Table of Contents

Table of Contents

Electronic and Electrical Engineering

Electromagnetic Propagation in
Multi-Mode Random Media
Author(s): Harrison E. Rowe
Series Editor(s): Kai Chang
ISBN: 0471200700 (Electronic) 0471110035 (Print)
Copyright © 1999 John Wiley & Sons, Inc.

i-xiii Frontmatter and Index

Harrison E. Rowe
Summary PDF Full Text (Size: 309K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.fmatter_indsub

Chapter 1: Introduction
1-3 Harrison E. Rowe
Summary PDF Full Text (Size: 42K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.ch1

Chapter 2: Coupled Line Equations

5-20 Harrison E. Rowe
Summary PDF Full Text (Size: 114K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.ch2

Chapter 3: Guides with White Random Coupling

21-48 Harrison E. Rowe
Summary PDF Full Text (Size: 163K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.ch3

Chapter 4: Examples - White Coupling

49-85 Harrison E. Rowe
Summary PDF Full Text (Size: 212K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.ch4

Chapter 5: Directional Coupler with White Propagation Parameters

87-97 Harrison E. Rowe
Summary PDF Full Text (Size: 99K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.ch5

Chapter 6: Guides with General Coupling Spectra

99-122 Harrison E. Rowe
Summary PDF Full Text (Size: 152K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.ch6

http://www3.interscience.wiley.com/cgi-bin/booktoc?ID=86512825 (1 of 3) [1/14/2002 3:55:12 PM]

 Table of Contents

Chapter 7: Four-Mode Guide with Exponential Coupling Covariance

123-127 Harrison E. Rowe
Summary PDF Full Text (Size: 59K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.ch7

Chapter 8: Random Square-Wave Coupling

129-143 Harrison E. Rowe
Summary PDF Full Text (Size: 119K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.ch8

Chapter 9: Multi-Layer Coatings with Random Optical Thickness

145-159 Harrison E. Rowe
Summary PDF Full Text (Size: 124K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.ch9

Chapter 10: Conclusion

161-162 Harrison E. Rowe
Summary PDF Full Text (Size: 36K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.ch10

163-168 Appendix A: Series Solution for the Coupled Line Equations

Harrison E. Rowe
Summary PDF Full Text (Size: 70K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.app0a

169-173 Appendix B: General Transmission Properties of Two-Mode Guide

Harrison E. Rowe
Summary PDF Full Text (Size: 62K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.app0b

175-177 Appendix C: Kronecker Products

Harrison E. Rowe
Summary PDF Full Text (Size: 49K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.app0c

179-190 Appendix D: Expected Values of Matrix Products

Harrison E. Rowe
Summary PDF Full Text (Size: 94K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.app0d

191-196 Appendix E: Time- and Frequency-Domain Statistics

Harrison E. Rowe
Summary PDF Full Text (Size: 75K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.app0e

197-202 Appendix F: Symmetric Slab Waveguide - Lossless TE Modes

Harrison E. Rowe
Summary PDF Full Text (Size: 76K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.app0f

http://www3.interscience.wiley.com/cgi-bin/booktoc?ID=86512825 (2 of 3) [1/14/2002 3:55:12 PM]

 Table of Contents

203-208 Appendix G: Equal Propagation Constants

Harrison E. Rowe
Summary PDF Full Text (Size: 66K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.app0g

209-210 Appendix H: Asymptotic Form of Coupled Power Equations

Harrison E. Rowe
Summary PDF Full Text (Size: 46K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.app0h

211-214 Appendix I: Differential Equations Corresponding to Matrix Equations

Harrison E. Rowe
Summary PDF Full Text (Size: 55K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.app0i

215-226 Appendix J: Random Square-Wave Coupling Statistics

Harrison E. Rowe
Summary PDF Full Text (Size: 84K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.app0j

227-229 Appendix K: Matrix for a Multi-Layer Structure

Harrison E. Rowe
Summary PDF Full Text (Size: 47K)
Published Online: 30 Oct 2001
DOI 10.1002/0471200700.app0k

[Wiley InterScience Home Page] [Personal Home Page] [Journal Finder] [Book Finder] [Search Wiley InterScience] [Reference Works] [Help]
[Contact Us] [Logout]

Copyright © 1999-2001 by John Wiley & Sons, Inc. All rights reserved.

http://www3.interscience.wiley.com/cgi-bin/booktoc?ID=86512825 (3 of 3) [1/14/2002 3:55:12 PM]

 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

Electromagnetic Propagation
in Multi-Mode Random Media
Electromagnetic
Propagation
in Multi-Mode
Random Media

HARRISON E. ROWE

A WILEY-INTERSCIENCE PUBLICATION
JOHN WILEY & SONS, INC.
NEW YORK / CHICHESTER / WEINHEIM / BRISBANE / SINGAPORE / TORONTO
Designations used by companies to distinguish their products are often claimed as trademarks. In all
instances where John Wiley & Sons, Inc., is aware of a claim, the product names appear in initial
capital or ALL CAPITAL LETTERS. Readers, however, should contact the appropriate companies for
more complete information regarding trademarks and registration.

Copyright © 1999 by John Wiley & Sons, Inc. All rights reserved.

No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form
or by any means, electronic or mechanical, including uploading, downloading, printing, decompiling,
recording or otherwise, except as permitted under Sections 107 or 108 of the 1976 United States
Copyright Act, without the prior written permission of the Publisher. Requests to the Publisher
for permission should be addressed to the Permissions Department, John Wiley & Sons, Inc.,
605 Third Avenue, New York, NY 10158-0012, (212) 850-6011, fax (212) 850-6008, E-Mail:
[email protected].

This publication is designed to provide accurate and authoritative information in regard to the subject
matter covered. It is sold with the understanding that the publisher is not engaged in rendering
professional services. If professional advice or other expert assistance is required, the services of a
competent professional person should be sought.

ISBN 0-471-20070-0

This title is also available in print as ISBN 0-471-11003-5.

For more information about Wiley products, visit our web site at www.Wiley.com.
 To Alicia,
Amy, Elizabeth, Edward, and Alison,
and to the memory of
Stephen O. Rice
Contents

1 Introduction 1
References 3

2 Coupled Line Equations 5

2.1 Introduction 5
2.2 Two-Mode Coupled Line Equations 6
2.3 Exact Solutions 8
2.4 Discrete Approximation 11
2.5 Perturbation Theory 13
2.6 Multi-Mode Coupled Line Equations 15
References 19

3 Guides with White Random Coupling 21

3.1 Introduction 21
3.2 Notation—Two-Mode Case 23
3.3 Average Transfer Functions 25
3.4 Coupled Power Equations 30
3.5 Power Fluctuations 35
3.6 Transfer Function Statistics 39
3.7 Impulse Response Statistics 43
3.8 Discussion 46
References 47

4 Examples—White Coupling 49
4.1 Introduction 49
vii
viii CONTENTS

4.1.1 Single-Mode Input 50

4.1.2 Multi-Mode Coherent Input 51
4.1.3 Multi-Mode Incoherent Input 51
4.2 Coupled Power Equations—Two-Mode Case 52
4.2.1 Single-Mode Input 52
4.2.2 Two-Mode Coherent Input 55
4.2.3 Two-Mode Incoherent Input 58
4.3 Power Fluctuations—Two-Mode Guide 58
4.3.1 Single-Mode Input 60
4.3.2 Two-Mode Coherent Input 60
4.3.3 Two-Mode Incoherent Input 61
4.3.4 Discussion 62
4.4 Impulse Response—Two-Mode Case 64
4.5 Coupled Power Equations—Four-Mode Case 67
4.5.1 Single-Mode Input 69
4.5.2 Multi-Mode Coherent Input 72
4.5.3 Multi-Mode Incoherent Input 72
4.6 Nondegenerate Case—Approximate Results 74
4.6.1 Average Transfer Functions 75
4.6.2 Coupled Power Equations 77
4.6.3 Power Fluctuations 81
4.6.4 Discussion 84
4.7 Discussion 84
References 85

5 Directional Coupler with White Propagation

Parameters 87
5.1 Introduction 87
5.2 Statistical Model 88
5.3 Average Transfer Functions 91
5.4 Coupled Power Equations 93
5.5 Discussion 95
References 97

6 Guides with General Coupling Spectra 99

6.1 Introduction 99
6.2 Almost-White Coupling Spectra 100
6.2.1 Two Modes 101
 CONTENTS ix

6.2.2 N Modes 104

6.3 General Coupling Spectra—Lossless Case 106
6.3.1 Two Modes 109
6.3.2 N Modes 110
6.4 General Coupling Spectra—Lossy Case 115
6.4.1 Two Modes 116
6.4.2 N Modes 118
6.5 Discussion 121
References 122

7 Four-Mode Guide with Exponential Coupling

Covariance 123
7.1 Introduction 123
7.2 Average Transfer Functions 126
7.3 Coupled Power Equations 127
7.4 Discussion 127

8 Random Square-Wave Coupling 129

8.1 Introduction 129
8.2 Two Modes—Binary Independent Sections 132
8.3 Two Modes—Binary Markov Sections 135
8.4 Four Modes—Multi-Level Markov Sections 138
8.5 Discussion 143

9 Multi-Layer Coatings with Random Optical Thickness 145

9.1 Introduction 145
9.2 Matrix Analysis 147
9.3 Kronecker Products 150
9.4 Example: 13-Layer Filter 152
9.4.1 Statistical Model 153
9.4.2 Transmittance 155
9.4.3 Two-Frequency Transmission Statistics 155
9.5 Discussion 158
References 159

10 Conclusion 161
References 162
x CONTENTS

Appendix A Series Solution for the Coupled Line

Equations 163
References 168

Appendix B General Transmission Properties of

Two-Mode Guide 169
References 173

Appendix C Kronecker Products 175

References 177

Appendix D Expected Values of Matrix Products 179

D.1 Independent Matrices 179
D.2 Markov Matrices 183
D.2.1 Markov Chains 183
D.2.2 Scalar Variables 184
D.2.3 Markov Matrix Products 186
References 190

Appendix E Time- and Frequency-Domain Statistics 191

E.1 Second-Order Impulse Response
Statistics 191
E.2 Time-Domain Analysis 195
References 196

Appendix F Symmetric Slab Waveguide—Lossless TE

Modes 197
F.1 General Results 197
F.2 Example 200
References 202

Appendix G Equal Propagation Constants 203

Appendix H Asymptotic Form of Coupled Power Equations 209

Appendix I Differential Equations Corresponding to

Matrix Equations 211
I.1 Scalar Case 211
I.2 Matrix Case 212
 CONTENTS xi

References 214

Appendix J Random Square-Wave Coupling Statistics 215

J.1 Introduction 215
J.2 Binary Sections 217
J.2.1 Independent 218
J.2.2 Markov 218
J.3 Multi-Level Markov Sections 219
J.3.1 Low-Pass—Six Levels 219
J.3.2 Band-Pass—Five Levels 224
References 226

Appendix K Matrix for a Multi-Layer Structure 227

Index 231
Electromagnetic Propagation
in Multi-Mode Random Media
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

WILEY SERIES IN MICROWAVE AND OPTICAL ENGINEERING

KAI CHANG, Editor

Texas A&M University

FIBER-OPTIC COMMUNICATION SYSTEMS, Second Edition l Covind P. Agrawal

COHERENT OPTICAL COMMUNICATIONS SYSTEMS l Silvello Betti, Ciancarlo De Marchis and
Eugenio lannone
HIGH-FREQUENCY ELECTROMAGNETIC TECHNIQUES: RECENT ADVANCES AND
APPLICATIONS l Asoke K. Bhattacharyya
COMPUTATIONAL METHODS FOR ELECTROMAGNETICS AND MICROWAVES.
Richard C. Booton, )r.
MICROWAVE RING CIRCUITS AND ANTENNAS . Kai Chang
MICROWAVE SOLID-STATE CIRCUITS AND APPLICATIONS l Kai Chang
DIODE LASERSAND PHOTONIC INTEGRATED CIRCUITS l Larry Coldren and Scott Corzine
MULTICONDUCTOR TRANSMISSION-LINE STRUCTURES: MODAL ANALYSIS TECHNIQUES .
1. A. Branddo Faria
PHASED ARRAY-BASED SYSTEMS AND APPLICATIONS l Nick Fourikis
FUNDAMENTALS OF MICROWAVE TRANSMISSION LINES l /on C. Freeman
MICROSTRIP CIRCUITS l Fred Cardiol
HIGH-SPEED VLSI INTERCONNECTIONS: MODELING, ANALYSIS, AND SIMULATION .
A. K. Gael
FUNDAMENTALS OF WAVELETS: THEORY, ALGORITHMS, AND APPLICATIONS .
Jaideva C. Coswami and Andrew K. Chan
PHASED ARRAY ANTENNAS l R. C. Hansen
HIGH-FREQUENCY ANALOG INTEGRATED CIRCUIT DESIGN l Ravender Goyal (ed.)
MICROWAVE APPROACH TO HIGHLY IRREGULAR FIBEROPTICS l Huang Hung-Chia
NONLINEAR OPTICAL COMMUNICATION NETWORKS l Eugenio lannone, Francesco Matera,
Antonio Mecozzi, and Marina Settembre
FINITE ELEMENT SOFTWARE FOR MICROWAVE ENGINEERING l Tatsuo Itoh, Giuseppe Pelosi
and Peter P. Silvester (eds.)
SUPERCONDUCTOR TECHNOLOGY: APPLICATIONS TO MICROWAVE, ELECTRO-OPTICS,
ELECTRICAL MACHINES, AND PROPULSION SYSTEMS l A. R. /ha
OPTICAL COMPUTING: AN INTRODUCTION l M. A. Karim and A. S. S. Awwal
INTRODUCTION TO ELECTROMAGNETIC AND MICROWAVE ENGINEERING l Paul R. Karmel,
Gabriel D. Colef, and Raymond L. Camisa
MILLIMETER WAVE OPTICAL DIELECTRIC INTEGRATED GUIDES AND CIRCUITS l

Shiban K. Koul
MICROWAVE DEVICES, CIRCUITS AND THEIR INTERACTION l Char/es A. lee and
G. Conrad Da/man
ADVANCES IN MICROSTRIP AND PRINTED ANTENNAS l Kai-Fong Lee and Wei Chen (eds.)
OPTICAL FILTER DESIGN AND ANALYSIS: A SIGNAL PROCESSING APPROACH l
C. K. Madsen and 1. H. Zhao
OPTOELECTRONIC PACKAGING l A. R. Mickelson, N. R. Basavanhaly, and Y. C. Lee feds.)
ANTENNAS FOR RADAR AND COMMUNICATIONS: A POLARIMETRIC APPROACH l
Harold Mott
INTEGRATED ACTIVE ANTENNAS AND SPATIAL POWER COMBINING l julio A. Navarro and
Kai Chang
FREQUENCY CONTROL OF SEMICONDUCTOR LASERS l Motoichi Ohtsu (ecf.)
SOLAR CELLS AND THEIR APPLICATIONS l Larry 0. Par&in (ed.)
ANALYSIS OF MULTICONDUCTOR TRANSMISSION LINES l Clayton R. Paul
INTRODUCTION TO ELECTROMAGNETIC COMPATIBILITY l Clayton R. Paul
INTRODUCTION TO HIGH-SPEED ELECTRONICS AND OPTOELECTRONICS .
Leonard M. Riaziat
NEW FRONTIERS IN MEDICAL DEVICE TECHNOLOGY l Arye Rosen and Hare/ Rosen (eds.)
ELECTROMAGNETIC PROPAGATION IN MULTI-MODE RANDOM MEDIA l Harrison E. Rowe
NONLINEAR OPTICS l E. C. Sauter
InP-BASED MATERIALS AND DEVICES: PHYSICS AND TECHNOLOGY l Osamu Wada
and Hideki Hasegawa (eds.)
FREQUENCY SELECTIVESURFACE AND GRID ARRAY l T. K. Wu (ed.)
ACTIVE AND QUASI-OPTICAL ARRAYS FOR SOLID-STATE POWER COMBINING l
Robert A. York and Zoya B. PopovZ (eds.)
OPTICAL SIGNAL PROCESSING, COMPUTING AND NEURAL NETWORKS l Francis T. S. Yu
and Suganda lutamulia
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0
232 INDEX

Coupling coefficients (continued) Markov matrices:

example,68 example,189
two modes,8,23 expectedvalues,187
constant,9-lo,88 secondmoments,188-l 89
discrete,delta-function,9-l 2 Modes, 1,5,6. Seealso Propagation
Cross-powers,seePowersand cross- parameters
powers backward,1,8,87
degenerate,8, 16
Dielectric slab waveguide,67-68, 124, delay,24-25
131,197-198 dispersion,25
example,200-202 forward, 1,6,8
Differential and differenceequations: nondegenerate, 30,43,49,74
matrix, 2 12-2 13 powers.Seealso Coupledpower
scalar,2 11 equations;Cross-powers;Power
Directional couplers,l-2,5. Seealso fluctuations
Randompropagationparameters multi-mode, 15,16
conservation,16
Filter: two modes,7
intensity impulseresponse,45, 195 conservation,7
wide-sensestationarytransferfunction, signal and spurious,7,23-24
44,191 velocities,24,43
correlation bandwidth,45, 194 Multi-layer coatings,1,2, 146
spectraldensity,44-45, 192 Kroneckerproducts,15l-l 52
matrix representation,146148,229
Impulseresponse,25, 196 reflectanceandtransmittance,145,149
statistics,45-46 reflection andtransmission,145,
example,64-66 147-148
Inputs: seriesexpansions,149-150
coherent,30,50-5 1 thirteen-layerfilter:
incoherent,30, 50-52 designtransmittance,154
parameters,153
Kroneckerproducts,3,30,33,40,42,84, transmissioncorrelationcoefficient,
96,99,121,147,158,161 157-158
properties,175-l 77 transmissionvariance,157
transmittance,156
MAPLE, 3,6,34-35,37,47,49,63,81, Multi-mode:
143,154,162 optical fibers, 1, 5
Marcuse,74,80 waveguides,1,5,
Markov:
coupling, 135,138-139,218-219,224 Optical coatings,seeMulti-layer coatings
matrices,183, 186
scalars,183-l 84 Personick,66
Matrix products,seealso Kronecker Perturbationtheory, 143,147,158, 161
products multi-mode, 18- 19, 108
independentmatrices: two modes,13-15,106107
expectedvalues, 181 Powerfluctuations,36, 58-59
secondmoments,182 multi-mode,37
 INDEX 233

nondegenerate,83 propagation,seeRandompropagation
two modes,37-38 parameters
examples,60-63 separationbetweenguides,2
Powersand cross-powers,31 straightness,2,5
multi-mode, 33-34,50-52 Randompropagationparameters,88-9 1
two modes,32,93-95 correlationlength,89
Propagationparameters,6. Seealso
Randompropagationparameters Schelkunoff,5
multi-mode, 15 Square-wavecoupling:
attenuationandphase,18 coupling function, 130,215
constant,17,22, 100 coupling spectra:
example,68 binary independent,133,218
degenerate,16,206 binary Markov, 136137,2 18
two modes,6-7 five-level, band-passMarkov, 141,
attenuationandphase,6-7, 11-12, 224
22,24-26,38,40 six-level, low-passMarkov, 140,219
constant,9-10,22,30,49,100 examples:
degenerate,8,203 four modes:
six-level, low-passMarkov
Randomcoupling coefficients,seealso sections,141-142
Square-wavecoupling two modes:
almost-whitespectra,1OO-101 binary independentsections,
correlation length, 13, 101 132-134
Gaussianor Poisson,2 1 binary Markov sections,135, 138
generalspectra: inputs, 130
band-pass,107
low-pass,100 Transferfunction(s),23-24
white spectrum,2 1,3 1,49 covariance:
Randomoptical thickness,2. SeeaZso multi-mode,43
Multi-layer coatings two modes,42
Randomparameters: Transmissionstatistics,1,2, 6. Seealso
coupling,seeRandomcoupling Averagetransferfunctions;
coefficients Coupledpowerequations;Impulse
guide width, 2 response,statistics;Power
layer thickness,seeMulti-layer coatings fluctuations;Square-wave
optical thickness,seeMulti-layer coupling;Transferfunctions,
coatings covariance;Multi-layer coatings
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

CHAPTER ONE

Introduction

This text presents analytic methods for calculating the transmission

statistics of microwave and optical components with random imper-
fections. Three general classes of devices are studied:

1. Multi-mode guides such as oversize waveguides or optical

fibers.
2. Directional couplers.
3. Multi-layer optical coatings used as windows, mirrors, or fil-
ters, with plane-wave excitation.

All of these various transmission media and devices are multi-

mode. For the first two categories the significant modes travel in
the same direction, that is, forward; coupling to backward modes
is neglected. In the third class of devices, the two modes are plane
waves traveling in opposite directions.
Electromagnetic calculations yield equations that describe their
performance in terms of coupling between modes. For multi-mode
guides and directional couplers these are called the coupled line
equations. For multi-layer coatings, they are the Fresnel reflection
and transmission coefficients between adjacent layers. The starting
point for all of our analyses will be these various equations; we as-
sume their coefficients have been determined elsewhere by electro-
magnetic theory, in terms of the geometry and dielectric constants of
the media comprising each device. No electromagnetic calculations
are contained in the present work.
1
2 INTRODUCTION

The performance of every such system is limited by random de-

partures of its physical parameters from their ideal design values.
These include both geometric and electrical parameters. Some
examples are:

1. The axis of a multi-mode guide may exhibit random straight-

ness deviations, or cross-sectional deformations such as slight
ellipticity in a nominally circular guide.
2. A directional coupler made of two microstrip lines may show
small random variations in the separation of the microstrip
lines or random variations in their individual widths.
3. Microscopic dielectric constant variations may exist in the
medium of either of the above two examples.
4. A multi-layer coating may have random errors in the optical
thickness of the different layers, caused by variations in either
the geometric thickness or the electrical parameters of the
layers.

The statistics of the parameters in the corresponding coupled mode

equations are determined by the statistics of these physical imper-
fections.
We characterize the transmission performance of each of these
various systems by their complex transfer functions and the corre-
sponding impulse responses. We determine the complex transfer
function statistics as functions of the statistics of the coupling
coefficients, propagation parameters, or the Fresnel coefficients
appropriate to each case, and of the design parameters of the ideal
system. These results in turn determine the corresponding time-
domain statistics. The treatment is exclusively analytic; no Monte
Carlo or other simulation methods are employed. Computer us-
age is restricted to symbolic operations, to evaluation of analytical
expressions, and to creating plots.
The present text has the following goals:

1. Teaching the analytic methods.

2. Showing the different types of problems to which they may be
applied.
3. Application to problems of significant practical interest.
 REFERENCES 3

Matrix techniques—in particular Kronecker products and related

methods—play a central role in this work. Their application is nat-
ural for multi-layer devices. For continuous random coupling, as an
intermediate step the guide is divided into statistically independent
sections, each described by a wave matrix. In each case, this approach
yields results with clarity and generality.
The present work has evolved from several early papers by the au-
thor and his colleagues [1–4]. These earlier results were obtained by
pencil-and-paper analysis. The availability of computer programs ca-
pable of symbolic algebra, calculus, and matrix operations has greatly
expanded the scope of these methods. MAPLE [5] has been used ex-
tensively throughout the present work. To the best of the author’s
knowledge many of the present results are new.
The main text is restricted to the analysis of transmission statistics
of various classes of devices described above. A number of related
topics are relegated to the appendices.
The present methods have application beyond the random mode
coupling problems treated here. Kronecker products apply directly
to the statistical analysis of any cascaded system characterized by a
matrix product with random elements.

REFERENCES

1. Harrison E. Rowe and D. T. Young, “Transmission Distortion in Multi-

mode Random Waveguides,” IEEE Transactions on Microwave Theory
and Techniques, Vol. MTT-20, June 1972, pp. 349–365.
2. D. T. Young and Harrison E. Rowe, “Optimum Coupling for Random
Guides with Frequency-Dependent Coupling,” IEEE Transactions on Mi-
crowave Theory and Techniques, Vol. MTT-20, June 1972, pp. 365–372.
3. Harrison E. Rowe and Iris M. Mack, “Coupled Modes with Ran-
dom Propagation Constants,” Radio Science, Vol. 16, July–August 1981,
pp. 485–493.
4. Harrison E. Rowe, “Waves with Random Coupling and Random Propa-
gation Constants,” Applied Scientific Research, Vol. 41, 1984, pp. 237–255.
5. André Heck, Introduction to Maple, Springer-Verlag, New York, 1993.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

CHAPTER TWO

Coupled Line Equations

2.1. INTRODUCTION

Optical and microwave transmission media of diverse physical forms

have a common mathematical description in terms of the coupled
line equations. Examples include oversize hollow metallic waveguide
of various cross sections, dielectric waveguide, and optical fibers.
Different frequency bands are appropriate to these various media,
ranging from microwave and millimeter wave through optical fre-
quencies. Single guides are used to carry signals from one place to
another; pairs of similar guides comprise directional couplers.
An ideal guide, with constant geometry and material properties,
transmits a set of modes that propagate independently of each other.
A closed guide (e.g., a hollow metallic waveguide with perfectly con-
ducting walls) supports an infinite discrete set of modes, of which
a finite number are propagating; open structures (e.g., dielectric
waveguides or optical fibers) have a finite number of discrete propa-
gating modes plus a continuum of modes corresponding to the radi-
ation field. We shall be concerned only with the propagating modes
in the present work.
Random imperfections in these structures can arise from geomet-
ric or from material parameter departures from ideal design. Ge-
ometric imperfections include random straightness or cross-section
variations; material imperfections arise from undesired dielectric
constant, or index of refraction variations. Schelkunoff [1] observed
that fields in an imperfect waveguide could be expressed as a sum
over modes of the corresponding ideal waveguide. In the absence
5
6 COUPLED LINE EQUATIONS

of imperfections, the modes of an ideal guide are uncoupled, i.e.,

propagate independently; imperfections cause coupling between the
modes. Directional couplers require intentional coupling between
two guides.
The primary coupling in such structures occurs between modes
traveling in the same direction; the coupling between modes travel-
ing in opposite directions is normally small, and is neglected through-
out the present work.
The coupled line equations serve as a common description for
all of these media. The quantities in these equations that charac-
terize the various transmission systems are the propagation parame-
ters of and the coupling coefficients between the different propagat-
ing modes. These quantities may exhibit statistical variations arising
from the geometric and material imperfections of the physical sys-
tems. Our task in following chapters is to determine transmission
statistics in terms of coupling coefficient and/or propagation param-
eter statistics.
In this chapter, we examine the general properties and the de-
terministic solutions of the coupled line equations, that will be of
use throughout the statistical treatment in several following chap-
ters. We describe the two-mode and multi-mode cases separately;
the analytical methods used in calculating transmission statistics are
more clearly illustrated in the two-mode case, where expressions can
be written out explicitly. Additional modes introduce only additional
algebraic complexity, treated here by MAPLE without explicit pre-
sentation of intermediate results.

2.2. TWO-MODE COUPLED LINE EQUATIONS

The coupled line equations for two forward-traveling modes are

[2–10]

I0
z = −0 zI0 z + jczI1 z

(2.1)
I1
z = jczI0 z − 1 zI1 z

0 z = α0 z + jβ0 z 1 z = α1 z + jβ1 z (2.2)

The complex wave amplitude Ii z is proportional to the transverse

electric field of the ith mode at point z along the guide, normalized
 2.2. TWO-MODE COUPLED LINE EQUATIONS 7

such that the power carried in this mode is given by

Pi z = Ii z2  i = 0 1 (2.3)

0 z and 1 z represent the propagation parameters; the attenu-

ation α0 z α1 z and phase β0 z β1 z are real and positive, and
the coupling coefficient cz is real. The form of coupling coeffi-
cient in Equation (2.1) is appropriate for systems whose elements
dz possess geometric symmetry, e.g., guides with random straight-
ness deviation and directional couplers. The ’s and c are functions
of the geometric and material parameters of the device, and of the
frequency.
Let

Pz = P0 z + P1 z (2.4)

In the lossless case powers of different modes add, and Pz repre-
sents the total power in the guide. More generally, Pz is simply the
sum of the mode powers computed individually. Substituting Equa-
tion (2.3) into (2.4), differentiating, and substituting Equation (2.1)
for the resulting derivatives, we obtain
dPz
= −2α0 zP0 z − 2α1 zP1 z (2.5)
dz
This result states that each mode contributes to the decrease in
Pz in proportion to the product of its attenuation constant and
the power it carries. For the lossless case, α0 = α1 = 0, Equation
(2.5) yields conservation of power.
Equations (2.1)–(2.2) approximate the response of a multi-mode
guide in which only two forward-traveling modes are significant, and
other modes may be neglected. We denote the signal (desired) mode
by I0 and the spurious (undesired) mode by I1 . For the ideal guide
(without imperfections) 0 and 1 are constant, independent of z,
and cz = 0. Since both modes travel in the forward +z direction
α0 , α1 , β0 , and β1 are all positive. We assume that the signal mode
has lower loss; α0 ≤ α1 . Random imperfections are modeled by re-
garding cz 0 z and 1 z as stationary random processes with
appropriate statistics. As a particular example, a two-mode guide
with random two-dimensional straightness deviations has constant
propagation parameters 0 and 1 and a coupling coefficient in-
versely proportional to the radius of curvature Rz of the guide
8 COUPLED LINE EQUATIONS

axis,
C
cz =  (2.6)
Rz
the constant C being a function of the two modes, determined by
electromagnetic theory. In the special case of “single-mode” fiber,
I0 and I1 represent the two polarizations of the dominant mode and
0 = 1 .
The coupled line equations can describe a directional coupler
where backward modes are neglected. For this case, random im-
perfections are modeled by taking the propagation parameters to be
stationary random processes. As a simple example, cz = constant
and the propagation parameters have equal mean values, 1 z =
2 z .
Alternatively, Equations (2.1) and (2.2) can be applied to a trans-
mission medium with a reflected wave. Denoting the forward wave
by I0 and the backward wave by I1 , 0 = −1 = α + jβ with positive
α and β; c represents the reflection coefficient. However, we will not
use these relations in our later treatment of multi-layer devices, but
will rather employ direct methods.
Direct solution of the coupled line equations for arbitrary cz
and/or 0 z 1 z is not possible. The matrix methods we employ
to determine the transmission statistics for random guides require
the solutions described in the remainder of this chapter.

2.3. EXACT SOLUTIONS

First, consider two degenerate forward-traveling modes, i.e., with

equal propagation parameters:

0 z = 1 z ≡ z (2.7)

The solutions to Equation (2.1) are

 z z 



I0 z z  cos 0 cxdx j sin 0 cxdx  I0 0
− 0 xdx 
=e ·
I1 z  z z  I 0 
j sin cxdx cos cxdx 1
0 0

(2.8)
 2.3. EXACT SOLUTIONS 9

In this case, only the integrals of z and cz matter, their de-
tailed functional behavior being unimportant. These results are a
consequence of the fact that the two modes have identical attenu-
ation and phase parameters; it doesn’t matter how the coupling is
distributed.
Equation (2.8) yields directly the wave matrix for delta-function
coupling. Set

cz = cδz (2.9)

where c is the magnitude of the delta function. Then,



I0 0+ I0 0
=S·  (2.10)
I1 0+ I1 0

where the wave matrix for the discrete coupler is given by


cos c j sin c
S=  (2.11)
j sin c cos c

Comparison with Equation (2.6) shows that Equations (2.9)–(2.11)

yield the response of a discrete tilt of angle c/C radians.
Next, consider the coupled line equations with constant cou-
pling and constant propagation parameters, now called propagation
constants:

cz = c0 0 z = 0  1 z = 1  (2.12)

Equations (2.1) become linear equations with constant coefficients,

solved by elementary means [5]:



I0 z I0 0
= Tz ·  (2.13)
I1 z I1 0

0 +1
Tz = e− 2 z /K+ − K− 

√ √ √ √ 
−K− e/2z + K+ e−/2z e/2z − e−/2z
· √ √ √ √ 
e/2z − e−/2z K+ e/2z − K− e−/2z
(2.14)
10 COUPLED LINE EQUATIONS
√
1±
K± = −j K+ K− = −1 (2.15)
2c0 /

1 c /
= j 0√  (2.16)
K+ − K−

√
= 1 − 2c0 /2  (2.17)

 = 0 − 1  (2.18)
Let us approach a discrete coupler, Equation (2.9), by setting
c
c0 = (2.19)
z
with fixed c, and taking the limit as z → 0 (and c0 → ∞) in Equa-
tions (2.12)–(2.18). Then [6],

lim /2z 1 − 2c/z2 = jc (2.20)

z→0

lim K± = ±1 (2.21)

z→0

1 1
lim =  (2.22)
z→0 K+ − K− 2
0 +1
lim e− 2 z
= 1 (2.23)
z→0

Substituting Equations (2.20)–(2.23) into Equations (2.14)–(2.17),

we obtain the limiting form of T z:


cos c j sin c
lim Tz =  (2.24)
z→0 j sin c cos c

Equation (2.24) is identical to Equation (2.11). This result has

the following physical interpretation. As the coupling becomes larger
over a shorter length of guide, in such a way that the integrated cou-
pling remains constant, the length becomes so short that the differ-
ential attenuation and phase shift between the two modes becomes
negligible. Moreover, this argument applies not only for constant 0
and 1 as in Equation (2.12), but holds more generally for propa-
 2.4. DISCRETE APPROXIMATION 11

gation parameters with arbitrary functional dependence. Hence, the

result of Equations (2.11) and (2.24) holds in general for a discrete
coupler.

2.4. DISCRETE APPROXIMATION

Next, we determine a discrete approximation to a guide with con-

tinuous coupling [7]. This is necessary to apply matrix techniques in
the calculation of transmission statistics for random guides. Divide
the guide into sections z long. The discussion of Section 2.3 sug-
gests that for small enough z we can lump all the coupling in each
section at the end of the section; i.e., we may replace cz by cδ z
as follows:

cδ z = ck δz − kz (2.25)
k=1

where
 kz
ck = czdz (2.26)
k−1z

The kth section is illustrated in Figure 2.1, which shows cz and
cδ z; this approximation places all of the coupling at the end of the
section.
The corresponding matrix description becomes



−γ 

I0 kz cos ck j sin ck e 0k 0 I0 k − 1z
= · · 
I1 kz j sin ck cos ck 0 e−γ1k I1 k − 1z
(2.27)
where ck is given by Equation (2.26) and the γk ’s give the total
attenuation and phase shift in the length z:
 kz  kz  kz
γ0k = 0 zdz = α0 zdz + j β0 zdz
k−1z k−1z k−1z
   (2.28)
kz kz kz
γ1k = 1 zdz = α1 zdz + j β1 zdz
k−1z k−1z k−1z

Equations (2.26)–(2.28) will approximate the solutions to the cou-

pled line equations if the differential attenuation and phase shift are
12 COUPLED LINE EQUATIONS

k z
area = c(z) dz = ck
(k – 1) z
c(z)

c (z)

ck (z – k z)

(k – 1) z k z

FIGURE 2.1. Discrete approximation for a guide section.

small in any section of guide of length less than z:

 z 
 2 
 α0 z − α1 zdz   1
 z 
1
 z  0 < z2 − z1 ≤ z (2.29)
 2 
 β0 z − β1 zdz   2π
z1


If αz and βz are constant, these restrictions become

1 2π
z   z   (2.30)
α0 − α1 β0 − β1 
 2.5. PERTURBATION THEORY 13

Note that no small-coupling assumptions have been necessary

here. Subject to Equation (2.29) or (2.30), the matrix result of
Equations (2.26)–(2.28) will yield a good approximation to the
solutions of Equations (2.1) and (2.2).
We will subsequently need to divide the guide into sections z
that are approximately statistically independent. Equations (2.26)–
(2.28) are appropriate for this purpose when z satisfying Equation
(2.29) or (2.30) is long compared to the correlation length of the
random coupling or propagation parameters, i.e., for random pa-
rameters having white or almost white spectra.

