ﬁeld components. The reader can prove as an exercise that, referring to their dependence on the radial coordinate, the transverse ﬁeld components, for the generic mode of this new set, are expressed, respectively, in the core by the upper line and in cladding by the lower line of the following formula: J (ur/a)/J (u) Eϕ = E0 F (ϕ) , Hr = −Eϕ /ηi , K (wr/a)/K (w) J (ur/a)/J (u) Er = E0 G (ϕ) , Hϕ = Er /ηi , (10.51) K (wr/a)/K (w) where E0 is an arbitrary constant, and the pair of functions of the azimuthal coordinate F (ϕ), G (ϕ) will be discussed shortly. In Eqs. (10.51) it is ηi = μ/i = η0 /ni , ni being the refractive index of the i-th medium (i = co, cl). This result indicates that, when the assumption Δ 1 holds, for any guided

© 2006 by Taylor & Francis Group, LLC

WEAKLY GUIDING RODS: THE LP MODES

365

Figure 10.6 Rectangular and cylindrical reference frames in a circular rod.

mode expressed by Eqs. (10.51) the wave impedance in the direction of propagation is equal to the intrinsic impedance of the medium. Consequently, comparing with Chapters 5 and 7, we see that we are dealing with a quasi-TEM propagation. Let us focus now on the azimuthal dependence of the transverse ﬁelds. Calculations show that the pair of functions in Eqs. (10.51) can be any among the following four choices (incidentally, note that the four-fold degeneracy has been preserved): F = sin( + 1) ϕ F = sin( − 1) ϕ ; ; G = − cos( + 1) ϕ G = cos( − 1) ϕ (10.52) F = cos( + 1) ϕ F = cos( − 1) ϕ ; . G = sin( + 1) ϕ G = − sin( − 1) ϕ At ﬁrst sight, these results may look disappointing, because the ϕ-dependence is not very simple and is actually not even the same for the Eϕ and Er components. Fortunately, these problems are very easy to circumvent. Indeed, the ﬁeld components in the rectangular coordinate frame shown in Figure 10.6 are related to Eqs. (10.51) as: Ex

= Er cos ϕ − Eϕ sin ϕ ,

Ey

= Er sin ϕ + Eϕ cos ϕ .

(10.53)

Starting from the four choices expressed by Eqs. (10.52), Eqs. (10.53) yield the following four possibilities: Ex ∝ cos ϕ Ex ∝ − cos ϕ ; ; (10.54) Ey ∝ sin ϕ Ey ∝ sin ϕ Ex ∝ sin ϕ Ex ∝ − sin ϕ ; . (10.55) Ey ∝ cos ϕ Ey ∝ cos ϕ Finally, summing and subtracting Eqs. (10.54), and Eqs. (10.55), we get the following four choices: Ex = 0 , Ex = 0 ,

© 2006 by Taylor & Francis Group, LLC

Ey = E0 R(r) sin ϕ , Ey = E0 R(r) cos ϕ ,

(10.56)

366

DIELECTRIC WAVEGUIDES x

x

|E| J0

K0

y

y

r/a

1

a

b

c

Figure 10.7 Schematic illustration of the fundamental (LP0,1 ) mode. Part a: xpolarization. Part b: y-polarization (continuous lines = electric ﬁeld; dashed lines = magnetic ﬁeld). Part c: transverse ﬁeld amplitude (arbitrary units) vs. normalized radial coordinate.

Ex = E0 R(r) cos ϕ Ex = E0 R(r) sin ϕ

, ,

Ey = 0 , Ey = 0 ,

(10.57)

where E0 is an arbitrary constant and R(r) is the quantity in brackets in Eqs. (10.51). Each of these electric ﬁelds is accompanied by a magnetic ﬁeld expressed by: Ex Ey , Hx = − . (10.58) Hy = ηi ηi Accordingly, we see that the transverse ﬁeld of each of these modes consists of one cartesian component for the electric ﬁeld, and one cartesian component for the magnetic ﬁeld orthogonal to the electric one and with the same radial distribution. This explains why these modes are said to be linearly polarized. The corresponding concise notation is LP,p , where = 0, 1, 2, . . . and p = 1, 2, . . . Note that: • the LP,p mode with > 1 is the combination of the HE+1,p mode and the EH−1,p mode; • the LP1,p mode is the combination of the HE2,p mode, the TE0,p mode and the TM0,p mode; • the LP0,p mode is the weakly-guiding approximation of the HE1,p mode; • any LP,p mode with > 0 exhibits a four-fold degeneracy, while an LP0,p mode is two-fold degenerate. Using the new terminology, the fundamental mode is the LP0,1 mode. Its ﬁeld lines and amplitude distribution are sketched in Figure 10.7. The ﬁgure illustrates why the fundamental mode is doubly degenerate: it is a matter of polarization. However, degeneracy is broken when circular symmetry is broken, either because the core cross-section is not circular (shape birefringence), or because the medium is anisotropic as a result of internal force distribution

© 2006 by Taylor & Francis Group, LLC

WEAKLY GUIDING RODS: THE LP MODES

367

Figure 10.8 Constant-amplitude contour plots for the ﬁrst higher-order (LP 1,1 ) mode.

(stress birefringence). Usually, in silica glass, the latter eﬀect dominates over the ﬁrst one. In some cases (high-birefringence ﬁbers), stress anisotropy is intentional, and meant to preserve the state of polarization of the guided ﬁeld. In typical telecommunication ﬁbers, birefringence is unwanted, and varies at random along the ﬁber. It results in distortion of the signal which travels along a single-mode ﬁber, and is usually referred to as polarization mode dispersion (Matera and Someda, 1992; Galtarossa and Menyuk, 2005), and probably easier to understand after reading the next section. The ﬁrst higher-order mode, whose normalized cut-oﬀ frequency remains vc = 2.405, is now called LP1,1 mode. Its ﬁeld distribution is sketched in Figure 10.8, which illustrates the sources of the four-fold degeneracy: one twofold degeneracy is related to polarization, while the other two-fold degeneracy is related to a rotation of the amplitude distribution around the z axis by an angle π/(2). The ﬁeld components of the LP modes are summarized in Table 10.6, which reconﬁrms that, as expected, longitudinal components become complicated while the transverse ones are simpliﬁed. The characteristic equation of the LP modes, Eq. (10.48), can be expressed in another way, which will be useful for further developments in the next section. It is also easier to correlate with physical intuition, arising from the ﬁelds shown in Table 10.6. Let us take Eq. (10.48) with the plus sign, the expression for jν which results from Eq. (10.37), and the expression for bν which results from: Kν (w) =

© 2006 by Taylor & Francis Group, LLC

ν Kν (w) − Kν+1 (w) . w

(10.59)

368

DIELECTRIC WAVEGUIDES Table 10.6 LP modes of the circular step-index rod.

The expressions shown in this table are related to the second among Eqs. (10.56). The other three cases can be obtained from these with the following replacements: π π π . ϕ1 = − ϕ ; ϕ3 = − ϕ ; ϕ4 = ϕ − ( − 1) 2 2 2 Core

Cladding

Ey = E 0

J (ur/a) cos ϕ J (u)

Hx = −nco Ez = −j

Ey η0

E0 u 2ka nco

Ez = −j

J−1 (ur/a) sin( − 1) ϕ J (u)

Hz = −j

J−1 (ur/a) cos( − 1) ϕ J (u)

E0 w 2ka ncl

K−1 (wr/a) sin( − 1) ϕ K (w)

Hz = −j

Ey η0

K+1 (wr/a) sin( + 1) ϕ− K (w) −

E0 u 2ka η0

J+1 (ur/a) cos( + 1) ϕ− J (u) −

K (wr/a) cos ϕ K (w)

Hx = −ncl

J+1 (ur/a) sin( + 1) ϕ+ J (u) +

E y = E0

E0 w 2ka η0

K+1 (wr/a) cos( + 1) ϕ+ K (w) +

K−1 (wr/a) cos( − 1) ϕ K (w)

Let us recall that we are now interested in the characteristic equation for the LP,p mode, corresponding to the EH−1,p , not to the EH,p mode; therefore, let ν = − 1. The ﬁnal result is: u

K−1 (w) J−1 (u) = −w J (u) K (w)

( = 0, 1, 2, . . .) ,

(10.60)

to be associated with Eq. (10.32). Looking carefully at Table 10.6, we see that Eq. (10.60) would be the characteristic equation of the system which we obtain if we stipulate that the radial components of the electric and of the magnetic ﬁeld are continuous at the core-cladding interface, r = a: Erco (r = a)

= Ercl (r = a) ,

Hrco (r = a)

= Hrcl (r = a) .

(10.61)

The second of Eqs. (10.61) is correct, because there is no discontinuity in magnetic permeability at r = a, but the ﬁrst one is not rigorous as long as

© 2006 by Taylor & Francis Group, LLC

DISPERSION IN DIELECTRIC WAVEGUIDES

369

co = cl . In fact, from Chapter 1 we know that the correct continuity condition at r = a is that the normal components of the electric displacement vector, D = E, are equal in the core and in the cladding. This comment indicates another implicit meaning of the weak-guidance approximation (or, equivalently, LP approximation): the approximation implies that we neglect refraction of the ﬁeld lines of the electric vector across the interface between the two media. Consequently, informations concerning polarization of the transverse ﬁelds, if drawn through LP modes, are not accurate. In reality, the linear polarization ratio of any guided mode is never inﬁnite. In spite of this limitation, LP modes are extremely useful in yielding practical information on optical ﬁbers, as we shall see in the following sections. 10.8

Dispersion in dielectric waveguides

In practical applications of dielectric waveguides (among which, optical telecommunication systems are by far and large the most important, nowadays) the main limits to performances come from attenuation and dispersion. Loss in a dielectric waveguide is a rather simple subject to deal with, and will be brieﬂy discussed in Section 10.11. On the contrary, dispersion deserves an accurate discussion, which may sound surprising. In fact, in the previous sections we have established, ﬁrstly, that the phase constant of any guided mode, β, must be in the range between βcl and βco , the intrinsic phase constants of the two dielectric media; secondly, that the diﬀerence in permittivity between these media is very small in all cases of practical interest. Comparing these results with those concerning guided modes in a conducting-wall waveguide, one might infer that in a dielectric waveguide variations of the phase constant vs. frequency are small and smooth, so that dispersion cannot be large. The reason why this conclusion is unacceptable is easy to explain. The lengths of practical microwave waveguides never exceed a few meters, which means, at most, a few hundreds of wavelengths. Optical ﬁbers can be more than hundred kilometers in length, which means more than 1010 wavelengths. Accuracy in modeling dispersion must then grow in proportion to these ﬁgures. Dispersion can be dealt with analytically in the weak-guidance approximation, i.e., when: nco − ncl 1 . (10.62) Δ ncl Given that βcl < β < βco , let us deﬁne a new quantity whose purpose is to expand this range and so emphasize the variations of β vs. frequency. Let: β2 β − ncl − n2cl 2 k k b≡ 2 nco − n2cl nco − ncl

.

(10.63)

Clearly, b is adimensional and varies between 0 and 1. The last approximate step in the previous equation comes from the fact that β/k + ncl nco + ncl . The ratio β/k is often referred to as the eﬀective index of the guided mode

© 2006 by Taylor & Francis Group, LLC

370

DIELECTRIC WAVEGUIDES

b

v Figure 10.9 Normalized phase constants vs. normalized frequency. The numbers refer to LP mode suﬃces.

characterized by the phase constant β. Clearly, its values are conﬁned in the range between ncl and nco . Eqs. (10.27) and (10.44) yield another expression for b, namely: b(v) = 1 −

u2 v2

,

(10.64)

where let us stress that u is also a function of v. Plots of Eq. (10.64) for some LP modes are shown in Figure 10.9 (from Gloge, 1971). Let us calculate, rather than the group velocity vg of the various LP modes, its inverse, referred to as the (per-unit-length) group delay, τg , following the deﬁnition given in Chapter 5, namely: τg =

1 dβ dβ = dω c0 dk

,

(10.65)

√ where c0 = 1/ μ0 0 is the speed of light in free space. In contrast to the usual situation in conducting-wall waveguides at microwave frequencies, contributions due to material dispersion, i.e., to d(μ)/dω = 0 (see Chapter 5), cannot be neglected in dielectric waveguides at optical frequencies. To account for them without making the notation too cumbersome,

© 2006 by Taylor & Francis Group, LLC

DISPERSION IN DIELECTRIC WAVEGUIDES

371

let us deﬁne the following quantities: d(ni k) dβi dni = = ni + k (i = co, cl) , (10.66) dk dk dk which are referred to as the group indices of the core and cladding materials. Let us deﬁne also the quantity: Ni ≡

Δ ≡

Nco − Ncl k d(nco − ncl ) =Δ+ ncl ncl dk

,

(10.67)

which may diﬀer signiﬁcantly from the relative index diﬀerence, Δ, only in waveguides where the core and cladding materials are completely diﬀerent (e.g., silica-plastic ﬁbers). Elaborating on β as a function of b, from Eq. (10.63), one ﬁnds the following rigorous result: dβ 1 1 db = ncl Ncl + (nco Nco − ncl Ncl ) b + v (1 + 2Δb)−1/2 . dk ncl 2 dv (10.68) This expression can be simpliﬁed, given that Δ 1. Let (1 + 2Δb)−1/2 1 − Δb, then neglect terms of order Δ2 , and use Eq. (10.67). The result is: d(vb) 1 dΔ db dβ Ncl + nco Δ − knco v . (10.69) dk dv 2 dk dv For a given waveguide, the ﬁrst term is independent of the particular mode. Accordingly, in multimode propagation regime, diﬀerences in group delay (referred to as intermodal dispersion) are due only to the second and third terms. The last one is essentially negligible for all modes, for which db/dv 0, except, at most, the fundamental mode, but we will return to this later on. The factor d(vb)/dv, which governs the second term, is plotted in Figure 10.10 for some low-order LP modes. It may be seen immediately that some of the modes (those of azimuthal order = 0, 1) are fast very close to cut-oﬀ, when most of their guided power travels in the cladding, where the speed of light is larger than in the core. For the modes of higher azimuthal order, even at cut-oﬀ the power fraction in the cladding is too small for the mode to be fast. In any case, as soon as they are suﬃciently above cut-oﬀ all the modes tend to become faster as the normalized frequency increases, for the same reason that the group velocity increases vs. frequency in a conducting-wall waveguide (see Section 7.4). Figure 10.10 also shows that the diﬀerence between the largest and the smallest normalized group delay depends, in its ﬁne details, on v in a complicated manner, but on the contrary its order of magnitude is unity, and remains essentially constant as soon as v 2.4, as it is in practical multimode ﬁbers. Therefore, the intermodal dispersion can be estimated to be: δτ int =

1 1 nco Δ (Nco − Ncl ) , c0 c0

(10.70)

with an approximation nco /ncl = 1 in the last step. Essentially, this result tells us that in a multimode step-index ﬁber intermodal dispersion equals the

© 2006 by Taylor & Francis Group, LLC

372

DIELECTRIC WAVEGUIDES

d ( vb ) dv

v Figure 10.10 Normalized group delays (waveguide contributions) vs. normalized frequency. The numbers refer to LP mode suﬃces.

diﬀerence between the group delays of two uniform plane waves, traveling in the core and in the cladding, respectively. It is left to the reader as an exercise to prove that an intermodal dispersion of this order of magnitude causes, over a ﬁber length of practical interest (at least 10 km), pulse spreading compatible only with very low bit rates. Actually, experimentally measured intermodal dispersions in step-index ﬁbers are somewhat smaller than these predictions, because the modes close to cutoﬀ are ﬁltered out by their losses, signiﬁcantly higher than those of low-order modes. However, they do not allow transmission of attractive bit rates. There are essentially two ways of improving this performance. One is to change the index proﬁle signiﬁcantly, and will be the subject of the following sections. The other one is to use single-mode ﬁbers. Let us now elaborate further on their dispersion properties. The very small values required for the index diﬀerence (see Section 10.6) implies that core and cladding materials diﬀer at most for a very small amount of dopants. This allows us to let Δ Δ, i.e., Δ independent of wavelength, thus canceling the third term in Eq. (10.69), and rewrite it (in the presence of just the fundamental mode LP0,1 ) as: dβ d(vb) Ncl + (Nco − Ncl ) . (10.71) dk dv In Chapter 5, we saw that a signal is distorted when the group velocity (or

© 2006 by Taylor & Francis Group, LLC

DISPERSION IN DIELECTRIC WAVEGUIDES

373

Figure 10.11 Normalized waveguide dispersion.

group delay) varies with frequency, within the signal spectral band. In the multimode regime, we overlooked this phenomenon, because it is dominated by the much larger intermodal dispersion. In the single-mode regime, we must account for the source being non-monochromatic, and study how the group delay of the guided ﬁeld varies vs. frequency, i.e., for intramodal dispersion, sometimes referred to also as chromatic dispersion. To calculate the derivative of the group delay with respect to angular frequency: s=

1 d2 β dτg = 2 dω c0 dk 2

,

(10.72)

starting with Eq. (10.71) we get: d(Nco − Ncl ) d(vb) v d2 (vb) d2 β ∼ dNcl + + (Nco − Ncl ) . (10.73) = 2 dk dk dk dv k dv 2 Thus, the intramodal dispersion of the LP0,1 mode consists of three terms. The ﬁrst originates from material dispersion. The third term indicates — similar to what we saw in Chapter 7 for conducting-wall waveguides — that a guided wave, being non-TEM, is inevitably dispersive, and therefore is referred to as waveguide dispersion. The second term is a combination of material and waveguide eﬀects. Usually, its magnitude is signiﬁcantly less than that of the ﬁrst and that of the third term. However, before neglecting it completely, we must investigate whether the ﬁrst and third term may cancel each other. Figure 10.11 shows that in the single-mode regime (v < 2.405) the last term in Eq. (10.73) is positive, for any choice of the ﬁber material and geometry. Note that this behavior is opposite to that of a metal-wall waveguide, where the group velocity increases as a function of frequency (anomalous dispersion). This indicates that, in the entire single-mode band, the phenomenon which governs waveguide dispersion is the growth in the power fraction that travels

© 2006 by Taylor & Francis Group, LLC

374

DIELECTRIC WAVEGUIDES

Table 10.7 International standards for intramodal dispersion. Source: International Telecommunication Union (formerly CCITT); G.652 and G.653. Unit: ps/(km·nm). Wavelength range Fiber optimized at 1300 nm Fiber optimized at 1550 nm

1285–1330 nm ≤ 3.5 —

1270–1430 nm ≤6 —

1525–1575 nm ≤ 20 ≤ 3.5

in the core, which was outlined previously. On the other hand, the sign of the ﬁrst term depends on the speciﬁc waveguide material, and on the wavelength, in a given material. In pure amorphous silica, material dispersion is positive at λ < 1.27 μm, negative at longer wavelengths. Consequently, it is possible to ∼ design a weakly guiding glass ﬁber so that its intramodal dispersion Eq. (10.73) goes through zero at one wavelength λ0 > 1.3 μm. The exact zero-dispersion ∼ wavelength, λ0 , is determined by the second term in Eq. (10.73), since the sum of the ﬁrst and third term is close to zero. The simplest and most widely used designs give ﬁbers whose λ0 is close to 1.3 μm, i.e., in the spectral range which is called, as we said before, “second window”. These ﬁbers are usually referred to as standard single-mode ﬁbers. However, the growing interest in exploiting the “third window,” where losses pass through their minimum value, and optical ampliﬁers are commercially available, has encouraged the design of ﬁbers whose zero-dispersion wavelength λ0 is close to 1.5 μm. These are referred to as dispersion shifted (DS) ﬁbers. Typically, their refractive index proﬁle is more complicated than the single step considered here. The reader who is interested in more details should consult many of the books quoted at the end of this chapter. More complicated ﬁbers have been envisaged, where total dispersion is kept below a given upper limit over the whole range from 1.3 μm to 1.55 μm. They are referred to as dispersion ﬂattened ﬁbers. However, interest in these appears to be low, in recent times. Units in current technical use, to measure intramodal dispersion of optical ﬁbers, are ps/(km·nm), i.e., picoseconds (of group delay) divided by kilometers (of ﬁber length) and by nanometers (of source spectral width). Some useful reference data are shown in Table 10.7. Before we close this section, we would like to remind the reader of the contents of Section 5.2. The quantity expressed by Eq. (10.72) is merely the ﬁrst coeﬃcient, in the expansion of the delay as a series of powers of Δλ. When the ﬁrst coeﬃcient is equal to zero, the signal distortion is governed by the next term, which is proportional to Δλ2 . So, at the zero-dispersion wavelength the ﬁber bandwidth is not inﬁnite, but is still orders of magnitude broader than at other wavelengths.

© 2006 by Taylor & Francis Group, LLC

GRADED-INDEX WAVEGUIDES

375

Figure 10.12 Ray model of propagation in a graded-index waveguide.

10.9

Graded-index waveguides

As we saw in the previous section, in a multimode step-index ﬁber intermodal dispersion can be too large for long-distance pulse transmission, even at moderate bit rates. A solution, which has been studied in depth theoretically and experimentally, and successfully applied during the early decades of optical communications, is to reduce intermodal dispersion by “equalizing” the speeds of propagation of the various modes. One can do this by acting on the refractive index distribution in the waveguide cross-section, as a function of the radial coordinate. This idea is easy to justify, qualitatively, in terms of ray optics: “times of ﬂight” along guided ray paths of diﬀerent geometrical lengths can be equalized by increasing the speed of propagation (i.e., decreasing the refractive index n) along the longer paths, namely further away from the waveguide axis (see Figure 10.12). Propagation along graded-index waveguides cannot be studied rigorously, in vector terms. However, most of the results of practical interest can be obtained by means of an analysis based on the WKBJ approximation, which was presented, in general, in Section 5.7. This will be the subject of the next section. Before we proceed in this direction, however, it is worthwhile to sketch brieﬂy, omitting some details, how one can pass from Maxwell’s equations to a rather scalar-looking problem, which can then be tackled using the WKBJ method. Let us see whether Eqs. (7.4), or Eqs. (7.11) and (7.12), which express transverse ﬁelds in terms of the longitudinal components Ez , Hz , are still valid when , and maybe μ, depend on the transverse coordinates. Using Maxwell’s equations, one can show that this is indeed the case. On the contrary, Ez and Hz do not satisfy Eq. (7.8), in an inhomogeneous medium. This implies that the theory of guided waves that we developed for conducting-wall waveguides, and then for step-index ﬁbers, does not hold for graded index waveguides, because it was based on the pair of scalar functions which were called prepotentials in Section 3.8. To replace Eq. (7.8) with two new equations, let us go back to Maxwell’s equations, calculate the curl of both sides, replace ∇×∇× with −∇2 +∇ (∇·), and account for Eqs. (1.35) and (1.38), which, for = constant, μ = constant, yield ∇ · H = −(1/μ)∇μ · H and ∇ · E = −(1/)∇ · E, respectively. The ﬁnal

© 2006 by Taylor & Francis Group, LLC

376

DIELECTRIC WAVEGUIDES

results are the following equations: 2 ∇t χ 2 ∇t μ ∇t Hz + χ2 Hz ζˆ + − × ∇t Hz × ζˆ = 2 χ μ

−γ ∇t χ2 × ∇t Ez , jωμ χ2 (10.74)

2 ∇t χ 2 ∇t − ∇t Ez + χ2 Ez ζˆ + × ∇t Ez × ζˆ = 2 χ

−γ ∇2t χ2 × ∇t Hz , jω χ2 (10.75)

where it is important not to forget that χ2 = γ 2 + ω 2 μ is now a function of the radial coordinate, r. Given our interest in dielectric waveguides, similar to those of Section 10.8, let μ = μ0 = constant, = (r) = 0 n2 (r), where: 2 n0 [1 − 2Δf (r)] , r

© 2006 by Taylor & Francis Group, LLC

Electromagnetic Waves Carlo G. Someda

Boca Raton London New York

A CRC title, part of the Taylor & Francis imprint, a member of the Taylor & Francis Group, the academic division of T&F Informa plc.

© 2006 by Taylor & Francis Group, LLC

9589_Discl.fm Page 1 Friday, December 2, 2005 11:06 AM

Published in 2006 by CRC Press Taylor & Francis Group 6000 Broken Sound Parkway NW, Suite 300 Boca Raton, FL 33487-2742 © 2006 by Taylor & Francis Group, LLC CRC Press is an imprint of Taylor & Francis Group No claim to original U.S. Government works Printed in the United States of America on acid-free paper 10 9 8 7 6 5 4 3 2 1 International Standard Book Number-10: 0-8493-9589-5 (Hardcover) International Standard Book Number-13: 978-0-8493-9589-5 (Hardcover) Library of Congress Card Number 2005043549 This book contains information obtained from authentic and highly regarded sources. Reprinted material is quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable efforts have been made to publish reliable data and information, but the author and the publisher cannot assume responsibility for the validity of all materials or for the consequences of their use. No part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers. For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC) 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged. Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe.

Library of Congress Cataloging-in-Publication Data Someda, Carlo G. Electromagnetic waves / Carlo G. Someda.--2nd ed. p. cm. Includes bibliographical references and index. ISBN 0-8493-9589-5 1. Electromagnetic waves. I. Title. QC661.S66 2006 539.2--dc22

2005043549

Visit the Taylor & Francis Web site at http://www.taylorandfrancis.com Taylor & Francis Group is the Academic Division of Informa plc.

© 2006 by Taylor & Francis Group, LLC

and the CRC Press Web site at http://www.crcpress.com

To Gian Carlo Corazza and Enrique A.J. Marcatili who taught me how to mix Maxwell’s equations and practical life

To my grandchildren, Matilde and Edoardo

© 2006 by Taylor & Francis Group, LLC

Contents

Preface

xiii

Acknowledgments

xv

Preface to the Second Edition

xvii

1 Basic equations for electromagnetic ﬁelds 1.1 Introduction: Experimental laws 1.2 Maxwell’s equations and the charge continuity equation 1.3 Constitutive relations 1.4 Imposed currents 1.5 Divergence equations 1.6 Continuity conditions 1.7 The wave equation. The Helmholtz equation 1.8 Magnetic vector potential 1.9 Fitzgerald electric vector potential 1.10 Hertz vector potential 1.11 Further applications and suggested reading

1 1 4 5 9 11 13 15 18 23 25 26

References

33

2 Polarization 2.1 Introduction 2.2 Steinmetz representation of time-harmonic vectors 2.3 Parallel and orthogonal complex vectors 2.4 Properties of time-harmonic vectors 2.5 Properties of the complex vectors 2.6 Linear polarization ratio 2.7 Circular polarization ratio 2.8 Stokes parameters 2.9 The Poincar´e sphere 2.10 Evolution of polarization in a linear medium: Jones matrix 2.11 Further applications and suggested reading

35 35 36 37 38 39 40 41 42 45 46 48

References

51 vii

© 2006 by Taylor & Francis Group, LLC

viii

CONTENTS

3 General theorems 3.1 Introduction 3.2 Poynting’s theorem. Wave impedance 3.3 Uniqueness theorem 3.4 Reciprocity theorem 3.5 Equivalence theorem 3.6 Induction theorem 3.7 Duality theorem 3.8 TE-TM ﬁeld decomposition theorem 3.9 Spatial symmetries. Reﬂection operators 3.10 Further applications and suggested reading

53 53 53 58 61 63 69 71 72 75 79

References

87

4 Plane waves in isotropic media 4.1 Separability of variables in the homogeneous Helmholtz equation 4.2 Solution of the homogeneous Helmholtz equation in Cartesian coordinates 4.3 Plane waves: Terminology and classiﬁcation 4.4 Traveling waves. Phase velocity 4.5 Standing waves 4.6 Poynting vector and wave impedance 4.7 Completeness of plane waves 4.8 Reﬂection and refraction of plane waves 4.9 Fresnel formulas 4.10 Reﬂection in multilayer structures 4.11 Total reﬂection 4.12 Reﬂection on the surface of a good conductor 4.13 Further applications and suggested reading

89

90 93 96 98 101 105 108 113 115 118 122 124

References

131

5 Plane wave packets and beams 5.1 Modulated waves. Group velocity 5.2 Dispersion 5.3 The scalar approximation 5.4 The equations of geometrical optics 5.5 Geometrical optics: Electromagnetic implications 5.6 Examples of ray tracing in radio propagation and in optics 5.7 The WKBJ method 5.8 Further comments on the WKBJ method 5.9 Gaussian beams 5.10 Hermite-Gauss and Laguerre-Gauss modes 5.11 Reﬂection and refraction of Gaussian beams

133 133 137 140 142 147 149 152 155 157 162 168

© 2006 by Taylor & Francis Group, LLC

89

CONTENTS 5.12 On the completeness of a series 5.13 Further comments on rays and beams 5.14 Further applications and suggested reading

ix 171 173 174

References

183

6 Plane waves in anisotropic media 6.1 General properties of anisotropic media 6.2 Wave equations and potentials in anisotropic media 6.3 Birefringent media 6.4 Fresnel’s equation of wave normals 6.5 An application: Phase matching of two waves 6.6 Gyrotropic media 6.7 The Appleton-Hartree formula 6.8 An example of permittivity dyadic 6.9 Second example of permeability dyadic 6.10 Faraday rotation 6.11 Further applications and suggested reading

185 185 188 189 193 197 201 203 206 210 211 216

References

221

7 Waveguides with conducting walls 7.1 Introduction 7.2 Homogeneously ﬁlled cylindrical structures: Simpliﬁed proof of the TE-TM decomposition theorem 7.3 Waveguides with ideal conducting walls 7.4 Transmission modes of lossless cylindrical structures 7.5 Mode orthogonality 7.6 Some remarks on completeness 7.7 Rectangular waveguides 7.8 Circular waveguides and coaxial cables 7.9 Waveguides with nonideal walls 7.10 On wall impedances 7.11 Hybrid modes 7.12 Further applications and suggested reading

223 223

References

267

8 Waves on transmission lines 8.1 Introduction 8.2 Uniform transmission lines 8.3 Impedance transformation along a transmission line 8.4 Lossless transmission lines 8.5 Low-loss transmission lines 8.6 Partially standing waves

269 269 270 272 273 275 276

© 2006 by Taylor & Francis Group, LLC

224 228 229 235 238 239 244 250 254 258 260

x

CONTENTS 8.7 8.8 8.9 8.10 8.11 8.12 8.13

The Smith chart Remote measurement of the load impedance Impedance matching Transmission-line equations: An alternative derivation TEM and quasi-TEM propagation in planar lines The coupled-mode equations Further applications and suggested reading

278 282 284 291 297 301 305

References

311

9 Resonant cavities 9.1 Introduction 9.2 Separable coordinate systems in three dimensions 9.3 Completeness of resonator modes 9.4 Mode orthogonality in a perfect resonator 9.5 Lossless cylindrical cavities 9.6 Simple examples 9.7 Lossy resonators: Perturbation analysis. Intrinsic Q-factor 9.8 Resonators coupled to external loads. Loaded Q-factor 9.9 Open resonators 9.10 Stability of open resonators 9.11 Q-factor of an open resonator 9.12 Further applications and suggested reading

313 313 314 315 317 318 321 325 327 328 330 333 335

References

341

10 Dielectric waveguides 343 10.1 Introduction 343 10.2 Waves guided by a surface of discontinuity. The characteristic equation 344 10.3 Guided modes of a slab waveguide 349 10.4 Radiation modes of a slab waveguide 354 10.5 The cylindrical rod: Exact modes 356 10.6 Modal cut-oﬀ in the cylindrical rod 360 10.7 Weakly guiding rods: The LP modes 363 10.8 Dispersion in dielectric waveguides 369 10.9 Graded-index waveguides 375 10.10 The alpha proﬁles: An important class of multimode gradedindex ﬁbers 380 10.11 Attenuation in optical ﬁbers 385 10.12 Further applications and suggested reading 389 References

395

11 Retarded potentials

397

© 2006 by Taylor & Francis Group, LLC

CONTENTS 11.1 11.2 11.3 11.4 11.5 11.6 11.7 11.8 11.9 11.10 11.11 11.12

Introduction Green’s functions for the scalar Helmholtz equation Lorentz-gauge vector potentials in a homogeneous medium Field vectors in terms of dyadic Green’s functions Inhomogeneous media: Polarization currents Time-domain interpretation of Green’s functions Green’s function expansion into orthogonal eigenfunctions An example: Field in a rectangular box Spherical harmonics Multipole expansion An introduction to cylindrical harmonics Further applications and suggested reading

xi 397 398 401 404 406 407 410 412 414 420 424 426

References

431

12 Fundamentals of antenna theory 12.1 Introduction 12.2 Equivalent dipole moment of an extended source 12.3 Far-ﬁeld approximations 12.4 First example: Short electric-current element 12.5 Characterization of antennas 12.6 Behavior of receiving antennas. Reciprocity 12.7 Examples 12.8 Antenna arrays 12.9 Broad-side and end-ﬁre arrays 12.10 Further applications and suggested reading

433 433 435 437 439 444 448 454 464 469 471

References

477

13 Diﬀraction 13.1 Introduction 13.2 The diﬀraction integral: The vector formulation 13.3 Illumination conditions. Babinet’s principle 13.4 The scalar theory of diﬀraction 13.5 Diﬀraction formulas and Rayleigh-Sommerfeld 13.6 The Fresnel diﬀraction region 13.7 The Fraunhofer diﬀraction region 13.8 Examples 13.9 The ﬁeld near a focus: First example of Fresnel diﬀraction 13.10 Diﬀraction from a straight edge: Second example of Fresnel diﬀraction 13.11 A short note on the geometrical theory of diﬀraction 13.12 Further applications and suggested reading

479 479 479 484 488 493 495 497 500 508

References

521

© 2006 by Taylor & Francis Group, LLC

510 515 516

xii

CONTENTS

14 An introduction to the theory of coherence 14.1 Background and purpose of the chapter 14.2 The analytical signal 14.3 Complex degree of coherence 14.4 Temporal coherence of a source 14.5 Spatial coherence of a source 14.6 Higher-order coherence: An introduction 14.7 An introduction to photocount distributions 14.8 Modal noise in optical-ﬁber transmission systems: A short outline 14.9 Further applications and suggested reading

523 523 523 526 527 529 531 535

References

545

Appendices

547

A Vector calculus: Deﬁnitions and fundamental theorems

549

539 540

B Vector diﬀerential operators in frequently used reference systems 553 C Vector identities

555

D Fundamentals on Bessel functions D.1 Bessel, Neumann and Hankel functions D.2 Modiﬁed Bessel functions D.3 Bessel function formulas

557 557 560 561

References

565

Further Suggested Reading

567

© 2006 by Taylor & Francis Group, LLC

Preface Is there a real need to publish, in 1997, a new book on electromagnetic waves ? Clearly, the answer to this question will come from the market, not from the author. What the author can do is just to try to explain why he accepted to take this challenge. In the late 1970’s, UTET, one of the most important and prestigious Italian publishers, decided to prepare a series of volumes in electrical and electronic engineering, written by Italian authors. The overall purpose was to provide a complete cross-section of how these disciplines were taught in Italian universities. Francesco Barozzi, a very distinguished professor, then head of the Electrical and Electronic Engineering Department at the University of Trieste, was appointed as the editor in chief. He was a charming gentleman, with an immense, ubiquitous background in humanities and arts, which he never showed oﬀ. With a friendly smile, he always took an extremely calm approach to any kind of problem, making everybody feel at ease. To be selected by him as one of the authors was so rewarding, that my decision to prepare a volume for the UTET series was taken in less than one second. Writing the book was not equally fast: it was published in 1986. After that, for the ﬁrst few years its presence on the market was timid, but later sales increased to a very satisfactory level, which is now steadily maintained. Of course, this statement must be calibrated with respect to the limited size of the potential market: unfortunately, there is just one country where electromagnetic waves are taught in Italian! There are, on the other hand, several scientists around the world, who can read Italian. One of them, professor Michael C. Sexton, from University College Cork, Ireland, saw my book during one of his frequent visits to Padua in relationship to his research in the ﬁeld of plasma physics and fusion engineering. He was the ﬁrst who suggested to write an English version. Shortly after, Dr. Richard J. Black, who was then teaching at the Ecole Polytechnique in Montr´eal, Canada, and therefore was familiarizing himself with technical literature in the Romance languages, expressed the same opinion. They said “there is good stuﬀ in there.” I do not know precisely what they meant. On my part, I can say, comparing the contents of my book with those of many others, that I was fortunate to work, in a relatively short time interval, in three ﬁelds: microwaves, lasers, and ﬁber optics. What I tried to do in teaching, and consequently in writing my book, was to strengthen the links, and unify the background. The reader will judge whether I achieved this goal. The role Professor Sexton played was not only to trigger my interest in xiii © 2006 by Taylor & Francis Group, LLC

xiv

PREFACE

an English version of my book. He also contacted Chapman & Hall. The project was approved by the publisher under the condition that some changes were made, with respect to the Italian edition. The suggestions made to this purpose by anonymous referees matched exactly my own opinions. Indeed, my experiences as a student indicated that in the Latin countries there is a long tradition — maybe even a philosophical attitude — to teach “by concepts.” Conversely, in English-speaking countries one teaches — and learns — “by examples.” As a result, the English book could not be a mere translation, but had to be enriched with examples, and with more problems. As a net result, this book is very large. Maybe it is too thick, but I would not know what to cut. For this reason, I am very grateful to the publisher, who never put on me any pressure in order to reduce the market risks that this unusual size entails. C.G. Someda

© 2006 by Taylor & Francis Group, LLC

Acknowledgments The list of those towards whom I became indebted, during the preparation of this book, has no upper limits. Let me begin by repeating two names that I already mentioned in the Preface: those of Professor Michael C. Sexton and Dr. Richard J. Black. Their contributions range from suggesting the material to be added to the Italian version, and choosing examples and problems, to polishing my English, and adding to the references. In one word, their roles are quite similar to those of co-authors. It was very generous on their part to leave only to me the honor to sign this book. Next, I wish to thank an entire team of graduate students, post-doctoral fellows, and research assistants, who spent so much of their time and energy doing the initial translations, and then reading the manuscript over and over again, looking for no other reward but to see me happy and thankful. They were so many, that the list of their names would be too long, and the probability of forgetting someone would be too high. I am sure they understand what I mean, and recognize themselves in these lines. I want to mention explicitly, and to thank enthusiastically, Ms. Barbara Sicoli, who polished with professionality and patience the ﬁnal version of the manuscript, and Ms. Flavia Bergamin, who did the computer processing of almost all the ﬁgures. Finally, let me thank UTET, the Publisher of the Italian text, for the generous permission to publish the English version as a totally independent book, and Chapman & Hall, for being so patient during the long preparation of the manuscript, and for encouraging me when the goal looked out of my reach.

xv © 2006 by Taylor & Francis Group, LLC

Preface to the Second Edition

The ﬁrst edition of my “Electromagnetic Waves” is now seven years old not so old, in a ﬁeld where basic knowledge has a long record of stability, and technological breakthroughs occur on a time scale of several years, if not decades. Why a second edition, then? In fact, the main motivation does not come from the evolution of the contents, but from facts that took place in the book company arena. The publisher of the ﬁrst edition, Chapman & Hall, was acquired by another group - a prominent one - long ago, shortly after publishing my book. What I learnt during the following years was that, when similar events occur, regardless of the size and of the prestige of the acquiring group, redistributing the titles of the acquired company is a very complicated issue, which does not necessarily entail happy endings for all authors. It turned out that, without anybody being speciﬁcally guilty for that, the highly positive, I would almost say enthusiastic, reviews that my book had enjoyed when it appeared, were not followed by a marketing campaign of comparable intensity. Loss of enthusiasm can be contagious. Just to mention one of the negative fallouts, I never completed the exercise manual. Sales declined. The reader can imagine, then, how grateful I am to the Taylor & Francis Group, and to its CRC Press branch in particular, for proposing a second edition at the right moment. The ﬁrst one was almost sold out, and I was facing a bifurcation: either restart with new motivations, or abandon the whole project. I believe that the whole team wants now to show that we made the right decision. For what refers to the contents, the list of changes is very short. I have added, here and there, a few deﬁnitions, which are not new, but whose importance I had underestimated in the past; just to quote an example, antenna eﬀective height, in Chapter 12. The reference list has been updated, as a consequence of the wide set of ﬁrst-class books that were published in these seven years. The text of several problems has been edited, to remove some ambiguities or to complete, here and there, insuﬃcient data. The discovery of little imperfections in the problems occurred while their solutions were being prepared; this entails that the manual should be on the market soon after the book. Last but not least, several tens of misprints and minor mistakes that, in spite of automatic correctors etc., appeared in the ﬁrst edition (an extremely frustrating experience) have been taken care of. This is a signiﬁcant improvement, although, I am afraid, typographical perfection remains a dream. I am xvii © 2006 by Taylor & Francis Group, LLC

xviii

PREFACE TO THE SECOND EDITION

very grateful to an innumerable list of colleagues, friends, students, who had an active role in identifying these errors. To conclude, let me thank collectively all those who contributed to the second edition. Two persons, however, deserve more than anonymous thanks, as they have been instrumental in preparing the manuscript. One is Mr. Emanuele Zattin, who, while completing his education as a computer engineer, has developed a terriﬁc experience as a professional electronic editor. It took all his skills to recover the previous ﬁles, and make them compatible with state-of-the-art software. The second one is a junior colleague of mine, Antonio Daniele Capobianco. Without Tony’s continuous dedication and unlimited patience, preparation of this manuscript would have been an endless, painful process, not only for me, but for many other people.

C.G. Someda

© 2006 by Taylor & Francis Group, LLC

Electromagnetic Waves

© 2006 by Taylor & Francis Group, LLC

Electromagnetic Waves Carlo G. Someda

Boca Raton London New York

A CRC title, part of the Taylor & Francis imprint, a member of the Taylor & Francis Group, the academic division of T&F Informa plc.

© 2006 by Taylor & Francis Group, LLC

9589_Discl.fm Page 1 Friday, December 2, 2005 11:06 AM

Published in 2006 by CRC Press Taylor & Francis Group 6000 Broken Sound Parkway NW, Suite 300 Boca Raton, FL 33487-2742 © 2006 by Taylor & Francis Group, LLC CRC Press is an imprint of Taylor & Francis Group No claim to original U.S. Government works Printed in the United States of America on acid-free paper 10 9 8 7 6 5 4 3 2 1 International Standard Book Number-10: 0-8493-9589-5 (Hardcover) International Standard Book Number-13: 978-0-8493-9589-5 (Hardcover) Library of Congress Card Number 2005043549 This book contains information obtained from authentic and highly regarded sources. Reprinted material is quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable efforts have been made to publish reliable data and information, but the author and the publisher cannot assume responsibility for the validity of all materials or for the consequences of their use. No part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers. For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC) 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged. Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe.

Library of Congress Cataloging-in-Publication Data Someda, Carlo G. Electromagnetic waves / Carlo G. Someda.--2nd ed. p. cm. Includes bibliographical references and index. ISBN 0-8493-9589-5 1. Electromagnetic waves. I. Title. QC661.S66 2006 539.2--dc22

2005043549

Visit the Taylor & Francis Web site at http://www.taylorandfrancis.com Taylor & Francis Group is the Academic Division of Informa plc.

© 2006 by Taylor & Francis Group, LLC

and the CRC Press Web site at http://www.crcpress.com

To Gian Carlo Corazza and Enrique A.J. Marcatili who taught me how to mix Maxwell’s equations and practical life

To my grandchildren, Matilde and Edoardo

© 2006 by Taylor & Francis Group, LLC

Contents

Preface

xiii

Acknowledgments

xv

Preface to the Second Edition

xvii

1 Basic equations for electromagnetic ﬁelds 1.1 Introduction: Experimental laws 1.2 Maxwell’s equations and the charge continuity equation 1.3 Constitutive relations 1.4 Imposed currents 1.5 Divergence equations 1.6 Continuity conditions 1.7 The wave equation. The Helmholtz equation 1.8 Magnetic vector potential 1.9 Fitzgerald electric vector potential 1.10 Hertz vector potential 1.11 Further applications and suggested reading

1 1 4 5 9 11 13 15 18 23 25 26

References

33

2 Polarization 2.1 Introduction 2.2 Steinmetz representation of time-harmonic vectors 2.3 Parallel and orthogonal complex vectors 2.4 Properties of time-harmonic vectors 2.5 Properties of the complex vectors 2.6 Linear polarization ratio 2.7 Circular polarization ratio 2.8 Stokes parameters 2.9 The Poincar´e sphere 2.10 Evolution of polarization in a linear medium: Jones matrix 2.11 Further applications and suggested reading

35 35 36 37 38 39 40 41 42 45 46 48

References

51 vii

© 2006 by Taylor & Francis Group, LLC

viii

CONTENTS

3 General theorems 3.1 Introduction 3.2 Poynting’s theorem. Wave impedance 3.3 Uniqueness theorem 3.4 Reciprocity theorem 3.5 Equivalence theorem 3.6 Induction theorem 3.7 Duality theorem 3.8 TE-TM ﬁeld decomposition theorem 3.9 Spatial symmetries. Reﬂection operators 3.10 Further applications and suggested reading

53 53 53 58 61 63 69 71 72 75 79

References

87

4 Plane waves in isotropic media 4.1 Separability of variables in the homogeneous Helmholtz equation 4.2 Solution of the homogeneous Helmholtz equation in Cartesian coordinates 4.3 Plane waves: Terminology and classiﬁcation 4.4 Traveling waves. Phase velocity 4.5 Standing waves 4.6 Poynting vector and wave impedance 4.7 Completeness of plane waves 4.8 Reﬂection and refraction of plane waves 4.9 Fresnel formulas 4.10 Reﬂection in multilayer structures 4.11 Total reﬂection 4.12 Reﬂection on the surface of a good conductor 4.13 Further applications and suggested reading

89

90 93 96 98 101 105 108 113 115 118 122 124

References

131

5 Plane wave packets and beams 5.1 Modulated waves. Group velocity 5.2 Dispersion 5.3 The scalar approximation 5.4 The equations of geometrical optics 5.5 Geometrical optics: Electromagnetic implications 5.6 Examples of ray tracing in radio propagation and in optics 5.7 The WKBJ method 5.8 Further comments on the WKBJ method 5.9 Gaussian beams 5.10 Hermite-Gauss and Laguerre-Gauss modes 5.11 Reﬂection and refraction of Gaussian beams

133 133 137 140 142 147 149 152 155 157 162 168

© 2006 by Taylor & Francis Group, LLC

89

CONTENTS 5.12 On the completeness of a series 5.13 Further comments on rays and beams 5.14 Further applications and suggested reading

ix 171 173 174

References

183

6 Plane waves in anisotropic media 6.1 General properties of anisotropic media 6.2 Wave equations and potentials in anisotropic media 6.3 Birefringent media 6.4 Fresnel’s equation of wave normals 6.5 An application: Phase matching of two waves 6.6 Gyrotropic media 6.7 The Appleton-Hartree formula 6.8 An example of permittivity dyadic 6.9 Second example of permeability dyadic 6.10 Faraday rotation 6.11 Further applications and suggested reading

185 185 188 189 193 197 201 203 206 210 211 216

References

221

7 Waveguides with conducting walls 7.1 Introduction 7.2 Homogeneously ﬁlled cylindrical structures: Simpliﬁed proof of the TE-TM decomposition theorem 7.3 Waveguides with ideal conducting walls 7.4 Transmission modes of lossless cylindrical structures 7.5 Mode orthogonality 7.6 Some remarks on completeness 7.7 Rectangular waveguides 7.8 Circular waveguides and coaxial cables 7.9 Waveguides with nonideal walls 7.10 On wall impedances 7.11 Hybrid modes 7.12 Further applications and suggested reading

223 223

References

267

8 Waves on transmission lines 8.1 Introduction 8.2 Uniform transmission lines 8.3 Impedance transformation along a transmission line 8.4 Lossless transmission lines 8.5 Low-loss transmission lines 8.6 Partially standing waves

269 269 270 272 273 275 276

© 2006 by Taylor & Francis Group, LLC

224 228 229 235 238 239 244 250 254 258 260

x

CONTENTS 8.7 8.8 8.9 8.10 8.11 8.12 8.13

The Smith chart Remote measurement of the load impedance Impedance matching Transmission-line equations: An alternative derivation TEM and quasi-TEM propagation in planar lines The coupled-mode equations Further applications and suggested reading

278 282 284 291 297 301 305

References

311

9 Resonant cavities 9.1 Introduction 9.2 Separable coordinate systems in three dimensions 9.3 Completeness of resonator modes 9.4 Mode orthogonality in a perfect resonator 9.5 Lossless cylindrical cavities 9.6 Simple examples 9.7 Lossy resonators: Perturbation analysis. Intrinsic Q-factor 9.8 Resonators coupled to external loads. Loaded Q-factor 9.9 Open resonators 9.10 Stability of open resonators 9.11 Q-factor of an open resonator 9.12 Further applications and suggested reading

313 313 314 315 317 318 321 325 327 328 330 333 335

References

341

10 Dielectric waveguides 343 10.1 Introduction 343 10.2 Waves guided by a surface of discontinuity. The characteristic equation 344 10.3 Guided modes of a slab waveguide 349 10.4 Radiation modes of a slab waveguide 354 10.5 The cylindrical rod: Exact modes 356 10.6 Modal cut-oﬀ in the cylindrical rod 360 10.7 Weakly guiding rods: The LP modes 363 10.8 Dispersion in dielectric waveguides 369 10.9 Graded-index waveguides 375 10.10 The alpha proﬁles: An important class of multimode gradedindex ﬁbers 380 10.11 Attenuation in optical ﬁbers 385 10.12 Further applications and suggested reading 389 References

395

11 Retarded potentials

397

© 2006 by Taylor & Francis Group, LLC

CONTENTS 11.1 11.2 11.3 11.4 11.5 11.6 11.7 11.8 11.9 11.10 11.11 11.12

Introduction Green’s functions for the scalar Helmholtz equation Lorentz-gauge vector potentials in a homogeneous medium Field vectors in terms of dyadic Green’s functions Inhomogeneous media: Polarization currents Time-domain interpretation of Green’s functions Green’s function expansion into orthogonal eigenfunctions An example: Field in a rectangular box Spherical harmonics Multipole expansion An introduction to cylindrical harmonics Further applications and suggested reading

xi 397 398 401 404 406 407 410 412 414 420 424 426

References

431

12 Fundamentals of antenna theory 12.1 Introduction 12.2 Equivalent dipole moment of an extended source 12.3 Far-ﬁeld approximations 12.4 First example: Short electric-current element 12.5 Characterization of antennas 12.6 Behavior of receiving antennas. Reciprocity 12.7 Examples 12.8 Antenna arrays 12.9 Broad-side and end-ﬁre arrays 12.10 Further applications and suggested reading

433 433 435 437 439 444 448 454 464 469 471

References

477

13 Diﬀraction 13.1 Introduction 13.2 The diﬀraction integral: The vector formulation 13.3 Illumination conditions. Babinet’s principle 13.4 The scalar theory of diﬀraction 13.5 Diﬀraction formulas and Rayleigh-Sommerfeld 13.6 The Fresnel diﬀraction region 13.7 The Fraunhofer diﬀraction region 13.8 Examples 13.9 The ﬁeld near a focus: First example of Fresnel diﬀraction 13.10 Diﬀraction from a straight edge: Second example of Fresnel diﬀraction 13.11 A short note on the geometrical theory of diﬀraction 13.12 Further applications and suggested reading

479 479 479 484 488 493 495 497 500 508

References

521

© 2006 by Taylor & Francis Group, LLC

510 515 516

xii

CONTENTS

14 An introduction to the theory of coherence 14.1 Background and purpose of the chapter 14.2 The analytical signal 14.3 Complex degree of coherence 14.4 Temporal coherence of a source 14.5 Spatial coherence of a source 14.6 Higher-order coherence: An introduction 14.7 An introduction to photocount distributions 14.8 Modal noise in optical-ﬁber transmission systems: A short outline 14.9 Further applications and suggested reading

523 523 523 526 527 529 531 535

References

545

Appendices

547

A Vector calculus: Deﬁnitions and fundamental theorems

549

539 540

B Vector diﬀerential operators in frequently used reference systems 553 C Vector identities

555

D Fundamentals on Bessel functions D.1 Bessel, Neumann and Hankel functions D.2 Modiﬁed Bessel functions D.3 Bessel function formulas

557 557 560 561

References

565

Further Suggested Reading

567

© 2006 by Taylor & Francis Group, LLC

Preface Is there a real need to publish, in 1997, a new book on electromagnetic waves ? Clearly, the answer to this question will come from the market, not from the author. What the author can do is just to try to explain why he accepted to take this challenge. In the late 1970’s, UTET, one of the most important and prestigious Italian publishers, decided to prepare a series of volumes in electrical and electronic engineering, written by Italian authors. The overall purpose was to provide a complete cross-section of how these disciplines were taught in Italian universities. Francesco Barozzi, a very distinguished professor, then head of the Electrical and Electronic Engineering Department at the University of Trieste, was appointed as the editor in chief. He was a charming gentleman, with an immense, ubiquitous background in humanities and arts, which he never showed oﬀ. With a friendly smile, he always took an extremely calm approach to any kind of problem, making everybody feel at ease. To be selected by him as one of the authors was so rewarding, that my decision to prepare a volume for the UTET series was taken in less than one second. Writing the book was not equally fast: it was published in 1986. After that, for the ﬁrst few years its presence on the market was timid, but later sales increased to a very satisfactory level, which is now steadily maintained. Of course, this statement must be calibrated with respect to the limited size of the potential market: unfortunately, there is just one country where electromagnetic waves are taught in Italian! There are, on the other hand, several scientists around the world, who can read Italian. One of them, professor Michael C. Sexton, from University College Cork, Ireland, saw my book during one of his frequent visits to Padua in relationship to his research in the ﬁeld of plasma physics and fusion engineering. He was the ﬁrst who suggested to write an English version. Shortly after, Dr. Richard J. Black, who was then teaching at the Ecole Polytechnique in Montr´eal, Canada, and therefore was familiarizing himself with technical literature in the Romance languages, expressed the same opinion. They said “there is good stuﬀ in there.” I do not know precisely what they meant. On my part, I can say, comparing the contents of my book with those of many others, that I was fortunate to work, in a relatively short time interval, in three ﬁelds: microwaves, lasers, and ﬁber optics. What I tried to do in teaching, and consequently in writing my book, was to strengthen the links, and unify the background. The reader will judge whether I achieved this goal. The role Professor Sexton played was not only to trigger my interest in xiii © 2006 by Taylor & Francis Group, LLC

xiv

PREFACE

an English version of my book. He also contacted Chapman & Hall. The project was approved by the publisher under the condition that some changes were made, with respect to the Italian edition. The suggestions made to this purpose by anonymous referees matched exactly my own opinions. Indeed, my experiences as a student indicated that in the Latin countries there is a long tradition — maybe even a philosophical attitude — to teach “by concepts.” Conversely, in English-speaking countries one teaches — and learns — “by examples.” As a result, the English book could not be a mere translation, but had to be enriched with examples, and with more problems. As a net result, this book is very large. Maybe it is too thick, but I would not know what to cut. For this reason, I am very grateful to the publisher, who never put on me any pressure in order to reduce the market risks that this unusual size entails. C.G. Someda

© 2006 by Taylor & Francis Group, LLC

Acknowledgments The list of those towards whom I became indebted, during the preparation of this book, has no upper limits. Let me begin by repeating two names that I already mentioned in the Preface: those of Professor Michael C. Sexton and Dr. Richard J. Black. Their contributions range from suggesting the material to be added to the Italian version, and choosing examples and problems, to polishing my English, and adding to the references. In one word, their roles are quite similar to those of co-authors. It was very generous on their part to leave only to me the honor to sign this book. Next, I wish to thank an entire team of graduate students, post-doctoral fellows, and research assistants, who spent so much of their time and energy doing the initial translations, and then reading the manuscript over and over again, looking for no other reward but to see me happy and thankful. They were so many, that the list of their names would be too long, and the probability of forgetting someone would be too high. I am sure they understand what I mean, and recognize themselves in these lines. I want to mention explicitly, and to thank enthusiastically, Ms. Barbara Sicoli, who polished with professionality and patience the ﬁnal version of the manuscript, and Ms. Flavia Bergamin, who did the computer processing of almost all the ﬁgures. Finally, let me thank UTET, the Publisher of the Italian text, for the generous permission to publish the English version as a totally independent book, and Chapman & Hall, for being so patient during the long preparation of the manuscript, and for encouraging me when the goal looked out of my reach.

xv © 2006 by Taylor & Francis Group, LLC

Preface to the Second Edition

The ﬁrst edition of my “Electromagnetic Waves” is now seven years old not so old, in a ﬁeld where basic knowledge has a long record of stability, and technological breakthroughs occur on a time scale of several years, if not decades. Why a second edition, then? In fact, the main motivation does not come from the evolution of the contents, but from facts that took place in the book company arena. The publisher of the ﬁrst edition, Chapman & Hall, was acquired by another group - a prominent one - long ago, shortly after publishing my book. What I learnt during the following years was that, when similar events occur, regardless of the size and of the prestige of the acquiring group, redistributing the titles of the acquired company is a very complicated issue, which does not necessarily entail happy endings for all authors. It turned out that, without anybody being speciﬁcally guilty for that, the highly positive, I would almost say enthusiastic, reviews that my book had enjoyed when it appeared, were not followed by a marketing campaign of comparable intensity. Loss of enthusiasm can be contagious. Just to mention one of the negative fallouts, I never completed the exercise manual. Sales declined. The reader can imagine, then, how grateful I am to the Taylor & Francis Group, and to its CRC Press branch in particular, for proposing a second edition at the right moment. The ﬁrst one was almost sold out, and I was facing a bifurcation: either restart with new motivations, or abandon the whole project. I believe that the whole team wants now to show that we made the right decision. For what refers to the contents, the list of changes is very short. I have added, here and there, a few deﬁnitions, which are not new, but whose importance I had underestimated in the past; just to quote an example, antenna eﬀective height, in Chapter 12. The reference list has been updated, as a consequence of the wide set of ﬁrst-class books that were published in these seven years. The text of several problems has been edited, to remove some ambiguities or to complete, here and there, insuﬃcient data. The discovery of little imperfections in the problems occurred while their solutions were being prepared; this entails that the manual should be on the market soon after the book. Last but not least, several tens of misprints and minor mistakes that, in spite of automatic correctors etc., appeared in the ﬁrst edition (an extremely frustrating experience) have been taken care of. This is a signiﬁcant improvement, although, I am afraid, typographical perfection remains a dream. I am xvii © 2006 by Taylor & Francis Group, LLC

xviii

PREFACE TO THE SECOND EDITION

very grateful to an innumerable list of colleagues, friends, students, who had an active role in identifying these errors. To conclude, let me thank collectively all those who contributed to the second edition. Two persons, however, deserve more than anonymous thanks, as they have been instrumental in preparing the manuscript. One is Mr. Emanuele Zattin, who, while completing his education as a computer engineer, has developed a terriﬁc experience as a professional electronic editor. It took all his skills to recover the previous ﬁles, and make them compatible with state-of-the-art software. The second one is a junior colleague of mine, Antonio Daniele Capobianco. Without Tony’s continuous dedication and unlimited patience, preparation of this manuscript would have been an endless, painful process, not only for me, but for many other people.

C.G. Someda

© 2006 by Taylor & Francis Group, LLC

CHAPTER 2

Polarization

2.1

Introduction

As we said at the beginning of the previous chapter, we take it for granted that the reader has been previously exposed to the fundamentals of circuit and network theory. In particular, we assume that the reader is familiar with the representation of time-harmonic (i.e., varying sinusoidally in time) scalar quantities by means of complex numbers, i.e., with the so-called Steinmetz method (or procedure). This assumption is based on typical electrical engineering and/or electronics curricula, and also applies to curricula in physics in many universities. Those readers for whom this assumption does not apply can grasp the fundamentals of Steinmetz procedure in the next section. However, it is highly recommended that they read books that treat this subject in depth, and especially that they familiarize themselves with the Steinmetz method by solving many problems on circuits. For those readers who, on the contrary, feel conﬁdent that they know the procedure, the ﬁrst part of this chapter may look rather tedious. Nevertheless, we recommend that they read carefully at least the second part, beginning with Section 2.5, as otherwise some subtle details in the following chapters might be diﬃcult to grasp. Throughout this book, substantial beneﬁt can be gained if we assume, whenever possible, a purely monochromatic sinusoidal time dependence of the electromagnetic quantities that we are interested in, such as ﬁelds, potentials, etc. However, in contrast to circuit theory, here most of our variables are vectors, in three-dimensional (3-D) real space. We must ﬁrst ask ourselves how the Steinmetz method can be generalized to time-harmonic vectors. As we will see in the next section, this entails some additional diﬃculty. The rest of this chapter is then devoted to demonstrating why these complications are indeed worth the time and eﬀort they require. The net result will indeed be, that the complex representation of a time-harmonic vector can convey, in compact forms that lend themselves to very elegant analytical and geometrical interpretations, information on an important physical feature of the timeharmonic vector itself, namely, its polarization. Applications are found in all the following chapters, ranging from plane waves to waveguides, from beams to antennas, from diﬀraction to coherence. 35 © 2006 by Taylor & Francis Group, LLC

36 2.2

POLARIZATION Steinmetz representation of time-harmonic vectors

In real 3-D space, let us introduce a completely arbitrary cartesian orthogonal reference frame, {x1 , x2 , x3 }. Let {ˆ a1 , a ˆ2 , a ˆ3 } be the unit-vectors of this frame. We call time-harmonic, with angular frequency ω, a vector whose components in that reference frame vary sinusoidally in time, with angular frequency ω. Let then Sv (ω) be the set of all time-harmonic vectors of angular frequency ω: 3 a ˆn an cos(ωt + ϕn ) ; an ∈ R ; ϕn ∈ R ; an ≥ 0 ; 0 ≤ ϕn < 2π . Sv (ω) ≡ n=1

(2.1) It is easy to show that Sv (ω) satisﬁes the axioms that deﬁne a vector space over the real ﬁeld. We then deﬁne a complex vector as a linear combination of time-independent vectors, each of which belongs to the real 3-D space, where the scalar coefﬁcients are arbitrary complex numbers. Let C3 be the set of all the complex vectors deﬁned in this way: 3 jϕn a ˆn an e ; an ∈ R ; ϕn ∈ R ; an ≥ 0 ; 0 ≤ ϕn < 2π . (2.2) C3 ≡ n=1

Then, it is easy to show that C3 is also a vector space over the real ﬁeld. The relation: 3 3

jϕn jωt a a ˆn an cos(ωt + ϕn ) = Re a ˆn an e e = Re a ejωt ∼ (t) = n=1

n=1

(2.3) is a one-to-one and onto mapping, τ , between Sv (ω) and C3 . This relationship forms the complex representation of time-harmonic vectors, and is also referred to as the vectorial Steinmetz method. Let us deﬁne, in C3 , a scalar product as a straightforward extension of two well-known operations, the inner product of two real vectors, and the product of complex numbers, i.e.: 3 3 3 jϕn jψn a ˆn an e a ˆ n bn e an bn ej(ϕn +ψn ) . (2.4) · = a·b = n=1

n=1

n=1

With this deﬁnition of scalar product, C3 is not a unitary space, according to the deﬁnition given in classical textbooks on linear algebra (e.g., Halmos, 1958), because some axioms are violated. Some consequences, in particular those that involve the deﬁnitions of parallel and orthogonal complex vectors, will be outlined in the next section. Another consequence that we wish to point out is that the mapping in Eq. (2.3) is not an isomorphism, as there is no one-to-one matching between scalar products in the two vector spaces.

© 2006 by Taylor & Francis Group, LLC

PARALLEL AND ORTHOGONAL COMPLEX VECTORS

37

The scalar products in the two spaces are related to each other as follows: 1 Re{a · b∗ } = a ∼·b ∼ , 2

(2.5)

where the bar above the factors on the right-hand side indicates a time average over a period T = 2π/ω. We may then call the mapping in Eq. (2.3) a pseudoisomorphism.

2.3

Parallel and orthogonal complex vectors

Two complex vectors, A and B, are said to be parallel when there exists a complex number c, such that the relationship: A = cB

(2.6)

is satisﬁed. We want to preserve the validity, for complex vectors, of the following relation, which is well known to hold for real vectors and for real numbers: |A| = |c| |B| .

(2.7)

Note that, in order to do that, it is necessary to deﬁne the modulus of a complex vector in the following way: √ (2.8) |A| = A · A∗ . Another notion that we want to preserve from the space of real vectors is that the projection of a complex vector onto its own direction must coincide with its modulus. Then, to be consistent, the projection of a complex vector, A, onto the direction of another, completely arbitrary complex vector, C, must be deﬁned as follows: Ac = A · C∗ /|C| = (C · A∗ )∗ /|C|

.

(2.9)

Consequently, two complex vectors A and C are said to be orthogonal to each other when the following relationship: A · C∗ = A∗ · C = 0

(2.10)

is satisﬁed. Note that in general Eq. (2.10) does not imply A · C = 0. These results are probably surprising, and the reader might believe that they lead, sooner or later, to inconsistencies. On the contrary, we will see that these new rules are instrumental in some steps. For example, in Chapter 4 they will help us to correctly calculate the per-unit-area power carried by an arbitrarily polarized plane wave.

© 2006 by Taylor & Francis Group, LLC

38 2.4

POLARIZATION Properties of time-harmonic vectors

Any element in the set deﬁned by Eq. (2.1) can be written in the following way: 3 3 A a ˆn an cos ϕn cos ωt − a ˆn an sin ϕn sin ωt ∼ = n=1

n=1

= A cos ωt − A sin ωt ,

(2.11)

where the two vectors: A =

3

a ˆn an cos ϕn

,

A =

n=1

3

a ˆn an sin ϕn

(2.12)

n=1

are real and time-independent. Eq. (2.11) implies that, at any time t, the vector A ∼ belongs to the plane deﬁned by the two vectors in Eq. (2.12). Hence, for any time-harmonic vector originating at a given point O, the trajectory described vs. time by its terminal point is a plane curve. Furthermore, as the two vectors in Eq. (2.12) are linearly combined with time-periodic coeﬃcients, cos ωt and sin ωt, that trajectory has to be a closed curve. Let us show that in fact Eq. (2.11) is the parametric equation of an ellipse. Indeed, looking for the equation of that trajectory in any orthogonal reference frame {x , y } on the ˆx + y a ˆy , where a ˆx {A , A } plane (see Figure 2.1), we may write A= ∼ x a and a ˆy are obviously the unit vectors along the axes. Then, as a consequence of Eq. (2.11), we have: x y

= ax cos ωt − ax sin ωt , =

ay

cos ωt −

ay

(2.13)

sin ωt .

(2.14)

We now multiply Eq. (2.13) by ay , and Eq. (2.14) by ax , and subtract. Then, we multiply Eq. (2.13) by ay , and Eq. (2.14) by ax , and subtract. Finally, we square the two expressions, and sum, making use of the identity sin2 ωt + cos2 ωt = 1. The ﬁnal result is: (x ay − y ax )2 + (x ay − y ax )2 = (ax ay − ay ax )2

,

(2.15)

which is the standard form for the equation of an ellipse on the {x , y } plane. This result is usually expressed saying that a generic time-harmonic vector is elliptically polarized. It is usually convenient to express the ellipse in terms of the coordinate system corresponding to what are known as its principal axes. The principal axes are labelled x and y in Figure 2.1, and are parallel to the major and minor axes of the ellipse, respectively. In this reference frame, the instantaneous vector (2.11) can be expressed in the following form: (2.16) A ∼ = Fx cos(ωt + ϕ) − Fy sin(ωt + ϕ) , where Fx is a real vector parallel to the x axis, and Fy is a real vector parallel to the y axis.

© 2006 by Taylor & Francis Group, LLC

PROPERTIES OF THE COMPLEX VECTORS

39

Figure 2.1 The polarization ellipse, its principal axes, and a generic cartesian reference frame.

To ﬁnd the principal axes, when a vector is given in the form Eq. (2.11), is perfectly equivalent to ﬁnding the phase angle ϕ in Eq. (2.16). The problem is much simpler to tackle if one makes use of the corresponding complex vector, and therefore its solution is left to the next section. 2.5

Properties of the complex vectors

From Eq. (2.11) it follows that:

jωt . + j A ) e A = Re (A ∼

(2.17)

Comparing this with Eq. (2.3), we see that the complex vector A, which corresponds, according to the Steinmetz procedure, to the time-harmonic vector A ∼ , is connected to the two time-independent real vectors that appear in Eq. (2.11) by the simple relationship: A = A + j A

.

(2.18)

If we need to express the vector (2.18) in a generic orthogonal cartesian frame {x , y }, we may write: A = Ax a ˆx + Ay a ˆy

,

(2.19)

and the complex numbers Ax , Ay are the components of A in this frame, which is real, so that we may write: Ax

= A·a ˆ∗x = A · a ˆx

,

Ay

= A·a ˆ∗y = A · a ˆy

.

(2.20)

If we want to introduce the reference frame of the principal axes, then, combining Eq. (2.16) with Eq. (2.3), we get: A = Ax a ˆ x + Ay a ˆy = |Fx | ejϕ a ˆx ± j |Fy | ejϕ a ˆy

© 2006 by Taylor & Francis Group, LLC

.

(2.21)

40

POLARIZATION

This result shows that there is another way to deﬁne the principal axes of the ellipse represented by Eq. (2.15). It is the basis in which the ratio between the cartesian components of the complex vector A is purely imaginary, that is, the ratio: Ay , (2.22) p=j Ax is real. On the other hand, in a generic orthogonal basis, the ratio: p = j

Ay Ax

,

(2.23)

is a complex number. As is well known, a rotation of a real reference frame by an angle δ (see again Figure 2.1) entails the following transformation for the vector components: cos δ sin δ Ax Ax = . (2.24) − sin δ cos δ Ay Ay Then, comparing Eq. (2.22) and Eq. (2.23), we see that a simple way to ﬁnd the principal axes, starting from the vector expressed in a generic reference frame, is to impose that the rotation yield a real value for Eq. (2.22), i.e., to impose p = p∗ . We leave it as an exercise for the reader to prove that the ﬁnal result is the following rule: tan 2δ = j

p − p∗ 1 − p p∗

.

(2.25)

This point will be exploited in the next two sections. 2.6

Linear polarization ratio

The complex number in Eq. (2.23) is referred to as the linear polarization ratio of either the time-harmonic vector A ∼ , or the corresponding complex vector A, in the generic orthogonal reference frame {x , y }. When the reference frame is rotated by a generic angle δ , then the components are transformed according to Eq. (2.24), with δ replaced by δ . It is easy to check that the corresponding transformation rule for the linear polarization ratio is: p + j tan δ . (2.26) p = 1 + j p tan δ Eq. (2.26) is a bilinear relationship between p and p . The reader may prove, as an exercise, that an invariant under this transformation is the sign of the real part of the linear polarization ratio. This suggests that this sign has a meaning which is intrinsic to the vector A, independent of the reference frame. Indeed, we may observe that: • if Re(p ) < 0, i.e., when the sign in front of the last term in Eq. (2.21) is a plus, then the time-harmonic vector A ∼ rotates clockwise for an observer watching at the positive face of the {x, y} plane — we refer to this as a left-handed elliptical polarization;

© 2006 by Taylor & Francis Group, LLC

CIRCULAR POLARIZATION RATIO

41

• if Re(p ) > 0, i.e., when the sign in front of the last term in Eq. (2.21) is a minus, then the time-harmonic vector A ∼ rotates counterclockwise for an observer watching at the positive face of the {x, y} plane — we refer to this as a right-handed elliptical polarization; • ﬁnally, if Re(p ) = 0, i.e., if Fy = 0 in Eqs. (2.21) and (2.16), then the ellipse described by the point P = O+ A ∼ (t) degenerates into a segment — we refer to this as a linear polarization. In particular, when p = ±1, then |Fx | = |Fy | in Eqs. (2.21) and (2.16). The ellipse becomes a circle, and we refer to these cases as circular polarizations, either right- or left-handed, obviously depending on the sign of p . 2.7

Circular polarization ratio

Consider the following pair of complex vectors: √ √ ˆ = (ˆ ˆ = (ˆ R ax − j a ˆy )/ 2 , L ax + j a ˆy )/ 2 .

(2.27)

From what we saw in the previous section, they represent two circularly polarized time-harmonic vectors, a right-handed and a left-handed one, respectively. These two complex vectors: • belong to the (x, y) plane; • have moduli equal to unity: ˆ·R ˆ∗ = 1 , R

ˆ·L ˆ∗ = 1 , L

(2.28)

so they can be called unit vectors; • are orthogonal to each other, according to the deﬁnition in Eq. (2.10): ˆ·L ˆ∗ = 0 . R

(2.29)

Therefore, they form an orthonormal basis, which we refer to as the rotating frame. Any complex vector A belonging to the plane of Figure 2.1 can be expanded into this basis, as: ˆ + AL L ˆ . A = AR R

(2.30)

The components of A in this basis, the rotating frame, are related to those in the principal axis frame {x, y} by: AR

ˆ ∗ = √1 (Ax + j Ay ) , = A·R 2

AL

∗ 1 = A·L ˆ = √ (Ax − j Ay ) 2

.

(2.31)

The complex number: AL (2.32) AR is called the circular polarization ratio of the time-harmonic vector A ∼ , or of the corresponding complex vector A. q=

© 2006 by Taylor & Francis Group, LLC

42

POLARIZATION

All the deﬁnitions that we gave in this section originated from quantities that were expressed in the principal axis reference frame, {x, y}. If one starts from a completely arbitrary real orthogonal frame, {x , y }, rotated by an angle δ with respect to the principal axes, one gets, analogously: q =

AL = q ej2δ AR

.

(2.33)

Hence, the quantity |q| is independent of the reference frame, and therefore has an intrinsic meaning. This is reconﬁrmed by the relationship which links the circular polarization ratio to the linear polarization ratio deﬁned in the previous section, namely: 1 − p . (2.34) q = 1 + p This is again a bilinear transformation, which maps the loci |q | = constant on the q complex plane into curves on the p complex plane along which the sign of Re(p ) remains constant. In more detail, we have: • the line Re(p ) = 0 is mapped onto the circle |q | = 1 (locus of linear polarizations); • the half-plane Re(p ) < 0 is mapped onto the region |q | > 1, where lefthanded elliptical polarizations are located. In particular, q = ∞ corresponds to left-handed circular polarization; • the half-plane Re(p ) > 0 is mapped onto the region |q | < 1, where righthanded elliptical polarizations are located. In particular, q = 0 corresponds to right-handed circular polarization. 2.8

Stokes parameters

In the last two sections, we stressed that some properties of the linear and circular polarization ratios are independent of the reference frame one adopts on the plane of the ellipse, Eq. (2.15). Those remarks show how to proceed further, searching for quantities that characterize polarization, and are invariant with respect to a change in the reference frame, or at least, if they can not be constant, change in simple ways. Among many possibilities that were suggested in the past decades, the so-called Stokes parameters became quite popular, being both elegant and useful at the same time. In order to link their deﬁnition to what we have seen so far, let us start from Eq. (2.15), which describes the elliptical trajectory of the point P = O+ A ∼, in the generic frame {x , y } (see Figure 2.1), where the components of the complex vector A are expressed by Eq. (2.20). From now on, the notation that we have been using so far may be simpliﬁed, just because we are looking for quantities that are insensitive to a change in the reference frame. Therefore, let us write the components of the complex vector A as follows, omitting primes, although in general we are not using the

© 2006 by Taylor & Francis Group, LLC

STOKES PARAMETERS

43

Figure 2.2 Geometrical illustration of the symbols X, Y and δ (orientation angle).

principal axes as the reference system: Ax = bx ejϕx ,

Ay = by ejϕy ,

ϕ = ϕ x − ϕy ,

jωt A ∼ as Re(A e ) , (2.35) where the parameters bx , by , ϕx and ϕy are positive real. The equation of the ellipse, Eq. (2.15), can then be rewritten, after a little manipulation, as follows: 2 2 y x x y + −2 cos ϕ = sin2 ϕ . (2.36) bx by bx by

From this, one sees that the angle δ between the major axis of the ellipse and the x axis, referred to as the orientation angle (see Figure 2.2), is given by: 2bx by tan 2δ = 2 cos ϕ , (2.37) bx − b2y while the ratio between the lengths Y and X of, respectively, the minor and major axes of the ellipse is related to the parameters of Eq. (2.36) as follows: 2bx by 1 π Y = ± tan γ , γ = arcsin 2 , (2.38) sin ϕ , |γ| ≤ X 2 bx + b2y 4 where: (i) in line with the previous deﬁnitions, the plus and minus signs apply for the right-handed and left-handed polarizations respectively, and (ii) the quantity γ is referred to as the eccentricity angle. The four Stokes parameters are deﬁned as follows: s1

= Ax A∗x − Ay A∗y = b2x − b2y

s2

= Ax A∗y + A∗x Ay = 2bx by cos ϕ ,

s3

= −j(Ax A∗y − A∗x Ay ) = 2bx by sin ϕ ,

s0

= Ax A∗x + Ay A∗y = b2x + b2y

,

.

(2.39)

All the information concerning the polarization ellipse of Figure 2.1 is concentrated in the two real numbers γ and δ, plus the sign in Eq. (2.38). One

© 2006 by Taylor & Francis Group, LLC

44

POLARIZATION

may then wonder why we introduced a set of four real parameters. The answer may be given as follows. First of all, squaring the expressions in Eq. (2.39), and summing, we ﬁnd: (2.40) s21 + s22 + s23 = s20 , which shows that in any case the four Stokes parameters are not independent. A usual way to express this link in words is to refer to S ≡ (s1 , s2 , s3 ) as the Stokes vector, and to state that |S| = s0 . Furthermore, if we want to keep information on polarization of a time-harmonic vector separate from that on vector amplitude, we may use normalized amplitudes, i.e., impose: s0 = 1 .

(2.41)

In this way, the number of independent Stokes parameters goes down to two, with a further ambiguity in the sign of the third one. This result matches precisely our statement after Eq. (2.39). However, let us emphasize that we were entitled to make the normalization (2.41) only because we were dealing with a purely monochromatic, time-harmonic ﬁeld. The deﬁnition of the Stokes parameters has a broader scope of validity. For example, it may be used for narrow-band signals, i.e., for time-harmonic carriers modulated by slowlyvarying envelopes, a subject with which we will deal in Chapter 5. Another very important example of application of Stokes parameters beyond the scope of the previous sections in this chapter is the representation of incoherent (or partially coherent) ﬁelds, that is, ﬁelds whose slowly varying envelopes are random functions of time. In this case, the products of the type Aj A∗ j that appear in Eq. (2.39) have to be replaced by time averages, and therefore the results are strongly dependent on the assumptions one makes about the correlation properties of the ﬁeld. In all these cases, the diﬀerence between s0 and unity becomes a measure of the “depolarization” of the ﬁeld A ∼ . We advise the reader to return to these comments and reconsider them after studying the last chapter in this book which deals with statistical properties of e.m. ﬁelds. Once the Stokes parameters have been determined in a completely arbitrary reference frame, then it is straightforward to identify, starting from them, the polarization ellipse deﬁned with respect to the same frame. In fact, from Eq. (2.39), one ﬁnds: b2x =

s0 + s1 2

s0 − s1 2

,

b2y =

,

2γ = arcsin

,

ϕ = arctan

s3 s2

,

(2.42)

or, alternatively: 2δ = arctan

s2 s1

s3 s0

,

X 2 + Y 2 = s0

.

(2.43)

If the reference frame is rotated by an angle δ on the plane of the ellipse, then, as the reader may check as an exercise, two parameters (namely, s3 and s0 ) are invariant, while the other two obey the simple transformation rule: s1

© 2006 by Taylor & Francis Group, LLC

= s1 cos 2δ + s2 sin 2δ

,

´ SPHERE THE POINCARE

45

Figure 2.3 Geometrical representation of the Stokes parameters: the Poincar´ e sphere.

s2

= s1 sin 2δ − s2 cos 2δ

,

(2.44)

which is typical of a rotation of the {s1 , s2 } plane by an angle 2δ. The invariance of s3 with respect to this kind of basis transformation is connected to the fact that s3 measures the content of circular polarization in the vector A ∼ , as the reader can infer from the deﬁnition of s3 itself, and as will be reconﬁrmed in the next section. Many of the most interesting properties of Stokes parameters become simple and evident if we adopt a geometrical construction, which will in fact be dealt with in the next section. 2.9

The Poincar´ e sphere

The relationship in Eq. (2.40) between the Stokes parameters lends itself to an immediate geometrical interpretation. If we introduce, in real 3-D space, three mutually orthogonal axes labeled {s1 , s2 , s3 }, then Eq. (2.40) can be read as the equation of a spherical surface of radius s0 centered at the origin. This is usually referred to as the Poincar´e sphere (see Figure 2.3). For a purely monochromatic time-harmonic ﬁeld, as noted in the previous section, one may separate information on polarization from that on amplitude, normalizing the ﬁeld so that s0 = 1. In general, as the reader will be able to appreciate more easily after reading the ﬁnal chapter of this book, the radius of the Poincar´e sphere will convey information regarding the degree of polarization of the ﬁeld one is dealing with. Unit-radius will imply completely polarized ﬁeld, i.e., the state of polarization is purely deterministic and known at any time without any uncertainty. At the opposite extreme, the radius will

© 2006 by Taylor & Francis Group, LLC

46

POLARIZATION

go to zero for a completely unpolarized ﬁeld, i.e., for a ﬁeld whose state of polarization varies randomly in time, with a correlation time shorter than the response time of the instrumentation that measures it, i.e., shorter than the time scale over which the quantities in Eq. (2.39) are averaged. The reader can now easily grasp the simple relationship that exists between the position of a point on the Poincar´e sphere and the parameters of the ellipse traced out by the point P = O+ A ∼ . Indeed, the Eqs. in (2.43) indicate that latitude and longitude on the Poincar´e sphere equal 2γ and 2δ respectively, where γ and δ are respectively the ellipse eccentricity and orientation angles. Consequently: • the upper hemisphere (s3 > 0) is the locus of right-handed polarizations, and the lower hemisphere (s3 < 0) is the locus of left-handed polarizations; • the equator (s3 = 0) is the locus of linear polarizations; • the two poles (s1 = s2 = 0) represent circular polarizations, right-handed at the north pole, and left-handed at the south pole; • mirror symmetry with respect to the plane of the equator reﬂects that, in real 3-D space where the vector A ∼ is deﬁned, a clockwise rotation becomes counterclockwise, and vice-versa, if one reverses the sign of the coordinate axis orthogonal to the plane of the ellipse. We think it might now be simpler for the reader to understand why the Stokes parameters transform according to Eq. (2.44) when the reference frame on the plane of the ellipse is rotated by an angle δ. Purely circular polarizations are completely insensitive to such a change in the reference frame. Linear polarizations remain linear, but their orientation angle, i.e., the angle between the reference frame and the direction of polarization, changes when the axes are rotated. The periodicity in such a change must be π, and that is the source of the factor of two in front of δ in Eq. (2.44). Similarly, implications for elliptical polarizations can be deduced in a straightforward manner. 2.10

Evolution of polarization in a linear medium: Jones matrix

All the deﬁnitions that we gave so far, and all the operations that were described in this chapter, dealt with polarization as a local property. They referred to a vector which was deﬁned at a given point in real 3-D space. On the other hand, it is self-explanatory why, in a book on electromagnetic waves, it is of interest to study how the polarization of a vector wave evolves as it propagates. This subject can be dealt with in depth, grasping the physical meaning of the equations one is writing, only after one familiarizes oneself with the contents of the next three chapters. Nevertheless, it appears to be useful to present here, in a rather abstract form, a general formalism, which is consistent with what we saw in this chapter, and will be applicable to the subjects to be dealt with later on. Consider two points in real 3-D space, Pin and Pout , both situated on the same path along which a wave is propagating. Pin is “upstream,” Pout is

© 2006 by Taylor & Francis Group, LLC

EVOLUTION OF POLARIZATION

47

“downstream”. As we saw in the previous sections, polarization at each point is completely characterized by a complex vector with two nonvanishing components. For simplicity, let us suppose that the two vectors at Pin and Pout , Ain and Aout , respectively, lay in parallel z = constant planes. If the medium between Pin and Pout is linear, then the two vectors Ain and Aout must be linearly related to each other. If we adopt the formalism of Section 2.5, and use the same reference frame {x, y} (not necessarily the principal axes) at the two points, then the relationship between the vectors can be expressed in matrix form, as: J11 J12 Ainx Aoutx = . (2.45) Aouty J21 J22 Ainy The matrix denoted as (J) in Eq. (2.45) is referred to as the Jones matrix. If one changes the reference frame, rotating it by an angle δ (see again Figure 2.1), then the Jones matrix in the new system, (J ), is related to that in the old system by: (J ) = (R) (J) (R)−1 , (2.46) where (R) is the rotation matrix that appears in Eq. (2.24). The physical reasons why polarization may change as a wave propagates along a linear system are not simple to understand at this stage. As previously mentioned, they will be clariﬁed in the following chapters. However, it is rather intuitive that, independent of the causes, the two components of the vector A in may experience either a diﬀerential loss (or gain, in an active medium), or a diﬀerential phase shift. A simple example of the ﬁrst kind is an ideal polarizer, whose Jones matrix, in a reference frame where the y axis is along the blocking direction, is: 1 0 . (2.47) (J0 ) = 0 0 A simple example of the second kind is a so-called quarter-wave retardation plate, i.e., a device which introduces a delay of π/2 radians on one component (say the one along y) with respect to the other. Then (except for an undeﬁned factor of unit modulus in front of the matrix, which represents an irrelevant phase shift, equal for all components), the Jones matrix reads: 1 0 . (2.48) (Jπ/2 ) = 0 j When two or more linear systems are cascaded, along the path of a given wave, then it is evident from the deﬁnition that the Jones matrix of the whole system is the product, in the appropriate order, of the Jones matrices of the individual components. Some of the problems at the end of this chapter deal with these subjects. Another point, which is probably much less obvious, and therefore must be emphasized here, is that the Jones matrix does not always lend itself to the inverse operation, namely, to reconstruct the input vector polarization from that of the output vector. An example in this sense is provided by the ideal polarizer: the matrix in Eq. (2.47) cannot be inverted,

© 2006 by Taylor & Francis Group, LLC

48

POLARIZATION

as it is singular. Indeed, this corresponds to the physical fact that the output of an ideal polarizer is, by deﬁnition, linearly polarized in a given direction, irrespective of the input ﬁeld. Similarly, in general one can not infer that the Jones matrix remains invariant when a wave travels through the same device in the opposite direction, i.e., when the roles of the points Pin and Pout are interchanged. Such a statement will have to be checked against the so-called reciprocity theorem, to be proved in the next chapter. 2.11

Further applications and suggested reading

For further details regarding the complex (phasor) representation of time harmonic vectors, we refer to any text on mathematical physics and Section 1.3 of Yariv and Yeh (1984) for some pertinent comments as well as Section 1.4 of Azzam and Bashara (1987). For an in-depth treatment of polarized light including the Poincar´e sphere and Jones matrices, we refer to Azzam and Bashara (1987) — see also the introductory level optics treatment on pp. 48– 49 and 544–545 of Guenther (1990). We also refer to Kliger et al. (1990) and Chapter 5 of Yariv and Yeh (1984) for further applications in optics as well as Someda and Stegeman (1992) and references therein for advanced waveguide theory applications, and Galtarossa and Menyuk (2005) for a thorough coverage of the now popular subject of polarization mode dispersion (PMD) in optical ﬁbers. Problems 2-1

Derive the rule expressed by Eq. (2.25).

2-2 Show that the sign of the real part of the linear polarization ratio is an invariant under the transformation deﬁned by Eq. (2.26). 2-3 Show that, if the reference frame is rotated by an angle δ on the plane of the polarization ellipse, two Stokes parameters (s3 and s0 ) are invariant, while the other two obey the transformation rule expressed by Eq. (2.44). 2-4 Consider the following vector ﬁeld which corresponds to the magnetic ﬁeld of the so-called TE10 mode of a rectangular waveguide (see Chapter 7) and is deﬁned with respect to rectangular coordinates (x, y, z) for 0 < x < a, 0 < y < b a s H x =Asin(πx/a)exp(−γz),H y =0,H z =A(γπ/a)cos(πx/a)exp(−γz). a) First assume that γ is real. Show that in this case the vector is linearly polarized at any point, and ﬁnd its direction of polarization as a function of the x coordinate. b) Next assume that γ = jβ is imaginary. Calculate the linear polarization ratio and the circular polarization ratio of the vector as a function of x. Find the values of x, if any, for which the vector is linearly polarized, and

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

49

those for which it is circularly polarized. Distinguish between right-hand and left-hand circular polarizations, for a given sign of β. 2-5 Refer to the vector deﬁned in the previous problem, and let (see Chapter 7, Eq. (7.24)): √ γ = jβ = jω μ0 0 1 − (ωc /ω)2 . Discuss the shape and the orientation of the polarization ellipse as a function of ω, over the range ωc < ω < 2ωc , at the points x = a/4 and x = 3a/4. 2-6 Calculate the Stokes parameters, vs. ω, for the vector of the previous problem, assuming that the constant A is real. Study the corresponding curve on the Poincar´e sphere. 2-7 Consider the following vector ﬁeld, E, which corresponds to the electric ﬁeld of the so-called TE11 mode of a circular waveguide (see Chapter 7) and is deﬁned within the circle of radius a centered at the origin, on the z = 0 plane of a cylindrical coordinate frame, (r, ϕ, z) as Er = E0 J1 (1.84 r/a) (1/r) sin ϕ, Eϕ = (1.84/a) E0 J1 (1.84 r/a) cos ϕ, Ez = 0, where E0 is a constant which is in general complex. (For the deﬁnition of the Bessel function J and of its derivative J , see Appendix D). a) Show that this vector is linearly polarized at any point where it is deﬁned. b) Find how its direction of polarization changes along a circle r = constant, for example the circle r = a/2. 2-8 Write in component form, using cylindrical and cartesian coordinates, a vector, E1 , such that the vector E + E1 is circularly polarized at any point, where E is the vector deﬁned in the previous problem. 2-9 Consider the following vector ﬁeld, E, which is deﬁned in all 3-D space with respect to a spherical coordinate system (r, ϑ, ϕ) and corresponds to the electric ﬁeld of the short electric dipole (see Chapter 12): 1 M 1 e−κ r E = cos ϑ rˆ κ + jω0 2π r r2 1 e−κ r κ 1 M + 2 sin ϑ ϑˆ , + κ2 + jω0 4π r r r √ where κ = ω 0 μ0 , and M is a constant. For simplicity, retain only the term of order 1/r 2 in the r-component, and the term of order 1/r in the ϑcomponent. Find the linear polarization ratio and the axes of the polarization ellipse, as a function of ϑ, for ω = 2π 106 rad/s, r = 300 m. 2-10

Write the equation of the ellipse described, in the time domain, by the

© 2006 by Taylor & Francis Group, LLC

50

POLARIZATION

instantaneous vector of unit amplitude and angular frequency ω = 2π 108 rad/s √ which corresponds to the following Stokes parameters: S1 = 0, S2 = 1/ 2, √ S3 = −1/ 2. Find the polar coordinates of this point on the Poincar´e sphere. 2-11 Write the Jones matrix for a half-wave retardation plate, i.e., an element which introduces a phase delay of π radians between its principal axes. First, write the matrix in a reference system which coincides with its principal axes, as was shown in the text for the quarter-wave plate. Then, rewrite it in a reference system rotated by an angle ϑ with respect to the previous one. 2-12 Write the Jones matrix for the following combinations of cascaded elements. a) A quarter-wave plate followed by an ideal polarizer (with an arbitrary rotation between the axes of the two elements). b) An ideal polarizer followed by a quarter-wave plate (again with an arbitrary rotation). Show that, for a generic input state of polarization, the output states of the two combinations are not equal. Discuss the reason for this situation to occur. 2-13 A birefringent plate (see Chapter 6) is characterized, as a function of frequency, by the following Jones matrix: J11 = 1, J22 = exp(jωΔn/c), J12 = J21 = 0, where c = 3 108 m/s (speed of light in free space), Δn = 10−3 (index diﬀerence, or birefringence), = 1 mm (plate thickness). Find all the frequencies (or wavelengths, measured in free space) at which this device acts as a quarter-wave plate. 2-14 Suppose that the device studied in the previous problem can, in practice, be considered as a quarter-wave plate when the phase shift between J 11 and J22 deviates by less than 1% from its nominal value of ±π/2. Calculate the relative bandwidths, Δω/ωi , (i = 1, 2, ...), over which this tolerance is satisﬁed, around the nominal values ωi found in the previous problem. Discuss how these bandwidths behave, as ωi increases.

© 2006 by Taylor & Francis Group, LLC

References Azzam, R.M.A. and Bashara, N.M. (1987) Ellipsometry and Polarized Light. Paperback edition, Elsevier, Amsterdam, Chapters 1–2. Galtarossa, A. and Menyuk, C.R. (Eds.) (2005) Polarization Mode Dispersion. Springer, New York. Guenther, R.D. (1990) Modern Optics. Wiley, New York. Halmos, P.R. (1958) Finite-Dimensional Vector Spaces. Van Nostrand Reinhold, New York. Kliger, D.S., Lewis, J.W. and Randall, C.E. (1990) Polarized Light in Optics and Spectroscopy. San Diego, Academic. Someda, C.G. and Stegeman, G.I. (Eds.) (1992) Anisotropic and Nonlinear Optical Waveguides. Elsevier, Amsterdam. Yariv, A. and Yeh, P. (1984) Optical Waves in Crystals: Propagation and Control of Laser Radiation. Wiley, New York, Sections 1.3 and 5.1.

51 © 2006 by Taylor & Francis Group, LLC

CHAPTER 3

General theorems 3.1

Introduction

Chapter 1 was devoted to the diﬀerential equations which the e.m. ﬁeld satisﬁes, and to some manipulations that allowed us to pass to other equations, either in the same unknowns or in other variables. The small number and the apparent simplicity of those equations could lead one to believe that they could be solved with one procedure, or at most with a few procedures. Still, the number of problems where Maxwell’s equations have exact closed-form solutions is very small. Furthermore, the parameters in Maxwell’s equations can vary by many orders of magnitude. For example, frequencies of technical interest range from zero to over 1014 Hertz; conductivity may vary between zero and 106 Ω−1 ·m−1 , and so forth. Therefore, approximate methods, which are necessary in many instances, are extremely diversiﬁed. As a result, ad hoc discussions are required in a multitude of cases. Indeed, that is what we will do starting with Chapter 4 and proceeding through until the end of the book. However before we begin this, it is advisable to prove some theorems which follow directly from the equations that were either postulated or derived in Chapter 1. Obviously, these theorems must be satisﬁed by all the solutions to be found later on. Their purposes are diﬀerent. Some of them clearly aim at “economy”. For example, the uniqueness theorem of Section 3.3 illustrates the conditions which are necessary in order to obtain one solution to Maxwell’s equations. Others (for example, the equivalence theorem of Section 3.5) are mainly pointers to approximate methods. In no case, however, does the matter discussed in this chapter yield an algorithm for solving Maxwell’s equations. For this, the reader should wait until the following chapters. In some cases, either the statements, or the proofs, depend on whether we consider a generic behavior in the time domain, or time-harmonic complex vectors. This will be emphasized by splitting those theorems into two subsections. Whenever that is not done, it is understood that the proofs in the two domains are substantially the same. 3.2

Poynting’s theorem. Wave impedance

Under this heading we will present, separately for the time and frequency domains, a relation between quantities with the physical dimensions of power. It has an immediate physical interpretation as an energy budget. Furthermore, 53 © 2006 by Taylor & Francis Group, LLC

54

GENERAL THEOREMS

Figure 3.1 The region of interest in the proof of Poynting’s theorem.

the next section will show that the same relation plays another important role, as the lemma of a subsequent theorem. The proofs presented here require a linear anisotropic medium, whose dyadics , μ, σ are constant with respect to time t and angular frequency ω. It is trivial to reduce them to an isotropic medium, whereas it can be a diﬃcult exercise to extend them to cases in which and μ depend on ω. This point will be reconsidered in Chapter 5. 3.2.1

Time domain

In the time domain, the theorem can be stated as follows. Consider a region V , in ordinary 3-D space, bounded by a closed, regular, time-invariant surface, S. Then, the total instantaneous power that the sources in V generate equals the sum of three terms: (i) the rate of change vs. time of the electromagnetic energy stored in V , (ii) the instantaneous dissipated power (i.e., the Joule eﬀect) in V , and (iii) the instantaneous power transmitted through the surface S. All those terms should be understood in the algebraic sense (i.e., they may be positive or negative), except for dissipation, which is non-negative by deﬁnition. In particular, the so-called sources can be either generators, i.e., imposed currents such that, as a time average, they supply a positive power to the ﬁeld, or utilizers, i.e., imposed currents which correspond to a time-averaged negative power, like for instance electrical motors. To prove the time domain version of the theorem, let us write, at any point in the region V (Figure 3.1), Maxwell’s equations; in the case of a linear anisotropic medium, they are straightforward generalizations of Eqs. (1.47) and (1.48): ∂H ∼ , ∇× E = −μ· ∼ ∂t ∂E ∼ + σ· E + J · ∇× H = ∼ ∼0 ∼ ∂t

© 2006 by Taylor & Francis Group, LLC

(3.1) .

