THEORY OF WAVEGUIDES AND

TRANSMISSION LINES
(Course Notes for ECEN 5114)

by
E. F. Kuester
Electromagnetics Laboratory
Department of Electrical and Computer Engineering
Campus Box 425
University of Colorado
Boulder, Colorado 80309

and
D. C. Chang
Chancellor and Past President
Polytechnic University
Brooklyn, New York 11201

c 2010 by
E. F. Kuester and D. C. Chang
ii
Contents

1 MODES OF A CLASSICAL TRANSMISSION LINE 1

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Classical Transmission Line (Distributed Constant Model) . . . . . . . . . . . . . . . . . . 3
1.2.1 Example: The conventional transmission line . . . . . . . . . . . . . . . . . . . . . 10
1.2.2 Example: An unconventional transmission line . . . . . . . . . . . . . . . . . . . . 10
1.3 Reflected Waves, the Smith Chart and Chain Parameters . . . . . . . . . . . . . . . . . . 12
1.3.1 Example: Reflection and transmission at a section of mismatched line . . . . . . . 15
1.4 Transmission Lines as Models for Plane Wave Propagation . . . . . . . . . . . . . . . . . . 17
1.4.1 Example: Reflection and transmission at an interface . . . . . . . . . . . . . . . . . 19
1.5 Nonuniform Transmission Lines and the WKB Approximation . . . . . . . . . . . . . . . . 20
1.5.1 Lowest-order WKB approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.5.2 First-order (Rayleigh-Bremmer) WKB approximation . . . . . . . . . . . . . . . . 24
1.5.3 Turning points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.6 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2 MULTIPORT NETWORK THEORY: MATRIX DESCRIPTIONS 37

2.1 Sourceless Linear Two-Port Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.1.1 Impedance, admittance and chain matrices; image parameters . . . . . . . . . . . . 38
2.1.2 The scattering matrix and related representations . . . . . . . . . . . . . . . . . . 43
2.1.3 Example: Equivalent network for a length of classical transmission line . . . . . . . 46
2.1.4 Example: Use of chain matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.1.5 Example: Realization of a 1:2 ideal transformer . . . . . . . . . . . . . . . . . . . . 49
2.1.6 Example: A two-terminal antenna as a two-port . . . . . . . . . . . . . . . . . . . 49
2.2 Multiport Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.2.1 Example: Analysis of baluns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.2.2 Example: Ring and hybrid networks . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.3 Power and energy relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.4 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3 CLASSICAL TRANSMISSION LINES: EXCITATION, COUPLING AND RES-

ONATORS 65
3.1 Induced and Impressed Currents and Voltages; Equivalent Circuits . . . . . . . . . . . . . 65
3.1.1 Thévenin and Norton equivalent circuits . . . . . . . . . . . . . . . . . . . . . . . . 66
3.1.2 Huygens (wave) sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.2 Excitation of a Classical Transmission Line . . . . . . . . . . . . . . . . . . . . . . . . . . 70
3.2.1 Lumped sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.3 Multiconductor Transmission Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

iii
 3.3.1 Matrix telegrapher’s equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.3.2 Modes of multiconductor lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.3.3 Total voltages and currents on a multiconductor line . . . . . . . . . . . . . . . . . 79
3.3.4 Reflection on a multiconductor line . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.3.5 Excitation of a multiconductor line . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.3.6 Example: The symmetric three-conductor line . . . . . . . . . . . . . . . . . . . . 81
3.4 Resonators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.4.1 Internal excitation of a resonator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.4.2 External excitation of a resonator . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.4.3 Example: Short-circuited section of lossy transmission line . . . . . . . . . . . . . . 89
3.5 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.6 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

4 PULSE PROPAGATION AND DISTORTION 101

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.2 The Fourier Transform and the Analytic Signal . . . . . . . . . . . . . . . . . . . . . . . . 101
4.2.1 Energy moments and pulse widths of signals . . . . . . . . . . . . . . . . . . . . . 103
4.2.2 Band-limited signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
4.3 Narrowband Pulse Distortion by a Linear Time-Invariant Network . . . . . . . . . . . . . 106
4.3.1 Quasistationary approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.3.2 Example: AM Gaussian pulse (quasistationary approximation) . . . . . . . . . . . 108
4.3.3 Quasi-Gaussian approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.3.4 Example: AM Gaussian pulse (quasi-Gaussian approximation) . . . . . . . . . . . 110
4.3.5 Change of pulse center and duration by a lossless network . . . . . . . . . . . . . . 111
4.4 Propagation and Distortion of Pulses; Group Velocity . . . . . . . . . . . . . . . . . . . . 112
4.4.1 Propagation of arbitrary pulses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.5 Multimode Broadening of Narrowband Pulses . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.6 Wavefront Distortion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.7 Moment Matching and AWE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
4.8 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

5 HOLLOW METALLIC WAVEGUIDES 127

5.1 The Parallel-Plate Waveguide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.2 Waveguide Modes in General . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.2.1 Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.2.2 Reduced Maxwell equations for modes . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.2.3 Forward and reverse modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
5.2.4 Normalized mode fields, power and stored energy . . . . . . . . . . . . . . . . . . . 134
5.2.5 Example: The parallel-plate waveguide . . . . . . . . . . . . . . . . . . . . . . . . . 136
5.3 “Hollow” Metallic Waveguides of Arbitrary Cross-Section . . . . . . . . . . . . . . . . . . 137
5.4 General Properties of TE and TM Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.4.1 TM modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.4.2 TE modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
5.4.3 The eigenvalues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
5.4.4 Power flow and stored energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
5.5 Circular Waveguides (Method of Separation of Variables) . . . . . . . . . . . . . . . . . . 142
5.5.1 Normalized mode fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
5.5.2 Example: A variable attenuator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
5.6 Rectangular Waveguides (Ray Description of the Modes) . . . . . . . . . . . . . . . . . . . 149
5.6.1 Normalized mode fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152

iv
 5.7 Dielectric Losses in Hollow Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.8 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156