2.5. PERTURBATION THEORY

For non-white coupling or propagation parameter spectra, z satis-

fying Equation (2.29) or (2.30) will be short compared to the cor-
relation length of the random parameters. In this case, Equations
(2.25)–(2.28) are replaced by results based on perturbation theory
to describe the response of a guide section long compared to the
correlation length, for small coupling [2–4, 6]. It is convenient to
normalize the complex wave amplitudes as follows:

I0 z = e−γ0 z G0 z

(2.31)
I1 z = e−γ1 z G1 z

where
 z  z
γ0 z = 0 xdx γ1 z = 1 xdx (2.32)
0 0

Then substituting into Equation (2.1) we find

G
0 z = jcze+γz G1 z

(2.33)
G
1 z = jcze−γz G0 z

where
 z
γz = xdx z = 0 z − 1 z (2.34)
0
14 COUPLED LINE EQUATIONS

Then the method of successive approximations given in Appendix A

yields the following approximate results:





G0 z G0 0 m00 m01 G0 0
≈M· = ·  (2.35)
G1 z G1 0 m10 m11 G1 0

where the elements of M are given as follows:

z x
m00 = 1 − cxe+γx dx cye−γy dy
0 0
 z
m01 = j cxe+γx dx
0
z
(2.36)
−γx
m10 = j cxe dx
0
 z  x
m11 = 1 − cxe−γx dx cye+γy dy
0 0

For constant propagation parameters,

0 z = 0  1 z = 1  z =  = 0 − 1  (2.37)

these results simplify by the substitution

γz → z (2.38)

In this case, two alternative forms for m00 and m11 are as follows:
z x
m00 = 1 − cxe+x dx cye−y dy
0 0
 z  z−ζ
=1− e+ζ dζ cxcx + ζdx
0 0

1 z z
=1− cxcye+x−y dxdy (2.39)
2 0 0
 z  x
m11 = 1 − cxe−x dx cye+y dy
0 0
 z  z−ζ
=1− e−ζ dζ cxcx + ζdx
0 0

1 z z
=1− cxcye−x−y dxdy (2.40)
2 0 0
 2.6. MULTI-MODE COUPLED LINE EQUATIONS 15

Each of these three expressions for m00 and for m11 has its uses. The
first yields a physical interpretation of these perturbation results. The
expected value of the middle expressions introduces the covariance
of the coupling coefficient for guides with random coupling, and
constant propagation parameters.
Equations (2.35)–(2.40) yield useful approximations for small cou-
pling. They are the first terms of infinite series given in Appendix A.
These series converge rapidly when
z
cxdx  1 (2.41)
0

2.6. MULTI-MODE COUPLED LINE EQUATIONS

The above relations for two modes are readily extended to many for-
ward modes [2–6] using matrix notation as follows. Equations (2.1)
and (2.2), the coupled line equations, become

I
z = −
z · Iz + jczC · Iz (2.42)

where
 
I T z = I0 z I1 z I2 z ···  (2.43)
 
0 z 0 0 ···
 
 0 1 z 0 ··· 
 

z =   (2.44)
 0 0 2 z ··· 
 
   
   
 
0 C01 C02 ···
 
 C01 0 C12 ··· 
 
C=  C = CT = C∗  (2.45)
 C02 C12 0 ··· 
 
   
   

The mode powers are

Pi z = Ii z2  (2.46)

16 COUPLED LINE EQUATIONS

Define

Pz = Pi z = I T z · I ∗ z (2.47)
i

Then,
dPz
= I T z · I
∗ z + I
T z · I ∗ z (2.48)
dz
Substituting Equation (2.42) into (2.48), the multi-mode generaliza-
tion of Equation (2.5) becomes
dPz 
= − I T z · 
z +
∗ z · I ∗ z = − 2 αi zPi z (2.49)
dz i

For the lossless case all αi = 0 and Equation (2.49) yields conser-
vation of power.
Exact solutions corresponding to those of Section 2.3 for two
modes are readily found for the multi-mode case. In the degenerate
case, all modes have identical propagation constants,

z = 0 z = 1 z = 2 z = · · · (2.50)

i.e., in matrix notation

z = zI  (2.51)

where I represents the unit matrix:

 
1 0 0 ···
 
0 1 0 ··· 
 
I =  (2.52)
0 0 1 ··· 
 
   
   

The solution to Equation (2.42) for this case is

z z
Iz = e− 0 xdx j
e 0 cxdx·C
· I0 (2.53)

For delta-function coupling, set

cz = cδz (2.54)

 2.6. MULTI-MODE COUPLED LINE EQUATIONS 17

in Equation (2.53). Then,

I0+ = S · I0 (2.55)

where the wave matrix for the multi-mode discrete coupler is given
by

S = ejcC  (2.56)

Finally, for constant coupling and constant propagation parame-

ters Equation (2.42) becomes

I
z = −
· Iz + jc0 C · Iz (2.57)

where
is the constant matrix

 
0 0 0 ···
 
 0 1 0 ··· 
 

=  (2.58)
 0 0 2 ··· 
 
   
   

The solution to Equations (2.57) and (2.58) in matrix notation is

Iz = e−
+jc0 Cz · I0 (2.59)

The exact multi-mode solutions given in Equations (2.50)–(2.59)

specialize to those for the two-mode case of Equations (2.7)–(2.24)
by evaluating the matrix exponentials. We perform this reduction for
the discrete coupler. Set


0 c
cC = (2.60)
c 0

in Equation (2.56). Diagonalizing this matrix,




−1
0 c 1 1 c 0 1 1
= · ·
c 0 1 −1 0 −c 1 −1




1 1 1 c 0 1 1
= · ·  (2.61)
2 1 −1 0 −c 1 −1
18 COUPLED LINE EQUATIONS

Then Equation (2.56) yields


jc 


1 1 1 e 0 1 1 cos c j sin c
ejcC = · · = 
2 1 −1 0 e−jc 1 −1 j sin c cos c
(2.62)

in agreement with Equation (2.11).

The discrete approximation of Section 2.4, used subsequently for
white or almost white coupling cz, generalizes to the multi-mode
case as follows:

Ikz = ejck C · e−k z · Ik − 1z (2.63)

where as in Equation (2.26),

 kz
ck = czdz (2.64)
k−1z

and
 kz
k =
zdz (2.65)
k−1z

The restrictions of Equation (2.29) generalize to

 z 
 2 
 αi z − αj zdz   1
 z 
1
 z  0 < z2 − z1 ≤ z i j = 0 1 2 · · · 
 2 
 βi z − βj zdz   2π
 z1

(2.66)

Finally, for multi-mode perturbation theory

Iz = e−z · Gz (2.67)

where
 z
z =
xdx (2.68)
0

Substituting into Equation (2.42), we find

G
z = jcz · e+z · C · e−z · Gz (2.69)

 REFERENCES 19

as the generalization of Equation (2.33). The method of successive

approximations in Appendix A yields the following:
 z
Gz ≈ I + j cxe+x · C · e−x dx
0
 z
− cxe+x · C · e−x dx
0
x 
· cye+y · C · e−y dy · G0 (2.70)
0

This result holds for small coupling, i.e., from Equations (A.22) and
(A.24)
z
cxdxC  1 (2.71)
0

where the matrix norm C is given by Equation (A.25). Equa-

tion (2.70) is required for the non-white case, where the correlation
lengths of cz or of the
i z are long compared to z satisfying
Equation (2.66).
Finally, substituting


0 cz
czC = (2.72)
cz 0

and
 z 
 0 xdx 0 
z =  
0
  z  (2.73)
0 1 xdx
0

into Equation (2.70) specializes this result to Equations (2.35) and

(2.36) for the two-mode case.

REFERENCES

1. S. A. Schelkunoff, “Conversion of Maxwell’s Equations into General-

ized Telegraphist’s Equations,” Bell System Technical Journal, Vol. 34,
September 1955, pp. 995–1043.
20 COUPLED LINE EQUATIONS

2. Huang Hung-chia, Coupled Mode Theory, VNU Science Press, Utrecht,

The Netherlands, 1984.
3. Dietrich Marcuse, Light Transmission Optics, 2nd ed., Robert E.
Krieger, Malabar, FL, 1989.
4. Dietrich Marcuse, Theory of Dielectric Optical Waveguides, 2nd ed., Aca-
demic Press, New York, 1991.
5. S. E. Miller, “Coupled Wave Theory and Waveguide Applications,” Bell
System Technical Journal, Vol. 33, May 1954, pp. 661–719.
6. H. E. Rowe and W. D. Warters, “Transmission in Multimode Waveguide
with Random Imperfections,” Bell System Technical Journal, Vol. 41,
May 1962, pp. 1031–1170.
7. Harrison E. Rowe and D. T. Young, “Transmission Distortion in Multi-
mode Random Waveguides,” IEEE Transactions on Microwave Theory
and Techniques, Vol. MTT-20, June 1972, pp. 349–365.
8. Harrison E. Rowe, “Waves with Random Coupling and Random Prop-
agation Constants,” Applied Scientific Research, Vol. 41, 1984, pp. 237–
255.
9. Hermann A. Haus and Weiping Huang, “Coupled-Mode Theory,” Pro-
ceedings of the IEEE, Vol. 79, October 1991, pp. 1505–1518.
10. Wei-Ping Huang, “Coupled-mode theory for optical waveguides: an
overview,” Journal of the Optical Society of America A, Vol. 11, March
1994, pp. 963–983.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

CHAPTER THREE

Guides with White Random

Coupling

3.1. INTRODUCTION

We consider the coupled line equations of Chapter 2 with white cou-

pling coefficient and constant propagation parameters in the present
chapter. This case is significant in that exact results for transmission
statistics are obtained. Assume the coupling coefficient c
z of Chap-
ter 2 is a zero-mean stationary random process with delta-function
covariance

Rc
ζ = c
z + ζc
z = S0 δ
ζ (3.1)

and white spectrum

∞
S
ν = Rc
ζe−j2πνζ dζ = S0 (3.2)
−∞

with spectral density S0 . We require the following additional property

for every pair of nonoverlapping intervals:

z2
z4
c
xdx and c
xdx are statistically independent for
z1 z3

z1 < z2 < z3 < z4 (3.3)

i.e., different sections of guide have statistically independent cou-

pling. In some cases, e.g., Gaussian or Poisson c
z, Equations (3.1)
21
22 GUIDES WITH WHITE RANDOM COUPLING

and (3.2) imply Equation (3.3). We require Equation (3.3) in any

case; it implies that c
z has a white spectrum.
For constant propagation parameters, Equations (2.1) and (2.2)
for two modes become

I0
z = −0 I0
z + jc
zI1
z

(3.4)
I1
z = jc
zI0
z − 1 I1
z

0 = α0 + jβ0 1 = α1 + jβ1  (3.5)

0 and 1 are independent of z. The normalized relations of Equa-

tions (2.31)–(2.34) become

I0
z = e−0 z G0
z

(3.6)
I1
z = e−1 z G1
z

G0
z = jc
ze+z G1
z

(3.7)
G1
z = jc
ze−z G0
z

 = 0 − 1 = α + jβ
(3.8)
α = α0 − α1 β = β0 − β1 

The corresponding results for the multi-mode case are obtained by

substituting Equation (2.58) into Equations (2.42) and (2.67)–(2.69):

I
z = −
· I
z + jc
zC · I
z (3.9)

 
0 0 0 ···
 
 0 1 0 ··· 
 

=  (3.10)
 0 0 2 ··· 
 
   
   

I
z = e−
z · G
z (3.11)

G
z = jc
z · e+
z · C · e−
z · G
z (3.12)

 3.2. NOTATION—TWO-MODE CASE 23
 
0 C01 C02 ···
 
 C01 0 C12 ··· 
 
C=  C = CT = C∗  (3.13)
 C02 C12 0 ··· 
 
   
   

While the i are independent of z, they are frequency dependent;

the coupling between modes, a random function of z, may also be
frequency dependent. The Gi are thus functions of z and of fre-
quency f . The frequency dependence is implicit in the above equa-
tions, and in subsequent solutions for single-frequency transmission
statistics, such as average mode powers. However, additional nota-
tion is required to display the frequency dependence explicitly in the
analysis of transfer-function frequency- and time-response statistics.

3.2. NOTATION—TWO-MODE CASE

We consider the solution to Equation (3.7) for the normalized sig-

nal transfer function G0 for some fixed length z = L, with initial
conditions

G0
0 = 1 G1
0 = 0 (3.14)

A unit sinusoidal signal is input at z = 0; the spurious mode is zero

at the input. G0
L and G1
L are the normalized signal and signal-
spurious mode transfer functions, respectively.
It is natural to suppress the L dependence, and instead display the
frequency dependence. We first separate the z and f dependence of
the coupling. The coupling coefficient c
z is proportional to some
geometric parameter, which we denote by d
z. For the two-mode
case,

c
z = Cd
z (3.15)

d
z does not depend on f , but is a function only of z; any fre-

quency dependence is contained in the parameter C. In the example
of Equation (2.6), d
z is the curvature of the guide axis; in other ap-
plications it might be ellipticity, etc. For the multi-mode case, we may
24 GUIDES WITH WHITE RANDOM COUPLING

absorb the frequency dependence into the elements of the matrix C

of Equation (3.13), and regard c
z as the frequency-independent
geometric parameter.
Let us now consider the solution to Equation (3.7) for the nor-
malized signal transfer function G0 , for some fixed length L and a
given geometric parameter d
z. G0 is a function of α, β, and C,
and through these parameters a function of frequency f . However,
the principal frequency dependence will normally occur through β.
This suggests investigating the properties of G0 as a function of β,
with α and C regarded as fixed. Toward this end, we define the
Fourier transform of G0 as follows [1]:

∞ 
−j2πτ
βL
2π 
βL
gα
τ = G0
α βe d
−∞ 2π
1
∞
= G
α βe−jτβL d
βL (3.16)
2π −∞ 0
The inverse transform is

∞
G0
α β = gα
τe+jτβL dτ (3.17)
−∞

The fixed parameters L, d
z, and C are suppressed in these two

relations.
Consider an idealized guide in which α and C are frequency
independent and β is strictly linear with f . In a real guide α is
an even function of f , β and C odd. Therefore for the idealized
guide C must satisfy

+C0 f >0
C = C0 sgn f = C0 a positive constant. (3.18)
−C0 f <0

The guide is dispersionless:


1 1
β = −2πf − v0 > v1 (3.19)
v1 v0
where v0 and v1 are the signal and spurious mode velocities, respec-
tively. Let

1 1
T = − L (3.20)
v1 v0
 3.3. AVERAGE TRANSFER FUNCTIONS 25

be the differential delay between signal and spurious mode transit

time over length L. Then substituting in Equation (3.16),

∞
gα
τ = T G0
α βej2πfτ·T df (3.21)
−∞

Denote the signal mode impulse response by ɡα
t:

∞
ɡα
t = G0
α βej2πft df (3.22)
−∞

Comparison with Equation (3.21) shows that


1 t
ɡα
t = gα  (3.23)
T T

gα
τ of Equations (3.16) and (3.17) is therefore the normalized im-
pulse response of the idealized dispersionless guide with frequency-
independent attenuation constants and coupling.
While Equation (3.23) is strictly true only for an idealized guide,
it will provide a useful approximation for a physical guide in many
cases. This approximation requires the frequency variation of α,
C, and of the group delay of the two modes to be small over the
frequency range of interest.
The general properties of G0
α β and gα
τ are described in
Appendix B. These properties are reflected in the statistical results
to follow.

3.3. AVERAGE TRANSFER FUNCTIONS

For the two-mode case with constant propagation parameters, Equa-

tion (2.27) becomes

I0
kz e−0 z cos ck e−1 z j sin ck I0
k − 1z

= ·
I1
kz e−0 z j sin ck e−1 z cos ck I1
k − 1z
(3.24)

where ck is given by Equation (2.26) and illustrated in Figure 2.1.

This approximation requires that the restrictions of Equation (2.30)
26 GUIDES WITH WHITE RANDOM COUPLING

are satisfied:
1 2π
z  z   (3.25)
α β
By Equations (3.1) and (3.2), c
z is white with zero mean; by Equa-
tion (3.3), the different sections remain independent for any z.
Therefore Equation (3.24) may be identified with Equation (D.8).
Define the average complex mode amplitudes as
I i
z = Ii
z G i
z = Gi
z (3.26)
by Equation (D.14), the expected value of Equation 3.24 yields
I 0
kz e−0 z cos ck  e−1 z jsin ck  I 0
k − 1z
= · 
I 1
kz e−0 z jsin ck  e−1 z cos ck  I 1
k − 1z
(3.27)
For sufficiently small z,1
1 1
cos ck  ≈ 1 − ck2  = 1 − S0 z
2 2 (3.28)
sin ck  ≈ ck  = 0
Substituting Equation (3.28) into Equation (3.27) and taking the
limit as z → 0,

S
I 0
z = − 0 + 0 I 0
z
2
 (3.29)
S0
I 1
z = − 1 + I 1
z
2
1
From Equation (2.26), since c
z has zero mean,

kz
ck  = c
z dz = 0

k−1z

Using Equations (2.26) and(3.1),

kz
kz
kz
kz
ck2  = c
zc
z  dzdz = S0 δ
z − z  dzdz

k−1z
k−1z
k−1z
k−1z

kz
= S0 dz = S0 z

k−1z

These results are used throughout Chapter 3.

 3.3. AVERAGE TRANSFER FUNCTIONS 27

Thus, the expected responses are [1]

S0
I 0
z = I0
z = e−0 z e− 2 z I0
0
S0
(3.30)
I 1
z = I1
z = e−1 z e− 2 z I 1
0

where I0
0 and I1
0 are, respectively, the sinusoidal signal and
spurious mode inputs. By Equation (3.6), the expected normalized
responses are
S0
G 0
z = G0
z = e− 2 z G0
0
S0
(3.31)
− z
G 1
z = G1
z = e 2 G
0
1

The expected complex wave amplitudes in the two-mode case de-

cay exponentially, and are uncoupled. The larger the white coupling
spectral density S0 , the more rapid the decay. For unit sinusoidal
signal input and zero spurious mode input,
I0
0 = G0
0 = 1 I1
0 = G1
0 = 0 (3.32)
the expected responses are
S0
I 0
z = I0
z = e−0 z e− 2 z I 1
z = I1
z = 0
S0
(3.33)
G 0
z = G0
z = e− 2 z G 1
z = G1
z = 0
Here the spurious mode has expected complex amplitude identically
zero for all z. Of course, zero expected value tells nothing about
higher order statistics; a complex vector with unit magnitude and
random uniform phase has zero average value.
The multi-mode case exhibits qualitative differences. From Equa-
tions (2.63), (2.58), and (2.45),
I
kz = ejck C · e−
z · I
k − 1z  (3.34)
 
0 C01 C02 ···
 
 C01 0 C12 ··· 
 
C=  C = CT = C∗  (3.35)
 C02 C12 0 ··· 
 
   
   
28 GUIDES WITH WHITE RANDOM COUPLING
 
0 0 0 ···
 
 0 1 0 ··· 
 

=  (3.36)
 0 0 2 ··· 
 
   
   

Extending the two-mode analysis above,


kz = ejck C · e−
z  · 
k − 1 z (3.37)

where from Equation 3.26


 T
z = I 0
z I 1
z · · · = I T
z

= I0
z I1
z · · ·  (3.38)

For small z,

1
ejck C ≈ I + jck C − ck2 C2
2 (3.39)
−
z
e ≈ I −
z

where I is the unit matrix. The first factor on the right-hand side of
Equation (3.37) becomes

1 1
ejck C · e−
z  ≈ I −
z − ck2 C2 = I −
z − S0 C2 z (3.40)
2 2

Substituting Equation (3.40) into Equation (3.37) and taking the

limit as z → 0, the multi-mode extension of Equation (3.29)
becomes

1

z = −
− S0 C2 · 
z 
z = I
z (3.41)
2

The substitution

0 1
C= C2 = I (3.42)
1 0
 3.3. AVERAGE TRANSFER FUNCTIONS 29

reduces Equation (3.41) to Equation (3.29) for the two-mode

case.
It is instructive to write explicit results for the three-mode case:
 
0 C01 C02
 
C=
 C01 0 C12 

C02 C12 0
  (3.43)
2 2
C01 + C02 C02 C12 C01 C12
 
C2 = 
 C02 C12
2
C01 2
+ C12 C01 C02 
2 2
C01 C12 C01 C02 C02 + C12

The differential equations for the expected complex mode ampli-

tudes are

1 2 2
I 0
z = − 0 + S0
C01 + C02  I 0
z
2
1 1
− S0 C02 C12 · I 1
z − S0 C01 C12 · I 2
z (3.44)
2 2


1 1 2 2
I 1
z = − S0 C02 C12 · I 0
z − 1 + S0
C01 + C12  I 1
z
2 2
1
− S0 C01 C02 · I 2
z (3.45)
2

1 1
I 2
z = − S0 C01 C12 · I 0
z − S0 C01 C02 · I 1
z
2 2

1 2 2
− 2 + S0
C02 + C12  I 2
z (3.46)
2

The expected complex mode amplitudes are no longer uncoupled.

Even for zero spurious mode inputs

I0
0 = 1 I1
0 = I2
0 = 0 (3.47)

30 GUIDES WITH WHITE RANDOM COUPLING

the spurious mode averages I 1
z = I1
z and I 2
z = I2
z
depart from zero for a coherent input signal.
We use coherent inputs to the coupled line equations to study
transfer function statistics, here the expected transfer functions.
For a coherent sinusoidal input with unit power to mode i we set
Ii
0 = 1, a deterministic quantity representing a complex wave ejωt .
Sources such as light-emitting diodes (LEDs) produce incoherent or
partially coherent inputs to multi-mode optical fibers. We model an
incoherent sinusoidal input of unit power with constant amplitude
and random phase as Ii
0 = ejθ , with θ a random variable uni-
formly distributed from 0 < θ < 2π, representing the complex wave
ejωt ejθ ; here the expected input is zero, I0
0 = 0 The present
results, Equations (3.29), (3.41), or (3.44)–(3.46), show that all ex-
pected mode amplitudes are zero for incoherent inputs; for coherent
or partially coherent inputs the expected mode amplitudes depart
from zero.
The present results simplify in the nondegenerate case, where no
two modes have the same propagation constants. We show in Section
4.6.1 that for large βij = βi − βj the coupling between expected
mode amplitudes approaches zero, as in the two-mode case.
The expected complex transfer functions give little information
about the transmission behavior of guides with random coupling. Ex-
act results for higher order statistics—average powers, power fluctua-
tions, transfer function vs. frequency statistics—are calculated below
using Kronecker products.

3.4. COUPLED POWER EQUATIONS

For two modes with constant propagation parameters, take the Kro-
necker product of Equation (3.24) with its complex conjugate, to
yield by Equation (C.7)
   
I0
kz2 I0
k − 1z 2
   
 I0
kzI1∗
kz   I
k − 1z I1∗
k − 1z 
  =
M ⊗ M  ·  0
∗ 
 ∗   ∗ 
 I1
kzI0
kz   I1
k − 1z I0
k − 1z 
I1
kz2 I1
k − 1z 2
(3.48)
 3.4. COUPLED POWER EQUATIONS 31

where M is the 2 × 2 matrix on the right-hand side of Equation

(3.24), and the elements of M ⊗ M∗ are given as follows:
∗

M ⊗ M∗ 11 = e−0 z cos ck · e−0 z cos ck
∗

M ⊗ M∗ 12 = −e−0 z cos ck · e−1 z j sin ck
∗

M ⊗ M∗ 13 = e−1 z j sin ck · e−0 z cos ck
∗

M ⊗ M∗ 14 = −e−1 z j sin ck · e−1 z j sin ck
∗

M ⊗ M∗ 21 = −e−0 z cos ck · e−0 z j sin ck
∗

M ⊗ M∗ 22 = e−0 z cos ck · e−1 z cos ck
∗

M ⊗ M∗ 23 = −e−1 z j sin ck · e−0 z j sin ck
∗

M ⊗ M∗ 24 = e−1 z j sin ck · e−1 z cos ck
∗
(3.49)

M ⊗ M∗ 31 = e−0 z j sin ck · e−0 z cos ck
∗

M ⊗ M∗ 32 = −e−0 z j sin ck · e−1 z j sin ck
∗

M ⊗ M∗ 33 = e−1 z cos ck · e−0 z cos ck
∗

M ⊗ M∗ 34 = −e−1 z cos ck · e−1 z j sin ck
∗

M ⊗ M∗ 41 = −e−0 z j sin ck · e−0 z j sin ck
∗

M ⊗ M∗ 42 = e−0 z j sin ck · e−1 z cos ck
∗

M ⊗ M∗ 43 = −e−1 z cos ck · e−0 z j sin ck
∗

M ⊗ M∗ 44 = e−1 z cos ck · e−1 z cos ck 

As before, the restrictions of Equation (3.25) must be satisfied.

For white coupling, different sections are independent for any z,
and the expected value of Equation (3.48) may be identified with
Equation (D.17). Define the average powers and cross-powers as
follows:2
P i
z = Pi
z = Ii
z2 
(3.50)
P ij
z = P ∗ji
z = Pij
z Pij
z = Ii
zIj∗
z
2
Pi
z is defined in Equations (2.3) and (2.46).
32 GUIDES WITH WHITE RANDOM COUPLING

Taking expected values in Equations (3.48) and (3.49), substituting

Equation (3.50), evaluating
cos2 ck  ≈ 1 − ck2  = 1 − S0 z
sin2 ck  ≈ ck2  = S0 z (3.51)
sin ck cos ck  ≈ ck  = 0
as in Equation (3.28), and taking the limit as z → 0,
 
P 0
z
 
 P 01
z 
 
 
P
 10
z 
P 1
z
 
−
2α0 + S0  0 0 S0
 
 0 −
0 + ∗1 + S0  S0 0 
=


 0 S0 −
∗0 + 1 + S0  0 
S0 0 0 −
2α1 + S0 
 
P 0
z
 
 P 01
z 

· 
352

P
 10
z 
P 1
z
Equation (3.52) separates into two sets of differential equations. The
average mode powers are [1, 2]:
P 0
z = −
2α0 + S0 P 0
z + S0 P 1
z
(3.53)
P 1
z = S0 P 0
z −
2α1 + S0 P 1
z
The cross-powers are [1]
P 01
z = −
0 + ∗1 + S0 P 01
z + S0 P 10
z
(3.54)
P 10
z = S0 P 01
z −
∗0 + 1 + S0 P 10
z
The two relations of Equation (3.54) are complex conjugates of each
other.
 3.4. COUPLED POWER EQUATIONS 33

The powers and cross-powers are decoupled only in the two-mode

case. In general, the expected value of the Kronecker product of
Equation (3.34) with its complex conjugate yields
I
kz ⊗ I ∗
kz = M ⊗ M∗  · I
k − 1z ⊗ I ∗
k − 1z 
(3.55)
where
M = ejck C · e−
z  (3.56)
For N modes, define the Kronecker product of Equation 2.43 with
its complex conjugate as
 T
z = P T
z PT
z = I T
z ⊗ I T∗
z
 T
z = P 0 P 01 · · · P 0 N−1 P 10 P 1 · · · P 1 N−1 ···

··· P N−1 0 P N−1 1 · · · P N−1 (3.57)
where the P’s represent the average powers and cross-powers de-
fined in Equation (3.50). Each of these elements is a function of
z, i.e., P
z; this dependence has been suppressed for compactness.
Then, Equation (3.55) becomes

kz = M ⊗ M∗  · 
k − 1z  (3.58)
From Equation (3.56),
∗
M ⊗ M∗ =
ejck C · e−
z  ⊗
e−jck C · e−
z 
∗
=
ejck C ⊗ e−jck C  ·
e−
z ⊗ e−
z  (3.59)
Substituting Equation (3.39) and taking expected values, for small z

M ⊗ M∗  ≈ 
I + jck C − 21 ck2 C2  ⊗
I − jck C − 21 ck2 C2 

·
I −
z ⊗
I −
∗ z
= I ⊗ I + S0 z
− 21 I ⊗ C2 + C ⊗ C − 21 C2 ⊗ I 
· I ⊗ I − z
I ⊗
∗ +
⊗ I 
= I ⊗ I + zS0
− 21 I ⊗ C2 + C ⊗ C − 21 C2 ⊗ I 
− I ⊗
∗ −
⊗ I 
(3.60)
34 GUIDES WITH WHITE RANDOM COUPLING

Substituting Equation (3.60) into Equation (3.58) and taking the

limit as z → 0, the multi-mode generalization of Equations (3.52)
to (3.54) is


z = S0
C ⊗ C − 21 I ⊗ C2 − 21 C2 ⊗ I  − I ⊗
∗ −
⊗ I  · 
z

(3.61)

where C and
are given by Equations (3.35) and (3.36) and I is the
N × N unit matrix. Equation (3.61) reduces to the two-mode case
by the substitution of Equation (3.42).
Even the two-mode case involves a disagreeable amount of hand
calculation, requiring the expected values of 16 terms of a 4 × 4
matrix. It is evident that greater numbers of modes require the use
of a computer algebra system capable of programmable symbolic
and numeric algebraic, matrix, and calculus operations. The N-mode
coupled power equations of Equation (3.61) comprise a set of N 2
differential equations; N of these are real, N 2 − N occur in complex
conjugate pairs.
The three-mode case of Equation (3.43) leads to a 9 × 9 matrix
with 81 terms, yielding 9 differential equations relating the average
powers and cross powers, each with 9 terms on the right-hand side.
These have been obtained with MAPLE; the first two of these 9
equations are given below:
2 2
P 0
z = − S0
C01 + C02  + 0 + ∗0 P 0
z − 5S0 C02 C12 P 01
z
2
− 5S0 C01 C12 P 02
z − 5S0 C02 C12 P 10
z + S0 C01 P 1
z
+ S0 C01 C02 P 12
z − 5S0 C01 C12 P 20
z + S0 C01 C02 P 21
z
2
+ S0 C02 P 2
z
P 01
z = −5S0 C02 C12 P 0
z
2 2 2
− S0
C01 + 5C12 + 5C02  + 0 + ∗1 P 01
z
2
− 5S0 C01 C02 P 02
z + S0 C01 P 10
z − 5S0 C02 C12 P 1
z
+ S0 C01 C12 P 12
z + S0 C01 C02 P 20
z − 5S0 C01 C12 P 21
z
+ S0 C02 C12 P 2
z
··· ··· ···················································
(3.62)
 3.5. POWER FLUCTUATIONS 35

It is unlikely that anyone will wish to deal with even three modes
by hand. Subsequent examples given in Chapters 4, 6, 7, and 8
demonstrate that MAPLE is capable of treating significant multi-
mode problems.
The coupled power equations of this section apply equally well
to coherent, incoherent, or partially coherent inputs; their solutions
yield the mode powers as a function of distance z along the guide in
terms of the input mode powers, whatever the state of coherence at
the input. For coherent or partially coherent inputs, the modes will
be correlated and will contain coherent components given by the
results of Section 3.3; but the total mode powers and cross-powers
depend only on their input values.
Strict decoupling of the average powers and cross-powers occurs
only in the two-mode case. However, if all propagation constants
are distinct (i.e., no degenerate modes), the powers and the cross-
powers are approximately decoupled. In this approximation, the av-
erage mode powers are described by N differential equations [2],
rather than the N 2 differential equations of Equation (3.61). Fur-
ther discussion is given in Section 4.6.2.

3.5. POWER FLUCTUATIONS

Higher order statistics can be readily obtained in a similar way. As

an example we calculate the power fluctuations for white coupling.
Define

Pijk%
z = Ii
zIj∗
zIk
zI%∗
z (3.63)

The average of this quantity is the fourth-order moment

P ijk%
z = Pijk%
z = Ii
zIj∗
zIk
zI%∗
z (3.64)

Powers and cross-powers previously defined in Equations (3.50),

(2.3), and (2.46) occur in some of these quantities. Of particular
present interest

Piiii
z = Pi2
z (3.65)

36 GUIDES WITH WHITE RANDOM COUPLING

is the square of the power in the ith mode,

P iiii
z = Piiii
z = Pi2
z (3.66)

is the second moment of the ith mode power, and

P iijj
z = Pi
zPj
z P ijij
z = Pij2
z

 2 (3.67)
P ijji
z = Pij
z  = Pi
zPj
z i = j

Define the mean-square power fluctuation in the ith mode as

2
 Pi
z 2  = Pi2
z − Pi
z = P iiii
z − P 2i
z (3.68)

P i
z was found in Section 3.4; we determine P iiii
z in this section,
and through Equation (3.68), the power fluctuation. From Equation
(3.34),

I
kz ⊗ I ∗
kz ⊗ I
kz ⊗ I ∗
kz

= ejck C ⊗ e−jck C ⊗ ejck C ⊗ e−jck C 
∗ ∗
·
e−
z ⊗ e−
z ⊗ e−
z ⊗ e−
z 
· I
k − 1z ⊗ I ∗
k − 1z
⊗ I
k − 1z ⊗ I ∗
k − 1z  (3.69)

Define

 T4
z = I T
z ⊗ I T∗
z ⊗ I T
z ⊗ I T∗ 

= P 0000
z P 0001
z · · · P ijk%
z · · · P N−1 N−1 N−1 N−1
z
(3.70)

where the subscripts, mod(N), are arranged in numerical order. Sub-

stituting Equation (3.70) into Equation (3.69),

 4
kz = ejck C ⊗ e−jck C ⊗ ejck C ⊗ e−jck C 

∗ ∗
·
e−
z ⊗ e−
z ⊗ e−
z ⊗ e−
z 
·  4
k − 1z  (3.71)
 3.5. POWER FLUCTUATIONS 37

Substituting Equation (3.39) and taking expected values, the limit as

z → 0 yields the following coupled differential equations for the
fourth moments, for N modes:

 4
z = S0 I ⊗ I ⊗ C ⊗ C − I ⊗ C ⊗ I ⊗ C + I ⊗ C ⊗ C ⊗ I

+C⊗I ⊗I ⊗C−C⊗I ⊗C⊗I +C⊗C⊗I ⊗I

1
−
I ⊗ I ⊗ I ⊗ C2 + I ⊗ I ⊗ C2 ⊗ I
2

2 2
+ I ⊗ C ⊗ I ⊗ I + C ⊗ I ⊗ I ⊗ I

− I ⊗ I ⊗ I ⊗
∗ − I ⊗ I ⊗
⊗ I

∗
− I ⊗
⊗ I ⊗ I −
⊗ I ⊗ I ⊗ I ·  4
z (3.72)

where I is the N × N unit matrix, and C and
are given by Equa-

tions (3.35) and (3.36).
Equation (3.72) is exact; it represents N 4 first-order differential
equations for the fourth moments, each with N 4 right-hand-side
terms. However, eliminating redundant equations yields a significant
reduction in the number of equations. By elementary combinatorics,
for N ≥ 4 eliminating identical equations reduces the number from
N 4 to N 4 /4 + N 3 /2 + N 2 /4. Further reductions occur in special cases,
illustrated below for N = 2 modes.
To determine the power fluctuations in the two-mode case we sub-
stitute Equation (3.42) into the right-hand side of Equation (3.72),
obtaining 16 simultaneous first-order differential equations. The re-
sulting matrix has been evaluated by MAPLE; significant simplifica-
tions occur. The 16 differential equations separate into two sets of 8
equations each, one set containing only the functions of Equations
(3.66) and (3.67) with i j = 0 1, the other containing mixed mo-
ments, i.e., quantities of Equation (3.64) with more than two distinct
subscripts. The first set, of current interest, contains 5 independent
equations. Define

QT
z = P 0000
z P 0011
z P 0101
z P 1010
z P 1111
z
(3.73)
38 GUIDES WITH WHITE RANDOM COUPLING

where from Equations (3.66) and (3.67)

P 0000
z = I0
zI0∗
zI0
zI0∗
z = P02
z

P 0011
z = I0
zI0∗
zI1
zI1∗
z = P0
zP1
z
P 0101
z = I0
zI1∗
zI0
zI1∗
z = P01
2

z (3.74)
P 1010
z = I1
zI0∗
zI1
zI0∗
z = P10
2 2

z = P01
z∗
P 1111
z = I1
zI1∗
zI1
zI1∗
z = P12
z

Then the fourth moments of interest are given by the differential

equations

Q
z = S0 N1 − N2  · Q
z (3.75)

where S0 , the coupling spectral density, is given in Equations (3.1)

and (3.2), and the matrices N1 and N2 are given as follows:
 
−2 4 −1 −1 0
 
 1 −4 1 1 1
 
 
N1 =  −1 4 −2 0 −1  (3.76)
 
 −1 4 0 −2 −1 
 
0 4 −1 −1 −2

 
−4α0 0 0 0 0
 
 0 −2α+ 0 0 0 
 
 
N2 =  0 0 −2
α+ + jβ 0 0 
 
 0 0 0 −2
α+ − jβ 0 
 
0 0 0 0 −4α1
(3.77)

where

α+ = α0 + α1 β = β0 − β1  (3.78)
 3.6. TRANSFER FUNCTION STATISTICS 39

The present calculations for power fluctuations with white cou-

pling are exact. Cross-powers and higher mixed moments are
included without approximations or assumptions of any kind. For
two modes the calculation for power fluctuations includes the cross-
correlation of the mode powers P0
zP1
z [2], as well as the
2
square of the cross-power P01
z. Finally, further simplification
occurs in the non-degenerate case, described in Section 4.6.3.