(3.2)

POYNTING’S THEOREM. WAVE IMPEDANCE

55

We then take the scalar product (from the left) of the ﬁrst equation by H ∼, and of the second by E . Next we subtract term by term, and use the vector ∼ identity (C.4), which reads: a·∇×b − b·∇×a = −∇·(a×b) . We obtain the following relation, holding at any point in V : ∂ 1 1 ∇·(E · E . (3.3) ∼×H ∼ ) = − ∂t 2 H ∼ ·μ· H ∼ ·· E ∼ +2 E ∼ −E ∼ ·σ· E ∼−J ∼0 ∼ Let us then integrate it over the whole region V . Taking account of the time-invariance of V , applying Gauss’ theorem (A.8) to the integral of the divergence of E ∼×H ∼ , and ﬁnally reordering the terms, we obtain: 1 ∂ − H ·μ· dV = J0 · E H + E ·· E ∼ ∼ ∼ dV ∼ ∼ ∂t V 2 ∼ V + E E n dS , (3.4) ∼ ·σ· E ∼ dV + ∼×H ∼ ·ˆ V

S

where n ˆ denotes the unit vector normal to S, directed outwards from V . Eq. (3.4) proves the above statement as can be rapidly veriﬁed by considering the physical meaning of each term on which we now comment. We ﬁrst consider the term on the left-hand side. The imposed current density J0 can be ∼ regarded as the motion of a charge distribution of density ρ0 and velocity v0, ∼ · J0 is the i.e., J0 = ρ0 v0. Then, if we recall Eq. (1.2), we establish that E ∼ ∼ ∼ ∼ per-unit-time and per-unit-volume work of the Lorentz force, when a current with density J0 circulates. Therefore, when we change its sign, i.e., we con∼ sider − E 0, this represents the per-unit-volume instantaneous power that ∼·J ∼ is transferred (in the algebraic sense) by the source to the e.m. ﬁeld {E ∼, H ∼}, at the expense of an energy source that supports the imposed current density J 0, ∼ but that does not appear in Maxwell’s equations, and that for this reason is often referred to as being of a “non-electrical nature”. Passing to the right-hand side, the reader should already know from the time-independent (or slowly varying) case, that the quadratic forms (H ∼ ∼ ·μ· H E )/2 are, respectively, the instantaneous magnetic and electric )/2 and (E ·· ∼ ∼ energy densities stored in V , measured in joules per cubic meter. The quadratic form E ∼ is the instantaneous dissipated power density measured in watts ∼ ·σ· E per cubic meter. The remaining term, consisting of the surface integral of the vector product E n, is the power transmitted through the surface S, as ∼×H ∼ ·ˆ one can deduce from two observations. The ﬁrst is that inside the region V there are no other possible uses for electromagnetic power, in addition to those already discussed. The second is that the sign of this surface integral changes its sign (while its absolute value remains the same) if we deal with the region that is complementary to V .

© 2006 by Taylor & Francis Group, LLC

56

GENERAL THEOREMS

Figure 3.2 Schematic diagram of a reﬂector antenna.

The vector function: 2

P (W/m ) , (3.5) ∼=E ∼×H ∼ is referred to as the Poynting vector. Its physical dimensions are those of a power density, i.e., watts per square meter. However, it is not legitimate to state that it has a physical meaning point by point. Only its ﬂux through a closed surface is physically identiﬁed as a power. Nonetheless, in the following chapters we will encounter many cases in which we consider the ﬂux of P ∼ through open surfaces, S , and attribute to it the physical meaning of power transmitted through S . Whenever we do so, this will be justiﬁed by the fact that there exists another surface, S , such that (i) S + S is a closed surface, and (ii) the ﬂux of P ∼ through S is zero. A typical example in this sense is a so-called aperture antenna, similar to a parabolic-reﬂector antenna (see Figure 3.2). The total power received by the antenna is usually computed as the Poynting vector ﬂux through its “aperture,” because the contribution through the paraboloidal surface, consisting of a good conductor, is supposed to be negligible. 3.2.2

Frequency domain

In the complex-vector representation of time-harmonic vectors, Eq. (3.4) is replaced by a relation involving complex quantities. We assume that the reader is familiar with the concept of complex power∗ for a lumped-element linear one-port network (Figure 3.3). Poynting’s theorem can then be stated as follows. Consider a region V in ordinary 3-D space, bounded by a regular closed time-invariant surface, S. Then, the total complex power delivered by the sources present in V equals ∗ For a linear one-port electrical device, this quantity (having physical dimensions of power) can be deﬁned as follows: (i) its real part is the active power, i.e., the time-average over a period of the instantaneous power; (ii) its imaginary coeﬃcient is the so-called reactive power, i.e., the product of twice the angular frequency, 2ω, times the amplitude of the sinusoidal component (at an angular frequency 2ω) of the energy stored in the device, with sign + or − depending on whether magnetic or electric energy dominates in the storage.

© 2006 by Taylor & Francis Group, LLC

POYNTING’S THEOREM. WAVE IMPEDANCE

57

Figure 3.3 The concept of complex power in linear circuit theory.

the sum of three terms: (i) the reactive power corresponding to the periodic variation in the energy stored in V , (ii) the active power dissipated in V , and (iii) the complex power transmitted through the surface S. The proof is quite similar to that of the previous case, but the results should not be mixed; let us stress the main diﬀerences. We start from the complex form of Maxwell’s equations, and multiply scalarly from the left the ﬁrst equation by H∗ , and the complex conjugate of the second equation by E. After subtracting, we use the identity (C.4). Then, we integrate over the region V . Finally, we divide all terms by 2, in order to correctly identify their physical meaning. We obtain: ∗ H ·μ·H E··E∗ J∗0 ·E E·σ·E∗ = dV + 2jω − − dV 2 2 4 4 V V V E×H∗ ·ˆ n dS . (3.6) + 2 S It is straightforward to check that Eq. (3.6) corresponds to what we stated. Notice that the ﬁrst term on the right-hand side is real, while the second is imaginary; the left-hand side and the ﬁnal term are, in general, complex. Also note that, as we pointed out previously in Chapter 2, the complex Poynting vector: E×H∗ , (3.7) P= 2 is not related to the time-domain Poynting vector, Eq. (3.5), by the Steinmetz method. Indeed, the one-to-one and onto mapping of complex vectors into time-harmonic vectors commutes only with linear operations in either of the two spaces, while Poynting vectors, being deﬁned via a vector product, are not linear with respect to the e.m. ﬁelds. We conclude this section giving a broad-scope deﬁnition, whose meaning probably looks obscure at this stage; hopefully, however, its application will clarify it in the following chapters. Let u ˆ be a unit-vector in ordinary 3-D real space; for any ﬁeld {E, H}, we deﬁne as the wave impedance in the direction of u ˆ, at any point Q, the dyadic operator η uˆ (Q) deﬁned by the following relationship: u) . (3.8) u ˆ×E×ˆ u = η uˆ (Q)·(H×ˆ We will see in the following chapters that the dyadic η uˆ (P ) reduces to a

© 2006 by Taylor & Francis Group, LLC

58

GENERAL THEOREMS

scalar in most cases of practical interest, where the components of E and H that are transverse with respect to u ˆ are orthogonal to each other. Hence, we may think of it as a generalized “ratio” of the components of E and H that are transverse with respect to u ˆ. Its usefulness consists in allowing us to express the component of the Poynting vector parallel to uˆ (and consequently, the ﬂux of P through a surface) in terms of only one of the two vectors E and H. The role is then the same as that of an impedance (or an admittance) in network theory, quantities which allow us to express the complex power ﬂowing into a port in terms of either only the voltage, or only the current at that port (Figure 3.3). In particular, if E and u ˆ are orthogonal to each other, we have: 1 1 (H∗ ×ˆ u) · η uˆ · (H×ˆ u) = E · (η u∗ˆ )−1 · (ˆ u×E∗ ×ˆ u) , (3.9) 2 2 where, to simplify the notation, the dependence on the point Q is implicitly understood. We leave it as an exercise for the reader to ﬁnd how this expression should be modiﬁed when H and u ˆ are orthogonal. P·u ˆ=

3.3

Uniqueness theorem

Whenever one has to solve diﬀerential equations, it is advisable, before trying to determine the unknowns, to identify conditions that guarantee, when they hold, the uniqueness of the solution. In this section we shall do that with reference to Maxwell’s equations. After that, we will be sure that, if we can solve the problem, there are no risks of other solutions, which might contradict experimentally our analytical (or numerical) results. At the same time, we will identify the minimum set of conditions that guarantee the uniqueness of the solution. However, we will also see later that there are situations, not at all rare, in which Maxwell’s equations can be solved only if one starts with a set of data larger than this minimum. What the uniqueness theorem tells us, is that the “extra” data can not be independent of the essential data. 3.3.1

Time domain

In the time domain, the theorem can be stated as follows. Consider a region V , in ordinary 3-D space, bounded by a closed, regular, time-invariant surface, S. Maxwell’s equations therein have a unique solution {E ∼, H ∼}, if the following quantities are known: • the ﬁeld sources, i.e., the imposed current density, J0, at each point of V , ∼ and at each time t after an initial time t0 , • the initial conditions, i.e., the values of E ∼ and H ∼ at each point of V at the initial time t0 , and • the boundary conditions, i.e., the components of either E ∼ or H ∼ tangential to the surface S, at each point on S, for all times t > t0 . To prove this, we proceed ab absurdo, i.e., we start by assuming that there are two solutions, {E ∼ ,H ∼ } and {E ∼, H ∼ }, and then we show that they must be

© 2006 by Taylor & Francis Group, LLC

UNIQUENESS THEOREM

59

the same in order to satisfy all the assumptions. Maxwell’s equations being linear, the diﬀerence between the two solutions:

e , h , (3.10) ∼=E ∼ −E ∼ ∼=H ∼ −H ∼ is in turn an e.m. ﬁeld. The term J0 being identical for the two solutions im∼ plies that the ﬁeld (3.10) satisﬁes homogeneous Maxwell’s equations, without sources, in V . Then, if we apply Poynting’s theorem in Eq. (3.4) to {e∼, h ∼}, we ﬁnd: 1 ∂ (h ·μ· h 0 = ∼+e ∼ ·· e ∼) dV ∂t V 2 ∼ + e e n dS . (3.11) ∼ ·σ· e ∼ dV + ∼×h ∼ ·ˆ V

S

It was assumed that we know unambiguously either Etan or Htan at each ∼ ∼ point on S; this allows us to determine that the last term of Eq. (3.11) is zero. Integrating the remaining terms with respect to time, from initial time t0 , and exploiting the initial-condition assumption for t = t0 at each point of V , we ﬁnally write: t 1 (h (3.12) ∼ ·μ· h ∼+e ∼ ·· e ∼) dV + t dt V e ∼ ·σ· e ∼ dV = 0 . V 2 0 The ﬁrst term on the left-hand side of Eq. (3.12) is an integral of deﬁnite positive quadratic forms; the second term is an integral of a semideﬁnite positive quadratic form. Therefore, this equation can only be satisﬁed in the case: e h (3.13) ∼= 0 , ∼= 0 , as we intended to show. 3.3.2

Frequency domain

In the complex-vector representation of time-harmonic vectors, the uniqueness theorem has to be modiﬁed, slightly but in a non-trivial manner. The assumptions, under which we will show that Maxwell’s equations have a unique solution in a region V , are the following: • the ﬁeld sources, i.e., the imposed current densities, J0 , are known at each point in V , • the boundary conditions, i.e., the components of either E or H tangential to the surface S, are known at each point on S, and • the medium contained in the region V has everywhere a nonzero conductivity, i.e., it is a lossy medium. The proof of this theorem starts oﬀ as does the previous one, but it diverges from it at the moment of applying Poynting’s theorem to the diﬀerence

© 2006 by Taylor & Francis Group, LLC

60

GENERAL THEOREMS

between the two solutions. Indeed, for complex vectors, Poynting’s theorem reads: h·μ·h∗ e··e∗ e·σ·e∗ dV + 2jω − 0= dV . (3.14) 2 4 4 V V We omitted the surface integral, equal to zero, for the same reason as in the previous case. The real and the imaginary parts of the right-hand side of Eq. (3.14) have to be both equal to zero, so we have: e·σ·e∗ dV = 0 , (3.15) 2 V h·μ·h∗ e··e∗ − dV = 0 . (3.16) 4 4 V It is straightforward to establish that if and only if we have σ = 0, these two relations imply e = h = 0, as required. This explains why it was necessary to add the third assumption. If σ = 0, i.e., in a lossless medium, we see that ﬁelds not sustained by any imposed current can indeed exist in V , simply provided that, as required by Eq. (3.16), the time-averaged energies stored in V in magnetic and electric forms are equal. Such an e.m. ﬁeld is referred to as a resonant ﬁeld, because of its clear analogy with a √ lumped-element LC circuit at resonance, i.e., at an angular frequency (ω = 1/ LC). It should be pointed out that assuming a lossless medium often simpliﬁes Maxwell’s equations remarkably. For this reason, we will use it quite often in the following chapters, but we must reconcile it with the uniqueness theorem. There is no contradiction, in those cases where the solution for σ = 0 is indeed the limit of the solution for the lossy case, as σ → 0. In fact, the third assumption of the uniqueness theorem requires only a non-vanishing loss, without requiring a minimum value for this loss, which can thus be arbitrarily small. Finally, let us mention that the uniqueness theorem can also be proved for a ﬁeld deﬁned throughout 3-D space, provided that the assumption on either Etan or Htan on a closed surface S is replaced by the following postulate (which is known as Sommerfeld’s radiation condition): lim r|E| = lim r|H| = 0 ,

r→∞

r→∞

(3.17)

where r is the distance from an arbitrarily chosen origin 0 (not at inﬁnity) to the point where the ﬁeld is calculated. The proof can be undertaken by the reader by following the guidelines of the previous proofs, applying Poynting’s theorem to a sphere of radius r centered at 0, and then letting r tend to inﬁnity. Comments on how the last proof should be adapted to problems where there are structures of cylindrical shape and inﬁnite length, such as waveguides, are postponed until Chapter 7.

© 2006 by Taylor & Francis Group, LLC

RECIPROCITY THEOREM 3.4

61

Reciprocity theorem

This section is devoted to a theorem that we will only prove for complex vectors representing time-harmonic ﬁelds: we leave as an exercise for the reader the version in the time domain. Before the proof, we need to introduce a deﬁnition, whose scope of applicability includes also circuit theory. Let J 0 be an imposed current density, deﬁned in a region V of ordinary 3-D space. Let {E, H} be an electromagnetic ﬁeld, deﬁned in the same region V , but not necessarily the ﬁeld sustained by the source J0 . We deﬁne the reaction of the ﬁeld {E, H} on the source J0 , in the region V , as the quantity: E·J0 dV . (3.18) R(E, J0 ) = 2 V Eq. (3.18) has the physical dimensions of power. However, even when {E, H} is the ﬁeld sustained by J0 , it is not the complex power supplied by the source J0 ; the latter, in fact, does not involve J0 , but −J∗0 (see Section 3.2). The main use of this new deﬁnition results in the theorem that we will prove shortly, which establishes under what conditions two diﬀerent current densities, J0a and J0b , deﬁned in the same region V of 3-D space, play interchangeable roles. Subscripts a and b will be used also to distinguish the ﬁelds sustained by these two diﬀerent sets of sources. For brevity, let us use terms like “source a, source b, ﬁeld a, ﬁeld b,” without fear of ambiguity. The Lorentz reciprocity theorem states that the reaction of the ﬁeld a on the source b is equal to that of the ﬁeld b on the source a, provided that: • the region V contains either a linear isotropic medium, or a linear anisotropic medium whose constitutive dyadics are all represented by diagonal matrices in a real orthogonal reference frame; • either V is the whole 3-D space, and then the ﬁelds a and b satisfy the Sommerfeld’s radiation conditions in Eq. (3.17), or, alternatively, V is bounded by a closed surface S, on which any e.m. ﬁeld deﬁned in V satisﬁes a boundary condition of the following type, whose physical meaning will become clear throughout the following chapters: n , Etan = Z Htan ×ˆ

(3.19)

where n ˆ is the unit vector normal to S ; Z [ohms] may be a function of spatial coordinates, but is independent of the ﬁeld {E, H}, and is referred to as the wall impedance. For simplicity, we will write in full detail the proof only for an isotropic medium in a bounded region, leaving as exercises the extensions to the other cases just stated. Maxwell’s equations for the source a and the associated ﬁeld read: ∇×Ea

= −jωμ Ha

∇×Ha

= jωc Ea + J0a

,

(3.20) .

(3.21)

We take the scalar product of the ﬁrst equation with Hb , and of the second with Eb ; then, we add term by term the two thus obtained relations. After

© 2006 by Taylor & Francis Group, LLC

62

GENERAL THEOREMS

that, we swap the roles of a and b, and repeat the same operations. Finally, we subtract term by term the two equations obtained in the two cases. The ﬁnal result is: Hb ·∇×Ea −Ea ·∇×Hb +Eb ·∇×Ha −Ha ·∇×Eb = J0a ·Eb −J0b ·Ea , (3.22) while all the other terms cancel out. We now divide Eq. (3.22) by 2 and integrate it over the region V . Then we apply to the two pairs of terms on the left-hand side the vector identity a · ∇×b − b · ∇×a = ∇·(b×a). We obtain: Ea ×Hb Eb ×Ha J0a ·Eb J0b ·Ea − dV − dV . (3.23) ·n ˆ dS = 2 2 2 2 S V V We can now apply the cross-product cyclic permutation rule to both terms in the integral on the left-hand side of Eq. (3.23). After that, we apply Eq. (3.19) to the ﬁeld a in one of the two terms, and to the ﬁeld b in the other term. We conclude that the left-hand side of Eq. (3.23) is zero. This ﬁnal result is exactly what we wanted to prove. As mentioned at the beginning of the section, this theorem means that, when its assumptions hold, one can interchange roles between two distinct sets of sources. A source a, taken alone, gives rise to a certain electric ﬁeld Ea at some points, where, in a diﬀerent situation, there is another source b; then the electric ﬁeld Eb generated by source b at the points that are sites of source a can not be independent of the ﬁeld Ea . It must in fact satisfy the equality between the two integrals on the right-hand side of Eq. (3.23). This theorem expresses, in more physical terms, the same concept as reciprocity in an electrical network consisting of linear, passive and reciprocal two-terminal components (resistors, inductors, mutually coupled inductors, and capacitors). In elementary circuit theory, this concept is often expressed in rather simplistic terms as follows: if we interchange the positions of a generator and of a current meter in a network, the reading of the latter remains unchanged. This familiarity with reciprocity may induce the reader to believe that all the consequences of the theorem match easily with intuition. This is not always true. To show this, let us consider the example of a so-called Y -junction (Figure 3.4). In simple intuitive terms, it is the conﬂuence of three inﬁnitely long waveguide structures, each of which, we will suppose, can propagate only one solution of Maxwell’s equations (apart from an arbitrary complex multiplicative constant). What is easy to reconcile with intuition, is that one can design such a structure so that, when a wave is coming in from arm 1, it is split in equal parts between arms 2 and 3. What, on the other hand, is not obvious at ﬁrst sight, is that, for the same structure, when a wave is coming in from arm 2, then, because of reciprocity, only one half of its power, at most, can be coupled to the output of arm 1. The reader may come back to this example at the end of Section 3.9, and rediscuss it in terms of symmetries. By far and large, the majority of the subjects that are dealt with in this

© 2006 by Taylor & Francis Group, LLC

EQUIVALENCE THEOREM

63

Figure 3.4 A microwave Y -junction.

book satisfy the assumptions of the reciprocity theorem. However, this should not induce the reader to draw the erroneous conclusion that all electromagnetic phenomena of practical interest are reciprocal. Nonreciprocal behavior is present in nature as well as in man-made devices. The most classical example of the ﬁrst class is ionospheric propagation. As for the second one, we may quote two-port and three-port devices that are widely used at microwave and optical frequencies. Isolators are two-port devices which, in their ideal version, transmit a signal without any loss in one sense, and introduce an inﬁnite attenuation in the opposite sense. Circulators are three-port devices which, again in their ideal version, pass all the power entering one port to the next one, and nothing to the other one. The usefulness of such devices in communications or radar systems is quite evident. From the proof that we gave in this section, it should be easy to grasp that, in order to get nonreciprocal behavior, it is strictly necessary to have a suitable anisotropic medium, and also to exploit it in a convenient way. In the ionosphere, anisotropy is caused by the Earth’s static magnetic ﬁeld. Isolators and circulators exploit either magnetized ferrites, or some types of crystals. We postpone discussion of more details regarding these subjects until Chapter 6. 3.5 3.5.1

Equivalence theorem Preliminaries

The contents of this section are more complicated than the previous ones. The path that we will take might initially look unnecessarily cautious and slow, but eventually it will turn out to be beneﬁcial to the reader. In Section 1.4, a known term was introduced into Maxwell’s equations, in order to model the physical sources of a ﬁeld. This procedure could lead one to believe that, in every problem, one must ﬁrst identify a unique, well-deﬁned function of space coordinates, J0 , that corresponds to the actual physical sources. The aim of this section is to illustrate that indeed the contrary is true: in order to determine the ﬁeld in a given region of 3-D space, multiple choices exist for replacing assigned sets of sources with other equivalent sets. Equivalent means that, if the boundary conditions are left unchanged,

© 2006 by Taylor & Francis Group, LLC

64

GENERAL THEOREMS

Figure 3.5 The region of interest in the proof of the equivalence theorem, and its subdivision as an internal and an external region.

the ﬁeld remains the same as that given by the original sources. The reader may correlate these results with others in circuit theory, in particular, with Th´evenin’s and Norton’s theorems on the equivalent generators (Kuo, 1966). All analogies between ﬁelds and circuits have deep logical roots, as circuits result always from an abstraction process that starts from the ﬁelds. The contents of this section apply only to media that are strictly linear, as all proofs involve superposition, and with losses, arbitrarily small but not strictly zero, as required by the uniqueness theorem of Section 3.3. The phrase “turn oﬀ the generators in a region of space R” will be used to indicate that, starting from a current density not identically zero in R, we pass to another one, which diﬀers from the previous one only because we impose J0 ≡ 0 in R. We will denote with R that region of 3-D space where the e.m. ﬁeld on which we elaborate is deﬁned, and with C the boundary of R. On C, we assume that the boundary conditions are such that, once we specify all the sources J 0 within R, then the ﬁeld is unique everywhere in R. Furthermore, we will assume that all the sources are localized within an internal subset of R, so that (see Figure 3.5) there are an inﬁnite number of regular closed surfaces completely contained inside R such that if S is any of these taken at random, then we have J0 = 0 in the region contained between S and C. We use the symbol Re to denote the portion of R external to S, and Ri to denote the portion internal to S. Of course, R can also be the whole space, in which case C is at inﬁnity and the ﬁeld satisﬁes the Sommerfeld’s radiation conditions in Eq. (3.17). Suppose that we aim only at determining the ﬁeld in the region Re , which does not contain sources. This case is quite frequent in practice. For example, we might be dealing with the far ﬁeld of a transmitting antenna, or with a diﬀraction problem. Then, according to the uniqueness theorem of Section 3.3,

© 2006 by Taylor & Francis Group, LLC

EQUIVALENCE THEOREM

65

Figure 3.6 The equivalence theorem, in the version which makes use of a perfect magnetic conductor.

any set of sources, situated either in Ri or on S, and capable of giving rise to the correct tangential component of H on the external face of S, would generate in Re a ﬁeld {E, H}, equal to that sustained by the true sources J0 . Recalling what we said in Section 1.6, we note that the problem may be reduced to that of being able to guarantee Htan = 0 on the internal face of S. In fact, once that is done, the correct Htan on the external face is supplied by a source consisting of a surface electric current, on S, having a (linear) density: JS = n ˆ×H (A/m) , (3.24) where n ˆ is the unit vector normal to S, oriented towards Re . The internal face of S does not belong to the region Re where we are interested in the ﬁeld. Hence, in order to ensure Htan = 0 on this face, we can take the liberty of modifying the medium in Ri . In nature, a material capable of guaranteeing Htan = 0 on its surface does not exist, not even as a rough approximation. Nevertheless, it is a useful mathematical device to deﬁne such a medium, and call it a perfect magnetic conductor. What we have obtained in this way is schematically illustrated in Figure 3.6. The generators in Ri are turned oﬀ, the current density of Eq. (3.24) is imposed at any point on S, and a “perfect magnetic conductor” has been placed all along the internal face of S. The ﬁeld {E, H} coincides in Re with that of Figure 3.5. In Ri , as the medium has been modiﬁed, the ﬁeld there is no longer deﬁned. Our goal has been, to a large extent, reached, because the physical sources have been replaced with equivalent ones. But the price we paid for this, inventing the ideal magnetic conductor, looks high. Another approach may start

© 2006 by Taylor & Francis Group, LLC

66

GENERAL THEOREMS

Figure 3.7 The equivalence theorem, in the version which makes use of a perfect electric conductor.

from the fact that the tangential magnetic and electric ﬁelds play perfectly interchangeable roles in the uniqueness theorem (see Section 3.3). Therefore, there is another recipe in order to get an e.m. ﬁeld {E, H} in Re which coincides with that of Figure 3.5. Its ingredients are sources which ensure the correct value of Etan on the external face of S. To obtain this, one can use another mathematical device, also without experimental correspondence: the deﬁnition of a magnetic surface current, whose (linear) density at any point of S is: MS = E×ˆ n (V/m) . (3.25) It ensures the correct value of Etan on the external face of S, provided we can impose Etan = 0 on the internal face. This can be done by postulating that in that place there is a perfect electric conductor, i.e., a medium having conductivity σ = ∞ (a subject that will be discussed again in Section 4.11). What we have obtained in this way is schematically illustrated in Figure 3.7. Once again, the ﬁeld {E, H} in Re coincides with that of Figure 3.5, whereas the ﬁeld in Ri is not deﬁned. Advantages and disadvantages are not the same as in the previous case, but comparable in level. 3.5.2

Theorem

None of the above-mentioned procedures is really attractive, mainly because they require modiﬁcation of the medium in the region Ri , so that the ﬁelds therein are no longer deﬁned. Hence the importance of the following statement, that for the time being will be proved for a source-free region. If we turn oﬀ the generators J0 in Ri , let both electric and magnetic surface currents in

© 2006 by Taylor & Francis Group, LLC

EQUIVALENCE THEOREM

67

Figure 3.8 The equivalence theorem, in the version which does not require any perfect conductor.

Eqs. (3.24) and (3.25) circulate on S, and leave the medium unchanged over the entire region R, then: • in the region Re , we have the same ﬁeld {E, H}, as that of the situation we started from, which is summarized in Figure 3.5, and • in the region Ri , the ﬁeld is identically zero (E = H = 0). The proof consists just of observing that the vector ﬁelds Eeq and Heq , deﬁned as: Eeq = E , Heq = H in Re , (3.26) Eeq = 0 , Heq = 0 in Ri , satisfy Maxwell’s equations in the whole region Ri , provided their sources are Eqs. (3.24) and (3.25), and also satisfy the boundary conditions on C that were satisﬁed by {E, H} in the case shown in Figure 3.5. Applying the uniqueness theorem, we conclude that the Eqs. (3.26) are indeed the solution of the problem schematically illustrated in Figure 3.8, as we intended to prove. It should be self-explanatory why one usually refers to the two surface densities, in Eqs. (3.24) and (3.25), taken together, as sources equivalent to J0 for the ﬁeld in region Re . 3.5.3

Absorbing currents

Let us study now what happens if, on the surface S, there are electric and magnetic surface currents with densities Ja and Ma opposite to the current densities JS and MS of Eqs. (3.24) and (3.25), i.e.: n×H , Ja = −ˆ

Ma = −E×ˆ n .

(3.27)

If the boundary conditions on C are invariant with respect to a change in sign of the ﬁeld (i.e., if the boundary conditions are satisﬁed by {−E, −H} when

© 2006 by Taylor & Francis Group, LLC

68

GENERAL THEOREMS

Figure 3.9 The absorbing currents.

they are by {E, H}, as is the case, for example, if they are the Sommerfeld’s radiation conditions in Eq. (3.17)), then the currents in Eq. (3.27), if alone, generate a ﬁeld which is everywhere the opposite of Eq. (3.26). Suppose now that the original sources J0 in Ri and the surface currents in Eq. (3.27) on S are present together. Then, by superposition, we get that the ﬁeld in such circumstances is: Ea = E − Eeq = E

,

Ha = H − Heq = H

in

Ri

,

Ea = E − Eeq = 0

,

Ha = H − Heq = 0

in

Re

.

(3.28)

So, this ﬁeld {Ea , Ha } coincides then with the original one, {E, H}, in Ri , where the sources are located. It is identically zero in the complementary region Re . This entails the name of absorbing currents (when they are looked at from region Re ) for Eqs. (3.27). We can also imagine having gone from the original situation (Figure 3.5) to that described by Eqs. (3.28) (Figure 3.9) turning oﬀ the generators in Re (which were zero), and replacing them by the surface current densities in Eq. (3.27). As we saw just now, in doing this, the ﬁeld in Ri remained unchanged, while that in Re became zero. Note that this result proves precisely that the equivalence theorem is also true for a region (Ri in our case) where there are sources. Another correct way to read these results is in the following terms. The equivalent sources are always given by Eqs. (3.24) and (3.25), provided that one takes the unit vector n ˆ oriented always away from the region where the equivalence theorem gives an identically zero ﬁeld. Consequently, Eqs. (3.24) and (3.25) can be interpreted also as absorbing currents, for the region Ri . This allows us to conclude by underlining the perfect symmetry in the roles that the two regions Ri and Re play in the equivalence theorem. Again making use of superposition, it is straightforward to treat an even

© 2006 by Taylor & Francis Group, LLC

INDUCTION THEOREM

69

more general case, in which one starts from a situation where there are sources both in Re and Ri . We leave this as an exercise for the reader. 3.5.4

Final remark

It has been shown that equivalent electric and magnetic surface currents can be treated on a par with known terms, if Maxwell’s equations are to be solved in a part of the region R where the ﬁeld is deﬁned. Unfortunately, however, these equivalent currents are not true known terms. Their expressions contain E and H on the surface S, quantities which are, in their turn, unknown. However, to use equivalent currents requires knowledge of E and H not in the entire region R, but only on a broadly arbitrary closed surface S. Consequently, the practical usage of the theorem resides in those numerous problems (see in particular Chapter 13), where the ﬁelds E and H on a suitable surface S can be either measured experimentally, or estimated via a satisfactory approximation. One can then start from the corresponding surface currents as working assumption, and elaborate on them. We will reconsider this at the beginning of Chapter 11. 3.6

Induction theorem

In this section we will prove a corollary of the equivalence theorem. It leads to relations that are per se exact. However, their practical use is in general limited to approximate approaches, where the theorem helps one to state a suitable “ansatz”. Let us start from the same assumptions as in Section 3.5 for the medium (linear, with arbitrarily small losses but never lossless). Let R denote the region of real 3-D space wherein the ﬁeld {E, H} is deﬁned. Suppose that on its contour surface, C, boundary conditions are such that the uniqueness theorem of Section 3.3 holds in R. Consider then a regular closed surface, S; it subdivides R into two domains, R1 and R2 (Figure 3.10). There are no imposed currents on it. The subscripts 1 and 2 underline that, as we saw in the previous section, the roles of the regions that were initially called “internal” and “external,” are interchangeable. The induction theorem states that the ﬁeld {E, H} in R can be expressed as the sum of two ﬁelds: E = Ei + E

,

H = H i + H

,

(3.29)

which are characterized by the following properties: • the ﬁeld {Ei , Hi }, known as the incident ﬁeld (or inducing, or imposed ), is diﬀerent from zero only in one of two regions — to illustrate the idea, in R1 — and identically zero in the other (which implies that all its sources are in R1 ); • the other ﬁeld {E , H }, referred to as the induced ﬁeld, is non-zero in all of R. It is sustained by those sources of {E, H} that are not sources of

© 2006 by Taylor & Francis Group, LLC

70

GENERAL THEOREMS

Figure 3.10 The induction theorem: schematic illustration.

{Ei , Hi }, and moreover by electric and magnetic surface currents, localized on S, whose densities are, respectively: ˆ , Ji = H i × n

Mi = n ˆ × Ei

,

(3.30)

where the unit vector normal to S, n ˆ , is directed towards the region where {Ei , Hi } is non-zero. To prove all this, consider, in addition to the actual sources of the total ﬁeld {E, H}, four surface current densities, localized on S. Two of them are given by Eq. (3.30), while the other two are the opposite of Eq. (3.30). Incidentally, note how arbitrary the ﬁeld {Ei , Hi } in Eq. (3.30) is, in doing this operation. The superposition of these surface currents is obviously equal to zero and therefore the e.m. ﬁeld, in this new situation, is identical to the ﬁeld {E, H}. We now subdivide all the currents present into two sets, S0 and S . The linearity of the medium allows us to consider the total ﬁeld {E, H} as the sum of the ﬁelds due separately to S0 and S . In the set S0 , we include the sources (all located in R1 ) that are able to generate the ﬁeld {Ei , Hi } on the surface S, and furthermore the surface currents opposite to those of Eq. (3.30). In the set S , we put Eq. (3.30), as well as all the sources of {E, H} that were not included in the set S0 . To reach our conclusion, it is enough to observe that the opposites to Eq. (3.30) are the absorbing currents of the ﬁeld {Ei , Hi } (i.e., they assure Ei = Hi = 0 in R2 ). Thus, S0 and S respectively sustain ﬁelds {Ei , Hi } and {E , H } that meet all the properties that were listed when we stated the theorem. A diﬀerent point of view regarding the current densities in Eq. (3.30) can help clarify the practical impact of this theorem. Let us use diﬀerent names

© 2006 by Taylor & Francis Group, LLC

DUALITY THEOREM

71

and symbols for the ﬁelds induced in the two regions R1 and R2 , for example: {E , H } = {Er , Hr } ,

“reﬂected ﬁeld” in R1

{E , H } = {Et , Ht } ,

“transmitted ﬁeld” in R2

; .

(3.31)

One of the assumptions was that the initial ﬁeld {E, H} does not have sources located on S. Therefore, its tangential components are continuous across that surface. With our new symbols, this continuity can be expressed as: n ˆ × Ei

= n ˆ × (Et − Er ) ,

Hi × n ˆ = (Ht − Hr ) × n ˆ .

(3.32)

Comparing Eq. (3.32) with Eq. (3.30), we see that the surface currents deﬁned by Eq. (3.30) are those that are required in order to have, in the passage across S, discontinuities in the induced ﬁeld {E , H } equal to the right-hand sides of Eqs. (3.32). As we said at the beginning, this theorem turns out to be really useful in practice only when one is clever in estimating the inducing ﬁeld. This must be a good approximation to the exact solution, and at the same time be simple to handle, in order to make the subsequent calculations simple and rapid. 3.7

Duality theorem

The proof of the equivalence theorem of Section 3.5 required the new concept of magnetic currents, which, as already stated, is a useful mathematical tool, with no experimental signiﬁcance. So far, we have only made use of surface magnetic currents, but nothing prevents us from also considering magnetic currents in a 3-D domain. Their use will be thoroughly justiﬁed much later, in Section 11.5. With this, Maxwell’s equation for complex vectors in a linear, time-invariant, anisotropic (or, as a special case, isotropic) medium, become symmetrical (apart from the signs) not only with respect to the unknowns, but also with respect to the imposed terms: ∇×E = −jω μ·H − M0 ∇×H = jω c ·E + J0

,

(3.33)

,

(3.34)

where the imposed magnetic current density is denoted as M0 . The so-called duality theorem is simply the following observation. If {E, H} is a solution of Eqs. (3.33) and (3.34) in a region R of real 3-D space, then another e.m. ﬁeld, also deﬁned in R, but in a diﬀerent medium and with different sources, is obtained via the following set of formal substitutions (where implicit factors of unitary magnitude are understood, to provide suitable conversion of physical dimensions): E = −H

,

H = E c

© 2006 by Taylor & Francis Group, LLC

=μ

J0 = −M0

, ,

μ = c

, .

M0 = J0

,

(3.35)

72

GENERAL THEOREMS

The ﬁeld {E , H } obtained in this way is referred to as the dual (or the conjugate) of the ﬁeld {E, H}. The same word is used also for sources and media. Obviously, the dual of the dual of a ﬁeld corresponds to the ﬁeld itself. Readers who have always seen Maxwell’s equations in the MKS SI system of units may be very surprised by this theorem, and wonder whether it bears any adherence to physical reality. If they consult a textbook (e.g., Jackson, 1975) where the so-called Gaussian CGS system of units is adopted, and pay attention to the fact that in this system, both and μ in free space are equal to unity, then it becomes rather intuitive that two dual ﬁelds may, at least in some cases, be very similar to each other. For example, there are cases where duality corresponds just to a change from one polarization to the orthogonal polarization. 3.8

TE-TM ﬁeld decomposition theorem

Electromagnetic potentials, in Chapter 1, gave us the feeling that there is a large degree of arbitrariness in the ways to express an e.m. ﬁeld in a linear medium. In particular, we saw that it may be expressed in many ways as the sum of two or more ﬁelds, via “superposition,” and that it is possible to impose some pre-selected requirements on each of these ﬁelds. The equivalence and induction theorems, of Sections 3.6 and 3.7, are two examples of how these degrees of freedom may be exploited. Another example will be supplied by this section, whose results will be used very often in Chapters 7, 9 and 10. We will show that in a simply connected region — for the relevance of this restriction, see again the footnote in Section 1.8 — without sources, containing a homogeneous linear isotropic medium, any e.m. ﬁeld {E, H} can be expressed as the sum of two e.m. ﬁelds (that will be labeled, respectively, with a subscript TE and a subscript TM), which, with respect to a unit vector a ˆ chosen in a completely arbitrary direction, are Transverse Electric (TE) and Transverse Magnetic (TM), namely: {E, H} = {ETE , HTE } + {ETM , HTM } , ETE ·ˆ a=0 ,

HTM ·ˆ a=0 .

(3.36)

Furthermore, we will prove that each of these two ﬁelds can be obtained, via standard vector-calculus operations, from a scalar function of spatial coordinates that satisﬁes the scalar homogeneous Helmholtz equation. Before we start showing this, let us warn the reader that the proof of this theorem is enormously simpliﬁed if one is entitled to make two further assumptions. The ﬁrst assumption is that we are entitled to disregard transverse electromagnetic (TEM) ﬁelds, i.e., solutions of Maxwell’s equations having E·ˆ a = 0 and H·ˆ a = 0. The second assumption is that one can guess the functional dependence of the ﬁeld on the coordinate along the direction of a ˆ, leaving only a complex parameter to be determined later on in this function. However, at this stage, we believe, the reader is totally unprepared to grasp the physical meaning and the relevance of these simplifying assumptions. For

© 2006 by Taylor & Francis Group, LLC

TE-TM FIELD DECOMPOSITION THEOREM

73

this reason, we will present here a very general, but also very demanding proof. If the use that is going to be made of this theorem, by a reader or by an instructor, is just to introduce guided-wave propagation (Chapters 7 and 10), then the rest of this section may be replaced by the simpliﬁed proof given in Section 7.2. The general proof begins observing that, given the assumptions on the sources and on the medium, a scalar electric potential ϕ = 0 is a Lorentz gauge. This implies, via Eq. (1.79), that ∇ · A = 0, and consequently, the magnetic vector potential satisﬁes an equation of the type: A = ∇×W

.

(3.37)

So we introduce another vector W (with arbitrary divergence), which in turn must satisfy the following equation, as one can show by inserting Eq. (3.37) into Eq. (1.80): (3.38) ∇×(∇×∇×W + κ2 W) = 0 . Incidentally, from the following vector identity (which can be proved by the reader as an exercise): ∇×(∇×∇×q) = −∇×(∇2 q) ,

(3.39)

(where q is any vector), it follows that Eq. (3.38) can also be replaced by: ∇×(∇2 W − κ2 W) = 0 .

(3.40)

Now, separating the longitudinal and transverse parts with respect to the direction (say z) of the unit vector a ˆ, we get: ˆ W=a ˆ×W×a ˆ + (W · a ˆ) a ˆ = WTM + L a

.

(3.41)

The divergence of W being arbitrary (while its curl is ﬁxed) implies that the scalar L is only speciﬁed to within the addition of an arbitrary function g, which must satisfy only the restriction ∇×(gˆ a) ≡ ∇g × a ˆ = 0. At the same time, the vector WTM is only speciﬁed to within the addition of a irrotational ˆ. vector W , perpendicular to a Now let us express the total e.m. ﬁeld as the sum of a contribution due ˆ. As for the latter, it may be immedito WTM , and another one due to L a ately shown that it is transverse electric (TE). In fact, from Eqs. (1.66) and (3.37), and from the vector identity (B.6), applied in the present case in which a ˆ = constant, we have: ˆ) = −jω∇L × a ˆ , ETE = −jωATE = −iω∇×(L a

(3.42)

ˆ = 0, as we wanted to show. and this relation requires ETE · a It is rather simple to prove now that L satisﬁes the scalar homogeneous Helmholtz equation: (3.43) ∇2 L − κ2 L = 0 . Actually, Eq. (3.43) does not follow necessarily from Eq. (3.40), but, as we said before, L contains an arbitrary additive function g, and this is enough for us to impose that the function L satisfy Eq. (3.43).

© 2006 by Taylor & Francis Group, LLC

74

GENERAL THEOREMS

At this point, we are able to prove that the ﬁeld contributed by WTM is transverse magnetic (TM). Namely, if we take into account Eq. (3.43), we see that WTM must also obey Eq. (3.40) and thus Eq. (3.38). Exploiting then the above mentioned existence of an arbitrary irrotational vector W , we can impose that WTM obeys the following equation: ∇×∇×WTM + κ2 WTM = 0 .

(3.44)

Incidentally, the steps from Eq. (3.38) to Eq. (3.44) are formally identical to those from Eq. (1.64) to Eq. (1.68), and thus equivalent to imposing ∇·WTM = 0. From Eqs. (1.60), (3.37) and (3.44), we get: HTM = (1/μ) ∇×ATM = (1/μ) ∇×∇×WTM = −(κ2 /μ) WTM

, (3.45)

a = 0, because WTM is perpendicular to a ˆ by construction. which yields HTM ·ˆ This is what we intended to prove. The last thing that remains to be shown is that the TM ﬁeld can also be obtained from a scalar function T satisfying the homogeneous Helmholtz equation: (3.46) ∇2 T − κ2 T = 0 . To do this, it is suﬃcient to invoke the duality theorem of Section 3.7, and start again from the very beginning of the present theorem, using the electric vector potential F instead of the magnetic vector potential A. The TE and TM ﬁelds interchange their roles, whence the conclusion. Summarizing, with the help of a few steps that follow simply from some vector identities of Appendix B, we obtain the following expressions for the TE ﬁeld: ETE HTE

= −jω∇L × a ˆ , 1 1 ∇×ETE ∂ ∂ = ∇L = ∇L , = a+ a+ −∇2 Lˆ −κ2 Lˆ −jωμ μ ∂z μ ∂z (3.47)

where L is a solution of Eq. (3.43). Dually, for the TM ﬁeld, we have: HTM

= −jω∇T × a ˆ ,

ETM

=

∇×HTM 1 = jωc c

ˆ− ∇2 T a

∂ ∇T ∂z

=

1 c

a− κ2 Lˆ

∂ ∇T , ∂z (3.48)

where T is a solution of Eq. (3.46). Comparison between the ﬁrst of Eqs. (3.48) and Eq. (3.45) illustrates how the scalar function T is connected with the vector WTM that was used in the ﬁrst part of the present proof, namely: WTM =

© 2006 by Taylor & Francis Group, LLC

1 ∇T × a ˆ . jωc

(3.49)

SPATIAL SYMMETRIES. REFLECTION OPERATORS

75

This relation also reveals where the residual liberty was hidden that was exploited when we imposed that T obeys Eq. (3.46). Indeed, Eq. (3.49) shows that the z-dependence of the function T is completely irrelevant for calculating the ﬁeld of Eq. (3.45), because the longitudinal component of ∇T disappears, when we take the cross product with the unit vector a ˆ. Hence, there was an arbitrary component of ∇T , up to when T was forced to satisfy Eq. (3.46). 3.9

Spatial symmetries. Reﬂection operators

Geometrical symmetries of the region where an electromagnetic ﬁeld is deﬁned, if properly exploited, can make its determination signiﬁcantly simpler. This statement has been known for a long time, but it became particularly attractive in recent times, when interactive numerical methods became widely used. Indeed, symmetries may permit great reductions in memory allocation and in processing time. The simplest case, as well as the most exploited one, is a symmetry plane. Other cases of considerable interest are symmetries with respect to an axis, and with respect to a point. To approach all these cases in a uniﬁed way, let us take as a starting point Maxwell’s equations written by components in a rectangular coordinate frame (x1 , x2 , x3 ). Our aim is to ﬁnd transformation rules which, when applied to one of their solutions, provide another solution, whenever a symmetry of one of those types is present. We will deal with media that are isotropic, but not necessarily homogeneous. Extensions to some classes of anisotropic media can be found in the literature, but the reader is advised to look for them only after studying Chapter 6. ˆi , Maxwell’s equations, in the generalExpanding the vectors as E = Σi Ei a ized form which also includes magnetic sources, are written by components, as: ∂E3 ∂E2 − ∂x2 ∂x3

= −jωμ H1 − M01

,

∂E1 ∂E3 − ∂x3 ∂x1

= −jωμ H2 − M02

,

∂E2 ∂E1 − ∂x1 ∂x2

= −jωμ H3 − M03

,

∂H3 ∂H2 − ∂x2 ∂x3

= jωc E1 + J01

,

∂H1 ∂H3 − ∂x3 ∂x1

= jωc E2 + J02

,

∂H2 ∂H1 − ∂x1 ∂x2

= jωc E3 + J03

.

(3.50)

Let P denote the generic point in the region R in which Eqs. (3.50) are

© 2006 by Taylor & Francis Group, LLC

76

GENERAL THEOREMS

satisﬁed. Consider now the following set of substitutions: P ≡ (x1 , x2 , x3 )

−→

P ≡ (x1 , x2 , −x3 )

,

E ≡ (E1 , E2 , E3 )

−→

E ≡ (E1 , E2 , −E3 ) ,

H ≡ (H1 , H2 , H3 )

−→

H ≡ (−H1 , −H2 , H3 ) ,

J0 ≡ (J01 , J02 , J03 )

−→

J0

M0 ≡ (M01 , M02 , M03 )

−→

M0 ≡ (−M01 , −M02 , M03 ) .

(3.51)

≡ (J01 , J02 , −J03 ) ,

One can verify immediately that if Eqs. (3.51) are applied all at once, then Eqs. (3.50) are unaltered. Therefore, if the point P belongs to the region R where Eqs. (3.50) are satisﬁed, then the vectors E (P ), H (P ) satisfy at the point P Maxwell’s equations, with sources {J0 (P ), M0 (P )} also located at the point P . Therefore, if the region R is symmetrical with respect to the plane x3 = 0, and if the boundary conditions are themselves symmetrical with respect to the same plane, Eqs. (3.51) deﬁne a new solution of Maxwell’s equations. We say that this new solution is obtained starting from the initial solution by reﬂection in the plane x3 = 0. It may be convenient to write Eqs. (3.51) in the following compact notation: {E , H } = R3 ∗ {E, H}

,

{J0 , M0 } = R3 ∗ {J0 , M0 } .

(3.52)

The operator represented by the symbol R3 ∗ is referred to as the reﬂection operator in the plane x3 = 0. Provided that the region R and the boundary conditions satisfy the appropriate symmetry conditions, also the quantities: {E, H } = −R3 ∗ {E, H}

,

{J0 , M0 } = −R3 ∗ {J0 , M0 } ,

(3.53)

which diﬀer from Eq. (3.51) only by a change in sign of all components, form a solution of Maxwell’s equations. Moreover, it can be veriﬁed immediately that R3 ∗ R3 ∗ = (−R3 ∗) (−R3 ∗) = I∗, where the last symbol stands for the identity operator. An e.m. ﬁeld which is invariant with respect to the application of the operator R3 ∗, namely, is such that in its entire domain of deﬁnition satisﬁes the following equalities: E1 (x1 , x2 , x3 )

= E1 (x1 , x2 , −x3 ) ,

E2 (x1 , x2 , x3 )

= E2 (x1 , x2 , −x3 )

E3 (x1 , x2 , x3 )

= −E3 (x1 , x2 , −x3 ) ,

H1 (x1 , x2 , x3 )

= −H1 (x1 , x2 , −x3 )

H2 (x1 , x2 , x3 )

= −H2 (x1 , x2 , −x3 ) ,

H3 (x1 , x2 , x3 )

= H3 (x1 , x2 , −x3 )

,

,

(3.54)

is called a ﬁeld of even symmetry (or, simply, an even ﬁeld) with respect to the plane x3 = 0. Analogously, a ﬁeld which obeys the relation {E, H} =

© 2006 by Taylor & Francis Group, LLC

SPATIAL SYMMETRIES. REFLECTION OPERATORS

77

−R3 ∗ {E, H}, i.e., satisﬁes a set of equations that diﬀer from Eqs. (3.54) only by the sign in each right-hand side, is referred to as a ﬁeld of odd symmetry (or, simply, an odd ﬁeld) with respect to the plane x3 = 0. If we let x3 = 0 in Eqs. (3.54), we see immediately that they imply E3 = 0 and H1 = H2 = 0 over the entire symmetry plane. Compare with what was said at the beginning of Section 3.5: the conclusion is that an even ﬁeld remains unaltered if an ideal magnetic conductor is introduced on the symmetry plane. Alternatively, if we let x3 = 0 in the relations that deﬁne an odd ﬁeld, then we establish in the same way that they imply, on the plane x3 = 0, zero tangential components of E, and zero normal component of H (E1 = E2 = 0, H3 = 0). Again comparing with Section 3.5, we conclude that an odd ﬁeld remains unaltered if an ideal electric conductor is introduced on the symmetry plane. A ﬁeld of odd symmetry, i.e., compatible with an ideal electric conductor wall, requires sources satisfying the relation: {J0 , M0 } = −R3 ∗ {J0 , M0 } .

(3.55)

This equation expresses in analytical terms the so-called method of images, which is probably known to the reader in the electrostatic case (Plonsey and Collin, 1961). Note that Eq. (3.55) involves a change in sign for some components (J01 , J02 , M03 ), passing from the “actual” sources to the “image” sources. It will be well-known to the reader that any scalar function of a single variable, deﬁned over an interval symmetrical with respect to the origin, can be expressed as the sum of its even and odd parts. Similarly, any e.m. ﬁeld, provided it is deﬁned in a region symmetrical with respect to the plane x3 = 0, may be split into its (i) even part:

{Ee , He } = (1/2) {E, H} + R3 ∗ {E, H} , (3.56) and its (ii) odd part:

{Eo , Ho } = (1/2) {E, H} − R3 ∗ {E, H}

.

(3.57)

Obviously: {E, H} = {Ee , He } + {Eo , Ho } . This property, if suitably exploited, is the key towards saving memory and computation time in numerical methods (see comments at the beginning of this section). However, since diﬀerent components in Eqs. (3.51) are treated diﬀerently in respect of their signs, Eqs. (3.56) and (3.57) are non-trivial extensions of the scalar case, and must be handled with care by a programmer. Let us go back to the beginning of this subject: it is evident that the coordinate x3 does not play a privileged role in Eqs. (3.50) with respect to the other two coordinates. Therefore, it is straightforward to deﬁne reﬂection operators with respect to the coordinate planes x1 and x2 . They will be denoted by R1 ∗

© 2006 by Taylor & Francis Group, LLC

78

GENERAL THEOREMS Table 3.1 Reﬂection operator multiplication table. I

R1

R2

R3

R1

R2

R3

R0

I R1 R2 R3

I R1 R2 R3

R1 I R3 R2

R2 R3 I R1

R3 R2 R1 I

R1 R0 R3 R2

R2 R3 R0 R1

R3 R2 R1 R0

R0 R1 R2 R3

R1 R2 R3 R0

R1 R2 R3 R0

R0 R3 R2 R1

R3 R0 R1 R2

R2 R1 R0 R3

I R3 R2 R1

R3 I R1 R2

R2 R1 I R3

R1 R2 R3 I

and R2 ∗, respectively. We leave it as an exercise for the reader to derive their explicit forms. All the previous considerations regarding even and odd parts, electric and magnetic walls, image sources, etc., can be repeated for these operators without any substantial change. Next we can deﬁne reﬂection operators with respect to coordinate axes. For example, the reﬂection operator with respect to the axis x3 is deﬁned by the following substitutions: P ≡ (x1 , x2 , x3 )

−→

P ≡ (−x1 , −x2 , x3 ) ,

E ≡ (E1 , E2 , E3 )

−→

E ≡ (−E1 , −E2 , E3 ) ,

H ≡ (H1 , H2 , H3 )

−→

H ≡ (−H1 , −H2 , H3 ) ,

J0 ≡ (J01 , J02 , J03 )

−→

J0 ≡ (−J01 , −J02 , J03 ) ,

M0 ≡ (M01 , M02 , M03 )

−→

M0 ≡ (−M01 , −M02 , M03 ) .

(3.58)

Finally, a reﬂection operator with respect to the origin can be deﬁned, as the following set of substitutions: P ≡ (x1 , x2 , x3 )

−→

P ≡ (−x1 , −x2 , −x3 ) ,

E ≡ (E1 , E2 , E3 )

−→

E ≡ (E1 , E2 , E3 ) ,

H ≡ (H1 , H2 , H3 )

−→

H ≡ (−H1 , −H2 , −H3 ) ,

(3.59)

J0 ≡ (J01 , J02 , J03 )

−→

J ≡ (J01 , J02 , J03 ) ,

M0 ≡ (M01 , M02 , M03 )

−→

M ≡ (−M01 , −M02 , −M03 ) .

Using Ri ∗ and R0 ∗ to denote the reﬂection operators with respect to the iaxis and to the origin, respectively, it can be shown that the seven so-deﬁned operators, plus the identity operator, form a set which is an Abelian group (e.g., Cracknell, 1968). Its elements are linked by the relationships summarized in Table 3.1. A simple and illustrative example of how useful symmetries can be is the fol-

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

79

lowing approach to a so-called directional coupler in waveguides, or striplines, etc., whose operation can be explained intuitively, without calculations. Although more formal deﬁnitions exist, let us consider here as a directional coupler a pair of inﬁnitely long transmission lines or waveguides, with a longitudinal symmetry plane, say x1 = 0 (see Figure 3.11). Because of symmetry, any ﬁeld in this structure can be decomposed into its even and odd part. Suppose now that the two lines or waveguides are “single-mode,” i.e., are designed in such a way, that each of them, if left alone, may support just one solution of Maxwell’s equations, uniquely characterized apart from a complex multiplicative constant. Then, at least as a ﬁrst approximation, the total ﬁeld in the coupler will be a superposition of the even (or in-phase) and odd (or out-of-phase) combinations of these “modes” of the individual waveguides. The even and odd combinations, corresponding to a magnetic and an electric wall in the x1 = 0 plane, respectively, are ﬁelds which satisfy Maxwell’s equations in two diﬀerent structures. Therefore, one should not be surprised if we state that they travel at diﬀerent velocities along the inﬁnitely long direction. Consequently, the total ﬁeld amplitude in each waveguide may evolve periodically, although the total power is conserved. The example shown in Figure 3.11 refers in particular to the case where all the power in one section was in the upper waveguide (for example, this was the “launching condition” at the input of the coupler). It shows that, perhaps in contrast with intuition, all the power may be transferred to the lower waveguide at suitable distances. The reader may try, as a demanding exercise, to explain in terms of symmetries (even and odd ﬁeld combinations) the reciprocity results for a Y -junction that we saw in Section 3.4. 3.10

Further applications and suggested reading

As all solutions of Maxwell’s equations must satisfy the theorems that form the core of this chapter, all the contents of the following chapters may be regarded as a set of further applications of the material in this chapter. In particular, Poynting’s theorem and the uniqueness theorem are quite pervasive, and therefore they may show up in any of the following chapters. The reciprocity theorem, as we already mentioned, will ﬁnd one of its main applications in Chapter 12, when we discuss the relationship between transmitting and receiving properties of a given antenna. Nonreciprocal electromagnetic phenomena, which occur when the assumptions of the reciprocity theorem are violated — namely, in some anisotropic media — will be introduced in Chapter 6. However, a much deeper knowledge of nonreciprocities, in terms of basic physics as well as of applications, is instrumental for those who intend to work in such areas as microwave devices and networks, optical telecommunications and sensors, ionospheric propagation, plasma physics. Many textbooks cover these subjects very well. The basic references can be found at the end of Chapter 6. The equivalence theorem is a fundamental starting point for retarded po-

© 2006 by Taylor & Francis Group, LLC

80

GENERAL THEOREMS

Figure 3.11 The principle of operation of a directional coupler.

tentials (Chapter 11) and for diﬀraction (Chapter 13). The induction theorem can prove very useful in the same context, but unfortunately space constraints oblige us to make its use minimal throughout this book. Interested readers can ﬁnd its applications in advanced books on diﬀraction, scattering, aperture antennas (e.g., Chapter 8 of Stutzman and Thiele, 1981; see also the list of references at the end of Chapter 13). Also, it may be useful to have a thorough understanding of the use of this theorem before getting involved in numerical methods (e.g., Ida, 1995; Press et al., 1992; Stutzman and Thiele, 1981, and references therein). As for the duality theorem, we said at the end of Section 3.7 that it may appear awkward, when presented, as herein, in the MKS system of units, where the orders of magnitudes of dual variables are, in most practical cases, quite diﬀerent. To appreciate why this theorem can be useful in practice, we insist on the point that the reader should deal, at least brieﬂy, with Maxwell’s equations written in the so-called Gaussian CGS system of units, where electric permittivity and magnetic permeability of free space are both equal to unity. These equations can be found, for example, in Jackson (1975). One important consequence was already pointed out in Section 3.7: there are many problems, in optics and in diﬀraction theory, where two dual ﬁelds diﬀer only in polarization. Another way to appreciate the potential uses of this theorem is to think, once again, of numerical methods for solving Maxwell’s equations. All

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

81

quantities handled by a numerical computer are adimensional, and must be normalized so that their numerical values fall within a suitable range. The TE–TM decomposition of an e.m. ﬁeld is the strategic starting point for waveguide theory, both at microwave and optical frequencies. The basic notions regarding guided-wave propagation, which are given in Chapters 7, 9 and 10, provide a very extensive set of examples of such applications. All of them, however, may be derived from the simpliﬁed proof to be given in Chapter 7. As we said in Section 3.8, the demanding proof that was given there broadens the theorem’s scope of validity. This entitles one to tackle more diﬃcult problems, compared to those of our chapters on waveguides. Advanced problems are dealt with in more specialized books, such as Collin (1991), Lewin et al. (1977), and Vassallo (1991). On these occasions, one needs ﬁelds that are TE and TM not with respect to a waveguide axis, but rather with respect to other directions. As for symmetries, several results which will be found in the following chapters are easy to reconcile with physical intuition if they are examined in terms of spatial symmetries of structures and, consequently, of the ﬁelds. The very well known example of the directional coupler, has already been mentioned, and will be brieﬂy reexamined in Chapter 8. Symmetries, as we pointed out in the previous section, are related to basic group theory notions. The advantages of a group-theory-based approach are beautifully outlined, for example, in Chapter 12 of Montgomery, Dicke and Purcell (1948), where many fundamental properties of microwave junctions are derived without getting into any detail of ﬁeld calculations. Similarly, fundamental properties of optical waveguides can be found in a very elegant way via group theory, as shown by Black and Gagnon (1995). For these reasons, the reader is strongly advised to come back and read again Section 3.9 after reading Chapter 10, particularly Section 10.7. Finally, we wish to stress that, from a tutorial viewpoint, it is very helpful for the reader to study the basic theorems of electric circuit theory, in parallel to these theorems on e.m. ﬁelds. Theorems in these two areas must overlap exactly, because circuit theory is another axiomatic theory whose postulates were derived from the same basic physical principles as those of our theory. Just to quote a few examples, uniqueness of voltages and/or currents in an electrical network may run into problems for perfectly lossless resonant circuits: compare with Section 3.3. Reciprocity is found in circuit theory to be intimately related with time invariance. Equivalent generator theorems (also called generalized Th´evenin’s theorems) are conceptually the same as the equivalence theorems of Section 3.5 (Corazza et al., 1969). The classical textbook by Newcomb (1966) remains a highly recommended starting point for those who look for advanced readings in circuit theory. Problems 3-1

For the ﬁeld of the TE10 mode of a rectangular waveguide (see Prob-

© 2006 by Taylor & Francis Group, LLC

82

GENERAL THEOREMS πx a

lems 1-23 and 2-4), E = Ey a ˆ y = E0 a ˆy sin exp(−jβz), H = Hx a ˆ x + Hz a ˆz , βE0 jE0 π πx πx with Hx = − ωμ sin a exp(−jβz), Hz = ωμ a cos a exp(−jβz), deﬁned for 0 < x < a, 0 < y < b, without sources, show that the ﬂux of the Poynting vector a) through the planes x = 0, x = a, y = 0, y = b, is equal to zero; b) through any plane z = constant, is real, and independent of z. Draw your conclusions. 3-2 Show that the results of part a) of the previous problem remain the same when jβ is replaced by γ, with real γ, in the argument of the exponentials and as a factor in the expression of Hx . On the other hand, this change modiﬁes deeply the conclusions of part b). Recalculate and draw suitable conclusions. 3-3 Calculate the wave impedance in the z direction, for the ﬁelds of the previous two problems. What happens to these wave impedances if the sign of β (or γ) is changed ? 3-4 The following ﬁeld represents a purely standing wave in free space, ˆy = E0 sin(2βz) a ˆ y , H = Hx a ˆx = to be studied in Chapter 4: E = Ey a E0 j η0 cos(2βz) a ˆx , where η0 = μ0 /0 . a) Write the Poynting vector. b) Find for which values of L the following equation is satisﬁed: L L 0 μ0 2 |E| dz = |H|2 dz . 4 4 0 0 c) Does this ﬁeld obey the uniqueness theorem ?

3-5 In the time domain, two ﬁelds, {E ∼ ,H ∼ } and {E ∼ ,H ∼ }, have identical sources and satisfy identical boundary conditions, but diﬀerent initial conditions. Show that, if the medium is lossy (i.e., for σ = 0), it must be limt→∞ (E ∼ −E ∼ ) = limt→∞ (H ∼ −H ∼ ) = 0, where t denotes time. 3-6 Consider a region, limited by two parallel conducting planes and extending to inﬁnity in all directions parallel to these planes. How should the Sommerfeld radiation conditions be modiﬁed, to ensure uniqueness of a ﬁeld deﬁned in this region ? (Hint. Introduce polar cylindrical coordinates.) 3-7 The ﬁeld of Problem 3-1 is incident on a source which consists of the following surface electric current density, imposed on the z = 0 plane: ˆy sin kx x [A/m] J0 = J a a

for 0 ≤ x ≤ a , 0 ≤ y ≤ b .

a) Calculate the reaction of the ﬁeld on this source, as a function of kx .

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

83

b) Discuss, in particular, what occurs when kx = nπ/a, with integer n, and suggest a physical interpretation of these results. 3-8 Two sources, equal to that of Problem 3-7, are separated by a distance L such that exp(−jβL) = −1. What is the total reaction of the ﬁeld of Problem 2-1 on this compound source ? Comment on the result. 3-9 In the deﬁnition of reaction, replace the imposed current densities with their complex conjugates. Retain all the other assumptions of the reciprocity theorem. Prove that a “modiﬁed” reciprocity theorem, involving “reactions” modiﬁed in this way, holds only if the medium is strictly lossless. Is this a meaningful, physically acceptable result ? (Hint. Reconsider what was said, about lossless media, after the uniqueness theorem.) 3-10 Consider the ﬁeld of Problem 3-1, deﬁned for −∞ < z < +∞. a) Calculate its equivalent electric and magnetic currents on the plane z = 0, i.e., the current densities on that plane such that the ﬁeld remains unchanged for z > 0, and is identically zero for z < 0. b) Calculate its absorbing currents on the plane z = 0, i.e., the current densities on that plane such that the ﬁeld remains unchanged for z < 0, and is identically zero for z > 0. 3-11 Repeat Problem 3-10, for a ﬁeld which diﬀers from the previous one in that β is replaced by −β. Comment on the results. 3-12 Consider two e.m. ﬁelds. The ﬁrst one is identical to that of Problem 31 for 0 < x < a, 0 < y < b, and identically zero outside that region. The second one has the same analytical expressions, which apply now for −∞ < x < +∞, −∞ < y < +∞. Calculate absorbing currents at x = 0, x = a, y = 0, y = b, which transform the second ﬁeld into the ﬁrst one. Compare the results with those of Problem 123. 3-13 Superimposing the ﬁeld of Problem 3-1 and a ﬁeld, of equal amplitude, which diﬀers from the former one in that β is replaced by −β, we get (assuming ˆy = 2E0 a ˆy sin πx ˆ x + Hz a ˆz , with that E0 is real): E = Ey a a cos(βz), H = Hx a βE0 jE0 π πx πx Hx = 2j ωμ sin a sin(βz), Hz = 2 ωμ a cos a cos(βz), for 0 < x < a, 0 < y < b. Calculate the equivalent currents in a generic plane z = constant, and discuss their behavior as functions of z. 3-14 A rectangular waveguide is deﬁned as an inﬁnitely long region of rectangular cross-section, ﬁlled by a homogeneous lossless dielectric and surrounded by a perfectly conducting wall (see Figure 7.4 in Chapter 7). Imagine

© 2006 by Taylor & Francis Group, LLC

84

GENERAL THEOREMS

a structure which consists of two identical waveguides, one above the other (i.e., 0 ≤ x ≤ a, −∞ < z < +∞ for both; 0 ≤ y ≤ b for waveguide 1, −b ≤ y ≤ 0 for waveguide 2). In the conducting wall between the two waveguides, there is a rectangular aperture S, at 0 ≤ x ≤ a, 0 ≤ z ≤ t. Assume that the “incident” ﬁeld, {E0 , H0 }, is deﬁned in waveguide 1, and coincides with that of Problem 3-1. Find the surface current densities of the induction theorem, Eqs. (3.30), on the aperture S. Next, suppose that between the two waveguides there are two rectangular apertures, each of the same size as the previous one, but with a distance d = π/(2β) between their centers. Calculate the current densities of the induction theorem on both apertures. What are the consequences which one should expect from this result, for the induced ﬁeld in region 2 ? 3-15 Write the dual of the ﬁeld of Problem 3-1, assuming, for simplicity, = 0 , μ = μ0 . 3-16 Consider the following e.m. ﬁeld, deﬁned for r < a in cylindrical coordinates: Er = Ez = 0, Eϕ = E0 J1 (3.83 r/a) exp(−jβz), Hr = E0 (β/ωμ) J1 (3.83 r/a) exp(−jβz), Hϕ = 0, Hz = j 3.83/(aωμ) E0 J0 (3.83 r/a) exp(−jβz), where E0 is a constant, and β is real (the so-called TE01 mode of a circular waveguide). (For the deﬁnition of Bessel functions, see Appendix D). Write explicit expressions for the components of the dual ﬁeld. 3-17

Consider again the ﬁeld of the previous problem.

a) Verify that it is TE with respect to the direction of a ˆz . b) Verify that its dual (see again Problem 3-16) is TM with respect to a ˆz . c) Find the corresponding pre-potential functions L and T , and ﬁnd the relationships which must be satisﬁed by β. (Hint. Exploit proportionality between longitudinal ﬁeld components and pre-potentials.) 3-18

Consider again the ﬁeld of Problem 3-1.

ˆy . a) Decompose it into its TE and TM parts with respect to a ˆ x and to a b) Carry out the same procedure for the ﬁeld given in Problem 3-16, and for its dual. 3-19 Show that the ﬁeld of Problem 3-13 (a mode of a so-called rectangular resonator, see Chapter 9) is: a) odd with respect to the planes x = 0, y = 0, z = 0; b) even with respect to reﬂection in the planes x = a/2, y = b/2, z = d/2, where d = π/β.

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

85

(Hint. Change reference frame, x = x − a/2, etc.) 3-20 Consider again a structure like that shown in Problem 3-14, but without any opening between the two waveguides. Consider a ﬁeld which is identical to that of Problem 3-1 in region 1, and identical to zero in region 2. Find its even and odd parts, with respect to the reﬂection operator in the plane y = 0. 3-21 Introduce a rectangular aperture between regions 1 and 2, as in Problem 3-14. Calculate the current densities of the induction theorem, assuming that the incident ﬁeld {E0 , H0 }, deﬁned now in the whole structure, is: a) the even ﬁeld as found in Problem 3-20; b) the odd ﬁeld as found in Problem 3-20. Comment on these results.

© 2006 by Taylor & Francis Group, LLC

References Black, R.J. and Gagnon, L. (1995) Optical Waveguides Modes: Symmetry, Coupling and Polarization. Elsevier, Amsterdam. Collin, R.E. (1991) Field Theory of Guided Waves. 2nd ed., IEEE Press, Piscataway, New Jersey. Corazza, G.C., Longo, G. and Someda, C.G. (1969) “Generalized Th´evenin’s Theorem for Linear n-Port Networks”. IEEE Trans. on Circuit Theory, 16 (4), 564–566. Cracknell, A.P. (1968) Applied Group Theory. Pergamon, Oxford. Ida, N. (1995) Numerical Modeling for Electromagnetic Non-Destructive Evaluation. Chapman & Hall, London. Jackson, J.D. (1975) Classical Electrodynamics. 2nd ed., Wiley, New York. Kuo, F.F. (1966) Network Analysis and Synthesis. 2nd ed., Wiley, New York, Section 7.2. Lewin, L., Chang, D.C. and Kuester, E.F. (1977) Electromagnetic Waves and Curved Structures. P. Peregrinus, Stevenage, UK. Montgomery, C.G., Dicke, R.H. and Purcell, E.M. (1948) Principles of Microwave Circuits. McGraw-Hill, New York. Newcomb, R.W. (1966) Linear Multiport Synthesis. McGraw-Hill, New York. Plonsey, R. and Collin, R.E. (1961) Principles and Applications of Electromagnetic Fields. McGraw-Hill, New York. Press, W.H., Flannery, B.P., Teukolsky, S.A. and Vetterling, W.T. (1992) Numerical Recipes in C: The Art of Scientiﬁc Computing. 2nd ed., Cambridge University Press, Cambridge, UK. Stutzman, W.L. and Thiele, G.A. (1981) Antenna Theory and Design. Wiley, New York. Vassallo, C. (1991) Optical Waveguide Concepts. Amsterdam, Elsevier.

87 © 2006 by Taylor & Francis Group, LLC

CHAPTER 4

Plane waves in isotropic media 4.1

Separability of variables in the homogeneous Helmholtz equation

The ﬁrst three chapters have provided a general framework. The next six chapters will be devoted to studying electromagnetic waves in regions which contain no sources, i.e., to the fundamentals of propagation. In this view, this section is devoted to some general remarks regarding the solution of the homogeneous Helmholtz equation. The preferential role that we attribute to this equation is related to the attempt made in this book to give a uniﬁed approach, based on electromagnetic potentials, to propagation (Chapters 4–10) and to radiation (Chapters 11–14). Note, though, that propagation phenomena could be studied by solving Maxwell’s equations directly, without using potentials. However, this loss of a uniﬁed approach would not entail any beneﬁt, in terms of less mathematical complexity. The content of this section applies to many of the partial diﬀerential equations of mathematical physics. For this reason, some of the following statements will look very generic, and possibly too vague, to the reader. To grasp their full meaning, we suggest to the reader to come back to this point after reading the rest of the present chapter. One of the most commonly used, and most deeply studied, techniques to solve partial diﬀerential equations, is the so-called method of separation of variables. Essentially, it consists of breaking a partial diﬀerential equation (PDE) into a set of ordinary diﬀerential equations (ODE’s), which can be solved separately from one another, by isolating each independent variable in a separate equation. Examples, presented in this chapter as well as in some of the following ones (Chapters 5, 7, 9, 10), will clarify many details on how this procedure works. However, before we proceed, it is worth brieﬂy discussing the scope of applicability of the technique. When an equation, such as the Helmholtz equation, contains partial derivatives with respect to ordinary coordinates in real space, then the main limitation stems from the fact that the equation is separable only in a few, welldeﬁned systems of orthogonal curvilinear coordinates. The phrase “the equation is separable” means that there is at least one set, S, of solutions obtained by separation of variables, which is dense enough, to allow one to express any possible solution of the PDE as a linear combination of the solutions belonging to S. Another way to express the same concept is to say that S is a complete set of solutions. 89 © 2006 by Taylor & Francis Group, LLC

90

PLANE WAVES IN ISOTROPIC MEDIA

The type of diﬃculties that may arise in practice from this limitation, can be understood as soon as one considers that solutions of any diﬀerential equation must always satisfy suitable boundary conditions. Indeed, it is easy to grasp that formal complications will arise as soon as the boundary is not a coordinate surface (or a set of coordinate surfaces), in one of the reference frames where the equation is separable. Nevertheless, separation of variables is a very widely used method. To a large extent, this is due to the drawbacks of the only method which oﬀers a broad-scope alternative, the so-called Green’s function method. This technique consists essentially of writing the solution of a PDE as a volume (or surface) integral of the ﬁeld sources multiplied by suitable functions, whose meaning will be identiﬁed in Chapter 11. It will be exploited in Chapters 11, 12 and 13, which are mainly devoted to the inhomogeneous Helmholtz equation, i.e., to ﬁelds with sources. We wish to point out that the Green’s function method is applicable also in the homogeneous case, but usually, in this case, it leads to more complication, compared with separation of variables, and this justiﬁes our preference. The relationship between the two methods will be discussed brieﬂy in Chapter 11. 4.2

Solution of the homogeneous Helmholtz equation in Cartesian coordinates

Let us suppose that the whole three-dimensional (3-D) real space is ﬁlled by a linear, isotropic and homogeneous medium, with parameters , μ, σ, and that imposed currents are zero everywhere. This, which will be assumed throughout this chapter, is, strictly speaking, a physically meaningless system. Nevertheless, it may build a model which applies with an adequate accuracy to systems which contain suﬃciently wide source-free homogeneous regions. A quantitative estimate of the expression “suﬃciently wide” will be given soon, through the deﬁnition of a length which characterizes the spatial evolution of e.m. phenomena, the wavelength. We aim at determining the electromagnetic ﬁelds which can exist in these conditions, using the potentials (A, ϕ) that were deﬁned in Section 1.8. Note that, under the present assumptions, ϕ = 0 is a Lorentz gauge. Therefore, the whole problem we want to tackle reduces to solving the homogeneous version of Eq. (1.80), i.e.: ∇2 A − κ2 A = 0 , where:

σ κ2 = −ω 2 μc = −ω 2 μ − j ω

(4.1)

.

(4.2)

Let us introduce a cartesian coordinate system (x1 , x2 , x3 ). In such a reference frame, the components of the Laplacian of a vector (see Appendix B) are the scalar Laplacians of the components of the vector. This enables us to

© 2006 by Taylor & Francis Group, LLC

SOLUTION OF THE HOMOGENEOUS HELMHOLTZ EQUATION

91

write Eq. (4.1) simply by components as: ∇2 Ai ≡

∂ 2 Ai ∂ 2 Ai ∂ 2 Ai + + = κ2 Ai 2 2 ∂x1 ∂x2 ∂x23

(i = 1, 2, 3)

.

(4.3)

As the three scalar equations (4.3), for the three components of A, are independent of each other, and the medium is linear, without loss of generality we can take A parallel to one of the axes, for example: A = Ax ˆ1

.

(4.4)

Separation of variables, anticipated in the previous section, consists of making the ansatz that A(x1 , x2 , x3 ) be written as a product of three functions, each of them depending on just one coordinate: A(x1 , x2 , x3 ) = f1 (x1 ) f2 (x2 ) f3 (x3 )

.

(4.5)

Substituting Eq. (4.5) in Eq. (4.3), and dividing by A (= 0), we get: f1 f f + 2 + 3 = κ2 f1 f2 f3

,

(4.6)

where fj denotes the second derivative of fj . The right-hand side of Eq. (4.6) is a constant. The left-hand side consists of three terms, each of them depending on a diﬀerent coordinate. Therefore, Eq. (4.6) can be satisﬁed if and only if the three terms are separately constant: fi = Si2 fi

,

(4.7)

with the condition:

(4.8) S12 + S22 + S32 = κ2 . 2 The quantities Si are the separation constants for Eq. (4.6). Eq. (4.8) is referred to as the separation condition. As we anticipated in Section 4.1, Eqs. (4.7) are three ordinary, mutually independent diﬀerential equations. In particular, in this case we found three harmonic equations, whose general integral, as well known, can be written in the form: (4.9) fi = Fai e−Si xi + Fbi eSi xi , where Fai , Fbi are arbitrary (and in general, complex) constants. Hence, any solution of the type in Eq. (4.5) could be written as a product of three functions of the type in Eq. (4.9). However we may notice that Eq. (4.8) deals with S i2 , not with Si = ± Si2 . Therefore, there is no loss of generality if we write Eq. (4.5) in the following form: A = A0 e−(S1 x1 +S2 x2 +S3 x3 ) ≡ A0 e−S·r

,

(4.10)

where A0 is an arbitrary constant. We have introduced the position vector ˆi , and deﬁned: r = Σi xi x ˆi (4.11) S = Σ i Si x as the propagation vector, whose components in our cartesian reference frame

© 2006 by Taylor & Francis Group, LLC

92

PLANE WAVES IN ISOTROPIC MEDIA

are Si (i = 1, 2, 3). Due to the constraint in Eq. (4.8) on the separation constants, this vector in Eq. (4.11) has to satisfy: S · S = κ2 = −ω 2 μc

.

(4.12)

Let us stress that, since S is a complex vector, S · S = |S|2 = S · S∗ , i.e., Eq. (4.12) is not the norm of the vector S. Let us introduce, for future use, explicit symbols for the real and the imaginary part of the propagation vector: S=a+jk

,

(4.13)

where a is referred to as the attenuation vector, and k as the phase vector. This notation permits us to rewrite Eq. (4.12) as follows: a2 − k2 + 2j a · k = κ2

.

(4.14)

Implications arising from Eq. (4.14) will be the subject of the next section. We will conclude this section by giving explicit expressions for the ﬁelds E and H that follow from the vector potential in Eq. (4.10). We emphasize that these ﬁelds may be derived directly, solving the Helmholtz equation for E or H by the method of separation variables. We leave this to the reader as an exercise. From Eq. (1.66), with ϕ = 0, it follows that: E = −jω A = E0 e−S·r

,

(4.15)

with E0 = −jω A0 x ˆ 1 = E0 x ˆ1 , which means that E0 is essentially an arbitrary constant. From Eq. (1.60), or from Maxwell’s equation for ∇ × E, using the vectorial identity (C.6), it follows that: H =

A0 −S·r 1 ∇ × A0 e−S·r x ∇ e ×x ˆ1 ˆ1 = μ μ

= −

S×E S × E0 −S·r S×A = = e μ jωμ jωμ

.

(4.16)

Note that the penultimate term in Eq. (4.16) gives a relation between E and H which is insensitive to the direction of the vector A. This means that this relationship between the electric and the magnetic ﬁeld vectors is always valid, and is not restricted to the case of a linear polarization, which was the restrictive meaning of Eq. (4.4). The last comment is a good starting point for a short discussion on how one should superimpose three solutions of the kind we just found, each of which gives one component of the vector A in the cartesian reference frame that we introduced before. From now on, unless the opposite is explicitly stated, we will restrict ourselves to superpositions where all three components of the vector A have the same propagation vector, S, i.e., to solutions of Eq. (4.1) of the kind: (4.17) A = A0 e−S·r ,

© 2006 by Taylor & Francis Group, LLC

PLANE WAVES: TERMINOLOGY AND CLASSIFICATION

93

where now A0 is a constant complex vector. The corresponding electromagnetic ﬁeld {E, H} is then: E = E0 e−S·r

,

H=

S × E0 −S·r e jωμ

,

(4.18)

where E0 = −jω A0 . Note that the previous steps provide a nontrivial generalization, compared with Eq. (4.4). Indeed, now we are in a position to change the reference frame in a completely arbitrary way. Should one be obliged to deal with a solution of Eq. (4.1) where the components Ai have diﬀerent propagation vectors (but always satisfying Eq. (4.12)), this ﬁeld could be thought of as a superposition of solutions of the type (4.17). A ﬁeld represented by Eq. (4.18) does not satisfy Sommerfeld’s radiation conditions. Indeed, for some directions of the position vector r (for example, when it is antiparallel to the attenuation vector), Eq. (3.17) is not satisﬁed. Therefore, the waves we just found (to be called plane waves) do not satisfy the uniqueness theorem, not even in the case of a lossy medium. On the other hand, this was an expected result, as we were looking for nontrivial solutions (i.e., solutions which do not coincide with E = H = 0) of Maxwell’s equations. However, in Section 4.7, where we use plane waves as a basis to expand more general ﬁelds, we will show that the solutions we just derived, although individually violating the uniqueness theorem, nevertheless can, as a set, be used in the construction of ﬁelds that satisfy rigorous physical constraints, such as Sommerfeld’s radiation conditions. Finally, inserting the second of Eq. (4.18) in Maxwell’s equation for ∇ × H, we get: S×E×S S×H =− , (4.19) E= −jωc ω 2 μc and, in the same way: S×H×S H=− . (4.20) ω 2 μc By similarities with real vectors, Eqs. (4.19) and (4.20) could make us think that the vectors E, H and S are three mutually orthogonal vectors. However the three vectors are complex, so that more attention has to be paid to orthogonality. The subject is postponed to the end of the next section for a more accurate discussion. 4.3

Plane waves: Terminology and classiﬁcation

Inserting Eq. (4.13), where a and k are real vectors, into Eq. (4.18), it follows that the loci where |E| = constant (and therefore |H| = constant, |A| = constant, also) are given by: a · r = constant

© 2006 by Taylor & Francis Group, LLC

,

(4.21)

94

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.1 Intrinsic propagation, phase and attenuation constants of a homogeneous medium, represented on the complex plane.

i.e., are the planes perpendicular to the attenuation vector a. They are called constant-amplitude or equiamplitude planes of the wave. We will comment on the term “wave” in the next section. In the same manner, it can be proved that loci where Ei = constant (the symbol f indicates the phase of the complex number f ), Ei being one of the components of the E vector in an arbitrary coordinate frame, are given by: k · r = constant

,

(4.22)

i.e., are the planes, called constant-phase or equiphase planes, perpendicular to the phase vector k. The equiphase surfaces of a wave are also called wavefronts. As its wavefronts are planes, the ﬁelds we found in the previous section are called plane waves. Properties of plane waves change drastically, depending on whether a and k are parallel (i.e., constant-phase and constant-amplitude planes coincide), or not. This subject will now be dealt with in detail. As in Section 1.7, let: κ = α + jβ κ2

,

(4.23)

2

be the root of = −ω μc which belongs to the ﬁrst quadrant of the complex plane (see Figure 4.1). Then, certainly we have β > 0, α ≥ 0, with the equality holding only if the medium is lossless. The quantities we denoted with α and β are called, respectively, the intrinsic attenuation constant and the intrinsic phase constant of the homogeneous medium we are dealing with. We can then rewrite the real and the imaginary part of Eq. (4.14) as follows: |a|2 − |k|2 2a·k

= Re{κ2 } = −ω 2 μ , = Im{κ2 } = ωμσ

(4.24)

.

(4.25) 2

Comparing signs on the two sides of Eq. (4.24), it follows that |k| > |a|2 , and therefore |k| > 0. Thus, solutions of Maxwell’s equations of the type in Eq. (4.18) can never have a constant phase throughout the region where they are deﬁned.

© 2006 by Taylor & Francis Group, LLC

PLANE WAVES: TERMINOLOGY AND CLASSIFICATION

95

From Eq. (4.25), we will draw diﬀerent information, depending on whether the medium is lossless (σ = 0) or lossy (σ = 0). Let us discuss separately the two cases. A) Lossless medium. From Eq. (4.25), it follows: a·k=0 ,

(4.26)

a=0 ,

(4.27)

which is satisﬁed either for: or for: a = 0 ,

a⊥k

,

(4.28)

where ⊥ means that a is perpendicular to k. When Eq. (4.27) holds, then we have |E| = constant and |H| = constant over the whole 3-D space, i.e., any plane is a constant-amplitude plane, for this kind of wave. Conventionally, we may decide (mainly by analogy with the following case of a lossy medium) to call constant-amplitude planes those which coincide with the constant-phase planes in Eq. (4.22). Such a wave is called a uniform plane wave. When, on the contrary, Eq. (4.28) holds — we notice, incidentally, that in this case, because of Eq. (4.24), |a| < |k| — then the constant-phase planes are orthogonal to the constant-amplitude planes. Such a wave is called evanescent. Many among the following sections and chapters will underline the physical relevance of this kind of waves, which, at ﬁrst sight, the reader might consider just as mathematical curiosities. B) Lossy medium. From Eq. (4.25) it follows that: a·k>0 .

(4.29)

This tells us ﬁrst, that |a| = 0, and second, that the angle between a and k is less than π/2. In the special case where a and k are parallel, i.e., the constant-phase planes coincide with the constant-amplitude planes, the wave is still called a uniform plane wave. When the angle between a and k is greater than zero, the wave is called dissociated. It is advisable to return to Eqs. (4.19) and (4.20), and discuss them in the light of the results we just found. In order to avoid a substantial amount of needless formal complication, let us restrict ourselves, for the rest of this section, to linearly polarized waves, i.e., waves whose ﬁeld vectors are proportional to real vectors. As a preliminary, note that, as a consequence of what was shown in Section 2.3, the identity B = (B · vˆ) vˆ + vˆ × B × vˆ, which, as well known, holds when B and vˆ belong to the real 3-D space, must be modiﬁed in the following way for the case of complex vectors: v × B) × vˆ∗ B = (B · vˆ∗ ) vˆ + (ˆ

.

(4.30)

Comparing now Eq. (4.30) with Eqs. (4.19) and (4.20), one sees that E, H and S are mutually orthogonal if and only if the wave is linearly polarized and, furthermore, S is parallel to S∗ . It is straightforward to check that this

© 2006 by Taylor & Francis Group, LLC

96

PLANE WAVES IN ISOTROPIC MEDIA

happens if and only if a and k are parallel (including also the particular case where a = 0). Therefore, both in a lossless and in a lossy medium, E, H and S are mutually orthogonal only for linearly polarized uniform plane waves. ∗ On the contrary, this is never possible for evanescent and for dissociated waves. This agrees with geometrical intuition: when a and k are not parallel, the vector propagation “covers” by itself two dimensions in space, so that it is impossible for E and H to be both mutually orthogonal, and perpendicular to the propagation vector as well. Finally it is worth noticing that Eqs. (4.19) and (4.20) always imply the following relations: E·S=0 , H·S=0 , (4.31) and similarly for their complex conjugates: E∗ · S∗ = 0 ,

H∗ · S∗ = 0 .

(4.32)

However, note that, because of what was proved in Section 2.3, these relations we just found are not orthogonality relations for E, H and S. 4.4

Traveling waves. Phase velocity

In order to clarify the basic physical features of plane waves, it is useful to transform, according to the usual Steinmetz procedure, complex ﬁeld vectors into the corresponding time-domain vectors. It is easy to check that, under this viewpoint, the quantities A, E and H given by Eqs. (4.17) and (4.18) have identical properties; therefore, we may concentrate on one of them. For instance, let us consider the electric ﬁeld E and, for the sake of simplicity, let us assume E0 to be real. The corresponding vectorial function of the time-space coordinates is then: −S·r jωt = E0 e−a·r cos(k · r − ωt) . e (4.33) E ∼ = Re E0 e This shows that E varies sinusoidally in time, and, if we leave aside the factor exp(−a · r), in space also. There is a well-deﬁned relationship between time and space dependencies, as the argument of the cosine function mixes the two coordinates. To underline this character, a wave of the kind in Eq. (4.33) is called a traveling wave. The factor exp(−a · r) is, as we said before, an attenuation term, which can be sensed in any direction, except for those perpendicular to a, i.e., those that lie in the constant-amplitude planes. The rate of change of this attenuation is maximum in the same direction as the vector a. This factor is not essential, when discussing the physical features of a traveling wave. Therefore, for the sake of simplicity, from now on it will be omitted. The factor cos(k · r − ωt) describes an undulatory motion. This can be veriﬁed by tracking, as a function of time, those points where the cosine factor ∗ Do not confuse orthogonality between instantaneous time-harmonic vectors, and orthogonality between the corresponding complex vectors. To clarify this point, solve Problem 4.2, at the end of this chapter.

© 2006 by Taylor & Francis Group, LLC

TRAVELING WAVES. PHASE VELOCITY

97

Figure 4.2 Understanding the behavior of the phase velocity.

is constant, i.e., points where the argument of the cosine is constant, that correspond to: d(k · r − ωt) = 0 , (4.34) where d( ) denotes the total diﬀerential of the expression in brackets. Since ω and k are constant, Eq. (4.34) is equivalent to: k · dr − ω dt = 0 .

(4.35)

If we let: dr = u ˆ dr

,

(4.36)

where the unit vector u ˆ has an arbitrary direction, and originates at the generic point P = O + r, then Eq. (4.35) can be written in terms of the following quantity: ω dr = , (4.37) u) ≡ vf (ˆ dt k·u ˆ whose dimensions are those of a velocity. This quantity is called phase velocity in the direction of u ˆ. It is easy to prove that vf (ˆ u) is not the same as a velocity of a point motion in kinematics. Indeed, its dependence on the angle ϑ between the direction of u ˆ, which we choose arbitrarily, and that of k, which is ﬁxed, is: ω . (4.38) u) = vf (ˆ |k| cos ϑ Accordingly, vf (ˆ u) is inversely proportional to cos ϑ, whereas in kinematics the component of a velocity vector v along the direction deﬁned by u ˆ would be directly proportional to cos ϑ. It can be seen with the help of Figure 4.2 that Eq. (4.38) is the velocity at which, to an observer, located in P and looking in the direction of u ˆ, would appear to move the intersection between the generic u) the speed of constant-phase plane and the direction of u ˆ. In no case is vf (ˆ a point mass. It will be shown in the following section that it is not even the velocity at which electromagnetic energy moves. So, it should not come as a surprise if it can be larger than the speed of light, eventually arbitrarily large, as is the case for ϑ → π/2. Normally, if u ˆ is not explicitly speciﬁed, then it is understood that the expression “phase velocity” refers to the value of Eq. (4.37) in the direction

© 2006 by Taylor & Francis Group, LLC

98

PLANE WAVES IN ISOTROPIC MEDIA

of k, i.e., the minimum value of Eq. (4.38), ω vf = . |k|

(4.39)

Now, following the scheme of the previous section, let us ﬁnd what are the values that Eq. (4.39) can take. A) Lossless medium. For a uniform plane wave, since a = 0, Eq. (4.24) gives: √ |k| = ω μ = β = 2π/λ , (4.40) where λ is the wavelength in the given medium, at the angular frequency ω. The modulus of the phase vector equals the intrinsic phase constant β, so that Eq. (4.39) yields: 1 (4.41) vf = √ = c . μ Therefore, for uniform plane waves in lossless media, the phase velocity (which is, in general, a feature of the wave) equals that quantity which in Section 1.7 was called the speed of light, which is, in turn, a feature of the medium. √ For an evanescent plane wave, since a = 0, Eq. (4.24) yields |k| > ω μ = β, where β denotes, as usual, the intrinsic phase constant of the medium. Hence, the equality (4.41) is replaced by the following inequality: vf < c .

(4.42)

Due to Eq. (4.42), an evanescent plane wave in a lossless medium is said to be a slow wave. B) Lossy medium. As in this case a = 0, the inequality in Eq. (4.42) holds, and therefore only slow waves can exist. Anyway, as σ → 0, we get vf → c for uniform plane waves, so that in media with small losses (σ ω), a ﬁrst-order approximation yields vf c. A higher-order approximation can be evaluated by the reader as an exercise. 4.5

Standing waves

The previous sections were devoted to the fundamental physical aspects of a solution of the Helmholtz equation of the kind in Eq. (4.10), assuming always that it was the only ﬁeld that existed in the region of interest. From previous experience with diﬀerential equations, and from elementary experimental observations on motion of waves (such as, for example, vibrating strings, or waves on the surface of a liquid), the reader can expect that as soon as boundary conditions are imposed on an e.m. ﬁeld, they can induce very substantial diﬀerences, with respect to what we saw up to now. At least one of these cases deserves to be investigated immediately. Consider two uniform plane waves which overlap in a lossless unbounded medium. Each wave is of the kind in Eq. (4.18). Their propagation vectors are

© 2006 by Taylor & Francis Group, LLC

STANDING WAVES

99

S = jk and −S = −jk = S∗ , respectively, so: E = E1 e−jk·r + E2 ejk·r

.

(4.43)

For simplicity, assume that E1 and E2 are real. Then, the corresponding time-dependent vector is: jωt E ∼ = Re(E e ) = 2E2 cos(k · r) cos ωt + (E1 − E2 ) cos(k · r − ωt) .

(4.44)

Whereas the second term in the ﬁnal expression of Eq. (4.44) is of the same type as Eq. (4.33), which has already been investigated in Section 4.4, the ﬁrst term is completely diﬀerent, as space and time coordinates appear in two distinct functions, which multiply each other, instead of being combined linearly in the argument of the same function. In particular, the time evolution is in synchronism throughout the whole space: the phase delays from point to point, that characterize Eq. (4.33), have disappeared in Eq. (4.44). What we see now is an oscillation, whose amplitude varies in space, according to a cos(k · r) rule. The points where amplitude reaches its maximum (called crests) are given by: k · r = mπ

(m = 0, ±1, ±2, . . .) ,

(4.45)

whereas the points where the oscillation amplitude is zero (called nodes) are given by: k · r = (2m + 1) π/2 (m = 0, ±1, . . .) . (4.46) Eqs. (4.45) and (4.46) deﬁne two families of planes, perpendicular to k. Comparison with Eq. (4.22) shows that, for each of the two traveling waves that overlap, these planes would have been constant-phase planes. Both families of planes in Eqs. (4.45) and (4.46) are periodic in space, with a period: Δr =

π ˆ λ ˆ k= k |k| 2

,

(4.47)

where kˆ is the unit vector of k and λ is the wavelength, deﬁned in Eq. (4.40). Planes belonging to the two families alternate with a constant spacing Δr = kˆ λ/4. The ﬁrst term in Eq. (4.44) is referred to as a pure standing wave, while the whole Eq. (4.44) is called a partially standing wave, given by superimposing a traveling wave with a pure standing wave. In Chapter 8 we will elaborate further on this point. In particular, we will see that the ratio between amplitudes of the traveling wave and of the pure standing one conveys very signiﬁcant information on boundary conditions where partial reﬂection of the traveling wave occurs. A partially standing wave, Eq. (4.44), is the most general case of superposition of two uniform plane waves propagating in opposite directions. To relax this last constraint, and to consider a superposition of two plane waves (of equal frequency) traveling in any possible direction, the factors exp(∓jk · r)

© 2006 by Taylor & Francis Group, LLC

100

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.3 Two uniform plane waves interfering at an angle: decomposition into a standing-wave factor and a traveling-wave factor.

have to be replaced by exp(−jk1 · r) and exp(−jk2 · r), where |k1 | = |k2 |. We leave it to the reader to develop the remaining steps, and reach the following conclusions: • the ﬁeld, in general, is still a superposition of a standing wave and a traveling wave; • the traveling wave is essentially the same as the one in Eq. (4.44); • the term representing the standing wave consists of a factor which is periodic in space, like: k1 − k 2 ·r , (4.48) cos 2 and has to be multiplied by another factor, like: e−j(k1 +k2 )·r/2

,

(4.49)

so that the wave is a standing one if we look at it in the direction of the vector k1 − k2 , as shown by Eq. (4.48); but at the same time, as shown by Eq. (4.49), it travels along the direction of the vector k1 +k2 , i.e., along the line that bisects the angle between the phase vectors of the two traveling waves (see Figure 4.3). Let us now return to the case described by Eq. (4.44), and, taking only the pure standing wave term, ﬁnd its instantaneous magnetic ﬁeld. It can be calculated by applying Eq. (4.18) to each of the terms that form the electric ﬁeld. The ﬁnal result reads: jωt k × E2 −jk·r jk·r e e − e H = Re j ∼ jωμ = 2

k × E2 sin(k · r) sin ωt ωμ

.