6 SURFACE WAVE MODES: BASIC OPTICAL WAVEGUIDES 161

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
6.2 Dielectric Slab Waveguides—The Ray Description . . . . . . . . . . . . . . . . . . . . . . 162
6.3 Dielectric Slab Waveguides—Field Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.4 Dielectric Slab Waveguides—The Eigenvalue Equation for TE Modes . . . . . . . . . . . . 172
6.4.1 Near-cutoff and far-from-cutoff approximations . . . . . . . . . . . . . . . . . . . . 177
6.5 TM Modes of the Dielectric Slab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
6.6 The Grounded Dielectric Slab . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
6.7 Circular Optical Fibers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
6.7.1 The weakly guiding fiber . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
6.8 Step-Index Fibers of Other Cross-Sections . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
6.9 The Mode Spectrum of Lossless Open Waveguides . . . . . . . . . . . . . . . . . . . . . . 192
6.9.1 The radiation modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
6.9.2 Location of mode propagation coefficients . . . . . . . . . . . . . . . . . . . . . . . 194
6.10 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
6.11 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197

7 TRANSVERSE RESONANCE IN GUIDED WAVE STRUCTURES 203

7.1 Multilayer Slab Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
7.1.1 TE modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
7.1.2 TM modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
7.2 Graded-Index Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
7.2.1 Example: A diffused slab waveguide . . . . . . . . . . . . . . . . . . . . . . . . . . 208
7.3 Loaded Rectangular Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
7.3.1 Ridged, grooved and finned rectangular metallic waveguides . . . . . . . . . . . . . 213
7.4 Quasi-Optical Microstrip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
7.5 Dielectric Rib Waveguide: The Effective Index Method . . . . . . . . . . . . . . . . . . . . 221
7.6 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
7.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227

8 TEM AND QUASI-TEM MODES: BASIC PLANAR TRANSMISSION LINES 231

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
8.2 TEM Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
8.2.1 Power, energy and normalization of TEM modes . . . . . . . . . . . . . . . . . . . 240
8.2.2 Dielectric losses of TEM modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
8.3 Some TEM Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
8.3.1 The coaxial transmission line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
8.3.2 Two-wire transmission line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
8.3.3 Open stripline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
8.4 Distributed-Circuit Model of Non-TEM Modes . . . . . . . . . . . . . . . . . . . . . . . . 245
8.5 Quasi-TEM Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
8.5.1 “Approximately” TEM modes and the quasi-TEM limit . . . . . . . . . . . . . . . 250
8.5.2 Quasi-TEM modes on lines having air and one other dielectric . . . . . . . . . . . 252
8.6 Example: Open Microstrip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
8.7 More Examples: Coplanar Lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
8.7.1 Open coplanar waveguide (CPW) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
8.7.2 Coplanar strip line (CPS) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259

v
 8.8 Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
8.8.1 Dispersion in microstrip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
8.8.2 Dispersion in coplanar transmission lines . . . . . . . . . . . . . . . . . . . . . . . 262
8.9 Pseudo-TEM Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
8.9.1 The Goubau line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
8.9.2 The slot line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
8.10 Qualitative Analysis of Open Planar Waveguides . . . . . . . . . . . . . . . . . . . . . . . 269
8.11 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
8.12 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274

9 ORTHOGONALITY, POWER FLOW AND WAVEGUIDE LOSSES 281

9.1 Reciprocity and Reaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
9.1.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284
9.2 Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
9.2.1 Orthogonality of modes on lossless waveguides . . . . . . . . . . . . . . . . . . . . 287
9.3 Power Flow in Lossless Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
9.4 Dielectric Losses in Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
9.4.1 Example: Dielectric loss for quasi-TEM modes . . . . . . . . . . . . . . . . . . . . 291
9.4.2 Example: Dielectric loss in step-index, weakly-guiding waveguides . . . . . . . . . 291
9.5 Wall Losses in Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
9.5.1 Example: The TE01 mode of a hollow circular waveguide . . . . . . . . . . . . . . 295
9.5.2 The incremental inductance rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
9.5.3 Example: Conductor loss in coaxial line . . . . . . . . . . . . . . . . . . . . . . . . 299
9.6 Conclusion: Loss Effects on Orthogonality and Power Flow . . . . . . . . . . . . . . . . . 300
9.7 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
9.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303

10 EXCITATION OF WAVEGUIDES 309

10.1 Impressed Sources versus Induced Currents for Electromagnetic Fields . . . . . . . . . . . 309
10.2 Equivalence Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
10.2.1 Example: Impressed and induced sources at a perfectly conducting surface . . . . 312
10.2.2 Example: Thévenin and Norton types of equivalence theorems for electromagnetic
fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
10.2.3 Example: Tangential magnetic and normal electric surface sources . . . . . . . . . 315
10.2.4 Example: Large apertures and the Kirchhoff approximation . . . . . . . . . . . . . 315
10.2.5 Example: Equivalent dipoles for electrically small apertures . . . . . . . . . . . . . 315
10.3 Equivalent Transmission Lines—Normalization and Characteristic Impedance . . . . . . . 322
10.4 Expansion of the Field as a Sum of Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
10.4.1 Representation of the longitudinal field . . . . . . . . . . . . . . . . . . . . . . . . 327
10.4.2 Optimizing excitation of a given mode . . . . . . . . . . . . . . . . . . . . . . . . . 327
10.5 Excitation by Given Arbitrary Aperture Fields . . . . . . . . . . . . . . . . . . . . . . . . 328
10.5.1 Example: Plane wave excitation of a dielectric slab . . . . . . . . . . . . . . . . . . 331
10.5.2 Prism coupler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
10.5.3 Example: Excitation of microstrip by a slot in the ground plane . . . . . . . . . . 332
10.6 Excitation by Electric and Magnetic Dipoles and Electrically Small Apertures . . . . . . . 333
10.6.1 Example: Dipole excitation of a coaxial cable . . . . . . . . . . . . . . . . . . . . . 334
10.6.2 Example: Diaphragm with a small aperture in a rectangular waveguide . . . . . . 335
10.7 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
10.8 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340

vi
11 NETWORK THEORY FOR GUIDED WAVES 347
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
11.2 The Mode Matching Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
11.2.1 Generalized Scattering Matrix (GSM) . . . . . . . . . . . . . . . . . . . . . . . . . 355
11.2.2 Example: Parallel-plate step in height . . . . . . . . . . . . . . . . . . . . . . . . . 356
11.3 One-Port Equivalent Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
11.4 Examples of One-Port Terminations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
11.4.1 The open-ended two-wire line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360
11.4.2 A microstrip patch antenna . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 362
11.4.3 Truncated dielectric slab waveguide . . . . . . . . . . . . . . . . . . . . . . . . . . 362
11.5 Two-Port and Multiport Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364
11.6 Examples of Two-Port Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
11.6.1 Transverse diaphragm with small aperture in a rectangular waveguide . . . . . . . 365
11.6.2 Capacitive diaphragm; Partitioning of equivalent networks . . . . . . . . . . . . . . 366
11.6.3 Right-angle bend in microstrip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
11.7 Examples of Multiport Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
11.7.1 Use of physical symmetry–The E-plane and H-plane T–junctions in a rectangular
waveguide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
11.7.2 90◦ Microstrip-slotline crossover . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
11.7.3 A microstrip coupler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
11.7.4 Intersecting dielectric waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
11.8 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
11.9 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379