3.6. TRANSFER FUNCTION STATISTICS

We have so far considered only single-frequency transmission statis-

tics, in Sections 3.3, 3.4, and 3.5. Multi-frequency statistics are nec-
essary in order to describe signal distortion. We therefore consider
the second-order gain-frequency statistics for a two-mode guide. We
adopt the notation of Section 3.2, with initial conditions given by
Equation (3.14); G0
z the normalized signal transfer function, is
a function of β. It proves convenient to define the following new
quantity:

G01
z = e0 z I1
z G01
0 = 0 (3.79)

G01
z is the normalized signal-spurious mode transfer function, also

a function of β. Then, Equation (3.24) becomes

G0
kz cos ck e·z j sin ck

=
G01
kz j sin ck e·z cos ck
G0
k − 1z
·  (3.80)
G01
k − 1z

We rewrite this relation as

G0
β cos ck e·z j sin ck

=
G01
β kz
j sin ck e·z cos ck
G0
β
· (3.81)
G01
β
k−1z
40 GUIDES WITH WHITE RANDOM COUPLING

in which the z dependence of the column vectors is indicated by

subscripts, and the α dependence is implicit.
We assume the frequency dependence occurs through β, and
neglect the frequency dependence of ck and α, following the dis-
cussion in Section 3.2. Set β → β + σ in Equation (3.81):

G0
β + σ cos ck ±e
+jσz j sin ck

=
G01
β + σ kz
±j sin ck e
+jσz cos ck

G0
β + σ
· (3.82)
G01
β + σ
k−1z

where the + signs are taken when β and β + σ have the same
sign (i.e., both frequencies are positive, as in the usual case, or both
are negative), the − signs when they have opposite signs, in accord
with Equations (3.15) and (3.18).
The analysis follows that of Section 3.4:

1. Take the Kronecker product of Equation (3.82) with the com-

plex conjugate of Equation (3.81).
2. Take the expected value of the result.
3. Substitute the following definitions for the resulting column
vectors:

   
R0
z G0
β + σG∗0
β
   
 R01
z   G0
β + σG∗01
β 
  =  (3.83)
   ∗ 
 R10
z   G01
β + σG0
β 
R1
z σ
G01
β + σG∗01
β z

The subscript σ on the left-hand side indicates the depen-

dence of the R’s on the normalized frequency interval σ; the
subscript z on the right-hand side indicates the z dependence
of the G’s, as before.
4. Substitute Equation (3.51).
5. Take the limit as z → 0.
 3.6. TRANSFER FUNCTION STATISTICS 41

The final results are as follows:

   
R0
z −S0 0 0 ±S0
   
 R01
z   0 ∗ − S0 ±S0 0 
 = 
   
R
 10
z   0 ±S0  + jσ − S0 0 
R1
z ±S0 0 0 2α + jσ − S0
 
R0
z
 
 R01
z 
·


 (3.84)
 R10
z 
R1
z

The initial conditions are obtained from Equation (3.14) and (3.79):

R0
σz=0 = 1 R1
σz=0 = 0 (3.85)

R01
σz=0 = R10
σz=0 = 0 (3.86)

The four equations again decouple into two sets of two equations,
in the two-mode case.
For the cross-moments, Equations (3.84) and (3.86) yield

R01
σz = R10
σz = 0 all z (3.87)

Signal and spurious modes remain uncorrelated for all z, for zero
spurious mode input.
The transfer function covariances are defined in Equation (3.83)
as

R0
σ = G0
β + σG∗0
β

(3.88)
R1
σ = G01
β + σG∗01
β

They are obtained from Equation (3.84) as solutions of the following

equations, with initial conditions given by Equation (3.85) [1]:

R0
z = −S0 R0
z ± S0 R1
z

(3.89)
R1
z = ±S0 R0
z +
2α + jσ − S0 R1
z
42 GUIDES WITH WHITE RANDOM COUPLING

The solutions yield exact expressions for the normalized signal and
signal-spurious mode transfer function covariances, for white cou-
pling, as follows:

R0
σ = exp −S0 z
1 − (

  
 sinh
S0 z 1 + (2 
2
· cosh
S0 z 1 + (  − (   (3.90)
1 + (2

sinh
S0 z 1 + (2 
R1
σ = ± exp −S0 z
1 − (   (3.91)
1 + (2
α + jσ/2
(=  (3.92)
S0
Finally, from Equation (3.33)
S0
G0
β = e− 2 z G01
β = 0 (3.93)

The transfer functions are wide-sense stationary functions of nor-

malized frequency β.
The multi-mode case is treated similarly. We assume as in Section
3.2 that the frequency dependence enters only through the phase
constants βi . In contrast, for the two-mode case only the single pa-
rameter β varies with frequency. We append subscripts x to ele-
ments of Equation (3.34) to indicate evaluation at frequency fx :

I x
kz = ejck C · e−
x z · I x
k − 1z (3.94)

with C and
given by Equations (3.35) and (3.36). Define the trans-
fer function co- and cross-variances as
 
Ri
zxy = Ii
zx Ii∗
zy
  (3.95)
Rij
zxy = Ii
zx Ij∗
zy = R∗ji
zyx i = j

Define the covariance vector 
zxy as the Kronecker product

 
T
zxy = I Tx
z ⊗ I ∗T
y
z

= R0 R01 · · · R0 N−1 R10 R1 · · · R1 N−1 ···


··· RN−1 0 · · · RN−1 xy
(3.96)
 3.7. IMPULSE RESPONSE STATISTICS 43

where each element of this vector represents one of the functions of

Equation (3.95) with abbreviated notation. Following the analysis of
Section 3.4:

1. Replace the subscript x in Equation (3.94) by the subscript y;

take the complex conjugate of the result.
2. Take the Kronecker product of Equation (3.94) with the result
of step 1.
3. Take the expected value of the result.
4. Substitute the definition of Equations (3.95) and (3.96) for the
resulting column vectors.
5. Substitute Equations (3.39) and (3.51).
6. Take the limit as z → 0.

The calculations are a straightforward extension to those of Equa-

tions (3.59) and 3.60). We obtain


zxy = S0
C ⊗ C − 21 I ⊗ C2 − 21 C2 ⊗ I 
− I ⊗
y∗ −
x ⊗ I  · 
zxy  (3.97)

The present results for σ = 0 or x = y, in the two-mode or

multi-mode cases respectively, yield the coupled power equations
of Section 3.4. Treatment of general frequency-dependent Cij , αi ,
and βi is straightforward by the present methods [3]. Both co- and
cross-variances of the transfer functions are obtained by this analy-
sis. We do not assume that different transfer functions are uncorre-
lated [2, 4]. Simplifications occur in the nondegenerate case (Section
4.6.2).

3.7. IMPULSE RESPONSE STATISTICS

The analysis of Section 3.6 for transfer-function statistics was carried

out in the frequency domain. We interpret these results in the time
domain for the signal-mode transfer function of the idealized two-
mode guide of Section 3.2, with frequency-independent attenuation
α1 and α2 , coupling coefficient C0 , and group velocities v1 and v2 .
44 GUIDES WITH WHITE RANDOM COUPLING

Consider a causal filter with real impulse response ɡ
t and trans-
fer function 
f :

∞
ɡ
t = 
f ej2πft df (3.98)
−∞

ɡ
t = ɡ∗
t ɡ
t = 0 t < 0 (3.99)

∞

f  = ɡ
te−j2πft dt (3.100)
0


f  = ∗
−f  (3.101)

Assume the transfer function is wide-sense stationary:


ν = 
f + ν∗
f   
ν = ∗
−ν (3.102)


f  =  (3.103)

The expected value and covariance of the transfer function are in-
dependent of the frequency f . The spectral density of the transfer
function is 
−t, where 
t is the Fourier transform of 
ν:

∞

t = 
νej2πtν dν 
t = ∗
t (3.104)
−∞

Cascade the transfer function 
f  with an ideal band-pass filter of

bandwidth B, to yield an overall transfer function 
f  as follows:


f  f − f0  < B

f  = B < f0  (3.105)
0 f − f0  > B

Let the impulse response of the overall filter be 
t, with envelope
 

t = 
t + j
t (3.106)



t is the Hilbert transform of 
t:
 1 1
∞ 
τ

t = + 
t = dτ (3.107)
πt π −∞ t − τ
 3.7. IMPULSE RESPONSE STATISTICS 45

where + denotes the convolution operator. Then, it is shown in Sec-

tion E.1 that

 2
t ≈ 8B
t B large, (3.108)
where B, the half-bandwidth of the ideal bandpass filter cascaded
with 
f , is large compared to the correlation bandwidth of

f  [1].
To summarize, 
−t is the spectral density of the transfer func-
tion 
f . 
t is proportional to the expected squared envelope of
the impulse response of this transfer function, for large bandwidth
at a high carrier frequency.
Let the filter input be the modulated white noise
t
t, where

t is a white noise of unit power:

t + τ
t = δ
τ (3.109)

t is a real deterministic pulse; we call 2
t the input intensity.
The filter output is

∞

t =
t
t + 
t =
τ
τ
t − τdτ (3.110)
−∞

The output expected squared envelope is given in Section E.1 as


t =
2 2


t +  2
t ≈ 8B 2

t + 
t B large. (3.111)

We call Equation (3.111) the output intensity, Equation (3.108) the

intensity impulse response. A linear stationary random transfer func-
tion is linear also in intensity [4].
These results apply to the normalized impulse response of Section
3.2 by straightforward change of variables. The Fourier transforms
of Equations (3.90) and (3.91) are [1]:

∞ 
−jτσz σz
P0
τ = R0
σe d = e−S0 z δ
τ + P0ac
τ (3.112)
−∞ 2π
where
 

 −S0 z 2αzτ
1−τ 
S0 z · e e I1
2S0 z τ
1 − τ  0<τ<1
P0ac
τ = τ

0 otherwise.
(3.113)
46 GUIDES WITH WHITE RANDOM COUPLING
 
S0 z · e−S0 z e2αzτ I0
2S0 z τ
1 − τ 0 < τ < 1
P1
τ = (3.114)
0 otherwise.

I0
 and I1
 represent modified Bessel functions of orders 0 and

1, respectively.
The first term of Equation (3.112) is the spectral density of the
d.c. component, and P0ac
−τ is the spectral density of the a.c. com-
ponent of the transfer function G0
β, as discussed in Section E.1.
The coefficient of δ
τ in Equation (3.112) is the square of G0  in
Equation (3.33). Moreover, 
t of Equations (3.104) and (3.108) is
related to P0
t and P1
t of Equations (3.112) and (3.114) by

1 t

t = P0 1 (3.115)
T T

where T is given in Equation (3.20). We call P0
τ the normalized

signal intensity impulse response. In Equation (3.112) e−S0 z · δ
τ
corresponds to the undistorted signal output. P0ac
τ corresponds to
the undesired signal-mode output due to the random white coupling;
we call P0ac
τ the normalized echo power. P1
τ is the normalized
signal-spurious intensity impulse response.3
Finally, a direct analysis of the coupled line equations in the time
domain, for the idealized two-mode case of Section 3.2, is given in
Section E.2. Limiting forms for some of the present results have
been obtained directly in the time domain [4].

3.8. DISCUSSION

We have obtained exact results for a number of transmission statis-

tics of the coupled line equations with white coupling. This analysis
contains no approximations of any kind; it can treat coherent and
incoherent, single and multi-mode inputs. The methods are straight-
forward, and may readily be applied to similar problems; as one
example, white random propagation constants are treated in Chap-
ter 5.
3
The usage of P
 for normalized intensity impulse response differs from similar
notation used to represent mode powers elsewhere in this text.
 REFERENCES 47

Closed-form analytical results are given above for two modes. A

large number of modes requires the use of one of the symbolic com-
puter algebra programs. Chapter 4 is devoted to applications of the
present results, using MAPLE where required for the more com-
plex examples. Cross-powers and sometimes higher mixed moments
appear in this work; but Section 4.6 demonstrates substantial simpli-
fication in the absence of degenerate modes.
An open guide, such as an optical fiber or the dielectric slab
waveguide of Appendix F, possesses radiation modes, in addition
to the guided modes studied here. We ignore radiation modes
throughout the present treatment; it has been suggested that their
effect may be included in a phenomenological way by an appropri-
ate contribution to the attenuation constant of each propagating
mode.
The present results for white coupling can be extended to general
coupling spectra under weak restrictions, by combining perturbation
theory of Chapter 2 with the matrix techniques of the present chap-
ter. This generalization is done in Chapter 6 for the coupled power
equations. These results are validated by comparison with exact re-
sults for random square-wave coupling in Chapter 8. The extended
results agree with the exact results of the present chapter for the
limiting case of white coupling, and show the utility of the present
results for almost white coupling spectra.
A number of related studies are contained in references [5–8].

REFERENCES

1. Harrison E. Rowe and D. T. Young, “Transmission Distortion in Mul-

timode Random Waveguides,” IEEE Transactions on Microwave Theory
and Techniques, Vol. MTT-20, June 1972, pp. 349–365.
2. Dietrich Marcuse, Theory of Dielectric Optical Waveguides, 2nd ed., Aca-
demic Press, New York, 1991.
3. D. T. Young and Harrison E. Rowe, “Optimum Coupling for Random
Guides with Frequency-Dependent Coupling,” IEEE Transactions on Mi-
crowave Theory and Techniques, Vol. MTT-20, June 1972, pp. 365–372.
4. S. D. Personick, “Time Dispersion in Dielectric Waveguides,” Bell System
Technical Journal, Vol. 50, March 1971, pp. 843–859.
5. S. E. Miller, “Waveguide as a Communication Medium,” Bell System
Technical Journal, Vol. 33, November 1954, pp. 1209–1265.
48 GUIDES WITH WHITE RANDOM COUPLING

6. J. A. Morrison and J. McKenna, “Coupled Line Equations with Random

Coupling,” Bell System Technical Journal, Vol. 51, January 1972, pp. 209–
228.
7. S. D. Personick, “Two Derivations of the Time-Dependent Coupled-
Power Equations,” Bell System Technical Journal, Vol. 54, January 1975,
pp. 47–52.
8. Richard Steinberg, “Pulse Propagation in Multimode Fibers with
Frequency-Dependent Coupling,” IEEE Transactions on Microwave The-
ory and Techniques, Vol. MTT-23, January 1975, pp. 121–122.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

CHAPTER FOUR

Examples—White Coupling

4.1. INTRODUCTION

We give a number of applications of the results of Chapter 3. This

study is not intended to be exhaustive, but rather to illustrate the
range of possibilities offered by the present analysis. White coupling
and constant propagation parameters are assumed, as described in
Section 3.1. In addition, all modes are assumed lossless in the fol-
lowing examples, to reduce the number of parameters. We will con-
sider coherent and incoherent, single- and multi-mode inputs for an
N-mode guide, with N = 2 and N = 4.
For two-mode guide closed-form expressions are evaluated for the
average mode powers and cross-powers, and for the second-order
impulse response statistics. Power fluctuations are determined by
MAPLE. The number of parameters describing the system is small
enough to permit presentation of the results in general terms.
For more than two modes, there are too many parameters for a
general treatment. We therefore consider a particular 4-mode guide
described in Appendix F, and use MAPLE to obtain the average
mode powers.
Finally, the nondegenerate case, in which all modes have dis-
tinct propagation constants, yields much simpler approximate results.
These approximate results are readily modified to include degener-
ate modes, while retaining substantial simplification from the general
exact results if most modes are nondegenerate.
The following inputs are used for these calculations.

49
50 EXAMPLES—WHITE COUPLING

4.1.1. Single-Mode Input

Consider first an input signal I0 of unit power, with other mode
inputs zero. All modes except the signal are zero at the input:
Ii 0 = 0 1 ≤ i ≤ N − 1 (4.1)
For a coherent signal, i.e., a unit sinusoidal input,
I0 0 = 1 (4.2)
We consider as an incoherent signal a sinusoid with unit amplitude
and random phase:1
I0 0 = ejθ0  (4.3)


 1  0 < θ0 < 2π
wθ0  = 2π (4.4)

0 otherwise.
Then,
I0 0 = 0  I0 0 2  = 1 (4.5)

Ii 0Ij 0 =  Ii 0 2 Ij 0 2  = Ii2 0Ij∗2 0 = 0

(4.6)
i = j 0 ≤ i j ≤ N − 1
For both cases, coherent and incoherent, the average input powers
and cross-powers are2
P 0 0 = 1 P i 0 = Pi 0 = 0 1 ≤ i ≤ N − 1 (4.7)

P ij 0 = Pij 0 = 0 i = j 0 ≤ i j ≤ N − 1 (4.8)

The higher moments are:
P02 0 = 1 Pi 0Pj 0 = Pij2 0 = 0
(4.9)
i = j 0 ≤ i j ≤ N − 1
1
Probability densities are denoted by w .
2
The mode powers Pi z are defined in Equations (2.3) and (2.46), the cross-powers
Pij z and the average mode powers P i z and cross-powers P ij z in Equations
(3.50).
 4.1. INTRODUCTION 51

4.1.2. Multi-Mode Coherent Input

A multi-mode coherent input, such as a plane wave, will excite all
modes with unit correlation coefficients at the input. We assume
unit sinusoidal inputs with zero phase for each of the N modes at
the input:

Ii 0 = 1 0 ≤ i ≤ N − 1 (4.10)

The corresponding average powers and cross-powers at the input

are:

P i 0 = Pi 0 = 1 0 ≤ i ≤ N − 1 (4.11)

P ij 0 = Pij 0 = 1 i = j 0 ≤ i j ≤ N − 1 (4.12)

The higher moments are:

Pi2 0 = Pi 0Pj 0 = Pij2 0 = 1 0 ≤ i j ≤ N − 1 (4.13)

4.1.3. Multi-Mode Incoherent Input

Finally, we consider a multi-mode incoherent input that excites unit
power in each mode, with all modes statistically independent. We as-
sume the input waves are unit sinusoids with independent, uniformly
distributed random phase:

Ii 0 = ejθi  0 ≤ i ≤ N − 1 (4.14)


 1  0 < θi < 2π
wθi  = 2π (4.15)

0 otherwise.

wθ1  θ2  · · · θN−1  = wθ1 wθ2  · · · wθN−1  (4.16)

Then,

Ii 0 = 0  Ii 0 2  =  Ii 0 4  = 1 0 ≤ i ≤ N − 1 (4.17)

52 EXAMPLES—WHITE COUPLING

 Ii 0 2 Ij 0 2  = 1 Ii 0Ij∗ 0 = Ii2 0Ij∗2 0 = 0

(4.18)
i = j 0 ≤ i j ≤ N − 1

The input power and cross-power statistics are

P i 0 = Pi 0 = 1 P ij 0 = Pij 0 = 0

(4.19)
i = j 0 ≤ i j ≤ N − 1

Higher moments become:

Pi2 0 = 1 Pi 0Pj 0 = 1 Pij2 0 = 0

(4.20)
i = j 0 ≤ i j ≤ N − 1

4.2. COUPLED POWER EQUATIONS—TWO-MODE CASE

4.2.1. Single-Mode Input

Consider Equations (3.4)–(3.6) for two coupled modes; assume zero
attenuation for both:

α0 = α1 = 0 (4.21)

Assume input only to the signal mode; i.e., from Equation (4.1) with
N = 2,

I1 0 = 0 (4.22)

The expected responses are given by Equations (3.32)–(3.33). For a

coherent signal input, Equation (4.2):
S0 S0
I 0 z = I0 z = e−jβ0 z e− 2 z  G 0 z = G0 z = e− 2 z 
(4.23)
I 1 z = I1 z = 0 G 1 z = G1 z = 0

S0 is given by Equations (3.1)–(3.2) as the spectral density of the

(white) coupling coefficient. The spurious mode has no coherent
component; the signal-mode coherent component decays exponen-
tially to zero at large distance z, the rate of decay increasing with S0 .
For an incoherent signal input, Equations (4.3)–(4.4), neither sig-
nal nor spurious mode has a coherent component.
 4.2. COUPLED POWER EQUATIONS—TWO-MODE CASE 53

The coupled power equations for the lossless case are, from Equa-
tions (3.53) and (4.21),

P 0 z = −S0 P 0 z + S0 P 1 z

(4.24)
P 1 z = S0 P 0 z − S0 P 1 z

The initial conditions, from Equations (4.7), are the same for both
coherent and incoherent signal-mode inputs; for N = 2,

P 0 0 = 1 P 1 0 = 0 (4.25)

The solutions to these equations are

1 1 −2S0 z
P 0 z = + e 
2 2
(4.26)
1 1
P 1 z = − e−2S0 z 
2 2
The results are plotted in Figure 4.1.
The cross-powers are given by Equation (3.54); in the lossless case
they become

P 01 z = −jβ + S0 P 01 z + S0 P 10 z

P 10 z = S0 P 01 z − −jβ + S0 P 10 z (4.27)
β = β0 − β1 

with initial conditions obtained from Equation (4.8) with N = 2 as

P 01 0 = P 10 0 = 0 (4.28)

Therefore,

P 01 z = P 10 z = 0 (4.29)

For signal only input the two modes are uncorrelated at the input,
and remain so for all z; this is true for both coherent and incoherent
signal mode inputs.
For a coherent signal input with zero spurious mode input, Equa-
tions (4.1) and (4.2), the signal mode contains a coherent component
given by Equation (4.23), plus a random component. The spurious
54 EXAMPLES—WHITE COUPLING

1.0

0.8
P0 (z)

0.6

0.4

P1 (z)
0.2

0
0 0.5 1 1.5 2 2.5 3
S0 z

FIGURE 4.1. Signal and spurious mode powers vs. distance, for coherent
or incoherent signal input.

mode contains only a random component. The signal power is there-

fore divided between coherent and incoherent components, the for-
mer given by I0 z 2 . We have
1 1 −2S0 z
P 0 zcoherent = e−S0 z  P 0 zincoherent = + e − e−S0 z  (4.30)
2 2
Figure 4.2 shows the total signal power and its two components.
 4.2. COUPLED POWER EQUATIONS—TWO-MODE CASE 55

1.0

0.8

P0 (z)

0.6

0.4

P0 (z) incoherent

0.2
P0 (z) coherent

0
0 0.5 1 1.5 2 2.5 3
S0 z

FIGURE 4.2. Coherent and incoherent signal mode powers vs. distance, for
coherent signal input.

As z increases the coherent signal component decays exponen-

tially to zero; for large z the total power divides equally between
signal and spurious modes. Further discussion is given in Section 4.4.

4.2.2. Two-Mode Coherent Input

Consider next the input conditions of Equations (4.10)–(4.12) for
N = 2. For the lossless case, Equation (4.21), the expected responses
56 EXAMPLES—WHITE COUPLING

given by Equations (3.30)–(3.31) become

S0 S0
I 0 z = I0 z = e−jβ0 z e− 2 z  G 0 z = G0 z = e− 2 z 
S0 S0
(4.31)
I 1 z = I1 z = e−jβ1 z e− 2 z  G 1 z = G1 z = e− 2 z 
Signal and spurious mode coherent components have equal magni-
tude, both decaying to zero at large distance z, with decay constant
S0 /2.
Equation (4.24), the lossless coupled power equations, now have
the trivial solutions
P 0 z = P 1 z = 1 (4.32)
These total powers are again composed of coherent and incoher-
ent components, the former given by the squared magnitudes of the
expected responses given in Equation (4.31). The incoherent com-
ponents are
P 0 zincoherent = P 1 zincoherent = 1 − e−S0 z  (4.33)
The cross-powers are given by Equation (4.27), with initial condi-
tions from Equation (4.12) with N = 2:
P 01 0 = P 10 0 = 1 (4.34)
The solutions are
e−S0 z
P 01 z = P ∗10 z =   2
2 1 − β S 0

 
2
β β −S0 1− β
S  z
2
(4.35)
· 1− −1+j e 0

S0 S0
 
2
β β +S0 1− β
S  z
2

+ 1− +1−j e 0

S0 S0
The following special cases are of interest. For equal propagation
constants,
P 01 z = P 10 z = 1 β = 0 (4.36)
this result is readily verified by elementary means. Next,
P 01 z = P ∗10 z = 1 + 1 − j · S0 ze−S0 z  β = S0  (4.37)
 4.2. COUPLED POWER EQUATIONS—TWO-MODE CASE 57

Finally, for large β we have

lim P 01 z = lim P 10 z = e−S0 z  (4.38)
β→∞ β→∞

Figure 4.3 shows the behavior of the cross-power magnitudes for

several values of β/S0 . The two modes become uncorrelated
for sufficiently large z. The four-mode guide described in Section 4.5

=0
S0
1

0.5

0.8

0.6

P01 (z)
5

P10 (z)
50 –
0.4

0.2

0
0 0.5 1 1.5 2 2.5 3
S0 z

FIGURE 4.3. Cross-powers vs. distance, for coherent signal and spurious
mode inputs.
58 EXAMPLES—WHITE COUPLING

has β/S0 ≈ 10+25 . This suggests that for non-degenerate modes

only the β/S0 = ∞ curve in Figure 4.3 will be of interest.

4.2.3. Two-Mode Incoherent Input

Consider finally the input conditions of Equations (4.14)–(4.19), with
N = 2. For the lossless case, Equations (3.30)–(3.31), (4.24), and
(4.27) yield the following trivial solutions:

I i z = Ii z = 0 G i z = Gi z = 0 i = 1 2 (4.39)

P 0 z = P 1 z = 1 (4.40)

P 01 z = P 10 z = 0 (4.41)

There is no coherent component in either signal or spurious mode.

Both modes have constant average powers. Finally, the cross-powers
are zero; consequently the two modes remain uncorrelated for all z.

4.3. POWER FLUCTUATIONS—TWO-MODE GUIDE

Next, we study the power fluctuations for two lossless coupled

modes. The fourth-order statistics for this case are given by Equa-
tions (3.73)–(3.78). For convenience, introduce the following nota-
tion in Equations (3.73) and (3.74):

T z = Q0 z Q1 z Q2 z Q3 z Q4 z  (4.42)

Q0 z = P02 z

Q1 z = P0 zP1 z
2
Q2 z = P01 z (4.43)
2 2
Q3 z = P10 z = P01 z∗ 
Q4 z = P12 z

The averages in Equation (4.43) are, respectively, the signal power

second moment, the cross-moment of signal and spurious mode pow-
 4.3. POWER FLUCTUATIONS—TWO-MODE GUIDE 59

ers, the two cross-power second moments, and the spurious power
second moment.
z is given by the solution of Equation (3.75),

 z = S0 N1 − N2  · z (4.44)

where N1 is given by Equation (3.76),

 
−2 4 −1 −1 0
 
 1 −4 1 1 1
 
 
N1 =  −1 4 −2 0 −1  (4.45)
 
 −1 4 0 −2 −1 
 
0 4 −1 −1 −2

and N2 is given by Equation (3.77) and (3.78) for the lossless case as
 
0 0 0 0 0
 
0 0 0 0 0
 
 
N2 =  0 0 −j2β 0 0  (4.46)
 
0 0 0 +j2β 0
 
0 0 0 0 0

The signal and spurious mode power variances are given by Equation
(3.68):

P0 z2  = Q0 z − P0 z2 = Q0 z − P 20 z

(4.47)
P1 z2  = Q4 z − P1 z2 = Q4 z − P 21 z

The corresponding standard deviations are

 
P 0 z = P0 z2  = Q0 z − P 20 z
  (4.48)
P 1 z = 2
P1 z  = Q4 z − P 21 z

where P 0 z and P 1 z are given in Section 4.2 for various inputs.
60 EXAMPLES—WHITE COUPLING

The initial conditions T 0 for Equations (4.44) are found by

substituting the relations of Sections 4.1.1–4.1.3 into Equations
(4.42)–(4.43) with z = 0. These differential equations are solved by
MAPLE, and the standard deviations for the signal and spurious
mode power fluctuations are determined by Equation (4.48). P 0 z
and P 1 z have the same units as the average powers, and indicate
the range of statistical fluctuation in the mode powers P0 z and
P1 z about their average values P 0 z and P 1 z due to random
coupling.
Q1 z Q2 z and Q3 z are not of principal interest here.

4.3.1. Single-Mode Input

The initial conditions T 0 are given by
T 0 = 1 0 0 0 0  (4.49)
This relation corresponds to zero spurious mode input and a con-
stant amplitude input signal of unit power, either with constant phase
(coherent) or random phase (incoherent). P 0 z and P 1 z are
shown in Figure 4.4.
The mode power second moments in the degenerate two-mode
case, β/S0 = 0 in Figure 4.4, are
3 1 −8S0 z 1 −2S0 z
Q0 z = P02 z = + e + e 
8 8 2
β = 0 (4.50)
3 1 1 −2S0 z
Q4 z = P12 z = + e−8S0 z − e 
8 8 2
The variances are
2 2 1 1 −8S0 z 1 −4S0 z
P 0 z = P 1 z = + e − e  β = 0 (4.51)
8 8 4

4.3.2. Two-Mode Coherent Input

The initial conditions for unit coherent inputs to signal and spurious
modes are
T 0 = 1 1 1 1 1  (4.52)
P 0 z and P 1 z are shown in Figure 4.5. These two quantities
are identically zero in the degenerate case, β = 0.
 4.3. POWER FLUCTUATIONS—TWO-MODE GUIDE 61

=0
S0
0.35

0.30

2
0.25

10 –
0.20
P0, 1 (z)

0.10

0.15

0.0

0
0 0.5 1 1.5 2 2.5 3

S0 z

FIGURE 4.4. Standard deviations of signal and spurious mode powers vs.
distance, with differential propagation constant as a parameter, for coher-
ent or incoherent signal input.

4.3.3. Two-Mode Incoherent Input

Finally, the initial conditions for unit incoherent inputs are
T 0 = 1 1 0 0 1  (4.53)
P 0 z and P 1 z are plotted in Figure 4.6. In the degenerate case

P 0 z = P 1 z = 5 1 − e−8S0 z  β = 0 (4.54)
62 EXAMPLES—WHITE COUPLING

100 –

0.6

0.5
10

0.4 2

P0, 1 (z)
0.3

=1
S0

0.2

0.1

=0
S0

0
0 0.5 1 1.5 2 2.5 3
S0 z

FIGURE 4.5. Standard deviations of signal and spurious mode powers vs.
distance, with differential propagation constant as a parameter, for coher-
ent signal and spurious mode inputs.

4.3.4. Discussion
The power fluctuations in the signal and spurious modes are equal
in each of the three cases of Figures 4.4–4.6. This is required by
conservation of power, Equation (2.5) in the lossless case. For large
z, the mode powers have standard deviations comparable to their
average values.
 4.3. POWER FLUCTUATIONS—TWO-MODE GUIDE 63

0.6

20 –
2
1
0.5

=0
S0

0.4

P0, 1 (z) 0.3

0.2

0.1

0
0 0.5 1 1.5 2 2.5 3
S0 z

FIGURE 4.6. Standard deviations of signal and spurious mode powers vs.
distance, with differential propagation constant as a parameter, for inco-
herent signal and spurious mode inputs.

The power fluctuations depend on β/S0 for small values of this

parameter. However, the example of Section 4.5 suggests that only
the β/S0 = ∞ curves in these three figures are significant for
nondegenerate modes.
The closed-form results given above for two degenerate modes,
obtained by MAPLE, are significant in that they are readily ob-
tained directly from the coupled line equations by elementary meth-
64 EXAMPLES—WHITE COUPLING

ods. This calculation is done in Appendix G, and serves as a partial

check on the present methods.

4.4. IMPULSE RESPONSE—TWO-MODE CASE

We next study some of the properties of the normalized signal-mode

intensity impulse response of a lossless, two-mode guide with white
coupling. From Equations (3.112) and (3.113),

P0 τ = e−S0 z δτ + P0ac τ (4.55)

where
   
 S z · e−S0 z 1 − τ I 2S z  τ1 − τ 

0 ≤ τ < 1
0
P0ac τ = τ 1 0


0 otherwise.
(4.56)

P0 t is proportional to the expected value of the squared enve-

lope of the normalized impulse response of Equations (3.16)–(3.17).
Recall that the real-time intensity impulse response
t for a
dispersionless guide is obtained from Equations (4.55)–(4.56) by
Equation (3.115), and that
t convolved with the input intensity
 2 t, is proportional to the corresponding output intensity (i.e., the
real-time output expected squared envelope) by Equation (3.111).
Equation (4.56) is plotted in Figure 4.7.
The normalized transfer function G0 β is wide-sense station-
ary by Equations (3.90) and (3.93), and therefore satisfies Equations
(E.1)–(E.15); it is the sum of a deterministic component G0  =
S0
G0 = e− 2 z and a random component G0ac β. Consequently, the
first term of Equation (4.55), e−S0 z δτ, represents the coherent part
of the response, due to the deterministic part of the transfer func-
tion. In Equation (4.56), P0ac τ represents the incoherent part of
the response, due to the random part of the transfer function, pre-
viously denoted the normalized echo power.
For zero coupling, S0 z = 0, P0ac τ = 0, and P0 τ = δτ. The
input signal travels undistorted, entirely in the signal mode, at veloc-
ity v0 = 2πf/β0 of Equations (3.19)–(3.20); since I0 z = e−jβ0 z G0 z,
 4.4. IMPULSE RESPONSE—TWO-MODE CASE 65

1.3

1.0
S0 z = 10

3
P0ac ( )

0.5

0.5

0
0 0.5 1.0

FIGURE 4.7. Normalized echo power, with coupling spectral density as a

parameter. α = 0 (From Harrison E. Rowe and D. T. Young, “Transmis-
sion Distortion in Multimode Random Waveguides,” IEEE Transactions on
Microwave Theory and Techniques, Vol. MTT-20, June 1972, pp. 349–365.
© 1972 IEEE.)

the output arrives at time L/v0 = β0 L/2πf  corresponding to τ = 0

in Figure 4.7.
For finite coupling, portions of the signal travel in both signal and
spurious modes, at velocities v0 and v1 , respectively, v0 > v1 . A signal
coupled into the spurious mode at z = 0, traveling to z = L in the
spurious mode, and coupled back into the signal mode at z = L,
will arrive at the output at time L/v1 = β1 L/2πf , corresponding
66 EXAMPLES—WHITE COUPLING

to τ = 1 in Figure 4.7. Generally, portions of the signal will travel in

both modes, with arrival times corresponding to values 0 < τ < 1 in
Figure 4.7.
We examine these results for small and for large coupling, to il-
lustrate the physics of these results. Using the small-argument ap-
proximation for the modified Bessel function in Equation (4.56):

P0ac τ ≈ S0 z2 e−S0 z 1 − τ S0 z  1 0 ≤ τ < 1 (4.57)

Using the asymptotic approximation for the Bessel function, and

LaPlace’s method:
  2
S0 z 1
P0ac τ ≈ exp −2S0 z τ −  S0 z  1 0 < τ < 1
2π 2
(4.58)

Equation (4.57) may be obtained directly from perturbation the-

ory, Equations (2.35) and (2.39), corresponding to the n = 1 term
of Equations (B.1)–(B.2). By Equation (E.33) this approximation in-
cludes only all pairs of transitions, from signal to spurious mode and
back. Since there are more pairs close together than far apart, the
normalized echo power P0ac τ in Equation (4.57) is triangular, as
shown in Figure 4.7 for S0 z = 05
Within the regime of perturbation theory, perhaps S0 z < 05,
Equations (4.55) and (4.57) or Figure 4.7 show that the larger the
coupling, the worse the performance; as S0 increases the coherent
component (i.e., the undistorted signal) decreases and the incoher-
ent component (i.e., the echo distortion) increases. Conventional
wisdom originally, and incorrectly, indicated that this behavior
continued beyond the perturbation region. S. D. Personick first
demonstrated that for very large coupling, increasing the coupling
improved the performance [1]. This remarkable and unexpected re-
sult is demonstrated by Equation (4.58) and Figure 4.7. The total
area under P0ac τ for large coupling is from Equation (4.58)
 
1 S0 z  ∞ −2S0 zτ2
P0ac τdt ≈ e dt = 05 S0 z  1 (4.59)
0 2π −∞
Therefore Equation (4.58) represents a narrow pulse of area 0.5
centered on τ = 05, approximating 05δτ for sufficiently large S0 z.
 4.5. COUPLED POWER EQUATIONS—FOUR-MODE CASE 67

As an example, for S0 z = 628, P0ac τ has a peak value P0ac 05 ≈ 10
and width ≈ 0032 at the 368% points, where P0ac 05 ± 0016 ≈
368; the coherent component has become insignificant.
As S0 increases, more terms of the summation in Equation (B.1)
become important. The partial impulse response corresponding to
Equation (B.2) is given in Equation (E.33), which must be integrated
over all possible sets of 2n transitions; for the great majority of these
the signal will have spent about half of the time traveling in the signal
and in the spurious modes. Consequently, the output will be a narrow
pulse containing half the input power, with delay corresponding to
the average velocity of signal and spurious modes. This behavior is
described by the results of Equations (4.58) and (4.59).

4.5. COUPLED POWER EQUATIONS—FOUR-MODE CASE

We apply the results of Sections 3.3 and 3.4, for the average transfer
functions and for the coupled power equations, to four lossless cou-
pled modes. The large number of parameters necessary—6 coupling
coefficients, 4 propagation constants, and the spectral density of the
geometric parameter—do not allow a compact treatment of the gen-
eral case, as was done for the two-mode case. Consequently, we con-
sider as a specific example TE modes in a symmetric slab waveguide,
with particular numerical parameters for dielectric constants, thick-
ness, and frequency/wavelength given in Appendix F. The average
transfer functions are given by Equation (3.41) as

1
 z = − − S0 C2 ·  z  z = Iz (4.60)
2

The coupled power equations are from Equation (3.61)

 
1 2 1 2 ∗
 z = S0 C ⊗ C − I ⊗ C − C ⊗ I − I ⊗  −  ⊗ I · z
2 2
(4.61)

The dielectric slab is illustrated in Figures F.1 and F.3, with index
1.01, thickness 123 · 10−5 m, and wavelength 10−6 m, given in Equa-
tions (F.14) and (F.17). The propagation and coupling matrices are
68 EXAMPLES—WHITE COUPLING

given by Equations (F.20) and (F.21) as follows:

 
63424 0 0 0
 
 0
6
63316 0 0  −1
 = j × 10  m  (4.62)

 0 0 63142 0 
0 0 0 62921
 
0 23225 0 44966
 
 23225
9
0 68737 0  −2
C = 10  m  (4.63)

 0 68737 0 133083 
44966 0 133083 0
With these substitutions Equation (4.60) yields four coupled differ-
ential equations for the expected transfer functions, while Equation
(4.61) yields 16 coupled differential equations for the average mode
powers and cross-powers.
S0 of Equations (4.60) and (4.61), defined in Equations (3.1)–(3.2),
is the spectral density of cz in Equation (3.9). In the present case,
cz is the straightness deviation of the dielectric slab illustrated in
Figure F.3. We need to select a numerical value for S0 consistent
with a physical model of this guide.
The present white model for cz has infinite straightness devia-
tion for any finite value of S0 . The mean-square straightness devia-
tion for a general coupling spectral density Sν is
∞
c 2 z = Sνdν (4.64)
−∞

We show in Chapter 6 that the present results, for white coupling

spectra, hold under reasonable restrictions for ideal low-pass Sν
that cover the range of spatial frequencies 0 < ν < νc , where
βi − βj max
νc   all i j (4.65)
2π
In the present case, Equation (4.62) requires that Sν be flat up to
the spatial frequency
β0 − β3 00503 × 106
νc  = ≈ 8 · 103 m−1  (4.66)
2π 2π
 4.5. COUPLED POWER EQUATIONS—FOUR-MODE CASE 69

We somewhat arbitrarily choose the following parameters for

Sν:

S0 = 10−21 m3  ν < 2 × 104 m−1 
Sν = (4.67)
0 ν > 2 × 104 m−1 

From Equation (4.64), this corresponds to an rms straightness devi-

ation of

c 2 z ≈ 6325 × 10−9 m, (4.68)

small compared to the guide width in Equation (F.17) of 123·10−5 m.