(4.50)

It shows that the nodes of the H ∼ ﬁeld coincide with the crests of the E ∼ ﬁeld, and vice-versa. In the time domain, E is in quadrature with respect to H ∼ ∼. The reader can prove that, for a pure standing wave, the time-averaged electric and magnetic energies, stored in a cylindrical volume of arbitrary

© 2006 by Taylor & Francis Group, LLC

POYNTING VECTOR AND WAVE IMPEDANCE

101

cross-section, and of length λ/2 along the direction of the phase vectors k of the superimposed plane waves, are equal to each other. 4.6

Poynting vector and wave impedance

For a traveling wave of the kind in Eq. (4.18), which satisﬁes Eq. (4.19), the Poynting vector P = E × H∗ /2 can be evaluated by means of the vector identity (C.11). As Eq. (4.13) yields S + S∗ = 2a, one gets: E0 × (S∗ × E∗0 ) −(S+S∗ )·r e 2jωμ |E0 |2 ∗ E0 · S∗ ∗ −2a·r S + E0 e = − 2jωμ 2jωμ |H0 |2 H∗ · S S− 0 H0 e−2a·r . = 2jω 2jω

P = −

(4.51)

So, P is a function of space coordinates only through the factor exp(−2a·r). It is then constant on any constant-amplitude plane in Eq. (4.21). This would imply that the ﬂux of the P vector through any plane in space is inﬁnite. This is physically untenable. Indeed, Eq. (4.18) is perfectly legitimate as a mathematical solution of Maxwell’s equations, but, as it requires an indeﬁnitely extended homogeneous medium, there are limitations to its physical validity. When these constraints are violated, one gets meaningless results, as the previous example shows. This point will be further clariﬁed in Section 4.7 and in Chapter 5. As for the direction of P, it is not easy to recognize it from Eq. (4.51), because, as we said in Section 4.3, E and S, or H and S, are not always orthogonal. It is advisable to analyze separately the peculiar aspects that Eq. (4.51) can assume for the various types of plane waves, following the scheme given in Section 4.3. A) Lossless medium, uniform plane wave. The propagation vector S = jk is orthogonal to both electric and magnetic ﬁelds. Moreover, since Eq. (4.40) holds, it is: |E0 |2 ˆ k . (4.52) P= μ 2 Poynting vector is hence real, so the complex power carried by a uniform plane wave consists only of active power. It is simple to verify that under the same circumstances, at any point in space it is: 1 1 |E|2 = μ |H|2 , (4.53) 4 4 as required by Poynting’s theorem (see Section 3.2) whenever P is real. Let us now evaluate the wave impedance, as deﬁned in Section 3.2. Along ˆ which is also called the direction of the the direction of the phase vector, k,

© 2006 by Taylor & Francis Group, LLC

102

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.4 The geometry of a TM plane wave.

wave normal, and can be seen from Eq. (4.52) to coincide with the direction of P, the wave impedance can be evaluated using Eq. (B.11). It reads: μ ≡η . (4.54) Zkˆ = Thus, the wave impedance in this case coincides with the quantity η = μ/ which is referred to as the intrinsic impedance of the medium, this name coming from the fact that it depends only on the medium parameters, not on the features of a speciﬁc wave propagating in it. It is very useful to memorize that the free-space intrinsic impedance η0 = μ0 /0 equals 120π 377 Ω. As for wave impedances in directions which diﬀer from that of k, it is of primary interest to derive and emphasize two results, to be used in Section 4.8. Both deal with linearly polarized waves, for which the electric and magnetic vectors can be assumed to be real. The ﬁrst one refers to a unit vector u ˆ lying in the plane deﬁned by E and k, and is referred to as the transverse magnetic or TM case. Application of Eq. (3.8) gives then (see Figure 4.4): ZTM = η cos ϑ ,

(4.55)

ˆ is the angle between u where ϑ = (ˆ u, k) ˆ and the wave normal. The second result refers to the case where u ˆ lies in the plane deﬁned by H and k, and is referred to as the transverse electric or TE case. It yields: ZTE = η/ cos ϑ ,

(4.56)

ˆ (see Figure 4.5). where, again, ϑ = (u, ˆ k) B) For an evanescent plane wave in a lossless medium, Eq. (4.51) becomes more complicated. Also for evanescent waves, it will become clear in the following sections that there is a special interest in the case where E is orthogonal to S (a case that, by analogy with the previous discussion, is called TE), and in the case where H is orthogonal to S (TM). Under these circumstances, we ﬁnd, respectively: P =

© 2006 by Taylor & Francis Group, LLC

|E0 |2 −2a·r e (k + ja) 2ωμ

(TE) ,

(4.57)

POYNTING VECTOR AND WAVE IMPEDANCE

103

Figure 4.5 The geometry of a TE plane wave.

|H0 |2 −2a·r e (k − ja) (TM) . (4.58) 2ω In both these cases, the Poynting vector is complex. Its real part is parallel to k, and hence perpendicular to the constant-phase planes. Its imaginary part is parallel to a, and hence lies in the constant-phase planes. There is a change in the sign of reactive power, passing from the TE to the TM case. P =

C) Lossy medium. In both the TE and TM cases, Eq. (4.51) still becomes Eqs. (4.57) and (4.58), respectively. However, now k is no longer orthogonal to a. This implies, on one hand, that along every direction lying in the constantamplitude planes there is a real component of P, but, on the other hand, that the wave normal direction can never lie on the constant-amplitude planes. Therefore, in the k direction there is always a ﬂow of both active and reactive power. D) Standing waves. Let us stress that the Poynting vector for these waves cannot be obtained by superposition of results calculated separately for the traveling waves which build up the standing wave, since the Poynting vector is derived through a nonlinear operation, with respect to the ﬁelds (the outer product of two vectors). We will deal only with the case of a pure standing wave in a lossless medium. We leave it to the reader to consider the other cases as problems, to be solved more conveniently after reading Chapter 8. Let us start from: E = E0 e−jk·r + E0 ejk·r

,

(4.59)

where E0 is orthogonal to S = jk, and H can be derived by means of Eq. (4.50). One then gets: P=

|E0 |2 |E0 |2 j2k·r e sin(2k · r) kˆ − e−j2k·r k = j 2ωμ η

.

(4.60)

Therefore the Poynting vector is parallel, as in the case A), to the k vector of a traveling wave, but it is purely imaginary, and it is varying periodically in space. It goes through zero on the planes: mπ (m = 0, ±1, ±2, . . .) , (4.61) k·r= 2

© 2006 by Taylor & Francis Group, LLC

104

PLANE WAVES IN ISOTROPIC MEDIA

i.e. (see Section 4.5), both at the nodes of E ∼ and at those of H. ∼ Its modulus reaches its maximum on the planes at half way between two successive nodes. It is suggested that the reader complete these results with a careful analysis of reactive power and of the distribution in space of stored electric and magnetic energy, relating this subject with the discussion at the end of Section 4.5. The wave impedance in the wave normal direction is purely reactive, and periodic in space: Zkˆ = jη cotg (k · r) .

(4.62)

This is strictly analogous to situations involving transmission lines, to be dealt with in Chapter 8. The reader should reconsider this section after dealing, in particular, with the input impedance of lossless lines terminated on purely reactive loads. E) Waves interfering at an angle. The last case we wish to discuss brieﬂy is that which was schematically shown in Figure 4.3, and discussed in the previous section for what refers to amplitudes and phases of the two ﬁelds. Once again, we insist on the fact that the Poynting vector for the overall ﬁeld cannot be inferred in an elementary way from those of the two interfering waves. We will brieﬂy summarize the main ﬁndings for the case in which the two interfering waves have equal amplitudes, leaving their derivation to the reader as an exercise. In such a case: • neither the Poynting vector, nor the wave impedance (the latter taken along any direction in space) change, as one moves along the direction of the vector k1 + k2 , because exponential factors like Eq. (4.49) and its complex conjugate disappear in the calculations; • the Poynting vector and the wave impedance vary periodically along the direction of the vector k1 − k2 . The laws for their variations in space are easily derived from Eq. (4.60) and from Eq. (4.62), respectively, by replacing k with (k1 − k2 )/2; • the component of the Poynting vector along the direction of k1 + k2 , and the wave impedance in the same direction, are real at any point in space, indicating that active power is ﬂowing in that direction; • the component of the Poynting vector along the direction of k1 − k2 , and the wave impedance in the same direction, are imaginary at any point in space, indicating that reactive power is ﬂowing in that direction; • in general, the Poynting vector and the wave impedance are polarization dependent. A more demanding exercise is to repeat all these calculations for the case where the two interfering waves are of diﬀerent amplitudes. At this stage, the reader should know how to do it. However, we suggest to wait until after reading Chapter 8, because analogies with transmission lines will be a source of self-conﬁdence, for this type of exercise.

© 2006 by Taylor & Francis Group, LLC

COMPLETENESS OF PLANE WAVES 4.7

105

Completeness of plane waves

In this section, which is linked to the introductory remarks of Section 4.1, we shall investigate the possibility of representing a generic electromagnetic ﬁeld as a linear superposition of plane waves. In order to proceed towards this aim, it is necessary ﬁrst to state explicit requirements that this ﬁeld must satisfy. This sentence is meant to underline that the question of what ﬁelds can be actually expanded into plane waves, does not have a unique answer. The more restrictive the requirements one places on the ﬁelds, the easier to prove the completeness of plane waves. Here, our main task is to show merely that it is possible to make such an expansion. Thus, we will start from fairly relaxed requirements, and consequently give a rather simple proof. More demanding ones can be found in some of the books that are listed among our references. Let us consider ﬁelds deﬁned in a region R which contains no sources, and is ﬁlled by an isotropic and linear medium. Moreover, let us suppose that this medium is stepwise homogeneous, i.e., the parameters c and μ, in R, vary in steps, at most, a ﬁnite number of times, as functions of the space coordinates. As we saw in Chapter 1, this implies that any ﬁeld {E, H} that may exist in R satisﬁes two requirements: a) both E and H satisfy almost everywhere (i.e., excluding at most a set of measure zero) the homogeneous Helmholtz equation, ∇2 a = κ2 a, where κ2 is stepwise constant; b) all the components of E and H in a ﬁxed reference frame are piecewise constant, i.e., have at most a ﬁnite number of discontinuities of the ﬁrst kind. We will now add a third assumption, whose purpose is to restrict our proof to physically meaningful ﬁelds, actually to those ﬁelds which store a ﬁnite energy, and carry a ﬁnite power. We will suppose that there is at least one direction in ordinary 3-D space, labeled z, such that the surface integral: 1 μ |H|2 + |E|2 dn , (4.63) N 4 is ﬁnite when N is any plane orthogonal to z. This is equivalent to saying that the ﬁeld {E, H} is square integrable, over any plane orthogonal to z. This assumption is well matched to situations (such as those treated in Chapters 7 and 9) where propagation takes place essentially in one direction. However, it is not diﬃcult to grasp that even for multidimensional propagation phenomena (such as, for example, waves of spherical symmetry), the physical need that the total energy must be ﬁnite can be translated into a mathematical requirement of the kind (4.63) on an arbitrary plane. We will show now that, under these assumptions, the ﬁeld {E, H} can be expanded as a linear superposition of an inﬁnite number of plane waves. Let us introduce a rectangular reference frame, (x1 , x2 , x3 ), whose x3 axis is parallel to the direction that we called z. Consider the generic Ei component in this system. Due to the assumptions, it is possible to Fourier transform E i with

© 2006 by Taylor & Francis Group, LLC

106

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.6 The loci of plane waves on the complex plane of the propagation constant.

respect to the “transverse” coordinates x1 , x2 : +∞ +∞ 1 dx1 dx2 Ei ej(k1 x1 +k2 x2 ) Ei (k1 , k2 , x3 ) = 2π −∞ −∞

.

(4.64)

Moreover, since Ei satisﬁes the scalar Helmholtz equation: ∇2 Ei ≡ Σl

∂ 2 Ei = κ2 Ei ∂x2l

,

(4.65)

exploiting the correspondence, in the Fourier transformation, between the ∂/∂xl operator and multiplication times jkl , the transformed component Ei satisﬁes: ∂ 2 Ei = −κ2 Ei . (4.66) (k12 + k22 ) Ei − ∂x23 Rearranging the terms, and letting: h2 = κ2 + k12 + k22

,

(4.67)

we see that Eq. (4.66) can be written as an ordinary diﬀerential equation: d2 Ei = h2 Ei dx23

,

(4.68)

whose general integral can be expressed as the linear superposition of exp(∓hx 3 ), if h = constant, and as follows, if h varies (independently of x3 ): Ei = fi (h) e−hx3

,

(4.69)

fi being an arbitrary function of h. From Eq. (4.67), noting that k1 and k2 are real and range from −∞ to +∞, it follows that, if κ2 = −ω 2 μ is real and negative (as is the case in lossless media), then h2 ranges from −ω 2 μ to +∞. Consequently, h spans the domain that consists of the real axis of the complex plane, plus the segment of the √ imaginary axis that lays between ∓jω μ (see Figure 4.6).

© 2006 by Taylor & Francis Group, LLC

COMPLETENESS OF PLANE WAVES

107

The inverse Fourier transformation then gives: +∞ +∞ 1 2 2 2 Ei = dk1 dk2 fi κ + k1 + k2 e−jk1 x1 −jk2 x2 −hx3 , 2π −∞ −∞ (4.70) which expresses Ei as a linear superposition of plane waves, each of which has ˆ1 + jk2 x ˆ2 + hˆ x3 . a propagation vector S = jk1 x Making use of Fourier transforms, we have proved plane wave completeness in the sense of the so-called averaged linear approximation: the right-hand side of Eq. (4.70) converges towards Ei , when the integration interval goes to inﬁnity, almost everywhere, i.e., at most with the exception of a set of discrete points where Ei has step discontinuities. This ends the proof that it is possible to expand E as a linear superposition of plane waves. It remains to be proved that the same expansion also yields the magnetic ﬁeld. It would not be enough to repeat the same procedure, starting from the Helmholtz equation for H. What has to be proved is that H can be expanded into plane waves with the same coeﬃcients as E. This step is necessary in order to prove that the whole ﬁeld {E, H} can be expressed as a linear combination of plane waves. Let us substitute E, by components expanded like Eq. (4.70), into Maxwell’s equation ∇×E = −jωμH. To prove that H is the linear superposition of same plane waves as E, is equivalent to showing that we can interchange the order between integration over k1 and k2 , and partial derivatives with respect to the coordinates. This is typically a question of uniform convergence of the integrals such as Eq. (4.70). It certainly applies where the components of E are continuous, i.e., within each one of the homogeneous media, which build up the piecewise homogeneous medium. So, this proves that the magnetic ﬁeld H can be expanded into plane waves with the same expansion coeﬃcients as E, except, at most, for what happens at the discrete points where the medium is discontinuous. The problem of what happens at these discontinuities is beyond the scope of this book, and the interested reader is referred to more advanced texts (e.g., Morse and Feshbach, 1953). The results of this section deserve some further comments. 1) Eq. (4.70), together with Eq. (4.67), shows that, in order to expand a generic ﬁeld {E, H}, uniform plane waves are not suﬃcient, as they cannot span the whole range of values which are taken by h. In particular, in lossless media (κ2 = −ω 2 μ < 0, and real) we see that it is necessary to take into account the whole set of the evanescent waves whose attenuation vector is along z, in addition to the set of plane waves traveling in all directions in 3-D real space. This point must be stressed, as it diﬀers signiﬁcantly from what the reader may know about Fourier transforms in the time domain, where the conjugate variable (frequency) spans only the real axis. 2) We have demonstrated that a square-integrable ﬁeld (which is a physically meaningful one) can be expanded as a superposition of non-squareintegrable waves. In this way, we have reconciled with physical intuition

© 2006 by Taylor & Francis Group, LLC

108

PLANE WAVES IN ISOTROPIC MEDIA

plane waves which, although physically absurd if taken literally, nevertheless are very powerful and simple mathematical tools. 3) A procedure which makes use of the Fourier transform automatically provides a quantitative criterion, to decide when a single plane wave is acceptable, at least as an approximation. Indeed, the “conjugate variables” relation between x1 , x2 and k1 , k2 allows us to write a well-known inequality, which links the minimum ranges where Ei and Ei have a nonnegligible amplitude: Δki Δxi ≥ 2π

(i = 1, 2)

.

(4.71)

Therefore, in order to express a ﬁeld as a single plane wave (i.e., in order to consider Δki ki ), it is necessary for this ﬁeld to have its amplitude nonnegligible over a region wide enough to satisfy the following condition: Δxi

2π ki

.

(4.72)

The components ki of the k vector depend, of course, on the direction of k itself. Nevertheless, a good term for comparison, which does not depend on the direction of k, can be easily found, as: λ=

2π |k|

.

(4.73)

This quantity was deﬁned in Section 4.4, and is called the wavelength of the plane wave of angular frequency ω. 4.8

Reﬂection and refraction of plane waves

Plane waves lend themselves to the description of how e.m. ﬁelds behave at a sharpe discontinuity in the parameters of the medium. The validity of the results one gets there in terms of single plane waves has to be checked with the criterion that we outlined in the previous section. For the sake of extreme simplicity, let us suppose that the whole 3-D space is divided by a plane surface, which separates two homogeneous media, whose parameters are labeled with a subscript m = 1, 2. Moreover, suppose that the medium 1 is lossless (σ1 = 0). Note, in particular, that the uniqueness theorem and the induction theorem do not hold under these assumptions. However, physical intuition, based on experimental evidence, suggests testing whether continuity of the tangential ﬁeld components on the discontinuity plane can be satisﬁed by superimposing an incident and a reﬂected wave in region 1, and assuming a transmitted wave in region 2. Let us suppose that the incident ﬁeld is a uniform plane wave. Let us also introduce a cartesian reference frame (Figure 4.7), whose plane x1 = 0 is placed at the discontinuity between the two media, and whose x3 axis is chosen so that the phase vector of the incident wave, ki , lies on the x3 = 0 plane. All these assumptions translate into the following formulas: for the incoming

© 2006 by Taylor & Francis Group, LLC

REFLECTION AND REFRACTION OF PLANE WAVES

109

Figure 4.7 Schematic representation of incident, reﬂected and transmitted waves, and the reference frame.

wave: Ei

= E0i e−jki ·r

Hi

=

Er

= E0r e−Sr ·r

Hr

=

,

kˆi × E0i −jki ·r e η1

.

(4.74)

,

(4.75)

.

(4.76)

For the reﬂected wave: ,

Sr × E0r −Sr ·r e jωμ1

and, for the transmitted wave: Et

= E0t e−St ·r

Ht

=

,

St × E0t −St ·r e jωμ2

Let us check whether these waves can satisfy, over the entire x1 = 0 plane, the continuity conditions in Eqs. (1.43) and (1.44), which can be written in the following form, usually referred to as Fresnel’s equations: x ˆ1 × (Ei + Er ) = x ˆ 1 × Et

,

ˆ 1 × Ht x ˆ1 × (Hi + Hr ) = x

.

(4.77)

In particular, these equations must hold at r = 0, i.e., we must have: x ˆ1 × (E0i + E0r )

= x ˆ1 × E0t

,

x ˆ1 × (H0i + H0r )

= x ˆ1 × H0t

.

(4.78)

In Eq. (4.77), each term can be written as the product of a factor like E0i , E0r , E0t , and of an exponential function, according to Eqs. (4.74), (4.75) and (4.76). Then, Eqs. (4.77) and (4.78) look like a system of four linear

© 2006 by Taylor & Francis Group, LLC

110

PLANE WAVES IN ISOTROPIC MEDIA

homogeneous equations in the three unknowns E0i , E0r , E0t . For this system to have nontrivial solutions, it is necessary that at most two of these equations are linearly independent. This happens if the three exponential functions have ˆ2 +x3 x ˆ3 , as in this case Eqs. (4.77) the same argument, for any value of r = x2 x coincide with Eqs. (4.78). This provides the following formulas, which describe the geometrical relationships of the three waves: jki · r = Sr · r = St · r .

(4.79)

For the reﬂected wave, writing Sr = ar + jkr , Eq. (4.79) yields: ar · r = 0 ,

k r · r = ki · r .

(4.80)

This pair of equations cannot be satisﬁed by an evanescent wave. Indeed, ar being orthogonal to kr , from the ﬁrst equation it follows that kr should lie on the plane x1 = 0; but then (remembering also that ki does not belong to √ this plane) the second equation would imply |kr | < |ki | = β1 = ω μ1 1 , thus violating Eq. (4.24). Therefore, we can conclude that the reﬂected wave is a uniform plane wave, with |kr | = |ki | = β1 . The second of Eqs. (4.80), for r = x ˆ3 , gives kr · x ˆ3 = 0, showing that kr lies on the plane x3 = 0, deﬁned by the vector ki and by the normal (x1 axis) to the discontinuity between the two media, and called the incidence plane. Moreover, as |kr | = |ki |, we must have: ϑr = ϑ i

,

(4.81)

where ϑr (ϑi ) is the angle between kr (ki ) and the x1 axis. They are called the reﬂection angle and the incidence angle, respectively. As for the transmitted wave, Eq. (4.79) reads: at · r = 0 ,

k t · r = ki · r .

(4.82)

A priori, there is no reason why this wave cannot be a dissociated one, or an evanescent one if medium 2 is lossless. We will see soon that, in some circumstances, this is indeed the case. In any case, the second of Eqs. (4.82) shows that also kt lies always in the incidence plane. Deﬁning then the transmission angle (or refraction angle) ϑt as the angle between kt and the x1 axis, the second of Eqs. (4.82) can be rewritten as: sin ϑt =

β1 sin ϑi |kt |

.

(4.83)

In order to evaluate |kt |, which is still unknown, it is necessary to distinguish whether medium 2 is lossless or lossy. A) Lossless medium. In Eqs. (4.82), we can have either at = 0, or at perpendicular to the x1 = 0 plane. For at = 0, the transmitted ﬁeld is a uniform plane √ wave, and is called the refracted wave. It has |kt | = β2 = ω μ2 2 . Eq. (4.83) can then be rewritten in the following form, which is known as Snell’s law, or

© 2006 by Taylor & Francis Group, LLC

REFLECTION AND REFRACTION OF PLANE WAVES sometimes as Descartes’ law : μ1 1 β1 sin ϑt = sin ϑi = sin ϑi β2 μ2 2

,

111

(4.84)

where:

μm m (m = 1, 2) , μ0 0 is the refractive index in the m-th medium. Eq. (4.84) then reads as: nm =

n2 sin ϑt = n1 sin ϑi

.

(4.85)

(4.86)

B) Again in a lossless medium, but considering now the possibility that: at · r = 0

(at = 0)

,

(4.87)

we see that this is an acceptable solution only if kt (which, because of what ˆ2 (i.e., was shown in Section 4.3, must be orthogonal to at ) is parallel to x if sin ϑt = 1). Moreover, we saw in Section 4.3 that |kt | > β2 , and therefore Eq. (4.83) tells us that an evanescent wave can exist in medium 2 only if the inequality: β2 n2 sin ϑi > = , (4.88) β1 n1 can be satisﬁed by a real value of the angle ϑi , as required by the initial assumption that the incident ﬁeld is a uniform plane wave. It is obvious that this condition can never be satisﬁed if n2 > n1 . What happens for n2 < n1 is extremely important, both conceptually and practically. It will be the subject of Section 4.11. C) Lossy medium. The transmitted wave has still to satisfy Eqs. (4.82). Hence, it is a dissociated wave, with constant-amplitude planes parallel to x 1 = 0 and constant-phase planes orthogonal to the direction deﬁned by Eq. (4.83). Note that in this case |kt | becomes a function of the angle ϑi . It can be proved (see Problems) that, instead of Eq. (4.83), one gets: 1/2 −1/2 σ22 n1 1 1+ 1+ 2 2 sin ϑt = sin ϑi . (4.89) n2 2 ω 2 cos2 ϑt This formula says that, for σ2 = 0, the ratio sin ϑt / sin ϑi does not depend only on the ratio between the refractive indices, but also on σ2 . Snell’s law has then to be looked at as a limit, which is approached when σ2 → 0. We will come back to this point in Section 4.12, when dealing with the case σ ω 2 . Before we close this section, let us brieﬂy mention what happens in the case (a very frequent one in practice) where there are more than two media, separated by parallel planes (see Figure 4.8). In all media but the last one, one wave is no longer suﬃcient: the number of continuity conditions to be satisﬁed by the ﬁeld in that medium doubled: in the previous case we had one plane of discontinuity in the parameters, in the

© 2006 by Taylor & Francis Group, LLC

112

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.8 A typical “multilayer” structure.

present case we have two such planes. Therefore, two waves are necessary in each medium, except the last one; they must diﬀer in the sign of the component along +x1 of their propagation vectors. Obviously, in general they will diﬀer in amplitude and phase too. When all continuity conditions are written explicitly, then one sees that the only way for this system to have nontrivial solutions is that conditions like Eqs. (4.81) and (4.83) be satisﬁed at each separation plane (the parameters in there must be those of the two media before and after that plane). However, at least as long as all the waves are uniform, to ﬁnd the transmission angle in the n-th medium it is not at all necessary to pass through explicit calculations which involve the other media. Indeed, exploiting the transitive law of equality, one may apply directly Snell’s law, Eq. (4.84), to the ﬁrst and to the n-th medium:

sin ϑn =

β1 sin ϑi = βn

μ1 1 sin ϑi μn n

.

(4.90)

The transitive law of equality makes the case where evanescent waves show up in one or more layers even more interesting than the present one. We will brieﬂy deal with it in Section 4.11.

© 2006 by Taylor & Francis Group, LLC

FRESNEL FORMULAS 4.9

113

Fresnel formulas

The previous section, devoted to the geometry of the reﬂection problem, has led us to conclude that nontrivial solutions of the system in Eq. (4.78) exist. We may then proceed to express the reﬂected and transmitted ﬁelds as function of the incident ﬁeld. These relationships depend on polarization of the incident ﬁeld. To determine them, the simplest way is to exploit some of the results of Chapter 2: we expand the incident wave, whose polarization is completely arbitrary, as the sum of two waves which have two preselected, linearly independent polarizations. Reﬂection and transmission of these two components can be studied separately, and the results can then be superimposed, to reconstruct what happens to the generic incident wave. Since Ei and Hi are, in any case, both transverse to ki , a convenient basis for the decomposition consists of two orthogonal linear polarizations, lying in the plane orthogonal to ki . The simplest results are obtained if we take the following pair: Ei perpendicular to the incidence plane, and Hi perpendicular to the incidence plane. They are referred to, respectively, as TE and TM polarizations.† We will discuss in full detail the case of lossless medium 2, and uniform plane transmitted wave. Suggestions will be given later on to the reader in order to extend the results to the other cases. For TM polarization (see Figure 4.9), Eqs. (4.78) read: E0i cos ϑi + E0r cos ϑr E0i E0r − η1 η1

= E0t cos ϑt E0t η2

=

,

.

(4.91)

Eliminating once E0t , and once E0r , we can ﬁnd the following expressions for the ratios E0r /E0i = ρ and E0t /E0i = τ , which are called reﬂection and transmission coeﬃcients, respectively: η2 cos ϑt − η1 cos ϑi η2 cos ϑt + η1 cos ϑi

ρTM

=

τTM

= (1 − ρTM ) η2 /η1 =

,

2η2 cos ϑt η2 cos ϑt + η1 cos ϑi

.

(4.92)

Similarly, for TE polarization, replacing the electric ﬁeld vectors in Figure 4.9 with the magnetic ones (and keeping track correctly of all the direction † There are other terminologies which are widely used for these two polarizations, like “parallel” and “perpendicular” polarizations, or “horizontal” and “vertical” polarizations. We prefer to avoid these terminologies because there is not a general consensus on them. Radio engineers tend to refer, when using these names, to the direction of the electric ﬁeld, while physicists, in particular in optics, tended, in the past, to refer to that of the magnetic ﬁeld. Recently (see, e.g., Born and Wolf, 1980), some very authoritative textbooks in optics adopted the convention related to the electric ﬁeld.

© 2006 by Taylor & Francis Group, LLC

114

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.9 The ﬁeld vectors of the incident, reﬂected and transmitted waves, in the TM case.

of the vectors that are orthogonal to the plane of the ﬁgure), we ﬁnd: η2 cos ϑi − η1 cos ϑt η2 sec ϑt − η1 sec ϑi = η2 cos ϑi + η1 cos ϑt η2 sec ϑt + η1 sec ϑi

ρTE

=

τTE

= 1 + ρTE =

2η2 sec ϑt η2 sec ϑt + η1 sec ϑi

.

, (4.93)

Eqs. (4.92) and (4.93) are known as Fresnel formulas. If we recall the TM and TE wave impedance formulas, Eqs. (4.55) and (4.56), the reﬂection coeﬃcients can be given a uniﬁed expression: ρ=

Z2 (ˆ x1 ) − Z1 (ˆ x1 ) Z2 (ˆ x1 ) + Z1 (ˆ x1 )

,

(4.94)

where Zm (ˆ x1 ) (m = 1, 2) is the TM or TE wave impedance in the m-th medium, in the direction normal to the plane that separates the two media. It can be proved that Eq. (4.94) holds also in the case of lossy medium 2. The reader who is familiar either with common transmission lines, or with the scattering matrix formalism in circuit theory, will recognize Eq. (4.94) as a very familiar formula. Those who see it for the ﬁrst time will understand the meaning of this comment as soon as they get to Chapter 8. Finally, in next section we will show that, when the transmitted wave in ˆ1 ) with medium 2 is evanescent, it is possible to replace the angle ϑt = (kˆt , x a generalized transmission angle, in general complex, such that Eqs. (4.92) and (4.93) remain formally valid even in this case, although the physics of the transmitted wave changes completely. An interesting question is whether there is any value for the angle of incidence, such that the reﬂection coeﬃcients vanish. To answer it, one has to remember that, under the assumptions which hold in this section, the transmission angle is related to the angle of incidence by Snell’s law, Eq. (4.84). Therefore, the larger angle is in the medium where the index is smaller. Before we proceed further, we must specify whether we deal with dielectric or with

© 2006 by Taylor & Francis Group, LLC

REFLECTION IN MULTILAYER STRUCTURES

115

magnetic media, because permittivity and permeability play equal roles in the refractive index but diﬀerent roles in the wave impedance. Let us deal with the ﬁrst case (i.e., μ1 = μ2 = μ0 ). Then, we see that the equation ρTE = 0 does not have any real solution, being impossible to make η2 cos ϑi = η1 cos ϑt . On the contrary, the equation ρTM = 0 has a solution, which, as the reader may verify by imposing this equation plus Snell’s law, Eq. (4.84), reads: sin ϑi = n2 /(n21 + n22 )1/2

.

(4.95)

This angle is referred to as the Brewster angle or the polarizing angle. The second name points out that when a plane wave in a completely arbitrary state of polarization impinges on a discontinuity at the Brewster angle, the reﬂected wave is linearly polarized in the TE state. This fact is exploited in practice, for example to build a laser which emits linearly polarized light (see Chapter 9). 4.10

Reﬂection in multilayer structures

At the end of Section 4.8, we stressed that if there are more than two media, separated by parallel planes (Figure 4.8), then two waves are necessary in each medium except the last one, to satisfy all the continuity conditions. We saw then that Eqs. (4.81) and (4.83) must be satisﬁed at each separation plane. This means that the geometrical part of the problem, i.e., the laws of reﬂection and refraction in a multilayer structure, are almost trivial extensions of the two semi-inﬁnite media. It is not so concerning the reﬂection and transmission coeﬃcients that we found in the previous section. In this section, we will just give a very schematic description of how one deals with this new problem, and then provide two simple examples. The reason why we postpone most of the applications, is because the same calculations, which would be very heavy now in terms of computational eﬀort, will be enormously simpliﬁed by the analogies that one can establish after reading Chapter 8. The procedure goes as follows. The Fresnel reﬂection coeﬃcients, ρTM and ρTE given by Eqs. (4.92) and (4.93), respectively, can still be used at the interface between the (n − 1)-th and the n-th medium, located (Figure 4.8) at x1 = 0. Then, one tracks the two waves in the (n − 1)-th medium, back to the plane x1 = −n−1 . Note that the up-going and the down-going waves undergo opposite phase shifts, in this operation. Therefore, the total ﬁeld in the (n − 1)-th medium at x1 = −n−1 will not be the same as that at x1 = 0. Still, at this stage we may think of it as a quantity which contains one undetermined complex multiplicative constant, because the relationship between the two waves is known. This means that we are facing again the same situation as at the beginning, namely, a plane of discontinuity where the ﬁeld has two degrees of freedom on the upstream side, and one degree of freedom on the downstream side. If we apply again the continuity conditions, we may now calculate the reﬂection coeﬃcient at the interface between the (n−2)-th and the (n−1)-th medium. We iterate this procedure as many times

© 2006 by Taylor & Francis Group, LLC

116

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.10 Schematic illustration of a Fabry-P´ erot interferometer.

as there are layers, and ﬁnally get the reﬂection coeﬃcients at the input of the structure. As we said before, the procedure that we outlined does often require very cumbersome calculations, especially at oblique incidence, i.e., for an incidence angle ϑi = 0, because Snell’s law has to used at the same time to ﬁnd the various ϑj ’s, j = 2, . . . , n, which in turn aﬀect the reﬂection and transmission coeﬃcients. The following examples are quite simple, as they refer to normal incidence, ϑi = 0, and to three media, i.e., two half-spaces and just one layer in between. Let us study ﬁrst a layer between two semi-inﬁnite media having equal refractive indices, n1 = n3 = n2 . The media are supposed to be non-magnetic dielectrics. The reﬂection coeﬃcient at x1 = 0, obviously independent of polarization, is: n2 − n1 η1 − η2 = . (4.96) ρ= η1 + η2 n2 + n1 We now track back the two waves in medium 2, from x1 = 0 to x1 = −T (Figure 4.10). At those frequencies at which the thickness T of the layer is an integer multiple of the half-wavelength in medium 2, i.e., T = mπ/β 2 , m = 1, 2, . . ., the relative phase of the two waves in medium 2 at x1 = −T is the same as at x1 = 0, exactly as if there were no gap between medium 1 and medium 3. As these are identical, the reﬂection coeﬃcient at x1 = −T , as seen from medium 1, is equal to zero. At these frequencies, all the power

© 2006 by Taylor & Francis Group, LLC

REFLECTION IN MULTILAYER STRUCTURES

117

Figure 4.11 Design of an anti-reﬂection layer.

incident on the ﬁrst discontinuity is transmitted through layer 2, and emerges in medium 3. At all other frequencies, the reﬂection coeﬃcient at x1 = −T is not zero. Its modulus reaches its maximum value at those frequencies at which T equals an odd number of quarters of the wavelength in medium 2, a result that the reader may interpret in rather intuitive terms. The details of how the reﬂection coeﬃcient varies as a function of frequency depend on other details that will be easy to understand after reading Chapters 8 and 9. In any case, however, it is a periodic function of frequency, and its minima are equal to zero. The device schematically shown in Figure 4.10 is known as a Fabry-P´erot interferometer or as an ´etalon. In our second example, Figure 4.11, the two semi-inﬁnite media have different refractive indices, n1 = n3 , and the purpose of the layer between them is to cancel reﬂection at the front of this layer, i.e., at x1 = −T . The media are again supposed to be non-magnetic dielectrics. Physical intuition tells us that one way to reach our goal is to design the thickness and the indices in such a way, that two backward waves, due to reﬂections at x1 = 0 and at x1 = −T , cancel each other. Aiming at this, we assume that the thickness of layer 2 is an odd integer multiple of a quarter of the wavelength in medium 2, i.e., T = (2m + 1)π/(2β2 ), m = 0, 1, . . .. This entails that the phase diﬀerence between the forward and backward traveling waves shifts by an odd integer multiple of π radians, as we move from the upper side of layer 2 to the lower side. The reﬂection coeﬃcient at x1 = 0, again independent of polarization, is: ρ0 =

η3 − η2 n2 − n3 = η3 + η2 n2 + n3

,

(4.97)

and then, from what we just said, it follows that the ratio between the backward and the forward traveling waves in medium 2 at x1 = −T is ρ = −ρ0 = (n3 − n2 )/(n3 + n2 ). What this implies is that, in the ﬁeld continuity conditions, Eqs. (4.78), if written at x1 = −T , the transmitted

© 2006 by Taylor & Francis Group, LLC

118

PLANE WAVES IN ISOTROPIC MEDIA

electric ﬁeld E0t must be replaced by E0t (1 + ρ ), and the transmitted magnetic ﬁeld H0t must be replaced by H0t (1 − ρ ). The ratio between these two ﬁelds is then not equal any more to the intrinsic impedance of medium 2, η 2 , but to: 1 + ρ n3 = η2 . (4.98) ηequiv = η2 1 − ρ n2 If we insert Eq. (4.98) into Eq. (4.94) as the impedance after the discontinuity plane, we ﬁnd a reﬂection coeﬃcient: ρ1 =

ηequiv − η1 n1 − n22 /n3 = ηequiv + η1 n1 + n22 /n3

,

(4.99)

which clearly indicates that there is no reﬂection there (ρ1 = 0) for: n2 = (n1 n3 )1/2

.

(4.100)

For reasons that will become clearer after reading Chapter 8, the anti-reﬂection device which has been illustrated in this example is referred to as a quarterwave impedance-matching transformer. 4.11

Total reﬂection

Let us now study what happens if, for medium 2 being lossless, we have, n2 n2 < n1 , ϑi > ϑc = arcsin , (4.101) n1 where the symbols are the same as in the previous section. The angle ϑ c deﬁned by Eq. (4.101) is referred to as the critical (or limiting) angle. In this case, Snell’s law Eq. (4.83) has no real solutions for ϑt , since it should require sin ϑt > 1. We may conclude that the transmitted wave cannot be a uniform plane wave. On the other hand, there is a solution which, as we mentioned in Section 4.8, consists of an evanescent wave. In this entire chapter, we are working in such conditions that the uniqueness theorem can never be invoked. Yet, the evanescent wave is the solution, as demonstrated experimentally. Its study provides the best possible example of why evanescent waves are important in practice. In the reference frame of Figure 4.7, as we saw in Section 4.8, the evanescent wave must have its attenuation vector at parallel to the x1 axis, and its phase vector kt parallel to the x2 axis. The modulus of kt is then given by Eq. (4.83) with ϑt = π/2, i.e.: |kt | = β1 sin ϑi (> β2 ) , (4.102) while from Eq. (4.24) we get: 1/2 2 1/2 |at | = |kt |2 − β22 = β1 sin2 ϑi − β22

.

(4.103)

Physically, it is evident that at must be oriented in the positive x1 sense; otherwise, the amplitude would grow exponentially as the ﬁeld penetrates into medium 2.

© 2006 by Taylor & Francis Group, LLC

TOTAL REFLECTION

119

Now let us show that, under these circumstances, the wave impedances, both TE and TM, in medium 2, are purely imaginary, i.e., pure reactances. Let us deal ﬁrst with the TE case. Because of the continuity condition in Eq. (1.46) on the normal component of the vector D across the plane x1 = 0, we see that Et has to be parallel to x3 , like Ei and Er . Moreover, since the propagation vector St = at + jkt has a purely real component along x1 , from the second of Eqs. (4.18) it follows that Htan = (at × Et )/jωμ is in quadrature with Et , so x1 ) is imaginary, q.e.d.. The TM case can be analyzed in the same that Z2TE (ˆ way, starting from the continuity condition for the normal component of B, and exploiting now Eq. (4.19). The two procedures yield: Z2TE (ˆ x1 ) =

jωμ |at |

,

Z2TM (ˆ x1 ) =

|at | jω

,

(4.104)

where |at | is given by Eq. (4.103). Given the physical meaning of the wave impedance, these results show that the Poynting vector component along x1 is imaginary, which is equivalent to saying that the evanescent wave does not carry any active power in the x1 direction. This result should be linked with what was said in Section 4.6. We will show now that, in these cases, both ρTE and ρTM , on the plane x1 = 0, are of unit modulus. Indeed, from Eq. (4.78), recalling that E0t and H0t are related by Eq. (4.104), we ﬁnd that, ρTM

=

(|at |/jω2 ) − η1 cos ϑi (|at |/jω2 ) + η1 cos ϑi

,

ρTE

=

(jωμ2 /|at |) − η1 sec ϑi (jωμ2 /|at |) + η1 sec ϑi

,

(4.105)

and hence: |ρTM | ≡ 1

,

|ρTE | ≡ 1

.

(4.106)

This implies that we have, for both the TE and the TM case (we shall therefore omit the subscript): Ei × H∗i Er × H∗r Ei × H∗i ·x ˆ1 ≡ −ρρ∗ ·x ˆ1 = − ·x ˆ1 2 2 2

.

(4.107)

This relation shows that the reﬂected wave carries, in the direction orthogonal to the separation plane, an active power density equal to that of the incoming wave. This yields the name total reﬂection for this phenomenon, which — let us emphasize it again — can occur only for n2 < n1 , i.e., when passing from a medium of “higher optical density” to a medium of “lower optical density” (such as, for example, from glass to air, but not vice-versa). The reader can prove that, when total reﬂection occurs, superimposing the incoming and the reﬂected wave one gets a ﬁeld like that which was described at the end of Section 4.5 and in Problem 3.7. This ﬁeld behaves as a purely standing wave along x1 (its nodes and crests are x1 = constant planes, spaced by λ/(4 cos ϑi )), and as a traveling wave along x2 , with phase velocity

© 2006 by Taylor & Francis Group, LLC

120

PLANE WAVES IN ISOTROPIC MEDIA

vf (ˆ x2 ) = c1 / sin ϑi = ω/(βi sin ϑi ), i.e., synchronous with the phase velocity in medium 2 where the wave normal is parallel to the x2 axis. In the rest of this section, we will show how total reﬂection can be described with a formalism which is very similar to that of Section 4.8, provided one deﬁnes a suitable complex transmission angle, ϑ¯t . Let us start from the postulate that Snell’s law holds also beyond the critical angle, i.e., for ϑi > ϑc . This implies: n1 sin ϑi > 1 . (4.108) sin ϑ¯t = n2 Then, let ϑ¯t = ϕ + jψ, with real ϕ and ψ. Exploiting the identity sin ϑ¯t = −j sinh(j ϑ¯t ), and expanding the sinh function into its real and imaginary parts, as: sin ϑ¯t = cosh ψ sin ϕ − j sinh ψ cos ϕ , (4.109) we see that Eq. (4.108) is satisﬁed when: π sin ϑ¯t = cosh ψ . (4.110) ϑ¯t = + jψ , 2 Moreover, also for ϑ¯t complex, we have cos2 ϑ¯t + sin2 ϑ¯t = 1; thus, sin ϑ¯t larger than 1 implies that cos ϑ¯t is imaginary if, as reconﬁrmed by an expansion of cos ϑ¯t analogous to Eq. (4.109), which, using Eq. (4.110), is: cos ϑ¯t = ±j sinh ψ

.

(4.111)

Now comparing Eq. (4.102) with Eq. (4.107), and using Eq. (4.110), it follows that: |kt | = β2 sin ϑ¯t = β2 cosh ψ . (4.112) From Eq. (4.103), we get: |at | = jβ2 cos ϑ¯t = β2 sinh ψ

.

(4.113)

Eqs. (4.112) and (4.113) allow us to say that ϑ¯t is the angle that a complex vector, of modulus β2 , must form with the x1 axis, in order that its components along x2 and x1 equal kt and jat , respectively. Note that ϑ¯t is not the angle between x1 and the propagation vector St = at + jkt , as, for an evanescent wave, |St | = β2 . Finally, thanks to Eq. (4.113) we see that there is a perfect formal matching between Eq. (4.105) on one side, and Eqs. (4.92), (4.93) on the other side. This means that the complex angle in Eq. (4.110) also provides correct results for the reﬂection coeﬃcients. All the contents of this section, up to now, hold provided that medium 2 extends to inﬁnity along the positive x1 half-axis. Let us now consider a case where the medium with refractive index n2 , smaller than n1 , has a ﬁnite thickness. For the sake of simplicity, let us conﬁne ourselves to the case, shown in Figure 4.12, where n3 = n1 ; more complicated examples can be found among the problems. Continuity conditions like in Eq. (4.77) must be satisﬁed, now, on two parallel planes, x1 = 0 and x1 = d. It is straightforward to check that the minimum

© 2006 by Taylor & Francis Group, LLC

TOTAL REFLECTION

121

Figure 4.12 Illustration of partially frustrated total reﬂection.

number of waves which are needed to solve this problem for an arbitrary value of the incidence angle is ﬁve: an incoming and a reﬂected wave in medium 1, two evanescent waves in medium 2 (with opposite directions, along +x 1 and −x1 , for their attenuation vectors), and a transmitted wave in medium 3. For n3 = n1 , it is easy to show, using Snell’s law and the transitive law of equality, that: (4.114) ϑ3 = ϑ i , so the transmitted wave goes in the same direction as the incident wave in medium 1. In medium 2, the ﬁeld is the sum of the two evanescent waves: ∗

E = E2+ e−S2 ·r + E2− eS2 ·r H =

,

S2 × E2+ −S2 ·r S∗2 × E2− S∗2 ·r e − e jωμ2 jωμ2

.

(4.115)

The reader can prove that, for this ﬁeld, the wave impedance in the x1 direction has a nonzero real part and depends on x1 . This is in deep contrast with a single evanescent waves, whereas it has analogies and diﬀerence with stationary waves arising as superposition of two uniform plane waves, discussed in Section 4.5. Indeed, one uniform traveling wave carries zero reactive power, while superposition of two counter-propagating ones results in a stationary wave with nonzero reactive power. For evanescent waves, each carries zero active power, while superposition of two counter-attenuating ones gives rise to a wave that carries an active power in the direction of the attenuation vector. This provides a physical explanation for the presence of a uniform plane wave, carrying active power, in medium 3, which otherwise would appear as an inconsistency under the viewpoint of conservation of energy. The physical phenomenon of Figure 4.12 is called partially frustrated total reﬂection or, by analogy with quantum mechanics, electromagnetic tunneling. The analogy with quantum mechanics is reinforced by the impossibility of explaining this phenomenon in terms of ray optics, which would always predict

© 2006 by Taylor & Francis Group, LLC

122

PLANE WAVES IN ISOTROPIC MEDIA

Figure 4.13 Plane-wave reﬂection at the surface of a good conductor.

zero power in medium 2 as soon as the critical angle is exceeded, independently of its thickness. In Chapter 10, it will be shown that electromagnetic tunneling plays a fundamental role in guided-wave propagation along a dielectric rod, such as an optical ﬁber. When the geometry is not planar (e.g., when it is cylindrical), there are interesting variations to be added on the same theme, as we will see later. 4.12

Reﬂection on the surface of a good conductor

In technical application of electromagnetic waves, very commonly one is required to ﬁnd a ﬁeld {E, H} deﬁned in a region ﬁlled by a low-loss (or lossless) dielectric medium, bordered by a high-conductivity medium. Usually, knowing the ﬁeld inside the conducting medium is not of primary interest. Thus, it is unnecessary to solve Maxwell’s equations in the conductor, and impose continuity conditions at its borders. Rather, one looks for suitable boundary conditions, to be imposed on the ﬁeld deﬁned in the dielectric medium, at the dielectric-conductor interface. It is our aim, in this section, to ﬁnd such conditions starting from Maxwell’s equations inside the conductor. Let us consider a plane surface between two semi-inﬁnite media, as in Figure 4.13. As we saw in Section 4.8, in medium 1 a uniform plane wave is reﬂected with ϑr = ϑi , independently of the nature of medium 2. What happens in medium 2 will determine the value of the reﬂection coeﬃcient. Suppose that medium 2 is a good conductor, i.e.: σ2 ω2

.

(4.116)

Before we proceed, let us point out that, in some cases, approximations following from Eq. (4.116) may cause errors: we are comparing real and imaginary quantities, and this can result in unacceptable consequences for the phases of complex numbers, although the results are correct for their moduli. As we saw in Section 4.8, the wave in medium 2 is in general dissociated.

© 2006 by Taylor & Francis Group, LLC

REFLECTION ON THE SURFACE OF A GOOD CONDUCTOR

123

Its attenuation vector at is parallel to x1 . Moreover, from Eq. (4.89) we see that: lim ϑt = 0 . (4.117) (σ2 /ω2 )→∞

Thus, when Eq. (4.116) holds, we can take ϑt 0, i.e., kt is almost parallel to at . Moreover, from Eqs. (4.116), (4.24) and (4.25), we get: |kt |2 − |at |2 2 at · kt

.

(4.118)

Eq. (4.118) means that |at | |kt |, so that Eq. (4.25) becomes: |at | |kt | (ωμ2 σ2 /2)1/2 = (πf μ2 σ2 )1/2

,

(4.119)

where f = ω/2π is the frequency of the incident wave. Therefore, the transmitted wave is attenuated along x1 as exp[−(πf μ2 σ2 )1/2 x1 ] as it penetrates into the conductor. The quantity: δ = 1/(πf μ2 σ2 )1/2

,

(4.120)

whose dimensions are those of a length, gives an estimate of the depth up to which the transmitted wave has a nonnegligible amplitude. It is referred to as the penetration depth (or, especially in the past, as the skin depth). In practice, the results that we found for the indeﬁnitely extended medium 2 are reliable if the conductor thickness is equal to a few times the penetration depth. The wave impedance along x1 in the conducting medium can be obtained from Eq. (4.18), taking into account that, because of what we just saw: ˆ1 St = at + jkt (1 + j) (πf μ2 σ2 )1/2 x

,

(4.121)

and noting that, since S is parallel to x ˆ1 , both Et and Ht are perpendicular to x ˆ1 , irrespective of the incidence angle ϑi . In conclusion, it reads: 1/2 πf μ2 jωμ2 Zxˆ1 = = (1 + j) . (4.122) St · x ˆ1 σ2 This impedance depends only on the frequency and on the medium parameters. In contrast to previous results, it does not depend on the direction of incidence of the incoming wave. Therefore, it is a characteristic of the conducting medium, not of the wave propagating in it. This justiﬁes its name, wall impedance, and entitles us to use Eq. (4.122) as a boundary condition on the plane which limits the dielectric medium. The scope of validity to this approach may be limited by curvature of the separation surface; this point will be clariﬁed in Chapter 7. From Eq. (4.122), it follows that: lim Zxˆ1 = 0 ,

σ2 →∞

(4.123)

where no quantity characterizing the incident wave appears. This reminds us of the concept of an ideal electric conductor, found in Chapter 3, which is a useful mathematical tool in many problems, and tells

© 2006 by Taylor & Francis Group, LLC

124

PLANE WAVES IN ISOTROPIC MEDIA

us that it can be regarded as a medium with inﬁnite conductivity, imposing the boundary condition: Etan = 0 (4.124) on its surface. Note that Eq. (4.120) gives δ → 0 as σ2 → ∞, so that the thickness of an ideal electric conductor can be arbitrarily small. 4.13

Further applications and suggested reading

In Chapter 1, we presented Maxwell’s equations, and other equations which follow from them. In Chapter 3, we derived theorems which are based on Maxwell’s equations. In Chapter 4, we began to solve Maxwell’s equations, and this process will now occupy the remainder of the book. Solutions, subject to boundary conditions and often to other limiting assumptions, never have a scope of validity as general as the fundamental equations and theorems of the ﬁrst two chapters. Nevertheless, we may say that, in this chapter, we started oﬀ with solutions whose validity and whose usefulness are very broad. Plane waves appear to be much more ubiquitous than any other time-varying electromagnetic ﬁeld. Rather than writing a list of applications, which, irrespective of its length, could never aim at being complete, this section will limit itself to a few basic recommendations to the reader, and to the explanation of why some subjects have been omitted in this chapter, although they may be very important. First, let us remind the reader that electromagnetic plane waves in a homogeneous medium, being solutions of one of the most widely used equations in mathematical physics — the Helmholtz scalar equation — share an extremely large number of properties with waves of completely diﬀerent physical natures (e.g., Elmore and Heald, 1969; Pain, 1993; Young, 1976), such as acoustic waves in ﬂuids (e.g., Ford, 1970; Frisk, 1994; Brekhovskikh, 1960; Brekhovskikh and Godin, 1990/1991; Meyer and Neumann, 1972; Pierce, 1981), elastic waves in solids (e.g., Ashcroft and Mermin, 1976; Kittel, 1986), wave functions associated with probability distributions of particles in Quantum Mechanics. An impressive example in this sense is provided by the Fresnel reﬂection and transmission formulas (Section 4.9). They were regarded, historically, as the ﬁrst irrefutable proof of the wave nature of light, and were published in 1816, i.e., more than 50 years before Maxwell’s equations. The wave nature of light is by no means a synonym for the electromagnetic nature, a concept which was ﬁrst proposed by Maxwell many decades later. Incidentally, the history of undulatorial theories in physics is dealt with in an ad hoc introductory chapter of Born and Wolf (1980). It may be very useful for the reader to compare the content of this chapter with that of suitable textbooks on acoustics (e.g., Ford, 1970), classical analytical mechanics (e.g., Goldstein, 1980), and quantum mechanics (e.g., Cohen-Tannoudji et al., 1977; Dicke and Wittke, 1960; Merzbacher, 1970;

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

125

Schiﬀ, 1968). As we will see later, very strong similarity between electromagnetic waves and other waves exists also when dealing with guided propagation phenomena. We recommend that the reader return to these references, after reading Chapters 7, 9 and 10. In particular, consulting the more-than-onecentury old Theory of Sound (1877) by Lord Rayleigh would, at the same time, satisfy cultural appetites from the viewpoint of physics, and make the reader more curious about history of science. Plane waves in unbounded isotropic media suﬀer from a number of very severe limitations, which make them inadequate to model some basic phenomena. Just to quote one, at the end of this chapter, the reader does not yet know at which velocity the energy associated with an electromagnetic wave travels. The weak points of plane waves have not been emphasized in this chapter just because it would have been pedagogically wrong and pointless not to insist ﬁrst on their strong points. Their disadvantages, and how to correct them, e.g., using “local plane waves” (Snyder and Love, 1983, Chapter 35), will be contents of the next chapter. We postpone more suggestions for further readings until the end of Chapter 5. Some experienced readers will probably be surprised not to ﬁnd in this chapter an analysis (or design) of multilayer dielectrics. Indeed, this is one of the most important problems that can be tackled using plane waves. The reason why it was not dealt with in detail now is, again, a pedagogical choice. We believe that problems of this kind are dealt with in a much simpler fashion, especially for what concerns calculations, by analogy with electrical transmission lines. These will be covered in depth in Chapter 8. So, it will be advisable to come back to multilayers, and solve again some of the problems presented at the end of this chapter, after studying Chapter 8. Of course, talking about multilayers, under the viewpoint of theory and under that of a broad choice of examples of application, a classical reference that it is highly recommended for consultation is Born and Wolf (1980). We also refer to the work of Brekhovskikh (1960) for waves in multilayer media. Finally, with regard to plane waves in good conductors, in telecommunications they are of primary importance for microwave transmission, to evaluate waveguide loss (see Chapter 5), and in radio propagation, to study penetration into natural conductors, such as salt water at low frequencies, the ionosphere in the MHz range (Budden, 1961), etc. Outside the telecom world, keeping in mind what was said in the ﬁnal section of Chapter 1, interaction of electromagnetic ﬁelds with living tissues, and applications to material diagnostics, are just two among the most eminent examples. The relevant bibliographical suggestions are those that were already quoted in Chapter 1.

Problems 4-1 Show that the Helmholtz equation for the ﬁeld vectors E and H can be solved directly by separation of variables. Show that a generic solution of

© 2006 by Taylor & Francis Group, LLC

126

PLANE WAVES IN ISOTROPIC MEDIA

the equation for E and one of the equation for H do not form a solution of Maxwell’s equations. 4-2 A plane wave traveling in air has an electric ﬁeld amplitude of 10 mV/m. Derive its magnetic ﬁeld in the following cases. a) The wave is uniform, travels in the x3 direction, and the electric ﬁeld is linearly polarized along the x1 axis. b) The wave is uniform, travels in the x3 direction, and the electric ﬁeld is circularly polarized in the (x1 , x2 ) plane. c) The wave is evanescent, with its phase vector in the x3 direction, its attenuation vector in the x2 direction, and the electric ﬁeld is linearly polarized along the x1 axis. Show that in the ﬁrst case only the electric ﬁeld and the magnetic ﬁeld are orthogonal complex vectors. Show that also in the second case the corresponding vectors in the time domain are orthogonal at any instant. 4-3 Find a better approximation than vf c for a uniform plane wave in a low-loss medium where σ ω. Show that the ﬁrst correction is proportional to the square of the loss tangent. 4-4 Derive explicitly all the features of the ﬁeld resulting from two waves of equal amplitude interfering at an angle, as described in Section 4.6. Repeat the same exercise for two waves of diﬀerent amplitudes. 4-5 Show that the following expressions for power reﬂectivity (i.e., ratio of reﬂected and incident power densities), for plane waves at the separation between two semi-inﬁnite media, are equivalent to those which can be derived from the Fresnel formulas: RTE =

sin2 (ϑi − ϑt ) sin2 (ϑi + ϑt )

RTM =

tan2 (ϑi − ϑt ) tan2 (ϑi + ϑt )

for TE waves, and:

for TM waves, respectively. Reconsider the Brewster angle problem in the framework of these new results.

4-6 A layer of a lossless medium, 1 m in thickness, with a refractive index n = 4, is placed between an air-ﬁlled half-space and an ideally conducting plane (located at x1 = 0). A uniform plane wave, at a frequency f = 6 GHz, impinges from air, travels through the layer, and is reﬂected back. Find the period of the standing wave in the dielectric layer, along the x1 direction, for the following cases:

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

127

a) incidence along the x1 axis; b) incidence along a direction such that the refracted wave in the layer travels at 45 degrees with respect to the x1 axis; c) incidence at an angle (in air) of 45 degrees with respect to the x1 axis. 4-7 A uniform plane wave, impinging at normal incidence on the plane (x1 = 0) of separation between two lossless dielectric media, conveys a power density of 1 mW/mm2 . The power density of the transmitted wave is 3 dB below that of the incident wave. Find the modulus of the electric ﬁeld of the reﬂected wave in the following two cases: a) the medium in which the incident and the reﬂected waves travel is air; b) the medium in which the transmitted wave travels is air. Repeat the problem assuming that the incidence angle is π/4 and that the power densities on the x1 = 0 plane remain the same as in the previous cases. 4-8 A TE-polarized uniform plane wave impinges with an incidence angle of π/4 radians on the plane x1 = 0, separating air from a lossless dielectric medium. The reﬂection coeﬃcient equals −0.5. Find the refractive index of the second medium, the reﬂection coeﬃcient for a TM-polarized uniform plane wave coming along the same direction, and the reﬂection coeﬃcients for a TE- and a TM-polarized plane wave coming from the second medium, and giving rise to refracted waves with a transmission angle (in air) equal to π/4 radians. 4-9 Two semi-inﬁnite lossless dielectric media have refractive indices equal to 3 and to 1, respectively. At a frequency of 3×1014 Hz, experimental observation of an evanescent wave in the second medium shows that it decays in the ratio 1/e over a distance of 1 μm, measured along the normal to the plane of separation. a) Find the corresponding angle of incidence, in medium 1. b) Evaluate the phases of the corresponding TE and TM reﬂection coeﬃcients in medium 1, on the separation plane. 4-10 Show that when a wave is incident at the Brewster angle, Eq. (4.95), then the transmitted wave travels at an angle ϑt = π/2 − ϑi . Hence, show that it is impossible to ﬁnd two media such that the Brewster angle at their interface equals π/4. √ 4-11 A lossless layer, with a refractive index equal to 3 and a thickness of 1/8 μm, has been sandwiched between the semi-inﬁnite media of Problem 4-9. a) At what frequencies does it operate as an impedance-matching layer (i.e., reﬂection is canceled at normal incidence) ?

© 2006 by Taylor & Francis Group, LLC

128

PLANE WAVES IN ISOTROPIC MEDIA

b) Which is, in this structure with an intermediate layer, the critical angle for total internal reﬂection of a plane wave coming from the medium whose index equals 3 ? 4-12 Calculate the Brewster angle for the two semi-inﬁnite media of Problem 4-9, and the reﬂection coeﬃcient for a TE polarized wave incident at that angle. Suppose now that a uniform plane wave of unspeciﬁed polarization is incident at the same angle as in the previous question of this problem, but subsequent to the introduction of the intermediate layer discussed in the previous problem. Is the reﬂected wave still TE-polarized ? 4-13 A lossless dielectric layer is located between two air-ﬁlled semi-inﬁnite half-spaces. A uniform plane wave impinges normally on it. The resulting partially-standing wave pattern has a period of 4 mm in air, of 1 mm in the dielectric layer. a) Find the frequency of the wave. b) Find the thicknesses of the layer for which the peak-to-valley ratio of the standing wave, in the half-space where the incident wave comes from, is maximum, and those for which it is minimum. c) Does the layer thickness aﬀect the peak-to-valley ratio of the partially standing wave in the layer ? 4-14 A uniform plane wave, coming from a lossless medium, impinges at oblique incidence onto the plane surface which delimits another semi-inﬁnite medium. Compare the following two cases: a) medium 2 is lossless, and the incident wave undergoes total reﬂection; b) medium 2 is a good conductor. 1. 2. 3. 4. 5. 6.

Are Are Are Are Are Are

the reﬂection coeﬃcients comparable in modulus ? they dependent on the polarization of the incident wave ? their phases identical, or in general not ? their phases polarization-dependent or not ? the attenuation vectors in medium 2 in the same direction ? the phase vectors in medium 2 in the same direction ?

4-15 A typical value for the conductivity of copper is 5.8 × 107 Ω−1 m−1 . Permittivity and permeability may be taken, as an approximation, equal to those of free space. a) Evaluate the order of magnitude of the maximum frequencies at which copper can still be considered to be a good conductor. Calculate the frequency at which the penetration depth in copper equals 1 mm. Find the value of the surface impedance of a ﬂat copper wall at that frequency.

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

129

b) Repeat these calculations for the case of graphite, considered as a good conductor at industrial frequencies. Its conductivity, strongly temperaturedependent, is of the order of 105 Ω−1 m−1 . Are the numbers obtained in this case consistent with each other ? If not, explain why. 4-16 To establish a radio communication link with an immersed submarine, one requires a wave penetration depth of at least 10 meters. Find the maximum usable frequency. For the conductivity of sea water, a typical value is 5 Ω−1 m−1 . Magnetic permeability is practically the same as that of free space. 4-17 Evaluate the modulus of the phase vector for a plane wave in sea water, at a frequency of 500 Hz. Compare it with the modulus that the wave vector has, at the same frequency, for a plane wave in a lossless medium whose dielectric permittivity is equal to that of sea water (whose relative permittivity is of the order of 80). Explain the origin of this very large diﬀerence. (Hint. Use a graphical representation of −ω 2 μc and of its roots on the complex plane.)

© 2006 by Taylor & Francis Group, LLC

References Ashcroft, N.W. and Mermin, N.D. (1976) Solid State Physics. Holt, Rinehart and Winston, New York. Born, M. and Wolf, E. (1980) Principles of Optics. 6th ed., Pergamon, Oxford. Brekhovskikh, L.M. (1960) Waves in Layered Media. Academic, New York. Brekhovskikh, L.M. and Godin, O.A. (1990) Acoustics of Layered Media I: Plane and Quasi-Plane Waves. Springer-Verlag, Berlin. Brekhovskikh, L.M. and Godin O.A. (1991) Acoustics of Layered Media II: Point Source and Bounded Beams. Springer-Verlag, Berlin. Budden, K.G. (1961) Radio Waves in the Ionosphere. Cambridge University Press, Cambridge, UK. Budden, K.G. (1988) The Propagation of Radio Waves: The Theory of Radio Waves of Low Power in the Ionosphere and Magnetosphere. 1st paperback ed. with corrections, Cambridge University Press, Cambridge, UK. Cohen-Tannoudji, C., Diu, B. and Lalo¨e, F. (1977) M´ecanique Quantique. Hermann, Paris. Dicke, R.H. and Wittke, J.P. (1960) Introduction to Quantum Mechanics. AddisonWesley, Reading, MA. Elmore, W.C. and Heald, M.A. (1969) Physics of Waves. McGraw-Hill, New York (Dover, New York, 1985). Ford, R.D. (1970) Introduction to Acoustics. Elsevier, Amsterdam. Frisk, G.V. (1994) Ocean and Seabed Acoustics: A Theory of Wave Propagation. Prentice-Hall, Englewood Cliﬀs, NJ. Goldstein, H. (1980) Classical Mechanics. 2nd ed., Addison-Wesley, Reading, MA. Kittel, C. (1986) Introduction to Solid State Physics. Wiley, New York. Merzbacher, E. (1970) Quantum Mechanics. Wiley, New York. Meyer, E. and Neumann, E.-G. (1972) Physical and Applied Acoustics: An Introduction. Academic, New York. Morse, P.M. and Feshbach, H. (1953) Methods of Theoretical Physics. McGraw-Hill, New York. Pain, H.J. (1993) The Physics of Vibrations and Waves. 4th ed., Wiley, New York. Pierce, A.D. (1981) Acoustics: An Introduction to its Physical Principles and Applications. McGraw-Hill, New York. Schiﬀ, L.I. (1968) Quantum Mechanics. McGraw-Hill, International. Snyder, A.W. and Love, J.D. (1983) Optical Waveguide Theory. Chapman & Hall, London. Young, H.D. (1976) Fundamentals of Waves Optics and Modern Physics. 2nd ed., McGraw-Hill, New York.

131 © 2006 by Taylor & Francis Group, LLC

CHAPTER 5

Plane wave packets and beams 5.1

Modulated waves. Group velocity

Throughout the previous chapter, the reader had to bear in mind two fundamental warnings. One was that a single plane wave of ﬁnite amplitude does not make physical sense, as it would ﬁll the whole space with an inﬁnite energy. The second was that, on the other hand, plane waves form a complete set. Consequently, combinations of plane waves may build up physically meaningful and, more signiﬁcantly, technically important results. Some of them are still mathematically simple, and fairly intuitive, and this chapter is devoted to them. At ﬁrst sight, the subjects which form the chapter might look rather heterogeneous, but this is more apparent than real, as we shall see while we proceed. As we just said, a single plane wave, with an angular frequency ω, invades the whole 3-D space. At any point, its ﬁeld varies sinusoidally vs. time t, for t going from −∞ to +∞; consequently, it does not convey information from one point to another. It is because of this that we accepted that the phase velocity vf , deﬁned in Section 4.4, could be arbitrarily large. This is not a violation of the fundamental postulate of the special theory of relativity, since vf can never be the speed of propagation for a signal that synchronizes two “clocks”. Those readers who have been exposed to the fundamentals of communication theory know that, in order to transmit information, one needs a signal with a ﬁnite spectral width, such as a modulated wave. In order to determine the speed at which the information content carried by an e.m. wave travels, we must then start from a non-monochromatic ﬁeld {E ∼, H ∼}. Actually, in Section 4.7, completeness was proved only for strictly monochromatic plane waves. However, when regularity of functions and derivatives assures that operators in the time domain and in the space domain commute, a ﬁeld having a ﬁnite spectral width in the frequency domain can also be expanded into plane waves, provided it satisﬁes, as a function of space coordinates, conditions similar to those given in Section 4.7. A very simple model, which enables us to determine the speed we are looking for, at least in a lossless medium, consists of two uniform plane waves, traveling in the same direction (i.e., having parallel phase vectors k), but diﬀerent in angular frequencies, ω0 ± Δω, respectively. This notation is reminiscent of an amplitude modulation, with a modulation angular frequency Δω. In the following Section 5.2, however, we will show that the results obtained in this 133 © 2006 by Taylor & Francis Group, LLC

134

PLANE WAVE PACKETS AND BEAMS

way are quite general. To simplify the calculations, without any loss in the scope of validity of the results, we will suppose that these waves have equal polarizations and equal amplitudes, represented by a vector E which, again for simplicity, we assume to be real. Let us introduce rectangular coordinates such that the waves propagate along the x3 axis. In the time domain, the total electric ﬁeld is then expressed by

−j(β+ x3 −ω0 t−Δωt) , (5.1) + E e−j(β− x3 −ω0 t+Δωt) E ∼ (r, t) = Re E e where β+ and β− are the phase constants β = ω μ(ω) (ω) at the angular frequencies ω0 + Δω and ω0 − Δω, respectively. One may always write β+ = βm + Δβ

,

β− = βm − Δβ

,

(5.2)

where βm = (β+ +β− )/2. Quite often, β(ω) is a smooth function of ω, between ω0 − Δω and ω0 + Δω. In that case, there is an angular frequency ωm , in that interval, such that βm = β(ωm ). In general, ωm diﬀers from ω0 , but if Δω is small enough, then we can let βm β(ω0 ). Then, inserting Eqs. (5.2) into Eq. (5.1), we ﬁnd (5.3) E ∼ = E cos(ω0 t − βm x3 ) 2 cos(Δβx3 − Δωt) . This expression looks like a traveling wave (with angular frequency ω0 and phase constant βm ), wrapped in an envelope cos(Δβx3 − Δωt). The loci of constant envelope amplitude are the solutions of d(Δβx3 − Δωt) = 0. They are planes perpendicular to the x3 axis, moving along that axis with a velocity v = Δω/Δβ

.

(5.4)

If the function β(ω) can be expanded as a Taylor series dβ 1 d2 β (ω − ω0 ) + (ω − ω0 )2 + . . . , (5.5) dω 2 dω 2 then, if we let the spectral width of the signal tend to zero, we see that the velocity (5.4) tends to 1 dω = . (5.6) vg (ω0 ) = dβ dβ ω=ω0 dω ω0 β(ω) = β(ω0 ) +

The quantity deﬁned by Eq. (5.6) is referred to as the group velocity of a uniform plane wave with angular frequency ω0 . The procedure that led us to this result indicates that, for a narrow-band signal (Δω ω0 ), traveling in a medium where β(ω) is a smooth function satisfying all the requirements that we introduced, the group velocity (5.6) equals indeed the velocity at which the signal travels. But let us stress that this is not always true. For example, in a case where dβ/dω = 0, Eq. (5.6) would yield an inﬁnite speed, whereas Eq. (5.4) shows that, for Δβ = 0, the envelope propagates at a

© 2006 by Taylor & Francis Group, LLC

MODULATED WAVES. GROUP VELOCITY

135

ﬁnite speed. Furthermore, this value must be less than the speed of light, c, to be compatible with the special theory of relativity. In all cases where these requirements are not satisﬁed by the quantity in Eq. (5.6), we will keep calling it “group velocity,” but we will not state that, in these cases, the signal envelope propagates at the group velocity. While we wait until the next section to show what happens to signals whose spectrum is more complicated than just two frequencies, let us show that Eq. (5.6) has also another physical meaning, namely that of the velocity of energy transport along the direction of propagation. For a uniform plane wave in a lossless medium, in Maxwell’s equations we may write ∇× = (−jk)× = ˆ (−jβ k)×. Then, diﬀerentiating all terms in Maxwell’s equations with respect to ω, we get the following relations: −j

∂(ωμ) ∂H ∂ (k × E) = −j H − jωμ ∂ω ∂ω ∂ω

−j

∂ ∂(ω) ∂E (k × H) = j E + jω ∂ω ∂ω ∂ω

,

.

(5.7)

We take then the scalar product of the ﬁrst one by H∗ , and that of the second one by E∗ . Then, we subtract the ﬁrst one from the second one, and we take into account also the following relationships, which stem directly from Maxwell’s equations and allow us to get rid of several terms: −j k ×

∂E · H∗ ∂ω

= −jω E∗ ·

−j k ×

∂H · E∗ ∂ω

= jωμ H∗ ·

∂E ∂ω

∂H ∂ω

, .

Finally, we obtain: ˆ ∂β = ∂(ωμ) H · H∗ + ∂(ω) E · E∗ (E × H∗ · kˆ + E∗ × H · k) ∂ω ∂ω ∂ω This equation can be rewritten as: 1 ∂(ωμ) ∂(ω) ∗ ∗ ˆ H·H + E·E Re{P · k} = vg 4 ∂ω ∂ω

.

.

(5.8)

(5.9)

It shows that the real part of the Poynting vector ﬂux through a unit-area surface, orthogonal to the direction of propagation, equals the group velocity times a term whose dimensions are those of an energy density. As in ﬂuid dynamics, this tells us that the group velocity is the speed at which energy is carried by this e.m. ﬁeld. We note, though, that the expression for the energy density in general is not the same as the one which appears in Poynting’s theorem, Section 3.2, which was: 1 (μ H · H∗ + E · E∗ ) . 4

© 2006 by Taylor & Francis Group, LLC

(5.10)

136

PLANE WAVE PACKETS AND BEAMS

The two expressions coincide if and only if: ∂ ∂μ =0 , =0 , (5.11) ∂ω ∂ω that is, in what is called a nondispersive medium. The apparent discrepancy between the results can be explained if we remember that in Section 3.2 we dealt with a purely sinusoidal time-harmonic ﬁeld, while now we are studying what happens when we superimpose two plane at diﬀerent frequencies. As powers and energies result from nonlinear operations, even a very simple superposition can deeply modify previous results. In Chapter 4, we saw that, for a monochromatic uniform plane wave in a lossless medium, energy is uniformly distributed throughout the whole space. We could not track an energy “carried” by the wave, and indeed, for such a wave, the divergence of Poynting vector, P = E × H∗ /2, is zero everywhere. On the contrary, the modulated wave that we saw in this section does carry an energy. Its Poynting vector has a nonvanishing divergence, as we can prove in the following way. Consider, as an extension of the previous argument, the following e.m. ﬁeld: 1 E = E(ω0 + Δω) ej(ω0 +Δω)t + c.c. ∼ 2 1 E(ω0 − Δω) ej(ω0 −Δω)t + c.c. + 2 1 ∂E j(ω0 +Δω)t Δω e + c.c. E(ω0 ) + 2 ∂ω ∂E 1 Δω ej(ω0 −Δω)t + c.c. (5.12) + E(ω0 ) − 2 ∂ω ∼ H ∼ = +

∂H Δω ej(ω0 +Δω)t + c.c. H(ω0 ) + ∂ω ∂H 1 Δω ej(ω0 −Δω)t + c.c. H(ω0 ) − 2 ∂ω 1 2

,

(5.13)

where c.c. means the complex conjugate of the preceding term. Suppose that {E(ω0 ), H(ω0 )} is a uniform plane wave, then, ∂E/∂ω and ∂H/∂ω are related to each other by two equations that can be obtained by diﬀerentiating Maxwell’s equations. Now, let us calculate the instantaneous Poynting vector which corresponds to the ﬁeld in Eqs. (5.12), (5.13), including also the products between ﬁelds at diﬀerent frequencies, because, as one lets Δω → 0, the time-averaged values of these products tend to become quantities that are not zero. Let us drop the terms of order (Δω)2 . The assumption that {E(ω0 ), H(ω0 )} is a uniform plane wave implies that the contribution E(ω0 ) × H∗ (ω0 ) and alike have zero divergence. Then, from Maxwell’s equations and from their derivatives with respect to ω, taking into account the relation (C.4) ∇ · (a × b) = (∇ × a) · b − a · (∇ × b), it follows

© 2006 by Taylor & Francis Group, LLC

DISPERSION that:

137

∂E∗ ∂H∗ + ×H ∇· E× ∂ω ∂ω

=j

∂(ωμ) ∂(ω) H · H∗ + E · E∗ ∂ω ∂ω

. (5.14)

The Poynting vector computed from Eqs. (5.12) and (5.13) contains the terms on the left-hand side of Eq. (5.14), minus their complex conjugates. If we subtract Eq. (5.14) from its complex conjugate, we ﬁnd twice the quantity on the right-hand side. So, we see that the energy density which appears on the right-hand side of Eq. (5.9) is indeed related, in the sense of Poynting’s theorem, to that part of the Poynting vector which originates from products of ﬁelds at two diﬀerent frequencies. As we said, only this part of the e.m. energy travels, so these ﬁndings reconﬁrm that the interpretation we gave, for the physical meaning of group velocity, is correct. 5.2

Dispersion

In the previous section, group velocity, on one hand, was calculated at a single frequency, according to Eq. (5.6). On the other hand, its physical meaning can be grasped only if we consider non-monochromatic waves. The whole picture is clear and unambiguous if Eq. (5.6) is independent of ω, over a ﬁnite frequency interval around ω0 . But if vg varies within the signal spectral range, then further developments become necessary. The phenomena which occur when dvg /dω = 0 go under the generic name of dispersion. The historical origin of this term comes from chromatic dispersion of prisms, which, as well known, separate spectral components of white light because glass has a wavelength dependent refractive index. Nowadays, the word dispersion, if used in a transmission context, is essentially a synonym to delay distortion, and means that diﬀerent spectral components travel at diﬀerent speed. Other phrases that are also of general use, regardless of their historical origins, are: • normal dispersion: dvg /dω < 0 (dvg /dλ > 0); • anomalous dispersion: dvg /dω > 0 (dvg /dλ < 0). In the following chapters we will see that these concepts and these names apply not only to plane waves, but to electromagnetic propagation in general. This section is devoted to an example which is based on plane waves, but the concepts we develop here have a much broader scope of validity. Let us consider a packet of uniform plane waves that travel along one direction, say z. Let their spectrum be centered on a “carrier” angular frequency ω0 , and their time-domain envelope, at an initial abscissa z = 0, be a Gaussian function, which simpliﬁes calculations and so allows us to stress physical results. We choose the time reference system such that at t = 0 the center of the Gaussian pulse is at z = 0. Any component of the e.m. ﬁeld can be written as: ψ(z = 0, t) = A exp(−t2 /τ 2 ) cos ω0 t

© 2006 by Taylor & Francis Group, LLC

.

(5.15)

138

PLANE WAVE PACKETS AND BEAMS

The symbol ψ is intended to underline that the following results are independent of the physical dimensions of the quantity one explicitly deals with. To follow the temporal and spatial evolution of the wave packet, it is convenient to use the (time-domain) Fourier transform of Eq. (5.15), Ψ(ω), which is easily calculated, as the Fourier transform of a Gaussian function is a Gaussian function also: √ +∞ 2 2 π −δ2 /(2α)2 e e−α x ejδx dx = . (5.16) α −∞ We get therefore: +∞ 1 Ψ(ω) = √ ψ e−jωt dt 2π −∞ 2 2 Aτ −(ω−ω0 )2 τ 2 /4 √ e . (5.17) = + e−(ω+ω0 ) τ /4 2 2 At any time t ≥ 0 and at any abscissa z ≥ 0, ψ (z, t) can be calculated as ∼ an inverse Fourier transform: +∞ 1 √ ψ (z, t) = Ψ(ω) e−j[k(ω) z−ωt] dω . (5.18) ∼ 2π −∞ Let us assume that the phase constant k, as a function of ω, can be expanded as a Taylor series∗ around ω0 : dk d2 k (ω − ω0 )2 (ω − ω0 ) + + ... . (5.19) dω dω 2 2 Suppose now that, throughout the spectral range where Ψ(ω) has a nonnegligible amplitude compared with Ψ(ω0 ), k varies slowly enough to neglect, in Eq. (5.19), all the terms of order larger than 2 in (ω − ω0 ). Moreover, take into account that, by deﬁnition, it is dk/dω = 1/vg , and introduce the symbol k = d2 k/dω 2 = −(1/vg2 ) dvg /dω. Substituting Eq. (5.19), within the approximation that we just made, into Eq. (5.18), yields: Aτ j[k(ω0 ) z−ω0 t] +∞ −ξ2 τ 2 /4 −j vzg −t ξ jk zξ2 /2 √ e e e e dξ ψ (z, t) = ∼ 4 π −∞ Aτ −j[k(ω0 ) z−ω0 t] +∞ −ξ2 τ 2 /4 −j vzg −t ξ −jk zξ2 /2 √ e + e e e dξ . 4 π −∞ (5.20) k(ω) = k(ω0 ) +

For k = 0, i.e., in a dispersionless medium, Eq. (5.20) becomes much simpler, and, using again Eq. (5.16), it gives: 2 2 . (5.21) ψ (z, t) = A cos ω0 t − k(ω0 ) z e−(t−z/vg ) /τ ∼ Hence, if there is no dispersion, at any z the wave packet is still modulated ∗ As for the part of the spectrum which is centered on −ω0 , it is enough to recall that, for uniform plane waves, k(−ω) = −k(ω).

© 2006 by Taylor & Francis Group, LLC

DISPERSION

139

Figure 5.1 An amplitude-modulated Gaussian pulse.

by a Gaussian envelope whose temporal width, measured at 1/e amplitude, is still 2τ (see Figure 5.1). Accordingly, with no dispersion the envelope travels undistorted, and its speed of propagation coincides with the group velocity, vg . In the presence of dispersion (i.e., for k = 0), the pulse envelope can no longer travel unchanged, as t grows. To show this, let us collect the ﬁrst and the third factor, in each integral in Eq. (5.20), into one complex exponential, and simplify the notation letting: t = t −

z vg

,

D=

2k z τ2

,

T 4 = τ 4 + 4k 2 z 2 = τ 4 (1 + D2 )

. (5.22)

Then, use again Eq. (5.16), which also holds for complex α2 . We get: 2 4 A √ 1 + jD ej[k(ω0 ) z−ω0 t−2k zt /T ] ψ (z, t) = 2 ∼ 2 1+D +

1 − jD e−j[k(ω0 ) z−ω0 t+2k

zt2 /T 4 ]

2

e−t

τ 2 /T 4

. (5.23)

This new expression diﬀers signiﬁcantly from Eq. (5.21), in the traveling wave factor at angular frequency ω0 as well as in the envelope. Changes in the traveling wave (terms in parenthesis) aﬀect mainly its phase, as the factor √ 1 ± jD is trivial, being constant in time. On the contrary, the change represented by the last term in the argument of the exponentials is signiﬁcant, because for z = constant it introduces a t2 time dependence, i.e., a distortion that can be detected with a phase-sensitive receiver. As for the envelope, changes due to dispersion do not aﬀect the motion of the pulse center, which travels again at the velocity vg , nor, in a qualitative sense, the pulse shape, which remains Gaussian. They aﬀect the pulse width, at 1/e amplitude, which is now given by: T2 = 2 τ 2 + 4k 2 z 2 /τ 2 = 2 τ 2 + 4k 2 vg2 t2M /τ 2 , 2Δt = 2 (5.24) τ where tM = z/vg is the time at which, at a given distance z, the pulse am-

© 2006 by Taylor & Francis Group, LLC

140

PLANE WAVE PACKETS AND BEAMS

plitude reaches its maximum. So, as it propagates, the wave packet spreads, and its width tends asymptotically to be proportional to the distance it covered: 2Δt → 4 |k | vg tM /τ . This occurs independently of the sign of k , i.e., both in the normal and in the anomalous dispersion regime. It is important to stress that this last point is a consequence of our choice for the wave packet model. There are other cases (typical ones are frequency-modulated pulses, or “chirped” ones, i.e., pulses that are both frequency and amplitude modulated), where a suitable choice in the dispersion sign may lead to pulse compression as propagation distance grows, until a minimum width is reached. After that, the pulse width starts growing again. Readers who are interested in this subject can ﬁnd further hints at the end of this chapter. The initial width τ appears in the denominator of the parameter which determines the rate at which the Gaussian pulse spreads vs. z. This is because, as one sees from Eq. (5.17), τ −1 is a measure of the spectral width of the pulse. The broader the pulse spectrum in the frequency domain, the higher its sensitivity to a given dispersion in the medium. The last comment should also be read as a warning, when one is dealing with pulses whose τ is very small. In such cases, the spectrum may be so broad, that one cannot neglect the higher-order terms in Eq. (5.19), and the whole content of this section may become inaccurate. Another warning is for the cases where k is not a smooth function of frequency, as happens often around resonant molecular or atomic transitions in material media. Also in these cases the results of this section may become invalid. 5.3

The scalar approximation

In the two previous sections, as well as in most of Chapter 4, we elaborated on one or more components of the e.m. ﬁeld, in a cartesian reference frame, in ways that were not signiﬁcantly aﬀected by the vector nature of these ﬁelds. It is self-explanatory why changing a vector problem into a scalar one is always a very attractive target, but in the case of e.m. waves, it can be reached in an exact way only in rare cases, like those of Chapter 4. Much more often, it can be reached through approximations, as we shall see, for instance, in Chapter 13. Most of the problems that we will deal with in the rest of this chapter belong to the latter class. Therefore, it may be useful, before considering individual cases, to discuss brieﬂy the reasons why a scalar treatment involves approximations, as well as the reasons why, notwithstanding the approximate nature of the approach, we use it; and ﬁnally, to ﬁnd criteria to assess the order of magnitude of the errors that this approach entails. The ﬁrst reason why, in most sections of this chapter, a scalar treatment cannot be rigorous, is because we deal often with waves in non-homogeneous media. In Sections 4.8 and 4.9, it was shown that, at a sharp discontinuity in either or μ, polarization of the incident ﬁeld plays a fundamental role. In a smoothly varying medium (i.e., where and μ are continuous and diﬀerentiable

© 2006 by Taylor & Francis Group, LLC

THE SCALAR APPROXIMATION

141

functions of space coordinates, as we shall suppose in the following sections), we must rediscuss the steps which in Chapter 1 led us from Maxwell’s equations to the Helmholtz equation. In general, ∇ · E = 0 and ∇ · H = 0, so we must take into account the terms ∇(∇ · E) and ∇(∇ · H) when we want to pass from ∇ × ∇× to ∇2 . On the other hand, we know that ∇ · (E) = 0 and ∇ · (μH) = 0, always. These relations can be rewritten in the following form, thanks to the vector identity (C.3): ∇ · E + ∇ · E = 0 , ∇μ · H + μ ∇ · H = 0 . From this we ﬁnd:

(5.25)

∇ · E ∇ × ∇ × E = −∇ E + ∇(∇ · E) = −∇ E − ∇ , ∇μ · H , (5.26) ∇ × ∇ × H = −∇2 H + ∇(∇ · H) = −∇2 H − ∇ μ 2

2

so in an inhomogeneous medium the Helmholtz equation must encompass one more term, which is sensitive to the direction of the vector E (or H) through the scalar product times ∇ (or ∇μ). This is an equation which is not easy to break down into scalar problems in an exact way. The reasons why, nonetheless, even in inhomogeneous media the scalar approximation is used very often, come on one hand from mathematical diﬃculties due to the additional terms in Eqs. (5.26); on the other hand, from the richness of the scientiﬁc literature on scalar waves (acoustic, elastic, quantummechanical waves, etc.), which obey the scalar Helmholtz equation also in inhomogeneous media. Results from these neighboring ﬁelds are often acceptable approximations for e.m. wave propagation also, especially when they lead to numerical methods. One consequence of the scalar approximation is that physical dimensions of the wave lose importance: all the cartesian components of ﬁelds and of the potentials satisfy the same equation, namely: ∇2 ψ − κ2 ψ = 0 .

(5.27)

So, the wave function ψ can have arbitrary physical dimensions. It is often convenient to assume that it is adimensional. Finally, let us look for conditions that must be satisﬁed in order to insure that errors due to the scalar approximation are acceptable. From Eqs. (5.25) and (5.26), we can extract a necessary condition: the medium has to be slowly varying, this phrase meaning that the moduli of the following vectors (whose dimensions are m−1 ): ∇ ∇μ , , (5.28) μ must be small, compared with the scale length over which E and H evolve in space. In some cases (as we will see in Section 5.6), one may ﬁnd additional,

© 2006 by Taylor & Francis Group, LLC

142

PLANE WAVE PACKETS AND BEAMS

more restrictive constraints, but whenever this is not the case, then the typical ﬁeld evolution scale is the wavelength, λ. Let us stress once more that this condition is only a necessary one. There are no suﬃcient conditions with a broad scope of validity; they must be assessed individually for each problem. 5.4

The equations of geometrical optics

Propagation of light, meaning electromagnetic radiation whose wavelength is small compared to the size of all the objects which are present along its path, is often approached, at the most elementary level, in a purely geometrical way. Let us see how we may connect rigorous e.m. theory, i.e., Maxwell’s equations, with simpliﬁed models in terms of “rays,” starting from the axiomatic theory of Chapter 1 and introducing suitable approximations. Taking completeness of plane waves as a generic hint, let us suppose that, in a lossless inhomogeneous medium, any solution of Eq. (5.27) can be written in the following form: (5.29) ψ(r) = C(r) e−jk0 S(r) , √ where k0 = ω μ0 0 is the intrinsic phase constant of free space, while C(r) and S(r) are two real functions of spatial coordinates, which we will call amplitude and phase functions, respectively. At a generic (but ﬁxed) point P = O + r0 , Eq. (5.29) looks like a uniform plane wave. For this reason, the assumption we have introduced now is usually referred to as a “ﬁeld expansion into local plane waves”.† The ansatz Eq. (5.29) enables us to introduce easily some approximations. Indeed, if we substitute Eq. (5.29) into Eq. (5.27), make use of some vector identities (see Appendix C), and divide by exp(−jk0 S(r)) = 0, we get:

(5.30) ∇2 C − jk0 (2∇C · ∇S + C∇2 S) − (∇S)2 k02 + κ2 C = 0 . Having assumed that C and S are real, we may equate to zero separately the second term, which is imaginary, and the sum of the ﬁrst and the third ones, which are real. Considering these, we have: (∇S)2 = −

κ2 1 ∇2 C + 2 2 k0 k0 C

.

(5.31)

The ﬁrst term on the right-hand side in this equation, equal to ω 2 μ/ω 2 μ0 0 , is always ≥ 1. The second term would be equal to 1, if C were a solution of the Helmholtz equation in free space. In such a case, C would vary in space with a period equal to λ0 = 2π/k0 . But in our case, remember that C is just the amplitude of the wave Eq. (5.29). If it varies appreciably on a length scale comparable to λ0 , then the applicability of the idea of local plane waves is not † If we allow the functions C(r) and S(r) to take complex values, then we encompass evanescent local plane waves in the expansion. This entails a remarkable broadening in the scope of validity, but the subject is now premature. We will come back to it in Section 5.13.

© 2006 by Taylor & Francis Group, LLC

THE EQUATIONS OF GEOMETRICAL OPTICS

143

justiﬁed. Hence, we may assume: |∇2 C| k02 |C|

.

(5.32)

Some constraints on the results that follow from this assumption will be outlined in Section 5.13. Eq. (5.31) is then approximated as follows: (∇S)2 = n2

,

(5.33)

where n = (μ/0 μ0 )1/2 is the refractive index, which was already deﬁned in Section 4.9, but now becomes, in general, a function of spatial coordinates. If one is able to solve Eq. (5.33), then one can determine the surfaces over which S(r) = constant, i.e., the constant-phase surfaces of the wave Eq. (5.29). As the reader may know from elementary optics, the shape of such surfaces is strictly related to the capability of a wave of creating “images” of its sources. For this reason, Eq. (5.33) is often referred to as the eikonal equation (or the equation of images, from Greek ικ˜ ω ν = image). Within the approximations that we have accepted when we wrote Eq. (5.32), geometry of the wave propagation is fully contained in Eq. (5.33). Indeed, if the ﬁeld can be expanded, at any point R = O + r, into local plane waves, then each of these waves, utilizing what we saw in Section 4.3, is characterized by a phase vector perpendicular to the constant-phase surface which passes through R. The lines which envelope the local phase vectors, i.e., the lines orthogonal to the surfaces S(r) = constant, are called ray paths. If what one aims at is to ﬁnd the ray paths, the procedure can be simpliﬁed, with respect to solving Eq. (5.33) and then calculating the orthogonal trajectories. We can ﬁnd from Eq. (5.33) another diﬀerential equation where the unknown is the position along the generic ray. To this purpose, we introduce (see Figure 5.2) a curvilinear coordinate u which varies along the ray with a metric coeﬃcient equal to unity (i.e., the ratio between any path length along the ray, and the corresponding change in the coordinate u, equals one). Then, u ˆ = dr/du equals the unit vector tangent to the ray, and d/du = u ˆ·∇ is the direction-derivative operator along the direction of the ray, orthogonal everywhere to a constant-phase surface. Thus we may write: ∇S = n

dr = nˆ u . du

(5.34)

To reach our goal, we must now get from Eq. (5.34) an equation where the function S does not appear explicitly. Let us consider the quantity ∇(n2 ) = 2n ∇n, which is also equal to ∇(nˆ u · nˆ u) and therefore can be expanded (see Eq. (C.2)) as follows: ∇(A · B) = (A · ∇) B + (B · ∇) A + A × (∇ × B) + B × (∇ × A) . Remembering that: ∇ × (nˆ u) = ∇ × ∇S = 0 ,

© 2006 by Taylor & Francis Group, LLC

(5.35)

144

PLANE WAVE PACKETS AND BEAMS

Figure 5.2 The curvilinear coordinate u, along the trajectory of a generic ray.

we get: 2n ∇n = 2n(ˆ u · ∇) (nˆ u) .

(5.36)

Dividing by 2n (diﬀerent from zero everywhere) and rewriting the direction derivative u ˆ · ∇ as d/du, we have: d dr n = ∇n . (5.37) du du Eq. (5.37) is called the ray equation. The procedures for its direct integration, frequently used for propagation in non-homogeneous media (like the atmosphere and the ionosphere at radio frequencies, graded-index lenses and optical ﬁbers at optical frequencies), are called ray tracing. Nowadays, in most cases they are implemented numerically, and many packages for computer-aided design based on ray tracing are commercially available. In several cases of practical interest, Eq. (5.37) is still too complicated to be integrated directly. Then, one makes use of a further simplifying approximation, based on the assumption that all the rays that matter in practice are nearly parallel to a direction, z, called the optical axis of the system. Examples in this sense can be optical instruments with several lenses, whose spacing is not small compared to their transverse dimensions, but they extend also to multiple-reﬂector antennas, together with their feeders. In all these cases we may introduce the so-called paraxial approximation: du dz

,

(5.38)

where z indicates now a cartesian coordinate, measured along the optical axis. So, Eq. (5.37) becomes: dr d n = ∇n . (5.39) dz dz This equation is much simpler than Eq. (5.37), as the derivative along the unknown direction of a ray has been replaced by an ordinary derivative. However, one has to memorize that Eq. (5.39) is a further approximation, compared with the eikonal equation. This can be proved, for instance, by computing the curl

© 2006 by Taylor & Francis Group, LLC

THE EQUATIONS OF GEOMETRICAL OPTICS

145

Figure 5.3 The ray equation as generalized Snell’s law: geometrical scheme. M = center of curvature.

of both sides of Eq. (5.39). Provided that all the continuity requirements be satisﬁed, so that the operators ∇× and d/dz commute, one obtains: d ∇ × (nˆ u) = 0 , (5.40) dz i.e., ∇ × (nˆ u) = constant along the z-axis; on the contrary, as we have shown shortly ago, the eikonal equation yields ∇ × (nˆ u) = 0 along a ray. These two results coincide exactly only for a ray strictly parallel to the z axis, whence the conclusion. Let us show now that Eq. (5.37) can be thought of as a generalized Snell’s law (Eq. (4.84)). Take the scalar product of Eq. (5.37) by any unit-vector vˆ lying in the plane orthogonal to ∇n, i.e., tangent to the n = constant surface that passes through the point R = O + r, which belongs to the ray we are dealing with (see Figure 5.3). We get: vˆ ·

dn dˆ u d (nˆ u) = vˆ · u ˆ +n · vˆ = ∇n · vˆ = 0 . du du du

(5.41)

In particular, if we let vˆ = ˆb (the unit vector orthogonal to the plane deﬁned by the vectors ∇n and u ˆ), then the scalar product u ˆ · ˆb vanishes. So (being n = 0), we obtain: dˆ u ˆ ·b=0 . (5.42) du This equation shows that, in the proximity of the point R, variations of the vector u ˆ along the ray must also be orthogonal to the unit vector ˆb, and consequently, co-planar with u ˆ and with ∇n. Hence, in the proximity of any point R, a ray is well ﬁtted by a planar curve. The restrictions which are expressed by the phrases “in the proximity of R” and “ﬁtted” can be relaxed, if, in the whole medium we are dealing with, the index distribution is such that the vector ∇n remains everywhere parallel to a given plane (see an example in Section 5.6). From what we said, we may conclude that dˆ u/du is the curvature vector of

© 2006 by Taylor & Francis Group, LLC

146

PLANE WAVE PACKETS AND BEAMS

the ray trajectory: dˆ u cˆ =c= du ρ

,

(5.43)

where ρ is the radius of curvature at the point R (see again Figure 5.3). Let ϕ now be the angle between u ˆ and ∇n, whose physical signiﬁcance is the same as that of the incidence angle of Section 4.8. Set now vˆ = a ˆ, the unit vector orthogonal to ∇n and belonging to the plane (ˆ u, ∇n). Then, from Eq. (5.43) we ﬁnally get: −

1 dn sin ϕ + n cos ϕ = 0 . du ρ

(5.44)

Note that, if the direction of ∇n remains constant in the neighborhood of R, then it is du = −ρ dϕ. Therefore, in such a case Eq. (5.44) can be rewritten in the form: d (n sin ϕ) = 0 , (5.45) du which is, as we anticipated, a generalized Snell’s law which holds, as we just said, when the direction of ∇n remains constant. This happens, for example, in the media where the refractive index depends only on one cartesian coordinate. In cylindrically symmetrical media, where ∇n is everywhere in the radial direction (see an example in the next section), this statement applies only to rays whose trajectories belong to planes passing through the symmetry axis, usually referred to as meridional rays. For skew rays, the reader can show as an exercise that what follows from the ray equation, Eq. (5.37), is n cos ϑz = constant, where ϑz is the angle between the symmetry axis of the medium and the tangent to the ray trajectory. This also can be though of as a generalized Snell’s law. The case of spherical symmetry is also discussed in some of the problems at the end of this chapter. In the paraxial approximation, direction derivatives are replaced by the ordinary derivative with respect to z. Restrictions on the consequent results have already been pointed out a shortwhile ago. Let us now go back to the equation that we get if we set the imaginary part of Eq. (5.30) equal to zero, namely: 2∇C · ∇S + C ∇2 S = 0 .

(5.46)

Multiply it by C, and make use of the identity (C.3) ∇·(ϕa) = ∇ϕ·a+ϕ∇·a, as well as of Eq. (5.34). The result is: u) = 0 . ∇ · (C 2 ∇S) = ∇ · (C 2 nˆ

(5.47)

In order to grasp the meaning of this relation, let us imagine a surface, T , which consists entirely of ray trajectories (see Figure 5.4). By deﬁnition, u ˆ is parallel to a ray everywhere. Hence, Eq. (5.47) indicates that this surface T is a “ﬂux tube” of the quantity C 2 n. In other words, if AS denotes that portion of a generic constant-phase surface (S = constant) which is inside T , then

© 2006 by Taylor & Francis Group, LLC

GEOMETRICAL OPTICS: ELECTROMAGNETIC IMPLICATIONS

147

Figure 5.4 A family of rays seen as a “ﬂow tube”.