12 COUPLED-MODE THEORY 383

12.1 Field Analysis of Coupled Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
12.2 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383

13 RESONANT ELEMENTS FOR WAVEGUIDES 385

13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
13.2 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
13.3 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385

A FOURIER AND LAPLACE TRANSFORMS 389

A.1 Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
A.2 Table of Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
A.3 Table of Laplace Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
A.4 Table of Rules for the Laplace Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
A.5 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392

B SPECIAL FUNCTIONS 393

B.1 Bessel Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
B.1.1 Modified Bessel functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
B.1.2 Recurrence relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
B.1.3 Integral formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 398
B.2 The Gamma Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
B.3 The Elliptic Integrals K and E . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400
B.4 A Function Arising in Microstrip Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 402
B.5 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402

vii
C NUMERICAL SOLUTION OF IMPLICIT EQUATIONS 405
C.1 The Bisection Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
C.2 Newton’s Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
C.3 Iterative Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
C.4 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407

D FORMULAS FROM VECTOR, MATRIX AND TENSOR ANALYSIS 409

D.1 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 409
D.1.1 Vector differentiation in various coordinate systems . . . . . . . . . . . . . . . . . . 410
D.2 Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
D.3 Dyadics or Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 412
D.4 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413

E PROPERTIES OF SOLUTIONS TO NETWORK AND TRANSMISSION LINE

EQUATIONS 415
E.1 A Standard Form for Network Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
E.2 Reciprocity for Lumped Circuits and Classical Transmission Lines . . . . . . . . . . . . . 418
E.2.1 Lumped networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
E.2.2 Transmission lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
E.3 Power Flow in Networks and Classical Transmission Lines . . . . . . . . . . . . . . . . . . 419
E.3.1 Lumped networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
E.3.2 Transmission lines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
E.3.3 Decibels (dB) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
E.4 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423

F PROPERTIES OF SOLUTIONS TO THE ELECTROMAGNETIC FIELD EQUA-

TIONS 425
F.1 Maxwell’s Equations and Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . 425
F.2 Plane Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
F.3 Poynting’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
F.4 Elementary Dipoles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 428
F.5 Image Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430
F.6 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431

G ELECTROMAGNETIC MATERIAL PROPERTIES 433

G.1 Permittivity and Conductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434
G.2 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434

H BESSEL LINES AND TURNING POINTS 437

H.1 The Bessel Line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
H.2 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439

I MODES IN HOLLOW WAVEGUIDES WITH ZERO CUTOFF FREQUENCY 441

J FIELD LINE PLOTTING 443

K INTEGRAL IDENTITIES FOR FIELDS OF GUIDED MODES 445

K.1 Identities for General, Lossy Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
K.2 Identities for Modes in Lossless Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . 448
K.3 The Partial Power Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450
K.4 Expressions for Filling Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450
K.5 Bounds on Phase and Group Velocities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452

viii
 K.6 Notes and References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 454

L DERIVATION OF SURFACE IMPEDANCE FOR A GOOD CONDUCTOR 457

M CHANGE IN INDUCTANCE DUE TO DEFORMATION OF BOUNDARY 459

N CORRECTION OF SMALL HOLE THEORY FOR ENERGY CONSERVATION 463

ix
• Proposed Remaining Sections of Theory of Waveguides and Transmission Lines
– Chapter 11: Coupled Mode Theory
∗ Coupled Mode Equations
∗ Quasi-TEM coupled lines
∗ Coupled Dielectric Waveguides
– Chapter 12: Resonant Elements for Waveguides
∗ Cavity Resonators
∗ Open Resonators
– Appendix L: Evaluation of Coupling Coefficients

x
Preface

No matter what you say, my kids think I’m brilliant.

—Spike Milligan

The first half of this book deals with various types of electromagnetic waveguides and transmission
lines—particularly with their mode properties. Chapters 1-4 deal with so-called “classical transmission
lines”, which are distributed networks intermediate between ordinary lumped-element circuits and full
electromagnetic field descriptions of waveguide modes. The basic properties of classical transmission lines
are the same as those of any waveguide, so we will study them, their connections to lumped-element
networks and the distortion of pulses by them at some length before going on to the full field analysis
of waveguide modes. Chapters 5 through 8 examine the modes of various specific kinds of waveguide,
including traditional hollow waveguides, dielectric (including optical) waveguides, printed transmission
lines such as microstrip, and more. As the treatment progresses, we hope that the features common to
all varieties of waveguide will begin to be apparent. With this background, the stage is set for the last
part of the book, in which the problems of excitation and scattering of waveguide modes are addressed.
These tools permit consideration of how waveguides can act as interconnecting portions of real systems,
and how they perform various functions in these systems.
The reader is assumed to have had some initial exposure to electromagnetic fields (typically one year
at the junior level in a US undergraduate curriculum) as well as a background in electric circuit analysis
and a basic introduction to transmission line analysis. These topics are reviewed briefly in this book as
needed for our purposes, but the reader is urged to refer to one of several texts on electromagnetic fields
recommended in the bibliography at the end of Chapter 1, and to one of the texts in circuit analysis
cited at the end of Appendix E for suitable foundation material in these subjects.
The authors are grateful to the hundreds of students in the graduate course on waveguides and
transmission lines at the University of Colorado at Boulder who have endured since 1980 the evolution
of the course notes which have become this book. Their questions and feedback have helped the quality
of presentation immeasurably. The book has also benefited from comments by Profs. Carl Johnk, Zoya
Popović, John Dunn and Branko Popović. The first author would also like to express his gratitude
to the Electromagnetics Laboratory at the Technical University of Delft, The Netherlands, and to the
Laboratoire d’Éléctromagnetisme et d’Acoustique at the École Polytechnique Fédérale de Lausanne,
Switzerland, for their hospitality during his temporary visits with them. Some of the material in this
book was developed in the generous atmospheres provided by these groups.