We use MAPLE to solve Equations (4.60) and (4.61) for the nu-
merical parameters of Equations (4.62), (4.63), and (4.67), for three
representative inputs.

4.5.1. Single-Mode Input

The initial conditions for unit signal input are given in Section 4.1.1.
For coherent input, we have for Equation (4.60)

 T 0 = I T 0 = 1 0 0 0 (4.69)

The magnitude of the average signal transfer function is shown in

Figure 4.8.
Modes 1 and 3 have zero coherent component

I1 z = I3 z = 0 (4.70)

While the expected values of modes 0 and 2 are coupled, in contrast

to the two-mode case of Section 4.2.1, the coupling is very weak:

I2 z < 3 × 10−5  (4.71)

too small to appear in Figure 4.8. The average signal transfer func-
tion appears to decay exponentially, similar to the two-mode case.
For incoherent signal input, all four modes have zero coherent
component.
70 EXAMPLES—WHITE COUPLING

0.8

0.6

I0 (z)

0.4

0.2

0
0 50 100 150 200

z - meters
FIGURE 4.8. Average signal transfer function magnitude, for coherent
single-mode input.

The initial conditions for Equation (4.61) for unit coherent or

incoherent signal input are
 
 T 0 = 1 0  0 0 0 0 0 0 0 0 0 0 0 0 0 ! 
0
15 zeros
(4.72)

The average mode powers for the four modes are shown in Fig-
ure 4.9.
 4.5. COUPLED POWER EQUATIONS—FOUR-MODE CASE 71

1.0

0.8

P0 (z)
0.6

0.4

P3 (z)
0.2

P2 (z)

P1 (z)

0
0 20 40 60 80 100
z - meters
FIGURE 4.9. Mode powers vs. distance, for coherent or incoherent signal
input.

For large z, the total power divides equally between all four
modes. If the input signal is coherent, P 0 z contains a coherent
component I0 z, whose magnitude is shown in Figure 4.8, with
power given by I0 z 2 . Modes 1 and 3 have zero coherent com-
ponent; the coherent component of mode 2 is too small to appear
on the scale of Figure 4.9.
72 EXAMPLES—WHITE COUPLING

4.5.2. Multi-Mode Coherent Input

The initial conditions for four-mode coherent input are given in Sec-
tion 4.1.2. For Equation (4.60),

 T 0 = I T 0 = 1 1 1 1  (4.73)

The magnitudes of the expected values of the four modes are shown
in Figure 4.10.
Each mode appears to decay exponentially, with its own decay
constant. Further study shows that the coupling between all four
expected mode amplitudes is weak, as in the preceding section.
The initial conditions for the coupled power equations, Equation
(4.61), for four-mode coherent input are
 
 T 0 = 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1! 
16 ones

(4.74)

The average powers, P 0 z P 1 z P 2 z and P 3 z, depart only
infinitesimally from their initial values of 1, by less than ±4 × 10−6
over the range 0 < z < 30, and imperceptibly for greater z.
Each contains a coherent component, given by the square mag-
nitude of the corresponding average mode amplitude, shown on
Figure 4.10.

4.5.3. Multi-Mode Incoherent Input

For four-mode incoherent input, we have the following initial con-
ditions from Section 4.1.3:

 T 0 = I T 0 = 0 0 0 0  (4.75)

 T 0 = 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 
(4.76)

Equations (4.75) and (4.60) yield

I0 z = I1 z = I2 z = I3 z = 0 (4.77)

 4.5. COUPLED POWER EQUATIONS—FOUR-MODE CASE 73

0.8

0.6

0.4

I0 (z)

I1 (z)
0.2

I3 (z)

I2 (z)
0
0 50 100 150 200

z - meters
FIGURE 4.10. Average mode amplitudes, for four-mode coherent input.

all four modes have zero coherent component. Equations (4.76) and
(4.61) yield

P 0 z = P 1 z = P 2 z = P 3 z = 1 (4.78)

For large z the average mode powers are equal, with zero coherent
components. Since the initial conditions satisfy this asymptotic be-
havior, the coupled power equations yield no further change in the
average powers.
74 EXAMPLES—WHITE COUPLING

4.6. NONDEGENERATE CASE—APPROXIMATE RESULTS

The two-mode case of Section 4.2 exhibits simplifications not present

in the four-mode results of Section 4.5:

1. The expected mode amplitudes are exponential, and uncou-

pled.
2. The average mode powers are uncoupled from the cross-
powers.

Nevertheless, in this four-mode case the expected mode ampli-

tudes appear at least approximately exponential, and are coupled
only very weakly. Moreover, Marcuse has presented multi-mode cou-
pled power equations in which cross-powers do not appear [2]; it will
appear in Section 4.6.2 below that his simpler relations, obtained
from 4 coupled differential equations, give results indistinguishable
from those of Figure 4.9, obtained from 16 coupled differential equa-
tions. These facts suggest further simplifications of the present gen-
eral results in special cases.
The approximations that we desire arise from the disparate nu-
merical values of the coefficients in the differential equations for
the average mode amplitudes and average powers, Equations (4.60)
and (4.61). In the particular example of Section 4.5, with numerical
values given by Equations (4.62), (4.63), and (4.67), the first term
on the right-hand side of Equation (4.60) is of order O10+6 , the
second O10−3 ; the first term on the right-hand side of Equation
(4.61) is O10−3 , the second and third terms O10+6 . The principal
contributions to the derivatives on the left-hand sides of these two
differential equations thus arise from the terms involving  in this
particular case. This suggests, as a first step, removing the determin-
istic portion of the solutions involving the propagation constants, by
the change of variables given in Equation (3.11):
Iz = e−z · Gz (4.79)
Then find the asymptotic solutions as all the components of , i.e.,
the mode propagation constants "0  "1  "2  · · · of Equation (3.10),
approach infinity. In the nondegenerate case no two "i ’s will be
equal, and all of the differential propagation constants "ij will
also approach infinity. Special account must be taken for degenerate
modes, for which " = 0.
 4.6. NONDEGENERATE CASE—APPROXIMATE RESULTS 75

4.6.1. Average Transfer Functions

Substituting Equation (4.79) into Equation (4.60), we find

1 " #
Gz = − S0 e+z · C2 · e−z · Gz (4.80)
2

The coupling matrix C is given by Equation (2.45) or (3.35); the

elements of C2 are as follows:3

$
N−1 $
N−1
C2 ij = Ci−1k Ckj−1 = Ci−1k Cj−1k  1 ≤ i j ≤ N (4.81)
k=0 k=0

For diagonal elements, this becomes

$
N−1 $
N−1
C2 ii = Ci−1k Cki−1 = 2
Ci−1k  1 ≤ i ≤ N (4.82)
k=0 k=0

Note that the k = i − 1 and k = j − 1 terms in Equation (4.81) are

zero, while the k = i − 1 terms in Equation (4.82) are zero. Equation
(3.10) yields
 
eα0 +jβ0 z 0 0 ···
 
 0 eα1 +jβ1 z 0 ··· 
 
ez = e+jz =  (4.83)
 0 0 eα2 +jβ2 z ··· 
 
   
   

Now the diagonal terms of the matrix on the right-hand side of

Equation (4.80) are identical to those of Equation (4.82):

" # $
N−1
e+z · C2 · e−z ii
= C2 ii = 2
Ci−1k  1 ≤ i ≤ N (4.84)
k=0

3
The matrix elements are numbered from 1 1 to N N, while the nonzero coupling
coefficients in Equation 2.45, with unequal subscripts, are numbered from 0 1 to
N − 2 N − 1.
76 EXAMPLES—WHITE COUPLING

However, for the off-diagonal terms the exponential factors do not

cancel out; using Equations (4.81) and (4.83),
" #
e+z · C2 · e−z ij
= e"i−1j−1 z C2 ij

"i−1j−1 z
$
N−1
=e Ci−1k Ckj−1  1 ≤ i j ≤ N (4.85)
k=0

where

"ij = "i − "j = αij + jβij = αi − αj  + jβi − βj 

(4.86)
0 ≤ i j ≤ N − 1

Consider now the solution of Equation (4.80) by the method of

successive approximations, similar to that of Appendix A. The right-
hand-side integrals in each step of the successive approximations
will approach zero for each of the off-diagonal terms in Equation
(4.85), for i = j, as βij → ∞, by the method of stationary phase
[3]. Therefore in the asymptotic limit the differential equations of
Equation (4.80) become uncoupled, and we have
1 %N−1 2
Gi z∞ = Gi 0e− 2 S0 z k=0 Cik
 0 ≤ i ≤ N − 1 (4.87)

where the subscript ∞ indicates the asymptotic limit. Finally, substi-

tuting Equations (4.79) and (4.83),
1 %N−1 2
Ii z∞ = Ii 0e−"i z e− 2 S0 z k=0 Cik
 0 ≤ i ≤ N − 1 (4.88)

The average transfer functions become uncoupled and exponential

for large βij .
For the four-mode example of Section 4.5, with single-mode input
given by Equation (4.69), Equation (4.88) becomes

I0 z ∞ = e−001281z  (4.89)

In this case, the exponential factor ejβ02 z = ejβ0 −β2 z has approxi-
mately 106 periods along the horizontal axis of Figure 4.8. Therefore
the asymptotic approximation of Equation (4.89) is a good approxi-
mation to the exact solution shown in Figure 4.8, obtained by solving
 4.6. NONDEGENERATE CASE—APPROXIMATE RESULTS 77

four simultaneous differential equations. The coefficient in the ex-

ponent for the ith mode in Equations (4.87) and (4.88) is the power
per unit length coupled from the ith mode to all of the other modes;
the k = i term in the summation is zero.

4.6.2. Coupled Power Equations

We perform the substitution of Equation (4.79) in the coupled
power equations, Equation (4.61). From Equations (2.43), (3.50),
and (3.57),

Pz = Iz ⊗ I ∗ z (4.90)

z = Pz = Iz ⊗ I ∗ z (4.91)

Substituting Equation (4.79) into Equation (4.90),

& ' & ∗ '
Pz = e−z · Gz ⊗ e− z · G∗ z
& ∗ '
= e−z ⊗ e− z · Gz ⊗ G∗ z  (4.92)

We require the following additional notation:

Ri z = Ri z Ri z = Gi z 2  (4.93)

Rij z = R∗ji z = Rij z Rij z = Gi zG∗j z (4.94)

z = Rz Rz = Gz ⊗ G∗ z (4.95)

T
&
z = R0 R01 · · · R0N−1 R10 R1 · · · R1N−1 ···
'
··· RN−10 RN−11 · · · RN−1  (4.96)

where the z dependence of the Ri and Rij has been suppressed.

Then, Equations (4.92) and (4.95) yield
" ∗ #
Pz = e−z ⊗ e− z · Rz (4.97)

Differentiating Equation (4.97),

" ∗ # " ∗ #
P z = e−z ⊗ e− z · Rz + e−z ⊗ e− z · R z (4.98)
78 EXAMPLES—WHITE COUPLING

Evaluating the first factor on the right-hand side,

" −z ∗ #
e ⊗ e− z
" # ∗ " ∗ #
= − · e−z ⊗ e− z + e−z ⊗ −∗ · e− z
" # " ∗ # " # " ∗ #
= −  · e−z ⊗ I · e− z + I · e−z ⊗ −∗ · e− z (4.99)
" ∗ # " ∗ #
= −  ⊗ I  · e−z ⊗ e− z − I ⊗ ∗  · e−z ⊗ e− z
" ∗ #
= −  ⊗ I + I ⊗ ∗  · e−z ⊗ e− z 
where I is the unit matrix. Substituting Equation (4.99) into Equa-
tion (4.98), and taking the expected value using Equations (4.91) and
(4.95),
" ∗ #
 z = −  ⊗ I + I ⊗ ∗  · e−z ⊗ e− z · z
" ∗ #
+ e−z ⊗ e− z · z (4.100)
Similarly, the expected value of Equation (4.97) yields
" ∗ #
z = e−z ⊗ e− z · z (4.101)
Finally, substituting Equations (4.100) and (4.101) into Equation
(4.61), we obtain the desired result:
" ∗ # 1 1
z = S0 ez ⊗ e z · C ⊗ C − I ⊗ C2 − C2 ⊗ I
2 2
" −z ∗ #
· e ⊗ e− z · z (4.102)
Using the transformation of Equation (C.8), this may be written
1 1
z = S0 M1 − M2 − M3 · z (4.103)
2 2
where
M1 = V ⊗ V ∗  M2 = I ⊗ W ∗  M3 = W ⊗ I  (4.104)
and
V = ez · C · e−z  W = ez · C2 · e−z  (4.105)
We require the asymptotic limit of these results as βij → ∞,
i = j Toward this end we divide z into subvectors of dimension
 4.6. NONDEGENERATE CASE—APPROXIMATE RESULTS 79

N as follows:
T
& T T T
'
z = 1 z 2 z ··· N z  (4.106)

T
&
i z = Ri−10 z Ri−11 z ··· Rii−2 z
'
Ri−1 z Ri−1i z ··· Ri−1N−1 z  (4.107)

We divide each of the three M matrices into N × N submatrices:

 
M1 11 M1 12 M1 13 · · · M1 1N
 
 M1 21 M1 22 M1 23 · · · M1 2N 
 
 
M1 =  M1 31 M1 32 M1 33 · · · M1 3N  (4.108)
 
     
 
     
M1 N1 M1 N2 M1 N3 ··· M1 NN

with a similar partition for M2 and M3 . Then the ith subvector of

z is found from Equations (4.103)–(4.107) as the solution of

$
N
1 1
i z = S0 M1 ik − M2 ik − M3 ik · k z 1 ≤ i ≤ N
k=1
2 2
(4.109)

Appendix H gives the asymptotic form for the M matrices and their
submatrices.
First, substitute Equations (H.5)–(H.7) into the ith row of Equa-
tion (4.109):

$
N
2
Ri−1 z∞ = S0 Ci−1k−1 e2αi−1k−1 z Rk−1 z∞ − Ri−1 z∞ 
k=1
1 ≤ i ≤ N (4.110)

Using Equations (4.79), (4.93), and (3.50), and relabeling the sum-
mation index in Equation (4.110), we have the final desired result:

$
N−1
2
$
N−1
2
P i z∞ = − 2αi + S0 Cik P i z∞ + S0 Cik P k z∞ 
k=0 k=0
0 ≤ i ≤ N − 1 (4.111)
80 EXAMPLES—WHITE COUPLING

The subscript ∞ again denotes the asymptotic limit as βij → ∞

Note from Equation (3.35) that the k = i terms in the summations
of %
Equation (4.111) are zero. As noted in Section 4.6.1, the quantity
S0 N−1 2
k=0 Cik is the power per unit length coupled from the ith mode
2
to all of the other modes; S0 Cik P k z∞ is the power per unit length
coupled from the kth mode to the ith mode.
The result of Equation (4.111) shows that for large βij the av-
erage power become uncoupled from the cross-powers. Marcuse’s
multi-mode coupled power equations [2], which contain only av-
erage powers but not cross-powers, are thus seen to represent the
asymptotic limit for large βij  For the four-mode case of Section 4.5
Equations (4.111) become
   
P 0 z∞ −00256 00054 0 00202
   
 P 1 z∞   00054 −00526 00472 0 
 = 
   
 P 2 z∞   0 00472 −02243 01771 
P 3 z∞ 00202 0 01771 −01973
 
P 0 z∞
 
 P 1 z∞ 
·


 (4.112)
 P 2 z∞ 
P 3 z∞

For single-mode input, with initial conditions

P 0 0∞ = 1 P 1 0∞ = P 2 0∞ = P 3 0∞ = 0 (4.113)

the solutions to the four simultaneous differential equations of Equa-

tion (4.112) are indistinguishable from those of Figure 4.9, obtained
from the sixteen simultaneous differential equations of Equations
(4.61)–(4.63) and (4.72).
Finally, consider rows other than the ith of Equation (4.109). For
i = j Equation (H.5) shows that M1 ik makes zero contribution to
this summation. Equations (H.6) and (H.7) yield

S0 $
N "
2 2
#
Ri−1j−1 z∞ = − Ci−1k−1 + Cj−1k−1 Ri−1j−1 z∞  (4.114)
2 k=1
 4.6. NONDEGENERATE CASE—APPROXIMATE RESULTS 81

From Equations (3.50), (4.79), and (4.94),

 
& ' $ 2
S0 N−1 2
P ij z∞ = exp αi + αj z exp − C + Cjk z P ij 0∞ 
2 k=0 ik
(4.115)

For the lossless two-mode case, Equation (3.42) yields the following
analytic result for the β/S0 = ∞ curve of Figure (4.3):

P 01 z∞ = P 10 z∞ = e−S0 z  (4.116)

This confirms the prior result of Equation (4.38), obtained directly

as the limit of Equation (4.35).

4.6.3. Power Fluctuations

As an alternative to a general treatment of power fluctuations, we
use MAPLE directly for the two-mode case. By substituting Equa-
tion (4.79) into Equations (3.64), (3.70), and (3.72), we have
" ∗ ∗ #

4 z = S0 ez ⊗ e z ⊗ ez ⊗ e z


· I ⊗I ⊗C⊗C−I ⊗C⊗I ⊗C+I ⊗C⊗C⊗I

+C⊗I ⊗I ⊗C−C⊗I ⊗C⊗I +C⊗C⊗I ⊗I

1
− I ⊗ I ⊗ I ⊗ C2 + I ⊗ I ⊗ C2 ⊗ I
2

2 2
+ I ⊗ C ⊗ I ⊗ I + C ⊗ I ⊗ I ⊗ I
" ∗ ∗ #
· e−z ⊗ e− z ⊗ e−z ⊗ e− z · 4 z (4.117)

where
" ∗ ∗ #
4 z = ez ⊗ e z ⊗ ez ⊗ e z · z (4.118)

Taking the asymptotic limit of Equation (4.117) as βij → ∞, i = j,

4 z∞ = S0 M∞ · 4 z∞  (4.119)

82 EXAMPLES—WHITE COUPLING

where M∞ is found by setting to zero every complex term of the

matrix on the right-hand side of Equation (4.117).
For the two-mode case, we have from Equation (3.42)

0 1
C=  C2 = I  (4.120)
1 0

Substituting Equation (4.120) into Equation (4.117) and taking the

asymptotic limit, MAPLE yields for the lossless case
 
−2 0 0 1 0 0 1 0 0 1 0 0 1 0 0 0
 
 0 −2 0 0 −1 0 0 1 0 0 0 0 0 1 0 0
 
 
 0 0 −2 0 0 0 0 0 −1 0 0 1 0 0 1 0
 
 
 1 0 0 −2 0 0 −1 0 0 −1 0 0 0 0 0 1
 
 
 0 −1 0 0 −2 0 0 1 0 0 0 0 0 1 0 0
 
 
 0 0 0 0 0 −2 0 0 0 0 0 0 0 0 0 0
 
 
 1 0 0 −1 0 0 −2 0 0 0 0 0 −1 0 0 1
 
 
 0 1 0 0 1 0 0 −2 0 0 0 0 0 −1 0 0
M∞ =



 0 0 −1 0 0 0 0 0 −2 0 0 1 0 0 1 0
 
 
 1 0 0 −1 0 0 0 0 0 −2 0 0 −1 0 0 1
 
 
 0 0 0 0 0 0 0 0 0 0 −2 0 0 0 0 0
 
 
 0 0 1 0 0 0 0 0 1 0 0 −2 0 0 −1 0
 
 
 1 0 0 0 0 0 −1 0 0 −1 0 0 −2 0 0 1
 
 
 0 1 0 0 1 0 0 −1 0 0 0 0 0 −2 0 0
 
 
 0 0 1 0 0 0 0 0 1 0 0 −1 0 0 −2 0
 
0 0 0 1 0 0 1 0 0 1 0 0 1 0 0 −2

(4.121)

The 16 differential equations of Equations (4.119) and (4.121) sep-

arate into two independent sets of 8 equations each, as in Section
3.5. The set containing rows 1, 4, 7, 10, 13, and 16 yields the mode
powers and their moments, of present interest. Rows 4, 7, 10, and 13
are equivalent. By eliminating this redundancy, and using Equations
(4.118), (3.64), (3.66), and (3.70), we obtain 3 differential equations
relating the fourth moments of the average mode powers for the
 4.6. NONDEGENERATE CASE—APPROXIMATE RESULTS 83

lossless two-mode case:

     
Q0 z∞ −2 4 0 Q0 z∞
     
 Q z∞  =  1 −4 1  
 1    ·  Q1 z∞  (4.122)
Q4 z∞ 0 4 −2 Q4 z∞

where

Q0 z∞ = P02 z∞ 

Q1 z∞ = P0 zP1 z∞  (4.123)

Q4 z∞ = P12 z∞ 

Equation (4.122) represents the asymptotic limit of Equation (4.43),

providing a valid approximation for large β12 .
Equations (4.122)–(4.123) yield analytic solutions for β/S0 =
∞. For single-mode input, Section 4.3.1, the initial conditions are

Q0 0∞ = 1 Q1 0∞ = 0 Q4 0∞ = 0 (4.124)

and the solutions are


1 1 1
P 0 z = P 1 z = − e−4z + e−6z  (4.125)
12 4 6

For two-mode coherent or incoherent input, Section 4.3.2 or 4.3.3,

the initial conditions are

Q0 0∞ = 1 Q1 0∞ = 1 Q4 0∞ = 1 (4.126)

with solutions

1 1 −6z
P 0 z = P 1 z = − e  (4.127)
3 3

Equations (4.125) and (4.127) yield the β/S0 = ∞ curves for Fig-
ures 4.4 and for Figures 4.5 and 4.6, respectively.
84 EXAMPLES—WHITE COUPLING

4.6.4. Discussion
The asymptotic approximation provides major simplification in the
solutions for transfer function statistics of multi-mode transmission
systems with white coupling. In the two-mode case, the differential
equations for power fluctuations are reduced in number from 16 to
3, and provide a simple closed-form solution.
The physical example of Appendix F, a dielectric slab waveguide
supporting four modes, suggests that this approximation will apply
to many practical guides. Departures from these approximate results
can occur only with strict degeneracy between one or more pairs of
modes, i.e. with mode pairs having identical propagation constants.
Under these special conditions additional non-zero terms in the var-
ious matrices are readily included, with corresponding modifications
to the corresponding differential equations.

4.7. DISCUSSION

The examples of the present chapter illustrate the range of trans-

mission statistics that can be obtained for multi-mode transmission
systems. We assume that all modes travel in the same (forward) di-
rection. The coupling spectrum is assumed white, i.e., the coupling
(due to imperfections) is assumed to have a short correlation dis-
tance; this assumption permits exact results, without approximation
of any kind. The further condition of distinct propagation constants
(the nondegenerate case of Section 4.6) yields simplification, and
excellent agreement with these exact results.
The present methods are indirect, in that the guide is divided
into discrete sections as an intermediate step, permitting the use of
Kronecker products, yielding a straightforward analysis. Kronecker
product methods are used throughout the following chapters in treat-
ments of other problems. We extend the present work to include
arbitrary coupling spectra in Chapter 6.
Successively higher order statistics give increased information
about the transfer functions of multi-mode guide. Average transfer
functions and average mode powers provide no information about
signal distortion, which first appears in the second-order (two-
frequency) transfer function statistics or the equivalent time-domain
result, i.e., the corresponding intensity impulse response.
 REFERENCES 85

The case of an N-mode guide with identical propagation constants

for all modes is treated in Appendix G. This obviously nonphysical
case is of interest in that it permits exact solution without using Kro-
necker products. These results provide confirmation of the present
methods.

REFERENCES

1. S. D. Personick, “Time Dispersion in Dielectric Waveguides,” Bell System

Technical Journal, Vol. 50, March 1971, pp. 843–859.
2. Dietrich Marcuse, Theory of Dielectric Optical Waveguides, 2nd ed., Aca-
demic Press, New York, 1991.
3. Athanasios Papoulis, The Fourier Integral and its Applications, McGraw-
Hill, New York, 1962.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

CHAPTER FIVE

Directional Coupler with

White Propagation Parameters

5.1. INTRODUCTION

We apply the coupled line equations for two forward-traveling

modes, Equations (2.1) and (2.2), to the study of directional cou-
plers with random imperfections. We repeat these equations for
convenience:
I0
z = −0 zI0 z + jczI1 z
(5.1)
I1
z = jczI0 z − 1 zI1 z

0 z = α0 z + jβ0 z 1 z = α1 z + jβ1 z (5.2)

An ideal directional coupler has constant coupling and constant

propagation parameters. We assume in addition that the ideal cou-
pler is lossless and symmetric; then Equations (5.1) and (5.2) become
[1–5]

I0
z = −jβI0 z + jcI1 z

(5.3)
I1
z = jcI0 z − jβI1 z

Backward-traveling modes are neglected. A typical coupler is shown

schematically in Figure 5.1, with inputs

I0 0 = 1 I1 0 = 0 (5.4)

87
88 DIRECTIONAL COUPLER WITH WHITE PROPAGATION PARAMETERS

I0 (z)

I1 (z)

FIGURE 5.1. Directional coupler.

This sketch could represent identical parallel optical fibers, micro-

strips, etc.; the coupling c increases as the two guides are brought
closer together. The solutions to Equations (5.3) with the inputs of
Equations (5.4) are [6]

I0 z = cos cze−jβz  I1 z = j sin cze−jβz  (5.5)

Complete power transfer from mode 0 to mode 1 occurs for z =

π/2c; this can occur only for a symmetric coupler, with identical
propagation constants, and constant coupling, as in Equation (5.3).
A real coupler will have random variations in geometric and ma-
terial parameters that will cause the coupling cz and the propa-
gation parameters 1 z and 2 z of Equations (5.1) and (5.2) to
depart from their design values. For a symmetric coupler, βz =
β0 z − β1 z will vary randomly around zero, and cz will vary
randomly about c of Equation (5.3).
It is plausible that the variations in βz will have the most sig-
nificant effect on the transfer functions of the coupler. The statistical
model of the following section assumes constant coupling and inde-
pendent white propagation parameters for the two coupled guides
[7]. The results of the present chapter are thus complementary to
those of Chapters 3 and 4 for two-mode guides with white coupling
and constant β.

5.2. STATISTICAL MODEL

Assume that the propagation parameters in Equations (5.1) and (5.2)

are lossless, statistically independent, and stationary:

β0 z = β + b0 z
(5.6)
β1 z = β + b1 z
 5.2. STATISTICAL MODEL 89

β is the common mean value of the two propagation constants, the

design value of Equation (5.3):

β = β0 z = β1 z (5.7)

The a.c. components of the propagation parameters, b0 z and b1 z

have zero mean, the same covariance, and zero cross-variance:

b0 z = b1 z = 0

Rb ζ = b0 z + ζb0 z = b1 z + ζb1 z (5.8)

b0 z + ζb1 z = 0

They have spectral density

∞
Bν = Rb ζe−j2πνζ dζ (5.9)
−∞

and zero cross-spectrum.

We assume the coupling coefficient of Equation (5.1) is constant,
equal to its design value of Equation (5.3):

cz = c (5.10)

We find the average transfer functions and average powers for

white propagation parameters:

Rb ζ = B0 δζ Bν = B0  (5.11)

While the white assumption is nonphysical, in that it corresponds

to an infinite variance for the propagation parameters, Chapter 6
shows that the results for white coupling are a good approximation
for almost white coupling. We expect the present results for white
Bν of Equation (5.11) will provide a good approximation for low-
pass Bν, under appropriate restrictions.
Let the bandwidth of Bν be denoted by νb , with the correspond-
ing correlation length λb :

1
λb =  (5.12)
νb
90 DIRECTIONAL COUPLER WITH WHITE PROPAGATION PARAMETERS

We divide the coupler into sections z. Then, the following assump-

tions must be satisfied for physical systems:

1. Independent sections.

1
z λb =  (5.13)
νb

2. Small phase shift per section, Equation (2.29).

k z
φk = βzdz (5.14)
k−1 z

Then following the footnote of Equation (3.28), from Equa-

tion (5.8)

k z
φ2k  = b20 z + b21 z dz ≈ 2B0 z  1 (5.15)
k−1 z

3. Small coupling.

c z  1 (5.16)

4. Small fractional variation in β0 z, β1 z.

2B0 νb  β2  (5.17)

5. No coupling to backward modes.

νb  β (5.18)

Equations (5.13) and (5.16) yield

1
c  νb  (5.19)
z
Equations (5.13) and (5.15) yield

1
2B0   νb  (5.20)
z
Equations (5.18) and (5.20) yield Equation (5.17).
 5.3. AVERAGE TRANSFER FUNCTIONS 91

Therefore the physical restrictions on the parameters of the fol-

lowing analysis based on the white spectrum assumed in Equation
(5.11) for the a.c. components of the propagation parameters are as
follows:
1
c  νb  β
z
(5.21)
1
2B0   νb  β
z
These conditions correspond to a symmetric coupler with imper-
fections that vary slowly compared to the optical wavelength, vary
rapidly compared to the coupler length, and are sufficiently small
that little relative phase shift occurs between the two modes in the
correlation length of the imperfections.

5.3. AVERAGE TRANSFER FUNCTIONS

An elementary section of a lossless symmetric coupler is described by

substituting Equations (5.6) and (5.10) into Equations (2.26)–(2.28):
   k z k z 
 k z
G e−j k−1 z b0 zdz cos c z e−j k−1 z b1 zdz j sin c z
0
 = k z k z

 k z
G e−j k−1 z b0 zdz
j sin c z e−j k−1 z b1 zdz
cos c z
1
 
 k − 1 z
G 0
·  (5.22)

G1 k − 1 z

 z and
We have defined for convenience the auxiliary quantities G 0

G1 z as follows:
  
I0 z  z
G 0
= e−jβz   (5.23)
I1 z  z
G 1

where the tildes distinguish these quantities from those of Equa-

tion (2.31). Define the expected values of the complex mode ampli-
tudes as

I i z = Ii z   z

G i z = G i = 0 1 (5.24)
i
92 DIRECTIONAL COUPLER WITH WHITE PROPAGATION PARAMETERS

The different sections of Equation (5.22) are independent for white

propagation parameters, Equation (5.11). Take the expected value
of Equation (5.22), expand the matrix elements in Taylor series for
small z, and use Equation (5.15):
 
 B0 

G 0 k z  1 − z jc z   G 0 k − 1 z
2
= 
·
  

G 1 k z B0 G k − 1 z
jc z 1− z 1
2
(5.25)

From Equations (D.8) and (D.14), for small z the difference equa-
tions of Equation (5.25) are approximated by the differential
equations
 

 B0

G 0 z −2 jc    G 0 z
 = ·  (5.26)


 B0  
G 1 z jc − G 1 z
2
Assume the initial conditions of Equation (5.4) and Figure 5.1:

I 0 0 = 
G 0 0 = 1 I 1 0 = 
G 1 0 = 0 (5.27)

Then the solutions to Equation (5.26) are

 

G 0 z cos cz
= e−B0 /2z (5.28)

G 1 z j sin cz

or, using Equations (5.23) and (5.24)

 
I 0 z cos cz
= e− jβ+B0 /2 z  (5.29)
I 1 z j sin cz

These results are an exponentially damped version of the results

for an ideal coupler, given in Equation (5.5), the damping increas-
ing as the variations in the propagation parameters increase. For
B0 = 0, Equation (5.29) is identical to Equation (5.5), with com-
plete power transfer at z = π/2c. For an imperfect coupler, with
 5.4. COUPLED POWER EQUATIONS 93

B0 > 0, I0 π/2c  = 0, but this of course does not imply com-
plete power transfer, since the angles  I0 π/2c and  I1 π/2c
are random variables.
For zero coupling, c = 0, Equation (5.29) becomes
 
I 0 z 1
= e− jβ+B0 /2 z  (5.30)
I 1 z 0

For this case the solution to Equations (5.1), (5.2), (5.4), and (5.6) is
  −j  z b xdx
I0 z e 0 0
= e−jβz  (5.31)
I1 z 0

The exponent in Equation (5.31) will be Gaussian if b0 z is Gaus-

sian; this exponent will be approximately Gaussian in many other
cases by the assumption of Equation (5.13). Then the expected value
of Equation (5.31) follows directly from the characteristic function
of a Gaussian random variable, yielding Equation (5.30). Thus, the
expected complex mode amplitude of mode 0, I0 z, decays ex-
ponentially because the total phase shift performs a random walk,
with variance proportional to distance z; the power in mode 0 re-
mains constant, I0 z2  = 1. The complex amplitude of mode 1 is
identically zero, I1 z = 0, with expected value and power both zero.

5.4. COUPLED POWER EQUATIONS

The average powers and cross-powers are defined in Equation (3.50).

Using this relation and Equation (5.23),
 
P 0 z
 
 P 01 z  ∗
z =    =  z ⊗  ∗ z = 
z ⊗ 
 z (5.32)

 P 10 z 
P 1 z

where from Equation (5.24)

 
I 0 z 
G 0 z
 z = = Iz 
z = 
= Gz (5.33)
I 1 z 
G 1 z
94 DIRECTIONAL COUPLER WITH WHITE PROPAGATION PARAMETERS

with
  
I0 z  z
G 0
Iz =  
Gz =  (5.34)
I1 z  z
G 1

Substitute Equation (5.22), use Equations (D.15) and (D.17), expand

the resulting matrix elements for small z, use Equation (5.15), and
let z → 0 to yield the coupled power equations:

    
P 0 z 0 −jc +jc 0 P 0 z

    
 P 01 z   −jc −B0 0 +jc   P 01 z 
 = ·  (5.35)

    
P
 10 z   +jc 0 −B 0 −jc   10P z 
P
1 z 0 +jc −jc 0 P 1 z
For two modes with white coupling, the powers and cross-powers in
Equations (3.52)–(3.54) were decoupled. Here, for random propa-
gation parameters, they are inextricably entwined.
The initial conditions corresponding to Equations (5.4) and (5.27)
become
   
P 0 0 1
   
 P 01 0   0 
  =   (5.36)
   
P
 10 0  0
P 1 0 0
Equations (5.35) and (5.36) yield the following results for the aver-
age powers and cross-powers:
P 0 z = 1 − P 1 z 



 

  2 

1 1 −B0 /2z B0
= + e cos 2c 1 − ·z
2 2 
 4c




   2 
B0 

sin 2c 1 − ·z 

4c 

B0
+   2  (5.37)
4c 

B0 

1− 

4c
 5.5. DISCUSSION 95
  
 2
B0
sin 2c 1 − · z
4c
j
P 01 z = P ∗10 z = − e−B0 /2z   2  (5.38)
2 B0
1−
4c
For B0 /4c > 1, the trigonometric functions may be rewritten as
hyperbolic functions. For the limiting case,

1 1 −2cz
P 0 z = 1 − P 1 z = + e 1 + 2cz  B0 = 4c (5.39)
2 2
P 01 z = P ∗10 z = −jcze−2cz  B0 = 4c (5.40)

Equation (5.37) is plotted in Figure 5.2, which shows the power in

the excited mode of Figure 5.1 vs. distance for various propagation
parameter imperfections.
With imperfections absent B0 /c = 0 we obtain the result for the
ideal coupler, given in Equation (5.5). As B0 /c increases the behavior
changes from oscillatory to monotonic. For large z, the power divides
equally between the two modes, except for the ideal case (B0 = 0).
Since the coupler has been assumed lossless, a plot for the power in
the unexcited mode, P 1 z, is obtained as the mirror image of Figure
5.2 about the ordinate 0.5. P 0 z and P 1 z, the total average powers
in the two coupled guides, contain both coherent and incoherent
components; the coherent components are the square magnitude of
the results in Equation (5.29):
 
I 0 z2 cos2 cz
−B0 z
=e  (5.41)
I 1 z2 sin2 cz

Thus, in Figure 5.2, P 0 z is completely coherent at z = 0, com-

pletely incoherent as z → ∞.

5.5. DISCUSSION

We have found the expected response and average powers for a

symmetric, uniform coupler with white random propagation param-
eters and constant coupling. The results are exact, and provide a
96 DIRECTIONAL COUPLER WITH WHITE PROPAGATION PARAMETERS

B0
=
c
1.0

10
0.8

0.6

P0 (z) = 1 – P1 (z)

1
0.4

0.2

0
0 1 2 3 4 5
2 · cz

FIGURE 5.2. Average power vs. distance for excited mode of symmetric
coupler.

good approximation for the almost white case. The analysis uses
Kronecker product methods, similar to those of Chapter 3 for white
coupling and constant propagation parameters. For two modes with
white coupling, the average powers were decoupled from the cross-
powers. Here the cross-powers are essential in the analysis.
These results are appropriate for investigating the necessary tol-
erances in two-branch couplers or switches, for which B0 /c  1, and
 REFERENCES 97

possible power dividers, where B0 /c ≈ 4. Higher order statistics may

be required to investigate the utility of the latter application.
The present methods may be applied to other related problems:
correlated propagation parameters and coupling, asymmetric cou-
plers (with different average propagation parameters for the two
coupled guides), couplers with loss, multi-branch couplers, transfer-
function and other higher order statistics, time-response statistics,
etc.