Eq. (5.47) is equivalent to writing: C 2 n dAS ≡ |ψ|2 n dAS = constant AS

,

(5.48)

AS

(constant meaning independent of the curvilinear coordinate u measured along one of the rays lying on T ). Eq. (5.47), or equivalently Eq. (5.48), express the intensity conservation law of geometrical optics. Its electromagnetic implications will be discussed in the next section. These equations provide a “ﬂuid-dynamics” model of wave propagation. They are also a good starting point if one wants to point out the limitations of geometrical optics. Indeed, let us consider a focus, that is a point where, according to Eqs. (5.37), an inﬁnite number of rays converge into an inﬁnitesimal cross-section As . Then, Eq. (5.48) implies C → ∞. This result is not only physically unrealistic, it is also in contrast with the approximations that we introduced at the beginning of this section, based on Eq. (5.32). This point will be re-examined in the following sections. Going back to Eq. (5.46), we can deduce from it another relationship: by simply taking into account Eq. (5.34), we get: 1 2 dC 1 = − ∇2 S = − ∇ · (nˆ u) . C du n n

(5.49)

This equation is easy to integrate along a generic ray path, yielding: ! u ∇2 S ! u ∇·(nu) ˆ − du − du = C 2 (0) e 0 n . (5.50) |ψ(u)|2 ≡ C 2 (u) = C 2 (0) e 0 n The last formula can be used while tracing the rays, in order to track, from one point to the next, along a ray, how the intensity C 2 (u) evolves. 5.5

Geometrical optics: Electromagnetic implications

The basic equations of geometrical optics, Eqs. (5.33), (5.37) and (5.48), were obtained after the scalar approximation. In fact, they can be obtained also from the e.m. ﬁeld vectors, provided one makes an ansatz similar to Eq. (5.29), namely: E(r) = e(r) e−jk0 S(r)

© 2006 by Taylor & Francis Group, LLC

,

H(r) = h(r) e−jk0 S(r)

,

(5.51)

148

PLANE WAVE PACKETS AND BEAMS

where the vectors e and h have to be supposed as real. If we insert Eqs. (5.51) into Maxwell’s equations (without losses and without sources), the conclusions we reach are the same as in the previous section. The main reason why we repeat this procedure is simply because we want to avoid a misunderstanding. We do not want the reader to believe that a “vectorial” theory of geometrical optics could be more accurate that the “scalar” one. Indeed, let us show that, when one arrives at the equations of geometrical optics, then the problem has already become a scalar one, perhaps without us noticing it, even if one starts from the ansatz in Eq. (5.51) and makes use of Maxwell’s equations. To show that, we introduce Eqs. (5.51) into Maxwell’s equations. We get: ∇ × E ≡ (∇ × e) e−jk0 S + (−jk0 ∇S) × e e−jk0 S = −jωμ h e−jk0 S , (5.52) and likewise for ∇ × H. Divide by k0 exp(−jk0 S) and neglect, for the same reasons that were outlined at the beginning of the previous section, the terms in k0−1 . We obtain: ∇S × e nη h , ∇S × h −n

e η

,

(5.53)

where the refractive index n = (μ/μ0 0 )1/2 and the wave impedance η = (μ/)1/2 are now functions of spatial coordinates, since in general the medium is not homogeneous. Taking into account Eq. (5.34), we get: e . (5.54) u ˆ×e ηh , u ˆ×h− η From Eqs. (5.54) we may deduce some interesting results, that were implicit in the scalar theory of the previous section. In particular, note that these results depend critically on S(r) being a real-valued function. A) The vectors e, h and ∇S (or u ˆ) are mutually orthogonal. Therefore, E and H are transverse, at any point R, with respect to the ray path passing through R. B) As η is real, e and h, and consequently E and H, are in phase, at any point R. C) As a consequence, the Poynting vector P = E × H∗ /2 is everywhere real and parallel to the ray direction. This is the reason why it became common, even in other contexts, to call “direction of the electromagnetic ray” the direction of the real part of the Poynting vector. We will speak again about this terminology in Chapter 6. D) The wave impedance in the ray direction is, at any point R, equal to η(R), the intrinsic impedance of the medium (or rather, its local value at the point R). E) From the previous points A) and D) we can infer that, at any point R, the time-averaged values of the electric and magnetic energy density equal

© 2006 by Taylor & Francis Group, LLC

EXAMPLES OF RAY TRACING

149

each other: 1 1 (R) E(R) · E∗ (R) = μ(R) H(R) · H∗ (R) . 4 4

(5.55)

F) From Eq. (5.55) and from Poynting’s theorem, as there are neither losses nor sources, it follows that the divergence of the Poynting vector equals zero everywhere. Based on this and on the previous point C), Eq. (5.48) indicates that the conserved quantity, referred to as the intensity, in the scalar approximation |ψ|2 n, is proportional to the modulus of the Poynting vector : |H|2 |E|2 ≡η . (5.56) |ψ|2 n ∝ |P| ≡ 2η 2 The physical dimensions of the proportionality constant in the previous relationship are not yet speciﬁed, since the physical dimensions of the scalar wave function ψ are unspeciﬁed as well. However, we may notice that Eq. (5.56) makes it necessary that the refractive index n shows up explicitly in the relationship between ψ and the e.m. vectors E, H. Furthermore, n is involved implicitly, through the local wave impedance η. This point will be raised again and clariﬁed in Section 5.7. In conclusion, we have shown in this section that the approximations of geometrical optics imply that, locally at any point, the e.m. ﬁeld has all the attributes of a uniform plane wave. Therefore, referring to such a ﬁeld as a set of local plane waves, as we did before (after Eq. (5.29)), appears to be a legitimate procedure. Clearly, this approach is intrinsically an approximation. Indeed, we have shown in Chapter 4 that a wave, when impinging on a steep change in parameters of the medium, must be reﬂected in part. The theory we developed here loses any track of reﬂection, as shown by the invariance of the ﬂux of Poynting vector along a ﬂux tube. This is one of its basic drawbacks, or at least limitations. 5.6

Examples of ray tracing in radio propagation and in optics

An isotropic spherically symmetric medium, whose refractive index decreases monotonically as the radial coordinate increases, is a reasonable model for studying radio wave propagation in the earth atmosphere, at heights below the ionosphere. Assume that above the ground (i.e., for r > r0 ) the refractive index varies as:

n(r) = n0 1 − a(r − r0 ) = n0 [1 − ah] , (5.57) where n0 = n(r0 ) is the refractive index at the ground level, and h = r − r0 is the height above the ground. For r < r0 , n(r) is not deﬁned. If we restrict ourselves to heights such that |a| h 1, then Eq. (5.57) can be looked at as a truncated Taylor series, for a widely arbitrary function n(r), being a =

© 2006 by Taylor & Francis Group, LLC

150

PLANE WAVE PACKETS AND BEAMS

Figure 5.5 A spherical coordinate reference frame, to solve the ray equation in a spherically symmetrical medium.

−(1/n0 ) (dn/dr)r=r0 . Therefore, the model is more general than one may think at ﬁrst sight. In spherical coordinates‡ (Figure 5.5), it is convenient to re-write the ray equation (5.37) recalling Eq. (5.43) and using the following identity: d ∂ 1 ∂ =u ˆ · ∇ = cos ϕ + sin ϕ du ∂r r ∂ϑ

.

We obtain then: ∂n 1 ∂n 1 d (nˆ u) ≡ cos ϕ u ˆ + sin ϕ u ˆ + n cˆ = ∇n ≡ −n0 aˆ r. du ∂r r ∂ϑ ρ

(5.58)

(5.59)

The term ∂n/∂ϑ is equal to zero, because of the spherical symmetry. Since cˆ and u ˆ are orthogonal, we see that u ˆ, cˆ and rˆ lay on the same plane and, consequently, the trajectory of any ray is on a plane passing through the origin. Integration of Eq. (5.59) becomes very simple in the cases (quite relevant for terrestrial radio links) where the ray trajectories have little inclination with respect to the horizon. Then, we may assume ϕ π/2, | cos ϕ| 1 everywhere, while ∂n/∂r and n/ρ are of the same order of magnitude. Therefore we get the following equation: n

1 cˆ = −n0 aˆ r , ρ

(5.60)

which shows that, for a > 0, the unit vector cˆ is oriented towards the center of the sphere, indicating that rays bend towards the ground as they propagate. Furthermore, if we assume n n0 on the left-hand side, Eq. (5.60) gives: ρ=

1 = constant . a

(5.61)

‡ Note that, to be consistent with symbols used in the previous section, ϕ is not a coordinate angle in the spherical reference frame, but the angle between the ray and the index gradient. Anyhow, thanks to symmetries in the problem, two spherical coordinates, r and ϑ, will be enough in this section.

© 2006 by Taylor & Francis Group, LLC

EXAMPLES OF RAY TRACING

151

Figure 5.6 Radio transmission beyond the geometrical horizon.

This means that, within this approximation, all the ray trajectories are circles of radius (1/a). If (1/a) > r0 (which is the normal situation in the atmosphere), then it is possible to draw circles of this family, which pass through two points above the ground, without intersecting the earth surface (see Figure 5.6). This result implies that a radio link can be established between two points each of which is beyond the geometrical horizon with respect to the other. As a second example, let us deal with cylindrical symmetry, i.e., a medium whose refractive index depends only on the distance r from an axis, which we take as the z axis of a cylindrical reference frame (Figure 5.7). This can be a model for some types of graded-index optical waveguides (or lenses), to be studied in more detail in Section 10.9 and in the following ones. In particular, we concentrate on the so-called parabolic index proﬁle:

, (5.62) n2 = n20 1 − (r/a)2 where n0 is the maximum index, on axis, and a is a suitable normalizing radius, which sets the scale on which the index decreases as we get away from the axis. The region of interest for ray propagation is where one can assume: r/a 1 . 2

(5.63) ) (n20 /n) rˆ,

in the radial direction The index gradient is found to be −(r/a and always inwards: the cylindrical medium is a focusing one. In the paraxial approximation du dz, no derivative of n appears on the left-hand side of the ray equation, Eq. (5.37), which then reads: dˆ u = −ˆ r(r/a2 ) (n0 /n)2 −ˆ r(r/a2 ) . (5.64) dz This shows that the ray is subject to a radial “pull-back” force, proportional to the elongation away from the z axis. Therefore, by analogy with mechanics of a simple pendulum, the reader may conclude that those solutions of Eq. (5.64) that are contained in a plane passing through the z axis are sinusoidal trajec-

© 2006 by Taylor & Francis Group, LLC

152

PLANE WAVE PACKETS AND BEAMS

r r

a

4

u

∇n

0

n20

n2

z ∇n

-4

Figure 5.7 Ray trajectories in a parabolic-index slab.

tories (see again Figure 5.7). They are called meridional rays. Combinations of sinusoidal motions with equal periods in two orthogonal planes give rise to skew rays, which are, in the most general case, helical trajectories. The reader should, as an exercise, explain why the assumption that the two periods are equal is justiﬁed by Eq. (5.64). For a signiﬁcant number of additional examples of ray tracing in structures which are of interest to optical communications, the reader may consult, for instance, van Etten and van der Plaats (1991). 5.7

The WKBJ method

The subject of this section, although closely related to the previous one, is not easy to link with intuition. It looks advisable to postpone the discussion of its physical meaning and of its relationship with the previous topics, and to outline its mathematical treatment ﬁrst. The technique which is currently referred to as the WKBJ method (from the initials of the four mathematicians, Wentzel, Kramers, Brillouin, and Jeﬀreys, who initiated it, independently of each other) will be outlined in this section in its version for uniform plane waves. The same method can be applied to many other circumstances; examples in this sense will be discussed in the next section and again in Chapter 10. Similarly to Section 5.4, let us suppose that the equation to be solved is (5.27). So, we assume that the scalar approximation holds. Consider a lossless medium, and suppose it is inhomogeneous just because and μ vary only in one direction which we take as the z axis of an orthogonal reference frame: = (z)

,

μ = μ(z) ,

κ2 = −ω 2 μ(z) (z) = −k 2 (z) .

(5.65)

Another assumption, which links the WKBJ method to plane waves, but makes an important diﬀerence with respect to the previous section, is that we

© 2006 by Taylor & Francis Group, LLC

THE WKBJ METHOD

153

want the wave function be separable, namely: ψ(x, y, z) = φ(x, y) f (z) .

(5.66)

Consequently, to solve the scalar Helmholtz equation: ∇2 ψ − κ2 ψ = 0 ,

(5.67)

we have merely to solve the following ordinary diﬀerential equation:

d2 f 2 2 2 = κ (z) + k , t f ≡ −β (z) f dz 2

(5.68)

where kt2 is a separation constant. We will restrict ourselves to the case kt2 < ω 2 μ, i.e., β 2 (z) > 0. The case β 2 (z) < 0 is thoroughly treated in the literature (e.g., Budden, 1961). For β 2 an arbitrary function of z, Eq. (5.68) does not admit an exact solution. The WKBJ approximate solution requires that β 2 (z) be slowly varying on the scale of a local characteristic length, Λ(z) = 2π/|β(z)|: 2 dβ |β| = β . (5.69) dz Λ 2π To clarify the meaning of this characteristic length, let us refer to the case where the unknown function is a plane wave; then, we have kt2 = 0, β 2 = −κ2 , and the condition in Eq. (5.69) is equivalent to assuming that the parameters of the medium vary slowly with respect to the local wavelength λ = 2π/k(z): d(μ) μ (5.70) dz λ . Note that Eq. (5.69) is more general than Eq. (5.70). This will be very useful in the following, mainly in Chapter 10. Let us now set: f (z) = eg(z) . (5.71) Diﬀerentiating with respect to z, we obtain: f = (g 2 + g ) f

.

(5.72)

g 2 + g = −β 2 (z)

.

(5.73)

Then, Eq. (5.68) becomes:

Now it is time to exploit the assumption in Eq. (5.69). Remember that, in a homogeneous medium (β = constant), Eq. (5.27) has uniform plane waves as solutions, for which g(z) = −jβz, and then g = −jβ. Therefore, let us presume now, and verify later, that, when the condition in Eq. (5.69) holds, we are entitled to say that: |g 2 | |g |

.

(5.74)

Then, a ﬁrst approximation to Eq. (5.73) reads: g 2 −β 2 (z) .

© 2006 by Taylor & Francis Group, LLC

(5.75)

154

PLANE WAVE PACKETS AND BEAMS

Anyway, it is easy to check that this approximation is not satisfactory. Eq. (5.75) would give an imaginary g , whence also an imaginary g(z), and so, through Eq. (5.71), |f (z)| = constant. The corresponding wave function ψ = φf would not even satisfy energy conservation, which is respected by geometrical optics. In conclusion, this approximation is actually worse than geometrical optics. In order to improve on Eq. (5.75), we can use a result which is a straightforward consequence of Eq. (5.75) itself, namely: dβ , (5.76) dz and substitute it into the exact equation (5.73). The binomial expansion of the square root, cut oﬀ at the second term (as allowed by Eq. (5.74)), gives then: dβ 1 d ln(β 1/2 ) . (5.77) g ±j β(z) 1 ± j = ±j β(z) − 2 2β (z) dz dz g ±j

Integrating Eq. (5.77), and substituting the result into Eq. (5.71), we get: ! A ±j z β(z ) dz 0 f (z) = 1/2 e , (5.78) β (A = arbitrary constant), which is the ﬁrst-order WKBJ solution, one of the key results of the method. Eq. (5.78) expresses the phase delay of the wave as the result of an inﬁnite number of inﬁnitesimal phase shifts. This is the reason why the integral in the previous exponential function is sometimes referred to as phase memory. The factor β −1/2 in front of the exponential means intensity conservation: in the one-dimensional case (i.e., for φ(x, y) = constant), this factor plays the same role as n−1/2 in the intensity conservation law of geometrical optics |ψ|2 n × area = constant. Hence, we have many indications that geometrical optics and the WKBJ method are strictly interconnected. Indeed, it can be shown that the two methods give results of equal accuracy, measured as the number of powers of Λ that are accounted for in a series expansion of the unknown wave functions and of the equations they must satisfy. Nevertheless, the two theories diﬀer in their scope of application. In principle, the assumption of separation of variables makes the WKBJ method less general than geometrical optics. On the other hand, in practice that assumption makes the WKBJ method more versatile than geometrical optics. For example, one can apply WKBJ to study the dependence of the wave function on a coordinate which, in the paraxial approximation (see Section 5.4), is orthogonal to the z direction. This point will be dealt with in detail in Section 10.10. It is easy to show that also the WKBJ method collapses under some critical conditions, which make at least its ﬁrst-order version, which we saw so far, not applicable. These conditions coincide with those that are critical for geometrical optics, that is to say, they occur near the focal points. Indeed, if we insert

© 2006 by Taylor & Francis Group, LLC

FURTHER COMMENTS ON THE WKBJ METHOD

155

the approximate solution in Eq. (5.78) into the diﬀerential equation (5.73), we ﬁnd that the validity of that solution is subject to the following necessary condition: 3 1 dβ 2 1 1 d2 β − (5.79) β2 . 4 β dz 2 β dz 2 This condition is not satisﬁed where β(z) ∼ = 0, i.e., where kt2 = −κ2 . The surfaces where β(z) = 0 are called caustics. It can be shown, with a rather lengthy procedure, that the same surfaces are loci of focal points when the same problem is examined in terms of ray tracing. In Section 5.13 we will quote brieﬂy some procedures to overcome these restrictions. Eq. (5.79) is more restrictive than Eq. (5.69), which was our starting point. Indeed, Eq. (5.69) is veriﬁed also in the proximity of the extrema of β(z), but in general near such points there are no restrictions on the second derivative of β(z), so Eq. (5.79) is not necessarily satisﬁed. One must then adopt additional care when applying the WKBJ solution near a maximum or minimum of β(z). Finally, let us stress that the WKBJ method is intrinsically a scalar method, like geometrical optics. It is not possible to gain in accuracy by going to a vectorial formulation of the same method. In fact, if one looks for a WKBJ solution for an e.m. ﬁeld, i.e., for a behavior like Eq. (5.78) for all components of the vectors E and H, then one must suppose that every component separately satisﬁes an equation like Eq. (5.73). This entails necessarily a scalar approximation, i.e., an implicit assumption that the double curl operator is replaced by a Laplacian operator even in non-homogeneous media, where this step is not rigorous. 5.8

Further comments on the WKBJ method

Several propagation experiments (for example, radio wave sensing of the ionosphere, or testing of graded-index optical waveguides) deal with slowly varying media of unknown characteristic. One tries to exploit the results of the WKBJ method to evaluate the parameters of the medium from measurements on waves that propagate through it (“inverse WKBJ”). If what is measured is a wave transmitted through the unknown medium, then the equations to be used are those of the previous section. In particular we may use Eq. (5.78), where the value of f (z) at the receiver (z = z0 ) is now a known quantity, and the quantity to be determined is β(z ) for 0 ≤ z ≤ z0 . It is evident that the known quantities are not enough to evaluate β(z ). One needs additional information. Usually, one starts from a model of β(z ) which contains a number of unknown parameters equal to the number that can be evaluated from Eq. (5.78). These unknown parameters are then ﬁtted to the experimental data. Often, however, transmitted waves are much more diﬃcult to measure than reﬂected waves. For example, ionospheric soundings become much simpler to perform if the receiver is placed near the transmitter. This requires the test waves to propagate vertically. However, the WKBJ method is not the most

© 2006 by Taylor & Francis Group, LLC

156

PLANE WAVE PACKETS AND BEAMS

suitable one in order to evaluate distributed reﬂections, due to slow variations in β(z). Indeed, the two solutions of Eq. (5.78), corresponding to waves traveling in opposite directions, look as if they were completely decoupled; but physical intuition says that a strong localized reﬂection should be expected to occur where the incident wave (e.g., Eq. (5.78) with a − sign) reaches a caustic, i.e., points where β = 0. This intuitive idea relies on the fact that beyond the caustic β(z) becomes imaginary, meaning that there our wave becomes evanescent, a situation which is reminiscent of total reﬂection (Section 4.11). To solve this problem with the inverse WKBJ method, one is obliged to work in a region where the approximations of the previous section fail. Anyway, it can be shown (although the proof is too time-consuming to be reported here) that the corrections required because Eq. (5.69) does not hold are small, and physically intuitive. We may then begin to discuss reﬂections starting from the results of the previous section, in particular from Eq. (5.78). Let us set z = 0 at the point of observation of the reﬂected wave, and call z0 the height of the caustic, i.e., β(z0 ) = 0. Because of what we said above, we suppose that Eq. (5.78) applies over the whole range 0, z0 . If fi and fr are the incident and the reﬂected wave, respectively, then we have: " ! β(0) −j z β(z ) dz 0 e fi (z) = fi (0) , (5.80) β(z) " fr (0)

β(z0 ) −j e β(0)

= fr (z0 ) " = fr (z0 )

β(z0 ) j e β(0)

!0 z0

! z0 0

β(z ) dz

β(z ) dz

.

(5.81)

Our goal is a relationship between the reﬂection coeﬃcient at z = 0, ρ = fr (0)/fi (0), and that at z0 . We need ﬁrst to ﬁnd a relationship between fi (z0 ) and fr (z0 ). This cannot be derived rigorously from Eq. (5.78), because this equation gives fi → ∞ and fr → ∞ as z → z0 . One way to overcome the problem could be to suppose that these two inﬁnites are equal: fi (z0 ) = fr (z0 ). Then we would obtain: ! z0 −j2 β(z ) dz 0 . (5.82) ρ=e In reality, this is an arbitrary statement. Eq. (5.78) says only that both wave amplitudes diverge when β(z) → 0. To be more accurate, one can show (e.g., Budden, 1961), through a more careful approximation, that the solutions of Eq. (5.73) near a caustic can be expressed as combinations of Airy functions (Abramowitz and Stegun, Eds., 1965). The only change that this entails with respect to Eq. (5.82) is simply a phase shift of π/2 radians. The ﬁnal result is then: ! z0 π ! z0 −j2 β(z ) dz −j 2 β(z ) dz + 2 0 0 =e . (5.83) ρ=je As we anticipated, this phase correction is easy to link to physical intuition.

© 2006 by Taylor & Francis Group, LLC

GAUSSIAN BEAMS

157

Indeed, it expresses, within the limits of the scalar approximation, the phase shift that we found in Section 4.11 for total reﬂection, which we called then the Goos-H¨ anchen shift. Here as well as there, the shift is a consequence of passing from a real wave impedance to an imaginary one, at z = z0 . 5.9

Gaussian beams

In the previous sections, we gained some experience on how to ﬁnd approximate solutions of Maxwell’s equations in slowly varying media. We can now exploit that experience, to ﬁnd some other approximate solutions, ﬁrst in a homogeneous medium, although they can be extended to some classes of nonhomogeneous media as well. They are physically more realistic than a single plane wave, particularly because they carry a ﬁnite power. At the same time they are still reasonably simple. In a homogeneous lossless medium, the vector Helmholtz equation is strictly valid. It can be written by components in any cartesian reference frame, like for plane waves. So, it is easy to tackle in scalar terms. Therefore, let us consider just one component of the electric ﬁeld, with an angular frequency ω. Let us call (z) the direction, for the time being chosen with total freedom, of a √ phase vector k (|k| = ω μ). We may always write, without any physical implication: E(r) = φ(r) e−jk·r

.

(5.84)

This is simply a change of variables, but soon it will allow us to give a physical meaning to some approximations that we shall make. For a uniform plane wave traveling in the direction of k, we know that φ(r) is a constant. Within the approximation of geometrical optics, the amplitude function C is slowly varying. Therefore, it appears legitimate to see what happens to Eq. (5.84) if the variations of φ vs. z are slow, on the length scale of λ = 2π/k. If we recall the ﬁeld expansion as an integral of plane waves (Section 4.7), then we understand why a ﬁeld like Eq. (5.84) with slowly varying φ is called a beam of plane waves propagating in the z direction. Sometimes it is also called a ray beam. Quantitatively, the approximation we need to make reads: 2 ∂ φ 2k ∂φ . (5.85) ∂z ∂z 2 If we insert Eq. (5.84) into the scalar Helmholtz equation, and neglect the term ∂ 2 φ/∂z 2 because of Eq. (5.85), we get: ∂2φ ∂2φ ∂φ + 2 = 2jk 2 ∂x ∂y ∂z

.

(5.86)

This equation is rather unusual in classical electrodynamics, but, on the contrary, is very well known in Quantum Mechanics, where it is referred to as the (two-dimensional) Schr¨ odinger equation for a free particle, and is usually

© 2006 by Taylor & Francis Group, LLC

158

PLANE WAVE PACKETS AND BEAMS

written as: 2m ∂φ , h ¯ ∂t where φ is the wave function associated with the particle, m is the particle mass, t is time, h ¯ = h/2π, h being Planck’s constant. This equation has been studied so widely, that we can adapt to our needs solutions that can be found in the literature. In particular, we will now study in depth the simplest element in a complete set of discrete, linearly independent solutions. The other elements of this set will be examined in Section 5.10, and shown to share many basic features of the ﬁrst solution. Let then φ be a complex function, whose amplitude, in the generic transverse plane z = constant, varies like a circularly symmetrical Gaussian function: k 2 φ = A exp −j P (z) + r , (5.87) 2q(z) ∇2 φ = −j

where r 2 = x2 + y 2 . The complex quantities P (z) and q(z) are still unknown. Inserting the ansatz (5.87) into Eq. (5.86), we ﬁnd: k2 j (5.88) 2k P + + r2 2 (1 − q ) = 0 , q q where the primes denote derivatives with respect to z. The quantities in brackets are functions only of z, the ﬁrst bracket is multiplied by a constant, while the second one is multiplied by a function of x and y. Therefore, Eq. (5.88) can be satisﬁed everywhere only if the two quantities in brackets are separately equal to zero: q

= 1 ,

P

= −

j q

(5.89) .

(5.90)

The general integrals of these two equations are simply found to be respectively: q P

= z + q0

, z = −j ln 1 + q0

(5.91) .

(5.92)

In the second integral, an arbitrary integration constant B has been omitted, since in Eq. (5.87) P appears in the argument of an exponential function, so the factor exp(B) can be incorporated into the arbitrary constant A. The function q(z), being complex, plays two roles, through Eq. (5.87). It aﬀects the length scale of the Gaussian dependence on x and y in the transverse plane z = constant, and also the shape of the constant-phase surfaces of the wave. As for P (z), it characterizes the slow variations in amplitude and phase of φ as a function of z. However, Eq. (5.87) becomes simpler to analyze if we replace P (z) and q(z), which were useful in making the diﬀerential equations

© 2006 by Taylor & Francis Group, LLC

GAUSSIAN BEAMS

159

Figure 5.8 Behavior of the complex function q vs. the longitudinal coordinate z.

easy, with new variables, whose physical meaning is more direct. First, let us separate real and imaginary parts of 1/q: 1 1 1 = − 2j q R kw2

,

where R and w are real function of z. Eq. (5.87) then becomes: kr2 r2 φ = A exp −j P (z) + . − 2 2R w

(5.93)

(5.94)

From this expression we can easily grasp the signiﬁcance of R(z) and w(z). Recalling Eq. (5.84), we see that the ﬁeld is given by φ multiplied by exp(−jkz); variations in the phase of φ due to P (z) are slow, with respect to λ (see Eq. (5.92)). Consequently, the constant-phase surface passing through the generic point of coordinates (x, y, z) satisﬁes the equation k(z +r 2 /2R) = constant. This says that the surface is an ellipsoid, but near that point it is well approximated by a spherical surface, centered on the z axis, and of radius R. So, physically R(z) is the radius of curvature of the wavefront at the point (0, 0, z). As for w(z), from Eq. (5.94) we see that it equals the radius of a circle, on the transverse plane z = constant, where the modulus of φ equals 1/e times its value on the z axis. It is clear why 2w is referred to as the Gaussian beam diameter, or the spot size. From Eq. (5.91) it is clear that there is one (and only one) value of z, for which the real part of q vanishes (while the imaginary part is independent of z). This is in fact a peculiar z = constant plane, because Eq. (5.93) shows that, when Re(q) = 0, then 1/R = 0. Furthermore (see Figure 5.8), w takes its minimum value on the same plane. Hence, this plane is referred to as the beam waist plane. From now on, it is convenient to choose it as the z = 0 plane. Let w0 denote the minimum of w. In the new reference frame, we have

© 2006 by Taylor & Francis Group, LLC

160

PLANE WAVE PACKETS AND BEAMS 5

R(z)/ z

w(z)/ w

0

4

3

2

1

0

0

1

2

3

4

λ | z|/(πw 20 ) Figure 5.9 Beam width and radius of curvature, vs. normalized distance from the beam waist.

q0 = q(z = 0), so that:

w02 . (5.95) 2 This simpliﬁes very much the z-dependencies of w and R. Indeed, from Eqs. (5.91), (5.93) and (5.95), letting k = 2π/λ, we get: 2 zλ w2 (z) = w02 1 + , (5.96) πw02 q0 = jk

R(z) = z 1 +

πw02 λz

2 .

(5.97)

We see that the wavefront curvature and the relative width of the beam depend on only one variable, namely the ratio between the area of the beam waist, πw02 , and the product λz. They are plotted in Figure 5.9. As we see, the narrower the spot on the beam waist, with respect to λ, the faster the beam diverges as we move away from the waist, and the larger the curvature of the wavefront at a given distance |z|. This is in perfect agreement with diﬀraction theory (see Chapter 13), but in sharp contrast with geometrical optics. This is the ﬁrst important indication that Gaussian beams are indeed an improved approximation, compared to ray optics, especially in the vicinity of focal planes. Eq. (5.96) is, on the plane (z, w), the equation of a hyperbola (Figure 5.10) whose asymptotes are given by: w=±

© 2006 by Taylor & Francis Group, LLC

λ z πw0

.

(5.98)

GAUSSIAN BEAMS

161

Figure 5.10 The 1/e hyperbola and its asymptotes.

Hence, far from the beam waist, i.e., for z πw02 /λ, the beam diverges linearly vs. z, with an angular aperture: 2ϑ = 2 arctan

λ 2 = 2 arctan πw0 kw0

.

(5.99)

This relation conﬁrms that a tightly focused Gaussian beam (small spot size) diverges rapidly, away from the waist plane, while if one wants a well collimated beam, i.e., a beam consisting of almost parallel rays, it is necessary that the beam cross-section be not too small. This comment is fundamental not only in optics, but also in the case of aperture antennas. It expresses the socalled “uncertainty principle” between “conjugate variables,” i.e., between two quantities which span the domains of two functions related to each other by a Fourier transform. It is, in this sense, the same as the relationship between widths of a signal in the time domain and in the frequency domain, which we recalled in Section 4.7. In the present case, the Fourier transform is the plane-wave expansion of the ﬁeld on the beam waist plane. Let us add a short comment, which stems from terminology, but whose real purpose is to stress the underlying links with other subjects, to be dealt with in some of the following chapters. The quantity: πw02 w02 = (5.100) 2 λ is called the Rayleigh range. This name is deliberately reminiscent of a result obtained by Lord Rayleigh, at the end of the 19th Century: in a focused pencil of rays, when the distance z from the focal plane is less than z0 , results based on geometrical optics are severely contradicted by experiment. The physical phenomenon that governs proagation in that range is diﬀraction. This is a further conﬁrmation that Gaussian beams are indeed the big improvement we were looking for, with respect to ray optics. To give Eq. (5.94) a ﬁnal shape, let us substitute Eq. (5.95) into Eq. (5.92). We get: 2z 2z 2z + j arctan − 2 j P (z) = ln 1 − j = ln 1 − j . (5.101) kw02 kw02 kw0 z0 = −jq0 = k

© 2006 by Taylor & Francis Group, LLC

162

PLANE WAVE PACKETS AND BEAMS

Note that:

2 1/2 2z λz w(z) 1 − j = 1+ − = kw02 πw02 w0

,

(5.102)

λz χ = − arctan − 2 . (5.103) πw0 Incidentally, we point out that for any z this quantity χ is between ±π/2. This reconﬁrms our previous statements, that contributions to the phase coming from the factor encompassing P (z) were much smaller than the others. We then get: w0 −jχ e−jP (z) = e , (5.104) w(z) and the whole solution of Eq. (5.86) ﬁnally reads:

kr2 w0 exp −j(kz + χ) − j E=A . (5.105) exp −r2 /w2 (z) w(z) 2R(z) and let:

Note that the factor w0 /w(z) is necessary in order to maintain the power ﬂux through the z = constant planes independent of z, as required by the medium being lossless and homogeneous. In order to strengthen the ties between the contents of this section and their technical fallouts, Table 5.1 shows a set of basic data regarding some of the most popular state-of-the-art laser sources. 5.10

Hermite-Gauss and Laguerre-Gauss modes

Before making the ansatz (5.87), we justiﬁed why the previous section was devoted entirely to one solution of Eq. (5.86). A countable inﬁnity of solutions, forming a complete set, share all the basic features of Eq. (5.105). The completeness of the set will be shortly discussed in Section 5.12. In this section, we will look for the other solutions of Eq. (5.86), starting with a less restrictive ansatz than Eq. (5.87), with more freedom for the dependence of φ(x, y, z) on the transverse coordinates x, y. As we aim at separating variables, we are tempted to set: k 2 r + ξ(z) , (5.106) φ = f1 (x) f2 (y) exp −j P + 2q where f1 (x), f2 (y) and ξ(z) are unknown functions of one coordinate each. Suppose that P and q remain the same functions of z as in the previous section. Then, inserting Eq. (5.106) into Eq. (5.86), exploiting Eqs. (5.89) and (5.90), and dividing by f1 f2 , we ﬁnd: f2 f1 2kx f1 2ky f2 −j −j (5.107) + = 2kξ , f1 q f1 f2 q f2 where the primes denote ordinary derivatives, without any ambiguity. Contrary to what may appear at ﬁrst sight, Eq. (5.107) does not allow

© 2006 by Taylor & Francis Group, LLC

HERMITE-GAUSS AND LAGUERRE-GAUSS MODES

163

Table 5.1 State-of-the-art lasers. Name

Active medium

Wavelength (μm)

Ruby

Crystal (Cr3+ in Al2 O3 )

0.6943

Neodymium: Yag

(Nd

3+

Crystal in Y3 Al5 O12 )

Typical angular divergence (radians)

Typical power range 2.5–25 kW (pulsed)

1.0641

0.5–2×10−3

1–100 W (cw)

Helium-Neon

Gas mixture

0.6328 (1.15, 3.39)

10−3

0.5–2.5 mW (cw)

Argon Ion

Ionized gas (Ar)

0.35–0.52 (typ. 0.488)

< 10−3

0.2–20 W

Carbon Dioxide

Gas mixture (CO2 )

10.4

1.5–2.5×10−3

1 W–1.5 kW (cw)

Gallium Arsenide

Semiconductor (GaAs)

∼ 0.85

∼ 0.3 (in the juction plane) ∼ 1 (normal to the juction)

< 100 mW ∼ (cw)

All-ﬁber

Er-doped glass ﬁber

1.55

10−1

< 100 mW ∼ (cw)

separation of variables. In fact, the two terms on the left-hand side contain q, which is a function of z. To reach our goal, rather than writing Eq. (5.106) we have to let: k 2 r + ξ(z) , (5.108) φ = g1 (t1 ) g2 (t2 ) exp −j P + 2q where:

√ x √ y , t2 = 2 , (5.109) 2 w w are transverse coordinates normalized to the width (5.96) of the Gaussian function, which depends on z. If we substitute Eq. (5.108) into Eq. (5.86), take correctly into account all the terms like (dgi /dti ) (dti /dz), and make use of the following relationship, which comes from Eqs. (5.96) and (5.97): t1 =

2 then we obtain:

g1 g − 2t1 1 g1 g1

d(w2 ) w2 = R dz

+

,

g2 g − 2t2 2 g2 g2

(5.110)

= kw2 ξ

,

(5.111)

where now primes denote derivatives with respect to ti on the left-hand side,

© 2006 by Taylor & Francis Group, LLC

164

PLANE WAVE PACKETS AND BEAMS Table 5.2 Hermite polynomials. Deﬁnition

(−1)n (2t)m−2n

[m/2]

Hm (t) = m!

n=0

n! (m − 2n)!

= (−1)m et

2

dm −t2 e dtm

where [m/2] is the integer part of m/2. Examples H0 (t) = 1 H1 (t) = 2t H2 (t) = 4t2 − 2 H3 (t) = 8t3 − 12t

and z on the right-hand side. The three independent variables are now t1 , t2 , z, and we can separate them. The equations for g1 and g2 are identical, namely: d2 gi dgi − 2ti + 2mi gi = 0 dt2i dti

(i = 1, 2)

.

(5.112)

When the separation constant mi equals a non-negative integer, then the previous equation has a polynomial solution, known as the Hermite polynomial of degree mi . Deﬁnition and examples of Hermite polynomials are shown in Table 5.2. When mi is not a non-negative integer, then Eq. (5.112) still has power-series solutions, but they diverge at inﬁnity faster than exp(t2i ). Hence, when multiplied by the Gaussian function, they would yield ﬁelds which are not modulus-square integrable. So they must be disregarded, as non-physical solutions for our problem. The corresponding equation for ξ(z) reads: 1 . kw2 Inserting Eq. (5.96) into it, its general integral is found to be: zλ ξ = −(m1 + m2 ) arctan + constant . πw02 ξ = −2(m1 + m2 )

(5.113)

(5.114)

Therefore, Eq. (5.86) has the following discrete set of solutions: √ x √ y

w0 Hm 1 Hm 2 exp −r2 /w2 E = A 2 2 w w w 2 λz πr − (m1 + m2 + 1) arctan · exp −j kz + . λR πw02 (5.115) One of their most interesting properties is that they all share the same behavior as functions of z, for what concerns the beam diameter, 2w(z), and the wavefront radius of curvature, R(z). This reconﬁrms that eventually it

© 2006 by Taylor & Francis Group, LLC

HERMITE-GAUSS AND LAGUERRE-GAUSS MODES

165

has been sensible to devote all the previous section to one particular solution, which belongs of course to this set and corresponds to m1 = m2 = 0 in the equations of this section. The solutions in Eq. (5.115) are called Hermite-Gauss modes. An interesting property that can be proven for the Hermite polynomial of any degree m is that its m zeros are all real. Therefore, the orders m1 and m2 in Eq. (5.115) have a simple physical meaning: they are the numbers of zeroes of the function E, Eq. (5.84), along the x and y axes, respectively. They are called the transverse wavenumbers of the Hermite-Gauss modes, which in turn (for reasons similar to those outlined in Section 5.5) are usually labelled TEMm1 ,m2 modes. The Hermite-Gauss modes are linearly independent solutions of Eq. (5.86). This stems from the following orthogonality relation between Hermite polynomials (Abramowitz and Stegun, Eds., 1965): +∞ 2 e−t Hm (t) H (t) dt = 0 for = m . (5.116) −∞

At this stage, let us recall that a Gaussian beam is an approximate solution of the Helmholtz equation that must satisfy the condition in Eq. (5.85). Indeed, not all the solutions of the paraxial equation (5.86) that we have found now, satisfy the restriction in Eq. (5.85). As m1 and m2 grow, the z-derivative of the exponent in Eq. (5.115) grows indeﬁnitely as well. It is left to the reader as an exercise to show that, taking into account Eqs. (5.113) and (5.114) (and assuming, for simplicity, zλ πw02 ), the inequality in Eq. (5.85) is satisﬁed only by those modes for which: w02 . (5.117) λ2 Fortunately, in all cases of practical interest (like laser beams, open resonators — see Chapter 9 — and aperture antennas), most of the ﬁeld power is in the lowest-order Hermite-Gauss modes. Thus, an expansion of the total ﬁeld as a sum of Hermite-Gauss modes does not contradict the initial assumption of slow variations in φ vs. z. For conceptual reasons, to be discussed in the next section, as well as for practical ones, we are driven to ﬁnd solutions of Eq. (5.86) in cylindrical coordinates (r, ϕ, z) as well. Passing from a cartesian system to a cylindrical one, the Gaussian function, circularly symmetrical around the z axis, remains unchanged. The other amplitude factors can be found following essentially the same way as in the previous case. The ﬁnal results are: r

w0 h exp −r2 /w2 E = A w(z) w λz πr2 − (2p + + 1) arctan · exp −j kz ± ϕ + , λR πw02 m1 + m2 4π 2

(5.118)

© 2006 by Taylor & Francis Group, LLC

166

PLANE WAVE PACKETS AND BEAMS Table 5.3 Associated Laguerre polynomials. Deﬁnitions Lp (t) =

p (−1)m p + m=0

where Ls (t) =

p−m

m!

s (−1)m s m=0

m!

m

tm = (−1)

tm = e t

d Lp+ (t) dt

ds (e−t ts ) dts

is the ordinary Laguerre polynomial of degree s. Examples L0 (t) = 1 L1 (t) = + 1 − t 1 1 L2 (t) = ( + 1) ( + 2) − ( + 2) t + t2 . 2 2

where and p are two non-negative integers. We set: 2 √ r 2r 2 Lp h = , w w2

(5.119)

where Lp (t) is the so-called associated Laguerre polynomial of orders , p (see Table 5.3), which are polynomial solutions of the following equation: dL d2 L + pL = 0 . (5.120) + ( + 1 − t) dt2 dt From Eq. (5.118) we see that, when ϕ varies from 0 to 2π, then E, Eq. (5.84), repeats periodically times. For this reason is called the azimuthal wavenumber, and must be an integer, otherwise we would get a multivalued function, which is physically unacceptable. From general properties of Laguerre polynomials (see Abramowitz and Stegun, Eds., 1965), it follows that p equals the number of zeroes of the polynomial Lp , which are all real and positive. For this reason p is called the radial wavenumber. Linear independence among the Laguerre-Gauss modes (also labeled TEMp ) is guaranteed by the following orthogonality relation between generalized Laguerre polynomials of equal azimuthal order: ∞ e−t t Lp (t) Lq (t) dt = 0 for p = q . (5.121) t

0

Well known orthogonality relations apply among the functions e−jϕ over the interval (0, 2π), and so entail modes of diﬀerent azimuthal order to be always orthogonal. Figure 5.11 shows some very famous experimental results of intensities (proportional to |E|2 ) on a z = constant plane, for some Hermite-Gauss modes. Figures 5.12 and 5.13 show the behavior of a few Hermite and Laguerre polynomials. They also indicate that there is no one-to-one matching between these

© 2006 by Taylor & Francis Group, LLC

HERMITE-GAUSS AND LAGUERRE-GAUSS MODES

167

Figure 5.11 Very famous early experimental results: a laser oscillating in various Hermite-Gauss modes (from Boyd, G.D. and Gordon, J.P. (1961) Bell System Technical Journal, 40, 489–508).

© 2006 by Taylor & Francis Group, LLC

168

PLANE WAVE PACKETS AND BEAMS

Hn n3 H5 125

H3 27

H4 64

H2 8

x

Figure 5.12 Examples of Hermite polynomials.

Ln ( x )

L2

L3 x

L5

L4

Figure 5.13 Examples of Laguerre polynomials.

two families of polynomials. Their only relationship is that any Hermite-Gauss mode can be expanded as a series of Laguerre-Gauss modes, and vice-versa. 5.11

Reﬂection and refraction of Gaussian beams

At the beginning of Section 5.9, we introduced an important assumption, namely, that the medium in all the region of 3-D space where a Gaussian

© 2006 by Taylor & Francis Group, LLC

REFLECTION AND REFRACTION OF GAUSSIAN BEAMS

169

Figure 5.14 Reﬂection of a wavefront on a spherical mirror.

beam is deﬁned, be homogeneous. It occurs often that one wants to know how a Gaussian beam behaves at a sharp discontinuity in the medium. This is the case when one analyzes the reﬂection of the beam on a mirror, for example in an optical resonator (see Chapter 9), or the transmission of a beam through a lens. To be rigorous, one could expand the incident beam into plane waves, and apply the results of Chapter 4 to each individual wave. This procedure is quite cumbersome. Frequently, the problem can be simpliﬁed through approximations based on ray optics. We shall discuss them now. Let us start from a perfectly reﬂecting spherical mirror (Figure 5.14) of focal length F , meaning that its curvature radius equals 2F . We take the focus of the mirror as the origin of coordinates. Let us deal explicitly only with the TEM00 mode. The results can be easily extended to higher-order modes, as well as to a thin lens. Suppose (this is the source of approximation) that the distance Ri between the waist of the incident Gaussian beam and the point where the mirror intersects the z axis is so large, that we may approximate the ﬁeld impinging on the mirror with a non-uniform (i.e., amplitude not constant on a constantphase surface) spherical wave, of radius Ri . In this case, the elementary laws of ray optics say that the reﬂected ﬁeld is also a spherical wave, whose curvature radius is related to Ri and to F as: 1 1 1 . (5.122) =− + Rr Ri F The reﬂected spherical wave is also non-uniform. We assumed a perfect reﬂector, so the reﬂected amplitude distribution coincides with that of the incident wave, whence is still Gaussian and its width equals that of the incident wave: wr (z = −F ) = wi (z = −F ). If Rr , given by Eq. (5.122), is also large enough, so that the reﬂected wave is in turn a good approximation to a Gaussian beam, then the complex parameter qr (z) of the reﬂected beam deﬁned by Eq. (5.93): 1 1 1 = − 2j , (5.123) qr Rr kw2

© 2006 by Taylor & Francis Group, LLC

170

PLANE WAVE PACKETS AND BEAMS

Figure 5.15 Reﬂection on a spherical mirror: the ray picture.

is related at the mirror (z = −F ) to that of the incident beam, qi , by the following equation: 1 1 1 = + . (5.124) qr (−F ) qi (−F ) F The diﬀerence in sign between Eq. (5.122) and Eq. (5.124) is simply a consequence of Eq. (5.97), which gives R < 0 for z < 0. Once the value (5.124) of the complex parameter qr (−F ) is known, then q can be calculated at any plane z = constant by means of Eq. (5.91), which can be rewritten as follows: q(z) = q(−F ) + (z + F ) .

(5.125)

It is of interest to see whether we can write how qr at the generic plane z = z2 is related to qi at another generic plane, z = z1 , skipping Eqs. (5.124) and (5.125). Looking at Figure 5.15, we see that R = (x/x ), where x = dx/dz (for paraxial rays). While traveling in the region between mirrors (or lenses), the transverse coordinate x(z) of the generic ray varies linearly vs. z, in proportion to Eq. (5.125), while x = dx/dz remains constant. At each reﬂection, x(z) remains constant while x/x follows Eq. (5.124). Hence, during the whole path from z = z1 to z = z2 , the parameter q undergoes exactly the same transformations as the ratio x/x . Now, it is well known that a spherical mirror maps, in a “one-to-one and onto” way, a family of rays leaving a source point, into another family of rays, passing through an image point. This operation performed by the mirror is fully characterized, in mathematical terms, by a 2 × 2 matrix: A B xi xr = . (5.126) C D xr xi From what we said, it follows that Gaussian beam reﬂections obey the following relationship: qr (z = z2 ) =

© 2006 by Taylor & Francis Group, LLC

A qi (z = z1 ) + B C qi (z = z1 ) + D

,

(5.127)

ON THE COMPLETENESS OF A SERIES

171

where the coeﬃcients A, B, C and D can be determined by elementary ray optics. Eq. (5.127) is usually referred to as the ABCD law. Also a thin lens maps a family of straight lines into another family of straight lines. So, refraction of Gaussian beams at thin lenses can also be described by Eq. (5.127). When using Eq. (5.124), to distinguish between a lens and a mirror it is important to pay attention to the sign of the focal length F . Many contents of this section will be used again in Section 9.10, which deals with stability of open resonators.

5.12

On the completeness of a series

In Section 4.7 we discussed completeness of plane waves and we showed that, in general, a monochromatic ﬁeld, which carries a ﬁnite power through a plane, can be expressed as a plane wave integral. In other words, the basis for expanding such a ﬁeld is a continuous inﬁnity of elements. On the other hand, in Section 5.10 we mentioned on several occasions that we were dealing with a countable set of functions, and yet we stated that it was a complete set. This means that the most general ﬁeld we can have under these circumstances is not an integral, but a series. There is an apparent decrease in generality, so we must discuss, at least qualitatively, the conditions the ﬁeld must satisfy in order to allow such an expansion. Completeness problems, related to a partial diﬀerential equation, are well deﬁned only when boundary conditions are speciﬁed. In general, the weaker the boundary conditions, the wider the set, in order to be complete. So, the proof given in Section 4.7 referred to rather weak boundary conditions, as the only requirement was that the ﬁeld be modulus-square integrable. In such a case, a set could be complete only if it was a continuum. One may ask an interesting question: how stringent must the boundary conditions become, in order to reach completeness with a countable set of basis elements ? If we recall how a Fourier integral is related to a Fourier series, it is easy to conclude that a suﬃcient condition for the spatial spectrum of a ﬁeld to become discrete, is that the ﬁeld should satisfy periodic boundary conditions; or, equivalently, that the ﬁeld be non-zero only over a ﬁnite spatial interval, I. In the latter case, as well known from signal analysis, the ﬁeld can be repeated periodically, all the way to inﬁnity.§ Sometimes, the series convergence could be nonuniform at the extremes of the interval I. When this happens, it is not possible to diﬀerentiate the series term by term. In this framework, we will speak of completeness of a set only § Maybe the terminology used in this section biases the reader towards rectangular coordinates. However, the same concepts, with minor changes, apply in other types of reference frames. For example, in cylindrical coordinates, we shall consider expansions in series of trigonometric functions of the azimuthal coordinate, and Bessel functions of the radial coordinate.

© 2006 by Taylor & Francis Group, LLC

172 in the sense that:

PLANE WAVE PACKETS AND BEAMS 2 n lim cm fm dI = 0 , ψ − n→∞ I

(5.128)

m=1

where ψ can be any solution of the partial diﬀerential equation, {fm } is the complete set, the {cm }’s are the expansion coeﬃcients, and I, as we said, is the interval where the ﬁeld is deﬁned. All the subjects of Chapters 7 and 9 (waveguides with conducting walls and resonant cavities) will enable us to use periodic extension of ﬁelds deﬁned over ﬁnite intervals. It is easy, in those cases, to expect complete sets to be discrete; but Gaussian beams do not belong to the same class. We infer from this that periodicity is a suﬃcient condition for passing from a continuum to a discrete basis, but not a necessary one. There must be another explanation which applies to Gaussian beams. The answer can be identiﬁed as a boundary condition which, although set at inﬁnity, is much more restrictive than simply to require that the ﬁeld be square-integrable. To simplify the proof of this statement, let us use a onedimensional notation, which applies to both Laguerre-Gauss and HermiteGauss modes. Assume that F is a function that can be expanded as a series of Gaussian beams: ∞ ∞ 2 2 Pi (u) e−u = e−u Pi (u) , (5.129) F (u) = i=0

i=0

where Pi (u) is the i-th polynomial of the suitable type. From the last equation it follows that F (u) can be written as the product of a Gaussian function times a power series in u. F (u) is deﬁned in a region which extends to inﬁnity, and Eq. (5.129) must hold in all that region. Then, as u grows, F (u) has to 2 decrease fast enough, so that the Taylor series for F (u) eu converges all the way to inﬁnity. In other words, it must be: n u2 (5.130) lim F (u) e − Pi (u) = 0 , n→∞ i=1

for any u ∈ (0, +∞). As we said above, this constraint is stronger than squareintegrability. In practical problems where Gaussian beams turn out to be convenient (typically, laser beams and aperture antennas), Eq. (5.130) is easily satisﬁed for very simple physical reasons. For example, in an open resonator (Chapter 9), i.e., in a system consisting of two mirrors facing each other, the ﬁnite size of the mirrors in the transverse directions ensures that the ﬁeld is conﬁned, in practice, within a ﬁnite region. For u greater than a certain ﬁnite value, it is F (u) 0. Similarly, for a laser beam, or for the ﬁeld emitted by an aperture antenna, physical intuition leads to the same conclusion. With these limitations and margins of caution, one can speak of completeness of Gaussian beams in all cases where we need to use them.

© 2006 by Taylor & Francis Group, LLC

FURTHER COMMENTS ON RAYS AND BEAMS 5.13

173

Further comments on rays and beams

There are many links between geometrical optics and Gaussian beams. The ﬁrst one is in the terminology. Indeed, at the beginning of Section 5.9 we said that a tentative solution of the type φ(x, y, z) exp(−jkz), where φ is a slowly varying function of z, could be called a beam of rays. In contrast to a single ray, however, such a beam satisﬁes a basic requirement, the “minimum uncertainty relationship,” ΔxΔkx ≥ 2π, between its location in real space and its location in the spatial frequency space. As the reader may know from previous studies, and which will be reconﬁrmed in Chapter 13, this relationship is also fundamental for diﬀraction. Geometrical optics misses completely this point because it deals with each ray as an entity that is totally independent of what surrounds it. It comes as a consequence that energy ﬂows in ﬂux tubes, according to a ﬂuid-dynamics pattern. A critical analysis would show that WKBJ results are also aﬀected by the same limitations since they neglect diﬀraction. This shows that Gaussian beams are more accurate solutions of Maxwell’s equations than WKBJ solutions. Consequently, while geometrical optics and the WKBJ method fail near foci and caustics, where ray densities grow indeﬁnitely and ﬂux-tube cross-sections collapse, Gaussian beams do not have critical points, and are acceptable — although approximate — solutions in the whole 3-D space. It takes a more accurate analysis (e.g., Marcatili and Someda, 1987) to point out some inconsistencies in tightly focused Gaussian beams, in the neighborhood of the waist. As for geometrical optics and the WKBJ method, we wish to point out that suitable “correction procedures” have been developed, which allow one to overcome part of their drawbacks. The results are interesting, but too specialized to be discussed here in detail. Let us just say that there are rules that enable one to track rays beyond a focus. What happens near the focus can never be described in terms of rays. Similarly, there are connection formulas which link to each other two WKBJ solutions deﬁned on each side of the return point, or caustic. They go under the generic name “enlarged WKBJ method,” and yield, among other things, the result on phase shift that was mentioned in Section 5.8. However, these connection formulas are still rather poor approximations. In particular, they fail when there two or more caustics are close to each other. However, it would be naive to think that beams are superior to rays, as solutions of Maxwell’s equations, only because a larger number of powers of λ are kept in the approximation. If this were the case, then one might improve very signiﬁcantly on geometrical optics just by keeping more powers of λ. So, there must be other reasons. In fact, the true cause is that, in geometrical optics, the amplitude and phase functions are assumed to be real. Following this assumption, we divided the Helmholtz equation into two equations, where no term of order k0−1 appears. At this stage, we have diminished the initial level of accuracy. Removing the

© 2006 by Taylor & Francis Group, LLC

174

PLANE WAVE PACKETS AND BEAMS

assumption of real amplitude and phase functions, one ﬁnds Gaussian beams. In this sense, beams can be looked at as complex rays. Physically, what makes the diﬀerence are evanescent waves, which are left out of geometrical optics on one side, but on the other hand play a crucial, although partly hidden, role in Gaussian beams, near the beam waist, where the Poynting vector changes gradually its direction. In the literature, the reader can ﬁnd very important examples (e.g., Felsen and Marcuvitz, 1973) that show how other methods can be pushed to any level of accuracy in terms of powers of 1/k0 , at the price of a rapid growth in computation complexity. 5.14

Further applications and suggested reading

The logical thread of this chapter can be summarized as follows. A monochromatic plane wave is a simple, exact solution (in an unbounded homogeneous medium) of Maxwell’s equations, but it can not be taken literally, because it is unphysical under several viewpoints. In order to ﬁnd similar but physically acceptable solutions, one has to introduce approximations. Diﬀerent approximations are required under diﬀerent circumstances. It is rather unusual for a textbook to present several approximations side by side in one chapter. Our purpose was to help the reader to make comparison, stressing the relationships that exist often among diﬀerent approximations. It happens much more frequently to ﬁnd books which deal in depth with just one, or with a few, of these approximate solutions. Therefore, it is quite easy, for the interested reader, to proceed, on each individual subject, well beyond the introductory level of this chapter. Readers who have been exposed to fundamentals of telecommunications know well that the contents of the ﬁrst two sections (group velocity and dispersion) are extremely important in that context. What happens to signals which travel through non-ideal transmission media must be studied in depth by telecommunication engineers, regardless of the physical mechanism which is responsible for non-ideal propagation. Just to quote one among many classical reference books in this area, in Sunde (1969) the subject of dispersion-related signal distortion is covered in detail, for both analog and digital signals, and well linked to the other crucial issues in transmission and detection. In Section 5.2 we brieﬂy mentioned that pulses may be compressed in the time domain by making a proper choice of the dispersion sign in the medium where they propagate. This property has been exploited for decades at microwave frequencies, where shortening radar pulses is intended to increase spatial resolution (Skolnik, 1980; Baden Fuller, 1990). In the last decades, it has been transferred, very successfully, to the optical frequency range (Rudolph and Wilhelmi, 1989). Grating-assisted compression of laser pulses has been the way to generate the shortest man-made events that are known today, i.e., optical pulses in the range of a few femtoseconds (1 fs = 10−15 s). In optical telecommunications, ingenious techniques have been demonstrated for pulse

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

175

compression such as reversing signal spectrum midway between the transmitter and the receiver with a chirped grating (Hill et al., 1994; Kashyap et al., 1996) or reversing the sign of the dispersion using a DISpersion COmpensating (“DISCO”) optical ﬁber. Section 5.3 was mainly intended to set the stage for the following ones, but it could also be looked at as a hint for those who are interested in understanding more in depth the relation between scalar waves (encountered in many other branches of Physics) and vector waves. Some chapters in the classical treatise by Morse and Feshbach (1953) are quite relevant in this respect. They should be at least sampled by our readers. We also refer to Black and Ankiewicz (1985) for a discussion of analogies between optics and other areas of physics. The following three sections formed a package devoted to a subject — geometrical optics — which has many applications, in basic physics, in optical engineering, in communication engineering, etc. Chapter III of Born and Wolf (1980) is an in-depth view of basic theoretical arguments, those in favor of this approximation as well as those that show its limits. Chapters IV, V and VI in the same book describe exhaustively applications to imaging systems, and to the theory of aberrations (Buchdahl, 1968/1970), a subject which is also of potential interest to engineers involved in aperture antenna design. Readers who want to become more familiar with recent developments in imaging systems should invest some of their time studying graded-index (GRIN) components, and their computer-aided design, based on numerical solutions of the ray equation. They may start from the basic concepts outlined in Part I of Snyder and Love (1983), and then proceed to consult more specialized books, like Marchand (1978), for detailed, sophisticated theories, or Iga et al. (1984) for descriptions of GRIN engineering applications. We also said that geometrical optics has been for a long time, and still is, an important background item also for communication engineers. Radio wave propagation in the slowly-varying (and also random-varying) troposphere is instrumental in designing radio-relay and radar systems. Just to mention a few basic references on this subject, Chapters 33 and 34 of the handbook edited by Jordan (1986) focus on radio transmission problems. On the other hand, Sunde (1969) deals mainly with consequences on transmitted signals. Passing from the troposphere to the ionosphere, in addition to being slowly varying this medium is anisotropic as well. Should one wait for Chapter 6 ? In part, the answer is yes, but we can say now that even in the ionosphere ray tracing remains important for radio engineers. Highly recommended reading is Budden (1961). Ray tracing application to telecom problems enjoyed a second youth in the 1970’s, when optical ﬁbers became practical, and most of the interest was concentrated, for about one decade, on multimode ﬁbers. Snyder and Love (1983) is one of the most complete — and most popular — among the books which distilled that knowledge. Even at times, like the present ones, when single-mode ﬁbers are much more widely used than multimode ones for long-haul systems, it is still mandatory for a telecommunication engineer to

© 2006 by Taylor & Francis Group, LLC

176

PLANE WAVE PACKETS AND BEAMS

be exposed to the geometrical-optics treatment of graded-index multimode ﬁbers. This subject will be dealt with, to some extent, in Chapter 10. Not surprisingly, ionospheric propagation and graded-index optical guides are also among the main examples of application of the WKBJ method to e.m. wave propagation. References given in the previous paragraph, including our own Chapter 10, are still valid for the contents of Sections 5.7 and 5.8. We should also point out that the WKBJ method is applied, very widely, in other branches of pure and applied physics, ranging from underwater propagation (Frisk, 1994) to solid-state electronics (e.g., Kittel, 1996). Gaussian beams, the subject of Sections 5.9 through 5.12, were invented in the early ’60s, to model laser beams and resonators accurately enough to account for their basic diﬀraction limits. It is strongly recommended to come back to this subject after reading Chapter 13. Gaussian beams were successfully used, later on, also for designing aperture antennas (see Chapter 12), especially those with multiple reﬂectors, each of which is in the near ﬁeld of at least another one. Still, their main applications remain in the laser area. Classical textbooks on optical oscillators (see, e.g., Svelto, 1989) devote several chapters to Gaussian beams. All the students who include Quantum Electronics in their curricula must master this subject, and be able to solve problems on Gaussian beams. A special problem, obviously related to the previous ones, is propagation of Gaussian beams in lens-like media, i.e., in media where a slight radial gradient in refractive index may counteract, at least in part, beam divergence caused by diﬀraction. Gaussian beams in media with gain or loss are also required in some very practical problems. Straightforward introductions to these subjects can be found in Yariv (1991) and in Adams (1981). Other important practical problems that have been modeled successfully in terms of Gaussian beams are those related to source-to-ﬁber and ﬁber-to-ﬁber coupling in optical communication systems (Senior, 1992; Tosco, Ed., 1990). In particular, in the case of multimode ﬁbers the latter problem was an interesting example where the application of the whole family of Hermite-Gauss modes (Rizzoli and Someda, 1981) yields fundamental improvements with respect to ray-based approaches. Finally, the content of Section 5.13 should be taken just as a hint. It suggests that the links between Gaussian beams and geometrical optics are deep and fundamental. Clarifying them was a task that was successfully accomplished by theoreticians who investigated e.m. ﬁelds in very abstract terms, without direct interest in designing devices. A very suitable reference, in order to appreciate this, although very demanding reading, is the previously quoted book by Felsen and Marcuvitz (1973). Problems 5-1 Evaluate the group velocity, vg , and the dispersion, k , as a function of the angular frequency ω, in a medium where the phase constant for a plane wave (see Section 6.9) is deﬁned as β+ = (ω/c) 1 + ωM /(ω0 + ω).

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

177

5-2 Repeat the previous exercise with the following expression for the phase constant (see again Section 6.9): β− = (ω/c) 1 + ωM /(ω0 − ω). Note that in this case there is a frequency band where the phase constant is imaginary, i.e., the plane wave is evanescent. The group velocity is not deﬁned over that band. 5-3 Evaluate the group velocity, vy , and the dispersion, k , as a function of frequency, in a dielectric medium, near an atomic or molecular resonant absorption peak at ω = ω0 , where the permittivity exhibits a “reactive tail” of the Lorentzian absorption line, i.e.: = 0 + Δ

2Δω ω02 + 4Q2 Δω 2

,

where Δω = ω − ω0 , and Δ, ω0 and Q are real and positive, while μ = μ0 . Find the frequency intervals where dispersion is normal and those where it is anomalous. (Hint. d/dω = d/(dΔω).) 5-4 A wave is characterized by a phase constant β = (ω/c) [1 − (ωc /ω)2 ]1/2 (deﬁned for ω > ωc > 0, with c = constant). Find its group velocity vs. ω. Show that vf vg = c2 . Show that β → 0 for ω → ∞. Discuss the physical interpretation of these results. √ 5-5 A wave is characterized by a phase constant β = ω μ [1 − (ωc /ω)2 ]1/2 , but now, in contrast to the previous problem, depends on frequency as in Problem 5-3. Assume ω0 ωc . Show that the contributions to dispersion due to d/dω = 0, and those due to the factor [1 − (ωc /ω)2 ]1/2 , have equal signs in a frequency range close to ω0 , where both yield anomalous dispersion; opposite signs far from ω0 , where the “material dispersion” (Problem 5-3) is normal while the “waveguide dispersion” (Problem 5-4) remains anomalous. 5-6 A transmission medium is characterized by a dispersion coeﬃcient |k | = 0.05 ns2 /km, and is 20 km in length. What is the required initial half-width of a Gaussian pulse, for its ﬁnal width to be 3 ns ? Why are there two solutions to this problem ? Which one is more convenient in practice ? 5-7 A simple model for a “ﬂat” atmosphere is a dielectric medium (μ = μ0 ) where the permittivity varies in the vertical direction according to = 1 (1 − az), where 1 = 1.05 0 , a = 10−5 m−1 . Estimate the minimum frequency at which this can be looked at as a slowly varying medium. Discuss whether polarization (horizontal or vertical electric ﬁeld) has an eﬀect on the validity of the slowly-varying approximation. 5-8

Explain why Eq. (5.64) enables us to assume that the two periods, in

© 2006 by Taylor & Francis Group, LLC

178

PLANE WAVE PACKETS AND BEAMS

two diﬀerent meridional planes over which a meridional ray can be projected, are equal. 5-9 Show that, in a spherically symmetrical medium, i.e., for a refractive index n = n(r), the generalized Snell’s law reads: d n(r) r sin ϕ = 0 , du where ϕ is the angle between the tangent to the ray and the position vector r, which, due to the spherical symmetry, is parallel to ∇n. Show that any ray propagating through such a medium is meridional, i.e., belongs entirely to a plane through the origin. (Hint. Show ﬁrst that d(r × nˆ u)/du = 0 and make use of the ray equation.) 5-10 The Luneburg lens is a sphere where the refractive index varies as n(r) = [2 − (r/a)2 ]1/2 for r ≤ a, while n = 1 outside the sphere. Using the result of the previous problem, show that a beam of parallel rays, incident on this lens, is focused “on the back” of the sphere, i.e., on the point where the surface r = a intercepts the diameter parallel to the beam. 5-11 The so-called ﬁsh-eye lens, ﬁrst investigated by Maxwell, is a spherically symmetrical medium where the refractive index varies, for any r, as n(r) = n0 [1 + (r/a)2 ]−1 (n0 is a real constant). a) Show that all the rays emitted by a point source located on the r = a spherical surface pass through the symmetrical point with respect to the center of the lens. b) Show that all the rays emitted by a source located at a point where r > a pass through a point, located on the same diameter, on the opposite side with respect to the origin, and having a radial coordinate r = a2 /r < a (and vice-versa). 5-12 The numerical aperture of an optical system, exhibiting rotational symmetry around an “optic axis” z, is deﬁned as N A = sin ϑM , where ϑM is the maximum angle (measured in air) which a ray propagating through the system can form with the z axis. a) Show that if the refractive index varies along z, either in steps or gradually, then N A = n(z) sin ϑM (z). b) For a rotationally symmetrical medium where n decreases parabolically vs. r (see Eq. (5.62)), show that the numerical aperture (deﬁned with reference to meridional rays) is maximum on the axis and decreases gradually for growing r. 5-13 Consider a parabolic graded-index slab, i.e., a medium where the refractive index is independent of one of the transverse coordinates, y, and varies

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

179

with the other one according to n2 = n20 [1 − 2Δ(x/a)2 ] for |x| < a, and then remains constant for x > a. Rays may reach the borderline r = a, but are not supposed to penetrate further. a) Discuss the relationship between the numerical aperture (deﬁned as in the previous problem, with reference to rays in the (x, z) plane) and the relative index diﬀerence, Δ (which is always supposed to be much less than unity). b) For n0 = 1.5, N A = 0.1, and a maximum index gradient 104 m−1 , ﬁnd the corresponding slab width, 2a. 5-14 Consider again the parabolic graded-index slab of the previous problem. Suppose that the scalar wave equation, Eq. (5.67), can be approached by separation of the variables x and z, and furthermore, that ∂ 2 /∂z 2 = kz2 = constant. Write explicitly the restriction which plays the role of Eq. (5.69), in order to use the WKBJ method, for the equation which governs the x-dependence. Discuss in detail whether it is satisﬁed in the region where |x| a. 5-15 A lossless slab whose refractive index n(z) varies slowly along z (normal to the slab faces), and is independent of x and y, is sandwiched between two semi-inﬁnite lossless media of refractive indices n1 and n2 , at z = 0 and z = L, respectively. Show that the WKBJ method yields the following expression for the reﬂection coeﬃcient at z = 0: n2 n(0) − n1 n(L) . ρ= n2 n(0) + n1 n(L) 5-16 To use a graded-index layer as an antireﬂection coating, one sets n(L) = n2 . Find how the expression of the previous problem changes in this case. Compare this device to the quarter-wavelength antireﬂection coating discussed in Chapter 4, in terms of practical parameters, such as thickness, fabrication tolerances, bandwidth. 5-17 Taking into account Eqs. (5.113) and (5.114) — and assuming, for simplicity, |z|λ πw02 — show that only those TEM modes which satisfy the restriction (5.117) do not violate the inequality (5.85). 5-18 Draw a comparison between |f (z)|2 for the ﬁrst-order WKBJ solution, and |φ(z)|2 , where φ is a TEM0,0 Gaussian beam expressed by Eq. (5.87). Discuss the relationship between these results and the conservation law of geometrical optics, Eq. (5.48). 5-19 The emitting surface of a heterojunction semiconductor laser is, roughly, a rectangular strip. Its width in the junction plane is Δx = 100 μm, and its thickness in the orthogonal direction is Δy = 10 μm. The operating wavelength is 1.3 μm. Suppose that the emitted beam can be modeled as a TEM 0,0

© 2006 by Taylor & Francis Group, LLC

180

PLANE WAVE PACKETS AND BEAMS

astigmatic (i.e., non-circularly-symmetrical) Gaussian beam. Find the beam angular widths in the (x, z) and (y, z) planes, assuming that the beam waist is located on the output facet of the device. After reading Chapter 13, compare these results with those based on diﬀraction theory. 5-20 In cylindrical coordinates, a CO2 laser emits a TEM0,2 beam at a wavelength λ = 10 μm. Far from the waist, the beam full angular divergence is 2 × 10−2 rad. The electric ﬁeld, measured on the beam axis at z = 1 m (distance from the beam waist) equals 100 V/m. Find: a) the spot size on the plane of the beam waist; b) the maximum electric ﬁeld on the beam waist; c) the total power of the beam; d) the coordinates of the points, on the waist plane, where the ﬁeld vanishes. 5-21 A Gaussian beam, emitted by a Nd:YAG laser source (λ = 1.06 μm) whose waist half-width equals 2 mm and whose angular divergence equals 10−4 radians, has to be focused through a circular aperture, 30 mm in diameter, at a distance of 1.4 m from the waist plane. Design a suitable lens (i.e., ﬁnd its distance from the beam waist and its focal length). As a design criterion, note that about 99% of the power of a Gaussian beam is transmitted through an aperture whose radius is 3 times that of the beam at 1/e its maximum value. 5-22 A circular aperture of radius a is illuminated by a uniform ﬁeld, E = E0 = constant (while E = 0 for r > a). Expand this ﬁeld into a set of Laguerre-Gauss modes. (Hint. Use the orthogonality relation Eq. (5.121), and the following identity (Abramowitz and Stegun, 1965, p. 786): ∞

(x) .) e−t Lp (t) dt = e−x L(x) − Lp−1 p x

5-23 Repeat the previous problem, assuming now that the illuminated surface is annular (i.e., b < r < a, b > 0) and that the ﬁeld on it is E = E0 r−1 cos ϕ (again E = 0 out of this region). 5-24 A successful way to model an oﬀset joint between two multimode optical ﬁbers (Rizzoli and Someda, 1981) was to express the ﬁeld in each ﬁber as a series of Hermite-Gauss modes. Show that the so-called overlap integral (i.e., the coeﬃcient in the expansion) between the TEMp,q and the TEMm,n Hermite-Gauss modes, of equal widths w, for an oﬀset d between the centers of the two Gaussian functions along the x axis, and perfect alignment along

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING the y axis, is expressed by d2 n δqn exp − 2w 2 d d tμv = Hm xi − wi Hp xi + (π2p+m p!m!)1/2 i=1 (2)w (2)w

181

,

where δqn = Kronecker’s symbol, a = core radius, s = p + q, [p, q] ≡ v = [s(s + 1)/2] + p + 1 is the so-called principal mode number, and [m, n] ≡ μ. For wi , xi see Abramowitz and Stegun (1965, pp. 890–924).

© 2006 by Taylor & Francis Group, LLC

References Abramowitz, M. and Stegun, I.A. (Eds.) (1965) Handbook of Mathematical Functions. Dover, New York. Adams, M.J. (1981) An Introduction to Optical Waveguides. Wiley, Chichester. Baden Fuller, A.J. (1990) Microwaves: An Introduction to Microwave Theory and Techniques. 3rd ed., Pergamon Press, Oxford. Black, R.J. and Ankiewicz, A. (1985) “Fiber-optic analogies with mechanics”. American Journal of Physics, 53, 554–563. Born, M. and Wolf, E. (1980) Principles of Optics: Electromagnetic Theory of Propagation, Interference and Diﬀraction of Light. 6th (corrected) ed., reprinted 1986, Pergamon Press, Oxford. Buchdahl, H.A. (1968) Optical Aberration Coeﬃcients. Dover, New York. Buchdahl, H.A. (1970) An Introduction to Hamiltonian Optics. Cambridge University Press, UK. Budden, K.G. (1961) Radio Waves in the Ionosphere. Cambridge University Press, Cambridge, UK. Etten, W., van der and Plaats, J., van der (1991) Fundamentals of Optical Fibers Communications. Prentice-Hall, New York. Felsen, L.B. and Marcuvitz, N. (1973) Radiation and Scattering of Waves. PrenticeHall, Englewood Cliﬀs, NJ. Frisk, G.V. (1994) Ocean and Seabed Acoustics: A Theory of Wave Propagation. Prentice-Hall, Englewood Cliﬀs, NJ. Hill, K.O. et al. (1994) “Chirped in-ﬁber Bragg gratings for compensation of opticalﬁber dispersion”. Optics Letters, 19 (17), 1314–1316. Iga, K., Kokubun, Y. and Oikawa, M. (1984) Fundamentals of Microoptics: Distributed-Index, Microlens, and Stacked Planar Optics. Academic, Tokyo and Orlando, FL. Jordan, E.C., (Ed.) (1986) Reference Data for Engineers: Radio, Electronics, Computer and Communications. 7th ed., Howard W. Sams & Co., Indianapolis. Kashyap, R. et al. (1996) “1.3 m long super-step-chirped ﬁbre Bragg grating with a continuous delay of 13.5 ns and bandwidth 10 nm for broadband dispersion compensation”. Electronics Letters, 32 (19), 1807–1809. Kittel, C. (1996) Introduction to Solid State Physics. 7th ed., Wiley, New York. Marcatili, E.A.J. and Someda, C.G. (1987) “Gaussian beam are fundamentally different from free-space modes”. IEEE J. of Quantum Electronics, 23 (2), 164–167. Marchand, E.W. (1978) Gradient Index Optics. Academic, New York. Morse, P.M. and Feshbach, H. (1953) Methods of Theoretical Physics. Vols. I and II, McGraw-Hill, New York. Rizzoli, V. and Someda, C.G. (1981) “Mutual inﬂuence among imperfect joints in multimode-ﬁbre link”. Electronics Letters, 24, 906–907. Rudolph, W. and Wilhelmi, B. (1989) Light Pulse Compression. Harwood, New 183 © 2006 by Taylor & Francis Group, LLC

184

REFERENCES

York. Saleh, B.E.A. and Teich, M.C. (1991) Fundamentals of Photonics. Wiley, New York. Senior, J.M. (1992) Optical Fiber Communications: Principles and Practice. 2nd ed., Prentice-Hall, New York. Skolnik, M.I. (1980) Introduction to Radar Systems. 2nd ed., McGraw-Hill, New York. Snyder, A.W. and Love, J.D. (1983) Optical Waveguide Theory. Chapman & Hall, London. Sunde, E.D. (1969) Communication System Engineering Theory. Wiley, New York. Svelto, O. (1989) Principles of Lasers. 3rd ed., translated from Italian and edited by D.C. Hanna, Plenum, New York. Tosco, F. (Ed.) (1990) Fiber Optic Communications Handbook. TAB Professional and Reference Books, Blue Ridge Summit, PA. Yariv, A. (1991) Optical Electronics. 4th ed., Saunders College Publishing, Philadelphia.

© 2006 by Taylor & Francis Group, LLC

CHAPTER 6

Plane waves in anisotropic media 6.1

General properties of anisotropic media

We brieﬂy mentioned in Section 1.3 that a medium is said to be anisotropic, from an electromagnetic point of view, when it has the following property: at any point, and considering generic directions for the vectors E, H, at least one of the following statements is true: • the total current density vector Jt = J + jωD is not parallel to the electric ﬁeld vector E; • the magnetic induction vector B is not parallel to the magnetic ﬁeld intensity vector H. Tackling anisotropic media analytically is not too complicated when the above mentioned vectors are linked to one another by linear relationships. As is well known, it is absolutely general to express a linear relationship between two vectors in terms of a dyadic operator, i.e., in terms of a second-rank tensor accompanied by the rule of taking its inner product with the vector on its right. If B and Jt are linked by a linear relationship to both E and H, i.e., if we can write: Jt

= g·E+λ·H ,

B

= ν ·E+μ·H ,

(6.1)

then one refers to this as a linear bianisotropic medium. The special case λ = ν = 0 is referred to as a linear anisotropic medium. In the latter case, it is a generalized habit to use the notation of Eqs. (1.23), (1.24) and (1.25), which we rewrite here for future convenience: = σ·E ,

(6.2)

D = ·E ,

(6.3)

J

B

= μ·H .

(6.4)

In a linear isotropic medium, one may always think that Eqs. (6.2), (6.3), and (6.4) still hold, with σ = σ¯i, = ¯i, μ = μ¯i, where ¯i is the identity dyadic. In the following, whenever we speak about anisotropic media, we will implicitly exclude this trivial case. An anisotropic conductor will be, in our terminology, a medium with σ = σ¯i, but with = ¯i and μ = μ¯i. The 185 © 2006 by Taylor & Francis Group, LLC

186

PLANE WAVES IN ANISOTROPIC MEDIA

deﬁnitions of anisotropic dielectric and anisotropic magnetic media are now obvious. Before studying how plane waves propagate in anisotropic media, let us see how some fundamental results, which we obtained in the previous chapters for isotropic media, can either be extended, or modiﬁed. A) Lossless media. Like in the isotropic case, in Maxwell’s equations: ∇ × H = jω · E + σ · E + J0 ∇ × E = −jωμ · H

,

,

(6.5) (6.6)

the right-hand side of Eq. (6.5) can be written in a more compact form if we deﬁne the following dyadic: 1 c = + σ . (6.7) jω By similarity with the isotropic case, we are inclined to refer to this quantity as the complex permittivity dyadic. However, there is a fundamental diﬀerence, with respect to the isotropic case. Here, in fact, for the medium to be lossless it is not required that all the elements of the dyadic c should be real. This conclusion can be drawn directly from Poynting’s theorem. If we introduce the dyadic in Eq. (6.7) into Maxwell’s equations, then the proof of the theorem (see Section 3.2) yields an expression which no longer contains an explicit dissipation term (i.e., a term proportional to the conductivity σ), but instead we ﬁnd that the other volume integral becomes now: E · ∗c · E∗ H∗ · μ · H −2jω − dV . (6.8) 4 4 V For this quantity to be imaginary (i.e., to represent a reactive power), it is not necessary that the dyadics c and μ be real. In fact, the real part of Eq. (6.8) is zero, regardless of the particular integration volume V (i.e., the medium is lossless), if and only if: E · ∗c · E∗ − E∗ · c · E + H · μ∗ · H∗ − H∗ · μ · H = 0 ,

(6.9)

where E, H are arbitrary. This occurs if and only if both dyadics c and μ are Hermitian, i.e.: c = †c , μ = μ† , (6.10) where the symbol † means transposed and conjugate. A medium which can be lossless even with σ = 0, i.e., a “lossless conductor,” sounds puzzling. In fact, we will see later in this chapter that, under some circumstances, charged particles can move, under the inﬂuence of an e.m. wave, in such a way that their convection current is in quadrature with the electric ﬁeld of the wave. The ﬁeld-current product is imaginary, giving rise to a purely reactive power, although the current is not a displacement one. It is evident, however, that this conduction without dissipation results from a model of particle motion where any friction is completely neglected, hence necessarily an approximate model.

© 2006 by Taylor & Francis Group, LLC

GENERAL PROPERTIES OF ANISOTROPIC MEDIA

187

After learning that complex elements in c do not necessarily imply that the medium is lossy, we could drop the distinction between the symbols and c . However, using two diﬀerent symbols may help to prevent the mistake of thinking that E∗ · c · E/4 is always the (time-averaged) electric energy density. Indeed, it can be shown that, in some lossless media, and for some values of the frequency, this quantity can even become negative, which is clearly unacceptable for an energy density. This apparent nonsense can be explained, taking into account that, in such a medium, there is also a storage of distributed kinetic energy, because of the motion of the charged particles. Without getting now into details that are diﬃcult to grasp at this stage, let us repeat that it is because of these problems that, in the following, we will preserve the distinction between , which refers to a zero-conductivity anisotropic dielectric medium, and c , for an anisotropic medium with a ﬁnite conductivity. Another surprise for the reader may be the fact that, similarly, some media can be characterized by a complex permeability dyadic μ. This will also be clariﬁed by means of an example, later on in this chapter. It is well known that, in a complex space, any Hermitian operator is described by a diagonal matrix in a suitable orthogonal reference frame (Smirnov, 1961). Therefore, we can state that for any lossless anisotropic dielectric or magnetic medium, there is a basis — in general, a complex one — where c or μ is represented by a diagonal matrix. However, in a medium characterized by c and μ that are both Hermitian, we cannot state for sure that there exists an orthogonal basis where both dyadics are represented by diagonal matrices. B) Reciprocity. Let us repeat the proof of the reciprocity theorem (Section 3.4), for a linear anisotropic medium, using a real rectangular reference frame, and writing all the vectors by components. Then, we see that the theorem holds if and only if c and μ are both represented by symmetric matrices: ˜ c , (6.11) (c ) = # , (μ) = μ where the symbol ˜ means transposed. As known from linear algebra (Smirnov, 1961), symmetry of a matrix is a property which is invariant under an orthogonal transformation, i.e., a change of basis in real 3-D space. In general, it is not invariant under a unitary transformation, i.e., the most general change of basis in a complex space. Incidentally, this is in contrast with the Hermitian property, which is preserved by unitary transformations. This implies that, while we could speak of lossless media without any need to specify what kind of reference frame we were taking, on the contrary, when the subject is reciprocity in an anisotropic medium, statements are unambiguous only as long as they are restricted to linearly polarized ﬁelds. In engineering contexts, this point is often taken for granted. However, let us stress that it may entail important consequences, that will be outlined in detail in Section 6.6. For the time being, in order to grasp the ﬁrst meaning of this discussion, the reader may verify that the reaction, deﬁned in Section 3.4 and intended to be a mea-

© 2006 by Taylor & Francis Group, LLC

188

PLANE WAVES IN ANISOTROPIC MEDIA

sure of reciprocity, is not an invariant with respect to an arbitrary change of basis in the complex ﬁeld. Indeed, even a very trivial change — multiply one of the basis vectors by a phase factor — modiﬁes the reaction. Mathematically, this surprising behavior is related to the fact that Steinmetz’s representation of time-harmonic quantities is not an isomorphism between the space of real time-harmonic vectors and the 3-D complex vector space, and does not preserve the inner product. 6.2

Wave equations and potentials in anisotropic media

Dyadic operators do not always commute with the curl operator, neither with the divergence one. Accordingly, Maxwell’s equations in anisotropic media cannot be handled in the same way as in the ﬁrst chapter. In general, it is not possible to obtain from Eqs. (6.5) and (6.6) a simple equation, like the Helmholtz equation, containing either E or H, but not both. For a dielectric anisotropic medium (scalar μ), it is possible to obtain an equation where only E appears. To do that, take the curl of Eq. (6.6), replace ∇ × H with Eq. (6.5), and make use of the deﬁnition of the Laplacian of a vector. This yields: ∇2 E − ∇(∇ · E) + ω 2 μ c · E = jωμ J0

.

(6.12)

On the other hand, in such a medium it is impossible to derive an equation where the only unknown is H, at least as long as further information on the dyadic c is not provided. Eq. (6.12) in general is not the same as the Helmholtz equation, because the identity (C.5) ∇ · ∇ × a = 0, applied to Eq. (6.5), gives: ∇ · (c · E) = −

1 ∇ · J0 jω

.

(6.13)

If there are no sources, or if the sources have zero divergence, this yields: ∇ · (c · E) = 0 .

(6.14)

In general the last equation does not entail ∇ · E = 0. Therefore, the second term on the left-hand side in Eq. (6.12) does not disappear. In the following sections, we will see that only in a few special cases Eq. (6.12) can undergo signiﬁcant simpliﬁcation. Duality tells us that in an anisotropic magnetic medium the governing equation is: (6.15) ∇2 H − ∇(∇ · H) + ω 2 c μ · H = 0 , where the second term cannot be eliminated, since Eq. (6.6) and the identity ∇ · ∇ × a = 0 entail: ∇ · (μ · H) = 0 , (6.16) which does not necessarily lead to ∇ · H = 0. In this type of medium, we are unable to write a simple and general equation for the vector E. These problems in writing a diﬀerential equation, either for E or for H, are

© 2006 by Taylor & Francis Group, LLC

BIREFRINGENT MEDIA

189

accompanied by parallel problems if we try to develop a theory of e.m. potentials, similar to the one we built in Chapter 1 for isotropic media. Indeed, let us omit the very complicated case where c and μ are both dyadics (fortunately, not a case of primary importance), and take the case of a homogeneous anisotropic dielectric. We try to repeat the initial steps of Section 1.8. So, we introduce a magnetic vector potential A, whose scope of deﬁnition can be very broad. The ﬁeld vectors can be derived from it, as follows: 1 1 E= c −1 · (∇ × ∇ × A − μ J0 ) , (6.17) H= ∇×A , μ jωμ where c −1 is the inverse of c . The vector A is easily shown to obey the equation: (6.18) ∇ × c −1 · ∇ × ∇ × A − ω 2 μA − μ c −1 · J0 = 0 . In a simply connected domain, we may try to solve Eq. (6.18) introducing an arbitrary scalar function ϕ, as in isotropic media. This yields an equation which diﬀers formally from Eq. (1.65) only because the dyadic c replaces the scalar c : ∇ × ∇ × A − ω 2 μ c · A − μJ0 = −jωμ c · ∇ϕ .

(6.19)

In reality, we run into problems as soon as we try to simplify Eq. (6.19) through the choice of ϕ. Indeed, Eq. (6.19) and the usual identity ∇·∇×a = 0 give us an equation which contains ∇ · (c · A), not ∇ · A. As a consequence, even in the simplest case where ∇ · J0 = 0 (let us recall that we showed that under this assumption, in an isotropic medium, ϕ = 0 is a Lorentz gauge), now we are unable to get the Helmholtz equation from Eq. (6.19), because the term ∇(∇ · A) in general is not zero. The only result that we can reach is that, in those special cases where Eq. (6.12) becomes the Helmholtz equation, then the same occurs to Eq. (6.19), provided the ϕ = 0 gauge is adopted. The duality theorem of Section 3.7 indicates how to cover the case of a magnetic anisotropic medium, in terms of an electric vector potential F of the same type as the one we saw in Section 1.9. The details are left to the reader as an exercise. In conclusion, we saw in this section that anisotropic media are impossible to deal with at the same level of generality as we did in previous chapters. Consequently, we will be often forced to introduce some “ansatz” on the ﬁelds. The most widely used assumption is to deal only with plane waves of a given polarization. Another consequence is that it appears convenient to classify anisotropic media into classes, which diﬀer from each other in the manner of simplifying the equations that the ﬁelds must obey. 6.3

Birefringent media

In this section we will deal with one of the most important classes of anisotropic media. We will assume that they are lossless dielectrics. Including loss would make the mathematics much heavier, without any major gain in understanding

© 2006 by Taylor & Francis Group, LLC

190

PLANE WAVES IN ANISOTROPIC MEDIA

the physics. The dual case of magnetic media with identical formal properties is left to the reader as an exercise. A lossless anisotropic dielectric medium is referred to as birefringent if its dyadic is real, i.e., if all the elements of the matrix () which represent the dyadic in a real reference frame, are real. As we said, we maintain a distinction between and c ; otherwise, some consequences that we are going to draw from Eq. (6.22) would no longer be valid. This deﬁnition, combined with the assumption that the medium is lossless, Eq. (6.10), implies that, in any real rectangular coordinate system (x1 , x2 , x3 ), the matrix () is symmetric: ij = ji

(i, j = 1, 2, 3)

.

(6.20)

Any (3 × 3) real symmetric matrix has three, mutually orthogonal, real eigenvectors (Smirnov, 1961). Therefore, in any birefringent dielectric, there is a particular rectangular reference frame — from now on, referred to as the (x, y, z) system — where its matrix () is diagonal: ⎤ ⎡ ⎡ ⎤⎡ ⎤ x 0 0 Dx Ex ⎣ Dy ⎦ = 0 ⎣ 0 y 0 ⎦ ⎣ Ey ⎦ . (6.21) Dz 0 0 z Ez If the medium is homogeneous, this reference frame is constant in space throughout the region occupied by the medium. In this case, it is called the system of principal axes of the dielectric. The matrix elements x , y , z in Eq. (6.21) are referred to as the principal relative permittivities. The steps one has to go through, to pass from permittivity in a generic frame to Eq. (6.21), lend themselves to a geometrical interpretation, which might help us later on in understanding e.m. wave propagation in such a medium. Let us start from a generic frame (x1 , x2 , x3 ), where is described by the matrix (ij ), and deﬁne the following quadratic form: 3 i,j=1

ij xi xj =

3

ij xi x∗j

,

(6.22)

i,j=1

where x∗j = xj , since the coordinates are real. This form is easily seen to be deﬁnite positive. In fact, if we replace xi with the component Ei of the electric ﬁeld, Eq. (6.22) becomes four times the time-averaged electric energy density, which must be positive by deﬁnition for any nonvanishing ﬁeld. Writing that a deﬁnite positive form like Eq. (6.22) is equal to a constant, one gets an equation which, in analytical geometry, is the equation of an ellipsoid. Passing from a generic reference frame to the coordinate system where Eq. (6.21) holds, the ellipsoid equation reads: x x2 + y y 2 + z z 2 = constant .

(6.23)

Using the language of geometry, we call this the ellipsoid equation in the frame of its principal axes. This explains the names that were introduced earlier. Eq. (6.21) reconﬁrms that, in general, in an anisotropic dielectric medium

© 2006 by Taylor & Francis Group, LLC

BIREFRINGENT MEDIA

191

the vectors D and E are not parallel. Further information on how these two vectors lay can be obtained if we make the assumption (to be conﬁrmed later on), that a linearly polarized plane wave is allowed to propagate in this medium. Assume then, on the basis of what we saw in Chapter 4, that all the ﬁeld vectors vary in space according to exp(−j k · r), where k is a constant real vector. Then, the curl operator may be replaced by (−j k)×. Maxwell’s Eqs. (6.5) and (6.6), in a region without sources, read: −j k × H = jω D ,

(6.24)

−j k × E

(6.25)

= −jωμ H .

From these we see that H is perpendicular to D, to E, and to k. Consequently, these three vectors must all lay on one plane. Eq. (6.24) also shows that D is perpendicular to k. On the other hand, Eqs. (6.24) and (6.25) do not yield any information on the direction of E, except for the previous point that it belongs to the plane deﬁned by k and D. This entails that, in general, the vector E has a nonvanishing component along the direction of propagation. Consistently with the previous assumption of linear polarization, let us assume H = H0 exp(−jk · r), with a real H0 . Then, the vectors E0 = E/ exp(−jk · r) and D0 = D/ exp(−jk · r) are also real, and can be depicted as in Figure 6.1. The Poynting vector P = E × H∗ /2 is also real, and furthermore, being perpendicular to H0 , it belongs to the same plane as D0 , E0 and k. By construction, it is also orthogonal to E0 , and therefore in general it is not parallel to the phase vector k. The angle that the Poynting vector makes with the phase vector equals the angle between E0 and D0 . Accordingly, for a plane wave in a birefringent medium power ﬂows, generally speaking, in a direction (called the ray direction) which is not the same as the normal to the constant-phase planes (the wave normal direction). In Section 4.11, we saw that superpositions of evanescent waves in isotropic media do convey power in a direction perpendicular to their phase vector, but let us stress, to avoid any ambiguity, that the phenomenon that we are discussing here is completely diﬀerent, from a physical point of view. Let us aim now at ﬁnding a link between the vectors D and E, for any direction of the phase vector. Let us deﬁne a refractive index n as a function of the direction of k, as follows: k = n β0 kˆ

,

(6.26)

where β0 = ω(μ0 0 )1/2 , while kˆ is the unit vector in the direction of k. Such a deﬁnition stems from one of the physical meanings of the refractive index in an isotropic medium, namely that of ratio between phase velocity in free space, and phase velocity in the medium. On the other hand, it does not entail anything about the relationship between D and E. Insert then Eq. (6.26) into Eqs. (6.24) and (6.25), assume for simplicity μ = μ0 , and eliminate H. The result is: (6.27) D = n2 0 kˆ × E × kˆ = n2 0 Et ,

© 2006 by Taylor & Francis Group, LLC

192

PLANE WAVES IN ANISOTROPIC MEDIA

Figure 6.1 A plane wave in an anisotropic medium: relationships between its electromagnetic vectors.

where Et is the transverse component of E, with respect to the direction of propagation. To determine n as a function of the parameters of the medium, it is convenient to exploit the inverse permittivity (also called impermittivity) dyadic b, deﬁned as: b = 0 −1

,

(6.28)

0 E = b · D .

(6.29)

such that: ˆ Let us project Eq. (6.29) on a constant-phase plane (perpendicular to k), and take into account that D belongs to this plane. Thus, we obtain: 0 Et = bt · D ,

(6.30)

ˆ where the dyadic bt is the projection of b onto a plane orthogonal to k, therefore it is a dyadic in a 2-D subspace and can be described by a 2 × 2 matrix (Smirnov, 1961). The system consisting of Eqs. (6.30) and (6.27) yields an eigenvalue equation: 1 bt · D = 2 D . (6.31) n Therefore, n2 can be identiﬁed as the inverse of one of the eigenvalues of the operator bt . This enables us, ﬁrst of all, to count how many values the refractive index can take for a given direction of propagation. Let us take a rectangular coordinate frame with an axis parallel to the direction of propˆ Then, thanks to our initial assumptions on , the 2 × 2 agation (i.e., to k). matrix (bij ) representing bt is real and symmetric. Consequently, it has two real eigenvalues, in general diﬀerent from each other. In conclusion, for waves traveling along a generic direction, the refractive index may take two diﬀerent values. It is self-explanatory, now, why these media are called birefringent. Eq. (6.22), as already said, is a positive deﬁnite quadratic form (in lossless

© 2006 by Taylor & Francis Group, LLC

FRESNEL’S EQUATION OF WAVE NORMALS

193

media). Hence, both eigenvalues are positive.∗ Each eigenvalue gives rise to two opposite values of n, which correspond, obviously, to two waves traveling ˆ in opposite directions, ±k. Under the assumptions that were made at the beginning of this section, the operator bt has two real orthogonal eigenvectors. Each of them belongs to one of the above mentioned eigenvalues. This means that there are two possible, mutually orthogonal linear states of polarization for the D vector of plane waves traveling in a birefringent medium. Any other state of polarization is a ﬁeld whose dependence on spatial coordinates cannot be written in the simple form exp(−jk · r). So, to see how a ﬁeld, known at a given point, in an unspeciﬁed state of polarization, evolves in space, ﬁrst we must decompose it as the sum of the two linear polarizations corresponding to the eigenstates that we have just encountered. These are then two waves which travel independently of each other. The total ﬁeld at any distance is to be written as the sum of those two linearly polarized waves, after they propagate over that distance. As these two waves have two diﬀerent phase constants, the net result is that the state of polarization of the total ﬁeld, in general, is a function of distance, along the direction of propagation. The reader should be able to explain, as an exercise, why this function is periodic in space. 6.4

Fresnel’s equation of wave normals

In the previous section we studied the fundamentals of wave propagation in birefringent media, but we did not arrive at operative rules. We will show now how to calculate the main features of plane waves which travel in these media. Solving the eigenvalue equation (6.31) requires two cumbersome operations, to invert the matrix which represents , and to change the coordinate system. There is a faster way to compute the refractive indices, for a given direction of propagation, starting from Eqs. (6.3) and (6.27), which (i being the identity dyadic) entail: ˆ kˆ . (6.32) ( − n2 0 i) · E = −n2 0 (E · k) In the principal-axes reference frame (where is diagonal), it becomes very simple to write Eq. (6.32) by components. Let kˆ ≡ (vx , vy , vz ). We get: (n2 − x ) Ex

= n2 vx (kˆ · E) ,

(n2 − y ) Ey

= n2 vy (kˆ · E)

(n2 − z ) Ez

= n2 vz (kˆ · E) .

, (6.33)

∗ Many results of this chapter can be extended to the case where c = + (1/jω) σ is real. In that case, however, as we saw in Section 6.2, the quadratic form E ∗ · c · E does not have a physical meaning which assures that n2 is always positive. When a negative value is found for n2 , thus an imaginary value is found for ±n, the corresponding plane wave becomes extremely similar, physically, to the evanescent waves of Chapter 4. Such a formalism can be very useful in modeling ionospheric reﬂections (see, e.g., Budden, 1961).