xi
xii
Bibliography

The universe (which others call the Library) is

composed of an indefinite and perhaps infinite
number of hexagonal galleries, with vast air shafts
between, surrounded by very low railings. From any
of the hexagons one can see, interminably, the upper
and lower floors. The distribution of the galleries is
invariable. Twenty shelves, five long shelves per side,
cover all the sides except two; their height, which is
the distance from floor to ceiling, scarcely exceeds
that of a normal bookcase. One of the free sides leads
to a narrow hallway which opens onto another
gallery, identical to the first and to all the rest.
. . . Also through here passes a spiral stairway, which
sinks abysmally and soars upwards to remote
distances. In the hallway there is a mirror which
faithfully duplicates all appearances. Men usually
infer from this mirror that the Library is not infinite
(if it really were, why this illusory duplication?); I
prefer to dream that its polished surfaces represent
and promise the infinite . . .
—Jorge Luis Borges,
The Library of Babel

This listing contains books to which the reader can refer for a deeper insight or alternative treatment
of the topics covered in this text. References to journal articles are made only at the end of chapters,
where full citations are given.
M. Abramowitz and I. A. Stegun (eds.), Handbook of Mathematical Functions. Washington,
D.C.: U.S. Government Printing Office, 1964.
S. F. Adam, Microwave Theory and Applications. Englewood Cliffs, NJ: Prentice-Hall, 1969.
M. J. Adams, An Introduction to Optical Waveguides. New York: Wiley, 1981.
R. B. Adler, L. J. Chu and R. M. Fano, Electromagnetic Energy Transmission and Radiation.
New York: Wiley, 1960.
G. P. Agrawal, Nonlinear Fiber Optics. San Diego: Academic Press, 2001.
J. L. Altman, Microwave Circuits. Princeton, NJ: Van Nostrand, 1964.
J. A. Arnaud, Beam and Fiber Optics. New York: Academic Press, 1976.
H. A. Atwater, Introduction to Microwave Theory. Malabar, FL: Krieger, 1981.
A. J. Baden Fuller, Ferrites at Microwave Frequencies. London: Peter Peregrinus, 1989.
E. J. Baghdady (ed.), Lectures on Communication System Theory. New York: McGraw-Hill,
1961.
N. Balabanian and T. A. Bickart, Electrical Network Theory. New York: Wiley, 1969.
C. A. Balanis, Antenna Theory: Analysis and Design. New York: Harper and Row, 1982.

xiii
J. A. Ball, Algorithms for RPN Calculators. New York: Wiley, 1978.
P. Bamberg and S. Sternberg, A Course in Mathematics for Students of Physics, vol. 2.
Cambridge, UK: Cambridge University Press, 1990.
L. V. Bewley, Traveling Waves on Transmission Systems. New York: Dover, 1963.
B. Bhat and S. K. Koul, Analysis, Design and Applications of Fin Lines. Norwood, MA:
Artech House, 1987.
J. van Bladel, Electromagnetic Fields. New York: McGraw-Hill, 1964.
M. Born and E. Wolf, Principles of Optics. Oxford: Pergamon Press, 1975.
L. M. Brekhovskikh, Waves in Layered Media. New York: Academic Press, 1960.
H. Bremmer, Terrestrial Radio Waves. Amsterdam: Elsevier, 1949.
L. Brillouin, Wave Propagation and Group Velocity. New York: Academic Press, 1960.
W. M. Brown, Analysis of Linear Time-Invariant Systems. New York: McGraw-Hill, 1963.
K. G. Budden, The Wave-Guide Mode Theory of Wave Propagation. London: Logos Press,
1961.
P. F. Byrd and M. D. Friedman, Handbook of Elliptic Integrals for Engineers and Scientists.
Berlin: Springer-Verlag, 1971.
H. J. Carlin and A. B. Giordano, Network Theory. Englewood Cliffs, NJ: Prentice-Hall, 1964.
H. S. Carslaw and J. C. Jaeger, Operational Methods in Applied Mathematics. New York:
Dover, 1963.
E. Chiprout and M. S. Nakhla, Asymptotic Waveform Evaluation and Moment Matching for
Interconnect Analysis. Boston: Kluwer, 1994.
R. E. Collin, Field Theory of Guided Waves, second edition. New York: IEEE Press, 1991.
R. E. Collin, Foundations for Microwave Engineering, second edition. New York: IEEE
Press, 2001.
G. Conciauro, M. Guglielmi and R. Sorrentino, Advanced Modal Analysis. Chichester, UK:
Wiley, 2000.
R. Courant and D. Hilbert, Methods of Mathematical Physics, vol. 1. New York: Interscience,
1953.
B. Davies, Integral Transforms and Their Applications. New York: Springer, 1978.
P. Delogne, Leaky Feeders and Subsurface Radio Communication. Stevenage, UK: Peter
Peregrinus, 1982.
C. A. Desoer and E. S. Kuh, Basic Circuit Theory. New York: McGraw-Hill, 1969.
L. N. Dworsky, Modern Transmission Line Theory and Applications. New York: Wiley, 1979.
T. C. Edwards, Foundations for Microstrip Circuit Design. New York: Wiley, 1981.
G. V. Eleftheriades and K. G. Balmain, Negative-Refraction Metamaterials. Hoboken, NJ:
Wiley, 2005.
R. S. Elliott, An Introduction to Guided Waves and Microwave Circuits. Englewood Cliffs,
NJ: Prentice-Hall, 1993.
N. Engheta and R. W. Ziolkowski, Metamaterials: Physics and Engineering Explorations.
Hoboken, NJ: Wiley, 2006.
A. Erdélyi (ed.), Tables of Integral Transforms, vol. I. New York: McGraw-Hill, 1954.
N. Faché, F. Olyslager and D. De Zutter, Electromagnetic and Circuit Modelling of Multi-
conductor Transmission Lines. Oxford: Clarendon Press, 1993.
L. B. Felsen and N. Marcuvitz, Radiation and Scattering of Waves. Englewood Cliffs, NJ:
Prentice-Hall, 1973.
N. H. Fletcher and T. D. Rossing, The Physics of Musical Instruments. New York: Springer-
Verlag, 1991.
S. Frankel, Multiconductor Transmission Line Analysis. Dedham, MA: Artech House, 1977.
L. Franks, Signal Theory. Englewood Cliffs, NJ: Prentice-Hall, 1969.
P. R. Garabedian, Partial Differential Equations. New York: Wiley, 1964.
F. E. Gardiol, Lossy Transmission Lines. Norwood, MA: Artech House, 1987.