REFERENCES

1. W. H. Louisell, Coupled-Mode and Parametric Electronics, Wiley, New

York, 1960.
2. Theodor Tamir, Ed., Guided-Wave Optoelectronics, Springer-Verlag, New
York, 1988.
3. Dietrich Marcuse, Theory of Dielectric Optical Waveguides, 2nd ed., Aca-
demic Press, New York, 1991.
4. Hermann A. Haus and Weiping Huang, “Coupled-Mode Theory,” Pro-
ceedings of the IEEE, Vol. 79, October 1991, pp. 1505–1518.
5. Wei-Ping Huang, “Coupled-mode theory for optical waveguides: an
overview,” J. Opt. Soc. Am. A, Vol. 11, March 1994, pp. 963–983.
6. S. E. Miller, “Coupled Wave Theory and Waveguide Applications,” Bell
System Technical Journal, Vol. 33, May 1954, pp. 661–719.
7. Harrison E. Rowe and Iris M. Mack, “Coupled modes with ran-
dom propagation constants,” Radio Science, Vol. 16, July–August 1981,
pp. 485–493.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

CHAPTER SIX

Guides with General

Coupling Spectra

6.1. INTRODUCTION

A variety of transmission statistics for two-mode and multi-mode

guides having random coupling or propagation parameters with
white spectra have been presented in Chapters 3–5. All of these re-
sults are exact; they provide a standard against which to compare
approximate results for general imperfection spectra.
The matrix methods employing Kronecker products, used in ear-
lier chapters, depend on dividing the guide into statistically inde-
pendent sections of length
z. For white coupling or propagation
parameters,
z can approach zero, different sections remain strictly
independent, the analysis for each section becomes exact, and linear
differential equations for the various transmission statistics follow
directly from the matrix products. No perturbation or other approx-
imations enter the analysis.
In order to apply Kronecker methods for general imperfection
spectra, we again require the guide to be divided into independent
sections. Now the section length
z can no longer approach zero,
but must be long compared to the correlation length of the random
imperfections. The different sections are no longer strictly indepen-
dent, but only approximately so. Exact analysis of the individual sec-
tions is no longer possible; we must now use perturbation theory
for this purpose. Hence, the coupling is restricted to be small; the

99
100 GUIDES WITH GENERAL COUPLING SPECTRA

longer the correlation length, i.e., the narrower the bandwidth of

the coupling or propagation parameter spectra, the smaller must be
the coupling. The transmission statistics, expressed as matrix powers,
are only approximate. Finally, differential equations for the approxi-
mate transmission statistics may not necessarily be obtained directly
by setting
z → 0 in the matrix results.
We derive the coupled power equations for two-mode and N-
mode guide with constant propagation parameters, for almost-white
and for general non-white coupling spectra, in the present chapter.
The range of validity of the present results is indicated. Other results
of preceding chapters may be similarly generalized.

6.2. ALMOST-WHITE COUPLING SPECTRA

The covariance and spectrum of the coupling coefficient cz are

defined as the left-hand equalities of Equations (3.1) and (3.2):

Rc ζ = cz + ζcz (6.1)

∞
Sν = Rc ζe−j2πνζ dζ (6.2)
−∞

A typical low-pass coupling spectrum is shown in Figure 6.1.

S( )

FIGURE 6.1. Low-pass coupling spectrum.

 6.2. ALMOST-WHITE COUPLING SPECTRA 101

The cutoff spatial frequency νc gives a rough indication of the

coupling bandwidth. The corresponding correlation distance is
1
λc =  (6.3)
νc
The covariance will be small for ζ λc ; consequently cz and
cz + ζ will be almost uncorrelated, and for Gaussian and many
other cz almost independent, for sufficiently large ζ We assume
that guide sections of length

z λc (6.4)
will be approximately independent, as required by the present Kro-
necker product methods.
In addition, we use the discrete approximation of Sections 2.4 and
2.6. The relative phase shift and relative attenuation between every
pair of modes must therefore be small in length dz.

6.2.1. Two Modes

From Equations (2.30) and (6.3) and (6.4) [1],
1 1 1 2π

z 
z  (6.5)
νc
α νc
β
Therefore,

β
νc
α νc  (6.6)
2π
A physical description of the almost-white case may be given in
two equivalent ways in the lossless case: 1. cz must vary rapidly in
a beat wavelength 2π/
β  2. The relative phase shift must be small
in a correlation interval λc . These assumptions were previously used
in Equations (4.65)–(4.68).
Equation (3.27) for the average transfer function, and the average
of Equations (3.48) and (3.49) for the average powers and cross-
powers, involve the statistical averages sin ck , cos ck , sin2 ck ,
cos2 ck , and sin ck cos ck , where ck is given by Equation (2.26).
We assume ck has a symmetric probability density, consistent with
our earlier assumption that ck has zero mean. Then,
ck  = 0 sin ck  = 0 sin ck cos ck  = 21 sin 2ck  = 0 (6.7)
102 GUIDES WITH GENERAL COUPLING SPECTRA

Under the present restrictions of Equation (6.5), ck will be approx-

imately Gaussian in many cases. Then, extending the footnote to
Equation (3.28),

k
z
k
z
ck2  = Rc z − z dz dz
k−1
z k−1
z

k
z
∞
≈ dz Rc ζdζ = S0
z (6.8)
k−1
z −∞

For ck approximately Gaussian,

1
cos ck  ≈ e− 2 S0
z  (6.9)

cos2 ck  ≈ 21 1 + e−2S0
z  sin2 ck  ≈ 21 1 − e−2S0
z  (6.10)

For simplicity, we restrict further analysis to the small coupling case.

Expanding the trigonometric functions in Taylor series:

cos ck  ≈ 1 − 21 ck2  ≈ 1 − 21 S0
z S0
z 1 (6.11)

cos2 ck  ≈ 1 − ck2  ≈ 1 − S0
z

(6.12)
sin2 ck  ≈ ck2  ≈ S0
z S0
z 1

Equation (6.5) yields the necessary condition

1
S0 = S0λc 1 (6.13)
νc
The condition of Equation (6.13) is well satisfied for the numerical
parameters of Equation (4.67).
Equations (6.5), (6.7), and (6.11) substituted in Equation (3.27)
yield the following difference equations for the average transfer
functions, where we have dropped explicit notation indicating that
these results are approximate:
 
S0
I 0 k
z − I 0 k − 1
z = − 0 + I 0 k − 1
z
z
2
 
S0
I 1 k
z − I 1 k − 1
z = − 1 + I 1 k − 1
z
z
2
S0
z 1 (6.14)
 6.2. ALMOST-WHITE COUPLING SPECTRA 103

Equations (6.5), (6.7), and (6.12) substituted in the expected value

of Equations (3.48) and (3.49) yield the following matrix difference
equations for the average powers and cross-powers:
   
P 0 k
z 1 − 2α0 + S0
z S0
z
=
P 1 k
z S0
z 1 − 2α1 + S0
z
 
P 0 k − 1
z
·  S0
z 1 (6.15)
P 1 k − 1
z
   
P 01 k
z 1 − 0 + ∗1 + S0
z S0
z
=
P 10 k
z S0
z 1 − ∗0 + 1 + S0
z
 
P 01 k − 1
z
·  S0
z 1 (6.16)
P 10 k − 1
z

z is restricted by Equation (6.5) in the present almost-white case;

it can no longer approach 0 as for strictly white coupling in Sections
3.3 and 3.4. Nevertheless, in the small coupling case Appendix I
shows that useful approximations are obtained by taking the limit as

z → 0 of Equations (6.14)–(6.16), to yield the following differential
equations.
Average transfer functions:
 
S0
I 0 z = − 0 + I 0 z
2
  S0
z 1 (6.17)
S0
I 1 z = − 1 + I 1 z
2
Coupled power equations:
P 0 z = −2α0 + S0P 0 z + S0P 1 z
S0
z 1 (6.18)
P 1 z = S0P 0 z − 2α1 + S0P 1 z
Cross-powers:
P 01 z = −0 + ∗1 + S0P 01 z + S0P 10 z
S0
z 1
P 10 z = S0P 01 z − ∗0 + 1 + S0P 10 z
(6.19)
104 GUIDES WITH GENERAL COUPLING SPECTRA

Equations (6.17)–(6.19) for almost-white coupling are similar to the

corresponding results of Equations (3.29), (3.53), and (3.54) for
strictly white coupling with S0 replaced by S0
For large coupling, we must use Equations (6.9) and (6.10) in
place of Equations (6.11) and (6.12) to obtain difference equations
for the average transfer functions, average powers, and cross-powers.
Equivalent differential equations are obtained by the more compli-
cated procedure given in Appendix I, but differ from those of Equa-
tions (6.17)–(6.19).

6.2.2. N Modes
Equations (2.66) for constant propagation parameters, and Equa-
tions (6.3) and (6.4) yield

1 1 1 2π

z 
z  0 ≤ i j ≤ N − 1
νc
αij νc
βij
(6.20)

We require the averages ejck C  and ejck C ⊗ e−jck C  in Equations

(3.37) and (3.58)–(3.59), where the coupling coefficient matrix C is
given by Equation (3.35). Toward this end, diagonalize C as follows:

C = L ·
· L−1  (6.21)

where
 

λ0
0 
 
 λ1 
 

=   (6.22)
  
 
 
0 λN−1

Then,

ck C = L · ck
· L −1  (6.23)
 6.2. ALMOST-WHITE COUPLING SPECTRA 105
 

ejck λ0
0 
 
 ejck λ1 
  −1
ejck C =L·  ·L  (6.24)
  
 
 
0 ejck λN−1

Under the assumptions of the previous section, ck is approximately

Gaussian with zero mean; therefore from Equation (6.8)
 
0
1 2
e− 2 λ0 S0
z
 
 − 21 λ21 S0
z 
 e 
  −1
ejck C  = L ·  ·L 
  
  
 
0 1 2
e− 2 λN−1 S0
z
(6.25)

For small coupling, we may expand the exponentials to obtain

ejck C 
 

1 − 21 λ20 S0
z
0 
 
 1 − 21 λ21 S0
z 
  −1
=L·  ·L
  
 
 
0 1 2
1 − 2 λN−1 S0
z
 

λ20 0 
 
1  λ21 
  −1
= I − S0
zL ·  ·L
2   
  
 
0 λ2N−1
1 
= I − S0
zC2  S0
zλ2i max 1 (6.26)
2
106 GUIDES WITH GENERAL COUPLING SPECTRA

where I is the N × N unit matrix. Alternatively, we may use the left-

hand equality of Equation (3.40) with Equation (6.8) to obtain this
small-coupling result directly. Equation (6.26) may be substituted
into Equation (3.37), and the limit taken as
z → 0 (Appendix I),
to yield the result of Equation (3.41) with S0 replaced by S0
Utilizing the results of Appendix C, Equations (6.8) and (6.24)
yield

ejck C ⊗ e−jck C  = L ⊗ L · D · L ⊗ L−1  (6.27)

where D is a diagonal matrix, with elements

1 2
e− 2 λi −λj  S0
z  0 ≤ i j ≤ N − 1 (6.28)

For small coupling, we may substitute Equation (6.8) into Equation

(3.60) and take the limit as in the preceding paragraph, obtaining
finally the result of Equation (3.61) with S0 → S0.
To summarize, for a low-pass coupling spectrum satisfying the con-
ditions of Equation (6.20) and the small-coupling condition of the
final line of Equation (6.26), the average transfer functions and cou-
pled power equations are obtained by the substitution S0 → S0 in
the corresponding results for white coupling. This holds true also for
the approximate results of Section 4.6 for the nondegenerate case,
i.e., for distinct
βij . Large coupling requires the use of Equations
(6.25) and (6.27)–(6.28), and the extended procedure of Appendix I.

6.3. GENERAL COUPLING SPECTRA—LOSSLESS CASE

We now consider general coupling spectra with small coupling. Rep-

resentative spectra are shown in Figures 6.1 and 6.2 for low-pass and
band-pass coupling, respectively.
The guide is divided in to sections of length
z long compared to
the coupling correlation length λc , Equation (6.4). The guide trans-
mission properties are described by perturbation theory, Sections 2.5
and 2.6 for two-modes and N-modes, respectively.
For two modes,
     
G0 z 1 + H0 jH01 G0 z −
z
= ·  (6.29)
G1 z jH10 1 + H1 G1 z −
z
 6.3. GENERAL COUPLING SPECTRA—LOSSLESS CASE 107

S( )

2 c

FIGURE 6.2. Band-pass coupling spectrum.

     
I0 z e−0
z 1 + H0  e−1
z e−
z jH01 I0 z −
z
= · 
I1 z e−0
z e+
z jH10 e−1
z 1 + H1  I1 z −
z
(6.30)

where


z
z−ζ
H0 = − e+
ζ dζ cxcx + ζdx
0 z−
z

z
H01 = cxe+
x dx
z−
z

z (6.31)
H10 = cxe−
x dx
z−
z


z
z−ζ
H1 = − e−
ζ dζ cxcx + ζdx
0 z−
z

Small coupling requires

z
cx dx 1 (6.32)
z−
z

This suggests the condition



∞
Sνdν
z 1 (6.33)
−∞
108 GUIDES WITH GENERAL COUPLING SPECTRA

For N modes,

Iz = e−z · Gz (6.34)

Gz = G · Gz −
z (6.35)

Iz = e−z · G · e+z−
z · Iz −
z (6.36)

where

z
G =I +j cxex · C · e−x dx
z−
z

z
x
− cxex · C · e−x dx cyey · C · e−y dy (6.37)
z−
z z−
z

For the off-diagonal terms,

z 
N
Gij = jCi−1j−1 cxe
i−1j−1 x dx − Ci−1k−1 Ck−1j−1
z−
z k=1


z
z−ζ

i−1k−1 ζ
· e dζ cxcx + ζe
i−1j−1 x dx
0 z−
z

1 ≤ i j ≤ N i = j (6.38)

The diagonal terms are


N

z
Gii = 1 − Ci−1k−1 Ck−1i−1 e
i−1k−1 ζ dζ
k=1 0

z−ζ
· cxcx + ζdx 1 ≤ i ≤ N (6.39)
z−
z

The small coupling condition becomes

z
cx dxC 1 (6.40)
z−
z

where C denotes the matrix norm, Equation (A.25). The restric-
tion corresponding to Equation (6.33) becomes


∞
Sνdν
zC 1 (6.41)
−∞
 6.3. GENERAL COUPLING SPECTRA—LOSSLESS CASE 109

We consider the lossless case,

αij = 0 0 ≤ i j ≤ N − 1 (6.42)

throughout the rest of the present section.

6.3.1. Two Modes

The average transfer functions are found from the expected values
of Equations (6.30)–(6.31). Substituting Equations (6.1)–(6.2) and
(6.7), in the lossless case


z   
−jβ0
z ζ +j
βζ
I 0 z = e 1 −
z Rc ζ 1 − e dζ I 0 z −
z
0
z
(6.43)


z   
−jβ1
z ζ −j
βζ
I 1 z = e 1 −
z Rc ζ 1 − e dζ I 1 z −
z
0
z
(6.44)

where we use the notation of Equation (3.26) for the average com-
plex mode amplitudes, I i z = Ii z. By Equation (6.4), Rc ζ is
narrow compared to 1 − ζ/
z, and we have approximately
      
−jβ0
z 1
β 
β
I 0 z = e 1− S + jS
z I 0 z −
z (6.45)
2 2π 2π
      
−jβ1 z 1
β 
β
I 1 z = e 1− S − jS
z I 1 z −
z (6.46)
2 2π 2π


Sν is the Hilbert transform of the coupling spectrum Sν of Equa-
tion (6.2),

 1 1 1
∞ Sµ
Sν = Sν  = dµ (6.47)
π ν π −∞ ν − µ

where  represents the convolution operator. Then, by the approxi-

mation of Appendix I,
    
1
β 
β
−jβ0 z − 2 z S 2π +j S 2π
I 0 z = e e I 0 0 (6.48)
110 GUIDES WITH GENERAL COUPLING SPECTRA
    
1
β 
β
−jβ1 z − 2 z S 2π −j S 2π
I 1 z = e e I 1 0 (6.49)

The coupled power equations are found by taking the Kronecker

product of Equation (6.30) with its complex conjugate, using Equa-
tion (6.31), and taking the expected value of the result. Recalling the
notation of Equation (3.50), we obtain the following approximation
in the lossless case:
     
P 0 z B11 0 0 B14 P 0 z −
z
     
 P 01 z   0 B22 B23 0   P 01 z −
z 
 = ·  (6.50)
     
 P 10 z   0 B32 B33 0   P 10 z −
z 
P 1 z B41 0 0 B44 P 1 z −
z

where

 

z ζ
B11 = B44 = 1 −
z Rc ζ 1 − e+j
βζ dζ
−
z
z


z   (6.51)
ζ
B14 = B41 =
z Rc ζ 1 − e+j
βζ dζ
−
z
z
The average powers P 0 , P 1 are uncoupled from the cross-powers
P 01  P 10 ; we confine our current interest to the average powers.
Equation (6.4) permits neglect of the factor 1 −  ζ /
z in Equa-
tion (6.51), and the approximation of Appendix I yields the coupled
power equations for two modes:
     
 
β
β  
P 0 z  −S 2π S
2π  P 0 z
 
=     ·  (6.52)
P 1 z 
β
β  P 1 z
S −S
2π 2π

We omit the corresponding treatment of cross-powers.

6.3.2. N Modes
We retain the assumption of Equation (6.4) requiring guide sections
to be approximately independent:

z λc  (6.53)
 6.3. GENERAL COUPLING SPECTRA—LOSSLESS CASE 111

We seek coupled power equations for the non-degenerate case of

Section 4.6, for large
βij . For this purpose, we require the addi-
tional assumption
2π

z  0 ≤ i j ≤ N − 1 (6.54)

βij

i.e., the guide sections contain many beat wavelengths.

For the average transfer functions, take the expected value of
Equation (6.35):

z = Gz = G · z −
z (6.55)

where G is the expected value of Equation (6.37), with elements

obtained as the expected values of Equations (6.38) and (6.39). Us-
ing Equations (6.1)–(6.2) and (6.7), we have the following in the
lossless case:

N
Gij = − Ci−1k−1 Ck−1j−1
k=1

 −j
βi−1j−1 ζ


z
j
βi−1j−1 z e − e−j
βi−1j−1
z
·
z Rc ζ e
0 j
βi−1j−1
z
· ej
βi−1k−1 ζ dζ 1 ≤ i j ≤ N i = j (6.56)

 

N
z ζ
Gii = 1 − Ci−1k−1 Ck−1i−1
z Rc ζ 1 − ej
βi−1k−1 ζ dζ
k=1 0
z
1 ≤ i ≤ N (6.57)

The off-diagonal terms of Equation (6.56) may be neglected by

Equation (6.54). The factor 1 − ζ/
z in Equation (6.57) may be
neglected by Equation (6.53), as in the corresponding derivation of
Equations (6.45) and (6.46) for two modes. Therefore the following
approximations hold in the non-degenerate case, indicated by the
subscript ∞ as in Section 4.6:
    
N
1
βi−1k−1
βi−1k−1
Gii∞ = 1 − 2
Ci−1k−1 S 
+ jS
z
k=1
2 2π 2π
(6.58)
112 GUIDES WITH GENERAL COUPLING SPECTRA

Gij∞ = 0 i = j (6.59)

The expected value of Equation (6.35) yields

z∞ = G∞ · z −
z∞  (6.60)

Changing notation,

G∞ = I − H
z (6.61)

where H is a diagonal matrix

 


H1
0 

 H2 
H= 

 (6.62)
  

 
0 HN

with elements
    

N
1
βi−1k−1
βi−1k−1
Hi = 2
Ci−1k−1 S 
+ jS  (6.63)
k=1
2 2π 2π

From Appendix I, the corresponding approximate differential equa-

tion is

 z∞ = −H · z∞  (6.64)

and from Equation (6.34)

 z∞ = −j + H ·  z∞  (6.65)

 is the propagation matrix of Equation (2.58) in the lossless case.

The expected complex amplitudes of the individual modes are

I i z∞ = Ii z∞ = Ii 0e−jβi z

     
 2
1 N−1
βik
βik
· exp − z C S 
+ jS 
2 k=0 ik 2π 2π
0 ≤ i ≤ N − 1 (6.66)
 6.3. GENERAL COUPLING SPECTRA—LOSSLESS CASE 113

The coupled power equations are obtained as the Kronecker prod-

uct of Equation (6.34) with its complex conjugate. We seek the
asymptotic results for large
βij , Equation (6.54). We use the no-
tation of Section 4.6.2. From Equations (4.91) and (4.101),
 ∗ 
z = e−z ⊗ e− z · z (6.67)

From Equation (4.95),

z = Gz ⊗ G∗ z (6.68)

Substituting Equation (6.35),

z = M · z −
z (6.69)

where

M = G ⊗ G∗  (6.70)

and G is given by Equations (6.37)–(6.39). We now partition the

matrices z by Equation (4.106), and M as in Equation (4.108):
 
M 11 M 12 M 13 ··· M 1N
 
 M 21 M 22 M 23 ··· M 2N 
 
 
M =  M 31 M 32 M 33 ··· M 3N  (6.71)
 
      
      
M N1 M N2 M N3 ··· M NN

The submatrices M ij are given by1

M ij = Gij G∗  (6.72)

The subvectors of z, defined by Equation (4.107), are obtained

from Equations (6.69) and (6.71):


N
i z = M ij · j z −
z (6.73)
j=1

1
Gij represents the ij element of the matrix G, previously denoted by the slightly
different notation of Equation (6.38).
114 GUIDES WITH GENERAL COUPLING SPECTRA

We seek the asymptotic properties of the Mij as in Section 4.6.2,

using similar calculations to those of Equations (6.56) and (6.57).
Consider the expected value of the off-diagonal submatrix M ij ; its
k element is

M ij k  = Gij G∗k  i = j (6.74)

All of the elements are small except for k = i  = j; we have

approximately

M ij k ∞ = 0 i = j k = i and/or  = j (6.75)

 
2

βi−1j−1
M ij ij ∞ = Ci−1j−1 S
z i = j (6.76)
2π

Consider next the expected value of the diagonal submatrix M ii ; its

k element is

M ii k  = Gii G∗k  (6.77)

The off-diagonal elements are small, and neglected in the nonde-

generate approximation:

M ii k ∞ = 0 k =  (6.78)

For the diagonal elements,

    

N
1
βi−1k−1
βi−1k−1
M ii  ∞ = 1 − 2
Ci−1k−1 S 
+ jS
z
k=1
2 2π 2π
    
N
1
β−1k−1
β−1k−1
− 2
C−1k−1 S 
− jS
z
k=1
2 2π 2π
(6.79)

The special case  = i will be the only one of interest in determining

the average powers:
 

N
2

βi−1k−1
M ii ii ∞ = 1 − Ci−1k−1 S
z (6.80)
k=1
2π
 6.4. GENERAL COUPLING SPECTRA—LOSSY CASE 115

Consider the ith component of the subvector i z of Equation

(6.73). Recalling the notation of Equation (4.107),


N
Ri−1 z∞ = M ii ii ∞ Ri−1 z −
z∞ + M ij ij ∞ Rj−1 z −
z∞ 
j=1
j=i
(6.81)
Substituting Equations (6.76) and (6.80),
   

N
2

βi−1k−1
Ri−1 z∞ = 1 − Ci−1k−1 S
z Ri−1 z −
z∞
k=1
2π
N   
 2
βi−1j−1
+ Ci−1j−1 S
z Rj−1 z −
z∞  (6.82)
j=1
2π

Relabeling the summation indices,

 

N−1
2
βik
P i z∞ = P i z −
z∞ − Cik S P i z −
z∞
z
k=0
2π
 

N−1
2
βik
+ Cik S P k z −
z∞
z (6.83)
k=0
2π

Finally, from Appendix I the coupled power equations are given

as follows in the nondegenerate case [2]:
 

N−1
2
βik
P i z∞ = − Cik S P i z∞
k=0
2π
 
 2
N−1

βik
+ Cik S P k z∞  0 ≤ i ≤ N − 1 (6.84)
k=0
2π

6.4. GENERAL COUPLING SPECTRA—LOSSY CASE

Finally, we consider general coupling spectra for lossy media with

small coupling. Equations (6.4) and (6.29)–(6.41) remain valid, but
Equation (6.42) no longer holds.
116 GUIDES WITH GENERAL COUPLING SPECTRA

6.4.1. Two Modes

The expected values of Equations (6.30)–(6.31) yield the following
generalizations of the lossless results, Equations (6.48)–(6.49):
    
1
β 
β
−0 z − z S 0 +j S 0
2 2π 2π
I 0 z = e e I 0 0 (6.85)
    
1
β 
β
−1 z − 2 z S1 2π −j S1 2π
I 1 z = e e I 1 0 (6.86)

where

∞
S0 ν = Rc ζe−
α ζ e−j2πνζ dζ (6.87)
−∞

∞
S1 ν = Rc ζe+
α ζ e+j2πνζ dζ (6.88)
−∞

and we assume that the signal mode has lower loss, i.e.,
α = α0 −
α1 < 0. Note that S0 ν = S1 ν, in contrast to the lossless case.
Rc ζ must decay more rapidly than e+
α ζ grows for large ζ , in
order for Equation (6.88) to hold.
For the lossy coupled power equations, Equation (6.51) generalize
as follows. For the diagonal elements,


z   
−2α0
z −
α ζ ζ +j
βζ
B11 = e 1 −
z Rc ζe 1− e dζ 
−
z
z


z    (6.89)
−2α1
z +
α ζ ζ +j
βζ
B44 = e 1 −
z Rc ζe 1− e dζ 
−
z
z
Since Rc ζ is narrow by Equation (6.4), we have the usual approx-
imations, neglecting the factor 1 −  ζ /
z in the integrands of
Equation (6.89):
   
−2α0
z
β
B11 = e 1 − S0
z 
2π
    (6.90)
−2α1
z
β
B44 = e 1 − S1
z 
2π
where S0   and S1   are given by Equations (6.87) and (6.88). Again,
the product Rc ζe+
α ζ must decay for large ζ in order for this
result for B44 to remain valid.
 6.4. GENERAL COUPLING SPECTRA—LOSSY CASE 117

For the off-diagonal elements,

z 1 − e−2
α
z1−
z 
ζ

−2α1
z −
α ζ j
βζ

B14 = e
z Rc ζe e dζ
−
z 2
α
z
(6.91)


z 1−e +2
α
z 1−
z
ζ

B41 = e−2α0
z
z Rc ζe+
α ζ ej
β·ζ dζ
−
z −2
α
z

These quantities can be expressed simply only for small differential

loss in the section length
z. In the low-loss case,

1 − e−2
α
z1−
z  1 − e+2
α
z1−
z 
ζ ζ

≈
2
α
z −2
α
z
 
ζ
≈ 1− 
α
z 1 (6.92)

z

Then the same approximations used in deriving Equation (6.90) yield


−2α1
z
β
B14 = e S1
z
2π
 
α
z 1 (6.93)
−2α0
z
β
B41 = e S0
z
2π

Combining these results,

 
P0 z
= e−α0 +α1 
z
P1 z
       
+
α
z
β −
α
z
β
e 1 − S0
2π

z e S1
2π

z 
 
·      

β
β 
e+
α
z S0
z e−
α
z 1 − S1
z
2π 2π
 
P0 z −
z
· 
α
z 1 (6.94)
P1 z −
z
118 GUIDES WITH GENERAL COUPLING SPECTRA

From the perturbation results of Appendix I, we obtain the coupled

power equations for small differential loss:
     
 
β
β
P0 z  −2α0 − S0 2π S1
2π 
 
=     
P1 z 
β
β 
S0 −2α1 − S1
2π 2π
 
P0 z
· 
α
z 1 (6.95)
P1 z

Since we have assumed λc
z in Equation (6.4), the exponential

factors in Equations (6.87) and (6.88) will be insignificant, and S0  
and S1   in Equation (6.95) differ only trivially from S  of Equation
(6.2).
For significant differential loss, the extended procedure of Ap-
pendix I must be applied to Equations (6.89) and (6.91) to obtain
coupled power equations.

6.4.2. N Modes
The lossy two-mode results of Section 6.4.1 are readily extended to
N modes in the nondegenerate case. The lossless result of Equation
(6.66) for the average transfer functions becomes

I i z∞ = Ii z∞ = Ii 0e−i z

     
 2
1 N−1
βik
βik
· exp − z C S 
+ j Sik 
2 k=0 ik ik 2π 2π
0 ≤ i ≤ N − 1 (6.96)

where

∞
Sik ν = Rc ζe
αik ζ ej2πνζ dζ (6.97)
−∞

The lossless coupled power equations of Section 6.3.2 may be sim-

ilarly extended to the lossy case. The quantity of Equation (6.76)
 6.4. GENERAL COUPLING SPECTRA—LOSSY CASE 119

becomes

2
M ij ij ∞ = Ci−1j−1 · e2
αi−1j−1 z
z

z 1 − e−2
αi−1j−1
z1− 2π 
ζ

−
αi−1j−1 ζ j
βi−1j−1 ζ
· Rc ζe e dζ
−
z 2
αi−1j−1
z

i = j (6.98)

For small differential loss in each section, the approximation of

Equation (6.92) yields

 
2 2
αi−1j−1 z

βi−1j−1
M ij ij ∞ = Ci−1j−1 ·e Sj−1i−1
z
2π

αi−1j−1
z 1 (6.99)

However, the assumption of Equation (6.4) renders the exponential

factor in Equation (6.97) unimportant, and Sj−1i−1   in Equation
(6.99) is well approximated by S  of Equation (6.2) in the low-loss
case.
The quantity of Equation (6.80) becomes


N
2
M ii ii ∞ = 1 −
z Ci−1k−1
k=1

 

z

αi−1k−1 ζ ζ
· Rc ζe 1− ej
βi−1k−1 ζ dζ (6.100)
−
z
z

Neglecting the quantity 1 −  ζ /
z by the assumption of Equation

(6.4), the usual approximations yield

 

N
2

βi−1k−1
M ii ii ∞ = 1 −
z Ci−1k−1 Si−1k−1  (6.101)
k=1
2π

These results do not require small loss; for the low-loss case
Si−1k−1   is again approximated by S  of Equation (6.2).
120 GUIDES WITH GENERAL COUPLING SPECTRA

Substituting Equations (6.98) and (6.101) into Equation (6.81),

the lossless results of Equations (6.82) and (6.83) are modified as
follows for the lossy case:
   
N
2

βi−1k−1
Ri−1 z∞ = 1 − Ci−1k−1 Si−1k−1
z Ri−1 z −
z∞
k=1
2π
 N
 2
+ Ci−1j−1 e2
αi−1j−1 z
z
j=1


z
· Rc ζe−
αi−1j−1 ζ ej
βi−1j−1 ζ
−
z

1 − e−2
αi−1j−1
z1− 2π 
ζ

· dζRj−1 z −
z∞  (6.102)

2
αi−1j−1
z

   
 2
N−1

βik
P i z∞ = 1 − Cik Sik
z e−2αi
z P i z −
z∞
k=0
2π
 N−1

z
 2
+ Cik
z Rc ζe−
αik ζ ej
βik ζ
k=0 −
z


1 − e−2
αik
z1− 2π 
ζ

−2αk
z
· dζ e P k z −
z∞  (6.103)
2
αik
z

For small differential loss in each section, Equation (6.99) replaces

Equation (6.98), and Equation (6.103) becomes
   
 2
N−1

βik
P i z∞ = 1 − Cik S
z e−2αi
z P i z −
z∞
k=0
2π
 
 2
N−1

βik −2αk
z
+ Cik
z · S e P k z −
z∞ 
k=0
2π

αik
z 1 0 ≤ i k ≤ N − 1 (6.104)

We have used the approximation that Sik   of Equation (6.97) is ap-

proximately equal to S  of Equation (6.2) in the low-loss case. The
 6.5. DISCUSSION 121

perturbation results of Appendix I yield the coupled power equa-

tions for small loss per section [2]:
  
 2
N−1

βik
P i z∞ = − 2αi + Cik S P i z∞
k=0
2π
 
 2
N−1

βik
+ Cik · S P k z∞ 
k=0
2π

αik
z 1 0 ≤ i k ≤ N − 1 (6.105)

The extended procedure of Appendix I must be applied to Equa-

tion (6.102) or (6.103) for significant differential loss.

6.5. DISCUSSION

Exact solutions for the transmission statistics of multi-mode guides

have been presented for white coupling in Chapters 3 and 4. These
results depend on the single parameter S0 , the coupling spectral
density.
This chapter extends these results to general coupling spectra
Sν, for average transfer functions and coupled power equations of
two-mode and nondegenerate N-mode guides. This generalization
is accomplished by combining the earlier Kronecker product meth-
ods with perturbation theory. Other transmission statistics treated
in earlier chapters may be similarly extended. These results are no
longer exact, but require the coupling to be small.2 However, the
small-coupling restrictions of Equations (6.33) and (6.41) are only
heuristic.
For the lossless case the present results replace S0 of the white-
coupling results by the general coupling spectral density S
βij /2π,
calculated at appropriate values of its argument. However, a simi-
lar result is true for the coupled power equations in the lossy case
only for low differential loss; otherwise the extended procedure of
Appendix I is required.
2
An exception occurs for the almost-white case of Section 6.2, which permits the
treatment of large coupling by the extended procedure of Appendix I. The limiting
case, completely degenerate guides, is treated in Appendix G.
122 GUIDES WITH GENERAL COUPLING SPECTRA

The necessity of this more general procedure is illustrated by con-

sidering the term S1 
β/2π in the matrix of Equation (6.95) for
the covariance

Rc ζ = 1 − ζ  ζ ≤ 1 (6.106)

with differential attenuation

α = 1 (6.107)

Substituting these relations into Equation (6.88),

S1 884 = −0163 (6.108)

Negative values clearly violate the physical requirement that powers

must be positive for any input conditions [3].

REFERENCES

1. Harrison E. Rowe, “Waves with Random Coupling and Random Propa-

gation Constants,” Applied Scientific Research, Vol. 41, 1984, pp. 237–255.
2. Dietrich Marcuse, Theory of Dielectric Optical Waveguides, 2nd ed., Aca-
demic Press, New York, 1991.
3. Neil A. Jackman, Limitations and Tolerances in Optical Devices, Appendix
C, Ph.D. Thesis, Stevens Institute of Technology, Castle Point, Hoboken,
NJ, 1994.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

CHAPTER SEVEN

Four-Mode Guide with

Exponential Coupling
Covariance

7.1. INTRODUCTION

We illustrate the results of Chapter 6 for general coupling spectra

with a single example:

1. Exponential coupling covariance.

2. Four modes.
3. Nondegenerate.
4. Lossless.

Other examples appear in Chapter 8.

The covariance and spectrum of the coupling coefficient, Equa-
tions (6.1) and (6.2), are given as follows:

Rc
ζ = Rc
0e−ζ/ζc  (7.1)

2/ζc
S
ν = Rc
0  (7.2)

1/ζc  +
2πν2
2

The Hilbert transform of the coupling spectrum, Equation (6.47), is

4πν
S
ν = Rc
0  (7.3)

1/ζc  +
2πν2
2

123
124 FOUR-MODE GUIDE WITH EXPONENTIAL COUPLING COVARIANCE

The following numerical parameters apply to the four-mode di-

electric slab waveguide of Appendix F.2. From Equations (F.20)
and (F.21),
 
63424 0 0 0
 
 0
6
63316 0 0  −1

= j = j × 10   m  (7.4)

 0 0 63142 0 
0 0 0 62921
 
0 23225 0 44966
 
 23225
9
0 68737 0  −2
C = 10  m  (7.5)

 0 68737 0 133083 
44966 0 133083 0
We obtain the following matrix of differential propagation constants
from Equation (7.4):
 
0 β01 β02 β03
 
 β10 0 β12 β13 
 = 



 β20 β21 0 β23 
β30 β31 β32 0
 
0 00108 00282 00503
 
 −00108
6
0 00174 00395  −1
= 10  m  (7.6)

 −00282 −00174 0 00221 
−00503 −00395 −00221 0
We next select numerical values for the parameters Rc
0 and ζc
of Equations (7.1) and (7.2). We take
ζc = 8 × 10−5 m. (7.7)
The cutoff frequency and wavelength of S
ν of Equation (7.2) are
then
1
νc = = 19894 × 103 m−1
2πζc (7.8)
−4
λc = 2πζc = 50265 × 10 m.
 7.1. INTRODUCTION 125

This choice places the coupling cutoff wavelength in the middle

of the beat wavelengths of the four modes, which range from
2π/ β03 = 1250 × 10−4 m. to 2π/ β01 = 5818 × 10−4 m.
z must satisfy two constraints, Equations (6.53) and (6.54). We
take

z = 0005 m. (7.9)

The small-coupling condition of Equation (6.41) becomes


Rc
0 zC 1 (7.10)

From Equations (7.5) and (A.25),

C = 20182 × 1010 m−2  (7.11)

Combining these parameters,


Rc
0 991 × 10−9 m. (7.12)

We take

Rc
0 = 5 × 10−10 m., Rc
0 = 25 × 10−19 m2  (7.13)

Substituting these numerical values, Equations (7.1) and (7.2)

yield the following for the present example:
+4
Rc
ζ = 25 × 10−19 e−125×10 ζ
 (7.14)

625 × 10−15
S
ν =  (7.15)
15625 × 10+8 +
2πν2

πν × 10−18
S
ν =  (7.16)
15625 × 10+8 +
2πν2

S
0 = 4 × 10−23 . The present low-pass coupling spectrum has much

smaller bandwidth, low-frequency spectral density, and rms coupling
126 FOUR-MODE GUIDE WITH EXPONENTIAL COUPLING COVARIANCE

and straightness deviation than the almost-white example of Equa-

tions (4.67) and (4.68).