© 2006 by Taylor & Francis Group, LLC

194

PLANE WAVES IN ANISOTROPIC MEDIA

Multiply the ﬁrst of these equations by vx , the second by vy , the third by vz , sum the three equations, and divide both sides by kˆ · E (leaving out isotropic media, where kˆ · E vanishes). We get the following equation: vy2 vx2 v2 1 + + 2 z = 2 2 2 n − x n − y n − z n

,

(6.34)

currently referred to as Fresnel’s equation of the wave normals. It allows an easy computation of n2 , once the principal permittivities and the propagation direction (vx , vy , vz ) are given. Eq. (6.34) is biquadratic in n, because of the identity vx2 + vy2 + vz2 = 1. Hence, it always gives two solutions for n2 , in agreement with what we saw in Section 6.3. When written in a rational form, it holds also in isotropic media (x = y = z = r ), and in this case it gives, √ as it must give, solutions n = ± r that are independent of the direction of propagation. The directions of the two linear polarizations allowed for D are now simple to determine by means of geometry. In the principal-axes reference frame (x, y, z), consider the ellipsoid, centered in the origin, deﬁned by the following equation (note that it is not the same as Eq. (6.23)): x2 y2 z2 + + =1 . x y z

(6.35)

This surface is called index ellipsoid, or the optical indicatrix. We will show now that, for a wave traveling parallel to a unit vector kˆ = (vx , vy , vz ), the two directions that are allowed for linearly polarized D vectors are the major and minor axes of the ellipse which is the line where the ellipsoid Eq. (6.35) intersects the plane orthogonal to kˆ passing through the origin whose equation reads: (6.36) vx x + v y y + v z z = 0 . We will also prove that, with a suitable normalization, the lengths of these two axes are equal to the indices n given by Eq. (6.34). To do this, let us ﬁrst point out that the axes we are looking for are the minimum and maximum diameter lengths for the ellipse. Therefore, they may be found as conditional extrema of the distance from the origin, r, or of a convenient power of r. The simplest choice is to deal with: r 2 = x2 + y 2 + z 2

.

(6.37)

2

The constraints we must impose on r are Eqs. (6.35) and (6.36), meaning that the ellipse axes terminate at points that belong to the ellipsoid and to the ˆ The problem can be tackled with the standard method plane orthogonal to k. of Lagrange multipliers. Namely, we write a linear combination of Eqs. (6.37), (6.36) and (6.35), with auxiliary parameters 2λ1 , λ2 , i.e.: 2 x y2 z2 + + F (x, y, z) = x2 +y 2 +z 2 +2λ1 (vx x+vy y +vz z)+λ2 ; (6.38) x y z next, we take the partial derivatives of Eq. (6.38) with respect to x, y, z, and

© 2006 by Taylor & Francis Group, LLC

FRESNEL’S EQUATION OF WAVE NORMALS

195

Figure 6.2 Geometrical construction of the refractive indices and the allowed states of polarization, using the optical indicatrix.

set them equal to zero. We get the following system of equations: m =0 (m = x, y, z) . m + λ 1 vm + λ 2 m

(6.39)

The parameters λ1 and λ2 can be eliminated combining Eqs. (6.35), (6.36) and (6.39). So, we get three equations, the ﬁrst of which reads: 2 vx x vy y vz z r 2 − 1 x = r vx + + . (6.40) x x y z The other two can be obtained simply by permuting x with y and with z. These equations deﬁne, by construction, the extrema of the ellipse axes. Now, we observe that the ﬁrst of Eqs. (6.33) becomes Eq. (6.40) if we set n2 = r2 , x = Dx = x Ex . The same occurs with the other two components and the other two coordinates. This proves that n is proportional to the axis length, and, at the same time, it proves that the directions of the axes coincide with those allowed for D (see Figure 6.2). In other words, this completes our proof. We run into a case of special interest when two out of the three principal permittivities are identical, i.e., for x = y = z . The ellipsoid described by Eq. (6.35) has then a rotational symmetry around the z axis. In this case, for ˆ one of the axes of the ellipse where the ellipsoid intersects any direction of k, the plane normal to kˆ is, at the same time, a diameter of the circle where the ellipsoid intersects the plane z = 0 (which, in turn, is either a minimum or a maximum in length among all the intersection ellipses, depending on whether x is smaller or larger than z ). Consequently, in these media one of the two values for n (usually referred to as the ordinary index, n0 ) is independent of the direction of propagation, while the other one (referred to as the extraordinary ˆ The names given to the indices index, ne ) does depend on the direction of k. reﬂect the habit of calling ordinary and extraordinary waves, respectively,

© 2006 by Taylor & Francis Group, LLC

196

PLANE WAVES IN ANISOTROPIC MEDIA

the plane waves associated with n0 and with ne . The term birefringence, in a given direction, is in current use to denote the absolute value of the diﬀerence, |n0 − ne |, for a pair of plane waves traveling in that direction. Note, however, that sometimes it is used for the relative diﬀerence, |n0 − ne |/n0 , and on other occasions it means the absolute diﬀerence between the two phase constants, |n0 − ne |β0 . The quantity LB = 2π/|n0 − ne |β0 is called the beat length. It is simple to show, as an exercise, that, if an ordinary wave and an extraordinary one travel in the same direction, then the state of polarization of the total ﬁeld, resulting from the overlap of the two waves, evolves periodically as a funtion of distance, and LB is its spatial period. The values of n are both independent of kˆ only in a medium where x = y = z , i.e., in an isotropic medium. In this case, the ellipsoid given by Eq. (6.35) becomes a sphere, and all directions orthogonal to kˆ are legitimate for D. If, on the contrary, z = x = y , then there is only one direction for kˆ — the z-direction — for which we ﬁnd ne = n0 . For waves traveling along this axis, the directions allowed for D are all those in the plane orthogonal to z. A medium of this kind is called a uniaxial medium. Trigonal, tetragonal and hexagonal crystals fall within this category (Nye, 1985). When the three principal permittivities diﬀer from each other, it is simple to prove that there are two (and only two) planes through the origin that intersect the ellipsoid in circles. Hence, there are two directions of propagation (normal to those planes), for each of which the two values of n2 are equal, and any direction normal to propagation is allowed for D. These media are referred to as biaxial. Refraction at the separation between an isotropic and an anisotropic medium can be now studied along the guidelines of Section 4.8. For the sake of brevity, let us omit evanescent waves, which can be dealt with by the reader as an exercise. Then, in general (i.e., except for the special cases where the refracted waves travel along one of the optic axes) there are two transmitted waves, traveling in two diﬀerent directions, and orthogonally polarized. This phenomenon is referred to as double refraction. Each refracted wave obeys Snell’s law, Eq. (4.84), with the value of n which is relevant to that wave (either the ordinary, or the extraordinary index). Reﬂection and transmission coeﬃcients can be evaluated by decomposing any incident wave into the two polarizations that correspond to the ordinary and extraordinary waves in the birefringent medium, respectively, and then applying Fresnel’s formulas of Section 4.9 to each of them. Another refraction phenomenon, diﬀerent from those we saw so far, is peculiar of biaxial media. It is referred to as conical refraction. To outline it, let us consider a biaxial slab, deﬁned by two planes normal to one of the optic axes, and illuminated by a uniform plane wave at normal incidence. By definition of optic axis, any direction in the normal plane is permissible for D, and the refractive index n is independent of the direction of D. Snell’s law in Eq. (4.84) shows that, for normal incidence, the direction of the phase vector does not change as the wave passes from one medium to the other. However,

© 2006 by Taylor & Francis Group, LLC

AN APPLICATION: PHASE MATCHING OF TWO WAVES

197

Figure 6.3 Conical refraction.

as we saw in Section 6.2, the vector E in general is not in the same direction as the vector D, having a longitudinal component (see Figure 6.1). Consequently, the Poynting vector is not parallel to the optic axis, and furthermore, its direction, belonging always to the plane orthogonal to the magnetic ﬁeld, depends on that of the vector D. Suppose now that the incident wave is depolarized, i.e., is a random superposition of waves whose D vector is linearly polarized in all directions orthogonal to the propagation direction, with a ﬂat probability distribution. The corresponding directions for the vector E form a surface, which can be shown (see Born and Wolf, 1980, Chapter 14) to be ˆ by a cone. Note, however, that the axis of this cone cannot be parallel to k: deﬁnition of biaxial medium (the three principal permittivities are diﬀerent), an optic axis can never coincide with one of the principal axes, and therefore the length of the longitudinal component of E depends on the direction of D. Consequently, the directions of the Poynting vector also form a cone, whose ˆ Hence, its intersection with the slab output plane is axis is not parallel to k. an ellipse. So, if the slab is illuminated over a ﬁnite round spot, and if the birefringent slab is thick enough, compared to the spot radius, then light (i.e., active power) emerges at the output face through an elliptical-ring-shaped area (see Figure 6.3). It is self-explanatory why this phenomenon is called, as we said before, conical refraction. Note that, as the wave normals remain always orthogonal to the input and output planes, there is no further refraction when the light leaves the slab. A few numerical data on typical uniaxial and biaxial crystals will be given at the end of the next section. 6.5

An application: Phase matching of two waves

Quite often, at least at optical frequencies, the nonlinear phenomena of highest practical interest (e.g., harmonic generation, frequency mixing, parametric ampliﬁcation, etc.) take place in birefringent media. Physics of nonlinear

© 2006 by Taylor & Francis Group, LLC

198

PLANE WAVES IN ANISOTROPIC MEDIA

phenomena in crystals, and nonlinear e.m. wave propagation, are outside the scope of this book. Those readers who are interested in such topics are strongly encouraged to consult several specialized books (Bloembergen, 1996; Zernike and Midwinter, 1973; Agrawal, 1995; Someda and Stegeman, Eds., 1992; etc.). However, most nonlinear wave interactions share an important problem that lends itself to be tackled now, as an application of what we developed in the previous section. For two waves to reach a signiﬁcant eﬃciency in a nonlinear interaction, they must travel with the same phase velocity. Otherwise, local contributions might still be strong, but would not add up coherently, and the average eﬀect over long distances would be essentially zero. Similar considerations apply to nonlinear processes involving three or more waves. Satisfying such constraints among interacting waves is referred to as phase matching. We will provide an example, in a birefringent medium, showing that phase matching is achieved only if the waves travel along selected directions. For simplicity, let us deal with a uniaxial crystal, and leave to the reader, as an exercise, biaxial ones. So, we let† n20 = x = y ; n2z = z . Suppose that, over the spectral range of interest, n0 and nz are both monotonic functions of frequency. Then, if f1 and f2 are the frequencies of two waves involved in a given nonlinear process, we have, inevitably, n0 (f1 ) = n0 (f2 ), and nz (f1 ) = nz (f2 ). Phase matching between two ordinary waves is not possible, regardless of their direction of propagation, neither it is possible between two extraordinary waves. One must try to phase match an ordinary wave and an extraordinary one. Let ϑ be the angle between the direction along which these two waves propagate, z , and the optic axis, z. We are looking for a value of ϑ such that the following equation: n0 (f1 ) = ne (f2 )

,

(6.41)

is satisﬁed, ne being the extraordinary index along z . To that purpose, let us express ne as a function of n0 , nz and ϑ, starting from Fresnel’s equation of the wave normal. To avoid a 0/0 form, let us ﬁrst rationalize Eq. (6.34), and then make use of the following symbols (see Figure 6.4): vx = sin ϑ ,

vy = 0 ,

vz = cos ϑ .

(6.42)

After simple algebra, we get: 1 = ne

cos2 ϑ sin2 ϑ + n20 n2z

1/2 .

(6.43)

† While n0 is called ordinary index without any ambiguity, the term “extraordinary index” is used by some authors to denote what we labelled here as nz , by other authors to denote the quantity (6.43). In this book, we will adhere consistently to the latter rule.

© 2006 by Taylor & Francis Group, LLC

AN APPLICATION: PHASE MATCHING OF TWO WAVES

199

Figure 6.4 Relationship between the propagation direction, z , and the optic axis, z.

So, the condition in Eq. (6.41) becomes an equation in ϑ which reads: 1 1 1 1 − 2 − sin2 ϑ 2 = . (6.44) n0 (f2 ) nz (f2 ) n20 (f2 ) n20 (f1 ) A real solution for ϑ, i.e., a direction of propagation satisfying Eq. (6.41), exists if the two quantities in brackets have the same sign, and if the one on the left-hand side is larger, in absolute value, then the one on the right-hand side. Hence, there are solutions to the problem either for: n0 (f2 ) > n0 (f1 ) > nz (f2 )

,

(6.45)

n0 (f2 ) < n0 (f1 ) < nz (f2 ) .

(6.46)

or for: In fact, in our formalism the two symbols f1 and f2 are completely interchangeable. (They are no longer interchangeable only if at a certain frequency, between f1 and f2 , the inequality between n0 and nz is reversed). So, the actual meaning of Eqs. (6.45) and (6.46) is that, given n0 and nz at one frequency (say f1 ), phase matching is possible at all those frequencies f2 at which n0 (f2 ) is within the range between n0 (f1 ) and nz (f1 ). The polarizations of the two waves, on the other hand, cannot be interchanged. It is easy to prove, as an exercise, that the extraordinary wave must be the higher frequency, if n0 is an increasing function of frequency f , and Eq. (6.45) applies; the lower frequency, if n0 is still an increasing function of frequency f , but Eq. (6.46) applies. The reverse is true if the ordinary index is a decreasing function of frequency. A numerical example may help clarifying how one proceeds. Let us deal with second-harmonic generation in KDP (Deuterated Potassium Phosphate), one of the most widely used materials in bulk integrated optics. It is a uniaxial crystal. Suppose that the light source is a ruby laser, emitting in the red, at λ1 = 694 nm. The second-harmonic wavelength is then λ2 = λ1 /2 = 347 nm, in the green. The measured values for the refractive indices can be found in

© 2006 by Taylor & Francis Group, LLC

200

PLANE WAVES IN ANISOTROPIC MEDIA Table 6.1 Typical values of KDP refractive indices at various wavelengths (T = 25◦ C). λ [nm] 266 355 532 694 1064

n0

nz

1.5593 1.5311 1.5123 1.5050 1.4938

1.5099 1.4858 1.4705 1.4653 1.4599

Typical values of KDP refractive indices at T = −100◦ C. λ [nm] 266 355 532 694 1064

n0

nz

1.5651 1.5363 1.5172 1.5097 1.4985

1.5133 1.4890 1.4736 1.4685 1.4630

the literature, and are: n0 (f1 ) = 1.506

,

nz (f1 ) = 1.466

,

n0 (f2 ) = 1.534

,

nz (f2 ) = 1.487 .

We see that n0 (f1 ) falls in the range between n0 (f2 ) and nz (f2 ). Hence, an ordinary wave at the fundamental frequency can be phase matched to an extraordinary second-harmonic wave. To ﬁnd the direction in which the two waves have to travel, we take Eq. (6.43) and rewrite it as: −1/2 cos2 ϑ 1 − cos2 ϑ + . 1.506 = (1.534)2 (1.487)2 The solution is easily found to be ϑ 50◦ . One of the features of birefringent media that the previous example is intended to point out is that, usually, the refractive indices and the birefringence (i.e, the diﬀerence between indices) depend signiﬁcantly on wavelength. They also depend strongly on temperature. Table 6.1 stresses this message, showing, as an example, ordinary and extraordinary indices in KDP at a few selected wavelengths and at two temperatures. On the other hand, Table 6.2 shows a list of birefringent crystals of practical interest. Because of what was pointed out just now, the ﬁgures in the table are indicative, and are meant to give the reader just a feeling of the orders of magnitude of the indices and of the birefringence. To solve actual technical problems, the reader should look for more detailed data vs. wavelength and vs. temperature, which can be found in many books and handbooks (e.g., Yariv and Yeh, 1984; Dmitriev et al., 1991; Tosco, Ed., 1990).

© 2006 by Taylor & Francis Group, LLC

GYROTROPIC MEDIA

201

Table 6.2 Some typical birefringent crystals. Uniaxial: positive birefringence Quartz Zircon Rutile ZnS

n0 1.54 1.92 2.62 2.35

nz 1.55 1.97 2.90 2.36

1.52 1.51 1.66 2.19 2.29 2.42

1.48 1.47 1.49 2.18 2.20 2.37

ny 1.526 1.582 1.620 1.938

nz 1.530 1.588 1.627 1.947

negative birefringence (ADP) (NH4 )H2 PO4 (KDP) KH2 PO4 Calcite LiTaO3 LiNbO3 BaTiO3 Biaxial Feldspar Mica Topaz YAlO3

6.6

nx 1.522 1.552 1.619 1.923

Gyrotropic media

An anisotropic dielectric is referred to as gyrotropic, if there is a real rectangular reference frame, (x1 , x2 , x3 ), where the dyadic c is represented by a matrix of the form: ⎞ ⎛ 11 12 0 (6.47) (c ) = ⎝ −12 11 0 ⎠ . 0 0 33 This deﬁnition makes use of a particular reference frame, and this may appear as a weak point. Indeed, we might replace it with other deﬁnitions based on intrinsic properties. The reason why this one was preferred, is because, in the cases of practical interest, physical considerations make it quite evident that one direction is “special” for that medium. Then, we choose it as the x3 axis of the reference frame. On the other hand, all the directions orthogonal to the special one are interchangeable. In fact, let us show that the matrix in Eq. (6.47) is invariant under any rotation of the reference frame around the x3 axis. Such a rotation does not aﬀect, obviously, the x3 components, while its eﬀects in the (x1 , x2 ) plane are described by the sub-matrices: cos ϑ sin ϑ cos ϑ − sin ϑ −1 (M ) = , (M ) = , (6.48) − sin ϑ cos ϑ sin ϑ cos ϑ

© 2006 by Taylor & Francis Group, LLC

202

PLANE WAVES IN ANISOTROPIC MEDIA

where ϑ is the angle of rotation around x3 , in the sense that in the new frame the upper-left (2 × 2) submatrix of Eq. (6.47) is replaced by: 11 12 11 12 −1 (M ) (M ) = , (6.49) −12 11 −12 11 whence we draw the conclusion that Eq. (6.47) does not change. In short, the choice of x1 and x2 , on the plane normal to x3 , is completely arbitrary. The examples given in the following sections will enlighten all these statements, and provide them with a stronger physical meaning. An anisotropic magnetic medium such that, in a real rectangular reference frame, its dyadic μ is represented by a matrix with the same features as Eq. (6.47), is also called gyrotropic, or rather gyromagnetic. Most of what we say in this section on dielectric gyrotropic media applies in a straightforward way to gyromagnetic media as well. In Section 6.9 we will discuss an example of such a medium. Combining Eq. (6.47) with Eq. (6.10), i.e., the condition (c ) = (c )† for an anisotropic medium to be lossless, we see that in a lossless gyrotropic medium all the diagonal elements of the matrix in Eq. (6.47) are real, whereas all the oﬀ-diagonal ones are imaginary: 11 = ∗11

,

33 = ∗33

,

12 = −∗12

.

(6.50)

In this section, similarly to what we did in the previous ones, we will focus on lossless media, gaining in simplicity without losing too much in generality. We will study plane waves which travel along one of the axes of the reference frame where Eq. (6.47) applies. We will postpone propagation in an arbitrary direction to the next section. First, let us see whether uniform plane waves can propagate in a direction orthogonal to x3 , for example along x1 . These results will apply, of course, to any direction in the (x1 , x2 ) plane, as a consequence of Eq. (6.49). By deﬁnition a1 ) a ˆ1 , of uniform plane waves, it must be ∇2 E = −k 2 E and ∇(∇·E) = −k 2 (E·ˆ where a ˆ1 is the unit vector of the x1 axis. Thus, Eq. (6.12) reads now:

ˆ1 ) a . (6.51) ˆ1 ω 2 μ · E = k 2 E − (E · a Note that this equation is reminiscent of Eq. (6.30); this is not a pure coincidence. Writing now Eq. (6.51) by components, we reach easily the conclusion that it has two solutions, namely: E = E3 a ˆ3 12 E = E2 − a ˆ1 + a ˆ2 11

, ,

k 2 = ω 2 μ 33 , 2 k 2 = ω 2 μ 11 + 12 11

.

(6.52)

This shows that there are two possible states of polarization, and that their phase velocities are diﬀerent. At ﬁrst sight, this is similar to birefringent media (see Section 6.3), but there is a subtle diﬀerence, as one of the allowed states of polarization (the one in the (x1 , x2 ) plane) is elliptical, since 12 is imaginary.

© 2006 by Taylor & Francis Group, LLC

THE APPLETON-HARTREE FORMULA

203

Furthermore, note that this wave can be evanescent, and actually is evanescent for |11 | < |12 |. When the phase vector k is parallel to x3 axis, the picture is completely diﬀerent. The matrix representation of Eq. (6.51) reads then: 11 12 E1 E1 ω2 μ = k2 . (6.53) −12 11 E2 E2 The characteristic equation of the matrix in Eq. (6.53) is: (11 − κ)2 = −212

,

(6.54)

whose solutions are: (6.55) κL,R = 11 ± j 12 . Because of the conditions in Eq. (6.50), or, equivalently, since the matrix () is Hermitian, both eigenvalues of (t ), in Eq. (6.55), are real. The corresponding normalized eigenvectors: 1 1 1 1 (L ) = √ , (R ) = √ (6.56) 2 j 2 −j represent the clockwise and counterclockwise rotating √ unit-vectors, respecˆ = 1/ 2 (ˆ ˆ = tively, well known from Chapter 2, namely L a1 + jˆ a2 ) and R √ a1 − jˆ a2 ). The net result is that uniform plane waves may travel along 1/ 2 (ˆ the x3 axis, provided they are circularly polarized. Their propagation constants are given by: 2 kL

= ω 2 μ(11 + j12 )

,

(6.57)

2 kR

= ω 2 μ(11 − j12 )

.

(6.58)

These results entail very important consequences, to be discussed in Section 6.10. 6.7

The Appleton-Hartree formula

Continuing along the guidelines of the previous section, let us deal now with waves traveling in an arbitrary direction, neither x3 nor orthogonal to it. It may be helpful to change reference frame. We take new rectangular coordinates, y1 , y2 , y3 , with the y3 axis parallel to the wave vector, and with the medium preferential direction x3 lying in the (y1 , y3 ) plane. Let 1 , 2 (= 0), and 3 = (1 − 21 )1/2 be the direction cosines of the “old” x3 axis with respect to the new axes yi . Then, the reader may check, as an exercise, that in the new frame the dyadic is represented by the following matrix: ⎞ ⎛ 2 3 12 1 3 (33 − 11 ) 3 11 + 21 33 ⎠ . ⎝ (6.59) 11 1 12 −3 12 2 2 1 3 (33 − 11 ) −1 12 1 11 + 3 33 We saw in Section 6.3 that for a uniform plane wave in a birefringent medium the vector D is orthogonal to y3 , and belongs to the plane deﬁned

© 2006 by Taylor & Francis Group, LLC

204

PLANE WAVES IN ANISOTROPIC MEDIA

Figure 6.5 The electric ﬁeld vector and the electric displacement vector in the (y1 , y2 , y3 ) reference system.

by y3 and by the vector E. This result still holds in the case at hand, as one can prove as an exercise. In the reference frame {yi }, we can express these statements through the following formulas (see Figure 6.5): D3 = 0 , D2 E2 = = −jp D1 E1

,

(6.60)

where p is the linear polarization ratio of the D ﬁeld, as deﬁned in Chapter 2. Combining Eq. (6.60) with the constitutive relation D = · E (written by components using the matrix in Eq. (6.59)), we get the following seconddegree equation in p: 3 12 33 p2 + j21 (11 33 − 211 − 212 ) p − 3 12 33 = 0 .

(6.61)

Solving Eq. (6.61) we get the polarizations of the two uniform plane waves which can propagate in the y3 direction. Consider a lossless gyrotropic medium (i.e., 12 imaginary and 11 , 33 real). Then, when both Eqs. (6.57) and (6.58) turn out to be positive (i.e., if |12 | < 11 ), Eq. (6.61) divided by the imaginary unit has real coeﬃcients, and a real and positive discriminant, which automatically entails two real roots. From the properties of the polarization ratio p discussed in Chapter 2, we conclude then that the axes of the polarization ellipse coincide with y1 and y2 . When the medium is lossy, the last points are no longer true. However, an interesting property can be derived in any case, as follows: the ﬁrst coeﬃcient and the last one in Eq. (6.61) are equal and opposite; hence, the product of the two roots of the equation has to be −1: p1 · p2 = −1 .

(6.62)

Again because of the properties of p seen in Chapter 2, Eq. (6.62) entails that the two corresponding time-dependent vectors D ∼ describe ellipses which

© 2006 by Taylor & Francis Group, LLC

THE APPLETON-HARTREE FORMULA

205

Figure 6.6 Schematic illustration of the two allowed states of elliptical polarization.

are mirror images of each other, with respect to the line y1 = y2 , i.e., the straight line at 45 degrees between the y1 and y2 axes (see Figure 6.6). We can derive now an expression for the refractive index, deﬁned always through Eq. (6.26). To this purpose, rewrite Eq. (6.27) as E1 D1 (6.63) = n2 0 , D3 = 0 . D2 E2 Next, write D1 , D2 and D3 (= 0) as functions of E1 , E2 and E3 , using the matrix in Eq. (6.59). The third equation obtained in this way allows us to express E3 as a function of E2 and E1 , namely:

1 ( . (6.64) − ) E + E E3 = 2 11 33 1 3 1 1 12 2 1 11 + 23 33 If we insert this into the expressions for D1 , and eliminate E2 by means of Eq. (6.60), we end up with the following general expression for n2 : n2 =

33 11 + j3 12 p 0 21 11 + 23 33

.

(6.65)

Replacing p with −1/p thanks to Eq. (6.62), we get n2 =

33 3 12 + j11 p 0 jp(21 11 + 23 33 )

.

(6.66)

If p is real (lossless medium, 12 imaginary), then n2 is real. If we suppose j3 12 p < 0, then, Eq. (6.66) is always positive, while Eq. (6.65) is either positive or negative, depending on whether 11 > 3 |12 p|, or 11 < 3 |12 p|. However, the complete opposite occurs if we suppose j3 12 p > 0. When n2 is negative, we have an evanescent wave, whose physical meaning will be outlined in Section 6.10. In a lossy medium, n2 is complex. Hence, any wave traveling in such a medium undergoes an attenuation.

© 2006 by Taylor & Francis Group, LLC

206

PLANE WAVES IN ANISOTROPIC MEDIA

What we said concerning waves traveling in an arbitrary direction in a gyrotropic medium is of practical interest to radio engineers, as it applies to ionospheric propagation, where the preferential direction x3 is that of the terrestrial magnetic ﬁeld. The direction of propagation, y3 , and the direction x3 deﬁne a plane (in our notation, it is the y2 = 0 plane) which is usually referred to as the “magnetic meridian plane”. The form taken by Eq. (6.65) or by Eq. (6.66), when one inserts values for ij which are typical of the ionosphere, is known in the literature as the Appleton-Hartree formula (Budden, 1961). 6.8

An example of permittivity dyadic

An e.m. wave traveling in an ionized gas is always accompanied by a convection current, because of the motion of electrons and, to a smaller extent, of positive ions, although they are much heavier than electrons. In this section we will see how, under suitable assumptions, this convection current can be accounted for, in a complex permittivity dyadic. The equation of motion for an individual electron in an e.m. ﬁeld, driven by the Lorentz force in Eq. (1.2), neglecting collisions, is: dv ∼

= −e(E (6.67) ∼+v ∼×B ∼) , dt where m = electron rest mass (= 9.1091 10−31 kg), e = electron charge magnitude (= 1.6021 10−19 C), v ∼ = instantaneous velocity. For a uniform plane wave in a medium whose intrinsic impedance does not diﬀer too much from that of free space, the ratio |B|/|E| is so small that, for non-relativistic particles (i.e., for | v ∼ | c), the second term on the right-hand side of Eq. (6.67) is negligible, compared to the ﬁrst one. From now on, this term will be omitted. However, it happens often, in plasma physics laboratories as well as in nature in the ionosphere, that an ionized gas is placed in a strong static magnetic ﬁeld, whose induction vector will be denoted as B0 . In the presence of this B0 and in the absence of a time-harmonic ﬁeld (i.e., for E ∼ = 0), Eq. (6.67) becomes the equation of motion along a helix, whose axis is parallel to the ﬂow lines of B0 . It is rather intuitive that propagation of a wave in such a medium is strongly anisotropic, with signiﬁcant diﬀerences between the direction of B 0 and all the orthogonal ones. Let us assume, for simplicity, that B0 is constant throughout the region of interest, and introduce rectangular coordinates with the x3 axis parallel to B0 . From Eq. (6.67) one sees that for B0 → ∞ the radius of curvature of the helical electron trajectories tends to zero. So, let us ﬁrst make a simpliﬁed analysis under the assumption B0 = ∞, which forbids the electrons to move in the plane transverse to x3 so that all quantities of interest depend only on x3 and time. In the time domain, the convection current density can then be written as: m

J =ρ v , ∼3 ∼ ∼3

© 2006 by Taylor & Francis Group, LLC

(6.68)

AN EXAMPLE OF PERMITTIVITY DYADIC

207

where ρ is the instantaneous electron charge density.‡ The electrons may move ∼ with a nonvanishing average velocity, v0 , referred to as the drift velocity. In the rest of this section, we will use a subscript 0 for the dc components, and a subscript a for the ac components. Then, Eq. (6.68) reads: J0 + Ja = (ρ0 + ρa) (v0 + va) . ∼ ∼ ∼

(6.69)

Clearly, J0 = ρ0 v0 . Assuming that ρa and va are both time-harmonic, with an ∼ ∼ angular frequency ω, then the time-dependent part contains terms at angular frequencies ω and 2ω. Assuming that the latter are negligibly small, we may write: (6.70) J ∼ = ρ0 va + ρa v0 . ∼a ∼ ∼ Our next, time-consuming task is to derive from Eq. (6.70) how the ac electric ﬁeld is related to the ac current density Ja. In fact, to link va with the ∼ ∼ electric ﬁeld, we note ﬁrst that, if we adopt a ﬂuid-dynamics viewpoint, we may write, in general: ∂ v3 ∂ v3 d v3 ∂ v3 ∂ v3 ∂x3 ∼= ∼+ ∼ = ∼ + v3 ∼ ∼ ∂x3 dt ∂t ∂x3 ∂t ∂t

.

(6.71)

The last term, a product of two quantities at ω, can be neglected as we did a moment ago with ρava. Introducing then new symbols (capital letters) such ∼∼ that va= Re(Va ejωt ), etc., the left-hand side of Eq. (6.67) can be replaced by: ∼ ∂Va m jω Va + v0 . (6.72) ∂x3 Assume now that all ac quantities depend on x3 as exp(−κx3 ), where κ is still unknown. Using calligraphic capital letters to indicate that Va = Va e−κx3 , etc., Eq. (6.67) becomes: e (6.73) (jω − κv0 ) Va = − E3 . m To reach our ﬁnal goal, we need now a suitable expression for ρa, to be used ∼ in Eq. (6.70). We get it from the continuity Eq. (1.29), namely: ∂ρ ∼ ∇· J ∼ = − ∂t which, under our assumptions, yields: κJa = −jω Pa

,

(6.74)

,

(6.75)

where ρa= Re{Pa ejωt e−κx3 }. Re-writing now Eq. (6.70) for the complex ∼ ‡ This analysis neglects electrostatic forces. This implies an underlying assumption, namely that the ionized gas, on average, is neutral, i.e., the average ion charge density is equal, in absolute value, to the electron charge density. When this assumption is not veriﬁed, as in a particle beam, then the results that we will ﬁnd here hold only as a ﬁrst approximation.

© 2006 by Taylor & Francis Group, LLC

208

PLANE WAVES IN ANISOTROPIC MEDIA

quantities represented by capital letters, and using for Va and Pa the expressions which follow from Eqs. (6.73) and (6.75), we get: ρa e Ja = −jω E3 . (6.76) m(jω − κv0 )2 Deﬁning the quantity: ωp =

N e2 0 m

1/2 ,

(6.77)

which is referred to as the plasma angular frequency, Eq. (6.76) becomes: Ja = −jω 0

ωp2 E3 (jω − κv0 )2

.

(6.78)

From our initial assumption that any electron motion perpendicular to x3 is forbidden, it follows that in the planes orthogonal to x3 the medium behaves like free space. So, we may write: ⎞ ⎛ 0 0 0 (c ) = ⎝ 0 0 0 ⎠ , (6.79) 0 0 33 with:

33 = 0

ωp2 1− (ω + jκv0 )2

.

(6.80)

In these expressions, ρava has been neglected with respect to (ρ0 va + ρa ∼∼ ∼ ∼ v0 ), and Va ∂Va /∂z has been neglected with respect to (jωVa + v0 ∂Va /∂z). Consequently, when v0 = 0, i.e., for a plasma at rest, these formulas are correct only if ρ0 ρa and |ω/κ| Va , i.e., if the wave phase velocity is large, ∼ compared with the electron speed. Now let us remove the assumption B0 = ∞, and take into account, in the electron motion, the components orthogonal to B0 . The starting point is again Eq. (6.67), where we still neglect the contribution v ∼×B ∼, due to the ac component of B . Eq. (6.68) is replaced by the complete vector equation: ∼ J (6.81) ∼=ρ ∼v ∼ . Assume, for simplicity, that the average velocity is equal to zero (v0 = 0). Then, the ac component of Eq. (6.81) can be written (neglecting, as usual, products of ac components on the right-hand side), as: (6.82) J = ρ 0 va . ∼a ∼ Suppose now that a uniform plane wave propagates in this medium, in an arbitrary direction. Then, all complex quantities representing time-harmonic functions are products of constants times a function of spatial coordinates of the type: (6.83) exp(−κ · r) = exp(−κ1 x1 − κ2 x2 − κ3 x3 ) .

© 2006 by Taylor & Francis Group, LLC

AN EXAMPLE OF PERMITTIVITY DYADIC

209

Under these circumstances, the left-hand side of Eq. (6.67) can be written as: m (jω − κ1 Va1 − κ2 Va2 − κ3 Va3 ) V a . (6.84) If the particle velocities are small compared to the wave phase velocity, then Eq. (6.84) can be approximated as: jω m V a

.

(6.85)

In the same reference frame as before (x3 parallel to B0 ), Eq. (6.67) may be written by components in a matrix form as: ⎛ ⎞ eB0 ⎛ ⎞ ⎛ ⎞ jω 0 E1 ⎜ ⎟ Va1 m ⎜ ⎟ e ⎜ ⎟ ⎜ −eB0 ⎟⎜ ⎟ (6.86) ⎜ ⎝ Va2 ⎠ = − ⎝ E2 ⎠ . jω 0 ⎟ ⎜ m ⎟ m ⎝ ⎠ V E3 a3 0 0 jω Inverting the matrix in Eq. (6.86), and setting: e B0 , (6.87) ωc = m (this quantity is referred to as the cyclotron angular frequency), we get: ⎛ ⎞ ω jωc 0 ⎜ ωc2 − ω 2 ωc2 − ω 2 ⎟⎛ ⎞ ⎛ ⎞ ⎜ ⎟ E1 Va1 ⎜ ⎟ ⎜ ⎟⎜ ⎟ ω e ⎜ −jωc ⎜ ⎟ ⎟ ⎝ E2 ⎠ . 0 V = −j (6.88) ⎝ a2 ⎠ 2 2 2 2 ⎜ ⎟ ωc − ω m ⎜ ωc − ω ⎟ ⎜ ⎟ Va3 1 ⎠ E3 ⎝ 0 0 − ω Going back to Eq. (6.82), and from this to Maxwell’s equations, we ﬁnally ﬁnd the dyadic c we were looking for. In our reference frame, it is represented by a matrix of the type in Eq. (6.47) with the following elements: ωp2 11 = 0 1 + 2 , ωc − ω 2 12

= −21 = j

33

= 0 1 −

ωp2 0 ωc 2 ω ωc − ω 2

ωp2 ω2

.

,

(6.89)

Note that the denominators in 11 , 12 and 21 , all vanish if ω = ωc . The medium has a resonant behavior when the angular frequency of the e.m. wave equals the angular velocity of the helical electron motion around the ﬂow lines of the vector B0 .

© 2006 by Taylor & Francis Group, LLC

210

PLANE WAVES IN ANISOTROPIC MEDIA

It is left to the reader as an exercise to show that, under suitable conditions, the medium exhibits an evanescence, as brieﬂy mentioned in Section 6.7. The reader should also elaborate on the physical relationship between this evanescence and the frictionless motion of charged particles. 6.9

Second example of permeability dyadic

In this section, we will be obliged to use concepts, terminology and notation taken from magnetostatics, a discipline which is not dealt with in this book. We take it for granted that the reader had been exposed to these notions in basic physics courses. An interest in refreshing them can easily be satisﬁed by consulting classical textbooks on static ﬁelds (e.g., Plonsey and Collin, 1961). In the so-called semiclassical treatment of ferromagnetic media (i.e., if a minimal amount of Quantum Mechanics is invoked), the magnetic dipole moment due to the individual electron spin, m ∼ , is looked at as the result of a rotation of the electron around its axis, with an angular momentum, Q ∼, proportional to m . In formulas: ∼ m ∼=gQ ∼ ,

(6.90)

where g is called the gyromagnetic constant of the electron (g = −1.76 × 1011 T/s). When immersed in a magnetic induction ﬁeld B ∼, the dipole moment m experiences a torque: ∼ τ =m ×B ∼ ∼ ∼

.

(6.91)

Hence, the dynamics of the vector m ∼ obeys the equation: dm ∼ dt

=gm ∼×B ∼

.

(6.92)

In a saturated magnetic material, if we neglect, as a ﬁrst approximation, demagnetizing factors related to the shape of the material itself, we may write: B ∼ = μ0 (H ∼×M ∼ ) = μ0 (H ∼ +N0 m ∼)

,

(6.93)

where N0 is the spin moment density (number per unit volume). Multiplying both sides of Eq. (6.92) times N0 , and using Eq. (6.93), we get: dM ∼ dt

= −g μ0 H ∼×M ∼ .

(6.94)

This is easily recognized as the equation of a precession around the vector H ∼. If the H ﬁeld has a dc component, called orienting ﬁeld, and if this com∼ ˆ3 , is parallel to the x3 axis, then, in the saturation regime, the ponent, H0 a ˆ3 , and use static component of M ˆ3 . We will call it M0 a ∼ is also parallel to a a subscript a to denote complex quantities which represent ac components.

© 2006 by Taylor & Francis Group, LLC

FARADAY ROTATION

211

From Eq. (6.94), we then get: ⎛ ⎞ ⎛ ⎞⎛ ⎞ jω −g μ0 H0 0 −Ha2 Ma1 ⎜ ⎟ ⎜ ⎟⎜ ⎟ jω 0 ⎠ ⎝ Ma2 ⎠ = g μ0 H0 ⎝ Ha1 ⎠ ⎝ g μ 0 H0 0

0

jω

Ma3

.

(6.95)

0

Solving this for the vector Ma , and replacing it into: Ba = μ0 (Ha + Ma ) ≡ μ · Ha

,

(6.96)

we get, in our reference frame, a matrix representation for the dyadic μ, whose elements are: ω0 ωM , μ11 = μ22 = μ0 1 + 2 ω0 − ω 2 μ12 μ33 = μ0

= −μ21 = j ,

μ0 ωωM ω02 − ω 2

,

μ13 = μ23 = μ31 = μ32 = 0 ,

(6.97)

where we set: ω0

= −g μ0 H0

,

ωM

= −g μ0 M0

.

(6.98)

The minus signs are due to the fact that g < 0 for an electron. Eqs. (6.97) show that, according to the deﬁnition given in Section 6.6, a magnetized ferrite is indeed a gyrotropic medium. The angular frequency ω 0 plays a role which is very similar to that of the cyclotron angular frequency ω c in a gyrotropic plasma. Physically, in fact, both these quantities are related to angular velocities (of rotation around B0 for ωc , of precession around H0 for ω0 ). Eqs. (6.97) exhibit a resonant behavior centered at ω = ω0 , like Eqs. (6.89). Examples of actual microwave devices which exploit gyromagnetic media, as well as of very similar optical devices, will be given at the end of the next section. 6.10

Faraday rotation

Some features of plane waves traveling along the preferential direction, x3 , in a gyrotropic medium, were derived in Section 6.6. However, the subject was left open for further discussion. We will see in this section that those waves provide a very interesting example — probably, the example which is most widely used in practice — of nonreciprocal electromagnetic behavior. Let us start oﬀ with two preliminary remarks, restricting ourselves, once again, to lossless media. A) As a matter of argument, assume that 12 is imaginary and positive. Then,

© 2006 by Taylor & Francis Group, LLC

212

PLANE WAVES IN ANISOTROPIC MEDIA

the right-hand side of Eq. (6.58) is always positive. The right-hand side of Eq. (6.57) is positive or negative, depending on whether 11 > |12 |, or 11 < |12 |. In the latter case, the root kL is imaginary: the corresponding wave undergoes an attenuation along x3 , while its phase does not change in this direction. The medium is assumed to be lossless, so an attenuation cannot be attributed to power dissipation. It must originate from an evanescence, similarly to what we saw in Chapter 4. In general, 11 and 12 are frequency dependent. So, if ω is changed, one can pass from a situation where both circularly polarized waves are uniform plane waves, to a situation where one of them is evanescent, while the other one is still uniform. B) The terminology “clockwise” or “counterclockwise,” that we adopted in Section 6.6 for the rotation of the instantaneous vectors represented either by ˆ or by L, ˆ Eqs. (6.56), refers to an observer looking at the positive face of the R x3 = 0 plane, regardless of whether the wave is traveling in the positive or negative sense of the x3 axis. This entails that: • as long as we deal with waves traveling in the sense of increasing x3 , the wave which rotates like a left-handed screw has a propagation constant equal to the positive square root of Eq. (6.57), the wave which rotates like a right-handed screw has a propagation constant equal to the positive square root of Eq. (6.58); • when we deal with waves traveling in the sense of decreasing x3 , the two propagation constants are interchanged, i.e., the negative square root of Eq. (6.58) for the left-handed screw, the negative square root of Eq. (6.57) for the right-handed screw. It is precisely this exchange between propagation constants when the sense in which the waves travel is reversed, that is responsible for nonreciprocity, as we will see in a shortwhile. Suppose now that, at the frequency of interest, kL and kR are both real. Let a uniform plane wave impinge on a plane orthogonal to x3 , for example on the x3 = 0 plane, traveling in the direction of increasing x3 , and assume that its electric ﬁeld is linearly polarized along a direction lying in the x3 = 0 plane, say x1 (Figure 6.7). This vector E can be written by components in the basis Eq. (6.56), which consists of the eigenvectors of the matrix () and therefore is easy to track as we move through the gyrotropic medium. In fact, the electric ﬁeld at any value of x3 can be simply written as: E0 ˆ −jkR x3 E0 ˆ −jkL x3 e e +√ R E= √ L 2 2

.

(6.99)

The vector in Eq. (6.99), sum of two circularly polarized vectors which are equal in modulus (for any x3 ), remains (see Chapter 2) linearly polarized everywhere. However, at a generic x3 , its direction no longer coincides with the input one. It is at an angle ϑ, with respect to the x1 axis, that can be

© 2006 by Taylor & Francis Group, LLC

FARADAY ROTATION

213

Figure 6.7 Schematic illustration of Faraday rotation.

found if we rewrite Eq. (6.99) in the form: kL +kR kL − kR kL − kR x3 + a x3 ˆ1 cos ˆ2 sin e−j 2 E = E0 a 2 2

x3

.

(6.100) From this, we ﬁnd: ϑ = arctan

E2 kL − kR x3 = E1 2

.

(6.101)

So, we see that the direction of polarization is rotated by an angle proportional to x3 , the proportionality factor (sometimes referred to as circular birefringence) being (kL −kR )/2 = 0 for 12 = 0, i.e., for a gyrotropic medium. This phenomenon is known as the Faraday rotation. As kL and kR are interchanged when propagation is reversed (see the previous comment B) this rotation is peculiar in that the sign of the rotation angle (with respect to a ﬁxed reference frame) does not change when the direction of propagation is reversed. Consequently, let us write, rather than Eq. (6.101): ϑ=

kL − kR |x3 | 2

.

(6.102)

This implies that the reciprocity theorem (Section 3.4) does not apply to the medium at hand. To prove this statement, let us deal with two sets of sources, satisfying the following requirements. The ﬁrst set generates a wave which travels in the positive x3 direction, and whose electric ﬁeld equals Eq. (6.99) on the x3 = 0 plane. The second set generates a wave which travels in the negative x3 direction, and whose electric ﬁeld on the plane x3 = equals Eq. (6.99) in amplitude, but is polarized at an angle ϑ0 = (kL − kR )/2 with respect to the x1 axis. Let us compute the reactions of these two ﬁelds on the sources, according to the deﬁnition given in Section 3.4. It is easily seen that, except for the very special cases where ϑ0 is an integer multiple of 2π, the two reactions are not equal. This proves, as we intended to do, that we are facing a nonreciprocal phenomenon.

© 2006 by Taylor & Francis Group, LLC

214

PLANE WAVES IN ANISOTROPIC MEDIA

Figure 6.8 Illustration of a microwave isolator.

Faraday rotation is widely exploited in practice, especially at microwave and optical frequencies. One of the simplest nonreciprocal devices, and at the same time one of the most useful ones, is the so-called isolator. Its function is to introduce a strongly nonreciprocal attenuation — ideally, no attenuation in one direction, an inﬁnite attenuation in the opposite one — aiming at making an oscillator insensitive to any change in its load. The most common types of isolators (although not the only ones) are based on Faraday rotation. Figure 6.8 provides a schematic illustration of its waveguide version at microwave frequencies, Figure 6.9 shows an optical implementation. The two are much more similar in their essence than at ﬁrst sight. Although the ﬁrst one was developed several decades before the second, we will brieﬂy comment only on the latter one, because understanding some features of the ﬁrst requires concepts to be developed in Chapter 7. The reader should come back to it after the chapter on conducting-wall waveguides. The optical isolator is a Faraday rotator (whose gyromagnetic medium is, usually, a garnet crystal), designed in such a way that the rotation angle ϑ equals π/4, and placed between two linear polarizers that are rotated with respect to each other also by π/4. As illustrated schematically in the ﬁgure, a backward wave is strongly attenuated by polarizer A, while a forward wave passes with very low attenuation through polarizer B. Another Faraday-rotation-based nonreciprocal device which is widely used at microwave frequencies is the so-called circulator. It is an n-port junction, with n ≥ 3, deﬁned, in its ideal version, as follows. When a wave impinges on the i-th port, all the incident power comes out from the i + 1-th port (understanding that port n + 1 is the same as port 1). Nothing is reﬂected, or transmitted to the remaining “decoupled” n − 2 ports. Figure 6.10 provides a schematic illustration of a 3-port circulator in a waveguide conﬁguration (part a), and the symbol used for it in circuit diagrams (part b). A qualitative explanation of how a Y -junction 3-port circulator operates may proceed as follows. Because of the conducting walls in the planes orthogonal to the static induction ﬁeld, the clockwise and counterclockwise waves of the previous analysis are no longer traveling, but standing waves. Their propagation constants along the vertical axis remain diﬀerent. As they are not in an unbounded medium, but in a closed environment, this diﬀerence in longitudinal propagation constant entails, through Maxwell’s equations, that their azimuthal phase velocities (i.e., the velocities at which they rotate around the static ﬁeld) are diﬀerent too. Then, a suitable design can make the two

© 2006 by Taylor & Francis Group, LLC

FARADAY ROTATION

215

Figure 6.9 Illustration of an optical isolator.

Figure 6.10 A 3-port microwave circulator. (a) View of a waveguide conﬁguration. (b) Symbol in use for circuit diagrams.

waves excited by a ﬁeld entering through port 1 meet in phase at port 2, in counterphase at port 3, whence the conclusion. A typical use of the 3-port circulator is in radars, where it allows the transmitter (connected to port 1) and the receiver (connected to port 3) to share the same antenna (connected to port 2) without “seeing” each other, which would be extremely detrimental since the transmitted power is many orders of magnitude larger than the received one. In addition to its technical relevance, which is illustrated also by some of

© 2006 by Taylor & Francis Group, LLC

216

PLANE WAVES IN ANISOTROPIC MEDIA

the problems at the end of the chapter, Faraday rotation is very important as a concept. It stresses that, if we want to avoid any ambiguity in discussing reciprocity in an anisotropic medium, it is strictly mandatory to use real reference frames, and to deal only with linearly polarized ﬁelds. To clarify the meaning of this statement, let us recall that, as pointed out in Section 6.6, the matrix () of a gyrotropic medium becomes diagonal (and therefore, symmetˆ ric) in a complex reference frame where two of the basis unit vectors are R ˆ Consequently, circularly polarized plane waves traveling in a gyrotropic and L. medium obey equations that are formally identical to those governing uniform plane waves in isotropic media. This might induce one to claim that circularly polarized waves in a gyrotropic medium behave reciprocally, but such a statement does not agree with the deﬁnition of reciprocity, based on the reaction, that we gave in Section 3.4. The origin of this apparent ambiguity is that the reaction, deﬁned through a scalar product between two complex vectors, without taking the complex conjugate of one of its factors, is not an invariant with respect to any change of basis in a complex space. In particular, the reaction does change, in general, when we pass from linearly polarized basis vectors to rotating basis vectors. 6.11

Further applications and suggested reading

This chapter contains only those fundamental notions about anisotropic media, that may be considered essential ingredients, nowadays, in the education of an advanced engineer who intends to work in the area of information technology. The level of knowledge which is necessary in order to become a specialist in propagation in anisotropic media is orders of magnitude larger than the contents of this short chapter. The list of further suggested readings is potentially unlimited. Similarly to what we said at the end of Chapter 4, we ﬁrst stress that linear anisotropic media are very pervasive in physics and in engineering. This statement applies, of course, also to the mathematical formalism which is normally used for their analysis, i.e., tensors and their matrix representations, that is shared by many diﬀerent ﬁelds of application (e.g., Babuska and Cara, 1991; Hijden, 1987; Nayfeh, 1995 and references that treat the propagation of mechanical, e.g., seismic waves in anisotropic media). To broaden one’s own general background in physical sciences, one should at least examine a broad-scope, comprehensive book, like, for instance, Nye (1985), where the mathematical properties of anisotropic media are presented in detail, and then followed by applications, ranging from magnetic and electric properties, to theory of elasticity, from transport phenomena to crystal optics. An extremely important concept that the reader can grasp from such a book is the physical relevance, in some problems, of tensors whose rank is higher than 2 (2 is, of course, the rank of dyadics, that we used in this chapter). Other treatments of crystal optics include Juretschke (1974). For what refers, more speciﬁcally, to further reading on electromagnetic

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

217

waves in anisotropic media, let us mention the following examples. Optics of crystals is the title of a famous chapter in Born and Wolf (1980), where the beauty of the presentation matches the richness of the scientiﬁc contents. Our emphasis is on electrical anisotropy. Materials with magnetic anisotropy are treated in, e.g., Tarling and Hrouda (1993) — see also Negi and Saraf (1989). For a treatment of electromagnetic wave propagation in bianisotropic media, we refer to Kong (1990) and Staelin et al. (1993). Propagation in magnetoplasmas, with emphasis on ionospheric phenomena, is thoroughly dealt with in Budden (1961). Those who are interested in the fundamentals of plasma physics and magnetohydrodynamics can ﬁnd them, e.g., in Jackson (1975), Chandrasekhar (1989), Sitenko and Malnev (1995), and Dendy, Ed., (1993), while Miyamoto (1989) focuses on applications of plasmas in thermonuclear fusion for energy production, and Sturrock (1994) includes material regarding astrophysical and geophysical as well as laboratory plasmas. As we already mentioned in Section 6.5, many among the experiments that opened, in the ’60s, the new ﬁeld of nonlinear optics — in bulk media, at that time — were performed in anisotropic media, birefringence being a crucial ingredient for obtaining phase matching. The fundamentals of nonlinear optics, as well as their links with propagation in anisotropic media, are well covered, e.g., in Bloembergen (1996) — see also Boyd (1992), Butcher and Cotter (1990), Newell and Moloney (1992), Shen (1984), and Yariv and Yeh (1984). Much more recently, thanks to the availability of very-low-loss optical ﬁbers, where interaction lengths could become as large as several kilometers, guidedwave nonlinear optics became much more attractive than its bulk precursor. Fundamentals on nonlinear phenomena in waveguides can be found, e.g., in Agrawal (1995). More speciﬁcally, Someda and Stegeman, Eds., (1992) covers the area where guided-wave propagation overlaps with anisotropy (for example, random birefringence in optical ﬁbers) and with nonlinearities. Propagation problems in anisotropic waveguides are an area where we may state — anticipating something to be seen in Chapter 10 — very few problems can be solved analytically in closed form. Numerical methods become then extremely important; a relevant example is provided by Zoboli and Bassi in Someda and Stegeman, Eds., (1992). The reader should remember what we said in Section 3.9 about symmetries: careful exploitation of spatial symmetries in numerical methods can yield enormous savings in processing time and in memory occupation. Notice, however, that in anisotropic media symmetry operations are legitimate only if the dyadic operators that characterize materials commute with the symmetry operators that characterize the geometry of the region where the ﬁeld is deﬁned. Problems 6-1 When the dyadics c and μ are non-singular, i.e., when both have an inverse, then Eqs. (6.3) and (6.4) can be inverted, expressing the electric and

© 2006 by Taylor & Francis Group, LLC

218

PLANE WAVES IN ANISOTROPIC MEDIA

magnetic ﬁeld vectors as functions of the displacement and induction vectors. Write the corresponding expression of the Poynting vector. Derive conditions under which the factors contained in this expression commute. 6-2 Show that if at least one of the dyadics c and μ is not Hermitian, then the quantity in Eq. (6.8) may have a nonvanishing real part. Find a way to distinguish media where this real part is positive (lossy media) and those where it is negative (media with gain). 6-3 Write Eq. (6.12) by components, in a rectangular reference frame, in a region without imposed currents. 6-4 Show that, in those cases where Eq. (6.12) becomes a Helmholtz equation, the same occurs to Eq. (6.19), provided the ϕ = 0 gauge is adopted. 6-5 Starting from the contents of Section 6.2, and applying the duality theorem of Section 3.7, discuss the case of a magnetic anisotropic medium in terms of an electric vector potential F, with the same physical dimensions as in Section 1.9. 6-6 Using duality, analyze a magnetic medium whose constitutive relations are the dual of those of the birefringent dielectric medium studied in Section 6.3. 6-7 Show that, if a plane wave is launched into a lossless birefringent medium in a generic direction and with a generic state of polarization, then its state of polarization evolves periodically as a function of the distance traveled by the wave. Hence, express the period as a function of the ordinary and extraordinary indices for that direction of propagation. Does the state of polarization still evolve periodically in the presence of loss ? 6-8 In a rutile crystal (a uniaxial dielectric medium whose refractive indices are on the list in Table 6.2), plane waves propagate along the straight line at 45 degrees between the x and z axes, z being the optic axis. Evaluate the values of the ordinary and extraordinary refractive indices. Use both approaches explained in the text: ﬁrst, ﬁnd the eigenvalues of the reduced inverse permittivity dyadic bt . Next, solve Fresnel’s equation of the wave normals as shown in Section 6.5. 6-9 A rutile slab is cut with planar parallel faces, spaced by 10 mm, orthogonal to the y axis. The surrounding medium is air. The incident wave impinges with an incidence angle ϑi = π/4. The plane of incidence is the (x, y)-plane. Applying Snell’s law and the Fresnel formulas of Chapter 4, ﬁnd the directions of propagation of the TE and TM refracted waves (note the socalled double refraction), and the transmission coeﬃcients. What is the angle

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

219

between the directions of propagation of ordinary and extraordinary waves when they emerge in air after the slab ? Is this result valid in general ? 6-10 Find the angles between the phase vectors and the Poynting vectors (ray directions) for the refracted waves within the slab of the previous problem.

6-11 Prove that, in a uniaxial medium where n0 is an increasing function of frequency f , phase matching between fundamental frequency and second harmonic can be achieved only if the extraordinary wave is the second harmonic when Eq. (6.45) applies; only if it is the fundamental frequency when Eq. (6.46) applies. Prove that the reverse is true, if the ordinary index is a decreasing function of frequency. 6-12 One method of fabricating an optical polarizer is to cut a prism of uniaxial medium in such a way that a plane wave, emerging from it towards air, undergoes total internal reﬂection for one polarization (either ordinary, or extraordinary), but not for the other one. Find the range of values for the angle of incidence for which this occurs, both for calcite and for zircon (see Table 6.2). Specify the totally reﬂected wave, in both cases. 6-13 As mentioned brieﬂy in Chapter 2, a quarter-wave plate is a slab of birefringent material which converts a linearly polarized input wave into a circularly polarized output wave. Consider three slabs of mica (see Table 6.2), each of them cut normally to one of the axes x, y, z. The incident wave is linearly polarized at 45 degrees with respect to the axes laying in the input plane of the slab. a) Calculate the (minimum) λ/4-slab thicknesses in the three conﬁgurations. b) In reality, if the incident wave (in air) is linearly polarized, the output polarization will be slightly elliptical because the reﬂection coeﬃcients along the two principal axes will be slightly diﬀerent. Evaluate the deviation from circular polarization, in terms of percentage. 6-14 Show that, if a plane wave is launched into a lossless gyrotropic medium in the x3 direction, with a generic state of polarization, then its state of polarization evolves periodically as a function of the distance traveled by the wave (provided both circularly polarized “eigenwaves” propagate at the frequency of interest). Express the period as a function of the propagation constants of the two circularly polarized eigenwaves. Does the state of polarization still evolve periodically with distance in the presence of loss ? 6-15 Prove that a lossless anisotropic medium, where the only plane waves that propagate in one direction (x3 ) preserving their state of polarization, are circularly polarized, is gyrotropic.

© 2006 by Taylor & Francis Group, LLC

220

PLANE WAVES IN ANISOTROPIC MEDIA

6-16 Prove that the change in reference frame described at the beginning of Section 6.7 yields Eq. (6.59). 6-17 Show that, analogous to what was shown in Section 6.3, for a uniform plane wave in a gyrotropic medium where the reference frame of Section 6.7 is used, the vector D is orthogonal to y3 and, if the wave is linearly polarized, it belongs to the plane deﬁned by y3 and by the electric ﬁeld vector E. 6-18 Study under which conditions one of the plane waves in the medium studied in Section 6.8 is evanescent. Can both waves be evanescent ? 6-19 In the highly ionized layers of the Earth’s ionosphere, electron densities are of the order of 1012 m−3 . A typical value for the terrestrial static magnetic induction ﬁeld is 3 × 10−5 T. a) Calculate the corresponding plasma frequency and cyclotron frequency. b) Repeat the calculation for a laboratory plasma with 1016 electrons per cubic meter and a 10 T induction ﬁeld. 6-20 Sketch the plots of the dispersion diagrams, k(ω) vs. ω, for the two circularly polarized waves whose phase constants are expressed by Eqs. (6.57) and (6.58), when the elements of the permittivity matrix are expressed by Eqs. (6.89). Consider separately the cases ωp > ωc , ωp < ωc . 6-21 Evaluate the intensity of the static magnetic ﬁeld H0 which is required in order to have a resonant frequency f0 = 1 GHz, in a saturated ferrite. 6-22 To build a microwave isolator (see Section 3.4) based on Faraday rotation, a saturated ferrite bar is used, with a static magnetic ﬁeld, H0 , applied in the direction along which the wave propagates. a) Approaching the problem in the dual way with respect to a gyrotropic dielectric medium, express the phase constants of the circularly polarized waves, and then the per-unit-length Faraday rotation, for a gyromagnetic medium with = 0 , and the permeability dyadic found in Section 6.8. Note: assume that the bar cross-section is wide enough to approximate the propagation constants in it with those of plane waves in an unbounded medium. b) Evaluate the intensity of the static magnetic ﬁeld required in order to have a Faraday rotation of π/4 radians over a length = 30 mm at a frequency f = 5 GHz. The saturation magnetization M0 equals 106 A/m.

© 2006 by Taylor & Francis Group, LLC

References Agrawal, G.P. (1995) Nonlinear Fiber Optics. 2nd ed., Academic, San Diego. Babuska, V. and Cara, M. (1991) Seismic Anisotropy in the Earth. Kluwer Academic, Boston. Bloembergen, N. (1996) Nonlinear optics. 4th ed., World Scientiﬁc, River Edge, NJ. Born, M. and Wolf, E. (1980) Principles of Optics: Electromagnetic Theory of Propagation, Interference and Diﬀraction of Light. 6th (corrected) ed. (reprinted 1986), Pergamon, Oxford. Boyd, R.W. (1992) Nonlinear Optics. Academic Press, Boston. Budden, K.G. (1961) Radio Waves in the Ionosphere. Cambridge University Press, Cambridge, UK. Butcher, P.N. and Cotter, D. (1990) The Elements of Nonlinear Optics. Cambridge University Press, Cambridge, UK. Chandrasekhar, S. (1989) Plasma Physics, Hydrodynamic and Hydromagnetic Stability, and Applications of the Tensor-Virial Theorem. University of Chicago Press, Chicago. Dendy, R. (Ed.) (1993) Plasma Physics: An Introductory Course. Cambridge University Press, UK. Dmitriev, V.G., Gurzadyan, G.G. and Nikogosian, D.N. (1991) Handbook of Nonlinear Optical Crystals. Springer-Verlag, Berlin. Hijden, J.H.M.T. van der (1987) Propagation of Transient Elastic Waves in Stratiﬁed Anisotropic Media. North Holland, Amsterdam. Jackson, J.D. (1975) Classical Electrodynamics. Wiley, New York. Juretschke, H.J. (1974) Crystal Physics: Macroscopic Physics of Anisotropic Solids. Benjamin, Advanced Book Program, Reading, MA. Kong, J.A. (1990) Electromagnetic Wave Theory. 2nd ed., Wiley, New York. Miyamoto, K. (1989) Plasma Physics for Nuclear Fusion. Rev. ed., MIT Press, Cambridge, MA. Nayfeh, A.H. (1995) Wave Propagation in Layered Anisotropic Media, with Applications to Composites. Elsevier, Amsterdam. Negi, J.G. and Saraf, P.D. (1989) Anisotropy in Geoelectromagnetism. Elsevier, Amsterdam. Newell, A.C. and Moloney, J.V. (1992) Nonlinear Optics. Advanced topics in the interdisciplinary mathematical science, Addison-Wesley, Redwood City, CA. Nye, J.F. (1985) Physical Properties of Crystals: Their Representation by Tensors and Matrices. 1st paperback edition with corrections and new material, Clarendon Press, Oxford, UK. Plonsey, R. and Collin, R.E. (1961) Principles and Applications of Electromagnetic Fields. McGraw-Hill, New York. Shen, Y.R. (1984) The Principles of Nonlinear Optics. Wiley, New York. Sitenko, A.G. and Malnev, V. (1995) Plasma Physics Theory. Chapman & Hall, 221 © 2006 by Taylor & Francis Group, LLC

222

REFERENCES

London. Smirnov, V.I. (1961) Linear Algebra and Group Theory. McGraw-Hill, New York. Someda, C.G. and Stegeman, G.I. (Eds.) (1992) Anisotropic and Nonlinear Waveguides. Elsevier, Amsterdam. Staelin, D.H., Morgenthaler, A.W. and Kong, J.A. (1993) Electromagnetic Waves. Prentice-Hall, Englewood Cliﬀs, NJ. Sturrock, P.A. (1994) Plasma Physics: An Introduction to the Theory of Astrophysical, Geophysical, and Laboratory Plasmas. Cambridge University Press, UK. Tarling, D.H. and Hrouda, F. (1993) The Magnetic Anisotropy of Rocks. Chapman & Hall, London. Tosco, F. (Ed.) (1990) Fiber Optic Communications Handbook. TAB Professional and Reference Books, Blue Ridge Summit, PA. Yariv, A. and Yeh, P. (1984) Optical Waves in Crystals: Propagation and Control of Laser Radiation. Wiley, New York. Zernike, F. and Midwinter, J.E. (1973) Applied Nonlinear Optics. Wiley, New York.

© 2006 by Taylor & Francis Group, LLC

CHAPTER 7

Waveguides with conducting walls

7.1

Introduction

What we saw in Chapters 4, 5 and 6 indicates that there are no physical reasons why in a homogeneous medium the energy carried by an electromagnetic wave should travel along a given direction. Indeed, Chapters 12 and 13, devoted to the fundamentals of antenna theory, will reconﬁrm that sources in a homogeneous medium cannot radiate just along one linear path, but necessarily in all directions, at least within a given cone. Guided propagation, which means sending almost all the energy conveyed by an electromagnetic wave along a chosen path, inevitably requires an inhomogeneous medium. However, depending on the applications and especially on the frequency in use, nonhomogeneities can either consist of combining dielectrics and conductors, or merely using diﬀerent dielectric media. In principle, electromagnetic ﬁelds penetrate in all media whose conductivity is ﬁnite, and satisfy continuity conditions on the surfaces between diﬀerent media. In practice, we may encounter two drastically diﬀerent situations. In a structure which consists only of dielectrics, in none of them is the ﬁeld negligible, even as a ﬁrst approximation. On the other hand, the step from a good conductor (deﬁned as in Section 4.12) to an ideal conductor, in which the ﬁeld is identically zero, is, at least in the ﬁrst approach to the problem, an excellent trade-oﬀ between being realistic and simplifying the mathematics. In fact, it allows replacement of the continuity conditions at the interface between dielectric and metal with boundary conditions for the ﬁeld, deﬁned only in the dielectric. For historical reasons, but above all for their mathematical simplicity which follows from what we said so far, conducting-wall waveguides will be dealt with ﬁrst, in this chapter. Dielectric waveguides will be studied later, in Chapter 10. The subject will not be tackled in terms of plane waves, although their completeness, proved in Section 4.7, would enable us to do that. In fact, a diﬀerent approach will yield more compact results, practically useful and formally elegant at the same time. We will invoke a plane wave approach only at some points where it provides an intuitive explanation for some of the results obtained in the other way. 223 © 2006 by Taylor & Francis Group, LLC

224

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.1 A waveguide of unspeciﬁed geometry, and inﬁnite length.

7.2

Homogeneously ﬁlled cylindrical structures: Simpliﬁed proof of the TE-TM decomposition theorem

Let us consider, in real 3-D space, a region ﬁlled with a homogeneous medium and surrounded by walls, whose physical nature does not have to be speciﬁed now, which extend to inﬁnity along one direction, z, and whose cross-section is constant vs. z. We will summarize these assumptions calling this region a “cylindrical structure in the z direction”. Let a ˆ z be the unit vector in the z direction, u1 , u2 be two generic orthogonal curvilinear coordinates in the plane orthogonal to z (see Figure 7.1). We will show that, under these circumstances, the TE-TM decomposition theorem can be proved in a greatly simpliﬁed way, compared with Section 3.8, while at the same time a physical identiﬁcation is possible for the rather mysterious pre-potentials that were introduced at that stage. As we anticipated in Section 3.8, there is some price to pay for this simpliﬁcation. One is the need to make an ansatz (easy to justify) for the z-dependence of the ﬁelds, the other one is a loss in generality, to be discussed and by-passed later in this section. As for the ansatz, it consists of the following assumptions, which can ﬁnd their justiﬁcation either in what we saw in Section 4.7 or, a posteriori, in the results that will follow from them, in particular for what concerns completeness of the solutions: E(u1 , u2 , z)

=

¯ 1 , u2 ) exp(−γz) E(u

,

H(u1 , u2 , z)

=

¯ 1 , u2 ) exp(−γz) H(u

.

(7.1)

In Section 3.8 it was shown that any ﬁeld in such a region can be expressed as the sum of two ﬁelds which are transverse electric (TE) and transverse magnetic (TM) with respect to an arbitrary real unit vector a ˆ. From now on, unless the opposite is explicitly stated, TE and TM will be understood with respect to a ˆz . To simplify the proof as much as possible, let us deal, for the time being, with rectangular coordinates (u1 = x, u2 = y), and write Maxwell’s equations by components using the symbolic determinant notation (see Appendix B) for the curl operator. We get: ∂Ez + γ Ey ∂y

© 2006 by Taylor & Francis Group, LLC

= −jωμ Hx

,

HOMOGENEOUSLY FILLED CYLINDRICAL STRUCTURES ∂Ez = −jωμ Hy , −γ Ex − ∂x ∂Hz + γ Hy ∂y

= jωc Ex

225

,

∂Hz = jωc Ey , (7.2) ∂x plus the two components along z, which we omit here as they are not relevant to the following. These equations can be looked at as an algebraic system of four linear equations in the four unknowns Ex , Ey , Hx , Hy , whose solution is simply found to be: ∂Hz 1 ∂Ez + jωμ Ex = − 2 γ , γ − κ2 ∂x ∂y ∂Hz 1 ∂Ez − jωμ γ Ey = − 2 , γ − κ2 ∂y ∂x ∂Hz ∂Ez 1 +γ −jωc Hx = − 2 , γ − κ2 ∂y ∂x ∂Hz ∂Ez 1 + γ jω , (7.3) Hy = − 2 c γ − κ2 ∂x ∂y −γ Hx −

where, as usual, κ2 = −ω 2 μc . These equations prove the ﬁrst part of the decomposition theorem, namely that any ﬁeld in the region of interest can be expressed as the sum of a TE and a TM ﬁeld. The transverse electric ﬁeld consists of Hz and of four transverse components expressed by the terms in the right column, on the right-hand side of Eqs. (7.3). The transverse magnetic ﬁeld consists of Ez and of four transverse components expressed by the terms in the left column, on the right-hand side of Eqs. (7.3). However, it is extremely important to point out that this proof holds only for γ 2 = κ2 . Otherwise, the quantity in the denominators on the right-hand side goes to zero. Implications of this restriction will be discussed shortly. Before we proceed to the second part of the decomposition theorem, we note, incidentally, that Eqs. (7.3) can be rewritten in an intrinsic form, independent of the (x, y) reference frame, as follows: E¯t

=

γ2

1 (−γ ∇t Ez + jωμ a ˆz × ∇t Hz ) , − κ2

¯t H

=

γ2

1 (−γ ∇t Hz − jωc a ˆz × ∇t Ez ) , − κ2

(7.4)

where the transverse gradient ∇t is deﬁned through the position: ∇ = ∇t + a ˆz

© 2006 by Taylor & Francis Group, LLC

∂ = ∇t − a ˆz γ ∂z

,

(7.5)

226

WAVEGUIDES WITH CONDUCTING WALLS

and can be expressed in a generic system of orthogonal curvilinear coordinates u1 , u2 as: tˆ1 ∂ tˆ2 ∂ ∇t = · + · , (7.6) h1 ∂u1 h2 ∂u2 where h1 , h2 are the metric coeﬃcients of the (u1 , u2 ) reference frame. The second part of the decomposition theorem says that the TE part of the ﬁeld can be derived from a scalar function which satisﬁes the scalar Helmholtz equation, and the same is true for the TM part. In the case at hand, it is straightforward to show that these two scalar functions are (apart from proportionality constants, which are needed to reconcile them with Section 3.8) the longitudinal ﬁeld components Hz and Ez , respectively. In fact, Eqs. (7.3) show, on one hand, that when these two components are known, the entire ﬁeld can be calculated by means of straightforward operations. On the other hand, the ﬁelds H and E, in a homogeneous medium without sources, satisfy the vector Helmholtz equation, as we saw in Chapter 1, and therefore their components along the straight axis z satisfy the scalar Helmholtz equation, as we had to prove. In more detail, if we deﬁne a transverse Laplacian ∇2t through the position: ∂2 = ∇2t + γ 2 , (7.7) ∂z 2 we see that the two functions of the two transverse coordinates u1 , u2 satisfy an equation of the type: ∇2t F = −χ2 F , (7.8) ∇2 = ∇2t +

where F is either Hz or Ez , and: χ2 = −(κ2 − γ 2 ) = 0 .

(7.9)

This is an eigenvalue equation, similar to the Helmholtz equation but different from it because in the 3-D case the eigenvalue χ2 = −κ2 is unique, and is known when the frequency and the medium are speciﬁed, whereas in Eq. (7.8) the eigenvalue is unknown. We will see shortly that indeed Eq. (7.8), at a given frequency, has an inﬁnity of discrete solutions, for suitable boundary conditions. Each eigenvalue corresponds to a value of γ, whose physical meaning is, according to Eq. (7.1), that of a propagation constant in the z direction. We understand then that diﬀerent solutions of Maxwell’s equations of the form in Eq. (7.1), corresponding to diﬀerent eigenvalues in Eq. (7.8), will have diﬀerent propagation properties. The “spatial spectrum” of the total ﬁeld which can exist in a cylindrical structure will be radically diﬀerent from a plane-wave spectrum. The link between the quantities and the symbols used in this section and those of Section 3.8 are simply found, by comparison, to be: L = μ Hz /χ2

,

T = −c Ez /χ2

.

(7.10)

Hence, if one follows the approach of Section 3.8, the expressions in Eq. (3.47) and (3.48) for the TE and TM ﬁelds, respectively, become the following ones,

© 2006 by Taylor & Francis Group, LLC

HOMOGENEOUSLY FILLED CYLINDRICAL STRUCTURES

227

where symbols like L and T indicate, as in the rest of this section, that the exponential dependence on z has been factored out: E¯TE

= −jω ∇t L × a ˆz

¯ TE H

= −

E¯TM

=

¯ TM H

,

1 2 γ (κ − γ 2 ) L a ˆz − ∇t L , μ μ

1 γ (κ2 − γ 2 ) T a ˆz + ∇t T c c

= −jω ∇t T × a ˆz

.

(7.11)

, (7.12)

Comparing Eqs. (7.12) (which, being based on the general proof of Section 3.8, are valid also for χ2 = 0) and Eqs. (7.4), we see that a ﬁeld corresponding to the eigenvalue χ2 = 0 can be non-identically zero only if: Ez ≡ 0

,

Hz ≡ 0

,

(7.13)

i.e., for a transverse electro-magnetic (TEM) ﬁeld. We will see shortly that the loss in generality caused by the restriction χ2 = 0 is not harmful, because the guided propagation of a TEM ﬁeld is quite peculiar. Indeed, its ﬁrst peculiarity, which can be derived starting either from Eqs.(7.11) or from Eqs. (7.12), is that: ¯ ×a E¯ = ±η H ˆz , (7.14) 1/2 is, as usual, the intrinsic impedance of the homogeneous where η = (μ/c ) medium in the cylindrical structure. The sign is either + or − depending on whether γ = +κ or γ = −κ . Hence, for a TEM ﬁeld the propagation constant and the wave impedance coincide with those of a uniform plane wave traveling in the z direction, in an unbounded medium with the same characteristics as those of the medium which ﬁlls the cylindrical structure. We will see in Section 7.4 that none of the other solutions of Maxwell’s equations in a waveguide can have these properties. In the same section we will also see the conditions under which a cylindrical structure can propagate a TEM ﬁeld. For the time being, let us point out that for a TEM wave, the ﬁelds, the potentials and the pre-potentials, as functions of the transverse coordinates, all obey an equation of the following type: ∇2t F = 0 ,

(7.15)

which is the 2-D Laplace equation. Then, we will be entitled to invoke all the uniqueness theorems which hold for this equation, with which the reader should be familiar from electrostatics. To close this section, let us emphasize that we proved that any ﬁeld in a homogeneous cylindrical structure can be expressed as the sum of a TE ﬁeld, a TM ﬁeld and, in some cases to be clariﬁed, a TEM ﬁeld. Whether one of these ﬁelds may exist in a given structure independently of the others, this can be said only when one knows what boundary conditions have to be satisﬁed in that structure. This will be the subject of Section 7.3 and again of Section 7.9.

© 2006 by Taylor & Francis Group, LLC

228 7.3

WAVEGUIDES WITH CONDUCTING WALLS Waveguides with ideal conducting walls

To get involved in the details of guided propagation, let us start from structures where metallic walls shield completely the region where the guided ﬁeld is conﬁned. This approach is convenient, for the reasons that were mentioned in the introduction, both in terms of concepts and of technicalities. The problem is further simpliﬁed in a drastic manner if we assume that the walls consist of an ideal electric conductor (see Section 4.12). Although, obviously, this is not realistic, nevertheless all the results of practical interest can be obtained from this model, at most by introducing further reﬁnements which are usually small and simple. Examples will be given in Section 7.9 and later on. It is well known that an ideal electric conductor imposes Etan = 0 on its surface. Then, through Maxwell’s equations, also the normal component of H, Hn , vanishes on that surface. Given the ansatz Eq. (7.1), these boundary conditions (not independent of each other) on the contour C of the cylindrical region where the guided ﬁeld is deﬁned (Figure 7.1) read: E¯tan = 0 ,

Hn = 0 .

(7.16)

We will show now that, if we leave aside the TEM case assuming χ2 = 0, then Eqs. (7.16) are satisﬁed on C if the following conditions are satisﬁed on the same surface: ⇐⇒ T = 0 , (7.17) Ez = 0 d Hz dn

= 0

⇐⇒

dL =0 dn

,

(7.18)

where d/dn is, as usual, the normal derivative on C. To prove this statement, we note ﬁrst that Eq. (7.17) implies Ez = constant on the surface C. Then ∇t Ez is orthogonal to C at any point. Conversely, Eq. (7.18) implies that the normal component of ∇t Hz vanishes on C, so that this vector is tangent to C at any point. Consequently, the general expressions in Eqs. (7.4) for the transverse components show that Et is normal to C and Ht is tangent to C everywhere. If, furthermore, we recall that Ez = 0 on C by assumption, then we conclude that Eqs. (7.16) are satisﬁed everywhere on C. Note that Eqs. (7.17) and (7.18) are independent of each other, unlike the two Eqs. (7.16). In particular, Eq. (7.17) is compatible with taking Hz ≡ 0 (L ≡ 0), which means it is applicable to a TM ﬁeld. Eq. (7.18) is compatible with taking Ez ≡ 0 (T ≡ 0), meaning it is applicable to a TE ﬁeld. A very important consequence that we may draw is that TE transmission modes and TM transmission modes exist in a cylindrical structure with ideal conducting walls. In Section 7.5 we will see that they form a complete set, in some cases provided that the TEM mode is included also. As we brieﬂy said at the end of Section 7.2, existence and completeness of TE and TM modes are properties which are not shared by all cylindrical structures. They follow from the speciﬁc assumptions in Eq. (7.16). Finally, note that Eqs. (7.17) and (7.18) are two scalar boundary conditions.

© 2006 by Taylor & Francis Group, LLC

TRANSMISSION MODES

229

We saw in Section 7.2 that the complete expressions for TE and TM ﬁelds may be derived from the solutions of two scalar partial diﬀerential equations. We may then conclude that guided propagation in cylindrical structures with ideal conducting walls can be thought of, under a mathematical point of view, as a scalar problem, in spite of the vector physical nature of the guided ﬁeld. 7.4

Transmission modes of lossless cylindrical structures

From now on, unless the opposite is explicitly stated, we will assume that the dielectric in the cylindrical structure is lossless. The walls were supposed in the previous section to be ideal conductors, so the whole structure is strictly lossless. The quantity κ2 = −ω 2 μ (constant, because the dielectric was supposed to be homogeneous at the beginning of Section 7.2) is real and negative. This property will be exploited several times in this section. 7.4.1

Properties of all transmission modes

A1. Real, positive eigenvalues All the eigenvalues in Eq. (7.9) of Eq. (7.8) are real and positive (except for the zero eigenvalue of the TEM mode). This follows from the 2-D form of the ﬁrst Green identity (A.10), namely: ∂ψ d , (7.19) ϕ ∇2t ψ + ∇t ϕ · ∇t ψ dS = ϕ ∂n S which we apply to the cross-section of the structure, S, and to its contour, , setting ϕ = F, ψ = F ∗ , F being any solution of Eq. (7.8). For all modes but the TEM, the boundary conditions in Eqs. (7.17) and (7.18) make the right-hand side of Eq. (7.19) equal to zero. Then, through Eq. (7.8) we get: 2 ∗ ∗ FF dS = ∇t F · ∇t F ∗ dS . (7.20) χ S

S

Both surface integrals in Eq. (7.20) are real and positive, whence the conclusion on χ2 . A2. Critical frequency For each transmission mode (labeled with an index n, shown later in Section 7.5 to be an integer) there is one positive value (zero for the TEM mode) of the angular frequency ω, at which the longitudinal propagation constant γ vanishes. Indeed, if we rewrite Eq. (7.9) as: γn2 = χ2n + κ2 = χ2n − ω 2 μ , and set it equal to zero, then it is satisﬁed for: χn ω = √ ≡ ωcn , μ

© 2006 by Taylor & Francis Group, LLC

(7.21)

(7.22)

230

WAVEGUIDES WITH CONDUCTING WALLS

which is a real and positive solution because of the previous property A1. The symbol ωcn underlines that this is called the critical angular frequency of the n-th mode. It is also called the cut-oﬀ angular frequency, for reasons to be explained in the next paragraph. A3. Cut-oﬀ of a transmission mode For the n-th transmission mode, the longitudinal propagation constant γ n is real at all frequencies smaller than its critical frequency, imaginary at all frequencies higher than its critical frequency. To prove it, we start from Eq. (7.21), which indicates that γn2 > 0 for ω < ωcn , γn2 < 0 for ω > ωcn . Hence, we may let: √ for ω < ωcn , (7.23) γn = αn = ω μ (ωcn /ω)2 − 1 ∈ Re √ for ω > ωcn . (7.24) γn = jβn = jω μ 1 − (ωcn /ω)2 ∈ Im This tells us that any transmission mode in a waveguide does not propagate (i.e., its phase does not change vs. z) at frequencies below its critical frequency (whence the name “cut-oﬀ frequency”). As the structure is strictly lossless, the attenuation described by Eq. (7.23) cannot be caused by energy dissipation. Its physical nature will be clariﬁed by looking also at the following properties. A4. Wave impedance The wave impedance in the z direction is imaginary for ω < ωcn , and real for ω > ωcn . To prove this, note that for a TE mode (Ez ≡ 0), the electric and magnetic nonvanishing terms in Eqs. (7.4) are in quadrature when Eq. (7.23) holds, are in phase when Eq. (7.24) holds. The same is true for the nonvanishing terms in the case of a TM mode (Hz ≡ 0). Further details will be added in the following subsections. Comparing with what we saw in Chapter 4 concerning propagation constants and wave impedances of plane waves, we may conclude that the physical phenomenon why a mode is attenuated vs. z below its cut-oﬀ frequency, is an evanescence. Then, the mode cannot convey any active power in the longitudinal direction of the cylindrical structure. However, in strict correspondence to what we saw in Chapter 4, this statement is correct only in the presence of one wave which attenuates in one sense along the z axis, and loses any validity for two superimposed waves which decay in the two opposite senses. A5. Phase and group velocities For any mode that can propagate in a cylindrical structure (i.e., at ω > ωcn ), the longitudinal phase velocity as a function of frequency can be obtained from Eq. (7.24) and reads: vpn =

© 2006 by Taylor & Francis Group, LLC

ω =c βn

1−

1 ω

cn

ω

2

,

(7.25)

TRANSMISSION MODES 231 √ where c = 1/ μ is the speed of light in the dielectric of the cylindrical structure. Eq. (7.25) indicates that vpn tends to inﬁnity for ω → ωcn , and decreases monotonically as ω increases, approaching c (from above) as ω → ∞. From what we saw in Chapter 4 on phase velocity of plane waves in any direction, this result suggests that at its cut-oﬀ frequency a guided mode consists of plane waves which travel along directions in the plane orthogonal to a ˆz . As ω is increased, i.e., as the wavelength becomes smaller with respect to the transverse geometrical dimensions of the structure, the plane waves the mode consists of, travel along directions which make smaller and smaller angles with that of a ˆz direction, gradually tending to a situation where they ignore the presence of walls. The group velocity: ω 2 1 cn =c 1− vgn = , (7.26) ∂βn ω ∂ω behaves, vs. frequency, in a way which agrees perfectly with the previous comment. Indeed, it equals zero at cut-oﬀ (ω = ωcn ), and increases as the frequency increases, tending asymptotically to c (from below) for ω → ∞. From Eqs. (7.25) and (7.26) it follows that: vgn vpn = c2

.

(7.27)

This relationship holds only in the structures that we are studying here. It cannot be extrapolated, unfortunately, to the dielectric waveguides of Chapter 10, where the behavior of group velocity vs. frequency is of paramount importance, and turns out to be much more complicated. The functions in Eqs. (7.23), (7.24), (7.25), (7.26) are sketched in Figure 7.2. According to the deﬁnitions given in Section 5.2, Eq. (7.26) shows that propagation of a guided mode in a metallic waveguide is aﬀected by anomalous dispersion (dvgn /dω > 0). Eq. (7.26) also shows that two guided modes having diﬀerent critical frequencies (i.e., diﬀerent eigenvalues) have, at the same frequency, diﬀerent group velocities. Eq. (7.25) shows that they have diﬀerent phase velocities also. Consequently, if several modes propagate together, a modulating signal is distorted because the group delay from the waveguide input to its output has as many values as the propagating modes. This cause of distortion, often referred to as “multipath,” in general will be superimposed onto the classical causes of signal distortion, which, as well known, for a two-port linear system read |H(jω)| = constant, H(jω) = aω + b, where H(jω) is the system transfer function. By aﬃnity with what we described in Section 5.2, phenomena related to multiple values of the group delay are also referred to as multimode (or inter-modal) dispersion. Eq. (7.27) shows that in a waveguide where the dielectric is homogeneous it is never possible, not even at a single value of ω, to have equal phase or group velocities for modes whose cut-oﬀ frequencies are diﬀerent. In Chapter 10 we will see that, on the contrary, curves of phase or group velocities vs. ω of

© 2006 by Taylor & Francis Group, LLC

232

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.2 Attenuation constant, phase constant, phase velocity, group velocity vs. frequency, for a generic mode of a lossless metallic waveguide.

diﬀerent modes may intersect each other in waveguide where the dielectric is not the same in the entire cross-section. On the other hand, the previous formulas show that two modes having equal eigenvalues have the same phase and group velocities at any frequency. Modes which are physically distinct but have equal eigenvalues are called degenerate. Their role is quite particular, and becomes extremely important in problems such as power exchange among modes of the same structure or of coupled structures, as we shall clarify later in this chapter and in the two following ones.

A6. The fundamental mode

The mode corresponding to the lowest eigenvalue compatible with the boundary conditions imposed by a given structure, χ20 , is called the fundamental mode of the structure. If it is not degenerate, it is the only mode which can propagate alone in that structure. This occurs over the frequency range between f0 = χ0 c/(2π) and the smallest among the cut-oﬀ frequencies of all the other modes. This is the only frequency range where the problems due to intermodal dispersion may be avoided.

© 2006 by Taylor & Francis Group, LLC

TRANSMISSION MODES

233

Figure 7.3 Behavior of the wave impedance of TE and TM modes as functions of frequency, plotted on the Smith chart.

7.4.2

Properties of TE modes

B1. Wave impedance The wave impedance in the direction of the unit vector a ˆ z , which can be calculated letting Ez = 0 in Eqs. (7.4), is: ηTE =

jωμ γn

.

(7.28)

It is constant over the cross-section of the structure. As for its frequency dependence, Figure 7.3 illustrates it on the Smith chart, to be deﬁned in the next chapter, but it can be easily described even without this graphical tool. One must distinguish, as we said in the previous subsection, whether the frequency is smaller or larger than fcn = ωcn /(2π). For f < fcn , Eq. (7.28) is an inductive reactance, equal to zero for f = 0, then increasing as f increases, becoming inﬁnite for f = fcn . Passing through this value, which corresponds to an ideal open circuit, Eq. (7.28) fcn , and becomes a resistance for f > √ then tends asymptotically to η = μ/ for f → ∞, while γn → jω μ, i.e., while the generic TE mode tends to become a plane wave traveling in the z direction. We ﬁnd an inﬁnite ηTE for f = fcn because at this frequency the ¯ vanishes, so that nothing changes if a cross-section transverse component of H of the structure is “metallized” by means of an ideal magnetic conductor (see Chapter 3). B2. Poynting vector The Poynting vector of the generic TE mode is given by: 1 1 1 ∗ −2αn z ∗ 2 ∗ jωμγn |∇t Hz | a ˆz − 2 jωμHz ∇t Hz , Pn = E × H n = e 2 2χ4n 2χn (7.29) where αn = 0 for f ≥ fcn . Its longitudinal part (the ﬁrst term) is imaginary ˆ z . Its for f < fcn , real for f > fcn , and is oriented in the same sense as a

© 2006 by Taylor & Francis Group, LLC

234

WAVEGUIDES WITH CONDUCTING WALLS

transverse component, parallel to ∇t Hz , can change in phase from one point to another in the cross-section only if the phase of Hz depends on the transverse coordinates. If this is not the case, then the phases of Hz∗ and ∇t Hz cancel each other, and we may state that the transverse component of the Poynting vector is imaginary at any frequency. In fact, we will see in the next section that Hz = constant is always a legitimate assumption. Therefore, we may indeed state that the transverse Poynting vector is indeed purely imaginary. 7.4.3

Properties of TM modes

These properties are “dual” of those of the TE modes, so they will be stated without proofs. C1. Wave impedance The wave impedance in the direction of a ˆ z is: γn ηTM = . (7.30) jω Its behavior vs. ω is also plotted schematically in Figure 7.3, and can be described as follows. For f = fcn we have ηTM = 0: at this frequency, E¯t = 0, so that nothing changes if the cross-section is short-circuited by means of an ideal electric conductor. C2. Poynting vector The Poynting vector of the generic TM mode is given by: 1 1 −2αn z 2 ∗ Pn = e ˆz − 2 jωEz ∇t Ez , − 4 jωγn |∇t Ez | a 2χn 2χn

(7.31)

where again αn = 0 for f ≥ fcn . For its longitudinal and transverse components, refer to Subsection B2. 7.4.4

Properties of the TEM mode

D1. Uniqueness of the TEM mode We saw in Section 7.2 that, when χ2 = 0, the ﬁeld must be TEM in order not to give inconsistent relationships. Now let us show also that the inverse is true, namely that a ﬁeld can be TEM only for χ2 = 0. In fact, for χ2 = 0, Eqs. (7.4) show that if we set Ez = 0, Hz = 0, then we necessarily have E¯t = 0, ¯ t = 0, i.e., the trivial solution of Maxwell’s equations in a region without H sources. Therefore the eigenvalue corresponding to a TEM ﬁeld is unique. For this reason we speak of the TEM mode. D2. Existence of the TEM mode The TEM mode may exist only in cylindrical structures where the crosssection of the dielectric is not a simply connected domain. Indeed, we saw

© 2006 by Taylor & Francis Group, LLC

MODE ORTHOGONALITY

235

that the TEM ﬁeld satisﬁes Eq. (7.15): ∇2t F = 0

(F = L, T ) ,

(7.32)

which is the Laplace equation. A well-known theorem ensures the uniqueness of its solution, when the values of either F or ∂F/∂n are known over the whole contour. Suppose that the border, , consists of only one ideal conductor, where Etan = 0. Then, elaborating on Eqs. (7.12) one may show that this assumption implies ∂L/∂n = 0 on and T = constant on . These conditions are compatible with a ﬁeld identically zero everywhere. Because of the uniqueness theorem that we have just quoted, the vanishing ﬁeld is the solution of the problem. In conclusion, the TEM ﬁeld cannot exist if the cross-section is a simply connected domain. Examples of non-simply connected regions of practical interest are two-wire lines and coaxial cables. The latter will be studied in this chapter. Most of the contents of the following chapter apply to both. One section in the next chapter will be devoted to striplines, which are also very important examples of structures with non-simply connected cross-sections. However, in striplines the dielectric is inevitably not homogeneous throughout the cross-section, and consequently they do not ﬁt rigorously into the patterns of this chapter. Their fundamental mode, as we shall see, is called “quasi-TEM,” and can be dealt with only by means of approximations. D3. TEM cut-oﬀ frequency In non-simply connected structures, the TEM mode is the fundamental mode. This is a corollary of the property D1 (the eigenvalue of the TEM mode is zero): through Eq. (7.22) we see that the cut-oﬀ frequency of the TEM mode is also equal to zero: (7.33) fcTEM = 0 . By deﬁnition of the fundamental mode, the conclusion is obvious. Eq. (7.33) tells us that the frequency range where only one mode propagates begins at zero frequency for those structures whose cross-sections are not simply connected. From direct current, it extends to the smallest among the critical frequencies of the higher-order modes. A typical example of higherorder modes in coaxial cable will be dealt with in Section 7.8. 7.5

Mode orthogonality

Propagation in waveguides without sources is described by homogeneous equations. In any mathematical problem of this type, it is important to discuss linear independence of solutions. In this section, we will prove some relationships which assess linear independence of diﬀerent solutions in its strongest possible form, namely orthogonality relationships. They can be looked at as extensions to an inﬁnite-dimensional vector space (the space of the solutions of Eq. (7.8)) of a property which is well known for eigenvalue problems of ﬁnite dimension, namely that eigenvectors belonging to diﬀerent eigenvalues are orthogonal.

© 2006 by Taylor & Francis Group, LLC

236 7.5.1

WAVEGUIDES WITH CONDUCTING WALLS Orthogonality between two TE modes or two TM modes

As usual, let S be the cross-section, normal to the unit vector a ˆ z , of a cylindrical structure surrounded by ideal conductors. Let be the border of S ( = S ∩ C, where C is the structure wall). Let us write on S the 2-D form of the Green theorem (A.11), where ϕ and ψ are two functions of the two transverse coordinates, regular over the domain S: ∂ϕ ∂ψ 2 2 −ψ ϕ ∇t ψ − ψ∇t ϕ dS = ϕ d . (7.34) ∂n ∂n S Let now ϕ and ψ be two functions satisfying Eq. (7.8). To be speciﬁc, let ϕ = Hzj , ψ = Hzk (i.e., we are dealing with the TEj and TEk modes). Both functions satisfy the boundary condition of Eq. (7.18), so the right-hand side of Eq. (7.34) is zero. The left-hand side can be modiﬁed exploiting Eq. (7.8), so we write: 2 2 χk − χj Hzj Hzk dS = 0 . (7.35) S

We get immediately the following orthogonality relationship: Hzj Hzk dS = 0 for χ2j = χ2k .

(7.36)

S

If we let ϕ = Ezj , ψ = Ezk (i.e., if we deal with two TM modes), satisfying the boundary condition of Eq. (7.17), then the right-hand side of Eq. (7.34) is again zero, and we reach the following result: Ezj Ezk dS = 0 for χ2j = χ2k . (7.37) S

Two additional orthogonality relationships can be proved starting from Eqs. (7.36) and (7.37), and using the ﬁrst Green identity written in the form in Eq. (7.19). In fact, a proportionality relationship similar to Eq. (7.20) yields then: E¯tj · E¯tk dS = 0 for χ2j = χ2k , (7.38)

S

¯ tj · H ¯ tk dS = 0 H

for

χ2j = χ2k

.

(7.39)

S

These last relationships hold both for TE and for TM modes. In the case of two (or more) degenerate modes (as we said, modes which are physically diﬀerent but belong to equal eigenvalues, χ2j = χ2k , j = k), what we may say for certainty is that there exist always two (or more) linear combinations of such modes, that satisfy the same orthogonality relationships that we proved now. These linear combinations are still propagation modes, as we will see in Section 7.7. The algorithm by means of which orthogonal modes can be obtained from non-orthogonal modes is, in abstract terms, the same as that which is known in linear algebra as the Schmidt orthogonalization procedure (Dicke and Wittke, 1960). It can be sketched as follows: one chooses,

© 2006 by Taylor & Francis Group, LLC

MODE ORTHOGONALITY

237

arbitrarily, one among the degenerate modes, and leaves it as it is. Then, one takes the second mode (the order is still arbitrary if the modes are more than two) and subtracts from it its projection onto the ﬁrst mode. From the third mode, one subtracts the projection onto the ﬁrst one and the projection onto the second one; one proceeds in the same manner with the following modes. 7.5.2

Orthogonality between a TE mode and a TM mode

To prove that the transverse ﬁelds of a TE and a TM mode also satisfy Eqs. (7.38) and (7.39), even when their eigenvalues are equal, let us apply Stokes’ theorem in Eq. (A.9) to the vector ψ∇ϕ, where ψ and ϕ are two regular scalar functions of space coordinates. Exploiting the identity (C.6), so that ∇ × (ψ∇ϕ) = ∇ψ × ∇ϕ, we have: (∇t ψ × ∇t ϕ) · a ˆz dS = ψ∇ϕ · d . (7.40) S

Let ψ = Ezk (a TM mode) and ϕ = Hzj (a TE mode). Then, it is ψ ≡ 0 on , so the right-hand side of Eq. (7.40) vanishes. The result is: ∇t Ezk × ∇t Hzj · a ˆz dS = 0 . (7.41) S

Thanks to the cyclic permutation rule for the mixed product of three vectors and to Eq. (7.5), from Eq. (7.41) it follows that Eqs. (7.38) and (7.39) still hold, without any restriction on the eigenvalues. 7.5.3

Consequences of mode orthogonality

The total power ﬂowing through the cross-section of a waveguide with ideal conducting walls is the sum of the powers carried by the individual modes, since all “cross terms”: 1 Etj × H∗tk · a ˆz dS (j = k) , (7.42) 2 S vanish. Strictly speaking, this comment follows not from Eqs. (7.38) and (7.39), but from two similar equations, where Etk and Htk are replaced by E∗tk and H∗tk . These relations are easily proved following the previous lines, and can be veriﬁed as an exercise. If a ﬁeld {E, H} in a waveguide is expressed as a series of propagation modes, then the expansion coeﬃcients can be simply calculated by an “orthogonal projection,” i.e., evaluating integrals of the type: E · E∗k dS , H · H∗k dS . (7.43) S

S

A third fundamental consequence of mode orthogonality will be outlined in the next section.