xiv
R. N. Ghose, Microwave Circuit Theory and Analysis. New York: McGraw-Hill, 1963.
D. Gloge (ed.), Optical Fiber Technology. New York: IEEE Press, 1976.
B. Gross and E. P. Braga, Singularities of Linear System Functions. Amsterdam: Elsevier,
1961.
E. A. Guillemin, Synthesis of Passive Networks. New York: Wiley, 1957.
M. A. R. Gunston, Microwave Transmission-Line Impedance Data. London: Van Nostrand
Reinhold, 1972.
K. C. Gupta, R. Garg and I. J. Bahl, Microstrip Lines and Slotlines. Dedham, MA: Artech
House, 1979.
K. C. Gupta, R. Garg and R. Chadha, Computer-Aided Design of Microwave Circuits. Ded-
ham, MA: Artech House, 1981.
K. C. Gupta and A. Singh, Microwave Integrated Circuits. New York: Wiley, 1974.
R. W. Hamming, Numerical Methods for Scientists and Engineers. New York: Dover, 1986.
R. F. Harrington, Time-Harmonic Electromagnetic Fields. New York: McGraw-Hill, 1961.
A. F. Harvey, Microwave Engineering. London: Academic Press, 1963.
H.A. Haus, Waves and Fields in Optoelectronics. Englewood Cliffs, NJ: Prentice-Hall, 1984.
H. A. Haus and R. B. Adler, Circuit Theory of Noisy Networks. New York: Wiley, 1959.
J. Helszajn, Nonreciprocal Microwave Junctions and Circulators. New York: Wiley, 1975.
J. Helszajn, Microwave Planar Passive Circuits and Filters. Chichester, UK: Wiley, 1994.
P. Henrici, Computational Analysis with the HP-25 Pocket Calculator. New York: Wiley,
1977.
W. Hilberg, Electrical Characteristics of Transmission Lines. Dedham, MA: Artech House,
1979.
R. K. Hoffmann, Handbook of Microwave Integrated Circuits. Norwood, MA: Artech House,
1987.
Huang Hung-Chia, Coupled-Mode Theory. Utrecht, The Netherlands: VNU Science Press,
1984.
A. S. Ilyinsky, G. Ya. Slepyan and A. Ya. Slepyan, Propagation, Scattering and Dissipation
of Electromagnetic Waves. Stevenage, UK: Peter Peregrinus, 1993.
ITT Staff, Reference Data for Radio Engineers. Indianapolis: Howard W. Sams, 1975.
J. D. Jackson, Classical Electrodynamics. New York: Wiley, 1975, second edition.
M. Javid and P. M. Brown, Field Analysis and Electromagnetics. New York: McGraw-Hill,
1963.
C. T. A. Johnk, Engineering Electromagnetic Fields and Waves. New York: Wiley, 1988.
C. C. Johnson, Field and Wave Electrodynamics. New York: McGraw-Hill, 1965.
D. S. Jones, The Theory of Electromagnetism. Oxford: Pergamon Press, 1964.
C. Kao (ed.), Optical Fiber Technology II. New York: IEEE Press, 1981.
N. S. Kapany and J.J. Burke, Optical Waveguides. New York: Academic Press, 1974.
J. J. Karakash, Transmission Lines and Filter Networks. New York: Macmillan, 1950.
B. Z. Katsenelenbaum, L. Mercader del Rio, M. Pereyaslavets, M. Sorolla Ayza and M.
Thumm, Theory of Nonuniform Waveguides. London: The Institution of Electrical En-
gineers, 1998.
D. M. Kerns and R. W. Beatty, Basic Theory of Waveguide Junctions and Introductory
Microwave Network Analysis. Oxford: Pergamon Press, 1967.
R. W. P. King, Transmission-Line Theory. New York: Dover, 1965.
R. W. P. King, H. R. Mimno and A. H. Wing, Transmission Lines, Antennas and Wave
Guides. New York: Dover, 1965.
J. Klapper, Selected Papers on Frequency Modulation. New York: Dover, 1970.
L. Koenigsberger, Hermann von Helmholtz. New York: Dover, 1965.
M. Koshiba, Optical Waveguide Analysis. New York: McGraw-Hill, 1992.
K. Kurokawa, An Introduction to the Theory of Microwave Circuits. New York: Academic

xv
 Press, 1969.
P. I. Kuznetsov and R. L. Stratonovich, Waves in Multiconductor Transmission Lines. New
York: Macmillan, 1964.
A. Lakhtakia, V. K. Varadan and V. V. Varadan, Time-Harmonic Electromagnetic Fields in
Chiral Media. Berlin: Springer-Verlag, 1989.
I. Lebedev, Microwave Engineering. Moscow: Mir, 1973.
N. N. Lebedev, Special Functions and Their Applications. New York: Dover, 1972.
D. L. Lee, Electromagnetic Principles of Integrated Optics. New York: Wiley, 1986.
L. Lewin, Theory of Waveguides. London: Newnes-Butterworths, 1975.
L. Lewin, D. C. Chang and E. F. Kuester Electromagnetic Waves and Curved Structures.
Stevenage, UK: Peter Peregrinus, 1977.
D. K. Linkhart, Microwave Circulator Design. Norwood, MA: Artech House, 1989.
J. T. Londergan, J. P. Carini and D. P. Murdock, Binding and Scattering in Two-Dimensional
Systems. Berlin: Springer-Verlag, 1999.
W. H. Louisell, Coupled Mode and Parametric Electronics. New York: Wiley, 1960.
S. A. Maas, The RF and Microwave Circuit Design Cookbook. Boston: Artech House, 1998.
L. M. Magid, Electromagnetic Fields, Energy, and Waves. New York: Wiley, 1972.
D. Marcuse (ed.), Integrated Optics. IEEE Press, 1973.
D. Marcuse, Theory of Dielectric Optical Waveguides. New York: Academic Press, 1974.
D. Marcuse, Light Transmission Optics. Princeton, NJ: Van Nostrand Reinhold, 1982.
N. Marcuvitz (ed.), Waveguide Handbook. New York: Dover, 1965.
J. Martin, Telecommunications and the Computer. Englewood Cliffs, NJ: Prentice-Hall, 1969.
J. Martin, Future Developments in Telecommunications. Englewood Cliffs, NJ: Prentice-Hall,
1971.
B. Mashkovtsev, K. N. Tsibizov and B. F. Emelin, Teoriya Volnovodov. Moscow: Nauka,
1966.
R. E. Matick, Transmission Lines for Digital and Communication Networks. New York:
McGraw-Hill, 1969.
A. Matsumoto (ed.), Microwave Filters and Circuits. New York: Academic Press, 1970.
G. L. Matthaei, L. Young and E. M. T. Jones (eds.), Microwave Filters, Impedance-Matching
Networks, and Coupling Structures. New York: McGraw-Hill, 1964.
A. R. Mickelson, Guided Wave Optics. New York: Van Nostrand Reinhold, 1993.
K. A. Milton and J. Schwinger, Electromagnetic Radiation: Variational Methods, Waveguides
and Accelerators. Berlin: Springer-Verlag, 2006.
R. Mittra and S. W. Lee, Analytical Techniques in the Theory of Guided Waves. New York:
Macmillan, 1971.
C. G. Montgomery, R. H. Dicke and E. M. Purcell (eds.), Principles of Microwave Circuits.
New York: Dover, 1965.
T. Moreno, Microwave Transmission Design Data. New York: Dover, 1958.
P. M. Morse and K. U. Ingard, Theoretical Acoustics. Princeton, NJ: Princeton University
Press, 1968.
M. Mrozowski, Guided Electromagnetic Waves: Properties and Analysis. Taunton, UK:
Research Studies Press, 1997.
E. I. Nefedov and A. T. Fialkovskii, Poloskovye Linii Peredachi. Moscow: Nauka, 1980.
R. W. Newcomb, Linear Multiport Synthesis. New York: McGraw-Hill, 1966.
F. Oberhettinger and L. Badii, Tables of Laplace Transforms. New York: Springer-Verlag,
1973.
F. Olyslager, Electromagnetic Waveguides and Transmission Lines. Oxford: Clarendon
Press, 1999.
K. E. Oughstun and G. C. Sherman, Electromagnetic Pulse Propagation in Causal Dielectrics.
Berlin: Springer-Verlag, 1994.