7.2. AVERAGE TRANSFER FUNCTIONS

Assume deterministic inputs Ii
0, 0 ≤ i ≤ 3 Take the magnitude of

Equation (6.66), to yield

1 3 2 βik
I i
z∞ = Ii
0 exp − z C S 0 ≤ i ≤ 3 (7.17)
2 k=0 ik 2π

The expected response magnitudes are found by substituting the nu-

merical parameters of Section 7.1 into this relation:

I 0
z∞ = I0
0 exp
−85295 × 10−5 z

I 1
z∞ = I1
0 exp
−38328 × 10−4 z
(7.18)
I 2
z∞ = I2
0 exp
−11792 × 10−3 z
I 3
z∞ = I3
0 exp
−88120 × 10−4 z

Finally, the angle of Equation (6.66) for real (zero phase) inputs is

1 3 2
βik
I i
z∞ = −z βi + C S  (7.19)
2 k=0 ik 2π

For the numerical parameters of Section 7.1, the expected response

angles are

I 0
z∞ = −z
β0 + 1480 × 10−4 

I 1
z∞ = −z
β1 + 3943 × 10−4 
(7.20)
I 2
z∞ = −z
β2 + 1070 × 10−3 
I 3
z∞ = −z
β3 + 1612 × 10−3 

The propagation constants βi , given in Equation (7.4), are large com-

pared to the contributions of the random coupling.
 7.4. DISCUSSION 127

7.3. COUPLED POWER EQUATIONS

We obtain the coupled power equations by substituting Equations

(7.5), (7.6), and (7.15) into Equation (6.84), with N = 4:
   
P 0
z∞ −17059 12358 0 04701
   
 P 1
z∞   12358 −76655 64297 0 
  = 10 
−4 
   
 P 2
z∞   0 64297 −235839 171542 
P 3
z∞ 04701 0 171542 −176243
 
P 0
z∞
 
 P 1
z∞ 
·


 (7.21)
 P 2
z∞ 
P 3
z∞

7.4. DISCUSSION

The results of the present chapter illustrate the coupled power equa-
tions of Chapter 6 for the particular dielectric slab waveguide of
Appendix F with straightness deviation spectrum of Equation (7.15).
Other results of prior chapters for white imperfections may be simi-
larly extended to general spectra.
Such results are only approximate; they require the imperfections
to be small. For the present problem this requirement is given in
Equation (7.12). However, such limits are only heuristic; no rigorous
bounds on errors are available.
Exact transmission statistics for non-white coupling or propaga-
tion parameters are available only for random square-wave imper-
fections. Chapter 8 derives results for random square-wave coupling.
Comparison with these exact results permits validation of the ap-
proximate transmission statistics of Chapter 6.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

CHAPTER EIGHT

Random Square-Wave
Coupling

8.1. INTRODUCTION

We derive exact transmission statistics for coupled modes with ran-

dom square-wave coupling coefficients having a variety of statistics.
Such transmission lines consist of cascaded sections, each having
constant coupling, with transfer functions described by modification
of Equations (2.13)–(2.18) and (2.59), for the two-mode and multi-
mode cases, respectively.
Chapters 3 and 4 have presented exact transmission statistics for
white coupling. However, general coupling spectra, treated in Chap-
ter 6, permit only approximate treatment, with heuristic estimates
for the range of validity. The present results yield exact transmission
statistics for a variety of coupling spectra corresponding to random
square-wave coupling models, and thus permit validation of these
approximate results in the present special cases.
The present methods have additional interest, in that they extend
the prior analysis, based on products of statistically independent ma-
trices, to products of Markov matrices.
A square-wave coupling function is illustrated in Figure 8.1. The
coupling c
z has constant value ck in the kth section:
c
z = ck 
k − 1 < z < k (8.1)
We give exact coupled power equations for a number of statisti-
cal models for ck , presented in Appendix J. The spectra for these
129
130 RANDOM SQUARE-WAVE COUPLING

c(z)

c1

c2
c4

z

2
3
4

c3

FIGURE 8.1. Square-wave coupling.

c
z, also presented in this appendix, permit comparison of these ex-

act results with the approximations of Chapter 6. Numerical results,
obtained using MAPLE, are presented for the signal power P 0
z,
for single-mode signal input with unit power:
P 0
0 = I0
02 = 1 P i
0 = 0 1 ≤ i
I0
0 = 1 coherent input; I0
0 = 0 incoherent input.
(8.2)
The coherent components (i.e., expected values) of all transfer func-
tions are identically zero for incoherent input.
The transmission matrix characterizing the kth section of a lossless
two-mode guide with square-wave coupling of Figure 8.1 is obtained
from Equations (2.13)–(2.18) as follows:



I0
k I0
k − 1
= T
ck  ·  (8.3)
I1
k I1
k − 1
β0 +β1
T
ck  = e−j 2  /
K+ − K− 

√ √ √ √ 
−K− ej
β/2 + K+ e−j
β/2 ej
β/2 − e−j
β/2
· √ √ √ √ 
ej
β/2 − e−j
β/2 K+ ej
β/2 − K− e−j
β/2
(8.4)
 8.1. INTRODUCTION 131
√
1±
K± = K+ K− = −1 (8.5)
2ck /β

1 c /β
= k√  (8.6)
K+ − K−

√
= 1 +
2ck /β2  (8.7)

β = β0 − β1  (8.8)

The corresponding result for the lossless multi-mode case follows

from Equations (2.43), (2.45), (2.58), and (2.59):

I
k = T
ck  · I
k − 1 (8.9)

T
ck  = ej−
+ck Cz  (8.10)

The four-mode example of Appendix F illustrates a more realistic

physical system;
and C are given by Equations (F.20) and (F.21):
 
63424 0 0 0
 
6
0 63316 0 0  −1
 = j
= j × 10   m  (8.11)

 0 0 63142 0 
0 0 0 62921

 
0 23225 0 44966
 
 23225
9
0 68737 0  −2
C = 10  m  (8.12)

 0 68737 0 133083 
44966 0 133083 0

For 2 modes, Equations (8.9) and (8.10) specialize as follows:


β0 0

=  (8.13)
0 β1
132 RANDOM SQUARE-WAVE COUPLING

and from Equation (2.60)


0 1
C=  (8.14)
1 0
These results are equivalent to Equations (8.3)–(8.8).

8.2. TWO MODES—BINARY INDEPENDENT SECTIONS

Equations (J.21) and (J.22) give the coupling statistics for indepen-
dent, symmetric, binary coupling. The coupling spectrum is given by
Equation (J.24), and plotted in Figure 8.2.
The small-coupling condition of Equations (6.32) and (6.33)
becomes
a  1 (8.15)
Equation (6.52) yields the following approximation for the signal
power with input given by Equation (8.2):
1 1 −2S
β z
P 0app
z = + e 2π  (8.16)
2 2
For n sections, total guide length z = n, Equations (8.3), (D.14),
and (D.17) yield the following exact solutions for the inputs of Equa-
tion (8.2):




I 0
n I0
n 1
= =T
ck  n ·  coherent input.
I 1
n I1
n 0
(8.17)




I 0
n I0
n 0
= =  incoherent input. (8.18)
I 1
n I1
n 0
   
P 0
n 1
   
 P 01
n  0
 = T
ck  ⊗ T∗
ck  n · 
   
 P 10
n  0
P 1
n 0
coherent or incoherent input, (8.19)
 8.2. TWO MODES—BINARY INDEPENDENT SECTIONS 133

1.0

0.8

0.6

1
S( )
a2

0.4

0.2

0
0 0.5 1 1.5 2


FIGURE 8.2. Coupling spectrum: binary independent square-wave cou-
pling.

where the powers and cross-powers are defined in Equation (3.50).

In these equations,
1
T
ck  = T
a + T
−a (8.20)
2
1
T
ck  ⊗ T∗
ck  = T
a ⊗ T∗
a + T
−a ⊗ T∗
−a (8.21)
2
where T
 is given by Equations (8.4)–(8.8).
134 RANDOM SQUARE-WAVE COUPLING

P 0app of Equation (8.16), and the exact P 0 determined from Equa-

tions (8.19) and (8.21) are plotted in Figure 8.3 for a = 05 and
β = 02π; the exact plot consists of discrete points since Equa-
tion (8.19) contains an integral number of sections. The value of a
for this example is less than 1, but is not much less than 1, as re-
quired by the heuristic condition of Equation (8.15). Therefore the
approximate and exact plots in Figure 8.3 are close, but not identical.
As a decreases these two results approach each other rapidly.

1.0

0.8
approximate

0.6 exact

P0

0.4

0.2

0
0 1 2 3 4
z/

FIGURE 8.3. Signal power vs. normalized distance; two modes; binary in-
dependent square-wave coupling. a = 05, β = 02π.
 8.3. TWO MODES—BINARY MARKOV SECTIONS 135

8.3. TWO MODES—BINARY MARKOV SECTIONS

A great variety of coupling spectra are available in the Markov case;

the coupling statistics are described in Section J.2.2. The parameter
p, the probability that two adjacent ck in Equation (8.1) and Figure
8.1 are different, determines the shape of the coupling spectrum.
The transition matrix, unconditional probability vector, and coupling
vector for the symmetric binary case are given in Equation (D.36).
Equation (J.27) for the coupling spectrum S
ν yields the result of
Figure 8.2 for p = 21 , the independent case. For small p, the coupling
approaches a constant and the spectrum approaches a δ function; as
p → 1 the coupling approaches a periodic square wave, and S
ν
approaches a line spectrum with fundamental frequency ν = 05/.
Figures 8.4 and 8.5 show typical S
ν for small and large p.
We choose the band-pass S
ν of Figure 8.5 with p = 08 for the
following example, with β = π corresponding to the point

ν = 05 S
ν = 1621a2  (8.22)

close to the peak of S
ν. The small-coupling requirement remains

as in Equation (8.15). The approximate signal power obtained by
substituting the above numerical value in Equation (8.16) is

1 1 −3242·
a2 z
P 0app
z = + e  (8.23)
2 2
The exact coupled power equations for unit signal input, Equation
(8.2), are given from Equations (D.58) and (D.59):
 
P 0
n
 

n−1
 P 01
n  1 02A1 ⊗ A1∗ 08A1 ⊗ A1∗
 = I I ·
 
 P 10
n  2 08A2 ⊗ A2∗ 02A2 ⊗ A2∗
P 1
n
 
1

  
A1 ⊗ A1∗ 0
· ∗
·
 
 (8.24)
A2 ⊗ A2 0
 
0
136 RANDOM SQUARE-WAVE COUPLING

2
1
S( )
a2

0
0 0.5 1 1.5 2


FIGURE 8.4. Coupling spectrum: binary Markov square-wave coupling.
p = 02

where n = z/, I is the 4 × 4 unit matrix, and

A1 = T
a A2 = T
−a (8.25)

where T
 is given by Equations (8.4)–(8.8).

The approximate and exact signal power, P 0app of Equation (8.23)
and P 0
n of Equation (8.24), are plotted in Figure 8.6 for a = 02
 8.4. FOUR MODES—MULTI-LEVEL MARKOV SECTIONS 137

1.6

1.4

1.2

1.0
1
S( )
a2

0.8

0.6

0.4

0.2

0
0 0.5 1 1.5 2


FIGURE 8.5. Coupling spectrum: binary Markov square-wave coupling.
p = 08

and β = π (as noted above, corresponding approximately to the

peak of the band-pass coupling spectrum of Figure 8.5).
As in the preceding section, we have chosen a value of a some-
what larger than indicated by Equation (8.15), in order to display a
significant difference between the approximate and exact solutions
in Figure 8.6.
138 RANDOM SQUARE-WAVE COUPLING

exact

0.8

approximate

0.6

P0

0.4

0.2

0
0 2 4 6 8 10 12 14

z/

FIGURE 8.6. Signal power vs. normalized distance; two modes; binary
Markov square-wave coupling. a = 02, β = π.

8.4. FOUR MODES—MULTI-LEVEL MARKOV SECTIONS

We take as a final example the two multi-level coupling coefficient

models described by Equations (J.9)–(J.17) and Section J.3. The pa-
rameters ck of Equation (8.1) and Figure 8.1 for these two models
may be described as follows:

ck = d dk−2 + dk−1 + dk + dk+1 + dk+2  Low-Pass. (8.26)

 8.4. FOUR MODES—MULTI-LEVEL MARKOV SECTIONS 139

ck = d dk−2 + dk−1 − dk+1 − dk+2  Band-Pass. (8.27)

The separation between coupling levels is 2d. The parameters dk 

are independent, symmetric, binary random variables taking on the
values ±1. The corresponding coupling spectra are given by Equa-
tions (J.29) and (J.46). Both generate pseudo-Gaussian square-wave
random coupling functions, with six and five levels, respectively. The
coupling spectra for these two cases are shown in Figures 8.7 and
8.8. We choose the low-pass case for the following numerical exam-
ple; the corresponding probability distribution is given in Equation
(J.30).
The larger number of parameters requires treatment of a specific
numerical example, in contrast to the two-mode case. We take the
four-mode guide of Appendix F, with parameters given in Equations
(8.9)–(8.12). The mean-square coupling coefficient is found from ei-
ther Equation (J.29) or (J.30) as
∞
ck2 = S
νdν = 5d 2  (8.28)
−∞

The matrix norm is found from Equations (A.25) and (8.12) as

C = 201820 × 109 m−2  (8.29)

We choose

d = 10−6 m, (8.30)

 = 2 × 10−5 m. (8.31)

Then the spectral width of S
ν in Figure 8.7 is 104 m−1 , roughly com-
parable to that of the low-pass example in Chapter 4, in Equation
(4.66). However, the low-pass spectral density, S
0 = 5 × 10−16 m3 ,
is much greater than that of Equation (4.67). The root mean square
(rms) straightness deviation is
 ∞
c 2
z = S
νdν = 2236 × 10−6 m, (8.32)
−∞

also much larger than that of Equation (4.68), but still smaller than
the guide width of 123×10−5 m from Equation (F.17). The left-hand
140 RANDOM SQUARE-WAVE COUPLING

25

20

15

1
S( )
d 2

10

0
0 0.2 0.4 0.6 0.8 1

FIGURE 8.7. Coupling spectrum: six-level Markov square-wave coupling.

side of the small coupling condition of Equation (6.41) becomes

∞
S
νdνC = 0903 (8.33)
−∞

This is not small compared to 1; we have made the coupling large

for the present example.
The coherent signal component for the coherent input of Equa-
tion (8.2) is given by Equations (6.66), (J.29), (8.11), and (8.12) for
 8.4. FOUR MODES—MULTI-LEVEL MARKOV SECTIONS 141

10

6
1
S( )
d 2

0
0 0.2 0.4 0.6 0.8 1


FIGURE 8.8. Coupling spectrum: five-level Markov square-wave coupling.

the small coupling approximation. Substituting the above numerical

parameters, we have
     
1 2 β01 2 β02
I 0
z app = exp − z C01 S + C02 S
2 2π 2π (8.34)
−1496z
=e 

The corresponding exact quantity I 0
z  is found from Equations

(D.50) and (D.39)–(D.43). The Ai of Equation (D.41) are given by
142 RANDOM SQUARE-WAVE COUPLING

Equation (8.10) as
Ai = T
ai  = ej−
+dai C  (8.35)
where the propagation and coupling matrices
and C are given by
Equations (8.11) and (8.12). These approximate and exact results
are plotted in Figure 8.9.
The small-coupling condition of Equation (6.41) is significantly
violated by Equation (8.33); the difference between the two results
in Figure 8.9 is not surprising.

0.8

approximate

0.6

I0 (z)

0.4
exact

0.2

0
0 2 ×10–4 4×10–4 6×10–4

z - meters
FIGURE 8.9. Coherent signal component vs. distance; four modes; six-level
Markov square-wave coupling.
 8.5. DISCUSSION 143

8.5. DISCUSSION

Exact results for the coupled line equations with random coupling or
propagation parameters are available for the limiting case of white
imperfection spectra. General coupling spectra, treated in Chapter
6, require the use of both Kronecker product methods, to find the
overall transmission statistics of cascaded sections, and perturbation
theory, to determine the performance of a typical section. These
results are therefore only approximate; heuristic conditions for small
coupling are given in Equations (6.33) and (6.41) for two modes and
N modes, respectively.
The present chapter has extended Kronecker product methods to
Markov matrices, permitting the treatment of correlated sections.
The choice of square-wave coupling permits exact treatment of the
individual sections, and therefore yields exact transmission statistics
for square-wave coupling with a variety of spectra. These results of-
fer partial validation of the intuitive small-coupling conditions of
Section 6.3.
The present results suggest the following physical interpretation
of the small-coupling requirement: perturbation theory must apply
in a correlation length of the coupling coefficient c
z.
The final example, of Section 8.4, represents an approach to a
more realistic model of a random coupling function. The example
chosen requires numerical computing with quite large matrices—
128 × 128 in these calculations for the expected transfer functions.
These results are within the capabilities of MAPLE on a Pentium
machine running at 233 MHz. Significant reduction in computing
time is achieved by taking advantage of the special properties of the
matrices in this particular example.
Higher order statistics require correspondingly increased comput-
ing resources. Determining the mode powers would require matrix
products 512 × 512 in size.
Finally, our choice of square-wave coupling for the individual sec-
tions permits exact results. However, the Markov–Kronecker product
methods may be applied to other coupling functions for the individ-
ual sections, greatly enlarging the scope of these methods. The in-
dividual sections would no longer permit exact treatment, as in the
present case, but would have to be solved by perturbation theory
or other numerical methods.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

CHAPTER NINE

Multi-Layer Coatings with

Random Optical Thickness1

9.1. INTRODUCTION

Figure 9.1 shows a multi-layer coating, designed to serve as an op-

tical filter, mirror, or window [3–6]. It consists
of
dielectric layers
with thickness dk and index of refraction nk = k /0 , 1 ≤ k ≤
. The
medium to the left of the coating has index ni , the output medium
has index nt ; as a special case either or both may be free space, with
n = 1.
An input plane wave with electric field Ei is incident upon the
coating, producing a reflected wave with field Er and an output wave
with field Et . The complex transfer functions for reflection and trans-
mission are
Er Et
R= T =  (9.1)
Ei Ei
The reflectance and transmittance are the corresponding ratios of
intensities:
n
R = R2 T = T 2 · t  (9.2)
ni
The design values for dk  and nk  are chosen to realize the
desired transmission and reflection performance. We study the re-
1
This chapter is based on thesis work of Neil Jackman [1, 2].
145
146 MULTI-LAYER COATINGS WITH RANDOM OPTICAL THICKNESS

Ei
d1 d2 dk d

Et

ni n1 n2 ... nk ... n
nt
Er

0 z1 z2 zk – 1 zk z
– 1 z

FIGURE 9.1. Multi-layer dielectric structure.

sulting transfer function statistics due to random variations in these

parameters. The following assumptions are made:

1. Infinite parallel plane geometry.

2. Dielectric layers of uniform thickness and dielectric constant.
3. Input wave linearly polarized, at normal incidence.
4. Ideal design given.
5. Lossless dielectrics.

Since the layers are lossless,

R + T = 1 (9.3)

The restriction to normal incidence is convenient for illustrating the

present methods, but not essential.
Forward and backward plane waves are present at the input, and
in each layer; the output is assumed matched, so only a forward
wave is present there. These two waves are coupled by the dielectric
layers; both can have significant magnitude. This problem therefore
corresponds to the analysis of previous chapters with two coupled
modes, but with a significant difference. Here the two modes travel
in opposite directions; in former chapters all modes traveled in the
same direction. The response of this system is given by Equation
(K.18) as
 
  1    
1
Ei 1
 ni 


1 1 Et
=  · Mk · · (9.4)
Er 2 1  k=1 nt −nt 0
1 −
ni
 9.2. MATRIX ANALYSIS 147

where Mk is given by Equations (K.15) and (K.16):

 
j sin φk
 cos φk nk  2π
Mk =   φk = n d  (9.5)
λ k k
jnk sin φk cos φk

λ is the free-space wavelength; the n’s and d’s are shown in Fig-
ure 9.1.
The circuit-theoretic model of a dominant-mode transmission line
with random parameters is closely related to the present problem
[7]. In both, input and output are related by matrix products, per-
mitting Kronecker product methods to be applied. However, unlike
the forward-mode case of preceding chapters, the outputs are known
while the inputs are not. Therefore direct analysis yields the statistics
for only the reciprocal transfer function 1/T . We desire statistics for
T and for R, corresponding to mixed boundary conditions at input
and output, i.e., a unit forward wave at the input and zero backward
wave at the output.
A heuristic solution for the average powers in two randomly cou-
pled modes traveling in opposite directions has been based on physi-
cal arguments [8, 9]. A general approach, based on the present meth-
ods, provides detailed results for various transfer function statistics
of multi-layer coatings. This is possible because useful devices of
the class studied here must exhibit behavior close to that of their
theoretical design. We combine perturbation theory with Kronecker
product methods to obtain accurate, general results for the trans-
mission and reflection statistics of multi-layer optical coatings whose
layers exhibit random optical thickness.

9.2. MATRIX ANALYSIS

For convenience set the output electric field equal to unity in Fig-
ure 9.1 and Equations (9.1) and (9.4):

Et = 1 (9.6)

Then Equation (9.1) becomes

Er 1
R= T =  (9.7)
Ei Ei
148 MULTI-LAYER COATINGS WITH RANDOM OPTICAL THICKNESS

Equations (9.2) and (9.3) are unchanged, and Equation (9.4)

becomes
 
  1    
1
Ei 1 ni 


1 1 1
=  · Mk · · (9.8)
Er 2 1  k=1 nt −nt 0
1 −
ni

with Mk given by Equation (9.5).

We determine the statistics of the transfer functions due to ran-
dom deviations in layer thickness. Let

dk = dk0 + δk (9.9)

where dk0 is the design thickness of the kth layer and δk is its ran-
dom thickness error. For the ideal design, substitute dk = dk0 into
Equation (9.5); denote the corresponding results of Equations (9.7)
and (9.2) as R0 T0 R0 and T 0 for the ideal device.
Next, let the different layers have random thickness variations
δk  of Equation (9.9). Equations (9.8) and (9.5) become
 
  1    
1
Ei + ei 1
 ni 


1 1 1
=  · Mk · · (9.10)
Er + er 2 1  k=1 nt −nt 0
1 −
ni

where
 
j sin φk
 cos φk nk  2π
Mk =  φk = n d + δk  (9.11)
λ k k0
jnk sin φk cos φk

The δk  are assumed to be independent random variables, with

specified probability densities. ei and er are the departures from de-
sign values Ei and Er , due to the random variations in layer thick-
ness; ei and er equal zero if all δk are zero. Then the various transfer
parameters are

Er + er 1
R= T =  (9.12)
Ei + ei Ei + ei
 9.2. MATRIX ANALYSIS 149
2 2
Er + er 1 nt
R= T =  (9.13)
Ei + ei Ei + ei ni

A useful device must exhibit performance close to the ideal de-

sign; the δk  must be small enough that this is so. Consequently,
Taylor series expansion of the denominators of Equations (9.12) and
(9.13) is appropriate. We use these series to second order. As an
example, consider the transmission transfer function T of Equa-
tion (9.12):

2 
1 e ei
T = 1− i +  (9.14)
Ei Ei Ei

We subsequently require expected values of T and the other quanti-

ties in Equations (9.12) and (9.13). The moments of ei in Equation
(9.14) can not be obtained directly. Therefore, we substitute

ei = Ei + ei − Ei (9.15)

into Equation (9.14), to obtain

2

1 3 E +e
T = 3− Ei + ei + i 2 i  (9.16)
Ei Ei Ei

The moments of Ei +ei are readily found by the Kronecker product

methods of the following section.
The following results are similarly obtained for the other quanti-
ties of Equations (9.12) and (9.13):

2

E + er 3 E +e
R= r 3− E + ei + i 2 i  (9.17)
Ei Ei i Ei


6 8
T = − Re Ei Ei∗ + e∗i
Ei 2 Ei 4

2 Ei + ei 2 nt
+ 6
Re Ei2 Ei∗ + e∗i 2
+  (9.18)
Ei  Ei 4 ni
150 MULTI-LAYER COATINGS WITH RANDOM OPTICAL THICKNESS

6 8
R = Er + er 2 · 2
− Re Ei Ei∗ + e∗i
Ei  Ei 4

2 Ei + ei 2
+ Re Ei2 Ei∗ + e∗i 2
+  (9.19)
Ei 6 Ei 4

Finally, the two-frequency transfer function statistics yield the im-

pulse response and signal distortion due to the random imperfections
of the device, as in Sections 3.6, 3.7, 4.4, and Appendix E. Let T1 and
T2 denote the complex transmission transfer function at free-space
wavelengths λ1 and λ2 ; denote the input plane wave electric field Ei
of Figure 9.1 at these two frequencies by Ei1 and Ei2  Then

nt 6 Ei2∗ + e∗i2 Ei1 + ei1 Ei2∗ + e∗i2 2

T1 T2∗ = − 4 − 4 +
ni Ei1 Ei2∗ Ei1 Ei2∗2 Ei12 Ei2∗ Ei1 Ei2∗3

Ei1 + ei1 2 Ei1 + ei1 Ei2∗ + e∗i2 nt
+ +  (9.20)
Ei13 Ei2∗ Ei12 Ei2∗2 ni

For the complex reflection transfer function,

R1 · R∗2 = Er1 + er1 Er2

∗
+ e∗r2
6 Ei2∗ + e∗i2 Ei1 + ei1 Ei2∗ + e∗i2 2
· − 4 − 4 +
Ei1 Ei2∗ Ei1 Ei2∗2 Ei12 Ei2∗ Ei1 Ei2∗3

Ei1 + ei1 2 Ei1 + ei1 Ei2∗ + e∗i2
+ +  (9.21)
Ei13 Ei2∗ Ei12 Ei2∗2

Ei1 and Ei2 denote Ei at the two frequencies, as in Equation (9.20),

while Er1 and Er2 denote the corresponding reflected plane wave
electric field Er of Figure 9.1.
The expected values of Equations (9.16)–(9.21) require the mo-
ments and cross-moments of the quantities Ei + ei and Er + er
on the left-hand side of Equation (9.10). These are found by the
Kronecker product methods of Section 9.3.

9.3. KRONECKER PRODUCTS

The following results, obtained from Equation (9.10), Section D.1

and Appendix C, are required to calculate the expected transmit-
 9.3. KRONECKER PRODUCTS 151

tance T  from Equation (9.18), and the two-frequency transfer

function statistics T1 T2∗  from Equation (9.20):
 
  1    
1
Ei + ei  1
 ni 

1 1 1
=   · Mk  · ·  (9.22)
Er + er  2 1  k=1 nt −nt 0
1 −
ni
 
 Ei + ei 2 
 
  Ei + ei Er + er  
 
 
  Er + er Ei + ei  
 Er + er 2 
   
1 1
 1  1
1 ni   ni 


=  ⊗ · Mk ⊗ Mk 
4 1   1  k=1
1 − 1 −
ni ni
 
1
     
1 1 1 1 0
· ⊗ ·
 
 (9.23)
nt −nt nt −nt  0
0
 2 
Ei + ei  
 
  Ei + ei Er∗ + e∗r  
 
 ∗ ∗ 
  Er + e r E i + e i  
Er + er 2 
   
1 1
 1  1
1 ni   ni 



=  ⊗  · Mk ⊗ Mk∗ 
4 1   1  k=1
1 − 1 −
ni ni
 
1
     
1 1 1 1 0
· ⊗ ·
 
 (9.24)
nt −nt nt −nt  0
0
152 MULTI-LAYER COATINGS WITH RANDOM OPTICAL THICKNESS
 
 Ei1 + ei1 Ei2∗ + e∗i2 
 ∗ 
  Ei1 + ei1 Er2 + e∗r2  
 
 ∗ ∗ 
  Er1 + e r1 E i2 + e i2  
∗
 Er1 + er1 Er2 + e∗r2 
   
1 1
 1   1 

1 ni   ni 
=  ⊗  · Mk λ1 ⊗ Mk∗ λ2 
4 1   1  k=1
1 − 1 −
ni ni
 
1
     
1 1 1 1 0
· ⊗ · 
  (9.25)
nt −nt nt −nt 0
0
The 1 1 elements of Equations (9.22)–(9.25) appear in T  and
T1 · T2∗ . Other quantities in these Equations are not of present in-
terest. Certain additional expected values that are required in or-
der to calculate other transmission and reflection statistics may be
found from appropriate Kronecker products; we omit these expres-
sions here, since they are not required for the example of the fol-
lowing section.
The Taylor series expansions of Section 9.2 provide good approx-
imations if
ei
 1 (9.26)
Ei
Therefore we require that

ei 2   Ei  (9.27)
where from Equation (9.15)
ei 2  = Ei + ei 2  − 2 · Re Ei Ei∗ + e∗i  + Ei 2  (9.28)

9.4. EXAMPLE: 13-LAYER FILTER

We illustrate the present methods with a 13-layer band-pass filter

centered on λ = 500 nm, with input and output media free-space.
 9.4. EXAMPLE: 13-LAYER FILTER 153

The design parameters of Equations (9.9)–(9.11) and Figure 9.1,

are [10]:

ni = nt = 1 (9.29)

k 1 2 3 4 5 6 7 8 9 10 11 12 13
nk 4 1.35 4 4 1.35 4 1.35 4 1.35 4 4 1.6 4
dk0 (nm)2 31.2 92.6 31.2 31.2 92.6 31.2 92.6 31.2 92.6 31.2 31.2 78.1 31.2 
π π π π π π π π π π π π π
φk
2 2 2 2 2 2 2 2 2 2 2 2 2
(9.30)

The design transmittance T 0 of this filter is found from Equations

(9.2), (9.5), and (9.8), and plotted in Figure 9.2.

9.4.1. Statistical Model

We assume that the random thickness parameters of Equation (9.11)
are identical, uniformly distributed, independent random variables,
with probability density


 1
 δ < δmax 
p δ = 2δmax (9.31)


0 δ > δmax 

While the different layers have design thickness varying by almost 3

to 1, they have the same maximum thickness error. This model would
correspond to a manufacturing process in which the measurement
error is independent of the thickness of the layer under study. These
assumptions are made only for convenience; other models are readily
accommodated in the following analysis.
We require the matrix expected values on the right-hand sides
of Equations (9.22)–(9.25), for Mk of Equation (9.11) and p δ of
Equation (9.31). These are used in the following calculations for
the expected transmittance and for the two-frequency transmission

2
These values are approximate. The exact values, given by dk = 125/nk correspond-
ing to φk = π/2 for this design, are used throughout the following calculations.
154 MULTI-LAYER COATINGS WITH RANDOM OPTICAL THICKNESS

1.0

0.8

0.6

T0

0.4

0.2

0
480 490 500 510 520
– nm

FIGURE 9.2. Design transmittance for 13-layer band-pass filter.

statistics. From Equations (9.11), (9.30), and (9.31),

 
λ 2πnk δmax
 0 j sin 
 2πn2k δmax λ 
Mk  =  (9.32)
 λ 2πnk δmax 
j sin 0
2πδmax λ

We omit similar results for Mk ⊗ Mk , Mk ⊗ Mk∗ , and Mk λ1 ⊗

Mk∗ λ2 . All these quantities are calculated by MAPLE for the fol-
lowing results.
 9.4. EXAMPLE: 13-LAYER FILTER 155

From Equation (9.28),

 2
ei Re Ei Ei∗ + e∗i  Ei + ei 2 
=1−2 +  (9.33)
Ei Ei 2 Ei 2
The expected values in the second and third terms of Equation (9.33)
are given by the 1 1 elements of Equations (9.22) and (9.24).
We choose
δmax = 01 nm (9.34)
for the examples of the following two sections; i.e., the different
layers of Equation (9.30) have independent, uniformly distributed,
thickness

variations with maximum deviation ±01 nm. For this case
ei /Ei 2  varies from a minimum of approximately 0025 at the
band edges to a maximum of approximately 01 near the band center.
The second-order Taylor series of Equations (9.14)–(9.21) should
provide reasonable approximations in this case.

9.4.2. Transmittance
Random thickness errors cause the average transmittance T  to
depart from the design value T 0 of Figure 9.2. Figure 9.3 shows
this deviation T 0 − T , determined from the expected value of
the second-order series expansion of Equation (9.18) and the re-
sults of Section 9.3, with random thickness errors having parameters
of Equations (9.31) and (9.34).
The in-band transmittance has been reduced by a small amount
near the band center, and increased slightly at the band edges. Other
results, not given here, show that the ratio T /T 0 is very close to 1
over the entire band 475–525 nm. Therefore the tolerances have not
reduced the out-of-band discrimination significantly. Calculation of
the in-band signal distortion requires in addition the two-frequency
transmission statistics, discussed in Section 9.4.3.

9.4.3. Two-Frequency Transmission Statistics

The average transmission transfer function is obtained from Equa-
tion (9.16) as

1 3  Ei + ei 2 
T  = 3 − Ei + ei  + (9.35)
Ei Ei Ei2
156 MULTI-LAYER COATINGS WITH RANDOM OPTICAL THICKNESS

T0 – T

0.008

0.006

0.004

0.002

0
480 490 500 510 520

– nm

–0.002

FIGURE 9.3. Deviation of average transmittance from design value. δmax =

01 nm.

using the results of Section 9.3. The a.c. component of T is

T = T − T  (9.36)

The variance of the transfer function fluctuation,

T 2  = T 2  − T 2 = T  − T 2 (9.37)

is plotted in Figure 9.4 for the mid-band region of the filter.

 9.4. EXAMPLE: 13-LAYER FILTER 157

0.010

0.008

0.006

0.004

0.002

0
490 495 500 505 510

– nm
FIGURE 9.4. Transmission transfer function variance.

Finally, the covariance is given by

T1 · T2∗  = T1 T2∗  − T1 T2 ∗ (9.38)

with T1 T2∗  given by the expected value of Equation (9.20). The
two-frequency correlation coefficient is given by

T1 · T2∗ 
C12 =

 (9.39)
T1 2  · T2 2 
158 MULTI-LAYER COATINGS WITH RANDOM OPTICAL THICKNESS

Consider the special case

λ1 = 500 +  λ2 = 500 −  0 <  < 10 nm, (9.40)

i.e., the correlation coefficient centered on 500 nm. C12 is so close

to 1 over the mid-band region that a plot is uninteresting; it varies
from its maximum value of 1 to a minimum of 0986. This suggests
that the transfer function fluctuations for different λ are strongly
correlated.

9.5. DISCUSSION

Direct application of Kronecker product methods to multi-layer

coatings would yield only statistics of the reciprocal transmission
transfer function. Combining these methods with perturbation the-
ory yields transmission and reflection transfer-function statistics for
small departures from the ideal design. This requires that the square
root of the parameter of Equation (9.33) be small compared
to 1.
For the 13-layer coating of Section 9.4, the parameter ei /Ei 2 
has a maximum value of about 0.1 for a peak thickness variation of
δmax = 01 nm.
As δmax increases the second-order series expansions of Sec-
tion 9.2 are no longer adequate. The analysis may be extended by
taking higher order terms, requiring correspondingly higher order
Kronecker products. However, for large enough δmax this method
will fail; one hopes that this will occur beyond the region of practical
interest.

For δmax = 1 nm in the present 13-layer example, ei /Ei 2  is
greater than 1 in the band center. The present method clearly fails
in this region.
Two methods may be used to verify the present results: 1. Higher
order series expansions, with higher order Kronecker products.
2. Simulation [1, 2]. If the probability density of Equation (9.31) is
replaced by a discrete distribution, exact results may be obtained.
For example, evaluation of 8192 matrix products of Equations (9.4)
and (9.5) yields exact transfer function statistics if δ takes on only
two values ±δmax with probabilities 21 
 REFERENCES 159

REFERENCES

1. Neil A. Jackman, Limitations and Tolerances in Optical Devices, Ph.

D. Dissertation, Stevens Institute of Technology, Hoboken, NJ, 1994;
Chapter 5.
2. Neil A. Jackman, “Multilayer Optical Filters with Random Errors,” Bell
Labs Technical Journal, January–March 1998, pp. 112–121.
3. Alfred Thelen, Design of Optical Interference Coatings, McGraw-Hill,
New York, 1988.
4. Pochi Yeh, Optical Waves in Layered Media, Wiley, New York, 1988.
5. H. A. Macleod, Thin-Film Optical Filters, McGraw-Hill, New York,
1989.
6. J. A. Dobrowolski, “Numerical Methods for Optical Thin Films,” Optics
and Photonics News, June 1997, pp. 23–33.
7. Harrison E. Rowe, “Propagation in One-Dimensional Random Media,”
IEEE Transactions on Microwave Theory and Techniques, Vol. MTT-19,
January 1971, pp. 73–80.
8. D. Marcuse, “Coupled Power Equations for Backward Waves,” IEEE
Transactions on Microwave Theory and Techniques, Vol. MTT-20, Au-
gust 1972, pp. 541–546.
9. Dietrich Marcuse, Theory of Dielectric Optical Waveguides, 2nd ed., Aca-
demic Press, New York, 1991.
10. Alfred Thelen, reference [3], page 201, Figure 10.3.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

CHAPTER TEN

Conclusion

The main thrust of the present work is the use of Kronecker prod-
uct methods in the calculation of transmission statistics of multi-
mode systems with random coupling. Such systems exhibit linearity
between complex mode amplitudes; Kronecker products yield linear
relations between corresponding moments of the mode amplitudes.
These methods require that the system must be regarded as a
cascade of sections, described by a matrix product. This descrip-
tion is natural for systems such as multi-layer coatings. For systems
with continuous coupling, a matrix description must be artificially
introduced.
For forward-traveling modes with white coupling, i.e., resulting
from imperfections with a very short correlation length, the sys-
tem may be divided into sections of infinitesimal length; exact linear
differential equations for the mode powers or for higher moments
are obtained. For more general spectra, the sections must be long
enough for the coupling in different sections to be approximately
statistically independent, and the coupling must be small enough for
perturbation theory to be applied to the individual sections; the re-
sulting coupled power or other differential equations for the higher
moments are only approximate.
Multi-layer coatings offer an important two-mode system with
modes traveling in opposite directions. A combination of Kronecker
product methods with perturbation theory yields approximate trans-
fer function statistics for systems with performance close to that of
the ideal design.