© 2006 by Taylor & Francis Group, LLC

238 7.6

WAVEGUIDES WITH CONDUCTING WALLS Some remarks on completeness

In Section 7.2, we made use (and gave a simpliﬁed proof) of the theorem shown in Section 3.8. It says that in a homogeneous cylindrical structure any e.m. ﬁeld can be split as the sum of a TE and a TM ﬁeld (leaving aside, for the time being, the peculiar TEM case). In Section 7.3, we proved that, in a structure where the boundary conditions of Eq. (7.16) hold, there are TE and TM ﬁelds such that each of them, individually, satisﬁes these boundary conditions. The scope of validity of this second result is much narrower than that of the ﬁrst one, which is insensitive to boundary conditions: in fact, Section 7.11 will provide an important counter-example, where the ﬁrst result holds but the second does not. There are many problems in which an important prerequisite is to be able to “count” the modes of a structure, i.e., to be able to assess whether a given set of solutions is complete or not. In the theory of partial diﬀerential equations, boundary conditions and completeness are always linked to each other. This statement matches, in our case, with what we did in Section 4.7, where the assumption of a square-integrable ﬁeld was instrumental to prove completeness of plane waves, and in Section 5.12, when we discussed completeness of Gaussian beams. Let us now take the case of closed cylindrical structures, where boundary conditions hold on a closed surface which is all at a ﬁnite distance from any point where the ﬁeld is deﬁned. Then, we may proceed by analogy with what happens to signals in the time domain, in particular with the well-known links between the Fourier integral (for boundary conditions at inﬁnity) and the Fourier series (for boundary conditions at the two ends of an interval). This analogy suggests that a complete set may consist of an inﬁnity of discrete elements. However, the waveguide problem is 2-D, and in two dimensions a discrete inﬁnity can be built up in an inﬁnite number of possible ways. There is no reason to be sure, a priori, that any of these procedures will eventually lead to a complete set. To give an example of a set which is obviously not complete, let us take an inﬁnity of solutions which diﬀer from each other only in their dependence on one coordinate. In summary, cardinality of a set (i.e., the number of its elements) is not enough to assure completeness. In any case, completeness of a countable set of solutions of a partial differential equation (PDE) is a subject that can be kept within tolerable limits of size and complexity only if the PDE is separable, that is to say, it can be reduced to ordinary diﬀerential equations (ODE’s), each of these in one unknown quantity, a function of only one coordinate. In a 2-D case, like the one at hand, a complete set of solutions of the PDE consists then of all the products of a countable inﬁnity of linearly independent solutions of the ODE with respect to the ﬁrst coordinate, multiplied by a countable inﬁnity of linearly independent solutions of the ODE with respect to the second coordinate. As for Eq. (7.8), it can be shown that, for positive real eigenvalues, it is separable only in four types of orthogonal coordinates: rectangular cartesian;

© 2006 by Taylor & Francis Group, LLC

RECTANGULAR WAVEGUIDES

239

circular polar (where the coordinate lines are straight half-lines in radial directions, and circles); parabolic (whose coordinate lines are two families of confocal, mutually orthogonal parabolas); elliptical (whose coordinate lines are confocal ellipses and confocal hyperbolas, mutually orthogonal).∗ When dealing with a cylindrical structure of any shape, one can always adopt a coordinate frame which belongs to this set, but clearly, such a reference frame is beneﬁcial only when the boundary conditions are imposed on coordinate lines, or at least on lines that can be broken into pieces of coordinate lines. When this is not the case, then Eq. (7.8) can still be separated into two ODE’s, but their solutions cannot satisfy the boundary conditions on an individual basis, and the ﬁnal result is that each mode is not expressed as the product of two functions, but as the product of two series of functions. Nowadays, many numerical methods have been developed on these premises, and can in fact study waveguides of any shape within reasonable processing times and memory occupation, but it remains extremely hard to assess the completeness of a set of solutions obtained in this way. All the structures dealt with in this book have cross-sections which not only allow the separation of variables, but also enable us to satisfy boundary conditions on an individual basis, i.e., with any single element in a countable inﬁnity of solutions of two ODE’s. Modes will be expressed as products of individual functions, not of series. No formal mathematical proof will be given for the completeness of these sets. However, the reader’s background on periodic functions will be enough to make him/her conﬁdent that we do not neglect any physically meaningful solution. However, one should not extrapolate from these examples, erroneously, that, given a separable 2-D PDE, the products of any two countable inﬁnities of linearly independent solutions of the two ODE’s form a complete set. This is not true in general even if all these functions satisfy the boundary conditions. A very simple counter-example goes as follows: given a complete set of linearly independent solutions, take away one of its elements. The set which is left is still a countable inﬁnity, and its elements are still linearly independent, but clearly it cannot be complete, because none of the linear combinations of the elements which are left can express the term which was taken away. 7.7

Rectangular waveguides

Waveguides of rectangular cross-section provide an excellent example of application for the previous results, being at the same time technically signiﬁcant and mathematically simple. Let us introduce rectangular coordinates (x, y, z), and place the origin in a corner of the cross-section (Figure 7.4), the x-axis along the longer side, of length a, b being the length of the shorter side. In this reference frame, the dif∗ The proof of this statement is too long to be developed here. Let us say just that it follows from the theory of analytical functions of a complex variable. It can be found rather easily in the literature (e.g., Morse and Feshbach, 1953).

© 2006 by Taylor & Francis Group, LLC

240

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.4 A rectangular waveguide, and its reference frame.

ferential operator which appears in Eq. (7.8) reads ∇2t = (∂ 2 /∂x2 )+(∂ 2 /∂y 2 ), and separation of variables is simply the 2-D version of the procedure we saw in Section 4.2. Let: F(x, y) = X(x) Y (y) . (7.44) Then, the same steps as in Section 4.2 yield a pair of harmonic equations: X = −kx2 X

,

Y = −ky2 Y

,

(7.45)

whose separation constants are linked to the eigenvalue −χ2n = (κ2 − γn2 ) as follows: (7.46) kx2 + ky2 = χ2n . We know from Section 7.4 that χ2n is real and positive, and from Section 7.3 that F must satisfy boundary conditions at ﬁnite distance. This suggests that it may be convenient to write the general integrals of Eq. (7.45) in terms of sine and cosine functions, i.e.: X

= A sin kx x + B cos kx x ,

Y

= C sin ky y + D cos ky y

,

(7.47)

where A, B, C and D are arbitrary constants.† The boundary conditions on the walls are diﬀerent, as we saw before, for TE and for TM modes. For TE modes, Eq. (7.18) holds, and once the variables are separated it reads: dX =0 dx

for

x=0, x=a ,

dY =0 dy

for

y=0, y=b .

(7.48)

Diﬀerentiating Eqs. (7.47) and imposing the conditions of Eqs. (7.48), we † So far, we cannot exclude that either kx or ky could be imaginary, so that Eqs. (7.47) contain one pair of hyperbolic functions with real argument. However, we will see soon that this is incompatible with the boundary conditions on the ideal conducting wall. Hyperbolic functions, on the other hand, cannot to be disregarded in open structures, like a pair of parallel conducting planes, in which case they represent ﬁelds which radiate in the direction along which the structure is open.

© 2006 by Taylor & Francis Group, LLC

RECTANGULAR WAVEGUIDES

241

Table 7.1 Field components of TE and TM modes in a rectangular waveguide. TE MODES nπ mπx nπy Ex = E o cos sin b a b mπ mπx nπy Ey = −Eo sin cos a a b Ez = 0 TM MODES Ex = ηTM Hy

nπ mπx nπy sin cos b a b mπ mπx nπy Hy = H o cos sin a a b Hx = −Ho

Ey = −ηTM Hx Ez = −

Ey ηTE Ex Hy = ηTE χ2 mπx nπy Hz = Eo cos cos jωμ a b Hx = −

χ2 mπx nπy Ho sin sin jω a b

Hz = 0

get immediately A = C = 0, and furthermore: mπ , a nπ sin ky b = 0 , (7.49) =⇒ ky = b where m and n are two integers. Accordingly, the longitudinal magnetic ﬁeld component of the generic transverse electric mode (TEm,n ) reads: nπy mπx cos , (7.50) Hz = H cos a b where H is an arbitrary constant. The other ﬁeld components, obtained from Eq. (7.50) through Eq. (7.3), are shown in Table 7.1, where, to simplify the notation for the transverse components, which are the most commonly used ones, we set H = Eo χ2 /(jωμ). The eigenvalue corresponding to Eq. (7.49) is given by: mπ 2 nπ 2 χ2m,n = + , (7.51) a b which lends itself to the simple geometrical interpretation of Figure 7.5. The corresponding cut-oﬀ frequency, according to Eq. (7.22), is: 1/2 m 2 n 2 1 + . (7.52) fcm,n = √ 2 μ a b sin kx a = 0

=⇒

kx =

Negative values for m and n should not be taken into account, since a change in sign does not aﬀect Eq. (7.52), nor Eq. (7.50). In the ﬁeld components shown in Table 7.1, changes in sign would not be signiﬁcant, as they can always be considered as part of the arbitrary constant Eo . Zero is a legitimate value

© 2006 by Taylor & Francis Group, LLC

242

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.5 Geometrical construction of the modal eigenvalues in a rectangular waveguide.

for either m or n, but not simultaneously for both, because in this case in fact the longitudinal component in Eq. (7.50) would become a constant, and consequently all the transverse components of E and H would vanish. This purely longitudinal magnetic ﬁeld, constant with respect to all coordinates, would be the magnetostatic ﬁeld of an inﬁnitely long rectangular solenoid, not a traveling guided wave. As for the TM modes, what we said for the TE modes remains valid until Eq. (7.47). The boundary condition on the wall is now Eq. (7.17), which once the variables are separated reads: X=0

for

x=0, x=a ,

Y =0

for

y=0, y=b .

(7.53)

Imposing these conditions on Eqs. (7.47) yields B = D = 0, and then Eqs. (7.49) hold again. So, the longitudinal electric ﬁeld component of the generic transverse mode TMm,n is: nπy mπx sin , (7.54) Ez = E sin a b where E is an arbitrary constant. The corresponding transverse ﬁeld components, which can be derived through Eq. (7.3), are also shown in Table 7.1, where, again to simplify the notation for the transverse components, we set E = −Ho χ2 /(jω). The eigenvalue corresponding to Eq. (7.54) is still given by Eq. (7.51), but now neither m nor n can take the zero value, because this would imply E z ≡ 0 in Eq. (7.54) and thus result in an identically zero ﬁeld. Negative values for m and n are also to be left out, for the same reasons as for TE modes. Eq. (7.51) holding for TM and for TE modes implies that, for m · n = 0, the TEm,n and TMm,n modes are degenerate. Any linear combination of two such modes is then a ﬁeld which not only satisﬁes the boundary conditions on the waveguide walls, but also depends exponentially on the longitudinal

© 2006 by Taylor & Francis Group, LLC

RECTANGULAR WAVEGUIDES

243

coordinate z and is characterized by the eigenvalue (7.51). Therefore, this ﬁeld in turn can be looked at as a mode of the waveguide. In Section 7.9 we will provide an example of a problem — losses in waveguides with nonideal metallic walls — where special attention has to be paid in the case of degenerate modes and of their linear combinations. As we said before, for TE modes Eq. (7.51) holds also for m = 0 or for n = 0. These modes are not degenerate with TM modes. Consequently, the smallest among all the values that are allowed for χ2m,n corresponds to m = 1 and n = 0, i.e., it belongs to the TE1,0 mode. This is, in conclusion, the fundamental mode of a rectangular waveguide. Letting m = 1 and n = 0 in Eq. (7.52), we get its cut-oﬀ frequency: fc =

c 2a

,

(7.55)

where c = (μ)−1/2 is the speed of light in the dielectric inside the waveguide. Eq. (7.55) shows that the distance a between the short walls of the waveguide equals a half wavelength (measured in the dielectric that ﬁlls the guide) at the cut-oﬀ frequency of the fundamental mode. This is a hint towards a physical interpretation of the “cut-oﬀ” phenomenon in terms of plane waves which impinge on the walls and are reﬂected on them, giving rise to standing waves in the waveguide. If the distance between two parallel metallic walls is smaller than one half of the plane-wave wavelength, then no standing wave resulting as a superposition of uniform plane waves can satisfy the condition of zero tangential ﬁeld on both walls, whose distance is shorter than the distance between two nodes of the electric ﬁeld, irrespectively of the direction of propagation of the two interfering traveling waves. The conditions on the two walls can be satisﬁed by the superposition of two evanescent waves whose attenuation vectors are parallel to the z axis and whose phase vectors (sat√ isfying |k| > ω μ) are orthogonal to the two walls. For these slow waves (see Chapter 4), two consecutive nodes in their interference pattern are less then λ/2 apart from one other. This interpretation reconﬁrms that the ﬁeld attenuation in the z direction, which cannot be caused by power dissipation since the structure is lossless by assumption, is indeed an evanescence. When, on the contrary, a half wavelength is shorter than the distance between the walls, say a, then two nodes of the electric ﬁeld at a distance a may indeed result from the superposition of two uniform plane waves, provided their phase vectors have a nonvanishing component in the z direction. The net result is a ﬁeld which propagates in the axial direction of the guide. Higher-order modes behave according to the same mechanism, diﬀering from each other in position and in number of the ﬁeld nodes in the interference pattern, which, for m = 0 and n = 0, results from reﬂections on the four walls. Examples of ﬁeld lines, for some of the simplest and yet most important modes of rectangular waveguides, are shown in Figure 7.6.

© 2006 by Taylor & Francis Group, LLC

244

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.6 Field lines of a set of signiﬁcant modes, in a rectangular waveguide.

7.8

Circular waveguides and coaxial cables

The subjects of this section are of technical interest and, at the same time, rather simple as far as calculations are concerned. The mathematical background, although not complicated, is much less dealt with, in standard curricula in electrical engineering and in physics, than that of the previous section. For this reason, Appendix D is devoted to outline it in some detail. Several contents of this section will be useful background points for very important subjects to be dealt with in Chapter 10. Consider an inﬁnitely long structure with a lossless dielectric either inside a circular-cylinder conductor (cylindrical waveguide, Figure 7.7), or between

© 2006 by Taylor & Francis Group, LLC

CIRCULAR WAVEGUIDES AND COAXIAL CABLES

245

Figure 7.7 A circular waveguide, and its reference frame.

Figure 7.8 A coaxial cable.

two circular-cylinder conductors with the same axis (coaxial cable, Figure 7.8). In either case, let us introduce a cylindrical coordinate system (r, ϕ, z), whose z axis is obviously the axis of the cylinders. The diﬀerential operator of Eq. (7.8) then reads: 1 ∂2 ∂2 1 ∂ 1 ∂ ∂ 1 ∂2 + 2 = . (7.56) ∇2t = 2 + r + 2 2 ∂r r ∂r r ∂ϕ r ∂r ∂r r ∂ϕ2 To separate the variables, let: F = R(r) φ(ϕ) ,

(7.57)

insert Eqs. (7.57) and (7.56) into Eq. (7.8), and then multiply both sides times r2 /Rφ. We obtain the following ODE’s, where ν 2 is the separation constant: φ = −ν 2 φ , 1 ν2 R + R + χ2m − 2 R = 0 . r r

(7.58) (7.59)

Eq. (7.58) is, once more, the harmonic equation, whose general integral can be written in either of the following forms: φ(ϕ)

© 2006 by Taylor & Francis Group, LLC

= G sin νϕ + G cos νϕ ,

(7.60)

246

WAVEGUIDES WITH CONDUCTING WALLS φ(ϕ)

= A ejνϕ + B e−jνϕ

,

(7.61)

where G , G (or A, B) are arbitrary constants. The ﬁrst form is more suitable if one wants to stress that the generic solution can be thought of as a superposition of two states of polarization, which correspond to G = 0 and to G = 0, respectively. The second form, on the other hand, represents φ(ϕ) as a superposition of two traveling waves which describe a helical path around the z axis, one in the clockwise sense (B = 0), the other one in the counter-clockwise sense (A = 0). This viewpoint becomes particularly attractive in the limit of geometrical optics, i.e., when a ray picture of propagation is associated with these two waves, as we shall see in Chapter 10. The two polarizations in Eq. (7.60) can then be looked at as standing waves which result from interference of two helical waves (Eq. (7.61)) of equal amplitudes. To be physically meaningful, Eq. (7.57) must be a single-valued function. Therefore, it must be: φ(ϕ + 2lπ) = φ(ϕ) , (7.62) where l can be any integer. Consequently, ν must be an integer. To stress this point, we will change the symbol, in agreement with standard conventions in mathematical handbooks and textbooks, and replace ν with n (= 0, 1, . . .). Negative values for n are not required, for the same reasons as in Section 7.7. Under the assumption χ2m = 0, i.e., if we leave aside the TEM mode, Eq. (7.59) is the Bessel equation of integer order ν = n. Its general integral can be expressed in several forms (see Appendix D). For the problems discussed in this section, the most convenient one is: R(r) = C Jn (χm r) + D Yn (χm r) ,

(7.63)

where Jn (x) and Yn (x) are, respectively, the Bessel function of the ﬁrst kind, of order n, and the Bessel function of the second kind (also called the Neumann function), of order n. Some examples are plotted in Figure D.1 and D.2 (see again Appendix D). Note in particular that, for any order n, Yn tends to −∞ for ν → 0. An inﬁnite value of the longitudinal ﬁeld component F is unphysical, so the term Yn (χm r) must be disregarded, i.e., we must set D = 0 in Eq. (7.63), whenever the z axis (where r = 0) belongs to the region where the e.m. ﬁeld is deﬁned. This occurs in cylindrical waveguides (Figure 7.7), but not in coaxial cables (Figure 7.8). From now on, it is convenient to deal separately with TE and TM modes. For the ﬁrst ones, the longitudinal ﬁeld component F = Hz must satisfy the boundary condition (7.18), which in the present case, keeping track of Eq. (7.57), reads: dR ≡ χm C Jn (χm a) = 0 , (7.64) dr r=a

where a is the radius of the cylindrical wall, and Jn (x) indicates the ﬁrst derivative of Jn (x) with respect to its argument. Let pn,m be the m-th zero of Jn (x). Tables of such zeros can be found in several handbooks. Usually, they

© 2006 by Taylor & Francis Group, LLC

CIRCULAR WAVEGUIDES AND COAXIAL CABLES

247

Table 7.2 Field components of TE and TM modes in a circular waveguide. ∇t = a ˆr

∂ 1 ∂ +a ˆϕ ∂r r ∂ϕ

TE MODES n sin nϕ Er = Eo Jn (χm r) − cos nϕ r cos nϕ Eϕ = Eo χm Jn (χm r) sin nϕ Ez = 0

Eϕ ηTE Er Hϕ = ηTE cos nϕ χ2 Hz = m Eo Jn (χm r) sin nϕ jωμ Hr = −

TM MODES n Jn (χm r) r

Er = ηTM Hϕ

Hr = H o

Eϕ = −ηTM Hr

Hϕ = Ho χm

Ez = −

χ2m Ho Jn (χm r) jω

cos nϕ sin nϕ

sin nϕ − cos nϕ cos nϕ Jn (χm r) sin nϕ

Hz = 0

are labeled also by a second integer, m = 1, 2, . . . (see, e.g., Abramowitz and Stegun, Eds., 1965). Then, the eigenvalue for the TEn,m mode is given by: pn,m , (7.65) a √ and the corresponding critical frequency is fcn,m = pn,m /(2πa μ). The longitudinal ﬁeld component can be written in the form: r cos nϕ Hz = H Jn pn,m (7.66) sin nϕ , a where H is an arbitrary constant. Eq. (7.66) stresses that, for n = 0, the two terms in Eq. (7.60) correspond, as we said before, to two distinct polarizations, giving two modes which are orthogonal (in the sense of Section 7.5) over the range 0 − 2π and must then be looked at as a pair of degenerate modes. The other ﬁeld components, obtained from Eq. (7.66) through Eq. (7.4), are shown in Table 7.2, which also shows the ∇t operator expressed in cylindrical coordinates. As in the previous section, to simplify the notation for the transverse components, which are the most commonly used ones, we set H = Eo χ2 /(jωμ). For the TM modes, the boundary condition of Eq. (7.17) reads, in the case at hand: (7.67) R(a) ≡ C Jn (χm a) = 0 . χn,m =

Let pn,m be the m-th zero of Jn (x). The eigenvalue of the TMn,m mode is then given by: pn,m , (7.68) χn,m = a

© 2006 by Taylor & Francis Group, LLC

248

WAVEGUIDES WITH CONDUCTING WALLS

where again a is the wall radius. The corresponding critical frequency is √ fcn,m = pn,m /(2πa μ), and the longitudinal ﬁeld component is: r cos nϕ Ez = E Jn pn,m , (7.69) sin nϕ a where E is an arbitrary constant. This expression, like Eq. (7.66), stresses that there are two orthogonal and degenerate polarizations for n = 0. The transverse ﬁeld components are also shown in Table 7.2, where again, for the usual reasons, we set E = −Ho χ2 /(jω). Among all the zeros of the functions Jn and of their derivatives Jn — incidentally, all of them are irrational numbers — the smallest one is p1,1 1.84. Hence, the fundamental mode of the circular guides is the TE1,1 mode, whose critical frequency is: fc =

1.84 √ 2πa μ

.

(7.70)

This is of the same order of magnitude as the cut-oﬀ frequency of the fundamental mode in a rectangular waveguide whose broad side equals in length the diameter of the circular waveguide, given by Eq. (7.55). This indicates that the physical nature of the cut-oﬀ phenomenon is not aﬀected by the waveguide geometry in a very signiﬁcant way. As for coaxial cables, we said before that in the presence of the inner conductor the r = 0 axis is out of the region where the ﬁeld is deﬁned, and this implies that the second term in Eq. (7.63) cannot be disregarded. As a matter of fact, it plays a fundamental role in the higher-order modes, but not in the fundamental mode, which, as the structure does not have a simply connected cross-section, is deﬁnitely the TEM mode, according to what we saw in Section 7.4. The TEM mode originates from the Laplace equation, whose solutions, as we said in Section 7.4, obey a well-known uniqueness theorem. We may exploit this fact, and derive the solution of Eq. (7.15) which corresponds to the TEM mode in a heuristic way, without any risk of loss in generality. Let us suppose ν = 0 in Eq. (7.58). This in turn gives φ = constant, and Eq. (7.59) becomes: 1 (7.71) R + R = 0 , r whose general integral is: R = C1 ln r + C2

,

(7.72)

where C1 , C2 are arbitrary constants. Let us set F = T in Eqs. (7.12). Then, we get the following expressions for the ﬁeld of the TEM mode: √ 1 ˆr e−jω μz , E = jω μ/ C1 a (7.73) r H = jω C1

© 2006 by Taylor & Francis Group, LLC

√ 1 a ˆϕ e−jω μz r

.

(7.74)

CIRCULAR WAVEGUIDES AND COAXIAL CABLES 249 √ Apart from the propagation factor exp(−jω μ z), Eq. (7.73) coincides with the electrostatic ﬁeld in a cylindrical capacitor where the per-unit-length √ charge is ±jω (2π) C1 μ on the outer and inner conductors, respectively. Eq. (7.74) coincides — again apart from the propagation factor — with the stationary magnetic ﬁeld in a cylindrical inductor where the longitudinal current intensity is ±2πω C1 in the inner and outer conductors, respectively. The wave impedance in the z direction, for the e.m. ﬁeld given by Eqs. (7.73) and (7.74), is η = μ/. This point may generate many comments, most of which will be made in Section 8.11. For what concerns higher-order modes in coaxial cables, the boundary conditions must be satisﬁed on two cylindrical surfaces, r = a and r = b, a and b being the radii of the outer and inner conductors, respectively. Consequently, one of the two arbitrary constants in the general integral in Eq. (7.63) can be expressed in terms of the other before one proceeds to ﬁnd the eigenvalues. For TE modes, the condition in Eq. (7.18) becomes dR/dr = 0 at r = a and r = b. Dividing side by side the two equations one gets in this way, we ﬁnd the following equation in the unknown χm : Jn (χm a) Y (χm a) = n Jn (χm b) Yn (χm b)

,

(7.75)

which is referred to as the characteristic equation of the TE modes. One can prove that it has a countable inﬁnity of discrete solutions. The corresponding longitudinal ﬁeld component can be expressed as:

cos nϕ Hz (r, ϕ) = H Yn (χm a) Jn (χm r) − Jn (χm a) Yn (χm r) , (7.76) sin nϕ where H is an arbitrary constant and, once again, two orthogonal polarizations for n = 0 are written explicitly. For TM modes, the boundary condition of Eq. (7.17), namely R(a) = R(b) = 0, leads to the following characteristic equation: Yn (χm a) Jn (χm a) = Jn (χm b) Yn (χm b)

,

and then to the following longitudinal ﬁeld component:

cos nϕ Ez (r, ϕ) = E Yn (χm a) Jn (χm r) − Jn (χm a) Yn (χm r) sin nϕ

(7.77)

,

(7.78)

which lends itself to the same comments as Eq. (7.76). It can be shown that, amongst all the solutions χm = 0 of Eqs. (7.75) and (7.77), the smallest one is the ﬁrst solution of Eq. (7.75) for n = 0. The corresponding TE0,1 mode is then the ﬁrst higher-order mode of the coaxial cable. Its eigenvalue lends itself to an approximation which leads to a physically enlightening expression for its critical frequency, namely: fc c/π(a + b) .

(7.79)

This corresponds to saying that at cut-oﬀ the wavelength is equal to the

© 2006 by Taylor & Francis Group, LLC

250

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.9 Field lines of a set of signiﬁcant modes, in a circular waveguide.

length of the mean circle between the conductors. Comments are then similar to those that were given on Eq. (7.55) in Section 7.7. Examples of ﬁeld lines, for some of the simplest and yet most important modes of circular waveguides, are shown in Figure 7.9. 7.9

Waveguides with nonideal walls

An ideal conducting wall is an abstraction, but nevertheless the theory of lossless waveguides is useful in practice. Most of its results apply well to real waveguides whose walls have a large but ﬁnite conductivity, σ ω. The only important quantity of practical interest which is absolutely impossible to evaluate by means of a lossless model is the power loss that a mode propagating in a real waveguide suﬀers because of dissipation, either in the nonideal conducting walls or in the imperfect dielectric. This inconvenience can be overcome using a procedure, to be described in this and in the following section, which is part of a broad-scope chapter of Mathematical Physics, known as “perturbation theory” (see, e.g., Morse and Feshbach, 1953). Let us make ﬁrst some general comment, which may turn out to be useful also in some of the following chapters. To ﬁnd an e.m. ﬁeld in a homogeneous dielectric surrounded by a good conductor in a rigorous way would require the following procedure: ﬁrst, to solve Maxwell’s equations in the dielectric and in the conductor; then, to apply continuity conditions to the tangential components at the interface between the two media. This is always cumbersome, and often impossible, as we may show with a surprisingly simple example. In a rectangular waveguide, continuity conditions should be imposed on four segments belonging to four diﬀerent coordinate lines. This problem has no analytical solution, as we will see again in the case of dielectric waveguides (Chapter 10). What we saw in Section 4.12 on plane waves in good conductors provides a

© 2006 by Taylor & Francis Group, LLC

WAVEGUIDES WITH NONIDEAL WALLS

251

good hint towards an approximate solution. Namely, if what we are actually interested in is mainly the ﬁeld in the dielectric, then we may characterize the conductor in terms of its wall impedance, that is to say in terms of the boundary condition it imposes on the ﬁeld at its surface. Indeed, one fundamental feature of a wall impedance is that it does not depend on the ﬁeld we are looking for in the dielectric. In a waveguide, it is not always simple to assess which value should be given to the surface impedance of walls which consist of a given material, as the geometrical shape can play an important role also. This statement will be clariﬁed by some means of examples in the next section. However, once a suitable wall impedance has been identiﬁed, then to ﬁnd the transmission modes of the waveguide means to solve an eigenvalue equation, similar to Eq. (7.8) but with some major changes. In fact, the boundary conditions are no longer Eqs. (7.17) and (7.18). As a consequence, the eigenvalues are not necessarily real and positive. Still, in most cases this new eigenvalue problem is very hard to solve rigorously. What is much simpler is to use perturbation theory, a procedure which yields a sequence of approximate solutions that converge towards the exact one. To do that, the basic requirement is to know, and to use as the starting point, the exact solution of another eigenvalue problem which is similar enough to the one that one cannot solve exactly. To ﬁnd a general criterion to assess what “similar enough” does actually mean, would require a discussion whose size and depth are well beyond the scope of this book. For the problem at hand, we will restrict ourselves to the following statement. In Section 4.12 we saw that the wall impedance of a good conductor tends to zero as σ → ∞. Accordingly, the exact solutions of a “similar problem” to start from will be, for us, the modes of the lossless waveguides that we studied in the previous sections. Furthermore, let us state that, for all practical purposes, attenuation constants of the modes above cut-oﬀ are obtained with suﬃcient accuracy as ﬁrst-order corrections to the eigenvalues of the lossless waveguide modes. This phrase means that calculation of attenuation constants requires only the knowledge of “zero order” ﬁelds, i.e., of the modes of the lossless guide. However, let us stress that the procedures to calculate ﬁrst-order corrections diﬀer drastically, depending on whether one is dealing with non-degenerate or degenerate modes. There are fundamental reasons behind this diﬀerence, which we will try to clarify by means of an example, later in this section. For a non-degenerate mode, we proceed as follows. We assume that the ﬁeld of the (m, n)-th mode of an ideal waveguide (Zw = 0, and lossless dielectric) is present in a waveguide with the same geometry, but with a ﬁnite wall impedance, Zw (later we will discuss also the case of a lossy dielectric). We calculate the per-unit-length active power which, under these conditions, would ﬂow into the walls. By deﬁnition of wall impedance, it equals |Htan m,n |2 d , (7.80) WP (m,n) = Re[Zw ] 2

© 2006 by Taylor & Francis Group, LLC

252

WAVEGUIDES WITH CONDUCTING WALLS

where is the contour of the waveguide cross-section, while the suﬃx “tan” denotes, as usual, the tangential component on the wall. The power lost in a piece of waveguide of inﬁnitesimal length dz, WP (m,n) dz, is proportional to the active power WT (m,n) ﬂowing through the waveguide cross-section (the “transmitted” power): ¯ tm,n |2 |H dS , (7.81) WT (m,n) = Zm,n 2 S where Zm,n is the wave impedance in the z direction for the (m, n) mode, ¯ tm,n is the transverse component of the magnetic ﬁeld. As the ﬁeld is a and H ¯ tan . ¯ t is proportional to H solution of homogeneous equations, H Proportionality between Eq. (7.80) and Eq. (7.81) entails that, as z increases, the transmitted power decreases exponentially. Hence the e.m. ﬁeld also decreases exponentially vs. z. Power being proportional to the squared modulus of the ﬁeld, the ﬁeld attenuation constant, αm,n , is one half of the power decay constant, so we ﬁnally ﬁnd: αm,n =

WP (m,n) 2 WT (m,n)

.

(7.82)

This result is accurate enough only if higher-order corrections are much smaller than Eq. (7.82). It can be shown that one necessary condition for this to occur is that Eq. (7.82) be much smaller than unity. Note that this requirement is violated as the (m, n) mode approaches its cut-oﬀ frequency, since then WT (m,n) goes to zero and Eq. (7.82) tends to inﬁnity. We may draw the conclusion that near cut-oﬀ it becomes demanding to calculate the attenuation constant accurately enough. Second-order corrections are, in general (see, e.g., Morse and Feshbach, 1953), much more cumbersome to calculate than ﬁrst-order ones, as they require summations which extend, at least in principle, over the whole inﬁnity of modes of the lossless waveguide. The reason why the ﬁrst-order procedure, as we described it, can be inadequate for degenerate modes is not too diﬃcult to explain at this stage. In fact, in each term of the summations involved in the second-order correction, the denominator contains the diﬀerence between the eigenvalue corresponding to the (m, n)-th mode which we are dealing with, and the eigenvalue of another mode. In the degenerate case, at least one among these diﬀerences vanishes, the corresponding term in the summation becomes inﬁnitely large, and it does not make sense any longer to call it a second-order correction. To avoid this, one has to make sure that the numerator of that term in the summation is also zero. In the case of the problem at hand, it can be shown that the numerator is of the type: ¯ tan m,n · H ¯∗ H tan q,r d . (7.83) WP (m,n;q,r) = Zw 2 Eq. (7.83) can be looked at as a measure of how much the two modes, (m, n) and (q, r), interact, i.e., loose their orthogonality, in the lossy structure. In

© 2006 by Taylor & Francis Group, LLC

WAVEGUIDES WITH NONIDEAL WALLS

253

precise terms, Eq. (7.83) is the per-unit-length complex power which ﬂows into the imperfect conductor because of the interplay between the two modes, i.e., is the diﬀerence between the total per-unit-length power loss and the sum of those which can be attributed to the two modes individually. If Eq. (7.83) is non-zero, then the two modes are not orthogonal. If, furthermore, they are degenerate, then one of them cannot propagate without being accompanied by the other. It becomes erroneous to calculate its loss as if it were traveling alone. The correct procedure for two degenerate modes, (m, n) and (q, r), is then the following one. First, one must ﬁnd two linear combinations of these modes, such that for them the integral in Eq. (7.83) is zero. They (see Section 7.5) are also modes of the ideal lossless guide, but, in contrast to the modes (m, n) and (q, r), they remain orthogonal in the lossy waveguide. One can then apply the procedure in Eq. (7.82) to each of these new modes. An example may help to clarify these points. Take the modes TE11 and TM11 in a rectangular waveguide, and suppose that Zw (to be discussed in more detail in the next section) is constant over the waveguide contour and ¯ , respectively, be ¯ and H can be taken out of the integral in Eq. (7.83). Let H the TE11 and the TM11 modal ﬁelds. From Table 7.1, setting all the arbitrary constants equal to one, as they are irrelevant in these calculations, we have: πx πy πy 1 πx ¯ = jβ11 π 1 sin πx cos πy a ˆ cos sin a ˆ cos a ˆz , + H x y + cos aχ211 b a b b a b a b (7.84) 1 πx πy πx πy jπk 1 ¯ = sin cos a ˆx − cos sin a ˆy H , (7.85) χ211 b a b a a b where: χ211

=

π 2 a

+

π 2

,

b

Calculating Eq. (7.83) as: WP = Z w

0

a

η=

μ

Hx Hx∗ dx + Zw

we ﬁnd: WP =

,

k=ω

0

b

√

μ .

Hy Hy∗ dy

Zw β11 kπ 2 (a − b) . 2 χ411 ab

,

(7.86)

(7.87)

(7.88)

Therefore, except for a square waveguide (a = b), the TE11 and TM11 modes are not orthogonal, but coupled to each other by the losses in the wall. To ﬁnd linear combinations of these modes for which the interaction term in Eq. (7.83) vanishes can be phrased as an eigenvalue problem, as it corresponds to diagonalizing the loss matrix: W WP , (7.89) W= WP W

© 2006 by Taylor & Francis Group, LLC

254

WAVEGUIDES WITH CONDUCTING WALLS

where W , W are Eq. (7.80) calculated for the TE11 and the TM11 modes, respectively, while WP is Eq. (7.88). The eigenvalues of the matrix in Eq. (7.89) are, clearly: 1 W + W ± (W − W )2 + WP2 . (7.90) λ1,2 = 2 2 These quantities have to be inserted in the numerator of Eq. (7.82) to obtain the correct values for the attenuation constants of the two orthogonalized modes. The magnitude of the error one makes by using Eq. (7.82) without accounting for the coupling between the degenerate modes can be calculated, as an exercise, using for Zw the value which was found in Section 4.11 for plane waves impinging on a ﬂat good conductor. The subject of wall impedance will be dealt with in more detail in the next section. To conclude this section, let us deal brieﬂy with loss due to an imperfect dielectric inside a waveguide. The concepts remain the same: one assumes that the modal ﬁelds of the lossless structure propagate in the lossy one, and evaluates the corresponding per-unit-length power dissipation. Dividing it by the transmitted power, and then by 2 to keep track of the relationship between ﬁeld amplitude and power, one ﬁnally gets the rate at which the ﬁeld decreases exponentially vs. z. Formulas, on the other hand, are much simpler than those of loss due to imperfect conductors. Indeed, the ﬁnal result turns out to be the same for TE and for TM modes, and also independent of whether the mode is degenerate or not. It reads: αd =

2

ω η 1 − (fc /f )2

,

(7.91)

where the notation c = −j has been adopted for the complex permittivity, η = μ/ is the intrinsic impedance of the dielectric in the lossless waveguide, and fc is the cut-oﬀ frequency of the mode we are dealing with. Necessary conditions for the validity of Eq. (7.91) are , and frequency not too close to the cut-oﬀ frequency.

7.10

On wall impedances

The right value for the surface impedance for good conducting walls of a waveguide can always be found using in the right way the same approach as in Section 4.11 for the similar problem of a plane wave impinging on the surface of a half-space good conductor. Namely, we will ﬁrst introduce in the Helmholtz equation those approximations which are derived from the deﬁnition of a good conductor (σ ω). We will see that then, as a consequence, the wave impedance along the normal to the conductor surface becomes independent of the conditions of incidence on the dielectric-conductor interface. Hence, its value may be taken as the wall impedance. We will see that often it is the same as that of Section 4.11; often, but, surprisingly perhaps, not always. Let us deal ﬁrst with a rectangular waveguide (Figure 7.10), and begin from

© 2006 by Taylor & Francis Group, LLC

ON WALL IMPEDANCES

255

Figure 7.10 Rectangular waveguide.

the walls orthogonal to the x axis. Separating variables as in Section 7.7, we get the following equation: d2 X = −(ω 2 μ − jωμσ + γ 2 − ky2 ) X ≡ −(kx2 − jωμσ) X dx2

.

(7.92)

In a lossless structure, kx = mπ/a can become large only if ω is large. Hence, good conductor means always ωμσ kx2 = |ω 2 μ + γ 2 − ky2 |. Consequently, the quantity in parenthesis on the right-hand side of Eq. (7.92) can be approximated as −jωμσ. (The same warnings as in Section 4.12 hold, for approximations where real numbers are compared with imaginary ones). From now on, everything ﬂows as in Section 4.12, and obviously the same occurs for the walls orthogonal to the y axis. We conclude that the wall impedance to be used in rectangular waveguide is Eq. (4.107), namely: 1/2 πμf . (7.93) Zw = (1 + j) σ For circular waveguides, the appropriate wall impedance is again given by Eq. (7.93). The proof, which requires separation of variables in cylindrical coordinates, is left to the reader as an exercise. One might then jump to the conclusion that Eq. (7.93) is valid in general, for any shape of the waveguide. This is erroneous. In most cases the error is numerically small, but not insigniﬁcant, as it may aﬀect, e.g., how the attenuation depends on frequency. As a counter-example, we will deal brieﬂy with waveguides with elliptical cross-section. We will show that in such a case the wall impedance Zw must be taken as a function of position on the waveguide contour, and furthermore, it takes diﬀerent values for TE and TM modes. The rather unexpected conclusion is that the wall impedance is anisotropic, although the wall is made of an isotropic conductor. To prove all these statements, we ﬁrst introduce (Figure 7.11) a system of elliptical-cylinder coordinates (ζ, η, z), related to cartesian coordinates (x, y, z) as follows: x = q cosh ζ cos η , y = q sinh ζ sin η , (7.94) where 2q is the distance between foci (which are the same for all the elliptical coordinate lines ζ = constant). Obviously, these foci are also those of the

© 2006 by Taylor & Francis Group, LLC

256

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.11 An elliptical coordinate system.

waveguide contour, which is then a coordinate line, ζ = ζ0 . Also, q is related to the ellipticity e, and to the lengths of the major and minor axes of the contour ellipse, 2a and 2b, respectively, as: 2 1/2 b q . (7.95) = e= 1− a a To ﬁnd the correct wall impedance, one must go from the beginning through the procedure we outlined before. Accordingly, the ﬁrst step is to solve Eq. (7.8) under the good-conductor approximation, σ ω. In elliptical coordinates, Eq. (7.8) reads (see Morse and Feshbach, 1953): ∂2F ∂2F + + χ2n q 2 (sinh2 ζ + sin2 η) F = 0 . ∂ζ 2 ∂η 2

(7.96)

We can ﬁnd its general integral (see, e.g., Abramowitz and Stegun, Eds., 1965), and then, similarly to what we did in Section 7.8, we can impose on it the restriction that it remains ﬁnite for ζ → 0. The net result can be written as:

(4) (7.97) F = M c(n) (ζ ; ρ) A ce(n) (η ; ρ) + B se(n) (η ; ρ) , where the symbols have the following meanings: • A, B = arbitrary constants; n = order of the functions; (4)

• M c(n) = radial Mathieu function of the fourth kind;

© 2006 by Taylor & Francis Group, LLC

ON WALL IMPEDANCES

257

• ce(n) = even angular Mathieu function of the ﬁrst kind; • se(n) = odd angular Mathieu function of the ﬁrst kind; • ρ = χ2n q 2 /4. The assumption of a good conductor implies χ2n −jωμσ in Eq. (7.96), and also, in most cases, |χn | a 1. Then, we can make use of an asymptotic ex(4) pression for M c(n) , which is valid for |χn q cosh ζ| → ∞. This is a necessary step in order to obtain wave impedances in the radial direction that are independent of the order n, i.e., independent of the mode under examination. Otherwise, we could not call it a wall impedance. The expressions one gets in this way are diﬀerent depending on whether we ˆz or Etan = Eη a ˆη . In detail, it can be shown (Falciasecca et take Etan = Ez a al., 1971; Lewin et al., 1977) that, for ζ = ζ0 , i.e., on the waveguide wall, it is: 1/2 πμf 1 Eη = (1 + j) , (7.98) ZPTE ≡ Hz σ (1 − e2 cos2 η)1/2 ZPTM

≡

Ez = (1 + j) Hη

πμf σ

1/2

(1 − e2 cos2 η)1/2

.

(7.99)

It is evident that these expressions are not the same as Eq. (7.93). Changes are due, from a mathematical viewpoint, to the fact that the metric coeﬃcient of the coordinate η changes along the coordinate line (ζ = ζ0 ) where we calculate partial derivatives in order to express the ﬁeld components. In a rectangular or circular waveguide, on the contrary, the metric coeﬃcient remains constant along the boundary. From a physical viewpoint, changes with respect to Eq. (7.93) can be explained by looking at the current ﬂow lines inside the conducting wall. We saw in Section 4.12 that, when a uniform plane wave impinges on a conducting half-space, the propagation vector of the transmitted wave is essentially perpendicular to planar interface between the two media. Consequently, the current density J = σE is parallel to the planar interface, at any depth inside the conductor. An equivalent surface current — which is a concept intimately related to that of surface impedance — is easy to deﬁne, given that all ﬂow lines are parallel. In a rectangular waveguide, if we neglect the corner eﬀects, the situation remains essentially the same. Current ﬂow lines in the walls are everywhere parallel to the wall surfaces. The same occurs again in a circular waveguide, where the current ﬂow lines have only an axial and an azimuthal component, always parallel to the surface. What we ﬁnd in an elliptical waveguide is completely diﬀerent. The current ﬂow lines, at the generic P point, have nonvanishing components along the coordinate elliptical line which passes through P , and consequently (see Figure 7.11) they are not parallel to the waveguide contour. At the same time, the distance between two given coordinate ellipses (which delimit a current ﬂow tube) changes as we move in the azimuthal direction. Also the Poynting vector inside the wall can be shown not to be perpendicular everywhere to the surface of the wall. All these comments justify why Eqs. (7.98) and (7.99)

© 2006 by Taylor & Francis Group, LLC

258

WAVEGUIDES WITH CONDUCTING WALLS

Figure 7.12 Schematic view of a helical waveguide.

are surprisingly diﬀerent from the wall impedances we have found so far. In particular, we note that: • none of the wall impedances in an elliptical waveguide is the same as Eq. (7.93); • the wall impedance is anisotropic, since Eqs. (7.98) and (7.99) are not the same; • all the impedances on the elliptical wall change as we move on the waveguide contour. The diﬀerence between Eqs. (7.98) and (7.99), on one hand, and Eq. (7.93), on the other hand, does not induce any relevant diﬀerence if the purpose is to ﬁnd the reﬂection coeﬃcient of an elliptical object (such as an antenna reﬂector, for example). On the contrary, with a “small” diﬀerence repeated many times, as in the case of a propagating mode which rebounds again and again on a waveguide wall, then the total diﬀerence on the attenuation constant can become quite signiﬁcant (Lewin et al., 1977). Experimental results on the frequency dependence of loss reconﬁrm that the correct values are those which one obtains with the procedure we outlined in this section. 7.11

Hybrid modes

In the two previous sections, the purpose of the wall impedance was to model more realistically structures which had been analyzed before with an idealized lossless model. The lossless model was a good starting point for the perturbation method. Consequently, we kept on using the terms TE and TM for the modes of waveguides with ohmic losses, although most of them were, strictly speaking, neither TE nor TM. Wall impedances are used in a slightly diﬀerent sense in modeling structures which do not have an ideal conducting wall, not even as a ﬁrst approximation. A typical example in this sense are the helical waveguides, structures (see Figure 7.12) whose walls are made of coiled wires or ribbons, the spacing between two coils being usually much smaller than the thickness of the wire or ribbon. These structures strongly discriminate the low attenuation of one family of modes whose currents ﬂow essentially along the wires or ribbons (the TE0,n family) from the high losses for all other modes. Their behavior can be studied using as a model an anisotropic wall, whose impedance, as a ﬁrst approximation, can be taken equal to zero in the azimuthal direction,

© 2006 by Taylor & Francis Group, LLC

HYBRID MODES

259

diﬀerent from zero in the longitudinal direction. Apart from the interest in their practical applications, which is nowadays much less than a few decades ago, what we wish to point out in this section is a concept, namely that, when wall impedances are ﬁnite, a complete set of solutions can no longer consist of only TE and TM modes. On the contrary, it is necessary to take into account other ﬁelds, which are called hybrid modes, are more complicated than TE and TM modes, and will play a dominant role in dielectric waveguides, to be studied in Chapter 10. To show this, let us start from the general expressions in Eqs. (7.11) and (7.12), and suppose that the anisotropic wall, with a circular cross-section of radius a, sets the following boundary conditions: Eϕ

= 0 ,

Ez

= ZHϕ

(7.100) ,

Z = 0 .

(7.101)

First of all, let us verify whether these conditions can both be satisﬁed by a TM ﬁeld, as given by Eq. (7.12). In cylindrical coordinates, after separation of variables, Eq. (7.100) becomes: ∂T ∝ R(a) φ (ϕ) = 0 , (7.102) ∂ϕ r=a

where, as usual, the prime indicates the derivative with respect to the argument, while Eq. (7.101) becomes: ∂T 2 2 . (7.103) (κ − γ ) T (a) = jω Z ∂r r=a

From what we saw in Section 7.8 it follows that, for an azimuthal order n = 0, φ (ϕ) is not identically zero. Therefore, Eq. (7.102) can be satisﬁed only for R(a) = 0. This condition implies T (a) = 0, and if we insert this into Eq. (7.103) we get R (a) = 0. Nontrivial solutions of the Bessel equation (7.59) do not have any value for their argument such that the function and its ﬁrst derivative, both vanish. Thus, Eqs. (7.102) and (7.103) are mutually incompatible, except for n = 0 (φ = constant). This means that TM0,n modes do exist in this type of waveguide, but, on the contrary, there are no TM modes that depend on the azimuthal coordinate. As for TE modes, we can show in a similar way that they cannot exist, except again for n = 0. In fact, assuming Ez ≡ 0 implies, via Eq. (7.101), Hϕ = 0, and because of Eq. (7.11) this entails: ∂L ∝ R(a) φ (ϕ) = 0 . (7.104) ∂ϕ r=a At the same time, the condition Eq. (7.100), again through Eq. (7.11), becomes: ∂L ∝ R (a) φ(ϕ) = 0 , (7.105) ∂r r=a and from now one can proceed as in the previous case.

© 2006 by Taylor & Francis Group, LLC

260

WAVEGUIDES WITH CONDUCTING WALLS

Evidently, TE0,n and TM0,n modes do not form a complete set, as the waveguide can certainly propagate ﬁelds that depend on ϕ. There must be further solutions of Maxwell’s equations. The decomposition theorem of Section 7.2 is quite general, and therefore one should be able to express any solution as a sum of a TE ﬁeld, of the type in Eq. (7.11), plus a TM ﬁeld, of the type in Eq. (7.12). Then, let L = A1 R1 φ1 and T = A2 R2 φ2 , with A1 and A2 arbitrary constants. The boundary conditions (7.100) and (7.101) become: nγ A2 R2 φ2 = 0 , (7.106) −A1 χ R1 φ1 + a jγ A1 n R1 φ1 , (7.107) −χ2 A2 R2 φ2 = Z jω A2 χ R2 φ2 + ωμa where primes indicate, as usual, derivatives with respect to arguments (note that the argument of Ri is χr, while that of φi is nϕ); furthermore, −χ2 = κ2 − γ 2 . From the system of Eqs. (7.106) and (7.107), we can: • eliminate one arbitrary constant Ai , showing that there is only one degree of freedom for amplitude and phase of each mode in this family; • derive an equation in implicit form which links the propagation constant γ with the angular frequency ω. It is usually referred to as the characteristic equation or the dispersion relationship of the structure. A simple way to ﬁnd it is to set the determinant of the system Eqs. (7.106) and (7.107), in the two unknowns Ai , equal to zero: jγZ R1 F −χ2 R2 − jωχ ZR2 F2 n 1 = a ωμaχ R1 F1 nγR2 F2

.

(7.108)

For this equation in the unknown χ to admit solutions, it is necessary that the dependence on ϕ be the same on both sides, i.e.: F1 F2 =± F1 F2

.

(7.109)

The two choices for the sign in Eq. (7.109) lead to two distinct sets of solutions for Eq. (7.108), resembling what we will see in Chapter 10. There is a need for a new terminology, as we cannot call TE and TM these new modes of helical guides. However, no general agreement on new names has been found throughout the technical literature. In Chapter 10, on the other hand, we will see that there is a general consensus on one nomenclature for the hybrid modes of cylindrical dielectric waveguides. 7.12

Further applications and suggested reading

We said at the beginning of this chapter that one of the reasons why we deal here with waveguides with conducting walls before tackling dielectric waveguides is because they came ﬁrst in time. In fact, waveguides surrounded

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

261

by metal walls were among the key ingredients in the development of radar, which was the ﬁrst example of practical use of microwaves, which were also referred to, at that time, as hyperfrequencies. A popular misconception is that those early developments were essentially empirical, based on a cut-and-try strategy. Those who want to see how advanced and sophisticated were, on the contrary, the ﬁrst major theoretical models in the area, should take at least a quick tour through the volumes known as the MIT Radiation Laboratory Series, published by McGraw-Hill in the late 1940’s. Readers who are familiar with the names of the most prominent theoretical physicists of that time will recognize some of them among the authors of these books. This was because during World War II most physicists in the United States became engaged in applied research for military aims — not only the very famous “Manhattan Project,” but also other developments, including of course radar. Accordingly, mathematical tools which were advanced, for that time, and were in use in ﬁelds such as quantum mechanics, were promptly and successfully transferred to modeling microwave devices. Because of this tidy and neat mathematical background, waveguides with ideal conducting walls are an ideal tutorial introduction to guided-wave propagation, much simpler than through any other similar structure. Furthermore, waveguides induce a very beneﬁcial step in the learning process of telecommunication engineers and other students with similar curricula. This step consists of the fact that students who have familiarized themselves with waveguides begin to look at ﬁelds, transmission lines, and electrical circuits, in a uniﬁed way, considering them as three ways to model the same physical reality. This cross-fertilization process can still be appreciated, as a major contribution, in at least one of the books in the previously quoted MIT Series, namely in Montgomery, Dicke and Purcell (1948). Readers who, at this stage, do not yet grasp the meaning and the importance of this comment, should come back to it after studying the next chapter. For decades, the word microwaves has had two meanings. It stood for a frequency range and, at the same time, for a set of technologies, which were then based mainly on waveguides. At present, this is no longer true. At those frequencies which are still called microwaves, up to an upper limit which is not sharply deﬁned, but is deﬁnitely higher than 10 GHz, and still growing, striplines dominate by far and large over waveguides. As we said when we introduced the TEM mode, striplines, although similar to those structures where the TEM mode propagates, are not easy to model analytically, as they require some approximation. For this reason, they have been deferred to the next chapter. While striplines were displacing waveguides from many applications, most of the electronics gradually became solid-state. Historically, microwave electronics consisted of special tubes which encompassed waveguiding structures, like sequences of cavities, helices, periodic combs, etc. The reader who wants to become familiar with modern microwave technology can be referred to textbooks like Baden Fuller (1990), Gardiol (1984) and Cronin (1995), and to more advanced books like Collin (1966/1991). Those who want to learn

© 2006 by Taylor & Francis Group, LLC

262

WAVEGUIDES WITH CONDUCTING WALLS

about classical microwave tubes, still in use for some applications, may refer to Pierce (1950). Conducting-wall waveguides have still some practical application, in addition to being pedagogically useful. Compared to striplines, their main advantage is that all the power they convey travels in a well deﬁned, closed region. This can be crucial in non-telecommunication applications, like industrial heating, domestic cooking, medical diagnosis and therapy, etc., where shielding humans from the ﬁeld is essential, to avoid unwanted interaction with biological tissues. Industrial, scientiﬁc and medical (ISM) applications of microwaves are widely described in the literature (e.g., see Roussy and Pearce, 1995, and references therein). Another case where waveguides are used, because they can convey very high powers with relatively low loss (especially if operated in the TE0n modes of either circular or helical waveguides), is plasma heating. Plasmas, as well known, have been and still are intensively explored, as the way towards generating energy by controlled nuclear fusion, and microwave heating is looked at as one of the possible ways to reach the threshold temperatures that are needed. The ﬁeld of plasma diagnostics also relies heavily on microwaves up to about 100 GHz. For these subjects, the reader is referred to Heald and Wharton (1978), Hutchinson (1987), Lochte-Holtgreven (1995) and references therein. Finally, some kinds of particle accelerators, for high-energy physics experiments, contain parts — typically, pipes where particles gain kinetic energy at the expenses of a radio-frequency or microwave ﬁeld — which can be looked at, and designed, as waveguiding structures. An excellent reading for those who are interested in these problems is Panofsky and Phillips (1962). Problems 7-1 A rectangular waveguide is interrupted on the plane z = L by a ﬂat ideal conductor, i.e., is a short-circuited waveguide. There are no sources between z = 0 and z = L. Assume that the waveguide wall and the dielectric inside it are all lossless. At a frequency such that only the TE10 mode is above its cut-oﬀ frequency, state whether the real and the imaginary parts of the ﬂux of the Poynting vector through the plane z = 0 vanish or not, specifying whether the answers depend on the length L. Repeat the problem assuming that the TE10 mode is below its cut-oﬀ frequency. 7-2 Derive two relationships which diﬀer from Eqs. (7.38) and (7.39) only because Etk and Htk are replaced by E∗tk and H∗tk , respectively. Comment on their physical meanings. 7-3 Estimate the error if Eq. (7.82) is used for a degenerate mode without accounting for its coupling with the other degenerate mode. Use for the wall impedance Zw the value found in Section 4.12 as the wave impedance for plane waves in a ﬂat conductor.

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

263

7-4 Return to Problem 1-17, and show that, in general, the ﬁeld of any TM mode can be derived from a purely longitudinal magnetic vector potential ˆz . A = Az a Derive the constant of proportionality between Az and Ez . Why is it diﬀerent from −jω ? 7-5 Return to Problem 1-18, and show that, in general, the ﬁeld of any TE mode can be derived from a purely longitudinal electric vector potential F = Fz a ˆz . Derive the constant of proportionality between Fz and Hz . Why is it diﬀerent from −jω ? 7-6 An important theorem of circuit theory, usually referred to as Foster’s theorem, states the following. Let X(ω) be the input reactance of a perfectly lossless two-terminal lumped-element network. Then, ∂X/∂ω > 0 for any network, where ω is, as usual, the angular frequency. a) Verify that the wave impedances of TE and TM modes below cut-oﬀ obey this rule, for waveguides of any shape and any size. b) For any mode of a rectangular waveguide, calculate the magnetic and electric energies stored in a semi-inﬁnite waveguide, z > 0, and, using the reactive power term in Poynting’s theorem, verify again that these results agree with Foster’s theorem. 7-7 Show that in the presence of loss or gain, i.e., when the real parts of the propagation constants of modes above cut-oﬀ are nonvanishing, two generic modes are not power-orthogonal, i.e., they do not satisfy Eq. (7.42). 7-8 Design an air-ﬁlled rectangular waveguide (i.e., calculate the lengths of its sides, a and b) so that the frequency f = 3 GHz is 1.3 times greater than the cut-oﬀ frequency of its fundamental mode, and 0.7 times the cut-oﬀ frequency of the ﬁrst higher-order mode. 7-9 A rectangular waveguide of width a = 23 mm and height b = 11.5 mm can propagate the TE10 mode at a frequency of 6 GHz, only if ﬁlled with a suitable dielectric. a) Find the range of values for the relative permittivity, r , for which the waveguide propagates only the fundamental mode. b) Study how the phase constant of the fundamental mode, as a function of frequency, varies when the permittivity of the medium varies within this range of values. 7-10 A rectangular waveguide whose dimensions are the same as in the previous problem, ﬁlled with air (r = 1), goes from a receiving antenna to a receiver for a total distance L = 5 m and conveys a signal whose spectrum is

© 2006 by Taylor & Francis Group, LLC

264

WAVEGUIDES WITH CONDUCTING WALLS

centered on f = 1.5 fc , where fc is the cut-oﬀ frequency of the fundamental mode. Evaluate the diﬀerential phase and group delays (i.e., the diﬀerences in velocities times the distance) over this length, for two frequencies which diﬀer from the center of the spectrum by ±5%. 7-11 A standard WR-15 rectangular waveguide (3.76 mm×1.88 mm) ends abruptly in an aperture. Assume that the wave impedance beyond the aperture equals the free-space intrinsic impedance. Find the reﬂection coeﬃcient for the TE10 mode at a frequency of 60 GHz, and the percentage of active power which is reﬂected back into the waveguide. (Comment. This problem is much simpler to solve after reading the following chapter.) 7-12 Two rectangular waveguides with diﬀerent major sides (a1 < a2 ) and equal minor sides (b1 = b2 ) are butt-jointed in the z = 0 plane. Waveguide 1 propagates the TE10 mode only, which travels in the positive z direction and impinges on the plane of discontinuity. Find which modes of waveguide 2 are excited, and which are their amplitudes, if a) the two waveguides are aligned in a corner (x = 0); b) they are aligned at their centers (the point x = a1 /2 of waveguide 1 is the same as the point x = a2 /2 of waveguide 2). (Hint. Expand an arc of cosine function into a set of harmonic functions with a diﬀerent period. Make use of orthogonality.) 7-13 Repeat the previous problem for two rectangular waveguides with equal major sides (a1 = a2 ) and diﬀerent minor sides (b1 < b2 ). Waveguide 1 propagates the TE10 mode only. a) The two waveguides are aligned in a corner (y = 0). b) They are aligned at their centers (the point y = b1 /2 of waveguide 1 is the same as the point y = b2 /2 of waveguide 2). 7-14

Consider a lossless circular waveguide.

a) Making use of Bessel function identities and integrals given in Appendix D, prove explicitly that its modes are orthogonal. b) Repeat the same problem for the modes of a coaxial cable. 7-15 Find the modes of a semicircular-waveguide, and those of a waveguide whose cross-section is a quarter of a circle, always assuming that the walls are ideal electric conductors. 7-16 Starting from the wave equation in a good conductor, and using separation of variables in cylindrical coordinates, show that the wall impedance

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

265

given by Eq. (7.93) is applicable when dealing with circular waveguides and coaxial cables. 7-17 Consider a circular waveguide with a low-loss wall, and use Eq. (7.93) for the wall impedance. a) Show that the perturbation formula for the attenuation, Eq. (7.80), indicates that the theoretical losses of the TE0,m modes are monotonically decreasing functions of the frequency, although the wall impedance increases in proportion to the square root of the frequency. b) Provide a physical explanation for this result. (Hint. Study the current density of these modes in the walls, as a function of frequency, for a constant transmitted power.) c) Why none of the modes of a rectangular waveguide behaves in the same way ? 7-18 A circular waveguide whose modes are all below their cut-oﬀ frequencies may be used as an attenuator in a coaxial cable: the internal conductor is removed over a certain length, and short-circuited to the external one at the two extremes of that length. a) Neglecting backward waves, i.e., assuming a purely exponential z-dependence for the TE11 mode below cut-oﬀ, write a “calibration chart” of an attenuator whose diameter is 20 mm and whose length is 50 mm, i.e., its attenuation (in decibels) vs. frequency, from dc to the cut-oﬀ frequency of the TE11 mode. b) In these calculations, all the structures have been assumed to be lossless. Consequently, attenuation in the device cannot be caused by power dissipation. Where does the lost power go ? Is it rigorous to accept, under these circumstances, the previous assumption that backward waves are negligible ? 7-19 In coaxial cables, the diameters of the inner and outer conductors which are in current use, as standards, in the design of the so-called Type N connectors, are 0.120 inches and 0.276 inches, respectively. Assuming that the dielectric between them has the same permittivity as air, evaluate (approximately) the uppermost frequency at which only the TEM mode propagates in the cable. Discuss how this cut-oﬀ frequency scales vs. relative permittivity of the dielectric. 7-20 A so-called microwave Y -junction is made of three identical rectangular waveguides, as sketched in Figure 3.4. Assume that they are lossless, and that only the TE10 mode is above cut-oﬀ. When waveguide 1 is connected to a generator, then the two waves at the outputs of waveguides 2 and 3 are equal in amplitude and in phase. By reciprocity, when two waves of equal amplitudes and phases are sent into waveguides 2 and 3, a wave in the TE10 mode emerges

© 2006 by Taylor & Francis Group, LLC

266

WAVEGUIDES WITH CONDUCTING WALLS

from waveguide 1. Using symmetry arguments (see Section 3.9), discuss what happens when: a) two waves of equal amplitude but phase-shifted by π radians are sent into waveguides 2 and 3; b) one wave is sent into waveguide 2 whereas no wave is sent into waveguide 3.

© 2006 by Taylor & Francis Group, LLC

References Abramowitz, M. and Stegun, I.A. (Eds.) (1965) Handbook of Mathematical Functions. Dover, New York. Baden Fuller, A.J. (1990) Microwaves: An Introduction to Microwave Theory and Techniques. 3rd ed., Pergamon, Oxford, New York. Collin, R.E. (1966) Foundations for Microwave Engineering. McGraw-Hill Physical and Quantum Electronics Series, McGraw-Hill, New York. Collin, R.E. (1991) Field Theory of Guided Waves. IEEE Press, New York. Cronin, N.J. (1995) Microwave and Optical Waveguides. Institute of Physics, Bristol, Philadelphia. Dicke, R.H. and Wittke, J.P. (1960) Introduction to Quantum Mechanics. AddisonWesley, Reading, MA. Falciasecca, G., Someda, C.G. and Valdoni, F. (1971) “Wall impedances and application to long-distance waveguides”. Alta Frequenza, 40, 426–434. Gardiol, F.E. (1984) Introduction to Microwaves. Artech House, Dedham, MA. Heald, M.A. and Wharton, C.B. (1978) Plasma Diagnostics with Microwaves. R.E. Krieger Pub. Co., Huntington, New York. Hutchinson, I.H. (1987) Principles of Plasma Diagnostics. Cambridge University Press, Cambridge, UK. Lewin, L., Chang, D.C. and Kuester, E.F. (1977) Electromagnetic Waves and Curved Structures. P. Peregrinus, Stevenage, UK. Lochte-Holtgreven, W. (1995) Plasma Diagnostics. American Vacuum Society Classics, AIP Press, New York. Montgomery, C.G., Dicke, R.H. and Purcell, E.M. (Eds.) (1948) Principles of Microwave Circuits. McGraw-Hill, New York. Reprinted by Peter Peregrinus, London, for IEE (1987). Morse, P.M. and Feshbach, H. (1953) Methods of Theoretical Physics. McGraw-Hill, New York. Panofsky, W.K.H. and Phillips, M. (1962) Classical Electricity and Magnetism. 2nd ed., Addison-Wesley, Reading, MA. Pierce, J.R. (1950) Traveling-Wave Tubes. Bell Telephone Laboratories Series, Van Nostrand, New York. Roussy, M.G. and Pearce, J.A. (1995) Foundations and Industrial Applications of Microwave and Radio Frequency Fields: Physical and Chemical Processes. Wiley, New York.

267 © 2006 by Taylor & Francis Group, LLC

CHAPTER 8

Waves on transmission lines 8.1

Introduction

We will use the phrase “transmission line” for any electrical system which, in one geometrical dimension, i.e., along the axis labeled z, extends signiﬁcantly more than along the others, and more than the wavelength. When we talk of “common” or “ordinary” transmission lines, they consist of two or more conductors, whose cross-sections, orthogonal to the preferential z direction, are separated by dielectrics. Simple examples are parallel wire pairs, and coaxial cables. More sophisticated examples are provided by microwave stripline circuits (see Section 8.11), the main diﬀerence coming from the fact that often their dielectric is not homogeneous. When the conductors are connected through lumped∗ passive components (resistors, capacitors, inductors, transformers), and fed by lumped voltage or current generators, then usually they can convey an electric current at arbitrarily low frequencies. This is the reason why the classical electricalengineering approach to transmission lines starts oﬀ as an extension of lumpedcircuit theory, i.e., begins by postulating Ohm’s law and Kirchhoﬀ’s laws. We will take it for granted that the readers know them from basic physics classes, and follow this standard approach in the ﬁrst part of the present chapter. We will shortly be able to show that waves may propagate along these lines. They are expressed in terms of voltages and currents, so they must be looked at as being electromagnetic waves. Our subsequent task will be to reconcile these waves on transmission lines with those e.m. waves that we have studied so far. This will be done also due to the fact that we have used very frequently, for our waves, one of the most typical concepts of circuit theory, that of impedance. Once the link is established between common transmission lines and what we saw in the previous chapters, then the transmission-line formalism, developed for actual lines, becomes suitable, and extremely useful, as an abstract model for transmission/reﬂection phenomena with any kind of e.m. wave, no matter whether free or guided. These subjects will be dealt with in the last sections of this chapter. ∗ “Lumped” means that the geometrical size of an element is negligible with respect to wavelength, at all frequencies of interest, so that none of the electrical quantities which can be deﬁned on that element, like voltage and current, experiences a signiﬁcant phase delay from one point to another within that element. 269 © 2006 by Taylor & Francis Group, LLC

270

WAVES ON TRANSMISSION LINES

Figure 8.1 Equivalent circuit for a section of inﬁnitesimal length of a uniform transmission line.

8.2

Uniform transmission lines

To retain simplicity, we will deal only with transmission lines whose characteristics are constant along the direction of propagation, z. We will refer to them as uniform transmission lines. In any transmission line, energy can be stored along its length under electric and under magnetic form. Some energy can also be dissipated along the line, under the form of ohmic losses in non-ideal conductors and in non-ideal dielectrics. Intuitively, then, it should be possible to model the electromagnetic behavior of any line in terms of distributed parameters. Magnetic energy stored in an inﬁnitesimal length dz can be thought of as being stored in an inﬁnitesimal inductance. Ohmic losses in the conductors, over an inﬁnitesimal length, can be attributed to an inﬁnitesimal resistance. The same can be said for the electric energy, stored in an inﬁnitesimal capacitance, and for losses in the dielectric, modeled as an inﬁnitesimal shunt conductance. In a uniform transmission line, these properties must, by deﬁnition, be constant in the z direction. Consequently, we can assume, as a starting point, that the suitable model for an inﬁnitesimal length dz is the “equivalent circuit” shown in Figure 8.1, where the parameters r, l, c, g are independent of z. We may then deﬁne a per-unit-length series impedance Z (expressed in ohms per meter), as the series of the per-unit-length inductance l (expressed in henrys per meter) and the per-unit-length resistance r (expressed in ohms per meter), Z = r + jωl. Similarly, we deﬁne a per-unit-length shunt admittance Y = g + jωc (expressed in inverse ohms per meter) as the parallel of the per-unit-length capacitance c (expressed in farads per meter) and the perunit-length conductance g (expressed in inverse ohms per meter). Note that this presentation of the “equivalent circuit” is rather phenomenological: at this stage, it should be regarded as a postulate. However, it will be justiﬁed in more physical terms in Section 8.10. We postpone this justiﬁcation because it is less urgent than the basic results which can be derived from the circuit itself. Applying Ohm’s law and Kirchhoﬀ’s laws (for voltages around a closed path and for currents in a node, respectively) for time-harmonic voltages and currents at steady state, from the circuit of Figure 8.1 we get the following

© 2006 by Taylor & Francis Group, LLC

UNIFORM TRANSMISSION LINES

271

diﬀerential equations, where the voltage V and the current I are unknown functions of the longitudinal coordinate z: dV = −ZI dz

,

dI = −Y V dz

,

(8.1)

where: Z = (r + jωl) , Y = (g + jωc) . (8.2) Eqs. (8.1) are frequently referred to as the telegraphist’s equations. Diﬀerentiating them again with respect to z and replacing in one of them the ﬁrst derivative by means of the other one, we get: d2 I d2 V = ZY V , = ZY I . (8.3) 2 dz dz 2 Eqs. (8.3) are sometimes referred to as the telephonist’s equations. They can also be looked at as one-dimensional forms of the Helmholtz equation. The general integral of either one can be expressed in terms of exponential functions: V (z) = V+ e−Γz + V− eΓz

I(z) = I+ e−Γz + I− eΓz

,

,

(8.4)

where:

Γ = + (r + jωl) (g + jωc) . (8.5) Since Γ is in general complex, each term in Eqs. (8.4) looks like a wave, traveling either in the forward or in the backward direction along the z axis. Consequently, Γ is referred to as the propagation constant of the transmission line. Current symbols are Γ = α + jβ, with α and β real, referred to as the attenuation constant and the phase constant of the transmission line. Eqs. (8.4) cannot be looked as the general integral of the system in Eqs. (8.1), as long as the four arbitrary constants are independent. Indeed, the order of the system in Eqs. (8.3) being higher, it has also spurious solutions, i.e., solutions which do not satisfy Eqs. (8.1). To restrict ourselves to the actual solutions, we must introduce Eqs. (8.4) into Eqs. (8.1). So, we ﬁnd that the constants must be related among themselves as follows: I+ =

V+ ZC

The quantity:

,

I− = −

V− ZC

.

(8.6)

" ZC =

(r + jωl) (g + jωc)

,

(8.7)

is referred to as the characteristic impedance of the transmission line. In general, it is complex, and this indicates that, in a uniform transmission line which is not further speciﬁed, a voltage and a current waves which travel in the same direction† are not in phase, at the same distance from the origin along the line, z. † To avoid an unnecessary cumbersome language, on many occasions throughout this chapter we will use the word “direction” with the meaning of “oriented direction,” or “sense”.

© 2006 by Taylor & Francis Group, LLC

272

WAVES ON TRANSMISSION LINES

The quantity: V (z) I ∗ (z) , (8.8) 2 where the star denotes as usual the complex conjugate, is the complex power across the generic cross-section of the line, ﬂowing in the positive direction of the z axis (i.e., active power is actually ﬂowing in the positive z direction if the real part of Eq. (8.8) is positive, in the negative direction if Re[Pc ] < 0). Inserting Eq. (8.6) into Eq. (8.8), we see that Pc consists of a term as∗ /2) exp(−2αz), plus a term associsociated with the forward wave, (V+ I+ ∗ ated with the backward wave, (V− I− /2) exp(2αz), plus two “cross” terms, ∗ ∗ /2) exp(−j2βz) and (V− I+ /2) exp(j2βz). (V+ I− The remaining two arbitrary constants in the expressions for voltage and current have to be determined by two boundary conditions. Normally, these are imposed by a generator at one end of the line, and a load at the other end. The rest of this chapter will be devoted mainly to how these boundary conditions aﬀect voltage and current along the entire transmission line. Pc (z) =

8.3

Impedance transformation along a transmission line

The ratio between voltage and current, at any value of z along the transmission line, deﬁnes an input impedance: Z(z) =

V+ e−Γz + V− eΓz V (z) = ZC I(z) V+ e−Γz − V− eΓz

.

(8.9)

The input admittance is, obviously, the inverse of Eq. (8.9). Suppose that the line is terminated, at z = 0, on a linear, passive twoterminal network, whose impedance ZL = V (0)/I(0) we refer to as the load impedance. Incidentally, note that normally we will take the positive z direction from the generator to the load, then, choosing the load terminal section as the origin for the z coordinate implies that the line is in the region z ≤ 0. The input impedance at any distance z can be obtained, as a function of z and of the load impedance, imposing Z(0) = ZL in Eq. (8.9). The result is easily derived as: Z(z) = ZC

ZL cosh Γz − ZC sinh Γz ZC cosh Γz − ZL sinh Γz

.

(8.10)

Another useful quantity to deﬁne at any distance z from the load is the (voltage) reﬂection coeﬃcient, equal to the ratio between the backward and the forward voltage waves at that z. Due, once again, to Eq. (8.9), it is expressed by: I− eΓz V− eΓz =− = ρL e2Γz , (8.11) ρ(z) = −Γz V+ e I+ e−Γz There will be no danger of ambiguity, as the waves considered in this chapter can only travel along the line.

© 2006 by Taylor & Francis Group, LLC

LOSSLESS TRANSMISSION LINES

273

where ρL is the (voltage) reﬂection coeﬃcient at z = 0, i.e., at the load terminals. To simplify the expression of ρ(z) as a function of the input impedance or admittance, and vice-versa, it is convenient to deﬁne the so-called normalized impedance and admittance, namely: ζ(z) =

Z(z) ZC

,

y(z) =

1 Y (z) ZC = = YC Z(z) ζ(z)

.

(8.12)

The relationships between these adimensional quantities and the reﬂection coeﬃcient (adimensional also) are easily found by combining the deﬁnitions in Eqs. (8.12) and (8.11), and read: ζ(z) =

1 + ρ(z) 1 − ρ(z)

,

y(z) =

1 − ρ(z) 1 + ρ(z)

,

whose inverse is: ζ(z) − 1 Z(z) − ZC YC − Y (z) 1 − y(z) ρ(z) = = = = ζ(z) + 1 Z(z) + ZC 1 + y(z) YC + Y (z)

(8.13)

.

(8.14)

Let us stress how similar these expressions are to Fresnel’s reﬂection formulas, Eqs. (4.92) and (4.93), which we derived for plane wave reﬂection in Section 4.9. This similarity will induce us to elaborate a broad-scope abstract interpretation for the formalism of transmission lines. This point will be developed in Section 8.10, where, following the same line of thought, we will deﬁne normalized voltages and currents, quantities which obey Ohm’s law with normalized (i.e., adimensional) impedances or admittances and therefore are broadly arbitrary in their physical dimensions. Coming back to common transmission lines, the previous relationships enable us to express the complex power, Eq. (8.8), in terms of the reﬂection coeﬃcient, as: ∗ V+ I+ Pc (z) = . (8.15) 1 − ρ(z) ρ∗ (z) + ρ − ρ∗ 2 An exercise which is left to the reader is to separate real and imaginary parts in the last equation, i.e., to learn how to express active and reactive power in terms of the reﬂection coeﬃcient. This should be done twice, initially assuming that the characteristic impedance is real, and then removing this assumption. The diﬀerence will help the reader to appreciate why the content of the following two sections is very valuable in practice. 8.4

Lossless transmission lines

All the relationships that were deﬁned or derived in the previous sections become much simpler in the ideal case of a transmission line whose conductors and dielectrics are lossless. The reader may easily obtain, setting the per-unit-length series resistance r and shunt conductance g equal to zero, the following results for propagation constant, characteristic impedance, nor-

© 2006 by Taylor & Francis Group, LLC

274

WAVES ON TRANSMISSION LINES

malized impedance and admittance, reﬂection coeﬃcient, voltage and current along a lossless transmission line: √ Γ = jβ = jω lc , (8.16) l , (8.17) ZC = c ζ(z)

ζL − j tan βz ζL cos βz − j sin βz = cos βz − jζL sin βz 1 − jζL tan βz

=

1 yL − j tan βz = ζ(z) 1 − jyL tan βz

y(z) =

V− j2βz e V+

ρ(z)

= ρL ej2βz =

V (z)

= V+ e−jβz + V− ejβz V+ −jβz V− jβz e − e ZC ZC

I(z) =

,

,

(8.18) (8.19)

,

(8.20)

,

(8.21) .

(8.22)

For a lossless transmission line the characteristic impedance Eq. (8.17) is real, like the intrinsic impedance in a lossless medium (and consequently the wave impedance in any direction in real space, for a uniform plane wave). To insist on this analogy, the quantity λ = 2π/β is called the wavelength on the transmission line. Voltage and current of either a forward or a backward traveling wave are in phase, at any z. Another result which, as the reader may easily check, is peculiar to lossless transmission lines, and holds irrespectively of their loads, is that the reﬂection coeﬃcient, the normalized input impedance and the normalized input admittance are periodic functions of z, with a period equal to λ/2. The modulus of the reﬂection coeﬃcient is constant along a lossless line. Voltage and current are periodic functions, of period λ. As for the complex power, Eq. (8.15) shows that along a lossless line its real part (i.e., the active power): Pa =

∗ V+ I+ 1 − ρ(z) ρ∗ (z) 2

,

(8.23)

is constant vs. z, while the coeﬃcient of its imaginary part (i.e., the reactive power): Q(z) = −j

∗ V+ I+ [ρ − ρ∗ ] , 2

(8.24)

varies sinusoidally, with a period equal to λ/2. These properties will be exploited to develop a graphical tool, the Smith chart, to be introduced in Section 8.7.

© 2006 by Taylor & Francis Group, LLC

LOW-LOSS TRANSMISSION LINES 8.5

275

Low-loss transmission lines

Strictly lossless transmission lines are not physical entities, in the same sense as the strictly lossless media of the previous chapters. In the real world, losses are inevitable. However, with very rare and special exceptions, like attenuators, transmission lines are of practical interest only if their losses are small enough, at least over a suitable frequency range. Consequently, the lossless model studied in the previous section may be looked at as a “zero-order” approximation for real lines, similarly to what we did for conducting-wall waveguides in Chapter 7. In fact, we will see soon that, for lossy lines, exact solutions are often unnecessarily cumbersome. We will show then how ﬁrst-order approximations for real lines can be derived by elaborating on the lossless model. The phrase “small loss,” still ambiguous up to this stage, will be used from now on to designate precisely the cases where these ﬁrst-order approximations are acceptable. As we said before, the propagation constant Eq. (8.5) is, in general, a complex quantity, Γ = α+jβ. One may always start from the following expression: Γ2 = (r + jωl) (g + jωc) = (rg − ω 2 lc) + jω(rc + lg) ,

(8.25)

and separate real and imaginary parts, obtaining: α2 − β 2

= rg − ω 2 lc ,

2αβ

= ω(rc + lg) .

(8.26)

Solving this as a system of two equations in two unknowns, we get: 1 α = √ (r2 + ω 2 l2 ) (g 2 + ω 2 c2 ) − (ω 2 lc − rg) , 2 1 β = √ (r2 + ω 2 l2 ) (g 2 + ω 2 c2 ) + (ω 2 lc − rg) . 2

(8.27)

As anticipated, these expressions are complicated, compared with Eq. (8.16), but become remarkably simpler for small losses, i.e., for r/ωl 1, g/ωc 1. Indeed, truncated binomial expansions for the square roots (neglecting terms of order rg/ω 2 lc and smaller) give: g r c l r g ∼ + = YC + ZC α = , 2 l 2 c 2 2 √ g 2 1r ∼ − β = ω lc 1 + . (8.28) 8 ωl ωc From Eq. (8.7) in a similar way we ﬁnd: g r 3g 2 rg r2 + j − − + ZC ∼ 1+ = ZC 8ω 2 l2 8ω 2 c2 4ω 2 lc 2ωc 2ωl

,

(8.29)

= 1/YC is the characteristic impedance of a lossless line having the where ZC same per-unit-length inductance and capacitance as the actual line.

© 2006 by Taylor & Francis Group, LLC

276

WAVES ON TRANSMISSION LINES

Note that, at ﬁrst order in r/ωl and g/ωc, the phase constant remains the same as for an ideal lossless line. The same occurs for the real part of the characteristic impedance. Accordingly, ﬁrst-order eﬀects are only a non-zero attenuation constant, and a reactive part in the characteristic impedance. As we said in Section 7.9, one of the most important results of perturbation theory in general (see, e.g., Morse and Feshbach, 1953, or Dicke and Wittke, 1960) is that n-th order corrections for an eigenvalue can be calculated by means of the n − 1-th order eigenvectors. What this means in our case is that (exactly like in Section 7.9) we can compute the ﬁrst-order attenuation constant from the zero-order voltages and currents, i.e., from what we found for lossless lines in the previous section. In our case, this statement is easy to prove. Insert the lossless-line voltage and current forward waves (the ﬁrst terms in Eqs. (8.21) and (8.22), respectively) into the following formulas, which express the per-unit-length power loss, Pl (in watts per meter) for a forward traveling wave, and the active power P (in watts) carried by that wave through the generic section of the line: 1 , (8.30) r |I+ (z)|2 + g |V+ (z)|2 Pl = 2 P

=

Y |V+ (z)|2 |I+ (z)|2 ZC = C 2 2

.

(8.31)

On a lossy line, where forward voltage and current decay like e−αz , active power has to vary like P (z) = P0 e−2αz , and power loss over an inﬁnitesimal length is then dP = −Pl dz = −2α P (z) dz. Then, from Eqs. (8.30) and (8.31), the attenuation constant turns out to be: gZC rY , (8.32) α= C + 2 2 which is the same as Eq. (8.28), as we intended to prove. 8.6

Partially standing waves

In this section we will refer to lossless lines, unless the opposite is explicitly stated. However, most of the results are not diﬃcult to extend to low-loss lines. The reader is encouraged to do that as an exercise. The general expression Eq. (8.4) for voltage along a lossless line can be rewritten as follows: V (z) = |V+ | ej(φ+ −βz) + |V− | ej(βz+φ− )

,

(8.33)

where φ+ and φ− are the phases of the complex numbers V+ and V− , respectively. Adding and subtracting |V− | e−jβz ejφ+ , Eq. (8.33) can be rewritten as:

. (8.34) V (z) = |V+ | − |V− | ej(φ+ −βz) + |V− | ej(βz+φ− ) + e−j(βz−φ+ ) The ﬁrst term on the right-hand side is a forward traveling wave, of ampli-

© 2006 by Taylor & Francis Group, LLC

PARTIALLY STANDING WAVES

277

tude |V+ | − |V− |. Comparison with Section 4.5 tells us that the second term is a pure standing wave, of maximum amplitude 2 |V− |. In practice, it occurs often that one can measure either voltage or current (or sometimes a ﬁeld related to these) on a transmission line by means of an instrument whose response is slow (i.e., it averages over a large number of periods, at the angular frequency ω) and quadratic (i.e., the output of the instrument is proportional to the modulus square of either voltage or current). In view of this, let us write explicitly:

(8.35) |V (z)|2 = |V+ |2 1 + |ρL |2 + 2 |V+ |2 |ρL | cos(2βz + φL ) , where |ρL | and φL are modulus and phase, respectively, of the load reﬂection coeﬃcient, ρL = ρ(0) = V− /V+ . The voltage modulus reaches its maximum value: (8.36) |V (zM )| = |V+ | + |V− | , in those sections where: 2β zM + φL = 2kπ

(k = 0, 1, 2, . . .) ,

(8.37)

and its minimum value: |V |m = |V+ | − |V− | ,

(8.38)

in those sections where: 2β zm + φL = (2k + 1) π

.

(8.39)

It is straightforward to check that the sections where the voltage modulus is maximum, Eq. (8.37), are also those where the current modulus is minimum, and vice-versa for the sections Eq. (8.39), since:

|I(z)|2 = |I+ |2 1 + |ρL |2 + 2 |I+ |2 |ρL | sin(2βz + φL ) . (8.40) Two adjacent minima (or maxima) are separated by a distance: (zM +1 − zM ) =

λ π = β 2

,

(8.41)

whereas the distance between any maximum and either one of the adjacent minima is λ/4. From now on, when using phrases like “voltage maximum” (or minimum) we will refer to the voltage modulus, without any danger of ambiguity. The same applies, of course, to the current. We might have derived the same results also from the following expressions: = |V+ | |1 + ρ(z)| ,

(8.42)

ZC |I(z)| = |V+ | |1 − ρ(z)| ,

(8.43)

|V (z)|

which lend themselves to the simple graphical interpretation shown in Figure 8.2, based on the fact that, as Eq. (8.20) shows, when z varies along a lossless line, ρ(z) describes a circle.

© 2006 by Taylor & Francis Group, LLC

278

WAVES ON TRANSMISSION LINES

Figure 8.2 Complex-plane illustration of the behavior of the quantities 1 ± ρ vs. distance, along a lossless transmission line.

The adimensional quantity: S=

1 + |ρ| |V |max |V+ | + |V− | = = |V |min |V+ | − |V− | 1 − |ρ|

,

(8.44)

is referred to as the voltage standing wave ratio, and frequently designated by its initials, VSWR. When a transmission line is closed on a passive load, i.e., on a two-terminal network such that, at any frequency of interest, active power ﬂowing into it is non-negative, then Eq. (8.21) shows that |ρ| ≤ 1, and consequently S ≥ 1. In the sections where voltage is maximum (and current is minimum), voltage and current are in phase, so the input impedance is real, and indeed it equals: Z(zmax ) =

|V (z)|max |V+ | + |V− | ZC = SZC = |I(z)|min |V+ | − |V− |

.

(8.45)

Similarly, in those sections where voltage is minimum (and current is maximum), voltage and current are again in phase, the input impedance is real, and equals: Z(zmin ) =

8.7

|V (z)|min |V+ | − |V− | ZC ZC = = |I(z)|max |V+ | + |V− | S

.

(8.46)

The Smith chart

In Section 8.3, we derived two formulas, Eq. (8.10) and its reciprocal, which enable us to calculate the input impedance and admittance in any section along a transmission line when the load impedance is speciﬁed. Those relationships are much more cumbersome than the corresponding transformation rule for the reﬂection coeﬃcient, Eq. (8.11). Reﬂection coeﬃcient and input impedance (or admittance) are related to each other by a one-to-one and onto mapping. Hence, it is self-explanatory why it is of interest to exploit this mapping and the simple rules which govern how ρ changes vs. z, to by-pass the cumbersome formulas for Z(z), Y (z).

© 2006 by Taylor & Francis Group, LLC

THE SMITH CHART

279

We derived in Section 8.3 the relationship between normalized input impedance (or admittance) and reﬂection coeﬃcient: ρ=

1−y ζ −1 = ζ +1 1+y

.

(8.47)

Relationships of the kind aτ η + bτ + cη + d = 0, where a, b, c, d are complex constants, and τ, η are two complex variables, are called bilinear transformations. It can be shown (Schwerdtfeger, 1961) that any bilinear transformation maps any circle of the τ complex plane into a circle on the η complex plane, and vice-versa. Furthermore, any bilinear transformation maps two mutually orthogonal pencils of circles of one of these complex planes into two mutually orthogonal pencils of circles on the other complex plane, and vice-versa. This statement can be extended to the case of two orthogonal sets of parallel straight lines, which can be looked at as degenerate pencils of circles, with base points at inﬁnity. In particular, this can be applied to the straight lines parallel to the real and to the imaginary axis of the ζ plane (or of the y plane), i.e., to the loci of constant real part of normalized impedance (or admittance), and of constant imaginary part of the normalized impedance (or admittance). In other words, we know a priori that these lines will be mapped by Eq. (8.47) into mutually orthogonal pencils of circles on the complex plane of ρ. To proceed to the calculations, let ζ = r + jx (y = g + jb), and ρ = u + jv, where r, x, g, b, u and v are all real.‡ Starting from: r + jx =

1 + (u + jv) 1+ρ = 1−ρ 1 − (u + jv)

,

(8.48)

and separating real and imaginary parts, we ﬁnd: r=

1 − (u2 + v 2 ) (1 − u)2 + v 2

,

x=

2v (1 − u)2 + v 2

These two equations can be rewritten as: 2 r 1 u− + v2 = 1+r (1 + r)2

,

.

(8.49)

(8.50)

and:

2 1 1 (u − 1)2 + v − = 2 . (8.51) x x Eq. (8.50) reconﬁrms that, on the ρ plane, the loci r = constant are circles. r , 0 and its radius is Each of them is centered at a point of coordinates 1+r given by 1/(1 + r) (see Figure 8.3). Similarly, Eq. (8.51) reconﬁrms that the ‡ Several textbooks on transmission lines use the symbols z and x for the normalized impedance and the normalized reactance, respectively. At the same time, they use one of these symbols to denote the longitudinal coordinate along the line. To avoid ambiguity, we call z the longitudinal coordinate, ζ the normalized impedance. Throughout this section, x means the normalized reactance. Elsewhere, it may indicate a transverse coordinate. The text will always make the symbol unambiguous.

© 2006 by Taylor & Francis Group, LLC

280

WAVES ON TRANSMISSION LINES

Figure 8.3 The constant-resistance loci, on the complex plane of the reﬂection coefﬁcient.

Figure 8.4 The constant-reactance loci, on the complex plane of the reﬂection coefﬁcient.

loci x = constant on the ρ plane are circles, each of which is centered at a point of coordinates (1, 1/x) and has a radius given by 1/|x| (see Figure 8.4). Their sets are indeed two orthogonal pencils. Each pencil is parabolic, i.e., instead of having two distinct base point as the more common elliptic pencils, it has one double base point (the same for both pencils). This base point is located at (1,0). “Double” means that all the circles belonging to the same pencil not only pass through that point, but also have the same tangent in that point. For the r = constant circles, their tangent in common is the line u = 1, parallel to the imaginary axis. For the x = constant circles, the tangent in common is the real axis, v = 0. As we saw before, for passive loads we have |ρ| ≤ 1, and correspondingly, r ≥ 0 in Eq. (8.50). Consequently, the region of the ρ plane where passive loads are mapped is the interior of the circle |ρ| = 1, including the circle itself, which is the locus of the lossless loads. The plot of the r = constant circles and of the portions of the x = constant which belong to this part of the ρ plane is usually referred to as the Smith impedance chart, and is shown in Figure 8.5. An equivalent chart in terms of admittance is easy to build, noting that

© 2006 by Taylor & Francis Group, LLC

THE SMITH CHART

281

70 (+ jX /Z

45

1.4

1.2

1.0

0.8

55

1.6

1.8

0.6 60

2.0

0.5

06 0. 44 0. 0 14

4

0.

0.0 5

25 0.4

0.2

20 3.0

0.6

9 0.2 30

4.0

15

5.0

0.2

IND UCT IVE

20 10 0.25 0.26 0.24 0.27 0.23 0.25 0.24 0.26 0.23 0.27 REFLECTION COEFFICIENT IN DEG REES LE OF ANG ISSION COEFFICIENT IN TRANSM DEGR LE OF EES ANG

8

0.

0.6

10

0.1

0.4

20

0.2

50

20

10

5.0

4.0

3.0

2.0

1.8

1.6

1.4

1.2

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

1.0

50

0.1

50

RESISTANCE COMPONENT (R/Zo), OR CONDUCTANCE COMPONENT (G/Yo) 0.2

20

0.4

10

-10 1.0

E IV CT DU IN

2.0

-4

5

0.14 -80 -4 0

0.15 0.35

0.13

TR S. RF S. A A W. L. W. TT N P L L EN SM EA O O . . C K SS [ SS C [dB O O (C dB O ] EF EF O ] EF F, F, NS F E P T. or P) I

2

0.12

0.37

40

-70

-1

06

0

0.0

1

0.4

0.4

0.39

0.38

RADIALLY SCALED PARAMETERS

∞ 40 30 0

10 20

1 0.9

5

0.8

3

15 2

0.7

4

0.6

2.5

10

3

4

0.5

0.4

5

2

8 6

0.3

7

1.8

6 8

0.2

9

1.6

5 10

0.1

1.4

4 12

3 14

0.05

1.2 1.1 1 2 20 0.01

1

15

TOWARD LOAD —> 10 7 5

1 1 30 ∞ 0

0.1

0 0

1.1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0 1

0.99

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

CENTER 1

1.1

1.2

1.3 1.4

0.4

0.6

1.2

1.3 0.95

1.4

0.8

1.5

0.9

WAVELE 0.49 NGTH S TOW ARD 0.0 0.49 AD WA 160 0 1. Let ρ = 1/ρ: in this way, the region outside the circle |ρ| = 1 is mapped onto the region inside that circle, and vice-versa. y(ρ) =

© 2006 by Taylor & Francis Group, LLC

282

WAVES ON TRANSMISSION LINES

Consequently, plots over Figure 8.5 can be used to represent active loads, provided one reads that plane as that of the inverse of the reﬂection coeﬃcient. Many commercially available modern instruments for testing active electronic devices, typically microwave transistors, encompass a function which allows passing from the ρ plane to the ρ−1 plane on their polar display, by simply pushing a button. 8.8

Remote measurement of the load impedance

An elegant and useful application of the theory that we developed in this chapter comes from a problem that is frequently encountered in actual telecommunication systems. It occurs often that an unknown load impedance, located at the end of a transmission line, is diﬃcult to access. For example, this is typical with antennas installed on the top of buildings, metallic towers, etc., where it is diﬃcult to carry the necessary measuring instrumentation and to operate it in a reliable way. Finding the unknown load impedance from data measured at a distance, along the transmission line, is a remarkable improvement. We saw in Section 8.3 that if the reﬂection coeﬃcient is known, in modulus and phase, at a known distance z from the load, then, in principle, it is straightforward to calculate the load impedance or admittance (normalized to the characteristic impedance of the line). We will show here that there is a rather simple experimental recipe to implement this theory. The required instrument is referred to as a voltage standing wave indicator (VSWI), and consists of: a voltage probe that can be moved in a controlled way along the transmission line (usually, it is inserted in a movable mount equipped with a calibrated vernier scale); an RF peak voltmeter (an instrument which measures the modulus of the RF voltage, when connected to the probe); ﬁnally, a known load impedance, normally a short circuit. The ﬁrst step is to ﬁnd the modulus of the reﬂection coeﬃcient, which we can calculate when we know the VSWR. Moving then the probe along the line, one looks for a maximum, and measures |V |max by means of the RF peak voltmeter. Next, one looks for an adjacent minimum, and measures |V |min . Assuming, at least during an initial phase, that the line is lossless, we ﬁnd: |ρ| = |ρL | =

|V |max − |V− |min S−1 = S+1 |V |max + |V− |min

.

(8.53)

If we measure the distance between two consecutive minima (or maxima) (equal to λ/2) by means of the sliding probe, and if the distance between the load and the n-th voltage maximum on the line, zL , is also known, then in principle it is easy to calculate the phase of the load reﬂection coeﬃcient ϕL = arg(ρL ) as follows: ϕL = 2β(zL − n λ/2) = 2π(2zL /λ − n)

.

(8.54)

In practice, however, this procedure is unsuitable, because the result would be aﬀected by a large error bar. The ﬁrst uncertainty is encountered in the

© 2006 by Taylor & Francis Group, LLC

REMOTE MEASUREMENT OF THE LOAD IMPEDANCE

283

|V|

a

b c

βz

Figure 8.6 Typical patterns of partially standing waves. a) VSWR = 1. b) VSWR = 2. c) VSWR = ∞.

location of a maximum. Indeed, in any partially standing wave, minima (of the modulus, see Figure 8.6) are sharper than maxima. Therefore, locating a minimum is more accurate than locating a maximum. This simply implies that ±π is added to Eq. (8.54). However, another more serious reason why this modiﬁed procedure is not yet adequate is because, as well known, an unknown quantity measured as the diﬀerence between two quantities of similar magnitude is inevitably aﬀected by an enlarged relative error bar. In the present case this inconvenience may be extremely signiﬁcant, because quite often it is extremely impractical to measure geometrically the distance z L . For example, this occurs whenever a line twists or bends, between the load and the measuring equipment. All these inconveniences are avoided if one measures ϕL by comparison with a reference standing wave, namely the wave which one ﬁnds on the same line when the unknown load is replaced by a known impedance. To be a good choice, this known load must yield a large VSWR, with sharply deﬁned voltage minima. An excellent combination of this requirement and simplicity is a short circuit. Incidentally, note that an open circuit would be equally good in principle, but in practice it is not, because at high frequency a transmission line whose remote terminals are open radiates, similar to an antenna. If a line is terminated on a short circuit, then the distance between the n-th voltage minimum and the load terminals is z¯SC = nλ/2. If we call z¯L the abscissa of the n-th voltage minimum when the line is terminated on the unknown load, then the unknown load phase is simply: zSC − z¯L ) ± π ϕL = 2β(¯

.

(8.55)

To perform the tests that were described in this section, there are commercially available VSWI’s which encompass a slotted line, i.e., a section of either coaxial cable or rectangular waveguide, whose outer wall is slotted in the longitudinal direction, parallel to the current ﬂow in the wall itself. The slot accommodates a movable voltage probe.