xvi
C. R. Paul, Introduction to Electromagnetic Compatibility. New York: Wiley, 1992.
C. R. Paul, Analysis of Multiconductor Lines. New York: Wiley, 1994.
L. T. Pillage, R. A. Rohrer and C. Visweswariah, Electronic Circuit and System Simulation
Methods. New York: McGraw-Hill, 1995.
G. Pólya and G. Szegö, Isoperimetric Inequalities in Mathematical Physics. Princeton, NJ:
Princeton University Press, 1951.
A. M. Portis, Electromagnetic Fields: Sources and Media. New York: Wiley, 1978.
D. M. Pozar, Microwave Engineering. Hoboken, NJ: Wiley, 2005.
W. H. Press, S. A. Teukolsky, W. T. Vettering and B. P. Flannery, Numerical Recipes in C.
Cambridge, UK: Cambridge University Press, 1992.
S. Ramo, J. R. Whinnery and T. van Duzer, Fields and Waves in Communication Electronics.
New York: Wiley, 1994.
H. J. Reich, P. F. Ordnung, H. L. Krauss and J. G. Skalnik, Microwave Theory and Tech-
niques. Princeton, NJ: Van Nostrand, 1953.
M. Remoissenet, Waves Called Solitons. Berlin: Springer-Verlag, 1996.
P. A. Rizzi, Microwave Engineering: Passive Circuits. Englewood Cliffs, NJ: Prentice-Hall,
1988.
H. E. Rowe, Signals and Noise in Communication Systems. Princeton, NJ: Van Nostrand,
1965.
T. Rozzi and M. Farina, Advanced Electromagnetic Analysis of Passive and Active Planar
Structures. London: Institution of Electrical Engineers, 1999.
T. Rozzi and M. Mongiardo, Open Electromagnetic Waveguides. London: Institution of
Electrical Engineers, 1997.
R. Rüdenberg, Electrical Shock Waves in Power Systems. Cambridge, MA: Harvard Univer-
sity Press, 1968.
T. S. Saad (ed.), Microwave Engineer’s Handbook. Dedham, MA: Artech House, 1971.
D. M. Sazonov, A.N. Gridin and B.A. Mishustin, Microwave Circuits. Moscow: Mir, 1982.
S. A. Schelkunoff, Electromagnetic Waves. New York: Van Nostrand, 1943.
J. Schwinger and D.S. Saxon, Discontinuities in Waveguides. New York: Gordon and Breach,
1968.
A. Scott, Active and Nonlinear Wave Propagation in Electronics. New York: Wiley, 1970.
S. Seshu and N. Balabanian, Linear Network Analysis. New York: Wiley, 1959.
J. Sevick, Transmission Line Transformers. Newington, CT: American Radio Relay League,
1990.
V. P. Shestopalov, Physical Foundations of the Millimeter and Submillimeter Waves Tech-
nique. Vol. 1: Open Structures. Utrecht, The Netherlands: VSP, 1997.
V. V. Shevchenko, Continuous Transitions in Open Waveguides. Boulder, CO: Golem Press,
1971.
A. S. Silbergleit and Yu. I. Kopilevich, Spectral Theory of Guided Waves. Bristol, UK:
Institute of Physics Publishing, 1996.
J. C. Slater, Microwave Transmission. New York: McGraw-Hill, 1942.
J. C. Slater, Microwave Electronics. New York: Van Nostrand, 1950.
A. A. Smith, Coupling of External Electromagnetic Fields to Transmission Lines. New York:
Wiley, 1977.
A. W. Snyder and J.D. Love, Optical Waveguide Theory. London: Chapman and Hall, 1983.
M. S. Sodha and A.K. Ghatak, Inhomogeneous Optical Waveguides. New York: Plenum
Press, 1977.
S. Solimeno, B. Crosignani and P. DiPorto, Guiding, Diffraction and Confinement of Optical
Radiation. Orlando, FL: Academic Press, 1986.
F. Sporleder and H.-G. Unger, Waveguide Tapers, Transitions and Couplers. Stevenage,
England: Peter Peregrinus, 1979.