161
162 CONCLUSION

The use of MAPLE or an alternative program capable of symbolic

computation is essential for the analysis described here.
The examples described here have been included to illustrate the
possibilities of the present methods; they are not intended to be
exhaustive. Other models, coupling statistics, input signal statistics
may be similarly treated.
The region of validity for the approximate results that have been
presented are only heuristic. Accuracy of these methods is confirmed
by studying special cases that admit exact solutions. These include
forward-traveling modes with random square-wave coupling having
Markov statistics, and multi-layer coatings with random thickness
having a discrete probability distribution. Some results for the former
case have been presented here.
Finally, similar matrix methods have appeared in the physics lit-
erature [1–3].

REFERENCES

1. A. R. McGurn, K. T. Christensen, F. M. Mueller, and A. A. Maradudin,

“Anderson localization in one-dimensional randomly disordered optical
systems that are periodic on average,” Physical Review B, Vol. 47, 15 May
1993-II, pp. 13 120–13 125.
2. J. B. Pendry, “Symmetry and transport of waves in one-dimensional dis-
ordered systems,” Advances in Physics, Vol. 43, 1994, pp. 461–542.
3. V. D. Freilikher, B. A. Liansky, I. V. Yurkevich, A. A. Maradudin, and A.
R. McGurn, “Enhanced transmission due to disorder,” Physical Review
E, Vol. 51, June 1995, pp. 6301–6304.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

APPENDIX A

Series Solution for the

Coupled Line Equations

We apply the method of successive approximations (Picard’s

method) [1, 2] to the normalized coupled line equations for two
modes, given in Equations (2.32)–(2.34) [3]. Recall that I0
z and
G0
z denote the signal mode, I1
z and G1
z denote the spurious
mode. The signal mode has lower loss; consequently,

z
α
z = α0
z − α1
z ≤ 0 Re γ
z = α
xdx < 0 (A.1)
0

Denote the exact solution to Equation (2.33) by

       
G0
z G0
0 p00 p01 G0
0
=P· = · (A.2)
G1
z G1
0 p10 p11 G1
0

The method of successive approximations yields infinite series for

the elements of P, of which the first two terms are the perturbation
solution M of Equations (2.35) and (2.36).
Consider first the initial conditions

G0
0 = 1 G1
0 = 0 (A.3)

in order to obtain p00 and p10 . Let G0
n
z and G1
n
z be the
nth approximation to the solution of Equation (2.33). Let the initial
approximation be given simply by the initial conditions of Equation
163
164 SERIES SOLUTION FOR THE COUPLED LINE EQUATIONS

(A.3). The successive approximations are:

G0
0
z = 1 G1
0
z = 0

G0
1
z = 1

z
G1
1
z = j c
xe−γ
x dx
0

z
x
G0
2
z = 1 − c
xe+γ
x dx c
ye−γ
y dy
0 0

z (A.4)
G1
2 = j c
xe−γ
x dx
0

··· ··· ···

z
G0
n
z = 1 + j c
xe+γ
x G1
n−1
xdx
0

z
G1
n
z = j c
xe−γ
x G0
n−1
xdx
0

The nth approximation is obtained by substituting the
n − 1th ap-

proximation in the right-hand side of Equation (2.33) and integrat-
ing. Define
G0
n
z − G0
n−1
z = p00
n
z
(A.5)
G1
n
z − G1
n−1
z = p10
n
z
Then,

n
G0
n
z = 1 + p00
k
z
k=1
(A.6)

n
G1
n
z = p10
k
z
k=1

The terms of Equation (A.6) satisfy the following relations:

z
p00
n
z = j c
xe+γ
x p10
n−1
xdx n ≥ 1
0

z
p10
n
z = j c
xe−γ
x p00
n−1
xdx n≥1 (A.7)
0

p00
0
z = 1 p10
0
z = 0

 SERIES SOLUTION FOR THE COUPLED LINE EQUATIONS 165

The limn→∞ of the summations in Equation (A.6) yields the first

column of the matrix P in Equation (A.2):


∞
p00 = 1 + p00
n
z
n=1
(A.8)

∞
p10 = p10
n
z
n=1

Note that:

p00
n
z = 0 n odd; p10
n
z = 0 n even. (A.9)

The first and second summations of Equations (A.6) and (A.8) con-
tain only even and odd terms, respectively.
We obtain the following bounds on the terms of Equations (A.6)
and (A.8):
 z

≤ 0
c
x dx n
 n even.
p00
n
z n! (A.10)


= 0 n odd.


 = 0
 n even.
p10
n
z z
c
x dx n − (A.11)

≤ 0
e
z
0 α
xdx
 n odd.
n!

Let Equation (A.10) hold for some even n. Then from the middle
relation of Equation (A.7),

z x
x
c
y dy n
0
p10
n+1
z ≤ c
x e− 0 α
ydy
dx
0 n!
e−
z
0 α
ydy
z
x n
x
≤ c
y dy d c
y dy
n! 0 0 0
z n+1
0
c
x dx z
= e− 0 α
xdx
(A.12)

n + 1!
166 SERIES SOLUTION FOR THE COUPLED LINE EQUATIONS

From Equation (A.12) and the first relation of Equation (A.7),

z x
x
c
y dy n+1
x
+ α
ydy 0
p00
n+2
z ≤ c
x e 0 e− 0 α
ydy
dx
0
n + 1!
1
z
x n+1
x
= c
y dy d c
y dy

n + 1! 0 0 0
z n+2
0
c
x dx
= (A.13)

n + 2!
The bounds of Equations (A.10) and (A.11) hold for all n by
induction.
The exact solutions for two degenerate modes, Equation (2.8),
and for a discrete coupler, Equation (2.11), attain the limits given
in Equations (A.10) and (A.11). Therefore these bounds cannot be
improved without additional restrictions on the coupling coefficient
c
z.
Similar results for the second column of the matrix P in Equation
(A.2) are readily obtained:

∞
p01 = p01
n
z
n=1
(A.14)

∞
p11 = 1 + p11
n
z
n=1
 z

≤ 0
c
x dx n − z
α
xdx
e 0  n even, n ≥ 2
p11
n
z n! (A.15)


= 0 n odd.

 = 0
 n even.
p01
n
z z
c
x dx n (A.16)

≤ 0
 n odd.
n!
p11
0
z = 1 p01
0
z = 0 (A.17)

The multi-mode case can be treated in a similar manner. We in-

troduce compact notation for Equation (2.69):

G
z = jc
zK
z · G
z (A.18)

 SERIES SOLUTION FOR THE COUPLED LINE EQUATIONS 167

where

K
z = e+

z · C · e−

z (A.19)

Define

z
γij
z = i
z − j
zdz (A.20)
0

Then writing out the matrix K
z

 
0 eγ01
z C01 eγ02
z C02 ···
 −γ
z 
 e 01 C01 0 eγ12
z C12 ··· 
 
K
z =  −γ
z  (A.21)
 e 02 C02 e−γ12
z C12 0 ··· 
 

The method of successive approximations yields


∞
G
z = G
0 + p
n
z (A.22)
n=1

where

z
p
n
z = j c
xK
x · p
n−1
xdx p
0
z = G
0 (A.23)
0

Truncating the summation of Equation (A.22) at n = 2 yields the

perturbation solution of Equation (2.70). Bounds on the terms of the
summation in Equation (A.22) for the lossless case are as follows:
z n
c
x dx
0
p
n
z < Cn G
0
n! (A.24)
αi
z = 0 i = 0 1 2 · · · 

where   denotes the vector or matrix norm [4, 5];

  
C = max Cij  G
0 = Gi
0 = Ii
0 (A.25)
j
i i i
168 SERIES SOLUTION FOR THE COUPLED LINE EQUATIONS

REFERENCES

1. E. L. Ince, Ordinary Differential Equations, Dover Publications, New

York, 1956.
2. R. Bellman, Stability Theory of Differential Equations, McGraw-Hill, New
York, 1953.
3. H. E. Rowe, “Approximate Solutions for the Coupled Line Equations,”
Bell System Technical Journal, Vol. 41, May 1962, pp. 1011–1029.
4. Assem S. Deif, Advanced Matrix Theory for Scientists and Engineers,
Wiley, New York, 1982.
5. Gilbert Strang, Linear Algebra and its Applications, 2nd ed., Academic
Press, New York, 1980.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

APPENDIX B

General Transmission
Properties of Two-Mode Guide

We obtain general transmission properties for the signal mode trans-

fer function and impulse response of two-mode guide from the exact
solution given in Appendix A [1]. The infinite series solution for the
G0 , given in Equations (A.6)–(A.9), takes on the following form for
a guide of length L with constant propagation functions:


∞
G0
 = 1 + p00
2n
 (B.1)
n=1

where p00
n
 is given in Equation (A.7), and may be rewritten as

 L  x1  x2n−1
p00
2n
 =
−1n dx1 dx2 · · · dx2n
0 0 0

· c
x1 c
x2  · · · c
x2n  · e
x1 −x2 +x3 −x4 +···−x2n  (B.2)

We have suppressed the L dependence in Equations (B.1) and (B.2),
as discussed in Section 3.2. Bounds on the terms of Equation (B.1)
are given in Equation (A.10):
L
c
xdx 2n
p00
2n
 ≤ 0  α ≤ 0 (B.3)

2n!
Under suitable conditions the individual terms p00
2n
 are
analytic in the  plane, the series in Equation (B.1) is uniformly
convergent, and G0
 is analytic in the  plane [2].
169
170 GENERAL TRANSMISSION PROPERTIES OF TWO-MODE GUIDE

Set

s = x1 − x2 + x3 − x4 + · · · − x2n (B.4)

in Equation (B.2) to yield

 L
p00
2n
α β = fn
seαs ejβs ds (B.5)
0

where fn
s is a real,
2n − 1-fold integral. Set

s = Lτ (B.6)

in
 Equation
 (B.5), substitute into Equation (B.1), and interchange
and ; then
 1 

∞ 
G0
α β = 1 + L fn
Lτ eαLτ ejβLτ dτ (B.7)
0 n=1

Finally, define

∞
f
Lτ = fn
Lτ (B.8)
n=1

to yield
 1
G0
α β = 1 + f
LτeαLτ ejβLτ (B.9)
0

Comparing Equation (B.9) with Equation (3.17),


δ
τ + Lf
LτeαLτ  0 ≤ τ ≤ 1
gα
τ = (B.10)
0 otherwise.

gα
τ is causal, time-limited, and real.

Define

G0
α β = 1 − A
α β + j
α β (B.11)

where A and  are real. A is the departure of Re G0 from unity,

and  is Im G0 , respectively. (For small coupling A ≈ loss and
 ≈ phase, but we do not make these approximations here.) Then
 GENERAL TRANSMISSION PROPERTIES OF TWO-MODE GUIDE 171

Equation (B.10) yields the following four properties of G0
α β

and gα
τ:

1. A and  are Hilbert transforms:


 = A (B.12)

where  denotes the Hilbert transform,

 1 1
A
α β = A
α β 
π β
1  ∞ A
α σ
= dσ (B.13)
π −∞ β − σ

and  represents the convolution operator. The inverse rela-

tionship is

 + A 1 B
A= A = lim A
α σdσ (B.14)
B→∞ 2B −B

A represents the d.c. component of A

2. From Equation (B.10),

gα
τ = e−αLτ gα=0
τ α ≤ 0 (B.15)

Therefore G0 and gα for any negative α are related to the

corresponding quantities for α = 0 as follows:

1 1
G0
α β = G0
0 β  (B.16)
2π α − jβ

1 α
A
α β = A
0 β  (B.17)
π α + β2
2

1 α

α β = 
0 β  (B.18)
π α + β2
2

Equations (B.17) and (B.18) show that as the differential loss

increases (α becomes more negative), the transfer function
fluctuations are reduced. The window function of these rela-
172 GENERAL TRANSMISSION PROPERTIES OF TWO-MODE GUIDE

tions has unit area:

1 ∞ 1
d
β = 1 (B.19)
πα −∞ 1 + β 2
α

This window function must approach a unit impulse for small

α:

1 1
lim 2
= δ
β (B.20)
α→0 πα
1 + β α

3. G0 is strictly band-limited, with sample points

2πn
βn = (B.21)
L

Then,

∞
2πn sin π
βL − n
G0
α β = ejπ
2π 
βL
2π
G0 α
−1n
n=−∞ L βL
π
2π − n
(B.22)

4. G0 is Hermitian, A symmetric, and  anti-symmetric:

G0
α −β = G∗0
α β

A
α −β = A
α β (B.23)

α −β = −
α β

G0
 is analytic in the entire  plane, and therefore has no

poles. Results
 ∞ for the location of 0’s of G0
 yield sufficient con-
ditions on −∞ c
xdx that guarantee G0 will be minimum-phase
[3]. We omit these results, since they do not appear in the present
work.
All of the present results are exact; in particular they are not
restricted to the small-coupling regime required by perturbation
theory.
 REFERENCES 173

REFERENCES

1. Harrison E. Rowe and D. T. Young, “Transmission Distortion in Mul-

timode Random Waveguides,” IEEE Transactions on Microwave Theory
and Techniques, Vol. MTT-20, June 1972, pp. 349–365.
2. E. A. Coddington and N. Levinson, Theory of Ordinary Differential Equa-
tions, McGraw-Hill, New York, 1955; p. 40, prob. 7.
3. Harrison E. Rowe and D. T. Young, “Minimum-Phase Behavior of Ran-
dom Media,” IEEE Transactions on Microwave Theory and Techniques,
Vol. MTT-23, May 1975, pp. 411–416.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

APPENDIX C

Kronecker Products

Consider two arbitrary matrices

   
a11 a12 · · · b11 b12 ···
   
A= 
 a21 a22 · · ·  B=
 b21 b22 ···  (C.1)
     
     

not necessarily square. The ordinary matrix product of two con-

formable matrices is denoted by A · B; the numbers of columns of
A and rows of B must match. The Kronecker product of these two
matrices is denoted by [1, 2]
 
a11 B a12 B · · ·
 
A⊗B=  a21 B a22 B · · · 
 (C.2)
  
  

There is no restriction on the dimensions of the two matrices. A

representative submatrix of Equation (C.2) is
 
aij b11 aij b12 · · ·
 
aij B =  
 aij b21 aij b22 · · ·  (C.3)
  
  

The Kronecker product A ⊗ B has dimensions equal to the products

of the dimensions of A and B.
175
176 KRONECKER PRODUCTS

The following properties are used throughout the present text

[1, 2]:

1. Both Kronecker products and ordinary matrix products are

not commutative:

A ⊗ B = B ⊗ A (C.4)

2. Kronecker products are associative and distributive:

A ⊗ B ⊗ C = A ⊗ B ⊗ C = A ⊗ B ⊗ C (C.5)
A + B ⊗ C + D = A ⊗ C + A ⊗ D + B ⊗ C + B ⊗ D (C.6)

Both A and B must be the same size, and C and D must be

the same (possibly different) size in Equation (C.6).
3. Interchange of Kronecker and ordinary matrix products is
essential to the present work:

A ⊗ B · C ⊗ D = A · C ⊗ B · D (C.7)

Both A and C must be conformable, and B and D must

be conformable, in Equation (C.7). This relation is readily
generalized;

A1 ⊗ A2 ⊗ A3 ⊗ · · · · B1 ⊗ B2 ⊗ B3 ⊗ · · ·
· C1 ⊗ C2 ⊗ C3 ⊗ · · · · · · ·
= A1 · B1 · C1 · · · · ⊗ A2 · B2 · C2 · · · ·
⊗ A3 · B3 · C3 · · · · ⊗ · · ·  (C.8)

The matrix dimensions in each of the parentheses in the sec-

ond line of Equation (C.8) must satisfy the conformability
requirements.
4. Inverse and transpose:

A ⊗ B−1 = A −1 ⊗ B−1  (C.9)

A ⊗ BT = A T ⊗ BT  (C.10)

Both A and B must be square in Equation (C.9). These rela-

tions generalize in an obvious way.
 REFERENCES 177

5. Denote the eigenvalues and eigenvectors of A and B by αi  βj

and ai  bj , respectively,

A · ai = αi ai  B · bj = βj bj  (C.11)

Define γk and ck by

A ⊗ B · ck = γk ck  (C.12)

Then from Equation (C.7)

γk = αi βj  ck = ai ⊗ bj  (C.13)

The eigenvalues of A ⊗ B are products of all pairs of eigen-

values, and the eigenvectors of A ⊗ B are Kronecker products
of all pairs of eigenvectors, of A and B.
6. Let a be a column vector and bT be a row vector.

a · bT = a ⊗ bT = bT ⊗ a (C.14)

REFERENCES

1. R. Bellman, Introduction to Matrix Analysis, McGraw-Hill, New York,

1970; Chapters 12 and 15.
2. F. A. Graybill, Introduction to Matrices with Applications to Statistics,
Wadsworth, Belmont, California, 1969; Section 8.8
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

APPENDIX D

Expected Values of
Matrix Products

D.1. INDEPENDENT MATRICES

Consider the scalar problem

I
n = M
nM
n − 1 · · · M
2M
1I
0 (D.1)

where the M
k are independent random variables. For example,

this relation may represent a chain of n noninteracting amplifiers
with random independent gains, with input I
0 and output I
n.
The expected output is

I
n = M
nM
n − 1 · · · M
2M
1I
0 (D.2)

Assume the different M
k have the same statistics; then

I
n = Mn I
0 (D.3)

Similarly the expected output power is

I
n2  = M2 n I
02  (D.4)

We require the matrix extension of these elementary results for

the N-mode case. First, the expected value of a matrix is defined
179
180 EXPECTED VALUES OF MATRIX PRODUCTS

as the matrix of expected values of its elements; i.e., the expected

value of
 
a11 a12 ···
 
A=  a21 a22 ···  (D.5)
  
  

is
 
a11  a12  ···
 
A = 
 a21  a22  ···  (D.6)
  
  

Consider an N-mode guide divided into n sections. Denote the col-

umn vector of mode complex amplitudes after k sections by I
k:

I T
k = I0
k I1
k I2
k ··· IN−1
k  (D.7)

Let M
k denote the wave matrix describing the kth section:

I
k = M
k · I
k − 1 (D.8)

Then the vector output after n sections is

I
n = M
n · M
n − 1 · · · · · M
2 · M
1 · I
0


n
= M
n + 1 − k · I
0 (D.9)
k=1

where I
0 is the vector input, and the M
k are N × N random
matrices.
Assume different M
k are independent. Any element of M
k
is statistically independent of any element of M
r, k = r; however,
different elements of any particular matrix are not independent, e.g.,
see Equation (3.24). Denote the elements of M
k by mij
k; then
mij
k is independent of mpq
r if k = r for any i j p q The th
element of Equation (D.9) is

I
n = mk1
nmk1 k2
n − 1 · · · mkn−1 kn Ikn
0 (D.10)
k1 ···kn
 D.1. INDEPENDENT MATRICES 181

Since different m’s in each term of the summation of Equation

(D.10) are statistically independent, the expected value of Equation
(D.10) yields the average outputs:

I
n = mk1
nmk1 k2
n − 1 · · · mkn−1 kn Ikn
0 (D.11)
k1 ···kn

In matrix notation, Equation (D.11) becomes

I
n = M
n · M
n − 1 · · · · · M
2 · M
1 · I
0


n
= M
n + 1 − k · I
0 (D.12)
k=1

The alternative result

I
k = M
k · I
k − 1 (D.13)

follows from Equation (D.12). Finally, if the M
k have the same
statistics,

I
n = Mn · I
0

(D.14)
I
k = M · I
k − 1

Second- and higher order statistics are obtained by the use of

Kronecker products, described in Appendix C [1–3]. For second or-
der, take the Kronecker product of Equation (D.9) with its complex
conjugate, and use Equation (C.8) to obtain


n
I
n ⊗ I ∗
n = M
n + 1 − k ⊗ M∗
n + 1 − k · I
0 ⊗ I ∗
0 
k=1

(D.15)

where the  denotes ordinary (·) matrix multiplication; the M
k ⊗

M∗
k are N 2 × N 2 random matrices. The expected value of Equa-
tion (D.15) is obtained by the same argument used above for the
182 EXPECTED VALUES OF MATRIX PRODUCTS

average output:

n
I
n ⊗ I ∗
n = M
n + 1− k ⊗ M∗
n + 1− k · I
0 ⊗ I ∗
0
k=1

(D.16)
If the M
k have the same statistics,
I
n ⊗ I ∗
n = M ⊗ M∗ n · I
0 ⊗ I ∗
0
(D.17)
I
k ⊗ I ∗
k = M ⊗ M∗  · I
k − 1 ⊗ I ∗
k − 1
We write out Equation (D.17) explicitly for the two-mode case:
   
I0
n2  I0
02 
   
 I0
nI1∗
n   I
0I1∗
0 
  = M ⊗ M∗ n ·  0  (D.18)
 ∗   ∗ 
 I1
nI0
n   I1
0I0
0 
I1
n2  I1
02 
where
 
m11 2  m11 m∗12  m12 m∗11  m12 2 
 
 m11 m∗21  m11 m∗22  m12 m∗21  m12 m∗22  
M ⊗ M  =
∗
 ∗ ∗ ∗ ∗
 (D.19)

 m21 m11  m21 m12  m22 m11  m22 m12  
m21 2  m21 m∗22  m22 m∗21  m22 2 
The input vector on the right-hand side of Equation (D.18) will nor-
mally be simplified. For example, if the signal mode I0 has determin-
istic unit input and the spurious mode I1 has zero input, the input
vector will be
 
1
 
0
I
0 ⊗ I ∗
0 = I
0 ⊗ I ∗
0 = 
 
 (D.20)
0
0
Higher order statistics are obtained in a similar manner. The ma-
trices in the analysis increase in size as the order and/or the number
of modes increases.
 D.2. MARKOV MATRICES 183

D.2. MARKOV MATRICES

D.2.1. Markov Chains

Let the random variables and matrices M
k and M
k, in Equa-
tions (D.1) and (D.9), respectively, each form a Markov chain. In
the independent case of Section D.1, these random variables were
continuous; in the present Markov case they are restricted to be dis-
crete. For each k, M
k takes on one of L scalar values

a1  a2  · · ·  aL  (D.21)

or M
k takes on one of L matrix values

A1  A2  · · ·  AL  (D.22)

We may treat scalars by specializing the matrix case to 1×1 matrices.

Define the unconditional and (conditional) transition probabilities
as [4]

pi
k = PM
k = Ai  (D.23)

pij = PM
k = Ai M
k − 1 = Aj  (D.24)

Note that the unconditional probability is in general a function of

k, while we assume the transition probability is independent of k.
Define the associated column vector and transition matrix as follows:

pT
k = p1
k p2
k · · · pL
k  (D.25)

 
p11 p12 · · · p1L
 
 p21 p22 · · · p2L 
P=   

  (D.26)

     
pL1 pL2 · · · pLL

Then,

p
k = P · p
k − 1 (D.27)

184 EXPECTED VALUES OF MATRIX PRODUCTS

We restrict the present treatment to the stationary case:

p
k = p (D.28)

From Equations (D.27) and (D.28), the solution for the unit eigen-
value of

P − I  · p = 0 (D.29)

where I is the L × L unit matrix and 0 is the zero L × 1 column

vector, yields the unconditional probabilities for the stationary case.

D.2.2. Scalar Variables

We shall require the spectrum of random square-wave coupling in
Chapter 8. Let the scalar M
k denote the coupling in the kth sec-
tion of the coupling function, n sections long. Then, we must deter-
mine M
nM
1 in order to calculate the spectrum of the coupling
function. We do so below for the Markov case.
Assume that M
k form a stationary Markov chain, with possi-
ble values Equation (D.21), transition matrix Equation (D.26), and
stationary unconditional probability vector Equations (D.25) and
(D.29). We require the following additional matrices:

   
a1 a1 0 ··· 0
   
 a2   0 a2 ··· 0 
a=
 

 aD = 
    

 (D.30)
        
aL 0 0 ··· aL
   
p1 p1 0 ··· 0
   
 p2   0 p2 ··· 0 
p=
 

 pD = 
    

 (D.31)
        
pL 0 0 ··· pL

1T = 1 1 ··· 1  (D.32)
L elements
 D.2. MARKOV MATRICES 185

The joint probability for M
n and M
1 is

PM
n = ai  M
1 = aj 

= pik1 pk1 k2 · · · pkn−3 kn−2 pkn−2 j pj  (D.33)
k1 ···kn−2

Therefore,

M
nM
1 = ai PM
n = ai  M
1 = aj aj  (D.34)
ij

In matrix notation, Equations (D.33) and (D.34) yield

M
nM
1 = 1T · aD · Pn−1 · pD · aD · 1 (D.35)
where the component matrices are given by Equations (D.25) and
(D.29), (D.26), and (D.30)–(D.32).
Consider the symmetric binary case as an example; i.e., M
k =
±a with p the probability that adjacent M
k are different:
 1
 
1−p p 2
+a
P=  p=  a=  (D.36)
1
p 1−p 2
−a
Diagonalizing P:
  −1
1 1 1 − 2p 0 1 1
P= · · 
−1 1 0 1 −1 1
−1 
1 1 1 1 −1
=  (D.37)
−1 1 2 1 1
Then Equation (D.35) becomes
 
1 a 0 1 1
M
nM
1 = 1 1 · ·
2 0 −a −1 1
 n−1 
1 − 2p 0 1 −1
· ·
0 1 1 1
1
  
2
0 a 0 1
· · ·
0 21 0 −a 1
= a2
1 − 2pn−1  (D.38)
186 EXPECTED VALUES OF MATRIX PRODUCTS

D.2.3. Markov Matrix Products

Assume the M
k of Equation (D.9) form a stationary Markov
chain with possible values Equation (D.22), transition matrix Equa-
tion (D.26), and stationary unconditional probability vector Equa-
tions (D.25) and (D.29). For N modes, M
k and Ai are N × N
matrices. We seek corresponding results to those of Equations
(D.12)–(D.20) for independent matrices. Calculation of M
n ·
M
n − 1 · · · M
2 · M
1 requires the following additional matrices
[5]; the size of each matrix is indicated by subscripts:
 
1 0 ··· 0
 
 0 1 ··· 0 

I N×N =      (D.39)
      

0 0 0 1
 
p11 I p12 I · · · p1L I
 
 p21 I p22 I · · · p2L I 
PLN×LN = PL×L ⊗ I N×N =    

  (D.40)
     
pL1 I pL2 I ··· pLL I
 
A1 0 ··· 0
 
 0 A2 ··· 0 
ALN×LN =
     
 (D.41)
     
0 0 · · · AL
 
p1 I
 
 p2 I 
pLN×N = pL×1 ⊗ I N×N =  

 (D.42)
  
pL I

IN×LN = I I ··· I  (D.43)

The joint probability for the n matrices of Equation (D.9) is
PM
n = Ai  M
n − 1 = Ak1  · · ·  M
2 = Akn−2  M
1 = Aj 
= pik1 pk1 k2 · · · pkn−2 j pj (D.44)
 D.2. MARKOV MATRICES 187

Therefore the expected value of the matrix product in Equation

(D.9) is

M
n · M
n − 1 · · · M
2 · M
1

 
 
= Ai pik1 · Ak1 pk1 k2 · · · Akn−3 pkn−3 kn−2 · Akn−2 pkn−2 j pj · Aj pj 
ij k1 ···kn−2

(D.45)

Now

A · Ppq = Ap ppq (D.46)

is the pqth submatrix of A·P. Consequently the quantity in
 inside

the first summation on the right side of Equation (D.45) is the ijth
submatrix of the
n − 1th power of A · P:
 

n−1
Ai pik1 · Ak1 pk1 k2 · · · Akn−3 pkn−3 kn−2 · Akn−2 pkn−2 j pj =
A · Pij 
k1 ···kn−2

(D.47)

Similarly,

A · pj = Aj pj (D.48)

is the jth submatrix of A · p. Therefore Equations (D.45)–(D.48)

yield

M
n · M
n − 1 · · · M
2 · M
1


n−1
=
A · Pij
A · pj
i j

=
A · P
n−1
A · p i
i

= I
A · P
n−1
A · p (D.49)

For the N-mode guide described in Section D.1, substitute Equa-

tion (D.49) into the expected value of Equation (D.9) to obtain

I
n = I ·
A · P
n−1 ·
A · p · I
0 (D.50)

188 EXPECTED VALUES OF MATRIX PRODUCTS

where I
0 is the vector input, and the other matrix quantities are
given in Equations (D.39)–(D.43).
Second- (and higher) order statistics are again found by use of
Kronecker products. For second-order, make the following substitu-
tions in the above analysis:
M
k −→ M
k ⊗ M∗
k
Ai −→ Ai ⊗ Ai∗
(D.51)
N −→ N 2
A −→ A⊗
Equations (D.39)–(D.43) become
 
1 0 ··· 0
 
0 1 ··· 0 
I N 2 ×N 2 = 
     
 (D.52)
     
0 0 0 1

PLN 2 ×LN 2 = PL×L ⊗ I N 2 ×N 2

 
p11 I p12 I ··· p1L I
 
 p21 I p22 I ··· p2L I 
=    

 (D.53)
     
pL1 I pL2 I ··· pLL I

 
A1 ⊗ A1∗ 0 0 0
 
 0 A2 ⊗ A2∗ ··· 0 
A⊗
LN 2 ×LN 2
=
    

 (D.54)
     
0 0 ··· AL ⊗ AL∗
 
p1 I
 
 p2 I 
pLN 2 ×N 2 = pL×1 ⊗ I N 2 ×N 2 =
 

 (D.55)
  
pL I
 D.2. MARKOV MATRICES 189

IN 2 ×LN 2 = I N 2 ×N 2 I N 2 ×N 2 ··· I N 2 ×N 2  (D.56)

L unit matrices

Note that A⊗
LN 2 ×LN 2
of Equation (D.54) is not equal to the Kronecker
product of ALN×LN of Equation (D.41) with itself. Then the expected
value of Equation (D.15) yields

I
n ⊗ I ∗
n = I ·
A⊗ · P
n−1 ·
A⊗ · p · I
0 ⊗ I ∗
0 (D.57)

where the matrix quantities are now given in Equations (D.52)–

(D.56). If the input consists of only a single deterministic (signal)
mode, the input vector I
0 ⊗ I ∗
0 will contain only a single nonzero
element.
We illustrate these results for the symmetric binary case; here the
transition matrix P and unconditional probability vector p are given
by Equation (D.36). Then, Equations (D.50) and (D.57) become


n−1 
1 
1 − pA1 pA1 A1
I
n = I I · · · I
0
2 pA2
1 − pA2 A2
(D.58)

where I is the 2 × 2 unit matrix, and

I
n ⊗ I ∗
n

n−1
1 
1 − pA1 ⊗ A1∗ pA1 ⊗ A1∗
= I I ·
2 pA2 ⊗ A2∗
1 − pA2 ⊗ A2∗

A1 ⊗ A1∗
· · I
0 ⊗ I ∗
0 (D.59)
A2 ⊗ A2∗

where I is the 4 × 4 unit matrix. For p = 21 , the matrices M
k

are independent; Equations (D.58) and (D.59) reduce to Equations
(D.14) and (D.17), respectively.
190 EXPECTED VALUES OF MATRIX PRODUCTS

REFERENCES

1. R. Bellman, Introduction to Matrix Analysis, McGraw-Hill, New York,

1970; Chapters 12 and 15.
2. Harrison E. Rowe and D. T. Young, “Transmission Distortion in Mul-
timode Random Waveguides,” IEEE Transactions on Microwave Theory
and Techniques, Vol. MTT-20, June 1972, pp. 349–365.
3. Harrison E. Rowe, “Waves with Random Coupling and Random Propa-
gation Constants,” Applied Scientific Research, Vol. 41, 1984, pp. 237–255.
4. Athanasios Papoulis, Probability, Random Variables, and Stochastic Pro-
cesses, McGraw-Hill, New York, 2nd. ed., 1984; Chapter 12.
5. Behram Homi Bharucha, On the Stability of Randomly Varying Systems,
Ph.D. Thesis, University of California, 1961; Appendix D.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

APPENDIX E

Time- and Frequency-Domain

Statistics

E.1. SECOND-ORDER IMPULSE RESPONSE STATISTICS

We derive the time-domain results used in Section 3.7 for a wide-

sense stationary random transfer function [1]. The filter impulse
response and transfer function of a real, causal filter are given in
Equations (3.98)–(3.101):

∞
ɡ
t = 
f ej2πft df (E.1)
−∞

ɡ
t = ɡ
t
∗
ɡ
t = 0 t < 0 (E.2)

∞

f  = ɡ
te−j2πft dt (E.3)
0


f  = ∗
−f  (E.4)

We denote the Fourier transform relation as follows:

ɡ
t ⇐⇒ 
f  (E.5)

The transfer function is wide-sense stationary by Equations

(3.102)–(3.103):


ν = 
f + ν∗
f   
ν = ∗
−ν (E.6)


f  =   (E.7)
191
192 TIME- AND FREQUENCY-DOMAIN STATISTICS

We assume that 
f  has no periodic components; then


∞ = 
−∞ =  2  (E.8)

It is convenient to divide 
f  into d.c. and a.c. components. To-

ward this end denote averages over frequency by . Then,
1
F
 = lim 
f df  
f  =  + ac
f  (E.9)
F→∞ 2F −F

We show that  is deterministic; i.e.,

=   (E.10)

Thus, from Equations (E.8) and (E.9),

2 1
F
F

 −   = lim 
f  
f   df  df 
F→∞
2F2 −F −F

−2   +  2
1
F
F
= lim 
f  − f  df  df  −  2
F→∞
2F2 −F −F

2
= 
∞ −  = 0 (E.11)

Therefore,


f  =  + ac
f  ac
f  = 0 (E.12)

2 ∗

ν =  + ac
ν ac
ν = ac
f + νac
f   (E.13)

From Equation (3.104),

∞

t = 
νej2πtν dν 
t = ∗
t (E.14)
−∞

where 
−t is the transfer function spectral density. Substituting

Equation (E.13):

∞

t =  2 δ
t + ac
t ac
t = ac
νe−j2πtν dν (E.15)
−∞

ac
t has no δ-function components.

 E.1. SECOND-ORDER IMPULSE RESPONSE STATISTICS 193

Additional properties of the real and imaginary parts of 
f  and


ν, not of direct interest here, are given in [1].
Cascade 
f  with an ideal band-pass filter, to yield the overall
transfer function 
f  of Equation (3.105):


f  f − f0  < B

f  = B < f0  (E.16)
0 f − f0  > B

The envelope of the overall response is given in terms of the impulse

response 
t and its Hilbert transform 
t by Equation (3.106),

t = 
t + j
t (E.17)

where 

t is given by Equation (3.107) in terms of the convolution
operator :

 1 1
∞ 
τ

t =  
t = dτ (E.18)
πt π −∞ t − τ

The positive and negative frequency content of 
t are:

1 1
+
t = 
t + j

t ⇐⇒ +
f  =
1 + sgn f 
f  (E.19)
2 2

1 1
−
t = 
t − j

t ⇐⇒ −
f  =
1 − sgn f 
f  (E.20)
2 2
where sgn
 is the signum function:

1 f > 0
sgn f = (E.21)
−1 f < 0

Figure E.1 shows a typical band-pass spectrum.

Then, using Equation (E.16),


t ⇐⇒ 4
2
+
f   −
f  = 4+
f   + ∗
f 

f0 +B−
f +f /2
=4 
a + f ∗
ada
f0 −B−
f −f /2

(E.22)
194 TIME- AND FREQUENCY-DOMAIN STATISTICS

( f )

–( f )
+( f )

f
– f0 – B – f0 – f0 + B f0 – B f0 f0 + B

FIGURE E.1. Band-pass spectrum.