© 2006 by Taylor & Francis Group, LLC

284

WAVES ON TRANSMISSION LINES

Figure 8.7 A uniform transmission line, connecting a generator and a load.

8.9

Impedance matching

Normally, a transmission line of ﬁnite length is connected to a generator at one end (the left side in Figure 8.7), and to a load at the other hand. Let ZG be the internal impedance of the generator (either in series to an ideal voltage generator whose open-circuit voltage is VG , or in parallel to an ideal current generator whose short-circuit current is IG = VG /ZG ). In general, there are reﬂected waves on the line. As we saw in the previous sections, and also in some of the previous chapters, reﬂected waves are responsible for unwanted eﬀects. First, the power delivered to the load becomes smaller than that carried by the forward wave. Second, keeping constant the active power delivered to the load P , the maximum voltage along the line can be expressed as: (8.56) |VM | = 2ZC P S , where ZC is, as usual, the characteristic impedance of the line, and S is the VSWR. The larger the reﬂected waves, the larger S, and, as consequence, |VM |. This must be accounted for, to avoid electrical discharges along the line. Last but not least, in general the reﬂection is frequency dependent, causing signal distortion in the case of ﬁnite spectral width. This intuition is reconﬁrmed by evaluating the voltage transfer function of a lossless transmission line of length l, i.e., the ratio between output and input voltages: H(jω) =

ZL VL = Vi ZL cos βl + jZC sin βl

.

(8.57)

It is easy to verify that in general Eq. (8.57) does not satisfy the distortionless transmission conditions (|H(jω)| = 1, H(jω) ∝ ω). However, if ZL = ZC , i.e., when there are no reﬂections, then H(jω) = e−jβl and the signal does not undergo any distortion. An impedance matching device is a passive (whenever possible, lossless) two-port network, whose purpose is to eliminate reﬂections at one of its ports, when the other one is connected to a given load (Figure 8.8). The purpose of this section is to illustrate three of the most commonly used impedance matching networks, which are based on lossless transmission lines. From the elementary theory of electrical circuits, we know that the active

© 2006 by Taylor & Francis Group, LLC

IMPEDANCE MATCHING

285

Figure 8.8 Deﬁnition of an impedance-matching network.

jXG

-jXG

RG

ZL

RG

VG

Figure 8.9 Power matching between a given generator and its load.

power delivered by a generator of given internal impedance, ZG = RG + jXG , reaches its maximum value when the load connected to the generator terminals, Zi (see Figure 8.9) equals the complex conjugate of ZG , i.e., for Zi = RG − jXG . This is referred to as the power matching condition. The complex power delivered by the generator under these circumstances is easily calculated from the deﬁnition of complex power (e.g., Eq. (8.8)). For a powermatched load it is I = VG /(2RG ), so the active power delivered to the load is: |VG |2 , (8.58) Pa = 8RG and the reactive power simply equals Pa times the ratio (−XG /RG ). Eq. (8.58) is referred to as the generator’s available power. The power matching condition, which is of paramount importance to electrical engineers involved in power systems, in general is not the best answer if we refer to signal transmission problems. In fact, power matching implies that the imaginary parts of the generator and load impedances have opposite signs. Because of the diﬀerent frequency dependence of reactances of opposite signs, their absolute values can be equal just at one frequency, not over a ﬁnite bandwidth, except for the case where they are equal to zero. Therefore, in general power matching entails signal distortion. The right recipe in a telecommunication context is not to look for the absolute maximum of the power delivered to the load, but for a conditional maximum, namely, under the restriction that it must be frequency-independent (at least over a narrow bandwidth). Again from elementary circuit theory, it can be shown that this

© 2006 by Taylor & Francis Group, LLC

286

WAVES ON TRANSMISSION LINES

corresponds to: Zi = ZG , (8.59) which is usually referred to as the uniformity matching condition. Note, however, that in the following we will restrict ourselves to lossless transmission lines, or at most to low-loss lines. It will be taken for granted that their characteristic impedance ZC is real. As the loads, and often the generators also, will be matched to this real ZC , the distinction between power matching and uniformity matching disappears. The purpose of an impedance-matching network is then to provide an input impedance which satisﬁes two independent requirements, if expressed in real numbers: real part equal to ZC , imaginary part equal to zero. We may then expect that the design of an impedance matching network should involve (at least) two real variables. All the following examples reconﬁrm this expectation. However, we will see that the physical nature of these variables changes completely as we pass from one network to another. 8.9.1

The quarter-wavelength transformer

In Section 4.10 we showed that, for plane waves, a lossless dielectric layer, whose thickness equals a quarter of the wavelength, can in fact cancel the Fresnel reﬂection at the interface between two semi-inﬁnite lossless media, provided its intrinsic impedance is given a suitable value. The Smith chart of Section 8.7, while on one hand it simpliﬁes the understanding of what we did in Section 4.10, on the other hand it suggests that the same results apply to transmission lines, provided the load impedance is real. Indeed, when a lossless line whose length equals an odd integer multiple of λ/4 is connected to a generic load impedance ZL , its input impedance equals: Zi =

2 ZC ZL

.

(8.60)

As the characteristic impedance of the lossless quarter-wavelength line, Z C , is real, Eq. (8.60) cannot be made equal to a real number (the characteristic impedance of the line to be matched) unless ZL is itself real. Using a , would mean introducing lossy quarter-wavelength line, with a complex ZC an unwanted attenuation. A suitable solution to the problem can be found if we simply accept that the quarter-wavelength line be inserted at a suitable distance d from the load (Figure 8.10), rather than connected directly to its terminals. Indeed, we know from Section 8.3 and the following ones that, as we move along the line, we ﬁnd periodically sections where the input impedance is real. They are either maximum-voltage sections, where:

Z(z) = SZC

,

(8.61)

,

(8.62)

or minimum-voltage sections, where: Z(z) = ZC /S

© 2006 by Taylor & Francis Group, LLC

IMPEDANCE MATCHING

287

Figure 8.10 Load matching by means of a quarter-wave line.

where S is the VSWR. On the Smith chart, these sections correspond to the intersections of the circle |ρ(z)| = constant, i.e., the locus of the reﬂection coeﬃcient along the line, with the real axis of the ρ plane, i.e., the locus of real impedances. In the ﬁrst case, combining Eqs. (8.60) and (8.61), the recipe for the intrinsic impedance of the quarter-wave line is found to be: √ = Z(z) ZC = ZC S , (8.63) ZC while in the second case it is found to be: √ = Z(z) ZC = ZC / S ZC

.

(8.64)

These formulas (even better if re-written in terms of normalized impedances, /ZC ) are strictly reminiscent of how one calculates the transforming ratio, ZC n : 1, for an ideal transformer to convert a resistive impedance Ro into an other resistive impedance Ri , n = Ri /Ro . For this reason, in the current terminology of telecommunications, lossless lines whose lengths equal an odd integer multiple of λ/4 are often referred to as quarter-wavelength transformers. Note, however, that the analogy disappears when the loads are not purely resistive. Indeed, the sign of the imaginary part of the impedance does not change as one goes through an ideal transformer, while, on the other hand, Eq. (8.60) converts an inductive load into a capacitive input impedance, and vice-versa. We said in the ﬁrst part of this section that we expected impedance-matching network designs to be based on two real parameters. In the case at hand, the two parameters are the characteristic impedance of the quarter-wave section, and its distance from the load, which equals either: d(S) = (λ/2) (ϕL /2π) + m(λ/2)

,

(8.65)

in the case of Eq. (8.63), and: d(1/S) = d(S) ± λ/4

,

(8.66)

in the case of Eq. (8.64), where ϕL is the phase angle of the load reﬂection coeﬃcient, and m = 0, 1, 2, . . .. As for the choice between the solutions in Eqs. (8.63) and (8.64), a suitable criterion is to make the line between the

© 2006 by Taylor & Francis Group, LLC

288

WAVES ON TRANSMISSION LINES

Figure 8.11 Single-stub impedance-matching network.

load and the λ/4-section as short as possible. Indeed, there are reﬂected waves on this piece of line, and from Eq. (8.64) one may check that the deviation from distortionless transmission grows as the line length grows. Sometimes, however, other more practical criteria become dominant with respect to this. For example, in a coaxial cable to diminish the characteristic impedance is much easier than to increase it, as discussed in one of the problems at the end of this chapter. 8.9.2

Impedance matching with one or two stubs

From the Smith chart, it is easy to check that if a lossless line is either shortcircuited, or open-circuited at its end, then its input impedance (and admittance, of course) is purely reactive, irrespectively of its length. Therefore, it is of great interest to study whether any load can be impedance-matched to a lossless transmission line by inserting along the line a pure reactance, either in parallel or in series. In principle, the answer is positive for all these combinations. In practice, however, open-circuited lines are disregarded, because they inevitably radiate an active power from their terminals, thus violating the requirement of being lossless. Insertions in series are also disregarded, normally, being more cumbersome than those in parallel. So, whenever we mention impedance-matching stubs, one should refer automatically to shortcircuited line sections in parallel to the main line, as shown schematically in Figure 8.11. From the general expression in Eq. (8.18), with ζL = 0, we ﬁnd immediately that the normalized input impedance of a stub is: ζi = 1/yi = j tan βl

,

(8.67)

where l is the stub length. Thus, any reactance, from −∞ to +∞, can be realized in this way. Therefore, any normalized admittance lying on the g = 1 circle of the −ρ plane (i.e., of the Smith admittance chart, see Figure 8.12), can be matched to a lossless transmission line. It is understood that normalization is with respect to the characteristic admittance of this line, YC = 1/ZC . So, the only requirement to be satisﬁed by what is located between the load and

© 2006 by Taylor & Francis Group, LLC

IMPEDANCE MATCHING

289

Figure 8.12 Illustration of the operation principle of a single-stub impedance matching network, on the Smith admittance chart.

the stub is to transform the point which represents the load into a point on the circle g = 1. Again from Figure 8.12 we see that one simple way to do that, applicable to any load, is to leave a suitable length of the main transmission line between the terminals of the load and those of the stub. Indeed, the circle |ρL | = constant (where ρL is the reﬂection coeﬃcient of the load) intersects the circle g = 1 at two points (called A and B in Figure 8.12). One of them is in the halfplane where b > 0, b being the normalized susceptance; the other one is where b < 0. To yield impedance matching, the stub must provide a susceptance which is equal in modulus and opposite in sign to that seen along the main line. Therefore, one solution (point A) entails an inductive stub (with a length l < λ/4) and the other one (point B) entails a capacitive stub (with a length λ/2 > l > λ/4). Once again the design of this impedance-matching network relies on two real parameters, namely the length of the piece of line between the load and the stub (to be calculated as ϑ/(2πβ), where the angle ϑ is shown in Figure 8.12), and the length of the stub. Note that in the last lines it was taken somewhat for granted that the stub was made with a line with the same characteristic admittance as the main line. This is not necessary at all. When this is not the case, the only change to be made with respect to what we said is the way to calculate the length of the stub, as illustrated in detail in some of the following problems. Incidentally, the reader may verify as an exercise that designing this type of device (which is currently referred to as a single-stub impedance matching

© 2006 by Taylor & Francis Group, LLC

290

WAVES ON TRANSMISSION LINES

Figure 8.13 Double-stub impedance-matching network.

network) using analytical tools, instead of the Smith chart, is possible but much more cumbersome. Another device which is of common use in practice is the double-stub impedance-matching network, schematically illustrated in Figure 8.13. We will call stub 1 the one which is connected to the same terminals as the load, stub 2 the one upstream towards the generator. Contrary to the single stub, distances along the main line are ﬁxed. The most commonly used values for the distance between the stubs are either λ/4, or λ/8, or 3λ/8. The two real variables exploited in the design of this network are the two lengths of the stubs. To outline how a double-stub network operates, note ﬁrst that the stub 2 must play the same role as the stub in the previously discussed single-stub network. Therefore, matching is possible only if the point which represents the input admittance on the main line, as seen from the terminals of the stub 2, lies on the circle g = 1 of the admittance Smith chart (see Figure 8.14). Consequently, as nothing can be modiﬁed on request between the stubs, the point which represents the input admittance immediately upstream of the stub 1 must lay on the curve where the circle g = 1 is mapped by the line section between the stubs. Figure 8.14 shows these curves for the three common cases of a distance equal to λ/4, to λ/8 and to 3λ/8. To avoid confusion, the rest of the procedure is illustrated in detail only for the case λ/4. The other ones can be completed by the reader as an exercise. The purpose of the stub 1 is to modify the input admittance, from the given load value to a value lying on the transformed circle, designated as the A-circle. A stub in parallel does not aﬀect the real part of the admittance. Therefore, we pass from the point yL to the point on the A-circle along a constant-g line. This intersects the A-circles at two points, indicating that, for a load like the one shown in Figure 8.14, there are two solutions. For each of them, the susceptance of the stub 1 is found as the diﬀerence between that at the point on the A-circle, and that of the load. The susceptance of the stub 2 is then computed in the same way as for a single-stub matching network. The main advantage of the double stub is to avoid sliding contacts along the

© 2006 by Taylor & Francis Group, LLC

TRANSMISSION-LINE EQUATIONS

291

Figure 8.14 Illustration of the operation principle of a double-stub impedance matching network, on the Smith admittance chart.

main line. Its main drawback is that it cannot match all possible loads. Indeed, if the normalized load admittance is located in the region whose g = constant lines do not intersect the transformed circle (to be clear, the A-circle in the previous example), then the procedure does not yield any solution. The main reason why double stubs spaced by either λ/8 or 3λ/8 are often preferred to those spaced by λ/4 is because their “non-match regions” are smaller. The further choice between them is a trade-oﬀ: the ﬁrst one is shorter, thus giving less distortion, but two closely spaced stubs may tend to interact with each other in a rather unpredictable way. A powerful solution which makes it possible to match any load is to use a triple stub. In such a case, there are several possible solutions for a generic load to match. The consequent main drawback, usually, is that the device is designed (or made empirically) using a time-consuming cut-and-try procedure. Double stubs could also be explained and designed in terms of line equations, but the Smith chart turns out to be much simpler, as the reader can verify as an exercise. 8.10

Transmission-line equations: An alternative derivation

In this section we will show how the equations which govern voltage and current waves along transmission lines can be derived from Maxwell’s equations, provided some suitable assumptions are satisﬁed. The ﬁrst purpose is to show that what we have seen so far in this chapter, although derived in a rather phenomenological way from the circuit of Figure 8.1, is not in contradiction

© 2006 by Taylor & Francis Group, LLC

292

WAVES ON TRANSMISSION LINES

with the axiomatic approach followed in the rest of the book. Another purpose, may be obscure now, but hopefully not so at the end of this section, is to show that transmission lines provide useful and elegant equivalent models for propagation phenomena in other guiding structures, like the conducting-wall waveguides of the previous chapter, and dielectric waveguides of Chapter 10. Let us consider a lossless homogeneous cylindrical structure whose crosssection is not a simply connected domain, so that it can propagate the TEM mode. Introduce rectangular coordinates where z is the longitudinal axis, and write homogeneous Maxwell’s equations by components. Under the assumptions that the ﬁeld is transverse electromagnetic, i.e.: Ez ≡ 0 ,

Hz ≡ 0 ,

(8.68)

the x and y components read: ∂Ey = jωμ Hx ∂z

,

∂Ex = −jωμ Hy ∂z

,

(8.69)

∂Hy ∂Hx = −jω Ex , = jω Ey . (8.70) ∂z ∂z The z components are omitted as they are not required in the following steps. The ﬁrst of Eqs. (8.69) and the second of Eqs. (8.70) form a system with two unknowns, Ey and Hx . The other two equations form another system (in Ex and Hy ), independent of the previous one and formally identical with it, except for a sign. Therefore, it is enough for us to investigate in detail only one of these systems, say the second one. All conclusions could then be repeated for the other one. Being familiar with separation of variables, we suggest to start oﬀ with the following ansatz: Ex (x, y, z)

= V (z) Ex (x, y)

,

Hy (x, y, z)

= I(z) Hy (x, y)

.

(8.71)

These two equations must be read as if they encompass two arbitrary multiplicative constants, A and B, i.e., they can always be replaced by:

1 Ex (x, y) , Ex = V (z) Ex (x, y) = A V (z) A

1 Hy (x, y) . (8.72) Hy = I (z) Hy (x, y) = B I(z) B Note that even the physical dimensions of A and B are arbitrary. Therefore, the physical dimensions of the quantities called V , I, Ex and Hy in Eqs. (8.71) are not yet deﬁned at this stage. Introducing Eqs. (8.71) into (8.69) and (8.70), we get immediately: dV Ex (x, y) dz

© 2006 by Taylor & Francis Group, LLC

= −jωμ I Hy (x, y) ,

TRANSMISSION-LINE EQUATIONS

293

Figure 8.15 Kirchhoﬀ ’s laws applied to the equivalent circuit are identical to Maxwell’s equations for TEM propagation.

dI Hy (x, y) = −jω V Ex (x, y) . (8.73) dz We saw in Section 7.2 that the wave impedance of a TEM ﬁeld traveling in the ±z direction is ±η. The same cannot be said, in general, for a superposition of TEM ﬁelds that propagate in both directions along the z axis. Indeed, on several occasions (in Chapter 4 and later) we saw that superimposing two waves is a delicate operation, from the viewpoint of the wave impedance. However, there was one point which was invariably true in all the cases that we examined: namely, the wave impedance did not depend on the transverse coordinates x and y. Consequently, we will assume now that the two functions of the transverse coordinates Ex (x, y) and Hy (x, y) in Eqs. (8.71) are proportional, i.e.: (8.74) Ex (x, y) = N Hy (x, y) . Clearly, the proportionality constant N is related to the arbitrary constants A and B of the previous comment. In particular, there is nothing wrong in letting Ex = Hy in Eq. (8.72), thus obtaining N = A/B. Because of Eq. (8.74), Eqs. (8.73) become: dV dz

= −

dI dz

= −jω N V

jωμ I N

, .

(8.75)

If we give the arbitrary constants in Eqs. (8.72) physical dimensions such that V and I get, respectively, dimensions of a voltage and of a current, then Eqs. (8.75) become identical with the telegraphist equations, Eqs. (8.1). Other choices for A and B do not entail any substantial change. The degrees of freedom in A and B simply mean that the circuit of Figure 8.15, where Eqs. (8.75) are satisﬁed as Kirchhoﬀ’s laws, should be looked at as an equivalent circuit for any TEM mode propagation, not necessarily as a phenomenological model, as it might have appeared after the initial derivation of the equations from the circuit shown in Figure 8.1. The circuit shown in Figure 8.15 may look as a simpliﬁed (or idealized) version of that of Figure 8.1. A natural question to ask then, is whether passing

© 2006 by Taylor & Francis Group, LLC

294

WAVES ON TRANSMISSION LINES

from a lossless medium to a lossy one in Maxwell’s equations, Eqs. (8.69) and (8.70), could yield an equivalent circuit like that of Figure 8.1. It is simple to verify that, if we replace by c , only the second of Eqs. (8.75) changes, so that its equivalent circuit encompasses now an inﬁnitesimal parallel conductance: g dz = Re [jωc N ] dz = N σ dz

.

(8.76)

No series resistance emerges in this way. The reason is because a series resistance is, strictly speaking, incompatible with the TEM assumption. Indeed, a current ﬂowing through conductors with ﬁnite conductivity necessarily entails a nonvanishing longitudinal component for the electric ﬁeld. As we saw at the beginning of this chapter, we may elaborate on Eqs. (8.75) so that each unknown, V (z) and I(z), is isolated. The order of the diﬀerential equations increases. Incidentally, note that this procedure is essentially the same as the one used in Chapter 1 to pass from Maxwell’s equations to the Helmholtz equation. The equations we get in this way are: d2 I d2 V 2 = −ω μ V , = −ω 2 μ I . (8.77) dz 2 dz 2 Eqs. (8.77) are essentially the same as the telephonist equations (8.3). Note that the arbitrary constant N of Eqs. (8.75) disappeared in the last equations. Their general integral is of the type: V = V+ e−jβz + V− e+jβz

,

(8.78)

where V+ and V− are arbitrary constants, and: √ β = ω μ .

(8.79)

Hence, for waves propagating along a TEM transmission line, phase constants are always equal to the intrinsic phase constant of the homogeneous lossless dielectric medium between the (ideal) conductors, independently of the geometrical shape of the structure cross-section. From Eq. (8.78) and the ﬁrst of Eqs. (8.75) we ﬁnd: I=

Nβ Nβ V+ e−jβz − V− ejβz = I+ e−jβz + I− ejβz ωμ ωμ

where now we let:

I± = ±V± N

V± =± μ (η/N )

.

,

(8.80)

(8.81)

The physical dimensions of V and I are still undeﬁned. So are those of the arbitrary constant N , and consequently of η/N , which appears in Eq. (8.81). Independently of the choice of N , this quantity plays the physical role of an impedance, and is proportional to the intrinsic impedance of the dielectric, η = μ/. Let us illustrate, by means of some examples, how one may exploit the still arbitrary constants so as to get a set of enlightening results. 1) If we give N the physical dimensions of an impedance, i.e.: [N ] = Ohm ,

© 2006 by Taylor & Francis Group, LLC

(8.82)

TRANSMISSION-LINE EQUATIONS

295

then η/N is adimensional, while V and I have the same dimensions. In particular, V and I can also be taken adimensional. In such a case we have [Ex ] = V/m, [Hy ] = A/m. Everything is ready for applying the so-called scattering-matrix formalism (Corazza, 1974). An interesting sub-case is to choose N = η. If we start from it, and extend the formalism of this section to TE and TM modes, as far as possible, then we ﬁnd again the results shown in Figure 7.3. 2) If we let N be adimensional, then it is possible (although not necessary) to let V have the physical dimensions of voltage, and I those of a current. Then, we have [Ex ] = [Hy ] = m−1 . If we require that for ω → 0 our V and I tend to become the same as voltage and current in the time-independent case, then N cannot be chosen arbitrarily, but must be derived from two deﬁnitions of the following types: V =− E · d1 , I= H · d2 , (8.83) 1

2

where 1 is an open path which belongs to the generic z = constant plane, and whose extrema fall on the two conductors of the line, while 2 is a closed path (also belonging to the generic z = constant plane), concatenated with only one of these conductors. Given that Eqs. (8.71) hold, Eqs. (8.83) lead to: Ex a ˆx · d1 = −1 , Hy a ˆy · d2 = 1 . (8.84) 1

2

3) If, independently of the physical dimensions one gives to Ex , we choose: N= |Ex |2 dx dy , (8.85) S

where S is the cross-section of the waveguiding structure, then the complex power which ﬂows through this cross-section is given by: E × H∗ V I∗ · aˆz dS = . (8.86) Pc ≡ 2 2 S This makes the relationship between complex power on one side, and V and I on the other side, formally identical with that which holds at the terminals of a one-port electrical network (see Figure 8.16). Note, however, that the dimensions of V and I are still undeﬁned. To make Eq. (8.86) really the same as the usual circuit-theory relationship, it is necessary that, in addition to Eq. (8.85), one of Eqs. (8.84) be valid too. One can then introduce the scattering-matrix formalism starting from any possible choice of N . To show that, let us rewrite Eqs. (8.75) in the following form: ⎞ ⎛ jωμ 0 d V V N ⎠ . (8.87) = −⎝ I dz I jωN 0 To diagonalize the matrix on the right-hand side of Eq. (8.87) (i.e., to ﬁnd its

© 2006 by Taylor & Francis Group, LLC

296

WAVES ON TRANSMISSION LINES

Figure 8.16 Relationship between voltage, current, and complex power.

√ eigenvalues ±iβ = ±iω μ and its eigenvectors) is straightforward algebra, yielding the following equations: da dz

= −jω

db dz

= jω

where: a =

1 2

b

1 2

=

√

√

V η/N V η/N

μ a

,

μ b ,

+

−

(8.88)

η/N I

,

η/N I

Their inverse relationships clearly are: η V = (a + b) , N " N (a − b) . I = η

.

(8.89)

(8.90)

The new variables Eqs. (8.88) are often referred to as the normal modes of the structure. Their practical importance is probably not evident, at this stage, because this whole section suﬀers from limitations due to some of the simplifying initial assumptions. In fact, after writing Eqs. (8.83), we always mentioned two conductors. However, what we proved in Section 7.4 was that the requirement for the existence of the TEM mode were at least two conductors, without an upper limit to their number. In a structure with n > 2 conductors, it is evident that the TEM mode, in spite of its eigenvalue being unique, has more degrees of freedom than what we assumed when we wrote Eqs. (8.71). Namely, in a multiwire transmission line, one can deﬁne n − 1 independent voltages (each of them is the voltage between one common reference conductor and the remaining n − 1 independent conductors) and n − 1 independent currents (the n-th current in the common reference being the “return” one, equal and

© 2006 by Taylor & Francis Group, LLC

TEM AND QUASI-TEM PROPAGATION IN PLANAR LINES

297

ε rε 0 t

2d w

ε0

t d

w

εrε0

Figure 8.17 Schematic view of a stripline and a microstrip.

opposite to the sum of the others). Dealing with such a structure in terms of 2(n − 1) independent diﬀerential equations would make the formalism unbearably heavy, unless one passes through diagonalizing a matrix like in Eq. (8.87). Further developments on this subject are left to Section 8.12. 8.11

TEM and quasi-TEM propagation in planar lines

For at least three decades, the whole world of electronics evolved in the direction of exploiting planar technologies more and more. Microwaves were not at all an exception to this rule. Waveguides of the types discussed in the previous chapters, coaxial cables, or wire pairs, are not very well matched to this trend. The tendency to build passive transmission lines which consisted of two or more ﬂat conductors, lying on parallel planes and separated by suitable dielectrics (e.g., Itoh, Ed., 1987) was then a natural evolution. Figure 8.17 shows schematically the most commonly used types of planar lines. The conducting planes whose edges are not shown in the drawings are, usually, much wider than all the other transverse geometrical dimensions. In the theoretical analyses they are usually taken as inﬁnitely wide. Unless the opposite is explicitly speciﬁed, they are at a common “ground” potential, also when they are more than one. Voltages of interest in the propagation along these structures are those of the other conductors, with respect to ground. The current in the ground electrodes is the opposite of the (algebraic) sum of currents in all the other conductors. The technological progress, from waveguides to planar lines, is not accompanied by a similar progress in modeling. On the contrary, none of the structures shown in Figure 8.17 is simple to analyze theoretically. Approximations are necessary, almost invariably, often for more than one reason, as we shall explain now. The theory of guided e.m. waves, as developed in the previous chapter, re-

© 2006 by Taylor & Francis Group, LLC

298

WAVES ON TRANSMISSION LINES

E field H field

ε0

εr ε0 Figure 8.18 Electric and magnetic ﬁeld lines in the cross-section of a typical microstrip.

quires a homogeneous medium in the region where the guided ﬁeld travels. This is the case, for example, for the symmetrical stripline shown in the upper part of Figure 8.17. Then, as the cross-section is not a simply connected domain, the general theory tells us that the fundamental mode is TEM. All the contents given so far in this chapter apply. The only diﬃculty that one still has to overcome is calculating the per-unit-length parameters l and c (plus, if required, r and g). Often, though, these entail approximations, because the geometry of the structure does not lend itself to a rigorous analysis. An example will be provided in a short while. In other cases — typically, the microstrip shown in the lower part of Figure 8.17, which is probably the most practical among all those examples — part of the ﬁeld travels in the dielectric between the conductors, and part in the air above them (see Figure 8.18). The theory of the previous chapter no longer applies. Intuition tells us that a structure with more than one conductor can operate at arbitrarily low frequencies, so its fundamental mode must have something in common with the TEM mode. However, it is easy to verify that, if we impose the condition that the longitudinal components of the electric and magnetic ﬁelds are zero everywhere, then we do not have enough degrees of freedom left to satisfy all the continuity conditions at the air-dielectric interface, plus the boundary conditions on the conductor surfaces. As a consequence, the fundamental mode is not TEM, but quasi-TEM. The whole content of this chapter then applies to it only as an approximation, superimposed upon the diﬃculty of ﬁnding the per-unit-length parameters. In the following we will discuss brieﬂy an example, to show what this can mean in practice. 8.11.1

First example: the stripline

For a lossless structure which can propagate the TEM mode, we know from Section 8.4 that its characteristic impedance and its phase constant are given √ by Eqs. (8.16) and (8.17), respectively, namely ZC = l/c and β = ω lc. Furthermore, from the previous section we know (see Eq. (8.79)) also that √ β = ω μ. Therefore, when the parameters μ and of the dielectric medium

© 2006 by Taylor & Francis Group, LLC

TEM AND QUASI-TEM PROPAGATION IN PLANAR LINES

299

are known, the phase constant is known, and to determine the characteristic impedance it is enough to determine either c, or l. As the reader may know from previous classes of static ﬁelds, a powerful technique to calculate capacitances, even for apparently complicated geometries, is known as conformal mapping. It is explained in several advanced textbooks on static ﬁelds (e.g., Plonsey and Collin, 1961), and cannot be dealt with in this book because of space limits. When applied to the geometry of Figure 8.17, part a), under the assumption of negligible thickness t, it yields a per-unit-length capacitance such that the corresponding characteristic impedance is: K(k) η √ , (8.91) Zc ≈ 4 K( 1 − k 2 ) where η = (μ/)1/2 is, as usual, the characteristic impedance of the homogeneous dielectric medium, K(k) indicates the so-called complete elliptic integral of the ﬁrst kind (Abramowitz and Stegun, Eds., 1965), and:

πw −1 k = cosh , (8.92) 4d the symbols w and d being explained in Figure 8.17, part a). The practical use of Eq. (8.91) can be simpliﬁed by using two approximate expressions for the elliptic integral, namely:

πw η ln 2 coth , (8.93) Zc 2π 8d for w/d ≤ 1.12, and: πη Zc , (8.94) 8[ln 2 + πw/(4d)] for w/d > 1.12. The assumption of negligible thickness, on the other hand, makes it impossible to calculate the ohmic loss for a ﬁnite conductivity and the consequent attenuation constant. Indeed, a conductor of zero cross-section and ﬁnite conductivity would entail an inﬁnitely large loss, which is obviously unrealistic. For a ﬁnite thickness t, a perturbation approach like that of Section 7.9 yields an attenuation constant: πw/2d + ln(8d/πt) Rs αc = , (8.95) 2Zc d ln 2 + πw/4d where RS is, once again, the surface resistance of a conducting half-space, which we found ﬁrst in Section 4.12. 8.11.2

Second example: the microstrip

It is easy to show that, as we said before, for all the structures shown in Figure 8.17 except for part a), the fundamental mode cannot be rigorously TEM, as they encompass more than one dielectric. Indeed, from the decomposition theorem of Section 3.8 we see that a nontrivial TEM solution (i.e.,

© 2006 by Taylor & Francis Group, LLC

300

WAVES ON TRANSMISSION LINES

with nonzero transverse components) requires that the phase constant be √ β = ω μi in each medium (i = 1, 2), which is clearly impossible if 1 = 2 . Approximations are required to begin with. We will outline the way of thinking with reference to the microstrip (Figure 8.17, part b) which is by far and large the most widely used in practice, but the essential features apply also to the other structures. At relatively low frequency, we may still assume that the longitudinal components are negligible compared to the transverse ones. (The reader may justify this, as an exercise, in quantitative, terms starting from the decomposition theorem). Within this approximation, the only problem we are left with is to calculate a per-unit-length capacitance, to replace Eq. (8.91). However, conformal mapping no longer provides exact results when the dielectric is not homogeneous. Another approximation is needed; the most widely accepted way to proceed is to calculate ﬁrst Eq. (8.91) as if the dielectric were air everywhere, and then correct it by means of an “equivalent” relative permittivity, deﬁned as that of a dielectric medium ﬁlling all the half-space above the ground plane, yielding the same per-unit-length capacitance as that of Figure 8.17, part b). Several answers can be found in the literature. The most commonly used one is given in Schneider (1969): (εr − 1) 1 . (8.96) 1+ εr = 1 + 2 1 + 10d/w At higher frequencies, neglecting the longitudinal components yields unacceptable results, in particular for the phase constant, which is of paramount importance in designing components and circuits. An exact theory, in terms of potentials, can be found in the literature (Collin, 1991), but is too long and demanding to be reported here. A widely used approximation is that of Yamashita et al. (1979): √ εr − εr β = + εr , (8.97) k0 1 + 4F −1.5 where k0 = 2π/λ0 is the free-space phase constant, and: √

4d εr − 1 w 2 F = 0.5 + 1 + 2 ln 1 + λ0 d

.

(8.98)

One essential feature is that the dispersion in this structure becomes larger when the relative permittivity becomes higher, and when the ratio w/d becomes smaller, i.e., when the structure becomes more compact. Therefore, design of microstrip circuits is inevitably a trade-oﬀ between contradictory requirements. As for attenuation due to ﬁnite conductivity, its order of magnitude can be estimated with a rather rough approximation, if a microstrip is considered as a pair of stripes of identical width w. The result of the perturbation approach is then simply: (8.99) αc = Rs /(wZc ) ,

© 2006 by Taylor & Francis Group, LLC

THE COUPLED-MODE EQUATIONS

301

where Rs is the usual surface resistance of a half-space good conductor, and Zc is the characteristic impedance of the microstrip. In agreement with intuition, Eq. (8.99) overestimates the attenuation, since, in reality, the width of the conducting backplane is much larger than w. More precise calculations are easily found in modern textbooks on microwaves, and are omitted here because of space limits. 8.12

The coupled-mode equations

Completeness of a set of modes in a perfect cylindrical structure was discussed in Section 7.6. The conclusion was that a suitable countable inﬁnity of orthogonal modes could form a basis for expanding any ﬁeld in such a structure. One must reconsider completeness, from another point of view, when dealing with an imperfect structure, I, which diﬀers slightly from a perfect one, P , whose modes are known, whilst those of the imperfect structure cannot be found exactly. A typical example in this sense is a waveguide whose metallic walls have a ﬁnite conductivity (see Section 7.9). In such a case, a set of modes {Ei , Hi } is said to be complete when it satisﬁes the following convergence requirement: n n ∗ ∗ ∗ lim Vi (z) Eti × Ht (z) − Ii (z) Hti · a Et (z) − ˆz dS = 0 , n→∞

S

i=1

i=1

(8.100) where {Et , Ht } are the transverse components of any ﬁeld which may exist in I; S is the cross-section of I; {Eti , Hti } are the (z-independent) transverse components of the i-th mode of the perfect structure P ; {Vi , Ii ; i = 1, 2, . . .} are complex coeﬃcients, which, in general, depend on z. For the time being, we may assume that these coeﬃcients are adimensional. The condition in Eq. (8.100) means that the two series Σ(Vi Eti ) and Σ(Ii Hti ) converge towards Et and Ht , respectively, at any point on S , except, at most, the points of a zero-measure subset. To state and to prove what are the most general conditions for Eq. (8.100) to be satisﬁed, is long and complicated, but may be found in the literature (Collin, 1991). We prefer to invest some space and eﬀort in introducing a formalism which is convenient to use whenever Eq. (8.100) holds. Through it, from Maxwell’s equations one obtains a set of ordinary diﬀerential equations (ODE’s) for {Vi (z), Ii (z)}. The derivation is formally reminiscent of that of Section 8.10, and therefore may be phrased in the typical terminology of transmission lines. The only problems we are going to face in this derivation come from the fact that, for the two series in Eq. (8.100), uniform convergence is not assured at every point. Thus, we are not guaranteed that they can be diﬀerentiated term by term. In the structure I, where there are no sources, Maxwell’s equations, projected on the z = constant plane, read: ∂Et ∂z

© 2006 by Taylor & Francis Group, LLC

= jωμ a ˆ z × H t + ∇ t Ez

,

302

WAVES ON TRANSMISSION LINES

∂Ht = −jω a ˆ z × Et + ∇ t H z . (8.101) ∂z When Eq. (8.100) holds, in the previously speciﬁed sense of this statement, we may set: Et =

∞

Vi (z) Eti

,

Ht =

i=1

∞

Ii (z) Hti

.

(8.102)

i=1

Let us recall that {Ei , Hi } is the i-th mode of the structure P . For the sake of generality, we will suppose that it is a hybrid mode (i.e., Ez = 0, Hz = 0). Then, using Eqs. (7.11) and (7.12), we may write: Eti Hti

jγi ∇t Ti ω

,

jγi ∇t Li + ∇t Ti × a ˆz ωμ

.

= ∇t Li × a ˆz − =

(8.103)

The functions Li and Ti satisfy: ∇2t Fi = −χ2i Fi

(χ2i = γi2 − κ2 ) .

(8.104)

To obtain the ODE’s in Vi (z) and Ii (z) we are looking for, at ﬁrst sight one would think it is enough to express Ez and Hz in terms of Vi and Ii in Eqs. (8.101). Another step, based on mode orthogonality in the perfect structure P , would then allow us to isolate individual coeﬃcients (i = 1, 2, . . .) in Eqs. (8.101). In reality, this procedure would not be correct. As the series do not converge uniformly, ∇t Ez and ∇t Hz in Eqs. (8.101) cannot be calculated by term-by-term diﬀerentiation of Eqs. (8.102). It is necessary to pass ﬁrst from Eqs. (8.101) to equations where Ez and Hz do not go through diﬀerential operators. To this purpose, let us take the scalar product of the ﬁrst of Eqs. (8.101) ∗ ∗ /2, and of the second one with a ˆz × Etm /2. Let us integrate with a ˆz × Htm over S , take into account mode orthogonality, and impose on each of them a normalization condition (power through S equal to 1 W). All these operations result in the following equation: ∗ Eti × Htm ·a ˆz dS = δi,m , (8.105) 2 S where δi,m is Kronecker’s delta. Let us write also the following identities, which can be proved by the reader as an exercise using Appendix C and taking into account Eq. (8.104): ∇t Ez · ∇t Tm ∇t Ez × ∇t Lm

= ∇t · (Ez ∇t Tm ) + χ2m Ez Tm = ∇t × (Ez ∇t Lm ) .

,

(8.106) (8.107)

Let us also consider two identities which diﬀer from the previous ones only because Ez is replaced by Hz , and because Tm and Lm have been interchanged. We omit writing them explicitly for the sake of brevity.

© 2006 by Taylor & Francis Group, LLC

THE COUPLED-MODE EQUATIONS

303

In the equations which were derived from Eqs. (8.101), all the terms involved in the latest identities show up inside integrals over the surface S . Gauss’ theorem (A.8) can be applied to the integral containing the ﬁrst term on the right-hand side in Eq. (8.106), and to the twin term, containing Hz . Stokes’ theorem (A.9) can be applied to the integral containing the right-hand side of Eq. (8.107), and to the twin term containing Hz . The results are the following equations: dVm jγ ∗ = −jωμ Im + m χ2m Ez Tm∗ dS Nm dz ω S ∗ ∂Tm∗ jγm ˆ d , (8.108) d + + Ez Ez (∇t L∗m · ) ω ∂n 1 dIm Nm dz

∗ jγm 2 = −jω Vm − χ Hz L∗m dS ωμ m S ∗ ∂L∗m jγm ˆ d , (8.109) d + − Hz Hz (∇t Tm∗ · ) ωμ ∂n

where: • ∂/∂n is the normal derivative; • = S ∩ C is the boundary of the section S , oriented clockwise with reference to the unit vector a ˆz ; ˆ • is the unit vector tangent to ; • Nm = |Etm |/|Htm | is a generalization, for the m-th mode, of the constant N of Eq. (8.74), and is proportional to the wave impedance of the m-th mode. At this stage, Ez and Hz in Eqs. (8.108) and (8.109) are no longer arguments of diﬀerential operators. Thus, non-uniform convergence of their series expansions does not cause problems any more. These series expansions can be obtained through the longitudinal components of Maxwell’s equations, if we let the transverse ﬁelds be expressed by Eqs. (8.102), namely: ∞

Ez

1 ∇t × Hti ∇ × Ht = Ii (z) jω jω i=1

=

=

∞

Ii (z) Ezi = −

i=1

∞ 1 Ii (z) χ2i Ti jω i=1

,

(8.110)

.

(8.111)

∞

Hz

= −

=

1 ∇t × Eti ∇ × Et = Vi (z) jωμ −jωμ i=1

∞ i=1

© 2006 by Taylor & Francis Group, LLC

Vi (z) Hzi =

∞ 1 Vi (z) χ2i Li jωμ i=1

304

WAVES ON TRANSMISSION LINES

Let us now place these expressions into Eqs. (8.108) and (8.109), and use commutation between the integral over S and the sum with respect to i. Taking into account further orthogonality relations, of the type in Eq. (7.35), between pre-potentials of the perfect structure P , the only terms among the integrals over S which survive are those whose arguments contain products of the type Lm L∗m and Tm Tm∗ . Because of the normalization in Eq. (8.105), the integrals containing these products are constants: Lm L∗m dS = Am , (8.112) S

S

Tm Tm∗ dS

= Bm

,

(8.113)

although it is impossible to specify their explicit values as long as the boundary conditions on the walls of P are not speciﬁed. In conclusion, Eqs. (8.108) and (8.109) become: dVm dz

∞ jωμ = − Im + cm Im + Cmn In Nm n=1

dIm dz

= −jω Nm Vm + dm Vm +

∞

,

Dmn Vn

,

(8.114)

n=1

whose coeﬃcients are shown in Table 8.1. In the perfect structure P , orthogonality assures that each mode may exist by itself. This implies that in P it must be Cmn = 0 and Dmn = 0 for m = n. Then, Eqs. (8.114) reduce to the standard transmission line equations that we studied for most of this chapter. This makes it self-explanatory why in general Eqs. (8.114), for the imperfect structure I, are referred to as the coupled-line equations, or also as the generalized telegraphist equations. Analogy with the previous section entails us to let, for each mode: " η Nm (am − bm ) . (am + bm ) , Im = (8.115) Vm = Nm η Introducing these new variables into Eqs. (8.104), we get the following equations, usually referred to as the coupled-mode equations: dam = −Γm am + (Fmn an + Gmn bn ) , dz n=m

dbm dz

= −Γm bm −

(Gmn an + Fmn bn ) ,

(8.116)

n=m

where Γm is the propagation constant of the m-th mode, provided it is alone (i.e., an = bn = 0 for n = m) in the imperfect structure I. In general, it is Γm = γm (the propagation constant of the same mode in the perfect structure), as Γm accounts also for the other terms in Eqs. (8.114) that correspond to n = m.

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

305

Table 8.1 Expressions for the coeﬃcients of the coupled-mode equations. Cm = −

∗ χ4m γm Bm ω 2 2 Nm

dm = −

χ4m ∗ γm Nm Am ω 2 μ2

Cmn = −

∗ 2 1 ! χn 1 ! ∂Tm∗ γm χ2n ˆ d T + Tn (∇t L∗m · ) d n ω 2 2 Nm ∂n −jω Nm

Dmn = −

∗ 2 ! ! χn ∂L∗m γm χ2n ˆ d N L + d Nm Ln (∇t L∗m · ) m n ω 2 μ2 ∂n jωμ

The mode coupling coeﬃcients are related to the quantities shown in Table 8.1 via: Nn Nm Cmn + Dmn , 2Fmn = Nm Nn Nn Nm 2Gmn = Cmn − Dmn . (8.117) Nm Nn Many examples of solutions of Eqs. (8.116) are available in the literature. Their physical meaning changes from problem to problem. Perhaps the most typical example is that of the directional coupler, which was brieﬂy illustrated in Section 3.9 when we discussed symmetries and their implications. Further suggestions to the reader will be given in the next section. 8.13

Further applications and suggested reading

Solving Maxwell’s equations, or using circuit theory, are, as we said on several occasions, two diﬀerent approaches to one physical reality. The diﬀerence is in the axioms. In circuit theory, the axioms are Ohm’s law and Kirchhoﬀ’s laws. In our approach to e.m. waves, the axioms are Maxwell’s equations and the charge continuity equation. In this chapter we have shown that, with an additional assumption (namely, that of TEM propagation), the ﬁeld approach leads to the same transmission lines equations as the circuit approach. The link between the two approaches is broader, in its scope of validity, than TEM waves, as we outlined brieﬂy in Section 8.10. Hence, one of the most important messages that the reader can extract from this chapter is that, in order to understand more in depth high-frequency e.m. ﬁelds, one should have a strong background in circuit theory also. We recommend reading the classical treatise by Newcomb (1966) that we have quoted on several occasions in this book, and at least one other text, more focused on distributed-parameter circuits, such as, for instance, Ghausi and Kelly (1968). In particular, the Smith chart of Section 8.7 is a crucial intersection between the circuit culture and the e.m. wave culture. On this front, further interest may be stimulated in those readers who like abstraction and mathematical approaches, by saying that bilinear transforms help to solve many other problems related to impedance transformations through various types of linear

© 2006 by Taylor & Francis Group, LLC

306

WAVES ON TRANSMISSION LINES

networks. The short but condensed booklet by Schwerdtfeger (1961), that we quoted when we introduced the Smith chart, is also a good starting point in this direction — see also Liboﬀ and Dalman (1985). Another valuable resource is Chapter 7 (by R. Beringer) of Montgomery, Dicke and Purcell (1948). The reader is encouraged to consult it after studying our Chapter 9 on resonant cavities. A point that might look trivial at ﬁrst sight, but in our opinion is not so, is that transmission-line based problems show up in a very large number, and in a very large variety, of occasions in the professional life of anybody in electronic engineering. Solving many problems is a valuable investment for an EE student, in view, for example, of an employment interview. For this reason, the collection of problems at the end of this chapter is wider than the average over the rest of the book. All other textbooks covering this subject are good sources of problems. The last comment has brought us towards rather practical issues. Continuing in the same direction, let us stress that what has been said in this chapter (and in the whole book) on striplines and microstrips, has to be thought of just as an elementary introduction. All telecommunication engineers should learn much more than that. Many ﬁrst-class textbooks cover the area both under the viewpoint of theory and that of applications. Just to quote a few, Collin (1991) is highly recommended in the ﬁrst category; Wadell (1991), Gupta et al. (1996), and Bhat and Koul (1989) are representative for the second one. Other useful texts include Freeman (1996) and Elliott (1993) (general microwave texts), and Christopoulos (1995), Fache et al. (1993), Faria (1993) (texts on mathematical modeling of microwave and strip transmission lines). On the other hand, the reader must be encouraged not to underestimate the power of the rather abstract approach of Sections 8.10 and 8.12. Without these connections, it would be very hard to understand why the transmission line equations and the coupled-mode equations show up in problems where there are no conductors at all, such as integrated optics, or where it remains hard to visualize two coupled lines, such as microwave tubes based on electron beams, or particle accelerators, or traveling-wave microwave masers. Suggested reading in these areas include Unger (1977) on planar dielectric waveguides, Tamir, Ed., (1979) on integrated optics, Gilmour (1986), Chodorow and Susskind (1964), and Sims and Stephenson (1963) on microwave tubes, Pierce (1950/1954/1974) on traveling wave tubes, electron beams, and waves as well as some excellent popularizations, Louisell (1960) on coupled modes, Roussy and Pearce (1995) on industrial microwave device applications. However, especially in the case of integrated optics, it is preferable for the reader to consult these books only after studying our Chapter 10. Let us conclude with a comment on the coupled-mode equations (CME’s). They are still a popular subject of discussion and investigation, especially in integrated optics. One of the reasons why they are popular is because some important devices (typically, directional couplers and ﬁlters) cannot be studied analytically without approximation. However, there is a more subtle reason.

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

307

It is because the “perfect” structure P one starts from, in order to study an “imperfect” structure I (see Section 8.12), is not unique. For example, to study a directional coupler which encompasses a grating, one can start either from the modes of the two isolated waveguides, or from those (sometimes designated as supermodes) of two parallel waveguides. These also are available in exact closed form, for some geometries. Looking back at the general derivation of Section 8.12, the reader, if familiar with perturbation theory, should be able to understand why the accuracy of the perturbation solutions depends on the perfect structure one starts from. For very instructive details on this point, recent research papers by Huang et al. (1996) should be consulted, since, to the best of our knowledge, the subject has not yet been published as a book — see Huang (1994) for an overview, and, e.g., Sarangan and Huang (1994), and Huang and Hong (1992) for example applications. Problems 8-1 Express the active and the reactive power in the generic section of a transmission line in terms of the reﬂection coeﬃcient, separating real and imaginary parts of Eq. (8.15). Assume ﬁrst that the characteristic impedance of the transmission line is real. Then, repeat the exercise assuming a complex characteristic impedance. 8-2 Extend all the results of Section 8.6 to the case of low-loss transmission lines. 8-3 For the lossless transmission lines shown in Figure 8.19, we have R G = Zc1 = 50 Ω, Zc2 = 100 Ω, VG = 10 V. In the line with characteristic impedance Zc1 , the VSWR equals 4/3. The voltage modulus goes through a minimum where there is a discontinuity in the characteristic impedance. a) Evaluate the load impedance ZL . b) Calculate the complex power delivered to the load impedance for XG = 0, for XG = RG , and for XG = −RG , assuming that the length of the line with characteristic impedance Zc1 equals an odd integer multiple of λ/4. 8-4 In the lossless transmission line shown in Figure 8.20, the phase constant is the same as in free space. The frequency is f = 100 MHz. We have RG = Zc = 50 Ω, RL = 60 Ω, VG = 20 V, C = (1/π) pF, L = 0.1/(4π) mH. The line lengths, 1 and 2 , are both equal to 2 m. a) Set R1 = 40 Ω, and calculate the active power delivered to the load resistance RL . b) Find the range of values of R1 for which the total active power, delivered to R1 and to RL , is more than one half of the generator’s available power. 8-5

For the lossless transmission line shown in Figure 8.21, at the frequency

© 2006 by Taylor & Francis Group, LLC

308

WAVES ON TRANSMISSION LINES d

jXG ZL

RG VG

ZC2

ZC1

Figure 8.19

ZG

R

VG

ZC

L

C

RL

ZC z=a

z=b

z=0

Figure 8.20

f = 600 MHz, we have Zc = 200 Ω, d1 = 15 m, XL = 100 Ω, XC = −400 Ω, VG = 10 V. In the Zc1 part of the line, the voltage modulus goes through a minimum 0.5 m on the left of the terminals 22 , while the VSWR is 4. The active power delivered to the load R equals 40 mW. a) Find all the values of the distance d2 which are compatible with these data. b) Find the unknown values of the resistances R and RG . c) Find voltage and current at a generic distance from the generator. 8-6 Reconsidering the transmission line shown in the ﬁgure of Problem 84, and retaining all the other parameter values, ﬁnd the internal impedance 1

RG

2

d1

XL

VG

XC

ZC 2’

Figure 8.21

© 2006 by Taylor & Francis Group, LLC

3

R

ZC 1’

d2

3’

FURTHER APPLICATIONS AND SUGGESTED READING

309

of the voltage generator which satisﬁes the power matching condition at the generator terminals. Repeat the problem modeling the generator as an ideal current generator in parallel with an internal admittance. 8-7 Verify that one can design a single-stub impedance-matching network using an analytical approach, instead of using the Smith chart, but that the procedure becomes quite cumbersome. 8-8 Work out in full detail the procedures to design a double-stub impedancematching network for the two cases shown in Figure 8.14 but not discussed in the text, i.e., when the distance between the stubs is equal either to λ/8 or to 3λ/8. 8-9 Enumerate relative advantages and disadvantages of single- and doublestub impedance-matching networks. Distinguish between two situations. First, consider a laboratory environment, where the load impedance is changed often, and in general is not known a priori. Second, consider a mass production environment, where many identical pieces are fabricated. 8-10 A low-loss transmission line, whose nominal characteristic impedance is 50 Ω, exhibits a loss of 2 dB over a length of 10 m. The loss contributed by the parallel conductance is negligibly small. Calculate the per-unit-length series resistance r, and the corresponding correction to the phase constant with respect to the lossless case, at 900 MHz, assuming a per-unit-length capacity of 50 pF/m. 8-11 For the network discussed in Problem 8-3, design a single-stub impedance-matching network at the frequency f = 600 MHz, to be placed upstream with respect to the discontinuity in the characteristic impedance, but as close as possible to it. Discuss two possibilities, namely a stub with characteristic impedance Zc1 and with characteristic impedance Zc2 . Calculate the complex power at the load, at the discontinuity, and at the terminals of the stub. 8-12 For the network discussed in Problem 8-4, design a quarter-wavelength impedance-matching network to be placed upstream with respect to the resistor R = 40 Ω. Find both solutions, one with an increase and the other with a decrease in characteristic impedance. Then, increasing the frequency to 200 MHz, and leaving the whole network unchanged, calculate the VSWR at the generator terminals. 8-13 A lossless transmission line, with phase constant equal to that of free space and characteristic impedance Zc = 100 Ω, is closed on a load impedance ZL = 100 − j50 Ω, and fed by a generator with an open-circuit voltage VG = 10 V and an internal impedance ZG = 100 + j50 Ω. The operating

© 2006 by Taylor & Francis Group, LLC

310

WAVES ON TRANSMISSION LINES

frequency is 300 MHz. Design a double-stub impedance-matching network connected to the load terminals, and with a spacing d = 0.25 m between the stubs. Study how the current and voltage vary with z, from the generator to the load. Calculate the complex power at the terminals of the load and at those of the two stubs. 8-14 For a lossless coaxial cable (whose outer and inner radii are denoted as a and b, respectively), show that if voltage and current are deﬁned as in the electrostatic case (i.e., according to Eqs. (8.83)), then the characteristic impedance of the cable equals Zc = (μ/)1/2 ln(a/b)/(2π). 8-15 In an air-ﬁlled coaxial cable, the inner conductor is of radius b = 2.5 mm. Making use of the formula that was proved in the previous problem, adjust the radius of the outer conductor such that Zc = 120 Ω. 8-16 The cable of the previous problem is connected to a 50 Ω resistive load impedance, at the frequency of 300 MHz. A metallic sleeve can be used as a λ/4 impedance-matching device. Find its length, its outer radius, and its distance from the load terminals. Repeat the same problem for a resistive load impedance of 300 Ω. 8-17 Show, using the decomposition theorem of Section 3.8 or of Section 7.2, that, within each homogeneous dielectric medium present in a microstrip, the longitudinal ﬁeld components are negligibly small compared to the transverse ones. 8-18 Using one of the approximations, Eqs. (8.93) or (8.94), design a stripline, in a homogeneous dielectric medium having r = 2.2 and a spacing d = 10 mm, so that its characteristic impedance is 50 Ω. For copper conductor whose thickness is t = 0.2 mm, evaluate the attenuation constant vs. frequency between 1 and 2 GHz. (Copper conductivity: 5.8 × 107 Ω−1 m−1 .) 8-19 A microstrip for which w/d = 4 is fabricated with copper on alumina, the dielectric for which r = 9.8. Evaluate the equivalent permittivity, Eq. (8.96), and then the corresponding phase constant, dispersion (i.e., the second derivative of β with respect to ω), and attenuation constant, at 3.5 GHz. 8-20 Using Appendix C and taking into account Eq. (8.104), derive Eqs. (8.106) and Eq. (8.107).

© 2006 by Taylor & Francis Group, LLC

References Abramowitz, M. and Stegun, I.A. (Eds.) (1965) Handbook of Mathematical Functions. Dover, New York. Bhat, B. and Koul, S.K. (1989) Stripline-Like Transmission Lines for Microwave Integrated Circuits. Wiley, New York. Chodorow, M. and Susskind, C. (1964) Fundamentals of Microwave Electronics. McGraw-Hill, New York. Christopoulos, C. (1995) The Transmission-Line Modeling Method: TLM. IEEE/OUP Series on Electromagnetic Wave Theory, Institute of Electrical and Electronics Engineers, New York/Oxford University Press, Oxford. Collin, R.E. (1991) Field Theory of Guided Waves. IEEE Press, New York. Corazza, G.C. (1974) Fondamenti di Campi Elettromagnetici e Circuiti. Vol. II, Patron, Bologna. Dicke, R.H. and Wittke, J.P. (1960) Introduction to Quantum Mechanics. AddisonWesley, Reading, MA. Elliott, R.S. (1993) An Introduction to Guided Waves and Microwave Circuits. Prentice-Hall, Englewood Cliﬀs, NJ. Fache, N., Olyslager, F. and de Zutter, D. (1993) Electromagnetic and Circuit Modelling of Multiconductor Transmission Lines. Oxford Engineering Science Series, Clarendon Press, Oxford/Oxford University Press, New York. Faria, J.A.B. (1993) Multiconductor Transmission-Line Structures. Wiley Series in Microwave and Optical Engineering, Wiley, New York. Freeman, J.C. (1996) Fundamentals of Microwave Transmission Lines. Wiley Series in Microwave and Optical Engineering, Wiley, New York. Ghausi, M.S. and Kelly, J.J. (1968) Introduction to Distributed-Parameter Networks: with Application to Integrated Circuits, Holt, Rinehart and Winston, New York. Gilmour, A.S., Jr. (1986) Microwave Tubes. Artech House Microwave Library, Artech House, Dedham, MA. Gupta, K.C. et al. (1996) Microstrip Lines and Slotlines. 2nd ed., Artech House Microwave Library, Artech House, Boston. Huang, W.-P. and Hong, J. (1992) “A coupled-waveguide grating resonator ﬁlter”. IEEE Photonics Technology Letters, 4 (8), 884–886. Huang, W.-P. (1994) “Coupled-mode theory for optical waveguides: an overview”. J. Optical Society of America A, 11 (3), 963–983. Huang, W.-P., Guo, Q. and Wu, C. (1996) “A polarization-independent distributed Bragg reﬂector based on phase-shifted grating structures”. J. Lightwave Technology, 14 (3), 469–473. Itoh, T. (Ed.) (1987) Planar Transmission Line Structures. IEEE Press Selected Reprint Series, IEEE Press, New York. Liboﬀ, R.L. and Dalman, G.C. (1985) Transmission Lines, Waveguides, and Smith Charts. Macmillan, New York. 311 © 2006 by Taylor & Francis Group, LLC

312

REFERENCES

Louisell, W.H. (1960) Coupled Mode and Parametric Electronics. Wiley, New York. Montgomery, C.G., Dicke, R.H. and Purcell, E.M. (Eds.) (1948) Principles of Microwave Circuits. McGraw-Hill, New York. Reprinted by Peter Peregrinus, London, for IEE (1987). Morse, P.M. and Feshbach, H. (1953) Methods of Theoretical Physics. McGraw-Hill, New York. Newcomb, R.W. (1966) Linear Multiport Synthesis. McGraw-Hill, New York. Pierce, J.R. (1950) Traveling-Wave Tubes. Bell Telephone Laboratories Series, Van Nostrand, New York. Pierce, J.R. (1954) Theory and Design of Electron Beams. Bell Telephone Laboratories Series, 2nd ed., Van Nostrand, New York. Pierce, J.R. (1974) Almost All About Waves. MIT Press, Cambridge, MA. Plonsey, R. and Collin, R.E. (1961) Principles and Applications of Electromagnetic Fields. McGraw-Hill, New York. Roussy, M.G. and Pearce, J.A. (1995) Foundations and Industrial Applications of Microwave and Radio Frequency Fields: Physical and Chemical Processes. Wiley, New York. Sarangan, A.M. and Huang, W.-P. (1994) “A coupled mode theory for electron wave directional couplers”. IEEE J. Quantum Electronics, 30 (12), 2803–2810. Schneider, M.V. (1969) “Microstrip lines for microwave integrated circuits”. Bell System Technical Journal, 48 (5), 1421–1444. Schwerdtfeger, H. (1961) Introduction to Linear Algebra and the Theory of Matrices. 2nd ed., Noordhoﬀ, Groningen. Sims, G.D. and Stephenson, I.M. (1963) Microwave Tubes and Semiconductor Devices. Interscience Publishers, New York. Tamir, T. (Ed.) (1979) Integrated Optics. 2nd corrected and updated ed., SpringerVerlag, Berlin. Unger, H.-G. (1977) Planar Optical Waveguides and Fibres. Clarendon Press, Oxford. Wadell, B.C. (1991) Transmission Line Design Handbook. Artech House Microwave Library, Artech House, Boston. Yamashita, E., Atsuki, K. and Ueda, T. (1979) “An approximate dispersion formula of microstrip lines for computer-aided design of microwave integrated circuits”. IEEE Trans. Microwave Theory & Techniques, MTT-27 (12), 1036–1038.

© 2006 by Taylor & Francis Group, LLC

CHAPTER 9

Resonant cavities 9.1

Introduction

In the previous chapters, we showed what theoretical background is necessary, in order to comply with a self-explanatory technical requirement: how to convey electromagnetic energy along a given path. Similarly, this chapter is devoted to the theoretical basis required to answer another very important technical requirement: how to store electromagnetic energy in a given region in 3-D space. In elementary circuit theory, as well known, one usually starts oﬀ by introducing two ideal components, the capacitor and the inductor. Their purpose is to model, in the most simplistic way, two physical realities, that consist of storing energy, electric and magnetic, respectively, within given regions. The next step is to combine these two elements, and introduce the concept of a resonant circuit. This approach is strongly biased by previous knowledge of electrostatics and magnetostatics. Indeed, the two ideal components are rather easy to deﬁne as long as electric and magnetic ﬁelds are time-invariant. But in a rapidly time-varying regime, electric and magnetic ﬁelds are inseparable. Then, as we will see in this chapter, if we start from Maxwell’s equations, we may run directly into some objects whose behavior is essentially that of resonant LC circuits; but the main diﬀerence is that it is impossible to split these objects into two separate L and C components. As we saw in the previous chapter (and also in terms of plane waves, in Chapter 4), the input impedance of a short-circuited single-mode line is purely reactive. It is inductive, if the length of the line is less than λg /4, capacitive if λg /4 < < λg /2. Let us consider a transmission line of length λg /2, shortcircuited at both ends (Figure 9.1). Any plane orthogonal to the z axis (except for the one z = λg /4) splits the line in two parts, one of them having an inductive input impedance, the other having a capacitive input impedance. The choice of this plane is, once again, completely arbitrary; this indicates that the resonant behavior is an intrinsic property of the whole system in itself. The deﬁnition of a capacitor and an inductor is an artifact, and was used just as a link with known concepts. In the example shown in Figure 9.1, a complete round trip, from one of the short-circuit planes to the other and back, along the z axis, has a total length of a wavelength. Consider then a wave that passes through the generic point P with a given phase φ. When it passes again through P , in the same sense as before, it has the same phase φ. In the remainder of this chapter we will see 313 © 2006 by Taylor & Francis Group, LLC

314

RESONANT CAVITIES

Figure 9.1 A short-circuited coaxial cable, half wavelength in length, seen as a resonant circuit.

that this comment has a very broad scope of validity. A resonant behavior always entails that there are closed paths (in general, three-dimensional paths) whose total length is an integer multiple of a wavelength. In Chapter 7, we saw that Maxwell’s equations with 2-D boundary conditions admit solutions at any frequency. In this chapter, on the contrary, we will see that 3-D boundary conditions can be satisﬁed only at discrete frequencies; this is a consequence of the phase conservation requirement along closed paths, that we just illustrated. This frequency selection process reinforces the analogy with networks consisting of lumped inductors and capacitors, which also may resonate only at discrete frequencies. 9.2

Separable coordinate systems in three dimensions

In some of the previous chapters, we saw that separation of variables is a powerful method to solve the Helmholtz equation. On the other hand, we saw in Chapter 7 that it suﬀers from intrinsic limitations, due to the ﬁnite number of reference systems in which it can be applied. We will see now brieﬂy how those limitations extend from the 2-D to the 3-D case. An electromagnetic resonator is, once again, a region in 3-D space, surrounded by a surface S, which can either be completely at a ﬁnite distance from an origin (closed resonators), or extend in part to inﬁnity (open resonators). On this surface, we impose boundary conditions that the ﬁeld must satisfy. When the Helmholtz equation is solved by separation of variables, in a given reference system, boundary conditions are simple to express only if the surface S is a coordinate surface, or can be subdivided into coordinate surfaces, in that reference system. When this is not the case, then we may still use a complete set of solutions obtained by separation of variables, but the ﬁeld must be written as a series expansion. Such a procedure is always a source of nuisance; sometimes, it causes conceptual problems also, as consequences of whether it is legitimate or not, to diﬀerentiate a series term by term. For these reasons, it is very important to know from the beginning that the 3-D Helmholtz equation is separable only in a ﬁnite number of coordinate systems. We must omit the proof for the sake of brevity; it is very similar to

© 2006 by Taylor & Francis Group, LLC

COMPLETENESS OF RESONATOR MODES

315

that of the 2-D case, and can be found in the same references. In any case, we will provide now the list of those coordinate systems (the so-called “separable coordinate systems”), where separation of variables is applicable. The most general case (in the sense that all the others may be looked at as its degenerate cases) is that of orthogonal ellipsoidal coordinates (ξ1 , ξ2 , ξ3 ), which are implicitly deﬁned by the following relations, where (x1 , x2 , x3 ) is an orthogonal cartesian system: 1/2 (ξ12 − a2i ) (ξ22 − a2i ) (ξ32 − a2i ) i = 1, 2 , xi = j = 2, 1 (= i) a2i (a2i − a2j ) x3

=

ξ1 ξ2 ξ3 a1 a2

,

ξ1 > a1 > ξ2 > a2 > ξ3 > 0 .

(9.1)

It can be proved (Morse and Feshbach, 1953) that its coordinate surfaces are confocal, mutually orthogonal quadric surfaces, namely: • ellipsoids: ξ1 = constant; • one-fold hyperboloids: ξ2 = constant; • two-fold hyperboloids: ξ3 = constant. The degenerate cases, which complete the set of separable coordinates for the Helmholtz equation, are the following (in brackets, the corresponding coordinate surfaces): 1. orthogonal cartesian coordinates (planes); 2. circular cylindrical coordinates (circular cylinders; half-planes; planes); 3. elliptical cylindrical coordinates (elliptical cylinders; hyperbolic cylinders; planes); 4. parabolic cylindrical coordinates (two families of parabolic cylinders; planes); 5. rotation parabolic coordinates (two families of rotation paraboloids; planes); 6. paraboloidal coordinates (two families of elliptical paraboloids; one family of hyperbolic paraboloids); 7. spherical coordinates (spherical surfaces; cones; planes); 8. prolate spheroidal coordinates (prolate spheroids; two-fold revolution hyperboloids; planes); 9. oblate spheroidal coordinates (oblate spheroids; one-fold revolution hyperboloids; planes); 10. conical coordinates (spherical surfaces; one-fold hyperboloids; cones with elliptical cross-section). 9.3

Completeness of resonator modes

Solving the Helmholtz equation by separation of variables is legitimate, similarly to what we saw in Chapters 4 and 7, provided we can prove that the

© 2006 by Taylor & Francis Group, LLC

316

RESONANT CAVITIES

solutions found in this way form a complete set. Completeness of resonator modes can be proved similarly to what we saw in Chapter 4 for plane waves, and in Chapter 7 for waveguide modes. When we compared Chapter 4 and Chapter 7, we noticed that the dimension of a complete set decreased (from a continuum to a countable inﬁnity), as we passed from the very mild restriction that the ﬁeld be square integrable, to boundary conditions at ﬁnite distance along two directions. At a given frequency, the component of the propagation vector along a given direction can vary continuously in an unbounded medium, but is discretized in a waveguide. We may wonder now whether in a resonator (for simplicity, a closed one) there is a further reduction in the size of a complete set, as we impose boundary conditions along three directions. The answer can be given intuitively, going back to the case (see Section 9.1) of a cavity that consists of a ﬁnite length of a lossless cylindrical structure, whose transmission modes we assume to know. Two perfect conductors close the structure at its ends. The i-th mode of the cylindrical guide can travel either way; so, we must associate it with two arbitrary complex constants, Pi , Qi . They represent amplitudes and phases of the two counterpropagating waves, on the plane z = 0. The boundary conditions that have been imposed at the ends of the resonator can be expressed as two linear homogeneous equations in Pi , Qi . This system has non-trivial solutions if and only if its determinant vanishes. From this condition we obtain an equation that admits solutions only for some discrete values of the longitudinal propagation constant of the i-th mode βi , and then, as βi is a function of frequency, only for suitable discrete values of ω, ωi,j = 2πfi,j (j = 1, 2, . . .). In conclusion, a lossless resonator, at a ﬁxed frequency, has always a ﬁnite number of resonant modes. This number is zero, when the frequency we are referring to is not one of those particular fi,j ’s that were deﬁned a few lines above, which are obviously called resonant frequencies. The corresponding modes of an ideal resonator form a complete set, in the same sense as those discussed in Chapter 7. If we take a structure whose geometrical shape is identical to that of the ideal resonator, but whose boundary conditions are diﬀerent from those on a perfect electric conductor, then we know that a suitable series of modes of the ideal resonator converges, in the linear mean sense, towards any ﬁeld that may exist in the non-ideal structure. This applies not only to resonators with ﬁnite conductivity walls, but also to resonators coupled to other structures, by means of holes, irises, or similar devices; they are referred to as loaded resonators. Actually, a completely closed resonant cavity, without any access, is an easy model to analyze; but it would be useless, if we had not noticed that, as a consequence of mode completeness, the ﬁeld in a loaded resonator may be expressed as a series of modes of the isolated resonator. As we saw in Chapter 7, in a non-ideal waveguide, mode orthogonality collapses; therefore, the ﬁeld in general cannot be thought of as being a single mode in a rigorous sense. Modes under their cut-oﬀ frequencies are necessary

© 2006 by Taylor & Francis Group, LLC

MODE ORTHOGONALITY IN A PERFECT RESONATOR

317

to fulﬁl the boundary conditions, where the real waveguide departs from the ideal one. In the same way, in a loaded resonator, a mode of the corresponding ideal resonator may be present at frequencies which diﬀer from its resonant frequency fn,m . Its amplitude coeﬃcient, if all the other parameters are kept constant, decreases as |f − fn,m | increases, but in principle it may diﬀer from zero at any frequency. This point will be clariﬁed, in more quantitative terms, in Section 9.8; but we thought it might be useful to notice now that it is strictly related to the notion of completeness. 9.4

Mode orthogonality in a perfect resonator

Similarly to what was carried out in Section 7.5, we will prove an orthogonality theorem for the modes of a perfectly lossless, completely isolated cavity. The last assumption entails that there are no imposed currents, neither inside the cavity, nor on its walls. The proof holds for two modes with diﬀerent resonant frequencies. For two modes with the same resonant frequency (called degenerate modes), one can always orthogonalize them, with a procedure based on projections and subtractions (known as Schmidt’s orthogonalization); this is essentially the same as we saw for transmission modes in a perfect waveguide, and therefore it will not be repeated here. The proof of the orthogonality theorem is similar to that of Poynting’s theorem (Section 3.2); the main diﬀerence comes from the fact that we have to deal with two modes, having resonant frequencies ωh , ωk , respectively. Both satisfy homogeneous Maxwell’s equations: (a)

∇ × Eh = −jωh μ Hh

(b)

∇ × Hh = jωh Eh

(c)

∇ × Ek = −jωk μ Hk

(d)

∇ × Hk = jωk Ek

,

, ,

.

(9.2)

Dot-multiply both sides of equation (a) times H∗k , and the complex conjugate of equation (d) by Eh . Then subtract side by side, and use the vector identity (C.4), which we rewrite here for convenience: ∇ · (Eh × H∗k ) = H∗k · ∇ × Eh − Eh · ∇ × H∗k

.

(9.3)

The net result is a formula, similar to one which one ﬁnds on the way towards Poynting’s theorem, before making a volume integration. We integrate it over the cavity volume V , and then apply Gauss’ theorem (A.8) to the integral of the left-hand side of Eq. (9.3) to convert it into a surface integral. It is easy to show that this integral vanishes. In fact, the perfect conductor surrounding the cavity imposes Etan = 0 at any point. Following again the same path as in Poynting’s theorem, we notice that two more terms disappear: the one due to sources (there are none in the cavity, by assumption), and the one due to

© 2006 by Taylor & Francis Group, LLC

318

RESONANT CAVITIES

losses (again zero by assumption). So, we are left with the following relation: ωh μ Hh · H∗k dV = ωk Eh · E∗k dV . (9.4) V

V

If we repeat the same steps starting from Eqs. (9.2) (b) and (c), and take the complex conjugate for each term in this new relation, we get: ∗ μ Hh · Hk dV = ωh Eh · E∗k dV . (9.5) ωk V

V

Now let us multiply Eq. (9.4) times ωk , Eq. (9.5) times ωh , and then subtract; we get the ﬁnal result, namely: Eh · E∗k dV = 0 . (9.6) (ωh2 − ωk2 ) V

Multiplying Eq. (9.4) times ωh and Eq. (9.5) times ωk , we obtain a similar result for the magnetic ﬁelds. Thus if the two modes have diﬀerent resonant frequencies, the following orthogonality relations hold: Eh · E∗k dV = 0 (ωh = ωk ) , (9.7) V

μ Hh · H∗k dV = 0

(ωh = ωk )

.

(9.8)

V

As we saw in Chapter 7, orthogonality is a strong form of linear independence, with links with the property of completeness. It is also of great help in practical problems, like expanding a generic ﬁeld into a series of modes. Just to quote one example of application, it may be useful in calculating the ﬁelds excited by a source, either in the cavity, or localized on one of its walls. For more details, the reader should see some of the problems. 9.5

Lossless cylindrical cavities

Let us start, like in Section 7.2, from a cylindrical structure. We call z its axis, and a ˆz its unit vector. Inside the structure there is a homogeneous lossless medium; the lateral surface, C, is a perfect electric conductor. Let us then cut a ﬁnite length of this structure, and close it by means of two perfectly conducting planes, orthogonal to the z axis, at z = 0 and z = d, respectively (Figure 9.2). We saw in Section 7.2, and in the following ones, that each transmission mode of the inﬁnitely long cylindrical structure can be obtained from a scalar pre-potential (either L or T , depending on whether the mode is TE or TM). It is of the form F(u1 , u2 )·exp(−γz). But now, we introduced two discontinuities, orthogonal to the z axis. They cause reﬂections, which couple forward and backward propagating waves. A reasonable ansatz is then to assume that the pre-potentials are of the following type: F = F(u1 , u2 ) f (z)

© 2006 by Taylor & Francis Group, LLC

(F = L, T ; f = , t)

.

(9.9)

LOSSLESS CYLINDRICAL CAVITIES

319

Figure 9.2 A cylindrical cavity obtained from a waveguide of unspeciﬁed geometry.

Here, L(u1 , u2 ) and T (u1 , u2 ), which depend on the transverse coordinates, must satisfy the boundary conditions on C, Eqs. (7.17) and (7.18). Consequently, they remain the same as for TE and TM modes of the inﬁnitely long cylindrical structure. The only quantities which are still unknown are (z) and t(z). The conducting planes at z = 0 and z = d impose the following conditions, to both TE and TM modes: Et = 0

for z = 0 , z = d

.

(9.10)

Incidentally, notice that Eq. (9.10) is not the same as Ht = 0: the total ﬁeld in the cavity is the sum of forward and backward waves, whose wave impedances have opposite signs. The reader should compare this with what we saw in Chapter 8, and also in Chapter 4 on standing plane waves. Let us take now Eqs. (3.47) and (3.48), which apply to any TE or TM ﬁeld in a homogeneous medium, and introduce Eq. (9.9) into them. We get the following expressions for the ﬁeld transverse components: ˆz ) (z) ETEt = (∇t L × a

,

ETMt =

j dt(z) ∇t T ωc dz

.

(9.11)

for TE modes ;

(9.12)

for TM modes .

(9.13)

This shows that an alternative way to express Eq. (9.10) is: (0) = (d) = 0 dt dt = =0 dz z=0 dz z=d

The TE/TM terminology refers, as usual, to the z axis. But it is worth noticing that, within a given resonator, modes can be classiﬁed in diﬀerent ways, depending on which direction the term “transverse” refers to. In particular, this becomes evident in those cavities where the cross-section remains constant along two or three axes. We know from Section 7.2 that the functions F satisfy the modiﬁed Helmholtz equation (7.8), and from Chapter 3 that the functions F satisfy the complete

© 2006 by Taylor & Francis Group, LLC

320

RESONANT CAVITIES

Helmholtz equation. Consequently, the functions f (z) must satisfy: d2 f = γ2f (γ 2 = χ2 − ω 2 μ) , (9.14) dz 2 where χ2 is real and positive (see Section 7.4). Solutions to Eq. (9.14) are compatible with Eqs. (9.12) and (9.13) only for γ 2 < 0. In this case, letting γ 2 = −k 2 , the solutions we are looking for are: π (p = 1, 2, . . .) , (9.15) (z) = A sin kz , k=p d π t(z) = B cos kz , k=p (p = 1, 2, . . .) , (9.16) d where A, B are arbitrary constants.∗ From the pre-potentials L and T that have been determined in this way, we may now calculate the ﬁelds of the cavity resonance modes, using the generalpurpose formulas in Eqs. (3.47) and (3.48). But the reader can prove, as an exercise, that they can be found in a faster way by changing slightly Eqs. (7.8) and (7.9). These changes consist simply of replacing the exponential functions in Eqs. (7.1) with sine and cosine functions, Eqs. (9.15) and (9.16). Let us keep the same symbols as in Chapter 7 for the z-independent factors ¯ TEt and H ¯ TM have in the waveguide modes; in particular, let E¯TE , E¯TMt , H the same meanings as in Eqs. (7.11) and (7.12). Then, ﬁnally we derive the following expressions, where the subscript t stands, as usual, for transverse components with respect to the z axis: TEm,n,p mode : Ez ≡ 0 ,

Hz = Hz sin p

Et = E¯TE sin p

πz d

,

πz d

,

¯ TEt cos p Ht = H

πz d

.

(9.17)

TMm,n,p mode : πz , Hz ≡ 0 , d πz ¯ TM cos p πz . (9.18) , Ht = H Et = E¯TMt sin p d d Since γ 2 and χ2m,n are linked by Eq. (9.14), it must be: pπ 2 ω 2 μ = χ2m,n + . (9.19) d The results of this section deserve further comments. First of all, one should notice that, to build a resonance mode in a cavity obtained from a waveguide, Ez = Ez cos p

∗ The case p = 0, in Eq. (9.16), which entails E TMt ≡ 0, is compatible with Eq. (9.13), but it is acceptable only for χ2 = 0, i.e., if T is not constant vs. u1 and/or vs. u2 . If this is not the case, then p = 0 must be disregarded, because (like for p = 0 in Eq. (9.15)) it would correspond to a ﬁeld identical to zero.

© 2006 by Taylor & Francis Group, LLC

SIMPLE EXAMPLES

321

a waveguide mode above its cut-oﬀ frequency is required. Indeed, no nontrivial solution to Eq. (9.14) is compatible with Eqs. (9.12) and (9.13) if we suppose γ 2 < 0 in Eq. (9.14): a resonant mode can never be a superposition of two evanescent waves. As anticipated in Section 9.1, it is a superposition of two traveling waves, one forward, the other one backward, and it is straightforward to detect that indeed there are two expressions of the type exp(−jkz) ∓ exp(jkz) in Eqs. (9.15) and (9.16). Resonance occurs when the round-trip length, for the traveling wave that undergoes reﬂections at both ends of the cavity, is an integer multiple of the wavelength. Notice that, as one can easily infer from Chapter 8, this result depends on the phase shift at the reﬂections, at both ends, being equal to π radians. Should the cavity be terminated by “mirrors” with a diﬀerent phase shift, this should be accounted for. However, if we recall from Chapter 7 that, for the (m, n)-th waveguide mode, βm,n = 2π/λg , where λg is the guided wavelength, then the condition k = pπ/d can be rewritten as: 2d = pλg

,

(9.20)

where p is an integer. This reconﬁrms the previous statement. Any solution like Eq. (9.17) or (9.18) obeys Eq. (9.19), where χ2m,n is determined by the shape and the size of the waveguide. This reconﬁrms that, for a given cavity (i.e., once the transverse eigenvalues are given, and the length d is also given), Eq. (9.9) is satisﬁed only at discrete frequencies, f m,n,p . If f = fm,n,p , the ﬁeld is identically zero. This result applies only to a strictly lossless cavity, perfectly isolated from the rest of the universe. What happens in a real cavity has been described in Section 9.3. 9.6 9.6.1

Simple examples Rectangular resonators

Given a rectangular box, the z axis (with the meaning given to it in the previous sections) can be taken parallel to any edge. Diﬀerent choices for the z axis give rise to diﬀerent nomenclatures for the resonator modes, but such changes do not correspond to any physical diﬀerence. ¯ t and Hz , to be inserted into Eqs. (9.17), TE modes. The expressions for E¯t , H come from Eq. (7.50), through Eqs. (7.4) and (7.5). For the TEm,n,p mode, we get: Ez

≡ 0 ,

Hz

= H cos

Ex

=

nπy pπz mπx cos sin a b d

,

mπx nπy pπz jωμ nπ cos sin sin H 2 kc b a b d

© 2006 by Taylor & Francis Group, LLC

,

322

RESONANT CAVITIES

Figure 9.3 Rectangular resonator.

Ey

= −

mπx nπy pπz jωμ mπ sin cos sin H 2 kc a a b d

,

Hx

= −

mπx nπy pπz H pπ mπ sin cos cos 2 kc d a a b d

,

Hy

= −

mπx nπy pπz H pπ nπ cos sin cos 2 kc d b a b d

,

(9.21)

where a, b and d are the lengths of the three sides of the cavity (Figure 9.3); H is an arbitrary constant; kc2 = (mπ/a)2 + (nπ/b)2 is a symbol taken from Chapter 7. ¯ t and Ez , to be inserted into Eqs. (9.18), TM modes. The expressions for E¯t , H come from Eq. (7.54), through Eqs. (7.4) and (7.5). For the TMm,n,p mode, we get: nπy pπz mπx sin cos , Ez = E sin a b d Hz

≡ 0 ,

Ex

= −

mπx nπy pπz E pπ mπ cos sin sin kc2 d a a b d

,

Ey

= −

mπx nπy pπz E pπ nπ sin cos sin kc2 d b a b d

,

Hx

=

Hy

= −

mπx nπy pπz jω nπ sin cos cos E 2 kc b a b d mπx nπy pπz jω mπ cos sin cos E 2 kc a a b d

, ,

(9.22)

where the notation is the same as above; E is an arbitrary constant. The reader can ﬁnd, as an exercise, which, among the TE and TM modes, can have one or two integer subscripts equal to zero. No mode can have all three subscripts equal to zero.

© 2006 by Taylor & Francis Group, LLC

SIMPLE EXAMPLES

323

Figure 9.4 Cylindrical resonator of circular cross-section.

9.6.2

Cavities of circular cross-section

For a cavity shaped like a circular cylinder, of radius a and length d (Figure 9.4), we use of course cylindrical coordinates r, ϕ, z, the z axis being the cylinder axis. In such a case there is no ambiguity in the mode nomenclature; but this has no physical meaning. Indeed, one might think that the resonator shown in Figure 9.4 is a piece of radial transmission line, short-circuited at r = a, or a structure where waves go in loops in the azimuth direction, closing their path at Δϕ = 2π. ¯ t and Hz , to be inserted into Eqs. (9.17), TE modes. The expressions for E¯t , H come from Eq. (7.66), through Eqs. (7.4) and (7.5). For the TEn,m,p mode, we get: Ez Hz Er Eϕ Hr

Hϕ

≡ 0 ,

r cos nϕ pπz , = H Jn pn,m sin a sin nϕ d sin nϕ pπz r jωμ a2 Jn pn,m , n sin = H 2 − cos nϕ pn,m r a d r cos nϕ a pπz , = −jωμ H J p sin pn,m n n,m a sin nϕ d r cos nϕ H pπ pπz p J , = a cos n n,m p2 d a sin nϕ d n,m sin nϕ pπz r H a2 pπ Jn pn,m n cos = − 2 − cos nϕ pn,m r d a d

, (9.23)

where Jn is the ﬁrst derivative of the Bessel function of the ﬁrst kind, Jn (u), with respect to its argument u (see Appendix D); H is an arbitrary constant; pn,m , the m-th zero of Jn (u), is linked by Eq. (7.65) to the eigenvalue of the TEn,m mode in the circular waveguide the cavity was derived from. The functions in brackets stand for the two possible polarizations of each mode

© 2006 by Taylor & Francis Group, LLC

324

RESONANT CAVITIES

with n = 0. Upper or lower rows must be taken together, to get the correct ﬁeld components.

¯ t and Ez , to be inserted into Eqs. (9.18), TM modes. The expressions for E¯t , H come from Eq. (7.69), through Eqs. (7.4) and (7.5). For the TEn,m,p mode, we get: r cos nϕ pπz , Ez = E Jn pn,m cos a d sin nϕ Hz

≡ 0 ,

Er

=

Eϕ

=

Hr

=

Hϕ

=

r cos nϕ pπ pπz Jn pn,m , − a sin pn,m d a d sin nϕ sin nϕ r pπz E a2 pπ Jn pn,m n , sin − cos nϕ p2n,m rd a d sin nϕ r pπz jωE a2 Jn pn,m , n cos − 2 − cos nϕ pn,m r a d r cos nϕ jωE pπz , − a Jn pn,m cos pn,m a d sin nϕ E

(9.24)

where the symbols are the same as for Eqs. (9.23), except for the arbitrary constant E and for pn,m (m-th zero of Jn (u)), linked to the eigenvalue of the TMn,m mode of the circular waveguide by Eq. (7.68). Some examples of ﬁeld lines, for a few, simple modes of rectangular and circular resonators, are shown in Figure 9.5.

9.6.3

Spherical cavities

In Chapter 11 we will solve the Helmholtz equation using separation of variables in spherical coordinates (r, ϑ, ϕ). The main reason why this subject is postponed, is because its main applications deal with ﬁeld sources (like antennas). After studying Chapter 11, the reader should come back to this section, and impose on each spherical harmonic the boundary condition Etan = 0 on a spherical surface of radius r = a. In this way, the modes of a spherical resonator will be found. Another reasonably simple exercise is to impose the same condition also on coordinate planes ϕ = constant, or on coordinate cones ϑ = constant. What one obtains in this way are the modes of more complicated structures, which are of some interest in problems related to feeders for aperture antennas (see Chapter 12 and 13), or to tapers between waveguides of diﬀerent cross-section.

© 2006 by Taylor & Francis Group, LLC

LOSSY RESONATORS

325

Figure 9.5 Field lines of a set of signiﬁcant modes, in rectangular and circular resonators.

9.7

Lossy resonators: Perturbation analysis. Intrinsic Q-factor

From the idealized model of perfectly conducting walls, to a more realistic one where the walls are good conductors (σ ω, but ﬁnite), the step is very similar to that of Chapter 7 about waveguides. On the grounds of what we saw in Section 4.12, the surface which surrounds the resonator can be characterized in terms of its wall impedance: πμf [Ω] , (9.25) Z0 = (1 + j) σ

© 2006 by Taylor & Francis Group, LLC

326

RESONANT CAVITIES

which was originally derived for a ﬂat surface. As we saw in Chapter 7, one can prove that this result is correct also for cylindrical or spherical walls. On the other hand, the wall impedance becomes much more complicated (involving metric coeﬃcients of the coordinate system) if the curvature is not constant along the wall itself. On these premises, the ﬁrst order perturbation theory is very similar to that of Chapter 7. We start oﬀ supposing that a mode of the ideal cavity can exist in the cavity with ﬁnite wall impedance, although it cannot be rigorously so. We calculate the ohmic losses under this assumption. As in Chapter 7, special caution is needed in the case of degenerate modes: among all the linear combinations of two or more degenerate modes, which are still modes of the ideal cavity, one must identify those that remain orthogonal in the presence of losses. From lumped-element circuit theory we know that power dissipation in a resonant circuit, WJ , is of little signiﬁcance, unless correlated with the amount of energy which is stored in the circuit, E. The link between these two quantities is, also for a resonant cavity, the adimensional number: Q0 =

ω0 E WJ

,

(9.26)

which is referred to as the quality factor, or factor of merit, or ﬁgure of merit (or, for short, the Q-factor) of the resonator mode characterized by the resonance angular frequency ω0 . Let us recall (from Chapter 3) that, for a resonant ﬁeld, the time-averaged energies stored in electric and in magnetic form are equal and the total energy, constant in time, is equal to twice those averages. If one continues along the same examples as in Section 9.6, one sees that, as a general trend, the Q-factor in a given cavity tends to increase with the order of the resonance mode. This can be intuitively justiﬁed as follows. Consider ﬁrst a cavity with lossless walls. In it, the electric ﬁeld components of a given mode have nodes, i.e., surfaces on which the tangential components of E are identically zero. Some of these nodes are located on the resonator walls, but some may be inside. The number of internal nodes grows as the mode order grows. All the nodal surfaces can be metallized with perfect electrical conductors, without aﬀecting the ﬁeld; this operation is equivalent to subdividing the resonator into several cavities, each of which resonates in a lower order mode, compared to that of the original cavity. When we calculate the Q-factor, only the real (external) walls, being non-ideal, contribute to loss. So, the power dissipated in the “large” resonator is less than the sum of the powers that would be dissipated in all the “small” resonators if they would be really separated by the same imperfect conductor the real walls are made of. On the other hand, the total stored energy obviously equals the sum of the stored energies. Consequently, the Q-factor is larger in the large resonator, which is equivalent to saying that (at least within a given “family” of modes) it grows as the order of the mode grows. From circuit theory we know that in a lumped-element resonator the Q-

© 2006 by Taylor & Francis Group, LLC

RESONATORS COUPLED TO EXTERNAL LOADS

327

factor determines the full width at half maximum of the so-called resonance line, i.e., of the Lorentzian function

f (ω) = 1 + jQ

1 ω0 ω − ω0 ω

.

(9.27)

In other words, the Q-factor of a resonant cavity mode is a measure how diﬃcult it is to inject into the cavity a signal whose frequency is diﬀerent from the resonant one. 9.8

Resonators coupled to external loads. Loaded Q-factor

A completely isolated resonant cavity does not bear any practical interest. Any use of a resonator requires coupling it to other structures, that allow to inject energy into it. Some typical coupling tools at microwave frequencies are: • connections either with waveguides or with other cavities through holes called irises or diaphragms; • connections with transmission lines, or with lumped-element circuits, by means of conductive wires or probes, either straight or loop-shaped; • connections with other elements (transmission lines or other cavities) by means of electron beams which pass through the resonator. To evaluate the external coupling of a lossless resonator, suppose that the cavity is at steady state: it is fed by a source which compensates exactly for the power that the cavity loses to the outside world. We deﬁne as the external quality factor of the resonator (or better, of its mode characterized by the resonance frequency ω0 ) the adimensional quantity: Qex =

ω0 E Wex

,

(9.28)

where E, as in the previous section, is the energy stored in the resonator at steady state, while Wex is the active power leaving the resonator to the outside world. If we consider a passive resonator, as we have always done so far, then the quantity in Eq. (9.28), if inserted into Eq. (9.27), determines, as in the previous section, the spectral width of the Lorentzian function which characterizes the mode under test. It is useful to add that for a resonator containing sources, the Q-factor deﬁned by Eq. (9.28) determines, again through Eq. (9.27), a linewidth which is not that which one measures when the source is running in the large-signal regime. To stress this point, the quantity in Eq. (9.28) is often referred to as the cold cavity Q-factor. In a practical resonator, losses and coupling with the outside world must be accounted for at the same time. In all cases of interest, Q0 and Qex are both much greater than one, meaning that the energy stored in the cavity is much larger than the energy lost in the resonator over one period T = 2π/ω0 ,

© 2006 by Taylor & Francis Group, LLC

328

RESONANT CAVITIES

as well as than the energy delivered to the external load during the same time interval. One consequence is that internal losses and energy given to the external load may be thought of as additive, although in general this is not rigorous. Consequently, it makes sense to deﬁne a loaded quality factor as: QL =

ω0 E WJ + Wex

.

(9.29)

Its link with the other quantities deﬁned in this section and in the previous one is: 1 1 1 = + , (9.30) QL Q0 Qex the loaded Q-factor Eq. (9.39) determines, as usual through Eq. (9.27), the spectral width of the frequency response of a cavity. 9.9

Open resonators

The examples shown in Section 9.6 reconﬁrm that resonance in a cavity means satisfying a relationship between the geometrical dimensions of the cavity itself and the wavelength at the resonant frequency. When the geometrical shape of the cavity is simple, this relation is simple also. If the shape of the cavity is complicated, this relation becomes complicated but the concept remains the same. At optical frequencies (of the order of 1014 –1015 Hz), to make geometrical dimensions comparable with wavelengths becomes a technological challenge. It is necessary to think of resonators which operate in high-order modes, in contrast to what is normally the case for microwave cavities. If we take a cavity of a given shape and given size, and increase the frequency, then the growth in the resonance mode order is accompanied by an increase in the relative density of resonances, on the frequency axis (deﬁned simply as the number of modes, n, whose resonance frequencies fall within a unit length on the f axis, Δf = 1 Hz, divided by the frequency at which this interval is centered, f0 ). Note that the relative density is much more relevant than the absolute one, n, because n/f0 is the quantity to compare with the cavity Q-factors in order to establish whether in practice it is possible to operate a cavity in a single mode. Single-mode operation requires at the same time a low relative density of modes and a high Q. In a cavity with metallic walls, of the same type as those we dealt with so far, it is impossible to meet these requirements at optical frequencies. This is because at these frequencies there is no material which satisﬁes σ ω. To by-pass the problem, one may proceed as follows: • replace the metallic walls with dielectric mirrors, which are dielectric multilayers (see Section 4.10) designed to yield high reﬂection coeﬃcient at the wavelength of interest; • decrease the relative density of modes by eliminating some of the walls, making an open resonator, i.e., a 3-D region conﬁned between two highly

© 2006 by Taylor & Francis Group, LLC

OPEN RESONATORS

329

Figure 9.6 Schematic examples of open resonators.

Figure 9.7 A Gaussian beam in an open resonator.

reﬂecting surfaces along one direction, z, but extending to inﬁnity in the planes orthogonal to z. Some examples of open resonators, with plane mirrors (Fabry-P´erot cavities, see Section 4.10) and with spherical mirrors, are sketched in Figure 9.6, parts a) and b), respectively. Some rays are also sketched in the same ﬁgure, to illustrate how one can justify in intuitive terms high-Q-factor modes in an open resonator. Indeed, for suitable geometries one ﬁnds rays that are trapped in the cavity and undergo an inﬁnite number of reﬂections on the two mirrors in spite of the fact that these have ﬁnite transverse dimensions. As we shall see in the next section, some quantitative information may be grasped from the intuitive ray optics model. However, a much larger amount of information can be extracted from Gaussian beams. In fact, we saw in Chapter 5 that, far from the beam waist (|z| πw02 /λ, see Figure 9.7), the wavefronts of a Gaussian beam are essentially spherical, and their radius of curvature, as given by Eq. (5.96), is: 2 2 πw0 z . (9.31) R(z) = z 1 + λz If a beam impinges on a spherical mirror whose radius of curvature equals that of the wavefront, reﬂection gives rise to another Gaussian beam whose parameters are identical to those of the incident beam, and travels in the opposite direction. Intuition indicates that suitable combinations of two mirrors can make the beam go up and down indeﬁnitely in a stable way. Predictions based on this model can be made quantitative, for mirrors having any radius of

© 2006 by Taylor & Francis Group, LLC

330

RESONANT CAVITIES

curvature, by means of the “ABCD law” of Section 5.11. All the main parameters of a mode of an open resonator can be calculated in terms of Gaussian beams: resonance frequency, total stored energy and its spatial distribution, and also, with a few further notions that will be given in Section 9.11, the Q-factor. In particular, the resonance frequency can be calculated ensuring that the total phase shift along a closed path, taking into account phase shifts due to reﬂections on the two mirrors, has to be an integer multiple of 2π. Making use of the results of Section 5.7, this requirement may be written as: arg ρ1 + arg ρ2 + 2kL + 2[χ(z2 ) + ξ(z2 ) − χ(z1 ) − ξ(z1 )] = 2nπ

,

(9.32)

where z1 and z2 are the coordinates of the centers of the two mirrors; L = (z1 − z2 ) is the cavity length; n is an integer; χ and ξ have the same meanings as in Sections 5.9 and 5.10; ρ1 and ρ2 are the reﬂection coeﬃcients of the two mirrors, whose arguments are normally independent of the mode one deals with. The fourth term on the left-hand side of Eq. (9.32) is usually much smaller than the third, so it is neglected in ﬁrst approximations; however, it is very important in principle, as it contains the term ξ(z) which, as we saw in Section 5.10, depends on the transverse order of the Gaussian beam, labeled in Section 5.10 as (m1 , m2 ) for the Hermite-Gauss modes and as (, p) for the Laguerre-Gauss modes. It is just because of this term that the resonance frequency of the TEMm1 ,m2 ,n or TEM,p,n depends slightly on the transverse order, for a given longitudinal order. This aﬀects signiﬁcantly the emission spectrum of laser sources, whenever they may oscillate on more than one “transverse mode”. Let us stress that such a result would be impossible to obtain using ray optics. 9.10

Stability of open resonators

In the previous section we ﬁrst gave the motivations for using open resonators at optical frequencies, and then we saw how they can be studied in terms of Gaussian beams. These beams, as we saw in Chapter 5, have wavefronts that in principle extend to inﬁnity in all directions orthogonal to the z axis. On the contrary, mirrors of open resonators are obviously of ﬁnite size. This contrast entails two diﬀerent types of consequences, that will be the subjects of this section and of the next one. Normally, mirrors are far away from focal planes. In other words, they are located in regions where evanescent waves do not contribute in a signiﬁcant way to the Gaussian beam. Therefore, as a ﬁrst approximation we may think in terms of ray optics, in order to ﬁnd under what conditions a ray remains trapped in the cavity when it undergoes multiple reﬂections on the two mirrors. Let us consider then what is sketched in Figure 9.8. The cavity consists of two spherical mirrors, with focal lengths R1 /2 and R2 /2, located at a distance L from each other. Let us restrict ourselves to meridional rays, i.e., rays belonging to planes passing through the cavity axis z, like those shown

© 2006 by Taylor & Francis Group, LLC

STABILITY OF OPEN RESONATORS

331

Figure 9.8 Deﬁnitions of the geometrical parameters of an open resonator.