xvii
G. Strang, Introduction to Applied Mathematics. Wellesley, MA: Wellesley-Cambridge Press,
1986.
F. G. Stremler, Introduction to Communication Systems. Reading, MA: Addison-Wesley,
1990.
J. B. Thomas, An Introduction to Statistical Communication Theory. New York: Wiley,
1969.
E. C. Titchmarsh, Introduction to the Theory of Fourier Integrals. New York: Chelsea, 1986.
H. C. Tuckwell, Introduction to Theoretical Neurobiology. Volume 1: Linear Cable Theory
and Dendritic Structure. New York: Cambridge University Press, 1988.
H. Uchida, Fundamentals of Coupled Lines and Multiwire Antennas. Sendai, Japan: Sasaki
Printing and Publishing Co., 1967.
J. Uher, J. Bornemann and U. Rosenberg, Waveguide Components for Antenna Feed Systems:
Theory and CAD. Boston: Artech House, 1993.
H.-G. Unger, Planar Optical Waveguides and Fibers. Oxford: Clarendon Press, 1977.
L. A. Vainshtein (Weinstein), Open Resonators and Open Waveguides. Boulder, CO: Golem
Press, 1969a.
L. A. Vainshtein (Weinstein), The Theory of Diffraction and the Factorization Method. Boul-
der, CO: Golem Press, 1969b.
L. A. Vainshtein (Weinstein), Elektromagnitnye Volny. Moscow: Radio i Svyaz’, 1988.
L. A. Vainshtein (Weinstein) and D. E. Vakman, Razdelenie Chastot v Teorii Kolebanii i
Voln. Moscow: Nauka, 1983.
D. E. Vakman, Signals, Oscillations and Waves. Boston: Artech House, 1998.
B. van der Pol and H. Bremmer, Operational Calculus. New York: Chelsea, 1987.
E. F. Vance, Coupling to Shielded Cables. New York: Wiley, 1978.
C. Vassallo, Théorie des Guides d’Ondes Électromagnétiques. Paris: Eyrolles, 1985, tomes 1
et 2.
C. Vassallo, Optical Waveguide Concepts. Amsterdam: Elsevier, 1991.
V. F. Vzyatyshev, Dielektricheskie Volnovody. Moscow: Sovetskoe Radio, 1970.
R. A. Waldron, Theory of Guided Electromagnetic Waves. London: Van Nostrand Reinhold,
1969.
D. A. Watkins, Topics in Electromagnetic Theory. New York: Wiley, 1958.
W. L. Weeks, Antenna Engineering. New York: McGraw-Hill, 1968.
L. A. Weinstein - see L. A. Vainshtein.
I. Wolff, Coplanar Microwave Integrated Circuits. Hoboken, NJ: Wiley, 2006.
T. T. Y. Wong, Fundamentals of Distributed Amplification. Boston: Artech House, 1993.
L. Young (ed.), Parallel Coupled Lines and Directional Couplers. Dedham, MA: Artech
House, 1972.
L. A. Zadeh and C. A. Desoer, Linear System Theory. New York: McGraw-Hill, 1963.

xviii
Chapter 1

MODES OF A CLASSICAL
TRANSMISSION LINE

The following story is true. There was a little boy,

and his father said, “Do try to be like other people.
Don’t frown.” And he tried and tried, but could not.
So his father beat him with a strap; and then he was
eaten up by lions.
Reader, if young, take warning by his sad life and
death. . . . if you are different, you had better hide it,
and pretend to be solemn and wooden-headed. Until
you make your fortune. . . . In particular, if you are
going to write a book, . . . be rigorous, that will cover
a multitude of sins. And do not frown.
—Oliver Heaviside,
Electromagnetic Theory

1.1 Introduction
In the broadest sense, all of electrical engineering deals with some sort of guided-wave system. Maxwell’s
equations and Poynting’s theorem indicate to us that, even in an elementary low-frequency circuit, energy
is stored and power is transferred, not by wires or circuit elements, but by the electromagnetic fields
surrounding them. The configurations themselves serve merely to “ arrange” or “ guide” the fields in
an advantageous manner. As operating frequency increases, this physical picture remains the same,
but the quasi-static approximation of circuit theory which was useful at lower frequencies breaks down.
Moreover, the electrical parameters of materials generally change at higher frequencies (conductors
become more lossy, atomic and molecular vibrations begin to influence the dielectric properties, and so
on). Circuits designed on quasi-static principles no longer operate efficiently, and new structures, along
with new methods to analyze them, must be found.
In this same broad context, of course, antennas could also be thought of as guided-wave structures,
especially so if designed with a highly directional pattern. The advantage of an antenna in a commu-
nication system is that no structure need be erected over a long distance. However, even in the best
antenna systems it is difficult to achieve a really good degree of “guidance”, and tens or hundreds of dB
of signal may be lost as a result. This is in part because no actual antenna guides a signal in a single
direction exclusively, and some sort of intermediate structure must be provided along the way if most or
all of the signal is to reach its destination. If such a structure had no material losses, it could in principle

1
 (µ,ε)
d

a
(a) (b)

(c)

Figure 1.1: (a) Rectangular metallic waveguide; (b) circular coaxial waveguide; (c) two-wire transmission
line.

guide all the energy from a source at one end to a load at the other, in sharp contrast to even a lossless
antenna system. It is such structures as this which we have in mind when we talk about waveguides,
and which will be the subject of this book.
Of course, the distinction is not a precise one. If an open waveguide (such as a two-wire transmis-
sion line) changes direction or encounters a discontinuity, energy can be radiated as with an antenna.
Therefore, for our purposes a waveguide will be defined to be any arrangement of matter in space
whose electrical properties (permittivity ǫ, conductivity σ, permeability µ) and geometrical structure
vary neither with the time t, nor with some given direction in space, say the z-direction. This axial or
longitudinal direction may be part of a cartesian (x, y, z) or cylindrical (ρ, φ, z) system of coordinates.
More general coordinate systems (u1 , u2 , z) can also be used, and are sometimes invoked to deal with
special situations. We have in mind artificial structures made for the purpose of guiding waves, but
there are naturally occurring cases of waveguides (such as the so-called earth-ionosphere waveguide) as
well. We will not explicitly deal with natural waveguides in this book.
You may already be familiar with some types of waveguides: the rectangular metallic waveguide
(Figure 1.1(a)), the circular coaxial waveguide (Figure 1.1(b)), and the two-wire transmission line (Figure
1.1(c)). Perhaps less familiar are planar guiding structures and those involving dissimilar dielectrics,
such as optical fibers, dielectric channel waveguides, microstrip transmission lines and fin-lines (Figures
1.2(a)-(d)). Each of these was developed in response to a particular need to provide wave guidance for a
certain frequency range or under other special conditions. Additionally, each is also tailored in response
to various technological constraints or requirements (such as the development of integrated microwave
circuits).
According to our definition, a waveguide is infinitely long and has a constant cross-section. In
practice, of course, only finite lengths of waveguides are actually used; we will refer to these as sections

2
 metal strip
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
ε1 εr
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
ε2 xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
dielectric
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
substrate

(a) (c)
ground plane

ε3
dielectric
ε1 substrate
ε2 metal
fins
εr
(b)
shield
(d)

Figure 1.2: (a) Optical fiber waveguide; (b) dielectric channel waveguide; (c) microstrip; (d) finline.

or segments of waveguide. Before we can treat segments of waveguide, however, we must first study
the properties of infinitely long waveguides. In this chapter, we will do this for a simple but important
structure—the so-called classical transmission line. In chapters 5-8, we will look at more elaborate types
of waveguiding structures.