Substituting Equation (E.6),

 
2B f 

t
2
= 8B
−2B
1−
2B

f ej2πft df (E.23)

From Equation (E.14),

  2 
sin 2πBt
 2

t = 8B 2B
2πBt
 
t (E.24)

From Equation (E.15),

 
2 
sin 2πBt

t
2
= 8B  2
2πBt
2B
  2 
sin 2πBt
+ 8B 2B  ac
t (E.25)
2πBt

Since
  2 
sin 2πBt
lim 2B = δ
t (E.26)
B→∞ 2πBt

we have the result stated in Equation (3.108):


t
2
≈ 8B · 
t B large, (E.27)

where 1/
2B is small compared to the width of ac
t, or equiva-

lently B is large compared to the correlation bandwidth of 
f .
 E.2. TIME-DOMAIN ANALYSIS 195

Finally, we show that the response of a wide-sense stationary

transfer function to white noise is linear in intensity [2]. From Equa-
tion (3.110)


t = 4
2
+

t−
t

∞
∞
=
τ1 
τ2 
τ1 
τ2 
−∞ −∞

· 4+
t − τ1 −
t − τ2 dτ1 dτ2  (E.28)

Averaging over the input and using Equation (3.109),

∞

t
2
=
−∞
2


τ 2
t − τdτ (E.29)

Averaging over the transfer function,


t
2
= 
t
2


∞
=
−∞
2

τ 
t − τ dτ
2

= 2

t  
t 
2
(E.30)

as given in Equation (3.111).

E.2. TIME-DOMAIN ANALYSIS

We derive the impulse response of the idealized two-mode guide of

Section 3.2 directly from the coupled line equations.
Consider a length of line dx for two coupled modes described by
Equations (3.4)–(3.5); the coupling coefficient for this section is

jc
xdx = j sgn f · C0 d
x dx (E.31)

from Equations (3.15) and (3.18). The corresponding impulse re-

sponse is the inverse Fourier transform of this quantity:

1
C0 d
xdx · j sgn f ⇐⇒ −C0 d
xdx  (E.32)
πt
196 TIME- AND FREQUENCY-DOMAIN STATISTICS

Thus, a signal δ
t at position x in one mode excites a response

−C0 d
xdx
1/πt at x in the other guide; similarly, a signal −1/πt
at x in one mode excites a response C0 d
xdxδ
t at x in the other
mode.
Assume a unit impulse incident on the signal mode at the guide
input z = 0, and apply the method of successive approximations in
the time domain. The elementary partial signal-mode impulse re-
sponse that has made n transitions from signal to spurious mode at
x2n  x2n−2  · · ·  x4  x2  and n transitions from spurious to signal
mode at x2n−1  x2n−3  · · ·  x3  x1  is given by
 
x − x2 + · · · + x2n−1 − x2n
δ t− 1 T · eα
x1 −x2 +···+x2n−1 −x2n 
L
· C02n d
x1 d
x2  · · · d
x2n−1 d
x2n dx1 dx2 · · · dx2n−1 dx2n 
L > x1 > x2 > · · · > x2n−1 > x2n  (E.33)

where T is given by Equation (3.20). Integrating Equation (E.33)

over the variables xi  1 ≤ i ≤ 2n yields the partial impulse re-
sponse composed of all components that have made exactly 2n tran-
sitions; note that this quantity normalized and Equation (B.2) are
Fourier transforms, as in Equations (3.16) and (3.17). Finally, sum-
ming over n 0 ≤ n ≤ ∞ yields the total signal mode impulse
response ɡα
t of Equations (3.22)–(3.23). The properties of Ap-
pendix B follow immediately.
A time-domain model that contained some of the essential fea-
tures of Equation (E.33) gave the first quantitative understanding of
the impulse response of random guides [2].

REFERENCES

1. Harrison E. Rowe and D. T. Young, “Transmission Distortion in Mul-

timode Random Waveguides,” IEEE Transactions on Microwave Theory
and Techniques, Vol. MTT-20, June 1972, pp. 349–365.
2. S. D. Personick, “Time Dispersion in Dielectric Waveguides,” Bell System
Technical Journal, Vol. 50, March 1971, pp. 843–859.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

APPENDIX F

Symmetric Slab Waveguide—

Lossless TE Modes

F.1. GENERAL RESULTS

Figure F.1 shows a symmetric dielectric slab waveguide of refractive

index n1 and thickness d suspended in a medium of index n2 .
Both media are nonmagnetic; i.e., they have the same permeability
as free space. We summarize the properties of this guide as fol-
lows [1]:
1
λ0 = √  free-space wavelength. (F.1)
f µ0 0

2π
k0 =  propagation constant in free space. (F.2)
λ0
2π
ki =  propagation constant in medium of index ni . (F.3)
λi


i λ
ni = = 0 n1 > n2 (F.4)
0 λi
The propagation constants of TE modes traveling in the z direction,
indexed by the subscript ν, are given by
 
βν = n21 k20 − κ2ν = k21 − κ2ν  (F.5)

197
198 SYMMETRIC SLAB WAVEGUIDE—LOSSLESS TE MODES

n2

n1
d
z

n2

n1 > n2

FIGURE F.1. Ideal symmetric slab waveguide.

where
 
γ= n21 − n22 k20 − κ2 = k21 − k22 − κ2 (F.6)

and κν is one of the solutions of the eigenvalue equations:

κd γ
tan =  even modes, (F.7)
2 κ

κd κ
tan =−  odd modes. (F.8)
2 γ

κν must be positive and γ real; hence,

 
0 < κν < κmax = n21 − n22 · k0 = k21 − k22 (F.9)

Equation (F.9) implies

k2 < βν < k1  (F.10)

the propagation constants of the TE modes in the waveguide lie

between the propagation constants of the core and cladding media.
 F.1. GENERAL RESULTS 199

It is helpful to rewrite Equation (F.6) in terms of κmax of Equa-

tion (F.9):

γ = κ2max − κ2 (F.11)
Equations (F.7) and (F.8) may be solved with Equation (F.11) for
the eigenvalues by plotting tan κd/2, γ/κ and −κ/γ on the same
graph and finding the intersections. Figure F.2 shows such a plot for
a special case of interest in Sections 4.5, 7.1, and F.2.
10

tan d
2
5

0
0 2 4 6 8 10 12 d

–5

–10

max d = 3.5

FIGURE F.2. Solution for the eigenvalues.

200 SYMMETRIC SLAB WAVEGUIDE—LOSSLESS TE MODES

n2

n1 c(z)
d z

n2

n1 > n2

FIGURE F.3. Symmetric slab waveguide with random straightness deviation.

For this case, κmax d = 3 5π, there are four TE modes, two even
and two odd. More generally, for N − 1π < κmax d < Nπ there are
N TE modes, N + 1/2 even and N/2 odd, where x represents
the greatest integer contained in x.
Finally, consider the slab guide with random straightness deviation
cz shown in Figure F.3. The coupling coefficients between two
forward-traveling TE modes are given as follows:
2κµ κν εµν
Cµν = −    1/2 (F.12)
1 1
βµ βν d+ d+
γµ γν

1 one mode even and the other odd.
εµν = (F.13)
0 both modes even or both odd.
Cµν is to be substituted into Equations (3.35) and (3.13); cz in
Equation (3.9) represents the straightness deviation of the random
guide, shown in Figure F.3.

F.2. EXAMPLE

We take the following parameters for the waveguide of Figures F.1

and F.3:
λ0 = 10−6 m.; n1 = 1 01 n2 = 1 (F.14)
 F.2. EXAMPLE 201

From Equation (F.9),

 2π
κmax = 1 012 − 1 m−1 (F.15)
10−6

Take

κmax d = 3 5π (F.16)

as in Figure F.2. Then, from Equations (F.15) and (F.16),

3 5 × 10−6
d= √ = 1 2343 × 10−5 m. (F.17)
2
2 1 01 − 1

Substituting Equation (F.16) into Equation (F.11),

γ 3 5π2
= −1 (F.18)
κ κd2

Equation (F.18) with Equations (F.7) and (F.8) yield the following
solutions for the eigenvalues, illustrated in Figure F.21 :

κ0 = 2 1501 × 105 m−1  κ1 = 4 2785 × 105 m−1

(F.19)
κ2 = 6 3496 × 105 m−1  κ3 = 8 2581 × 105 m−1

Substituting these values into Equations (F.5), (F.6), and (F.12), the
mode propagation constants and coupling coefficients to be substi-
tuted into Equations (3.35) and (3.36) are
 
6 3424 0 0 0
 
 6
0 6 3316 0 0  −1

= j × 10  m (F.20)

 0 0 6 3142 0 
0 0 0 6 2921

1
We number the eigenvalues 0 1 2 3 · · ·  to correspond with the notation used
throughout the text.
202 SYMMETRIC SLAB WAVEGUIDE—LOSSLESS TE MODES
 
0 2 3225 0 4 4966
 
 2 3225
9
0 6 8737 0  −2
C = 10  m (F.21)

 0 6 8737 0 13 3083 
4 4966 0 13 3083 0

REFERENCE

1. Dietrich Marcuse, Theory of Dielectric Optical Waveguides, 2nd ed., Aca-

demic Press, New York, 1991.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

APPENDIX G

Equal Propagation Constants

We treat completely degenerate guides, in which all modes have

identical propagation constants, by elementary methods. These re-
sults corroborate the general results obtained by Kronecker product
methods in Chapters 3, 4, 6, and 7. The present degenerate case is
without physical significance.
Consider the two-mode case with zero loss and constant propaga-
tion parameters:

0 =
1 = jβ (G.1)

Assume white coupling, Equations (3.1)–(3.3), with δ-function

covariance:

Rc ζ = cz + ζcz = S0 δζ

∞ (G.2)
Sν = Rc ζe−j2πνζ dζ = S0 
−∞

Assume unit coherent single-mode input:

I0 0 = 1 I1 0 = 0 (G.3)

From Equations (2.7) and (2.8),

I0 z = e−jβz cos Cz 

(G.4)
I1 z = e−jβz j sin Cz 
203
204 EQUAL PROPAGATION CONSTANTS

Cz is the integrated coupling coefficient:

z
Cz = cxdx (G.5)
0

For white coupling, Cz is Gaussian with zero mean

Cz = 0 (G.6)

and variance

z
z
C 2 z = cxcydx dy
0 0

z
z
= S0 dy δx − ydx = S0 z (G.7)
0 0

The following results are obtained by using the characteristic func-

tion of a Gaussian random variable to evaluate the trigonometric
expected values:
1
I 0 z = I0 z = e−jβz cos Cz  = e−jβz e− 2 S0 z 
(G.8)
I 1 z = I1 z = e−jβz sin Cz  = 0

1 
P 0 z = P0 z =  I0 z 2  = cos2 Cz  =1 + e−2S0 z 
2
(G.9)
2 2 1 −2S0 z

P 1 z = P1 z =  I1 z  = sin Cz  = 1 − e 
2

Q0 z = P02 z =  I0 z 4  = cos4 Cz 

3 1 −8S0 z 1 −2S0 z
= + e + e 
8 8 2
(G.10)
Q4 z = P12 z =  I0 z 4  = sin4 Cz 
3 1 −8S0 z 1 −2S0 z
= + e − e 
8 8 2
Equations (G.8) and (G.9) agree with Equations (4.23) and (4.26).
Note that the former were obtained here for equal propagation con-
stants; the latter are more general, valid for arbitrary propagation
constants. Equation (G.10) agrees with Equation (4.50).
 EQUAL PROPAGATION CONSTANTS 205

Next, assume two-mode coherent input, replacing Equation (G.3)

by

I0 0 = 1 I1 0 = 1 (G.11)

Equations (2.7) and (2.8) yield

I0 z = I1 z = e−jβz ejCz  (G.12)

where Cz is again given by Equation (G.5). Then

1
I 0 z = I 1 z = e−jβz ejCz  = e−jβz e− 2 S0 z  (G.13)

Equation (G.13) agrees with Equation (4.31). Here

P0 z = I0 z 2 = 1 P1 z = I1 z 2 = 1 (G.14)

the mode powers (but not the complex mode amplitudes) are deter-
ministic in the degenerate case, and consequently have zero variance,
in agreement with the results of Section 4.3.2 for the β = 0 case.
For two-mode incoherent input,

I0 0 = ejθ0  I1 0 = ejθ1  (G.15)

with θ0 and θ1 independent random variables, uniformly distributed

from 0 to 2π. Equations (2.7) and (2.8) become
 
I0 z = e−jβz ejθ0 cos Cz + jejθ1 sin Cz  (G.16)

with a corresponding result for I1 , obtained by interchanging the

subscripts 0 and 1. Then,

Q0 z = Q4 z = P02 z =  I0 z 4 

=  1 + sin 2Cz sinθ0 − θ1 2 
5 1 −8S0 z
= − e  (G.17)
4 4

Equations (G.17) and (4.40) substituted into Equation (4.48) confirm

the result of Equation (4.54).
206 EQUAL PROPAGATION CONSTANTS

The N-mode degenerate case may be treated in a similar di-

rect manner. For constant propagation parameter Equation (2.51)
becomes

z =
I  (G.18)

where
is the common propagation constant of all N modes,

= α + jβ (G.19)

with I the unit matrix of Equation (2.52). Then from Equation (2.53)

Iz = e−
z ejCzC · I0 (G.20)

C is the coupling coefficient matrix of Equation (2.45) and Cz is

the integrated coupling coefficient of Equation (G.5). Then using
the notation of Equations (3.38), (3.57), and (3.70):

 z = Iz = e−
z ejCzC  ·  0 (G.21)

 
z = Iz ⊗ I ∗ z = e−2αz ejCzC ⊗ e−jCzC · 0 (G.22)

 4 z = Iz ⊗ I ∗ z ⊗ Iz ⊗ I ∗ z

 
= e−4αz ejCzC ⊗ e−jCzC ⊗ ejCzC ⊗ e−jCzC ·  4 0
(G.23)

The general results of Chapter 3 for arbitrary propagation constants

specialize to the degenerate case by substituting Equation (G.18)
into Equations (3.41) and (3.61), to yield the following solutions:
1 2
 z = e−
z e− 2 S0 zC ·  0 (G.24)
1 2 2
z = e−2αz e− 2 S0 zI ⊗C +C ⊗I −2·C⊗C · 0 (G.25)

We omit the corresponding result obtained from Equation (3.72), for

brevity. Equations (G.24) and (G.25) must be identical to Equations
(G.21) and (G.22), respectively; therefore,
 jCzC  1 2
e = e− 2 S0 zC  (G.26)
 
ejCzC ⊗ e−jCzC = e− 2 S0 zI ⊗C +C ⊗I −2C⊗C 
1 2 2
(G.27)
 EQUAL PROPAGATION CONSTANTS 207

Equations (G.26) and (G.27) may be verified by series expansion of

the matrix exponentials and use of the Gaussian property of Cz
to obtain its moments. We perform this calculation for Equation
(G.26), where the second moment of Cz is given by Equation (G.7)
for white cz:
∞
 jCzC  jCzC n ∞
−1n C 2n z 2n
e = = C
n=0
n! n=0
2n!
n
∞
− 21 S0 zC2 1 2
= = e− 2 S0 zC  (G.28)
n=0
n!

Similar treatments verify Equation (G.27), and show the exponent

of the expected value in Equation (G.23) to be identical to the ex-
pression in [ ] on the right-hand side of Equation (3.72) with the
substitution of Equation (G.18).
Finally, the above results for white coupling may be extended to
completely degenerate guides with general coupling spectra Sν and
covariance Rc ζ as follows:

∞
Rc ζ = cz + ζcz Sν = Rc ζe−j2πνζ dζ (G.29)
−∞

Assume that Sν is low pass, or has a low-pass component; i.e.,

S0 = 0 (G.30)
Let νc represent the bandwidth of the coupling and λc its correlation
distance:
1
λc =  (G.31)
νc
Most of the power of coupling cz is contained in the band ν <
νc ; cz and cz + ζ are essentially uncorrelated for ζ  λc . The
integrated coupling coefficient of Equation (G.5) has zero mean,
Equation (G.6). For its variance, Equation (G.7) is replaced by

z
z
C 2 z = Rc x − ydx dy (G.32)
0 0

Calculation of this quantity requires complete knowledge of the

covariance; if the coupling cz is Gaussian, so is the integrated
coupling Cz, and its variance determines all of its higher mo-
208 EQUAL PROPAGATION CONSTANTS

ments, as in the prior results for white coupling. For simplicity, the
following treatment is restricted to the large z case. Then, Equation
(G.32) yields

C 2 z ≈ S0z z  λc  (G.33)

In this case, Cz will be approximately Gaussian, whatever the

statistics of cz. The prior results of the present appendix, for com-
pletely degenerate guides with white coupling, apply to general cou-
pling spectra by replacing the white spectral density S0 of Equation
(G.2) by the low-frequency spectral density S0 of Equation (G.29),
if z is sufficiently large.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

APPENDIX H

Asymptotic Form of Coupled

Power Equations

Consider M1 of Equations (4.104)–(4.105):

M1 = V ⊗ V ∗
(H.1)
∗ ∗
V = e
z · C · e−
z  V ∗ = e
z · C · e−
z
(H.2)

Recalling the notation of Equation (2.45) or (3.35), we have the

following results for the matrix elements:

Vii = Ci−1i−1 = 0 



Vij = e  i−1j−1 z
Ci−1j−1 = eαi−1j−1 +jβi−1j−1 z
Ci−1j−1 


V ∗ ii = Ci−1i−1 = 0 



 ∗i−1j−1 z


∗ αi−1j−1 −jβi−1j−1 z
V ij = e Ci−1j−1 = e Ci−1j−1

1 ≤ i j ≤ N
(H.3)

Now the submatrix M1 ij defined by Equation (4.108) is given by

M1 ij = Vij V ∗ = e i−1j−1 z
Ci−1j−1 V ∗  1 ≤ i j ≤ N
(H.4)

We find the asymptotic results by setting to zero each element of

Equation (H.4) that contains an exponential factor. Denote the ele-
ments of M1 ij by M1 ij k ; then assuming no degenerate modes, only
209
210 ASYMPTOTIC FORM OF COUPLED POWER EQUATIONS

the following elements survive:

M1 ij ij = e2αi−1j−1 z Ci−1j−1

2
 1 ≤ i j ≤ N (H.5)

this expression is zero for i = j. The submatrices M1 ii on the main

diagonal of M1 are zero; the submatrices M1 ij , i = j off of the main
diagonal of M1 have only a single nonzero element, in the ith row
and jth column of M1 ij .
The asymptotic form of M2 and M3 of Equations (4.104)–(4.105)
is found directly; only the diagonal terms survive. By using Equation
(4.82), the diagonal submatrices have only the following nonzero
elements in the asymptotic case:


N−1
M2 ii kk = C2 kk = 2
Ck−1  1 ≤ i k ≤ N
(H.6)
=0


N−1
M3 ii kk = C2 ii = 2
Ci−1  1 ≤ i k ≤ N
(H.7)
=0

The  = k − 1 and  = i − 1 terms in the respective summations of

these two relations are zero.
The results of Equations (H.5)–(H.7) are substituted into Equa-
tion (4.109).
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

APPENDIX I

Differential Equations
Corresponding to
Matrix Equations

I.1. SCALAR CASE

Consider the difference equation

y
kz = My
k − 1z (I.1)

M and z are fixed parameters. The solution to Equation (I.1) is

ykz = M k y0 (I.2)

We seek an equivalent differential equation:

dy
= Dy (I.3)
dz
The solution to Equation (I.3), evaluated at the points z = kz, is

ykz = eDkz y0 (I.4)

Equations (I.2) and (I.4) will be identical if

ln M
D=  (I.5)
z
211
212 DIFFERENTIAL EQUATIONS CORRESPONDING TO MATRIX EQUATIONS

In the perturbation case,

M ≈ 1 + Ez E z 1 (I.6)

Here Equation (I.5) becomes

D ≈ E (I.7)

I.2. MATRIX CASE

Consider the vector–matrix difference equation

y
kz = M · y
k − 1z (I.8)

where y is a column vector, M a fixed matrix, and z a fixed param-

eter. The solution to Equation (I.8) is

ykz = Mk · y0 (I.9)

The equivalent vector–matrix differential equation is

d
yz = D · yz (I.10)
dz
Its solution is

ykz = eDkz · y0 (I.11)

The two solutions agree if

eDz = M (I.12)

Diagonalizing
 
λM1 0 ··· 0
 
 0 λM2 ··· 0 
M = LM ·
M · −1
LM
M = 
    

 (I.13)
     
0 0 ··· λMN
 I.2. MATRIX CASE 213
 
λD1 0 ··· 0
 
 0 λD2 ··· 0 
D = LD ·
D · LD−1
D = 
    

 (I.14)
     
0 0 ··· λDN

Then,
 
eλD1 z 0 ··· 0
 
 0 eλD2 z ··· 0  −1
Dz
e = LD · 
     ·L 
 D (I.15)
     
0 0 ··· eλDN z

Equation (I.12) requires that

ln λMi
LD = LM λDi = i = 1 · · · N (I.16)
z

Simplifying notation, diagonalize the matrix M of the difference

equation, Equation (I.8):
 
λ1 0 ··· 0
 
 0 λ2 ··· 0  −1
M=L·
    
·L 
 (I.17)
     
0 0 ··· λN

Then, the equivalent differential equation is given by Equation (I.10)

with [1]
 
ln λ1
 z 0 ··· 0 
 
 ln λ2 
 
 0 ··· 0  −1
D=L·
 
z ·L 
 (I.18)
     
     
 
 ln λN 
0 0 ···
z
214 DIFFERENTIAL EQUATIONS CORRESPONDING TO MATRIX EQUATIONS

For perturbation theory,

M ≈ I + Ez E z 1 (I.19)

where I is the N × N unit matrix and denotes the matrix norm,

as in Equation (A.25). Here Equation (I.18) becomes

D ≈ E (I.20)

REFERENCE

1. Neil A. Jackman, Limitations and Tolerances in Optical Devices, Appendix

C, Ph.D. Thesis, Stevens Institute of Technology, Castle Point, Hoboken,
NJ, 1994.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

APPENDIX J

Random Square-Wave
Coupling Statistics

J.1. INTRODUCTION

The coupling function of Equation (8.1) and Figure 8.1 may be writ-
ten as follows:
 

z 1
c
z = ck rect − −k  (J.1)
 2
where
  
z 1 1 k < z <
k + 1
rect − −k = (J.2)
 2 0 otherwise.

The fundamental rect
 function has Fourier transform expressed

in terms of the sinc function as follows:
∞  
z 1
rect − · e−j2πνz dz = e−jπν  sinc ν
−∞  2
sin πν
= e−jπν   (J.3)
πν
The sequence ck has been called a discrete function [1] or a
discrete-time function [2]. Assume that ck is stationary, with dis-
crete covariance

φc
κ = ck+κ ck  (J.4)

215
216 RANDOM SQUARE-WAVE COUPLING STATISTICS

We assume the ck have zero mean:

ck = 0 (J.5)

Then S
ν, the spectral density of the coupling c
z defined in Equa-
tions (3.2) or (6.2), is given by
 2
sin πν

∞
S
ν =  φc
κe−j2πνκ  (J.6)
πν κ=−∞

Two special cases are of interest.

For independent ck ,

φc
0 = ck2 ≡ c 2 φc
κ = 0 κ = 0 (J.7)

Equation (J.6) becomes

 2
sin πν
2
S
ν = c   (J.8)
πν

Finally, consider a sequence ck that is a weighted average of a

stationary, independent, zero-mean, binary sequence d · dk :

∞

∞
ck = dak  ak = dn h
k − n = h
ndk−n  (J.9)
n=−∞ n=−∞

where h
 is the weight function,

dk = ±1 dk = 0 (J.10)

and the covariance of dk is given as follows:

φd
0 = dk2 = 1 φd
κ = dk+κ dk = 0 κ = 0 (J.11)

The autocorrelation of the weight function is defined as

∞
ϕh
κ = h
k + κh
k (J.12)
k=−∞
 J.2. BINARY SECTIONS 217

Then ck has covariance

φc
κ = d 2 ϕh
κ (J.13)

to be substituted in Equation (J.6).

The following two weight functions generate multi-level, pseudo-
Gaussian, square-wave coupling coefficients c
z, with low-pass and
band-pass coupling spectra S
ν, respectively:

1 k ≤ K
hLP
k = (J.14)
0 k > K

 −1 −K ≤ k ≤ −1
hBP
k = 0 k = 0 (J.15)

1 1 ≤ k ≤ K

c
z has 2
K + 1 or 2K + 1 equally spaced levels, for weight func-

tions hLP
k and hBP
k, respectively. The corresponding autocorre-
lation functions, for substitution in Equation (J.13), are

2K + 1 − κ κ ≤ 2K + 1
ϕLP
κ = (J.16)
0 κ ≥ 2K + 1


 2K κ = 0

 2K − 3κ + 1 1 ≤ κ ≤ K
ϕBP
κ = (J.17)

 −2K + κ − 1 K + 1 ≤ κ ≤ 2K + 1

 0 2K + 1 ≤ κ

J.2. BINARY SECTIONS

Let the ck of Equations (8.1) and (J.1) take on the values ai ,
1 ≤ i ≤ L, with unconditional probabilities

pi
k = pi = Pck = ai  (J.18)

independent of k for stationary ck . The joint probabilities are

Pck = ai  cn = aj  k = n (J.19)
218 RANDOM SQUARE-WAVE COUPLING STATISTICS

Denote the transition probabilities for adjacent values of k by

Pck−1 = aj  ck = ai
pij = Pck = ai ck−1 = aj =  (J.20)
Pck−1 = aj

again independent of k. We assume the symmetric binary case, L = 2,

in the following two subsections:
1
a1 = a a2 = −a p1 = p2 =  (J.21)
2

J.2.1. Independent
For independent ck ,

Pck = ai  cn = aj = pi pj pij = pi  (J.22)

The covariance, Equation (J.7), becomes

φc
0 = a2 φc
κ = 0 κ = 0 (J.23)

The coupling spectrum, Equation (J.8), is

 2
sin πν
2
S
ν = a   (J.24)
πν

J.2.2. Markov
Let p be the probability that adjacent ck are different:

p = Pck−1 = ck  1 − p = Pck−1 = ck  (J.25)

The transition matrix for adjacent ck is given by P of Equation

(D.36); the covariance of Equation (J.4) is obtained from Equa-
tion (D.38) as

φc
κ = a2
1 − 2pκ  (J.26)

By substituting Equation (J.26) into Equation (J.6), the coupling

spectrum is
 2
2 sin πν 1 −
1 − 2p2
S
ν = a   (J.27)
πν 1 +
1 − 2p2 − 2
1 − 2p cos 2πν
 J.3. MULTI-LEVEL MARKOV SECTIONS 219

For p = 21 , these results specialize to those of the preceding section

for the independent case.

J.3. MULTI-LEVEL MARKOV SECTIONS

J.3.1. Low-Pass—Six Levels

Consider a six-level, low-pass ck described by Equations (J.9)–(J.14).
Setting K = 2 in Equation (J.16):

κ 0 1 2 3 4 ≥5
 (J.28)
ϕLP
κ 5 4 3 2 1 0

Substituting in Equation (J.13), the coupling spectrum Equation

(J.6) becomes

 2
2 sin πν

4
S
ν = d  5+2
5 − κ cos
2πνκ  (J.29)
πν κ=1

We omit the closed form for the summation in this equation, subse-
quently using MAPLE to evaluate it directly.
The ck for this model are correlated, pseudo-Gaussian random
variables, with the following probability distribution:

ak −5 −3 −1 1 3 5
ck = dak 1 5 10 10 5 1  (J.30)
Pak
32 32 32 32 32 32

We need to describe this example as a Markov process to make

use of the results of Section D.2. From Equations (J.9) and (J.14),

ck = dak  ak = dk−2 + dk−1 + dk + dk+1 + dk+2  (J.31)

We denote the state of the system at time k by the nota-

tion STATE
k, characterized by the five-dimensional vector
220 RANDOM SQUARE-WAVE COUPLING STATISTICS

dk−2 dk−1 dk dk+1 dk+2 . The state can take on the (vector) val-
ues Si  1 ≤ i ≤ 32 given in the following table:
i Si aiLP
1 −1 −1 −1 −1 −1 −5
2 −1 −1 −1 −1 1 −3
3 −1 −1 −1 1 −1 −3
4 −1 −1 −1 1 1 −1
5 −1 −1 1 −1 −1 −3
6 −1 −1 1 −1 1 −1
7 −1 −1 1 1 −1 −1
8 −1 −1 1 1 1 1
9 −1 1 −1 −1 −1 −3
10 −1 1 −1 −1 1 −1
11 −1 1 −1 1 −1 −1
12 −1 1 −1 1 1 1
13 −1 1 1 −1 −1 −1
14 −1 1 1 −1 1 1
15 −1 1 1 1 −1 1
16 −1 1 1 1 1 3 (J.32)
17 1 −1 −1 −1 −1 −3
18 1 −1 −1 −1 1 −1
19 1 −1 −1 1 −1 −1
20 1 −1 −1 1 1 1
21 1 −1 1 −1 −1 −1
22 1 −1 1 −1 1 1
23 1 −1 1 1 −1 1
24 1 −1 1 1 1 3
25 1 1 −1 −1 −1 −1
26 1 1 −1 −1 1 1
27 1 1 −1 1 −1 1
28 1 1 −1 1 1 3
29 1 1 1 −1 −1 1
30 1 1 1 −1 1 3
31 1 1 1 1 −1 3
32 1 1 1 1 1 5

The left column i indexes the states; the middle column lists the state
vectors Si ; the final column aiLP , calculated from Si and Equation
(J.31), determines the coupling constant ck of the kth section as
ck = daiLP  STATE
k = Si  (J.33)
for the ith state at time k.
 J.3. MULTI-LEVEL MARKOV SECTIONS 221

We require the matrix quantities pD , aD , and P in Equation (D.35);

these three matrices have dimension 32 × 32 in the present example.
The 32 states have equal probability, yielding the diagonal matrix
 

1
0
1  1 
pD = ·
  
 (J.34)
32   
0 1

The diagonal matrix aD of Equation (D.35) is found from the right-

hand column of Equations (J.32) and (J.33), expressed as four 16 ×
16 submatrices as follows:
 

aD = d 
a 0
D11
 (J.35)
0 a D22

0 represents the zero submatrix, and the diagonal submatrices are

 


−5
−3
0 
 
 
 −3 
 
 −1 
 
 
 −3 
 
 
 −1 
 
 −1 
 
 
a
 1 
 
D11 = 



 −3 
 
 −1 
 
 −1 
 
 
 1 
 
 
 −1 
 
 
 1 
 
 1 
 
0 3

(J.36)
222 RANDOM SQUARE-WAVE COUPLING STATISTICS
 


−3
−1
0 
 
 
 −1 
 
 1 
 
 
 −1 
 
 
 1 
 
 1 
 
 
a
 3 
 
D22 = 



 −1 
 
 1 
 
 1 
 
 
 3 
 
 
 1 
 
 
 3 
 
 3 
 
0 5

(J.37)

Finally, the elements pij of the matrix P of Equation (D.26) are

 
pij = P STATE
K = Si STATE
K − 1 = Sj  (J.38)

These conditional probabilities follow directly from the middle

columns of Equation (J.32). For example, suppose that

dk−3 dk−2 dk−1 dk dk+1

STATE
k − 1 = S9 = −1 1 −1 −1 −1 (J.39)

Then since dk+2 can take on only the values ±1, STATE
K can have
only the two possible values

dk−2 dk−1 dk dk+1 dk+2

STATE
k = S17 =  1 −1 −1 −1 −1 (J.40)

and
dk−2 dk−1 dk dk+1 dk+2

STATE
k = S18 =  1 −1 −1 −1 1  (J.41)

 J.3. MULTI-LEVEL MARKOV SECTIONS 223

Therefore only two elements of the ninth column of P are nonzero:

1
p179 = p189 = pi9 = 0 i = 17 or 18 (J.42)
2
P is given as two submatrices as follows:
 
P= P P
 (J.43)
where
 
1
 1

0 
 1 
 
 
 1 
 
 1 
 
 1 
 
 1 
 
 
 1 
 
 1 
 
 1 
 
 
 1 
 
 1 
 
 1 
 
 1 
 
 
 1 
1 
P= 

2

1
1




 
 1 
 
 1 
 
 1 
 
 1 
 
 
 1 
 
 1 
 
 1 
 
 1 
 
 
 1 
 
 1 
 
 1 
 
 
 1 
 
 1 
 
 1 
0 1
(J.44)
224 RANDOM SQUARE-WAVE COUPLING STATISTICS

The coupling covariance is given by Equations (J.13), (J.16), and

(J.28); the same result has alternatively been obtained from the
Markov model, i.e., Equations (D.35), (J.34)–(J.37), and (J.44), us-
ing MAPLE for the calculations.

J.3.2. Band-Pass—Five Levels

Set K = 2 in Equation (J.17) to yield for a five-level, band-pass ck

κ 0 1 2 3 4 ≥5
 (J.45)
ϕBP
κ 4 2 −1 −2 −1 0

From Equations (J.13) and (J.6),

 2
sin πν
2
S
ν = d  4 + 22 cos
2πν − cos
4πν
πν
− 2 cos
6πν − cos
8πν  (J.46)

Equations (J.9) and (J.15) yield

ck = dak  ak = −dk−2 − dk−1 + dk+1 + dk+2  (J.47)

Equation (J.32) remains valid with the exception of the last column,
relabeled aiBP , calculated from the Si of Equations (J.32) and (J.47):

i 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
aiBP 0 2 2 4 0 2 2 4 −2 0 0 2 −2 0 0 2

i 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
aiBP −2 0 0 2 −2 0 0 2 −4 −2 −2 0 −4 −2 −2 0

(J.48)

The matrix pD remains the same as Equation (J.34). The diagonal

elements of aD are now given by aiBP of Equation (J.48); Equations
 J.3. MULTI-LEVEL MARKOV SECTIONS 225

(J.36) and (J.37), for the diagonal submatrices of Equation (J.35),

now become
 


0
2
0 
 
 2 
 
 4 
 
 
 0 
 
 2 
 
 
 2 
 
a
 4 
 
D11 = 



−2
 
 0 
 
 
 0 
 
 2 
 
 
 −2 
 
 0 
 
 
 0 
0 2
(J.49)
 


−2
0
0 
 
 0 
 
 2 
 
 
 −2 
 
 0 
 
 
 0 
 
a
 2 
 
D22 = 



−4
 
 −2 
 
 
 −2 
 
 0 
 
 
 −4 
 
 −2 
 
 
 −2 
0 0
(J.50)
Finally, Equations (J.43) and (J.44) remain valid for P.
226 RANDOM SQUARE-WAVE COUPLING STATISTICS

The coupling covariance computed from the Markov model—

Equations (D.35), (J.34), (J.35), (J.49), and (J.50)—agrees with that
obtained directly from Equations (J.13), (J.17), and (J.45).

REFERENCES

1. H. E. Rowe, Signals and Noise in Communication Systems, D. Van

Nostrand, New York, 1965.
2. Alan V. Oppenheim and Ronald W. Schafer, Discrete-Time Signal Pro-
cessing, Prentice Hall, Englewood Cliffs, NJ, 1989.
 Electromagnetic Propagation in Multi-Mode Random Media. Harrison E. Rowe
Copyright © 1999 John Wiley & Sons, Inc.
Print ISBN 0-471-11003-5; Electronic ISBN 0-471-20070-0

APPENDIX K

Matrix for a Multi-Layer

Structure

The electric and magnetic fields for the linearly polarized wave in
Figure 9.1 consist of forward and backward waves:
E
z = E +
z + E −
z H
z = H +
z + H −
z (K.1)
where
E +
z E −
z
H +
z =  H −
z = −  (K.2)
Z Z
and Z is the characteristic impedance of the dielectric medium:


µ0
Z=  (K.3)
ε
The forward and backward waves have the following z dependence:
E +
z = E + e−jβz  E −
z = E − e+jβz  (K.4)
where the propagation constant is
2πn
β=  (K.5)
λ
λ is the free-space wavelength,
1
λ= √  (K.6)
f µ0 ε0
227
228 MATRIX FOR A MULTI-LAYER STRUCTURE

and n is the index of refraction of the dielectric medium,

ε
n=  (K.7)
ε0
The total electric field and its forward and backward components
are related by the following matrix equations:
     + 
E
z 1 1 E
z
= ·  (K.8)
Z0 H
z n −n E −
z
 
  1  
+
E
z 1 E
z
1 n 
=  ·  (K.9)
E −
z 2 1  Z0 H
z
1 −
n
Here Z0 is the characteristic impedance of free space:


µ0
Z0 =  (K.10)
ε0
Apply these relations to the kth layer in Figure 9.1. The forward
and backward electric fields at zk and zk−1 are related as follows:
 +   jβ d   + 
E
zk−1  e k k 0 E
zk 
= ·  (K.11)
E −
zk−1  0 e−jβk dk E −
zk 

where from Equation (K.5)

2πnk
βk =  (K.12)
λ
From Equations (K.8), (K.9), and (K.11),
   
E
zk−1  E
zk 
= Mk ·  (K.13)
Z0 H
zk−1  Z0 H
zk 
where
 
    1
1 1 1 ejβk dk 0 1 n 
Mk = · ·

 (K.14)
2 n −n 0 e−jβk dk 1 
1 −
n
 MATRIX FOR A MULTI-LAYER STRUCTURE 229

Evaluating this matrix product,

 
j sin φk
 cos φk nk 
Mk =   (K.15)
jnk sin φk cos φk

where φk is the phase shift associated with the kth layer,

2π
φk = βk dk = n d  (K.16)
λ k k

The electric and magnetic fields are continuous across the bound-
aries at 0 z1  · · ·  z  Therefore,
   
E
z0 −  E
z +
= Mk ·  (K.17)
Z0 H
z0 − k=1 Z0 H
z +

By using Equation (K.9), the input and output quantities in Fig-

ure 9.1 are related as follows:
 
  1    
1
Ei 1 ni 

 1 1 Et
=  · Mk · ·  (K.18)
Er 2 1  k=1 nt −nt 0
1 −
ni