Figure 9.9 Ray reﬂection at a mirror at one end of an open resonator.

in Figure 9.8 which belong to the (x, z) plane. Just to ensure that there is no ambiguity, suppose that the n-th reﬂection of a given ray takes place on mirror 2, the (n + 1)-th on mirror 1, and so on. Let xn be the value of the transverse coordinate x at the site of the n-th reﬂection, and xi = (dx/dz)i (i = . . . , n − 1, n, n + 1, . . .) be the slope of the ray after the i-th reﬂection, with respect to the positive direction of the z axis.† The law of reﬂection, if applied on mirror 1 (Figure 9.9), gives the following equation (where the angles are small enough to be approximated by their tangents): xn+1 = −xn −

xn+1 f1

.

(9.33)

Furthermore, as the slope of a ray remains constant between two consecutive reﬂections, it is straightforward to write that: xn+2 − xn+1 = L xn+1

,

xn+1 − xn = −L xn

.

(9.34)

Replacing Eq. (9.33) into the ﬁrst of Eqs. (9.34) and eliminating xn by means of the second of Eqs. (9.34), rearranging the terms, we obtain the † Note that the signs of x depend on the sense in which the ray propagates: x n+1 > 0 i means that, traveling from mirror 1 towards mirror 2, x is increasing. On the contrary, xn > 0 means that, traveling from mirror 2 towards mirror 1, x is decreasing.

© 2006 by Taylor & Francis Group, LLC

332

RESONANT CAVITIES

following equation, without derivatives:

L xn+2 + xn = xn+1 2 − f1

.

(9.35)

It is self-explanatory why the same equation holds for the previous trips, those in one direction: L xn + xn−2 = xn−1 2 − , (9.36) f1 as well as those in the other direction:

L xn+1 + xn−1 = xn 2 − f2

.

(9.37)

Adding Eqs. (9.35) and (9.36) term by term, and eliminating from the righthand side the term xn+1 +xn−1 by means of Eq. (9.37), we obtain the following ﬁnite diﬀerence equation: xn+2 − 2b xn + xn−2 = 0 , where we set: b=1−

L L L2 L L − + =2 1− 1− −1 . f1 f2 2f1 f2 2f1 2f2

(9.38)

(9.39)

The ﬁrst thing of interest is to know whether Eq. (9.38) has stable solutions, that is whether there are ray trajectories for which the value of the transverse coordinate x does not diverge as the number of reﬂections increases indeﬁnitely. For this purpose we let: xn = x0 ejnϑ

,

(9.40)

where the unknown ϑ has to be thought of, for the sake of generality, as being complex. Substituting Eq. (9.40) into Eq. (9.38), we ﬁnd: ej2ϑ + e−j2ϑ = 2b ,

(9.41)

2ϑ = arccos b .

(9.42)

whose solution is: Hence, ϑ is real if |b| < 1. In this case, Eq. (9.40) is a periodic solution of Eq. (9.38). As long as a mode of the cavity can be reasonably approximated by plane waves obeying the ray-optics laws, the previous result indicates that these waves remain conﬁned inside the cavity for |b| < 1. For |b| > 1, on the contrary, ϑ becomes imaginary, and so the transverse coordinate Eq. (9.40) diverges exponentially as the number of round trips through the cavity increases. This indicates that, after a ﬁnite number of reﬂections, the generic ray will miss the mirrors and leave the cavity. The ﬁrst case is referred to as a stable resonator, the second one as an unstable resonator. It is useful to plot the regions where |b| > 1 or |b| < 1 on the cartesian

© 2006 by Taylor & Francis Group, LLC

Q-FACTOR OF AN OPEN RESONATOR

333

Figure 9.10 The stability diagram for open resonators.

plane {L/2f1 , L/2f2 }. In Figure 9.10, the blank regions correspond to geometries that yield stable resonators, while the dashed regions correspond to unstable resonators. Negative values of one of the coordinates correspond to a convex mirror; so, points whose coordinates are both negative (a pair of convex mirrors) are all in the region of instability, in agreement with intuition. They do not ﬁnd any practical application. On the other hand, cavities which are moderately unstable (|b| slightly larger than 1) are used in laser technology, when high-power output beams are required. Indeed, for a given beam power, the energy stored in an unstable cavity is signiﬁcantly smaller than in a stable one, and this can simplify problems related to materials. The price one pays when the stability is lowered (that is, when |b| is increased) is a reduced mode selectivity, that is, a lower quality (both vs. time and vs. spatial coordinates) of the output beam. 9.11

Q-factor of an open resonator

Also for open cavities it is of great interest to deﬁne an adimensional quantity, called merit or quality factor or, for short the Q-factor. It can still be deﬁned by multiplying the resonance frequency by the energy stored in the resonator, and dividing it by the time-averaged power loss. There are no diﬀerences with respect to what we saw in other sections from the viewpoint of principles, but now it is more diﬃcult to separate power lost in the resonator from that delivered outside (see Sections 9.7 and 9.8), a distinction which allowed us to deﬁne an unloaded Q-factor and a loaded Q-factor for closed resonators. In an open resonator, both stable and unstable, mirrors have ﬁnite diameters, while, as we said before, wavefronts of Gaussian beam extend indeﬁnitely. Therefore, every reﬂection is accompanied by loss of that amount of

© 2006 by Taylor & Francis Group, LLC

334

RESONANT CAVITIES

power which corresponds to the part of the beam which travels at distances from the cavity axis larger than the mirror size. This power goes obviously out of the cavity, but must be treated conceptually as lost power, because it is not delivered outside in any usable way. Consequently, it is included (together with absorption either within the cavity or on the mirrors) in the unloaded Q-factor. The case of one (or two) partially transparent mirror is completely diﬀerent. If a mirror reﬂects only a fraction of the incident power, the complementary part can be delivered to an external beam. This power passing through the partially transparent mirror has to be accounted for in the deﬁnition of an external Q-factor. When the losses due to ﬁnite size of the mirrors and those due to absorption are small with respect to the energy stored in the resonator, the two contributions to the unloaded factor Q0 can be added, following the rule used in Section 9.8. The low-loss assumption ﬁts always well stable cavities, provided the mirrors are made with low-loss materials. Then, we may write: 1 1 1 = + Q0 Qa Qd

,

(9.43)

where Qa = ω0 E/Wa is the Q-factor due to absorption (Wa = time-averaged absorbed power), while Qd is the contribution due to losses at mirrors edges. To ﬁnd a way to express these losses, one needs diﬀraction theory, to be dealt with in Chapter 13. Here, we will just quote the formulas we need now, without any proof. All the quantities that we need can be expressed as functions of an adimensional parameter, usually referred to as the Fresnel number (Born and Wolf, 1980) which, for two spherical mirrors, can be deﬁned as: a1 a2 , (9.44) N= λL where λ is the wavelength, L is the cavity length, a1 and a2 are the radii of the cross-sections of the mirrors (see Figure 9.8). The fraction of the power of a Gaussian beam, τ , which is lost in one reﬂection on the mirror, as a function of N , is well approximated by simple expressions for N 1 (diﬀraction losses small compared to the other losses of interest) and for N 1 (that means diﬀraction losses dominant on the others). Respectively, we have: τ 16π 2 N e−4πN and:

τ 1 − π2N 2

(N 1) (N 1) .

,

(9.45) (9.46)

A simple expression for Qd as a function of τ (Ramo et al., 1994) can be easily linked to intuitive properties of Gaussian beams that we saw in Chapter 5. If the radii of curvature of the mirrors are large enough, then the energy distribution in the cavity, on a transverse plane, is almost independent of the longitudinal coordinate, and the ratio between average and maximum values of the energy density depends only on the order of the mode. The

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

335

Poynting vector is equal to the product of the energy density times the group velocity, that we can assume equal to the speed of light in the medium which ﬁlls the cavity. Therefore, the energy stored in the cavity: E = πw02 L U

,

(9.47)

(w0 = waist width of the Gaussian beam) and the power incident on the mirror: (9.48) Wi = πw02 c U , can be easily expressed as functions of the average energy density, U . Taking now into account the fact that power loss at any reﬂection, Wd , is the product of τ times Eq. (9.48), we ﬁnally get: Qd =

2πL ω0 E ω0 L = = Wd τc τ λ0

,

(9.49)

where λ0 = 2πc/ω0 is the resonance wavelength. So, the quality factor due to diﬀraction losses is, if we keep all other conditions constant, proportional to the number of wavelengths L/λ0 contained in the cavity in the axial direction, i.e., to the integer number n of Eq. (9.32) (“longitudinal order” of the resonance mode). This agrees with the statements of Section 9.7, about the Q-factor growing when the order of the mode grows, and justiﬁes why very high values can be obtained without too much diﬃculty for Q-factors at optical frequencies. To ﬁnish this section, it may be useful to recall again that all the deﬁnitions and calculations in the sections on open resonators (Section 9.9 and following) refer to cavities ﬁlled with a passive medium. In the language of laser technology, these are called cold cavity modes. When cavities are ﬁlled with an active medium, capable of amplifying the resonant wave (the necessary energy comes from other sources), then there are important changes both in the Q-factor, with respect to that deﬁned in this section, and in the oscillation frequencies, which are “pulled” by the active medium to diﬀerent values with respect to those given by Eq. (9.32). These problems are largely discussed in books on lasers (Svelto, 1989; Yariv, 1976). 9.12

Further applications and suggested reading

We take it for granted that the background in basic physics of all the readers includes resonances in other types of passive systems, typically mechanical ones. The relationship between them and what we have seen in this chapter is not just a vague analogy. The links are very fundamental. Just to quote one of them, a mechanical system can be resonant only if it can store energy under two diﬀerent forms at diﬀerent times in its periodic motion (e.g., kinetic and potential in a pendulum, kinetic and elastic in a spring). Resonance occurs when these two energies balance each other, exactly like electric and magnetic energies in resonant cavities. It may be of some interest, at least for those readers who have an incli-

© 2006 by Taylor & Francis Group, LLC

336

RESONANT CAVITIES

nation for the History of Science, to know that the mathematical treatment of resonant cavities, which is rather sophisticated compared to that of simple mechanical systems, has been developed long ago, up to a very advanced level, in the framework of acoustics. The very famous treatise Theory of Sound by Lord Rayleigh, now more than one century old, remains a masterpiece of scientiﬁc literature as well as a very important marker in the advancement of science. It is still a highly recommended, very instructive reading. After these suggestions related to classical physics, we also invite the readers to strengthen their background on resonant systems in Quantum Mechanics, typically, harmonic oscillators (Louisell, 1960). An important analogy in this direction is that between discrete states in a quantized system, and discrete modes in a resonator. The two systems are eigenvalue problems, and it is immediately noted that energy and frequency are proportional, through the Planck constant. Concepts like mode degeneracy and perturbation theory (like the one we used for calculating Q-factors), both in this chapter and in the previous one, may become much more familiar through a cross-correlation with elementary Quantum Mechanics. Going back to e.m. resonators, it is rather surprising that microwave resonators and optical resonators were studied in completely separate textbooks for decades, in spite of the fact that both are easy to explain, at least qualitatively, in terms of transmission lines (Chapter 8). It is even more surprising since one of the main applications of resonators, ﬁltering, was shared by the two frequency ranges. Probably, this was because most of the experts in microwaves, being electrical engineers, had a strong background in circuit theory, and so they designed their devices using mathematical tools originating from this area, typically those based on topological matrices and on Laplace transforms. On the other side, most of the physicists who worked in optics preferred to use frequency-domain approaches. To see this contrast, one may consult, for example, the sections devoted to ﬁltering properties of Fabry-P´erot ´etalons in Born and Wolf (1980), on one side, and Chapters 4 and 5 of Montgomery, Dicke and Purcell, Eds., (1948) on the other side. Nowadays, thanks to the success of optics in telecommunications, it is much easier to ﬁnd the two subjects side by side. However, this comment was not inserted simply as a historical detour. It was intended to remind that, if one has to design resonators for ﬁltering purposes, then it may be advisable, irrespectively of the frequency band, to know at least something in the area of network theory. A sequence of readings in this sense may consist of Desoer and Kuh (1969), Collin (1966) and Howe (1974), to fulﬁl the basic needs and possibly trigger further interest. Expert readers may include dielectric resonators, a subject that was omitted here only because of space limits. One may consult for example Kajfez and Guillon, Eds., (1986). Other readers may wish to be introduced to computer-aided design of ﬁltering devices. At microwave frequencies, one may then consult Gupta et al. (1981). Another practical application of resonant cavities, comparable in importance to ﬁltering, is positive feedback, as required in order to build an oscil-

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

337

lator. This extends to microwave and optical frequencies the role played in conventional electronics by lumped-element LC circuits. However, it has been stressed many times in the previous sections that an oscillating device can no longer be modeled as a purely electromagnetic system. Indeed, the ﬁeld in it is coupled to something (e.g., an active medium in lasers, or a chargedparticle beam in microwave tubes) which provides gain to the ﬁeld. One must be warned that this coupling aﬀects some of the most important characteristics of the ﬁeld. Ramo et al. (1994) is a recommended source of introductory information on this point, for the microwave case. Textbooks on lasers (e.g., Svelto, 1989) should be consulted for the optical-frequency case. In this framework, it may be interesting to look again at the relationship between microwave and optical resonators. Section 10.15 of Ramo et al. (1994) outlines why it took the work of several outstanding scientists, to grasp the advantages of open resonators, compared to closed ones, when the ﬁrst lasers were built. Another important point is that the type and the quality of the resonator aﬀect both temporal and spatial coherence of the sources. As the subject of coherence will be introduced in Chapter 14, readers are advised to come back to the present chapter after reading that one, and link the two together. Problems 9-1 Show that for any resonant mode of a closed cavity with perfectly conducting walls, all the ﬁeld components can be obtained following the same approach as for the propagation modes in waveguides, provided one replaces the exponential functions in Eqs. (7.1) with either sine or cosine functions, following the indications of Eqs. (9.15) and (9.16). 9-2 Discuss which among the TE and the TM modes of a rectangular resonator can have one or two subscripts equal to zero. Repeat the problem for the modes of a resonator with circular cross-section. Establish a link between these two answers, describing intuitively how the lines of force of the electric and magnetic ﬁelds of a given mode in a rectangular resonator deform into those of the corresponding mode in a circular resonator, using a cavity of elliptical cross-section as an intermediate step. 9-3 If one or two ratios between side lengths, in a rectangular resonator, are rational numbers, then we may expect some modes to become degenerate. a) Find which modes are degenerate when a = 2b = d/2. b) Among these sets of degenerate modes, only some satisfy orthogonality conditions like Eqs. (9.7) and (9.8). Discuss which do and which do not. 9-4 The fundamental mode in a resonator is clearly deﬁned as the mode with the lowest resonance frequency. In a rectangular cavity, the ordering of the resonant frequencies depends on the inequalities between the side lengths.

© 2006 by Taylor & Francis Group, LLC

338

RESONANT CAVITIES

a) Keeping a > b, discuss which is the fundamental mode as a function of d. b) Find a design rule that maximizes the spacing between the resonance frequency of the fundamental mode and that of the next higher-order mode. 9-5 In a resonator of circular cross-section, resonant frequencies are ordered in a way which depends on the ratio between the length and the diameter. Repeat, for this case, the two questions of the previous problem. 9-6 In a spherical coordinate frame, impose the boundary condition Etan = 0 ﬁrst on the coordinate planes ϕ = constant, and then on the coordinate cones ϑ = constant. Find the corresponding solutions of Maxwell’s equations. Discuss the physical structures in which these solutions exist as resonant modes or as propagation modes. (Hint. Restrict the discussion to TE and TM waves, for which partial derivatives with respect to ϕ are identically zero). 9-7 Using the contents of Section 7.10, and with the help of a suitable handbook dealing with Mathieu’s functions (e.g., Abramowitz and Stegun, 1965), express the longitudinal ﬁeld components for the resonant modes of a cylindrical cavity with elliptical cross-section, with an ideal conducting wall. 9-8 Express the intrinsic Q-factors of the generic TE and TM mode of a cubic resonator (a = b = d) whose wall impedance is expressed by Eq. (9.25). In particular, discuss how the Q-factor of a given mode depends on the resonant frequency, which in turn depends on the cavity size. 9-9 As sketched in Section 9.1, a piece of coaxial cable whose ends are sortcircuited resonates at those frequencies where its length equals a half wavelength. a) Express the intrinsic Q-factor due to losses in the good conductor which is used to make the two cylinders and the two ﬂat terminal surfaces. b) What happens to the intrinsic Q-factor if the cavity resonates in a higherorder longitudinal mode ? (Hint. Do not forget the loss contributed by the two terminal ﬂat walls.) 9-10 Calculate the intrinsic Q-factor of a coaxial cavity, for an outer radius a = 5 mm, a characteristic impedance of the cable of 120 Ω, a dielectric permittivity inside the cable equal to 0 , copper walls (conductivity: 5.8 × 107 Ω−1 m−1 ), resonant frequency 600 MHz. 9-11 On some occasions, losses in the dielectric which ﬁlls a resonant cavity may contribute signiﬁcantly to the intrinsic Q-factor. a) Assuming that the conducting walls are ideal, express the intrinsic Q-factor due to a dielectric with a small loss tangent, δ, for the modes of the cubic resonator of Problem 9-8, and for the TEM modes of the coaxial cavity of Problem 9-9.

© 2006 by Taylor & Francis Group, LLC

FURTHER APPLICATIONS AND SUGGESTED READING

339

b) Which is the expression of the intrinsic Q-factor that accounts for both conductor and dielectric losses, assuming that they are small enough to be additive ? 9-12 Tunable resonators are often used to measure frequencies in current microwave practice. They are usually referred to as wavemeters. The higher the Q-factor, the sharper the resonance, and so, quite obviously, the more accurate the frequency measurement. This is one of the reasons why cavities of circular cross-section, resonating in a low-loss TE0,1,p mode (see Section 7.11), are usually preferred to rectangular cavities. To support this statement, calculate the intrinsic Q-factor of the fundamental mode of a cubic cavity (see the previous problem) resonating at 3 GHz, and that of a cylindrical cavity whose radius equals the side length of the previous cavity and resonates in the TE0,1,1 mode at the same frequency. Assume that the wall surfaces are covered with a galvanic silver layer (conductivity) thicker than the skin depth. 9-13 Another reason why cavities of circular cross-section are preferred as wavemeters is because there are no currents ﬂowing through the contact between the cylindrical wall and the sliding piston which makes the resonator tunable. To support this statement, ﬁnd the current densities in all the walls of a circular-cylinder cavity, in the TE0,1,1 mode. Are there other modes of the same cavity which enjoy the same property of not having currents ﬂowing through that contact ? 9-14 At microwave frequencies, one method of measuring the permittivity of a dielectric material is to observe the shift in the resonance frequency, Δω, when a small sample of the material is placed inside a given cavity. In fact, following a variational approach, Slater proved that Δω/ω0 = ΔU/U0 , where ΔU is the change in the energy stored in the cavity caused by the presence of the sample, while ω0 and U0 are the values which correspond to the cavity without the sample. a) To enhance the sensitivity of the measurement, is it convenient to place a given dielectric sample near a maximum of the electric ﬁeld or near one of the magnetic ﬁeld ? b) Assuming that the ﬁeld remains unperturbed when the small sample is introduced, evaluate Δω for a cubic sample having r = 3, 3 mm in side, placed in the center of a rectangular cavity with a = 2b = d = 30 mm, ﬁlled with a material having r = 1.5 and resonating in the TE101 mode. 9-15 In Chapter 4 we discussed a device, shown in Figure 4.10, which was referred to as the Fabry-P´erot interferometer. Actually, this name is usually given to any structure which consists of two ﬂat parallel mirrors, separated by a homogeneous medium. It is often used as a frequency (or rather, wavelength) meter at optical frequencies, by making it possible to vary continuously the distance between the two mirrors.

© 2006 by Taylor & Francis Group, LLC

340

RESONANT CAVITIES

a) To increase the resolution of the frequency measurement, the device is fabricated with dielectrics which exhibit very low loss. What determines then the spectral width of the resonance line, Eq. (9.27) ? b) Evaluate this external Q-factor of the open resonator. (Hint. Model the device as a piece of lossless transmission line, with a characteristic impedance Z1 , and a length equal to a multiple of λ/2, sandwiched between two inﬁnitely long transmission lines, with characteristic impedances Z2 .) c) How does the external Q-factor vary as a function of the diﬀerence between Z1 and Z2 ? 9-16 The mirrors which deﬁne the cavity in a typical semiconductor laser are simply the discontinuities between the high-index semiconductor chip, and air. a) Assuming a refractive index n = 3 (note that in Eq. (9.32) it is k = nk0 , where k0 is deﬁned in free space), calculate the resonant frequencies and the corresponding free-space wavelengths of the longitudinal modes (i.e., neglect the χ and ξ terms) for a chip length L = 300 μm. b) For the lowest-order answer to the previous question, estimate the frequency separation between two consecutive Hermite-Gauss modes, taking w0 = 5 μm at the beam waist. 9-17 Find one or more physical arguments to justify why (see Figure 9.10): a) an open resonator with two convex mirrors (f1 and f2 both negatives) is always unstable; b) a convex mirror and a concave one can make a stable resonator, provided their distance L is not too large. 9-18 Consider a Helium-Neon laser, emitting at λ = 0.633 μm. a) Show that, even when the mirrors are so small that diﬀraction losses are extremely high (i.e., for τ close to unity), the estimated Q-factor, based on Eq. (9.49), is quite high if the cavity length is in the range of 100 mm. b) Show that mirror radii in the millimeter range are suﬃcient to let Qd grow by many orders of magnitude.

© 2006 by Taylor & Francis Group, LLC

References Abramowitz, M. and Stegun, I.A. (Eds.) (1965) Handbook of Mathematical Functions. Dover, New York. Born, M. and Wolf, E. (1980) Principles of Optics: Electromagnetic Theory of Propagation, Interference and Diﬀraction of Light. 6th (corrected) ed., reprinted 1986, Pergamon, Oxford. Collin, R.E. (1966) Foundations for Microwave Engineering. McGraw-Hill, New York. Desoer, C.A. and Kuh, E.S. (1969) Basic Circuit Theory. McGraw-Hill, New York. Gupta, K.C., Garg, R. and Chadha, R. (1981) Computer-Aided Design of Microwave Circuits. Artech House, Dedham, MA. Howe, H. (1974) Stripline Circuit Design. Artech House, Dedham, MA. Kajfez, D. and Guillon, P. (Eds.) (1986) Dielectric Resonators. Artech House, Dedham, MA. Louisell, W.H. (1960) Coupled Mode and Parametric Electronics. Wiley, New York. Montgomery, C.G., Dicke, R.H. and Purcell, E.M. (Eds.) (1948) Principles of Microwave Circuits. McGraw-Hill, New York. Reprinted by Peter Peregrinus, London, for IEE (1987). Morse, P.M. and Feshbach, H. (1953) Methods of Theoretical Physics. McGraw-Hill, New York. Ramo, S., Whinnery, J.R. and Van Duzer, T. (1994) Fields and Waves in Communication Electronics. 3rd ed., Wiley, New York. Svelto, O. (1989) Principles of Lasers. 3rd ed., translated from Italian and edited by D.C. Hanna, Plenum, New York. Yariv, A. (1976) Introduction to Optical Electronics. Holt, Rinehart and Winston, New York.

341 © 2006 by Taylor & Francis Group, LLC

CHAPTER 10

Dielectric waveguides 10.1

Introduction

In all the waveguides dealt with in Chapter 7, the dielectric medium where the guided ﬁeld propagates was assumed to be homogeneous. In other structures, which may also be of practical interest, the medium is not homogeneous; for example, this is the case for microwave devices made with ferrite-loaded waveguides, of the same kind as those that were mentioned in Chapter 6. Still, these guides have a boundary deﬁned by a cylindrical conducting surface. Their analysis is similar, at least with reference to the fundamentals, to what we saw in Chapter 7, although some changes are not trivial. There is still another set of wave guiding structures, whose essential feature is no metallic wall to conﬁne the ﬁeld within a closed region. The dielectric medium does in principle extend to inﬁnity, in the directions orthogonal to the waveguide axis. What drives the ﬁeld energy along a given direction is a suitable set of inhomogeneities — either sharp discontinuities, or smooth variations — in the permittivity of the medium. These structures diﬀer radically from those of Chapter 7, and we must approach them in another way. The theoretical foundations of dielectric waveguides were established in the early decades of the 20th century. However, the space devoted to them in standard Electrical Engineering curricula was extremely marginal until 1970. What made them quite popular, subsequently, was the explosion of ﬁber optics in long-haul telecommunications, and the strictly correlated research eﬀort in integrated optics. Interest in exploiting optical frequencies for telecommunications had been stimulated, in the early ’60’s, by the invention of the laser. If one thinks of guiding waves at optical frequencies along a given path, one easily sees that there are at least three very important reasons why the metallicwall waveguides of Chapter 7 are absolutely unsuitable. Firstly, the size of the guide would be too small, both for manufacturing and handling it. Secondly, conductors would introduce too much loss; actually, the correct physical picture of the interaction between an e.m. ﬁeld and charge carriers at the surface of a conductor, at optical frequencies, is not simple, and requires quantummechanical methods. Thirdly, for a link length and a bandwidth of technical relevance, dispersion in a metallic waveguide would be prohibitively large. In this chapter, we will lay down the theoretical foundations of how e.m. waves are guided either by step discontinuities, or by graded inhomogeneities. When this can be done, we will make the analytical approach as similar as possible to that of Chapter 7. In some cases, however, we will be forced to 343 © 2006 by Taylor & Francis Group, LLC

344

DIELECTRIC WAVEGUIDES

adopt approximate techniques. In these instances, we will follow as closely as possible some ideas that were outlined in Chapter 5, e.g., the WKBJ technique and geometrical optics. However, we will point out that in general all these approximations become accurate only when the transverse dimensions of the waveguides are fairly large compared to the wavelength. Consequently, we will be reasonably satisﬁed with some analytical solutions only for highly multimode waveguides, not for single-mode or few-mode ones. In these cases, we will face a rather “black or white” situation, where either we are able to ﬁnd exact solutions, or we have to resort to numerical methods. Because of the wide technical interest in optical ﬁbers, we will deal explicitly with waveguides where the magnetic permeability is constant throughout the structure, and normally equal to that of free space. However, all the results of this chapter can easily be translated, by duality, into results which apply to structures where the dielectric permittivity is constant and the guiding discontinuities aﬀect the magnetic permeability μ. This chapter is organized as follows. We will commence with an unspeciﬁed geometry, where the dielectric permittivity exhibits just one step discontinuity. We will show that a structure of this kind supports guided modes (a ﬁnite number, at any ﬁnite frequency) and unbounded modes (a continuum). With the help of two examples — slab waveguides and cylindrical rods — we will show that only a subset of the guided modes are TE and TM, while the others modes are hybrid. Then, we will outline why the exact analytical modes of step-index circular waveguides are too cumbersome, and how they can be replaced by approximate solutions, whose most interesting feature is that their transverse ﬁelds are linearly polarized. Next, we will elaborate further on dispersion in dielectric rods, a very important issue in characterizing optical ﬁbers. We will subsequently proceed to graded-index waveguides, show how they can be studied with approximate analytical techniques, and discuss brieﬂy their performance limitations. Finally, we will just touch upon losses in dielectric waveguides. 10.2

Waves guided by a surface of discontinuity. The characteristic equation

The mathematical approach to an inhomogeneous dielectric waveguide varies, depending on whether the structure consists of a ﬁnite number of homogeneous media, or of a medium whose permittivity varies continuously as function of transverse coordinates. It is convenient to deal ﬁrst with waveguides made of several homogeneous materials, starting from the simplest case, that of two lossless dielectric media, referred to as core and cladding, separated by a cylindrical surface, C, parallel to the z axis. Suppose that the intersection between C and the plane z = constant is a regular closed curve, . At each point on we may deﬁne in an unambiguous way three mutually orthogonal unit vectors, tangent, normal and binormal to , denoted by tˆ, n ˆ, a ˆz , respectively. The last one is also the unit vector in the direction of the z axis. They are related as

© 2006 by Taylor & Francis Group, LLC

WAVES GUIDED BY A SURFACE OF DISCONTINUITY

345

Figure 10.1 A dielectric waveguide of unspeciﬁed geometry, consisting of two homogeneous media.

tˆ × n ˆ = a ˆz (see Figure 10.1). Let co , cl be the permittivities of the inner and outer medium, with co > cl , and μ be the permeability of both media (when not speciﬁed otherwise, suppose μ = μ0 ). We will also use the following symbols, whose meanings are simple extensions of other symbols used in the √ previous chapters: −κ2i = ω 2 μi , βi = ω μi (i = co, cl). Within each of the two regions, since the medium in it is homogeneous and there are no sources, we are entitled to apply the decomposition theorem of a generic ﬁeld into a TE and a TM ﬁeld, which was proved in Section 3.8 and, in a simpliﬁed way which is suitable for the case at hand, in Section 7.2. Indeed, we will retain the basic ansatz of Chapter 7, that all the unknown functions (pre-potentials and ﬁelds) depend on the longitudinal coordinate through an exponential function, in general of complex argument, exp(−γz). Italic symbols will be used, as in Chapter 7, for the functions of the transverse coordinates where the exponential dependence on z has been factored out. The decomposition theorem also tells us that the longitudinal ﬁeld components obey eigenvalue equations like Eq. (7.8), namely: ∇2t Fi = −χ2i Fi

(i = co, cl)

,

(10.1)

χ2i = (γ 2 + ω 2 μi ) = (γ 2 − κ2i )

.

(10.2)

where Fi is either Hiz or Eiz , and:

The logical steps to take from now on are: • to write the total ﬁeld, in the core and in the cladding, as the sum of the TE and the TM ﬁelds; • to express the continuity conditions of the tangential ﬁeld components, Eqs. (1.43) and (1.44), on the core-cladding interface surface, in terms of these ﬁelds; • to study whether this system has solutions, and whether they are discrete, or form a continuum. The ﬁrst step is elementary, regardless of the waveguide shape: E¯i = E¯TEi + E¯TMi

,

¯ i = E¯TEi + E¯TMi H

(i = co, cl)

.

(10.3)

The second step is simple as long as we do it using the general-purpose

© 2006 by Taylor & Francis Group, LLC

346

DIELECTRIC WAVEGUIDES

symbols that have been introduced so far: tˆ · (E¯TEco + E¯TMco )

=

tˆ · (E¯TEcl + E¯TMcl ) ,

ζˆ · (E¯TEco + E¯TMco )

=

ζˆ · (E¯TEcl + E¯TMcl ) ,

¯ TEco + H ¯ TMco ) tˆ · (H

=

¯ TEcl + H ¯ TMcl ) , tˆ · (H

¯ TEco + H ¯ TMco ) ζˆ · (H

=

¯ TEcl + H ¯ TMcl ) ζˆ · (H

.

(10.4)

Comparison between this set of four equations and the boundary conditions in a metallic waveguide, Eqs. (7.17) and (7.18), shows immediately that the mathematical complexity of this problem is much greater than that of Chapter 7. For all the ﬁeld components involved in Eqs. (10.4) we might write the general expressions, either Eqs. (7.4) or Eqs. (7.11) and (7.12). This would give us the correct perception that the system in Eqs. (10.4) contains four unknown functions, Hiz and Eiz , in addition to the parameter γ which is still undetermined. However, from a practical viewpoint this is an insigniﬁcant progress, unless the following condition is satisﬁed: the interface surface between core and cladding, C, is a coordinate surface in a reference frame where the Helmholtz equation is separable. To grasp the actual meaning of this point, the reader should return to Section 7.6. It tells us that the geometries which can be studied in closed form are very few: planar slabs, circular-core and elliptical-core waveguides (parabolic waveguides are never found in practice). Suppose then that we are in one of these fortunate cases, and let (u1 , u2 , z) be the suitable coordinates, (h1 , h2 , h3 = 1) be their metric coeﬃcients, and u1 = a = constant be the equation of the coordinate surface C. Then, each of Eqs. (10.1) can be split, as we saw in Chapter 7, into two second-order ordinary diﬀerential equations (ODE’s). So, in general every function Fi is a product of a linear combination of two functions of u1 , multiplied by a linear combination of two functions of u2 , and therefore it contains four arbitrary constants. At this stage, we must draw a sharp distinction between two possible cases, which will be clariﬁed by means of examples in the following sections. Let us proceed as follows. A) In the ﬁrst case, purely physical considerations enable us either to break a general integral into two parts (e.g., separating even and odd symmetry solutions), or to eliminate, as unacceptable, some of the functions of u 1 or u2 which show up in the general integrals (e.g., disregard functions which diverge to inﬁnity within their domain of deﬁnition). On some occasions, as we will see later, a procedure of this kind is capable of cutting the number of arbitrary constants in each function Fi from four to one, hence the total number of arbitrary constants in the problem from sixteen to four. In those cases, we may write: Hiz (u1 , u2 )

© 2006 by Taylor & Francis Group, LLC

= Ai Qi (u1 ) Mi (u2 )

(i = co, cl) ,

WAVES GUIDED BY A SURFACE OF DISCONTINUITY Eiz (u1 , u2 ) = Bi Ri (u1 ) Ni (u2 )

(i = co, cl) ,

347 (10.5)

where Ai , Bi (i = co, cl) are the remaining four arbitrary constants. Using either Eqs. (7.4) or Eqs. (7.11) and (7.12), Eqs. (10.4) can be rewritten in the following form, where it is understood that all the functions of u1 are evaluated at u1 = a: 1 dQco γ 1 dNco Aco Mco + Bco Rco h1 du1 jωco h2 du2 1 dQcl γ 1 dNcl = Acl Mcl + Bcl Rcl , h1 du1 jωcl h2 du2 Bco

2 κ2co − γ 2 κco − γ 2 Rco Nco = Bcl Rcl Ncl , jωco jωcl

1 dRco dMco γ 1 Aco − Qco Nco + Bco jωμ h2 du2 h1 du1 1 dRcl dMcl γ 1 = Acl − Qcl Ncl , + Bcl jωμ h2 du2 h1 du1

Aco (κ2co − γ 2 ) Qco Mco = Acl (κ2cl − γ 2 ) Qcl Mcl .

(10.6)

For this system of four homogeneous equations in the four variables Ai , Bi to have non-trivial solutions, the determinant of the coeﬃcients (in square brackets) in Eqs. (10.6) must be equal to zero. This yields a characteristic equation, where the unknown is the propagation constant γ. In general, it is a transcendental equation in the complex ﬁeld, since in the initial ansatz γ was supposed to be complex, and other parameters involved in the arguments of the functions M, N, Q, R can be complex too. However, it is strictly mandatory to check whether the solutions of the characteristic equation are consistent with the physical arguments that were used in order to arrive at Eqs. (10.6), eliminating some of the mathematically correct solutions of the ODE’s. The examples will show that this requirement is satisﬁed, for any ﬁnite value of the angular frequency ω, only by a ﬁnite number of solutions of the characteristic equation, all of which correspond to imaginary values of γ, and to e.m. ﬁelds whose power travels along the z direction. We may express these ﬁndings by saying that the structure has a ﬁnite number of guided modes. It can also be shown that, quite generally, the values of β = −jγ, which correspond to solutions of this kind, fall within the range deﬁned by the intrinsic phase constants of the cladding, βcl , and the core, βco , namely: βcl < β < βco

.

(10.7)

Clearly, from what we saw in the previous chapters, we may infer that this ﬁnite number of solutions cannot be a complete set of modes. To build a

© 2006 by Taylor & Francis Group, LLC

348

DIELECTRIC WAVEGUIDES

complete set, we may account also for solutions of a diﬀerent physical type, to be sketched as the next point. B) In the second case, on the contrary, one is not able to collect additional physical information, allowing us to write Eqs. (10.5) and so reducing the number of arbitrary constants. To quote just an example, this occurs for a cylindrical rod when one wants to describe ﬁelds traveling radially in the cladding. In general, they must be expressed as superpositions of Hankel functions of the ﬁrst and the second kind (see Appendix D). Under these circumstances, Eqs. (10.4) and the separation of variables yield, instead of Eqs. (10.6), a set of equations where the number of unknowns is larger than the number of equations. According to the Rouch´e-Capelli theorem of algebra, such a set has nontrivial solutions for any value of the propagation constant γ. Therefore, in addition to the ﬁnite number of guided modes, any dielectric waveguides has a continuum of modes. To grasp their physical nature, one has to check what values of γ are compatible with Maxwell’s equations in the core and in the cladding. It can be shown that, regardless of the waveguide shape, in a lossless structure the net result is as follows: √ 1. γ can take imaginary values, provided that |γ| < βcl = ω μcl . This continuum of solutions is referred to as the set of the radiation modes. Each of them consists of uniform plane waves, traveling in oblique directions with respect to the waveguide axis; 2. γ can take real values. This is also a continuum of solutions, referred to as the set of the evanescent modes. Each of them consists of evanescent plane waves whose attenuation vector is parallel to the waveguide axis, and consequently whose phase vector is orthogonal to the z axis. Such modes may reach an appreciable amplitude only in the immediate neighborhood of the ends of a waveguide of ﬁnite length. To summarize, at any ﬁnite value of the frequency a dielectric waveguide supports: • a ﬁnite number of guided modes; • a continuum of radiation modes; • a continuum of evanescent modes. The number of guided modes depends on the frequency, similarly to the number of propagating modes in a metallic waveguide. However, there is a major diﬀerence between metallic and dielectric waveguides from the viewpoint of cut-oﬀ. A mode below cut-oﬀ in a metallic waveguide cannot propagate, but does maintain its individual features. When a guided mode goes to its cut-oﬀ frequency in a dielectric waveguide, it loses its discrete nature, and merges into the continuum of radiation modes. Let us point out that in general the characteristic equation of the system (10.6) has other solutions, in addition to the ﬁnite number of guided modes, if γ is permitted to scan the whole complex plane, without restrictions. More precisely, these additional solutions form a countable inﬁnity, and

© 2006 by Taylor & Francis Group, LLC

GUIDED MODES OF A SLAB WAVEGUIDE

349

all of them are characterized by complex γ’s, with nonvanishing real and imaginary parts. At ﬁrst sight, each of these solutions might look like a generalized version of a guided mode, the attenuation accounting for some power radiated laterally. For this reason, ﬁelds of this kind, satisfying Eq. (10.1) and the following equations, have been called leaky modes of the dielectric waveguide. However, it is easy to show that they cannot be true rigorous modes of a lossless structure. Indeed, the characteristic equation can be derived through a crucial step, the selection based on physical requirements among the mathematically correct solutions of Eq. (10.1). Leaky modes violate these physical requirements, and because of this contradiction they cannot be looked at as physically meaningful solutions of the problem. Still, they can be quite useful in practice (Snyder and Love, 1983), to approximate the radiation ﬁeld (i.e., the total ﬁeld minus the guided modes) of a dielectric waveguide as a summation (in general, over a small number of terms) rather than as a line integral in the complex plane. In particular, in a structure aﬀected by loss, where the cut-oﬀ mechanism (i.e., a guided mode becoming radiating) is not simple to describe, leaky modes can be helpful. We will brieﬂy come back to this subject at the end of Section 10.4. Another important diﬀerence between dielectric and metallic waveguides is that, in general, the set of Eqs. (10.6) does not break into two sets of two equations, one including only Hiz (TE modes), and one including only Eiz (TM modes). Hence, in general dielectric waveguides have hybrid modes. TE and TM modes may exist, but are not enough to form a complete set. From this viewpoint, dielectric waveguides may look similar to waveguides surrounded by walls with a non-zero impedance (see Chapter 7). The main diﬀerence is that in a dielectric waveguide the continuity conditions may be expressed as a matching between mode-dependent wave impedances in the direction normal to C, in the core and in the cladding, but not in terms of a ﬁeld-independent wall-impedance. 10.3

Guided modes of a slab waveguide

We choose the ﬁrst example of application, for the broad-scope approach of the previous section, such that we may use rectangular coordinates, which simplify separation of variables and the following steps. However, we cannot apply the contents of the previous section to a rectangular waveguide, since its contour is not one coordinate surface, as it consists of four segments belonging to diﬀerent coordinates surfaces. In contrast to rectangular metallic waveguides, rectangular dielectric waveguides cannot be solved analytically in exact terms (see Snyder and Love, 1983). Therefore, our ﬁrst example will deal with a slab waveguide, which extends indeﬁnitely in one direction (say y) orthogonal to that of propagation (say z) (see Figure 10.2). Let 2a be the slab width along the x direction. We will consider only ﬁelds which are independent of the y coordinate. This assumption will allow us to clarify some basic ideas, but implies a loss in gen-

© 2006 by Taylor & Francis Group, LLC

350

DIELECTRIC WAVEGUIDES

Figure 10.2 Symmetrical step-index slab waveguide.

erality, whose consequences might appear to invalidate some of the statements of the previous section (in particular, the point referring to hybrid modes). The overall picture will be fully clariﬁed only after Section 10.5. Let us get into the details of the general approach outlined in Section 10.2, writing: Qi (u1 ) (i = co, cl) , (10.8) = Xi (x) Ri (u1 ) and following the same path as in Section 7.7, starting from the Helmholtz equation in the core and in the cladding. One gets that the ODE for Xi reads: Xi = −ki2 Xi

,

(10.9)

which is, once again, the harmonic equation, where: ki2 = −Γ2i = γ 2 − κ2i = γ 2 + ω 2 μi

(i = co, cl)

The general integral of Eq. (10.9) can be written as: sin kco x (even modes) Xco = C (odd modes) cos kco x in the core, and∗ :

⎧ x ⎨ e−Γcl |x| |x| Xcl = D ⎩ −Γ |x| e cl

.

(10.10)

(10.11)

(even modes) (10.12) (odd modes)

in the cladding, with C and D arbitrary constants. Solutions of the form exp(Γcl |x|) in the cladding have been disregarded as they diverge when x goes to inﬁnity, therefore are not physically meaningful. Assuming the ﬁeld is independent of the y-coordinate is equivalent to setting ∗ Calling even and odd the modes of a symmetrical slab waveguide is not a uniquely deﬁned process. In fact, when the longitudinal ﬁeld components in Eqs. (10.11) and (10.12) are even functions of x, then the transverse ﬁeld components, which descend from the longitudinal ones through a derivative with respect to x, are odd, and vice-versa. This fact is strictly correlated to what we saw about symmetries in Chapter 3. Denominations adopted here are not consistent with those given in Chapter 3, but are the most widely adopted in optical waveguide theory, and reﬂect the fact that transverse components are more important (and larger) than the longitudinal ones.

© 2006 by Taylor & Francis Group, LLC

GUIDED MODES OF A SLAB WAVEGUIDE

351

Mi ≡ Ni ≡ 0 in Eqs. (10.6). Then, this system breaks down into two sets, of two equations each in two unknowns. One of these sets identiﬁes TE modes, the other one, TM modes. To be more speciﬁc, continuity equations corresponding to even TE modes (where the word even is used according to the previous footnote) read: C kco cos kco a = −D Γcl e−Γcl a

,

2 C kco sin kco a = −D Γ2cl e−Γcl a

.

(10.13)

Setting equal to zero the determinant of the coeﬃcients of Eqs. (10.13) yields the characteristic equation (also called the dispersion relation): kco tan kco a = Γcl

,

(10.14)

which must be combined with the following equation, which stems from Eq. (10.10): 2 kco + Γ2cl = ω 2 μ(co − cl )

.

(10.15)

Any solution of Eq. (10.14) corresponds to a nontrivial solution of Eqs. (10.9), which is the longitudinal component, Hz , of an even TE mode. The corresponding transverse ﬁeld components can be obtained from it by means of Eqs. (7.4). In detail, if we let Hz = Xi (i = co, cl), we obtain the following ﬁeld in the core: jωμ −γ Ey , Hy = 0 , Ex = 0 , Ey = C cos kco x , Hx = kco jωμ (10.16) and in the cladding: Ex = 0 ,

Ey =

jωμ D e−Γcl |x| Γcl

,

Hx =

−γ Ey jωμ

,

Hy = 0 . (10.17)

As we anticipated in the previous footnote, the transverse ﬁeld components are even, for the reasons explained in Section 3.9. The characteristic equation and the ﬁeld components of the odd TE modes, and those of the even and odd TM modes, can be obtained in similar ways, and are summarized in Tables 10.1 and 10.2. All the characteristic equations shown in those tables are to be combined with Eq. (10.15). One of the main points in the theory of the previous section was that at any ﬁnite frequency there is a ﬁnite number of guided modes. So, we should prove that each of the four characteristic equations has, for a ﬁxed ﬁnite value of ω, a ﬁnite number of solutions. Let us prove it for Eq. (10.14), leaving to the reader as an exercise the very similar proofs for the other three equations. By means of Eq. (10.15), and imposing that Γcl is real in order not to lose the physical features of the guided modes, one derives: 2 − β2 |kco | ≤ ω μ(co − cl ) = βco , (10.18) cl where the equal sign holds only for ω = 0. Furthermore, by means of Eq. (10.15),

© 2006 by Taylor & Francis Group, LLC

352

DIELECTRIC WAVEGUIDES Table 10.1 TE modes of the symmetrical slab waveguide.

EVEN TE MODES Characteristic equation: kco tan kco a = Γcl Field components: Core Cladding Hz = C sin kco a

Hz = C sin kco x Ey =

jωμ C cos kco x kco

Hx = −

Ey =

β Ey ωμ

x −Γcl (|x|−a) e |x|

jωμ C sin kco a e−Γcl (|x|−a) Γcl

Hx = − Ex = H y = 0

β Ey ωμ

ODD TE MODES Characteristic equation: kco cot kco a = −Γcl Field components: Core Cladding Hz = C cos kco a e−Γcl (|x|−a)

Hz = C cos kco x Ey = −

jωμ C sin kco x kco

Hx = −

β Ey ωμ

Ey = −

jωμ x C cos kco a e−Γcl (|x|−a) Γcl |x|

Hx = − Ex = H y = 0

Eq. (10.14) can be rewritten as: tan kco a =

β Ey ωμ

2 − β 2 − k2 βco co cl kco

.

(10.19)

The left-hand side of Eq. (10.19) is a periodic function of kco , the right-hand side is a monotonically decreasing function of kco (see Figure 10.3). Within the range deﬁned by Eq. (10.18), a periodic curve and a monotonical one can intersect each other only at a ﬁnite number of points, hence the number of guided modes is ﬁnite. Eq. (10.10) and the inequality in Eq. (10.18) entail that the phase constant of any guided mode, β = −jγ, satisﬁes, for ω = 0, the following constraints: βcl < β < βco

,

(10.20)

which were also anticipated in the general discussion of Section 10.2. As ω tends to zero, Eq. (10.14) always has one solution, the corresponding phase constant also going to zero, together with kco and Γcl . This result indicates that one guided TE mode may propagate in the slab at an arbitrarily low frequency, which makes a fundamental diﬀerence with respect to metallic waveguides. The same is true for the ﬁrst even TM mode as well, but not for any odd mode, neither TE nor TM. Physically, these lowest-order guided

© 2006 by Taylor & Francis Group, LLC

GUIDED MODES OF A SLAB WAVEGUIDE

353

Table 10.2 TM modes of the symmetrical slab waveguide. EVEN TM MODES Characteristic equation: Field components:

1 1 kco tan kco a = Γcl co cl

Core

Cladding

Ez = C sin kco x

Ez = C sin kco a

Hy = − Ex =

jωco C cos kco x kco

β Hy ωco

Hy = − Ex = Ey = H x = 0

x −Γcl (|x|−a) e |x|

jωcl C sin kco a e−Γcl (|x|−a) Γcl

β Hy ωcl

ODD TM MODES Characteristic equation: Field components:

1 1 kco cot kco a = − Γcl co cl

Core

Cladding

Ez = C cos kco x

Ez = C cos kco a e−Γcl (|x|−a)

Hy = Ex =

jωco C sin kco x kco β Hy ωco

Hy = Ex = Ey = H x = 0

jωcl x C cos kco a e−Γcl (|x|−a) Γcl |x| β Hy ωcl

k co a

Figure 10.3 Graphical solution of the characteristic equation of the even TE modes.

© 2006 by Taylor & Francis Group, LLC

354

DIELECTRIC WAVEGUIDES x TE0

TE1

TE2

a Ey

0

-a

a

b

c

Figure 10.4 Examples of low-order TE modes in a slab waveguide. a) Lowest-order even mode. b) Lowest-order odd mode. c) First higher-order even mode.

modes tend to become uniform plane waves in the cladding when ω tends to zero, thus becoming less and less sensitive to the presence of a core. This can never be the case for an odd mode, where the ﬁelds in the two half-spaces which form the cladding, have opposite phase. To reconﬁrm that this is the correct physical explanation of the diﬀerence in cut-oﬀ behavior between the lowest-order even and odd modes, the reader may prove, as an exercise, that in an asymmetrical slab waveguides where the upper and lower half-spaces have diﬀerent permittivities, all the guided modes have a non-zero cut-oﬀ frequency. Amplitude distributions of the various ﬁeld components, and ﬁeld lines, for the lowest-order even and odd guided modes of a slab waveguide, are sketched in Figure 10.4. 10.4

Radiation modes of a slab waveguide

For the radiation modes of a slab waveguide, there are no physical grounds to eliminate, in the cladding, any of the oscillatory solutions (or of the two exponential solutions with imaginary argument) of Eq. (10.9). Therefore, we let: (10.21) Xcl = q ejkcl x + p e−jkcl x , where p and q are two arbitrary constants. Eq. (10.15) is replaced by: 2 2 − kcl = ω 2 μ(co − cl ) . kco

(10.22)

Continuity of tangential components at x = ±a yields, both for TE and for TM modes, two linear homogeneous equations in three unknowns, p, q (from Eq. (10.21)) and C (from Eq. (10.11)). The Rouch´e-Capelli theorem says that such a system has an inﬁnity of nontrivial solutions for any values of the coeﬃcients. Hence, we cannot ﬁnd additional conditions on kco and kcl . The radiation modes are a continuum, whose transverse ﬁeld parameters are

© 2006 by Taylor & Francis Group, LLC

RADIATION MODES OF A SLAB WAVEGUIDE

355

related only by Eq. (10.22). They must satisfy only one restriction, coming 2 from Eq. (10.10) as a consequence of kcl > 0, which is necessary in order to have, as required, an oscillatory behavior in the cladding. The restriction reads: β ≤ βcl , (10.23) where β is now the phase constant of the generic radiation mode. We leave it to the reader as an exercise to study the evanescent modes, namely the solutions with real γ, which must satisfy the restriction ki ≥ βi (i = co, cl), coming again from Eq. (10.10). Let us strengthen the links between the mathematical description of radiation modes and physical intuition, and try to clarify why Eq. (10.21) contains two terms, one corresponding to a wave traveling away from the slab, the other one to a wave traveling towards the slab. It would be erroneous to believe that radiation modes must be able to describe only ﬁelds excited by a source located at |x| < a, inferring from this starting point that waves traveling towards the slab are not physically meaningful. There are at least three reasons why this viewpoint would be wrong; a) To be complete, a set of modes must be able to describe a ﬁeld radiated by a source located on one side of the slab, at an arbitrary distance from it. This also requires waves which travel towards the slab; b) A cladding extending indeﬁnitely in the x direction is a physically acceptable model only if it represents the limit of a cladding with a ﬁnite thickness, d, when d goes to inﬁnity. Centripetal waves are needed in order to describe that part of the radiated ﬁeld which is reﬂected back by the external edge of the cladding. This also tells us that it would be erroneous to disregard centripetal waves; c) The whole centripetal ﬁeld, expressed as an integral over the continuum of radiation modes, can carry zero power, without implying zero amplitude for the individual centripetal waves. This is a well known property of the Fourier integral, and the reader is probably familiar with it from timedomain signal analysis. All these comments reconﬁrm that, as we remarked in Section 10.2, dielectric and metallic waveguides are totally diﬀerent concerning that part of the ﬁeld which does not propagate under the form of guided modes. In dielectric guides, each transmission mode loses its individual features as soon as it is cut-oﬀ. The “invention” of leaky modes can be thought of as an attempt to eliminate this diﬀerence, using another description for that part of the ﬁeld which complements the guided modes. A non-real solution of the characteristic equation, in the complex ﬁeld, could be looked at as an “extension” of a real solution, when a guided mode has gone through a cut-oﬀ condition, i.e., when the angular frequency is below a critical value ωcn , similarly to what occurs in Chapter 7. Every complex solution of the characteristic equation would entail Re γ = 0, and thus an attenuation which is an increasing function of ωcn − ω. These solutions form

© 2006 by Taylor & Francis Group, LLC

356

DIELECTRIC WAVEGUIDES

Figure 10.5 Step-index circular dielectric rod.

a countable inﬁnity and, as mentioned in Section 10.2, are often referred to as leaky modes. However, let us stress once again that the term “modes” might be misleading. As we said in general in Section 10.2, and as we saw in detail in Section 10.3, the characteristic equation was obtained on the grounds of a well deﬁned physical assumption — a guided mode — and there is no reason why it should hold when the physics changes. To appreciate the usefulness of leaky modes as an approximation, and to appreciate the mathematical subtleties of the procedure, the reader is referred to specialized books (Snyder and Love, 1983).

10.5

The cylindrical rod: Exact modes

This section is devoted to the guided modes of a cylindrical dielectric rod, of radius a (Figure 10.5). This example is very important under a practical viewpoint, because of the impact of optical ﬁbers in modern telecommunication technology and in other related ﬁelds, like sensors. We will show that these guided modes can be identiﬁed rigorously, but using their exact expressions makes it complicated to express some important quantities, like power density ﬂowing along the waveguide. For this reason, we will develop an approximate theory in the following section. Radiation modes of a cylindrical waveguide do not exhibit signiﬁcant diﬀerences with respect to those of a slab waveguide, except for mathematical details (Hankel functions replacing exponentials of imaginary argument) which can be left to the reader as an exercise. Let us follow the general procedure of Section 10.2, taking into account the contents of Section 7.8 about the Helmholtz equation in cylindrical coordinates. A physical selection criterion is to disregard those solutions of Bessel’s equations which diverge, either for r < a or for r > a. Accordingly, we can set: i = co, cl ; Mi (10.24) = ejνϕ ν = 0, ±1, ±2, . . . ; Ni

© 2006 by Taylor & Francis Group, LLC

THE CYLINDRICAL ROD: EXACT MODES ur Qco = Jν (kco r) = Jν Rco a

357 ,

(10.25)

where Jν is the Bessel function of the ﬁrst kind and of integer order ν, and: wr Qcl , (10.26) = Kν (Γcl r) = Kν Rcl a where Kν is the modiﬁed Bessel function of the second kind and of integer order ν. The following adimensional quantities have been deﬁned in these equations: u = kco a = a ω 2 μco − β 2 = a ω 2 μco + γ 2 = a γ 2 − κ2co ,

w = Γcl a = a

−ω 2 μcl + β 2 = a

−(γ 2 + ω 2 μcl ) = a

κ2cl − γ 2 , (10.27)

where β = −jγ is the phase constant of the generic guided mode. As for the harmonic equation which governs the dependence on the azimuthal coordinate, we chose to express its solutions in terms of exponential functions, Eq. (10.24), rather than in terms of cos(νϕ) and sin(νϕ), because the continuity conditions, Eqs. (10.6), involve the functions and also their derivatives, so Eqs. (10.24) end up making the calculations easier. The metric coeﬃcients are h1 = 1, h2 = r. Following a standard notation, let us use Jν and Kν for the derivatives of Bessel functions with respect to their arguments. Then, Eqs. (10.6) read as follows: Aco u Jν (u) − Bco

βν βν Jν (u) = Acl w Kν (w) − Bcl Kν (w) , jωco jωcl

Bco Aco

u2 Jν (u) jωco

= −Bcl

w2 Kν (w) jωcl

,

βν Jν (u) βν Kν (w) + Bco u Jν (u) = Acl + Bcl w Kν (w) , jωμ jωμ Aco u2 Jν (u) = −Acl w2 Kν (w)

.

(10.28)

For ν = 0, we can separate Eqs. (10.28) into two sets of two equations, one in the two unknowns (Aco , Acl ), the other in the two unknowns (Bco , Bcl ). In fact, it is not surprising that these modes, independent of the azimuthal coordinate ϕ, behave similarly to those independent of the y coordinate in the slab waveguide of Section 10.3. The set which consists of the ﬁrst and the fourth of Eqs. (10.28) corresponds to TE modes, and yields the following characteristic equation: K (w) J0 (u) =− 0 u J0 (u) w K0 (w)

(TE0,p ) .

(10.29)

The set which consists of the second and the third of Eqs. (10.28) corresponds

© 2006 by Taylor & Francis Group, LLC

358

DIELECTRIC WAVEGUIDES

to TM modes, and yields the following characteristic equation: co

J0 (u) K0 (w) = −cl u J0 (u) w K0 (w)

(TM0,p ) .

(10.30)

We see from these characteristic equations that the TE0,p and the TM0,p modes are not degenerate (co = cl ), but tend to become degenerate as co → cl . This comment will be expanded and used in the following section. Modes whose azimuthal order is not equal to zero are hybrid modes, as the set of Eqs. (10.28), for ν = 0, does not break down into two separate sets. Their characteristic equation can be written in a compact form if we set: jν (u) =

Jν (u) u Jν (u)

,

bν (w) =

Kν (w) w Kν (w)

,

(10.31)

and introduce an adimensional quantity, v, proportional to the angular frequency ω, deﬁned as follows: v 2 = u2 + w2 = a2 ω 2 μ(co − cl )

.

(10.32)

Dividing the ﬁrst and the third of Eqs. (10.28) by Jν Kν , eliminating Acl and Bcl by means of the second and the forth equation and, ﬁnally, setting the determinant of the remaining system of two equations in Aco , Bco equal to zero, the characteristic equation then reads: 2 1 1 2 2 2 2 (jν + bν ) (κco jν + κcl bν ) = −ν β + 2 , (10.33) u2 w where u, w and β are related through Eqs. (10.27). Another, perhaps more elegant way to write the characteristic equation is: 1 co 1 cl 2 , (10.34) + + (jν + bν ) (co jν + cl bν ) = ν u2 w2 u2 w2 which must be associated with Eq. (10.32). To reach this result, we divide Eq. (10.33) by v 2 , write 1/u2 +1/w2 = v 2 /(u2 w2 ), replace v 2 with Eq. (10.32), and ﬁnally, make use of the following expression, which can be obtained taking the squares of the two Eqs. (10.27), multiplying the ﬁrst by cl and the second by co , and summing them: β 2 a2 (co − cl ) = co w2 + cl u2

.

(10.35)

Note that Eq. (10.34) breaks into Eqs. (10.29) and (10.30) for ν = 0. For ν = 0, Eq. (10.34) is invariant with respect to the change ν → −ν, since (see , bν = b−ν . Therefore, each solution of Eq. (10.34) for Appendix D) jν = j−ν ν = 0 corresponds to a pair of degenerate modes. This agrees with physical intuition: waves propagating helically in an isotropic medium cannot distinguish between traveling clockwise or counterclockwise. This enables us to assume ν > 0 from now on, to simplify our calculations. Our next aim is to show that the solutions of the characteristic equation may be separated into two sets, which are physically diﬀerent from each other.

© 2006 by Taylor & Francis Group, LLC

THE CYLINDRICAL ROD: EXACT MODES

359

A suitable way to reach this result is simply to rewrite Eq. (10.34) as a secondorder algebraic equation in jν , and to solve it, while introducing, for future beneﬁts, two new variables, namely: co + cl , Δ = co − cl , = (10.36) 2 and using a concise notation, fν2 (u, w), for the right-hand side of Eq. (10.34). Make use of the following expressions for derivatives of Bessel functions (see Appendix D): ν Jν (u) − Jν+1 (u) , (10.37) Jν (u) = u when taking the positive square root of the discriminant, and: ν Jν (u) + Jν−1 (u) , u when taking the negative one. The ﬁnal results are, respectively: 1/2 Jν+1 (u) u Δ2 2 ν 2 = b + co fν (u, w) ub + − Jν (u) co ν u co 4 ν Jν (u) = −

for the plus sign in front of the discriminant, and: 1/2 Jν−1 (u) u Δ2 2 ν =− bν + co fν2 (u, w) ubν + − Jν (u) co u co 4

(10.38)

(10.39)

,

(10.40)

for the negative one. Let us emphasize once again that Eqs. (10.39) and (10.40) assume ν > 0, and each solution of either one corresponds to a pair of degenerate modes having identical behaviors vs. the radial coordinate, given by Eqs. (10.25) and (10.26), while they diﬀer in the sign in front of the azimuthal coordinate, exp(±jνϕ). The solutions of Eq. (10.39) are referred to as the EHν,p modes, those of Eq. (10.40) as the HEν,p modes. Radial and azimuthal components of the e.m. ﬁeld may be obtained from the longitudinal components, i.e., from Eqs. (10.24), (10.25) and (10.26), expressing Eqs. (7.4) in cylindrical coordinates. They are summarized in Table 10.3. The reader can verify, as an exercise, that radial components satisfy the following continuity conditions (which, as shown in Chapter 1, are not independent of those which hold for tangential components): co Erco (r = a) = cl Ercl (r = a) Hrco (r = a) = Hrcl (r = a) .

, (10.41)

Table 10.3 shows that the longitudinal components of these exact modal ﬁelds are fairly simple but, on the other hand, transverse components are complicated, mainly because Bessel functions of integer order ν can be written as superpositions of two Bessel functions of diﬀerent orders (to choose between ν, ν + 1, and ν − 1). Transverse components are fundamental in practice, e.g., to calculate the Poynting vector. Furthermore, it can be shown (see the rest of the chapter) that, in agreement with physical intuition, when the diﬀerence

© 2006 by Taylor & Francis Group, LLC

360

DIELECTRIC WAVEGUIDES Table 10.3 Exact modes of the circular step-index rod. Core (ρ ≤ 1)

Cladding (ρ ≥ 1)

Hz

A Jν (uρ) ejνϕ

A

Jν (u) Kν (wρ) ejνϕ Kν (w)

Ez

B Jν (uρ) ejνϕ

B

Jν (u) Kν (wρ) ejνϕ Kν (w)

Er

−

j jνϕ e · u2

· B Eϕ

−

βau Jν (uρ)

jωμ0 νa +A Jν (uρ) ρ

j jνϕ e · u2

Hr Hϕ

jβaν · B Jν (uρ) − A ωμ0 aw Jν (wρ) ρ ωco − Eϕ β ωco Er β

μ

HE modes:

B=j

EH modes:

B = −j

0

co

μ

j Jν (u) jνϕ e · w2 Kν (w)

· B

βaw Kν (wρ)

jωμ0 νa +A Kν (wρ) ρ

j Jν (u) jνϕ e · w2 Kν (w)

jβaν · B Kν (wρ) − A ωμ0 aw Kν (wρ) ρ ωcl − Eϕ β ωcl Er β

A 0

co

A

co − cl is not too large, the magnitude of the transverse components is much larger than that of longitudinal components. The practical need to simplify transverse ﬁeld components is a valid reason why the approximate theory of Section 10.7 is often preferred to the exact theory presented in this section. 10.6

Modal cut-oﬀ in the cylindrical rod

A suitable way to clarify the physical diﬀerence between EH and HE modes, is to analyze their cut-oﬀ conditions. As we found in general in Section 10.2, and saw reconﬁrmed by the example of Section 10.3, a guided mode is cut-oﬀ when its ﬁeld does not decay, in the radial direction, fast enough to prevent power from ﬂowing in the radial direction. This is translated into mathematical terms as w → 0, to be replaced in the previous equations. To proceed further, we make use of the following approximations (Abramowitz and Stegun, Eds., 1965) for small values of w: bν (w) −ν/w in the case of Eq. (10.39), and bν (w) −(ν/w) − 1/[2(ν − 1)] in the case of Eq. (10.40). Keeping also track that Eq. (10.32) implies that at cut-oﬀ (w = 0) it is u = v = vc (critical value of v parameter), after some algebra one ﬁnds that the cut-oﬀ condition arising

© 2006 by Taylor & Francis Group, LLC

MODAL CUT-OFF IN THE CYLINDRICAL ROD

361

Table 10.4 Cut-oﬀ normalized frequencies for some low-order modes. vc

0

2.405

3.832

5.136

5.520

Modes

HE11

TE01 , TM01

EH11 , HE12

EH21

TE02 , TM02

from Eq. (10.39) reads: Jν (vc ) = 0

(EHν,p modes ; ν = 1, 2, . . .) ,

(10.42)

while that arising from Eq. (10.40) reads: vc

2(ν − 1) ¯ Jν (vc ) = Jν−1 (vc ) cl

(HEν,p modes ; ν = 1, 2, . . .)

.

(10.43)

In particular, the latter equation, for ν = 1 and vc = 0, gives Jν (vc ) = 0, showing that at the cut-oﬀ (but only at cut-oﬀ) HE1,p modes and EH1,p−1 modes are degenerate. Eqs. (10.42) and (10.43) show that, for any ﬁnite value of the normalized frequency v deﬁned by Eq. (10.32), each family of guided modes is a ﬁnite set. This result could also be predicted similarly to what was carried out in Section 10.3 on the characteristic equation of the slab modes, i.e., observing that one side of these equations is monotonic, while the other one is a function which goes, sequentially, through zero and through inﬁnity. Eq. (10.43) also shows that the fundamental mode is the HE1,1 mode, which can propagate at an arbitrarily small frequency (vc = 0) like the fundamental mode of a slab. We will see in the following sections that fundamental modes of a slab and of a cylindrical rod are physically very similar, in spite of their labels appearing to be completely diﬀerent from each other. A very practical matter is to establish which are the ﬁrst higher-order modes. The question can be answered letting w = 0 in Eqs. (10.29) and (10.30), and this yields J0 (vc ) = 0. Therefore, the next modes are the TE0,1 and the TM0,1 . They are degenerate only at cut-oﬀ. The cut-oﬀ value for the normalized frequency is vc = 2.405. Cut-oﬀ values of the normalized frequency for other modes are shown in Table 10.4. For a better understanding of the role played by various parameters in the design of practical single-mode and multimode ﬁbers, let us rewrite the normalized frequency, deﬁned by Eq. (10.32), as follows, in terms of the refractive indices of core and cladding: √ √ 2πa ncl 2Δ , (10.44) v = ka ncl 2Δ = λ0 √ where k = ω μ0 0 = 2π/λ0 is the free-space wave number, and: 2Δ ≡

n2co − n2cl n2cl

(10.45)

is the relative index diﬀerence. In comparison with Chapter 7, we notice that, while in a conducting-wall waveguide the cut-oﬀ frequency of a given mode

© 2006 by Taylor & Francis Group, LLC

362

DIELECTRIC WAVEGUIDES Table 10.5 Typical values of refractive index for glasses in common use.

Pure silica (SiO2 ) GeO2 -doped silica F-doped silica Halide glasses Chalcogenide glasses

1.4585 1.4585 – 1.465 1.445 – 1.4585 1.41 – 1.72 1.7 – 2.5

depends on the guide size and on the refractive index of the dielectric medium which ﬁlls it, in a dielectric waveguide it depends on these parameters plus another one, namely the core-cladding index relative diﬀerence. For an optical ﬁber to be single-mode, v must be less than 2.405. Wavelengths of highest practical interest (measured in free space) are in the near infrared, more precisely in the range between 0.8 μm (the so-called “ﬁrst window”) and 1.55 μm (“third window”; these names are related to historical breakthroughs in the development of optical communications) where silica glass is very transparent (see Section 10.11). The refractive index of pure silica, at those wavelengths, is somewhat less than 1.5 (see Table 10.5). These √ facts indicate that the product a 2Δ must be smaller than a few units (depending on the exact wavelength) multiplied by 10−7 , if a is measured in meters. Actual choices for a and 2Δ are dictated by other, contradictory requirements (see Snyder and Love, 1983), typically bending loss and coupling loss. However, for all ﬁbers which are standard commercial products, a is a few micrometers, and 2Δ is less than 0.01. These are highly demanding requirements from a technological viewpoint. This point explains why highly multimode ﬁbers, where a is of the order of tens of micrometers and 2Δ reaches a few percent so that v 2.405, have attracted extremely serious technical attention for decades, in spite of some major limitation that will be outlined in Section 10.8. On the other hand, few-mode ﬁbers, where v > 2.405, but not v 2.405, are of interest only for some special applications, like sensors, which are beyond the scope of this book (Dakin and Culshaw, 1988). Incidentally, let us point out that, as 2Δ 1 in all cases of practical interest, we have nco + ncl 2ncl , so that Eq. (10.45) can be rewritten as Δ

nco − ncl ncl

( 1)

.

(10.46)

The practical importance of having Δ 1 will be thoroughly appreciated in the next three sections. Finally, let us try to explain why the cut-oﬀ conditions which were derived from the characteristic equations justify the strange terminology which is in current use for the dielectric waveguide modes. In fact, Eq. (10.42) appears to be the same as the equation which, in a circular waveguide with conducting walls, yields cut-oﬀ frequencies for those modes which are most frequently referred to as TM, but are called E modes by other authors. As we saw here, modes of a dielectric rod are hybrid; still, as a result of this purely formal

© 2006 by Taylor & Francis Group, LLC

WEAKLY GUIDING RODS: THE LP MODES

363

similarity, one set was named EH, and then, by analogy, the other one was named HE. 10.7

Weakly guiding rods: The LP modes

We saw in the previous section that, for most dielectric waveguides used in practice, and in particular for typical optical ﬁbres, if we return to the notation of Eq. (10.36), we ﬁnd: Δ ¯ . (10.47) A dielectric structure where Eq. (10.47) is satisﬁed is referred to as weakly guiding. Introducing this assumption into Eq. (10.34), we ﬁnd that both characteristic equations, for HE and for EH modes, result in the following approximate form: 1 1 ∼ + 2 jν + bν = ±ν . (10.48) u2 w This tells us that HE and EH modes tend to become degenerate as Δ → 0. Knowing from Section 10.5 that there is a strict two-fold degeneracy related to replacing ν with −ν, we see that for Δ → 0 we tend towards a four-fold degeneracy of the guided modes, in a dielectric rod. In mathematical terms, the idea of a “quasi-degeneracy” is not well founded. In any eigenvalue problem, degenerate solutions are critical, in the sense that some diﬃculties show up when two eigenvalues are equal, but disappear as soon as they are not equal, irrespectively of how close they are, provided they are distinct. If we adopt this viewpoint, then two modes which correspond to two diﬀerent eigenvalues can never be linearly combined to yield another mode, even if the two eigenvalues are very close to each other. However, in a weakly guiding dielectric waveguide of circular cross-section, the errors we make if we treat as degenerate EH and HE modes whose eigenvalues are not strictly identical can be shown (Snyder and Love, 1983) to be small, compared to errors that we make neglecting other perturbations, like slight ellipticity, absorption and scattering losses, bends, slight unintentional anisotropy, etc. As a result, the ﬁelds that we build up in this section are not exact solutions of Maxwell’s equations, and therefore should be called pseudomodes, rather than modes, but they match very well experimental results on actual ﬁbers aﬀected by imperfections, and in this sense can be referred to as modes. Let us proceed as if the two families of modes were degenerate. It is easy to show that, using the same notation as in the previous section, degeneracy does not occur between the HEν,m and the EHν,m mode, but between the HEν,m and the EHν−2,m mode. Indeed, using the recurrence formula: 2(ν − 1) Jν−1 (u) = Jν (u) + Jν−2 (u) . (10.49) u Eq. (10.43) shows that for ¯/cl → 1 the cut-oﬀ equation for the HEν,m mode becomes: (10.50) Jν−2 (vc ) = 0 ,

© 2006 by Taylor & Francis Group, LLC

364

DIELECTRIC WAVEGUIDES

which coincides with the cut-oﬀ condition for the EHν−2,m mode, Eq. (10.42), which is insensitive to the approximation cl ¯. This four-fold degeneracy can be exploited to derive drastically simpler transverse components. The price to pay is to have more complicated longitudinal components, but, as we said before, these are far less used in practical calculations. Furthermore, the reader can prove as an exercise (working, to simplify calculations, on the modes of a slab waveguide or on the TE and TM modes of a cylindrical rod) that the longitudinal components are much smaller than the transverse ones. In a weakly guiding structure, this ratio can be shown to be of order Δ1/2 , where Δ is still deﬁned by Eq. (10.45). The procedure which yields the simpliﬁed transverse ﬁelds can be summarized as follows, omitting calculations for the sake of brevity. To start with, let = 1, 2, . . .. (For = 0, there is nothing to add to what was said in Section 10.5). Then: • take the transverse components of the HE+1,m mode and those of the EH−1,m mode (see Table 10.3); • use Bessel and modiﬁed Bessel function recurrence relations to express all these components as linear combinations of Bessel and modiﬁed Bessel functions of two orders, to be chosen among , + 1, − 1; for example, let us take those of orders and + 1; • look for a linear combination of the HE+1,m mode and the EH−1,m mode ensuring that it does not contain Bessel functions of one of these remaining orders, say +1. Note that this step is not trivial at all, as the HE+1,m mode and the EH−1,m mode have diﬀerent dependences on the azimuthal coordinate, expressed by exp[j( ± 1) ϕ], respectively. The reason why one can ﬁnd a positive answer to this requirement is precisely the four-fold mode degeneracy, i.e., the possibility of combining linearly not only the above mentioned functions of ϕ, but also exp[−j( ± 1) ϕ]. The number of independent coeﬃcients one can play with then becomes suﬃcient to eliminate J+1 and K+1 from all the transverse ﬁeld components. The reader can prove as an exercise that, referring to their dependence on the radial coordinate, the transverse ﬁeld components, for the generic mode of this new set, are expressed, respectively, in the core by the upper line and in cladding by the lower line of the following formula: J (ur/a)/J (u) Eϕ = E0 F (ϕ) , Hr = −Eϕ /ηi , K (wr/a)/K (w) J (ur/a)/J (u) Er = E0 G (ϕ) , Hϕ = Er /ηi , (10.51) K (wr/a)/K (w) where E0 is an arbitrary constant, and the pair of functions of the azimuthal coordinate F (ϕ), G (ϕ) will be discussed shortly. In Eqs. (10.51) it is ηi = μ/i = η0 /ni , ni being the refractive index of the i-th medium (i = co, cl). This result indicates that, when the assumption Δ 1 holds, for any guided

© 2006 by Taylor & Francis Group, LLC

WEAKLY GUIDING RODS: THE LP MODES

365

Figure 10.6 Rectangular and cylindrical reference frames in a circular rod.

mode expressed by Eqs. (10.51) the wave impedance in the direction of propagation is equal to the intrinsic impedance of the medium. Consequently, comparing with Chapters 5 and 7, we see that we are dealing with a quasi-TEM propagation. Let us focus now on the azimuthal dependence of the transverse ﬁelds. Calculations show that the pair of functions in Eqs. (10.51) can be any among the following four choices (incidentally, note that the four-fold degeneracy has been preserved): F = sin( + 1) ϕ F = sin( − 1) ϕ ; ; G = − cos( + 1) ϕ G = cos( − 1) ϕ (10.52) F = cos( + 1) ϕ F = cos( − 1) ϕ ; . G = sin( + 1) ϕ G = − sin( − 1) ϕ At ﬁrst sight, these results may look disappointing, because the ϕ-dependence is not very simple and is actually not even the same for the Eϕ and Er components. Fortunately, these problems are very easy to circumvent. Indeed, the ﬁeld components in the rectangular coordinate frame shown in Figure 10.6 are related to Eqs. (10.51) as: Ex

= Er cos ϕ − Eϕ sin ϕ ,

Ey

= Er sin ϕ + Eϕ cos ϕ .

(10.53)

Starting from the four choices expressed by Eqs. (10.52), Eqs. (10.53) yield the following four possibilities: Ex ∝ cos ϕ Ex ∝ − cos ϕ ; ; (10.54) Ey ∝ sin ϕ Ey ∝ sin ϕ Ex ∝ sin ϕ Ex ∝ − sin ϕ ; . (10.55) Ey ∝ cos ϕ Ey ∝ cos ϕ Finally, summing and subtracting Eqs. (10.54), and Eqs. (10.55), we get the following four choices: Ex = 0 , Ex = 0 ,

© 2006 by Taylor & Francis Group, LLC

Ey = E0 R(r) sin ϕ , Ey = E0 R(r) cos ϕ ,

(10.56)

366

DIELECTRIC WAVEGUIDES x

x

|E| J0

K0

y

y

r/a

1

a

b

c

Figure 10.7 Schematic illustration of the fundamental (LP0,1 ) mode. Part a: xpolarization. Part b: y-polarization (continuous lines = electric ﬁeld; dashed lines = magnetic ﬁeld). Part c: transverse ﬁeld amplitude (arbitrary units) vs. normalized radial coordinate.

Ex = E0 R(r) cos ϕ Ex = E0 R(r) sin ϕ

, ,

Ey = 0 , Ey = 0 ,

(10.57)

where E0 is an arbitrary constant and R(r) is the quantity in brackets in Eqs. (10.51). Each of these electric ﬁelds is accompanied by a magnetic ﬁeld expressed by: Ex Ey , Hx = − . (10.58) Hy = ηi ηi Accordingly, we see that the transverse ﬁeld of each of these modes consists of one cartesian component for the electric ﬁeld, and one cartesian component for the magnetic ﬁeld orthogonal to the electric one and with the same radial distribution. This explains why these modes are said to be linearly polarized. The corresponding concise notation is LP,p , where = 0, 1, 2, . . . and p = 1, 2, . . . Note that: • the LP,p mode with > 1 is the combination of the HE+1,p mode and the EH−1,p mode; • the LP1,p mode is the combination of the HE2,p mode, the TE0,p mode and the TM0,p mode; • the LP0,p mode is the weakly-guiding approximation of the HE1,p mode; • any LP,p mode with > 0 exhibits a four-fold degeneracy, while an LP0,p mode is two-fold degenerate. Using the new terminology, the fundamental mode is the LP0,1 mode. Its ﬁeld lines and amplitude distribution are sketched in Figure 10.7. The ﬁgure illustrates why the fundamental mode is doubly degenerate: it is a matter of polarization. However, degeneracy is broken when circular symmetry is broken, either because the core cross-section is not circular (shape birefringence), or because the medium is anisotropic as a result of internal force distribution

© 2006 by Taylor & Francis Group, LLC

WEAKLY GUIDING RODS: THE LP MODES

367

Figure 10.8 Constant-amplitude contour plots for the ﬁrst higher-order (LP 1,1 ) mode.

(stress birefringence). Usually, in silica glass, the latter eﬀect dominates over the ﬁrst one. In some cases (high-birefringence ﬁbers), stress anisotropy is intentional, and meant to preserve the state of polarization of the guided ﬁeld. In typical telecommunication ﬁbers, birefringence is unwanted, and varies at random along the ﬁber. It results in distortion of the signal which travels along a single-mode ﬁber, and is usually referred to as polarization mode dispersion (Matera and Someda, 1992; Galtarossa and Menyuk, 2005), and probably easier to understand after reading the next section. The ﬁrst higher-order mode, whose normalized cut-oﬀ frequency remains vc = 2.405, is now called LP1,1 mode. Its ﬁeld distribution is sketched in Figure 10.8, which illustrates the sources of the four-fold degeneracy: one twofold degeneracy is related to polarization, while the other two-fold degeneracy is related to a rotation of the amplitude distribution around the z axis by an angle π/(2). The ﬁeld components of the LP modes are summarized in Table 10.6, which reconﬁrms that, as expected, longitudinal components become complicated while the transverse ones are simpliﬁed. The characteristic equation of the LP modes, Eq. (10.48), can be expressed in another way, which will be useful for further developments in the next section. It is also easier to correlate with physical intuition, arising from the ﬁelds shown in Table 10.6. Let us take Eq. (10.48) with the plus sign, the expression for jν which results from Eq. (10.37), and the expression for bν which results from: Kν (w) =

© 2006 by Taylor & Francis Group, LLC

ν Kν (w) − Kν+1 (w) . w

(10.59)

368

DIELECTRIC WAVEGUIDES Table 10.6 LP modes of the circular step-index rod.

The expressions shown in this table are related to the second among Eqs. (10.56). The other three cases can be obtained from these with the following replacements: π π π . ϕ1 = − ϕ ; ϕ3 = − ϕ ; ϕ4 = ϕ − ( − 1) 2 2 2 Core

Cladding

Ey = E 0

J (ur/a) cos ϕ J (u)

Hx = −nco Ez = −j

Ey η0

E0 u 2ka nco

Ez = −j

J−1 (ur/a) sin( − 1) ϕ J (u)

Hz = −j

J−1 (ur/a) cos( − 1) ϕ J (u)

E0 w 2ka ncl

K−1 (wr/a) sin( − 1) ϕ K (w)

Hz = −j

Ey η0

K+1 (wr/a) sin( + 1) ϕ− K (w) −

E0 u 2ka η0

J+1 (ur/a) cos( + 1) ϕ− J (u) −

K (wr/a) cos ϕ K (w)

Hx = −ncl

J+1 (ur/a) sin( + 1) ϕ+ J (u) +

E y = E0

E0 w 2ka η0

K+1 (wr/a) cos( + 1) ϕ+ K (w) +

K−1 (wr/a) cos( − 1) ϕ K (w)

Let us recall that we are now interested in the characteristic equation for the LP,p mode, corresponding to the EH−1,p , not to the EH,p mode; therefore, let ν = − 1. The ﬁnal result is: u

K−1 (w) J−1 (u) = −w J (u) K (w)

( = 0, 1, 2, . . .) ,

(10.60)

to be associated with Eq. (10.32). Looking carefully at Table 10.6, we see that Eq. (10.60) would be the characteristic equation of the system which we obtain if we stipulate that the radial components of the electric and of the magnetic ﬁeld are continuous at the core-cladding interface, r = a: Erco (r = a)

= Ercl (r = a) ,

Hrco (r = a)

= Hrcl (r = a) .

(10.61)

The second of Eqs. (10.61) is correct, because there is no discontinuity in magnetic permeability at r = a, but the ﬁrst one is not rigorous as long as

© 2006 by Taylor & Francis Group, LLC

DISPERSION IN DIELECTRIC WAVEGUIDES

369

co = cl . In fact, from Chapter 1 we know that the correct continuity condition at r = a is that the normal components of the electric displacement vector, D = E, are equal in the core and in the cladding. This comment indicates another implicit meaning of the weak-guidance approximation (or, equivalently, LP approximation): the approximation implies that we neglect refraction of the ﬁeld lines of the electric vector across the interface between the two media. Consequently, informations concerning polarization of the transverse ﬁelds, if drawn through LP modes, are not accurate. In reality, the linear polarization ratio of any guided mode is never inﬁnite. In spite of this limitation, LP modes are extremely useful in yielding practical information on optical ﬁbers, as we shall see in the following sections. 10.8

Dispersion in dielectric waveguides

In practical applications of dielectric waveguides (among which, optical telecommunication systems are by far and large the most important, nowadays) the main limits to performances come from attenuation and dispersion. Loss in a dielectric waveguide is a rather simple subject to deal with, and will be brieﬂy discussed in Section 10.11. On the contrary, dispersion deserves an accurate discussion, which may sound surprising. In fact, in the previous sections we have established, ﬁrstly, that the phase constant of any guided mode, β, must be in the range between βcl and βco , the intrinsic phase constants of the two dielectric media; secondly, that the diﬀerence in permittivity between these media is very small in all cases of practical interest. Comparing these results with those concerning guided modes in a conducting-wall waveguide, one might infer that in a dielectric waveguide variations of the phase constant vs. frequency are small and smooth, so that dispersion cannot be large. The reason why this conclusion is unacceptable is easy to explain. The lengths of practical microwave waveguides never exceed a few meters, which means, at most, a few hundreds of wavelengths. Optical ﬁbers can be more than hundred kilometers in length, which means more than 1010 wavelengths. Accuracy in modeling dispersion must then grow in proportion to these ﬁgures. Dispersion can be dealt with analytically in the weak-guidance approximation, i.e., when: nco − ncl 1 . (10.62) Δ ncl Given that βcl < β < βco , let us deﬁne a new quantity whose purpose is to expand this range and so emphasize the variations of β vs. frequency. Let: β2 β − ncl − n2cl 2 k k b≡ 2 nco − n2cl nco − ncl

.

(10.63)

Clearly, b is adimensional and varies between 0 and 1. The last approximate step in the previous equation comes from the fact that β/k + ncl nco + ncl . The ratio β/k is often referred to as the eﬀective index of the guided mode

© 2006 by Taylor & Francis Group, LLC

370

DIELECTRIC WAVEGUIDES

b

v Figure 10.9 Normalized phase constants vs. normalized frequency. The numbers refer to LP mode suﬃces.

characterized by the phase constant β. Clearly, its values are conﬁned in the range between ncl and nco . Eqs. (10.27) and (10.44) yield another expression for b, namely: b(v) = 1 −

u2 v2

,

(10.64)

where let us stress that u is also a function of v. Plots of Eq. (10.64) for some LP modes are shown in Figure 10.9 (from Gloge, 1971). Let us calculate, rather than the group velocity vg of the various LP modes, its inverse, referred to as the (per-unit-length) group delay, τg , following the deﬁnition given in Chapter 5, namely: τg =

1 dβ dβ = dω c0 dk

,

(10.65)

√ where c0 = 1/ μ0 0 is the speed of light in free space. In contrast to the usual situation in conducting-wall waveguides at microwave frequencies, contributions due to material dispersion, i.e., to d(μ)/dω = 0 (see Chapter 5), cannot be neglected in dielectric waveguides at optical frequencies. To account for them without making the notation too cumbersome,

© 2006 by Taylor & Francis Group, LLC

DISPERSION IN DIELECTRIC WAVEGUIDES

371

let us deﬁne the following quantities: d(ni k) dβi dni = = ni + k (i = co, cl) , (10.66) dk dk dk which are referred to as the group indices of the core and cladding materials. Let us deﬁne also the quantity: Ni ≡

Δ ≡

Nco − Ncl k d(nco − ncl ) =Δ+ ncl ncl dk

,

(10.67)

which may diﬀer signiﬁcantly from the relative index diﬀerence, Δ, only in waveguides where the core and cladding materials are completely diﬀerent (e.g., silica-plastic ﬁbers). Elaborating on β as a function of b, from Eq. (10.63), one ﬁnds the following rigorous result: dβ 1 1 db = ncl Ncl + (nco Nco − ncl Ncl ) b + v (1 + 2Δb)−1/2 . dk ncl 2 dv (10.68) This expression can be simpliﬁed, given that Δ 1. Let (1 + 2Δb)−1/2 1 − Δb, then neglect terms of order Δ2 , and use Eq. (10.67). The result is: d(vb) 1 dΔ db dβ Ncl + nco Δ − knco v . (10.69) dk dv 2 dk dv For a given waveguide, the ﬁrst term is independent of the particular mode. Accordingly, in multimode propagation regime, diﬀerences in group delay (referred to as intermodal dispersion) are due only to the second and third terms. The last one is essentially negligible for all modes, for which db/dv 0, except, at most, the fundamental mode, but we will return to this later on. The factor d(vb)/dv, which governs the second term, is plotted in Figure 10.10 for some low-order LP modes. It may be seen immediately that some of the modes (those of azimuthal order = 0, 1) are fast very close to cut-oﬀ, when most of their guided power travels in the cladding, where the speed of light is larger than in the core. For the modes of higher azimuthal order, even at cut-oﬀ the power fraction in the cladding is too small for the mode to be fast. In any case, as soon as they are suﬃciently above cut-oﬀ all the modes tend to become faster as the normalized frequency increases, for the same reason that the group velocity increases vs. frequency in a conducting-wall waveguide (see Section 7.4). Figure 10.10 also shows that the diﬀerence between the largest and the smallest normalized group delay depends, in its ﬁne details, on v in a complicated manner, but on the contrary its order of magnitude is unity, and remains essentially constant as soon as v 2.4, as it is in practical multimode ﬁbers. Therefore, the intermodal dispersion can be estimated to be: δτ int =

1 1 nco Δ (Nco − Ncl ) , c0 c0

(10.70)

with an approximation nco /ncl = 1 in the last step. Essentially, this result tells us that in a multimode step-index ﬁber intermodal dispersion equals the

© 2006 by Taylor & Francis Group, LLC

372

DIELECTRIC WAVEGUIDES

d ( vb ) dv

v Figure 10.10 Normalized group delays (waveguide contributions) vs. normalized frequency. The numbers refer to LP mode suﬃces.

diﬀerence between the group delays of two uniform plane waves, traveling in the core and in the cladding, respectively. It is left to the reader as an exercise to prove that an intermodal dispersion of this order of magnitude causes, over a ﬁber length of practical interest (at least 10 km), pulse spreading compatible only with very low bit rates. Actually, experimentally measured intermodal dispersions in step-index ﬁbers are somewhat smaller than these predictions, because the modes close to cutoﬀ are ﬁltered out by their losses, signiﬁcantly higher than those of low-order modes. However, they do not allow transmission of attractive bit rates. There are essentially two ways of improving this performance. One is to change the index proﬁle signiﬁcantly, and will be the subject of the following sections. The other one is to use single-mode ﬁbers. Let us now elaborate further on their dispersion properties. The very small values required for the index diﬀerence (see Section 10.6) implies that core and cladding materials diﬀer at most for a very small amount of dopants. This allows us to let Δ Δ, i.e., Δ independent of wavelength, thus canceling the third term in Eq. (10.69), and rewrite it (in the presence of just the fundamental mode LP0,1 ) as: dβ d(vb) Ncl + (Nco − Ncl ) . (10.71) dk dv In Chapter 5, we saw that a signal is distorted when the group velocity (or

© 2006 by Taylor & Francis Group, LLC

DISPERSION IN DIELECTRIC WAVEGUIDES

373

Figure 10.11 Normalized waveguide dispersion.

group delay) varies with frequency, within the signal spectral band. In the multimode regime, we overlooked this phenomenon, because it is dominated by the much larger intermodal dispersion. In the single-mode regime, we must account for the source being non-monochromatic, and study how the group delay of the guided ﬁeld varies vs. frequency, i.e., for intramodal dispersion, sometimes referred to also as chromatic dispersion. To calculate the derivative of the group delay with respect to angular frequency: s=

1 d2 β dτg = 2 dω c0 dk 2

,

(10.72)

starting with Eq. (10.71) we get: d(Nco − Ncl ) d(vb) v d2 (vb) d2 β ∼ dNcl + + (Nco − Ncl ) . (10.73) = 2 dk dk dk dv k dv 2 Thus, the intramodal dispersion of the LP0,1 mode consists of three terms. The ﬁrst originates from material dispersion. The third term indicates — similar to what we saw in Chapter 7 for conducting-wall waveguides — that a guided wave, being non-TEM, is inevitably dispersive, and therefore is referred to as waveguide dispersion. The second term is a combination of material and waveguide eﬀects. Usually, its magnitude is signiﬁcantly less than that of the ﬁrst and that of the third term. However, before neglecting it completely, we must investigate whether the ﬁrst and third term may cancel each other. Figure 10.11 shows that in the single-mode regime (v < 2.405) the last term in Eq. (10.73) is positive, for any choice of the ﬁber material and geometry. Note that this behavior is opposite to that of a metal-wall waveguide, where the group velocity increases as a function of frequency (anomalous dispersion). This indicates that, in the entire single-mode band, the phenomenon which governs waveguide dispersion is the growth in the power fraction that travels

© 2006 by Taylor & Francis Group, LLC

374

DIELECTRIC WAVEGUIDES

Table 10.7 International standards for intramodal dispersion. Source: International Telecommunication Union (formerly CCITT); G.652 and G.653. Unit: ps/(km·nm). Wavelength range Fiber optimized at 1300 nm Fiber optimized at 1550 nm

1285–1330 nm ≤ 3.5 —

1270–1430 nm ≤6 —

1525–1575 nm ≤ 20 ≤ 3.5

in the core, which was outlined previously. On the other hand, the sign of the ﬁrst term depends on the speciﬁc waveguide material, and on the wavelength, in a given material. In pure amorphous silica, material dispersion is positive at λ < 1.27 μm, negative at longer wavelengths. Consequently, it is possible to ∼ design a weakly guiding glass ﬁber so that its intramodal dispersion Eq. (10.73) goes through zero at one wavelength λ0 > 1.3 μm. The exact zero-dispersion ∼ wavelength, λ0 , is determined by the second term in Eq. (10.73), since the sum of the ﬁrst and third term is close to zero. The simplest and most widely used designs give ﬁbers whose λ0 is close to 1.3 μm, i.e., in the spectral range which is called, as we said before, “second window”. These ﬁbers are usually referred to as standard single-mode ﬁbers. However, the growing interest in exploiting the “third window,” where losses pass through their minimum value, and optical ampliﬁers are commercially available, has encouraged the design of ﬁbers whose zero-dispersion wavelength λ0 is close to 1.5 μm. These are referred to as dispersion shifted (DS) ﬁbers. Typically, their refractive index proﬁle is more complicated than the single step considered here. The reader who is interested in more details should consult many of the books quoted at the end of this chapter. More complicated ﬁbers have been envisaged, where total dispersion is kept below a given upper limit over the whole range from 1.3 μm to 1.55 μm. They are referred to as dispersion ﬂattened ﬁbers. However, interest in these appears to be low, in recent times. Units in current technical use, to measure intramodal dispersion of optical ﬁbers, are ps/(km·nm), i.e., picoseconds (of group delay) divided by kilometers (of ﬁber length) and by nanometers (of source spectral width). Some useful reference data are shown in Table 10.7. Before we close this section, we would like to remind the reader of the contents of Section 5.2. The quantity expressed by Eq. (10.72) is merely the ﬁrst coeﬃcient, in the expansion of the delay as a series of powers of Δλ. When the ﬁrst coeﬃcient is equal to zero, the signal distortion is governed by the next term, which is proportional to Δλ2 . So, at the zero-dispersion wavelength the ﬁber bandwidth is not inﬁnite, but is still orders of magnitude broader than at other wavelengths.

© 2006 by Taylor & Francis Group, LLC

GRADED-INDEX WAVEGUIDES

375

Figure 10.12 Ray model of propagation in a graded-index waveguide.

10.9

Graded-index waveguides

As we saw in the previous section, in a multimode step-index ﬁber intermodal dispersion can be too large for long-distance pulse transmission, even at moderate bit rates. A solution, which has been studied in depth theoretically and experimentally, and successfully applied during the early decades of optical communications, is to reduce intermodal dispersion by “equalizing” the speeds of propagation of the various modes. One can do this by acting on the refractive index distribution in the waveguide cross-section, as a function of the radial coordinate. This idea is easy to justify, qualitatively, in terms of ray optics: “times of ﬂight” along guided ray paths of diﬀerent geometrical lengths can be equalized by increasing the speed of propagation (i.e., decreasing the refractive index n) along the longer paths, namely further away from the waveguide axis (see Figure 10.12). Propagation along graded-index waveguides cannot be studied rigorously, in vector terms. However, most of the results of practical interest can be obtained by means of an analysis based on the WKBJ approximation, which was presented, in general, in Section 5.7. This will be the subject of the next section. Before we proceed in this direction, however, it is worthwhile to sketch brieﬂy, omitting some details, how one can pass from Maxwell’s equations to a rather scalar-looking problem, which can then be tackled using the WKBJ method. Let us see whether Eqs. (7.4), or Eqs. (7.11) and (7.12), which express transverse ﬁelds in terms of the longitudinal components Ez , Hz , are still valid when , and maybe μ, depend on the transverse coordinates. Using Maxwell’s equations, one can show that this is indeed the case. On the contrary, Ez and Hz do not satisfy Eq. (7.8), in an inhomogeneous medium. This implies that the theory of guided waves that we developed for conducting-wall waveguides, and then for step-index ﬁbers, does not hold for graded index waveguides, because it was based on the pair of scalar functions which were called prepotentials in Section 3.8. To replace Eq. (7.8) with two new equations, let us go back to Maxwell’s equations, calculate the curl of both sides, replace ∇×∇× with −∇2 +∇ (∇·), and account for Eqs. (1.35) and (1.38), which, for = constant, μ = constant, yield ∇ · H = −(1/μ)∇μ · H and ∇ · E = −(1/)∇ · E, respectively. The ﬁnal

© 2006 by Taylor & Francis Group, LLC

376

DIELECTRIC WAVEGUIDES

results are the following equations: 2 ∇t χ 2 ∇t μ ∇t Hz + χ2 Hz ζˆ + − × ∇t Hz × ζˆ = 2 χ μ

−γ ∇t χ2 × ∇t Ez , jωμ χ2 (10.74)

2 ∇t χ 2 ∇t − ∇t Ez + χ2 Ez ζˆ + × ∇t Ez × ζˆ = 2 χ

−γ ∇2t χ2 × ∇t Hz , jω χ2 (10.75)

where it is important not to forget that χ2 = γ 2 + ω 2 μ is now a function of the radial coordinate, r. Given our interest in dielectric waveguides, similar to those of Section 10.8, let μ = μ0 = constant, = (r) = 0 n2 (r), where: 2 n0 [1 − 2Δf (r)] , r

Our partners will collect data and use cookies for ad personalization and measurement. Learn how we and our ad partner Google, collect and use data. Agree & close