1.2 Classical Transmission Line (Distributed Constant Model)

The most readily understood waveguide is perhaps the classical transmission line. This line is not really
associated with any particular physical structure, but rather serves as an approximate model for a wide
variety of waveguides met with in practice. The classical line will also be an important concept in the
exact representation of fields in any waveguide, as will be seen in Chapter 10, and serves as an analogy
for virtually any form of one-dimensional wave propagation.
The classical transmission line can be described by a distributed circuit model as shown in Figure 1.3.
Consider an incremental length of line (which for definiteness can be thought of as a pair of wires) lying
between the locations z and z + ∆z along the z-axis. We assume that the wires present a series resistance
of r (Ω/m) and a series inductance of l (H/m) per unit length along z, and a shunt capacitance c (F/m)
and shunt conductance g (S/m) per unit length as well. We will further assume that all connections of
this transmission line to external circuit elements (such as generators and loads) are such that at any
point z on the line, the current on the “positive” (upper) conductor is equal but opposite to the current
in the “reference” or “ground” (lower) conductor, as indicated in Figure 1.3. This restricts the type of
external circuit connection that can be made between different points z on the line (especially of the
“feedback” type), and forces the pair of terminals at any point of a transmission line to behave as a port
(see section 2.1).
The equivalent circuit of Figure 1.3 which results implies the relations

3
 r∆z l∆z i(z+∆z,t)
+ i(z,t) +
... v(z,t) g∆z c∆z v(z+∆z,t) . . .
- -
i(z,t) i(z+∆z,t)
z z+∆z

Figure 1.3: Distributed network model of the classical transmission line.

rA∆z lA∆z

rB∆z lB∆z g∆z c∆z

z z+∆z

Figure 1.4: Alternative distributed network model of the classical transmission line.

∂i(z, t)
0 = (r∆z)i(z, t) + (l∆z) + v(z + ∆z, t) − v(z, t) (1.1)
∂t
∂v(z + ∆z, t)
0 = (g∆z)v(z + ∆z, t) + (c∆z) + i(z + ∆z, t) − i(z, t)
∂t
by Kirchhoff’s circuit laws. Dividing these equations by ∆z and letting ∆z → 0, we obtain the equations
∂v(z, t) ∂i(z, t)
= −ri(z, t) − l
∂z ∂t
∂i(z, t) ∂v(z, t)
= −gv(z, t) − c (1.2)
∂z ∂t
which are known as the telegrapher’s equations, or transmission-line equations. They provide a relation-
ship between the voltage v and current i at any point along the line. The parameters r, l, g and c are
called the distributed parameters of the line.
It should be noted that other distributed circuit models for the transmission line will also result in
equations (1.2). If we can associate the resistance per unit length rA and inductance per unit length
lA with the upper conductor, and rB and lB with the lower conductor as shown in Figure 1.4, the
same telegrapher’s equations (1.2) result as do from the circuit of Figure 1.3, if we put r = rA + rB
and l = lA + lB . The difference is that in the model of Fig. 1.3, the “ground” conductor is at the
same voltage for all values of z, while in Fig. 1.4, this is not necessarily the case. This means that
the telegrapher’s equations by themselves are incapable of giving any information about the voltage
difference between points at different values of z on a transmission line. Either this information will

4
 z∆z

y∆z

z z+∆z

Figure 1.5: General distributed network model of the classical transmission line.

be unobtainable (and may, in fact, be irrelevant under certain conditions), or it will follow only from
information about external circuit connections.
Except for chapter 4, where pulse phenomena will be discussed, we will be concerned in this book
with time-harmonic phenomena at an angular frequency ω = 2πf . Voltages, currents, fields or sources
of this type can be obtained from complex (phasor) functions which are independent of t. For example,

v(z, t) = Re ejωt V (z)



i(z, t) = Re ejωt I(z)


(1.3)
and so on. We will adopt the convention that, except where noted, a quantity denoted by a capital letter
is the phasor counterpart of the function of√time denoted by the corresponding small letter. We follow
electrical engineering practice and use j = −1 to represent the imaginary unit.
For time-harmonic voltage and current, equations (1.2) become ordinary differential equations:

dV (z)
= −zI(z)
dz
dI(z)
= −yV (z) (1.4)
dz
where z and y represent respectively the distributed series impedance and shunt admittance per unit
length of the line:
z = r + jωl; y = g + jωc (1.5)
It is possible to model transmission lines with more complicated distributed networks using generalized
versions of z and y:
z = r(ω) + jx(ω); y = g(ω) + jb(ω) (1.6)
where x is the series reactance per unit length, and b is the shunt susceptance per unit length (Figure 1.5).
All these line parameters may depend on frequency, possibly in complicated ways. Examples of such
behavior are illustrated in example 1.2.2 and problems p1-3 and p1-10.
When z and y are independent of z (the line is then said to be uniform), these equations have a
well-known solution:

V (z) = V0+ e−γz + V0− eγz

V0+ −γz V0− γz
I(z) = e − e (1.7)
Zc Zc

5
where V0+ and V0− are complex constants which will be determined by the strengths of the sources
(generators) connected elsewhere on the line. The quantity
z γ
Zc = = = |Zc |ejφZ (Ω) (1.8)
γ y

is called the characteristic impedance of the line. We sometimes also use the characteristic admittance
Yc = 1/Zc . The quantity
√
γ = α + jβ = zy (m−1 ) (1.9)
is called the complex propagation coefficient of the line, with its real part α denoted as the attenuation
coefficient, and β the phase coefficient (or sometimes, casually, the propagation constant). The terms
proportional to V0+ in (1.7) are called the forward wave, while those proportional to V0− are called the
reverse wave, for reasons that we will clarify below.
For a transmission line (and in later chapters, for any waveguide mode) whose complex propagation
coefficient γ has an imaginary part β not equal to zero, it is customary to define several auxiliary
quantities in terms of β:

1. Guide wavelength:
2π
λg ≡ (1.10)
|β|

2. Phase velocity:
ω
vp = (1.11)
β

3. Group velocity:
1
vg = (1.12)
dβ/dω

The group velocity will be seen in chapter 4 to describe the velocity with which the envelope of a
modulated signal travels along the line. Under many conditions, it is also the velocity with which energy
travels along the line. The phase velocity on the other hand, is generally not the velocity of any physically
observable quantity.
The sign of the square roots taken when calculating γ (which in turn will affect the value of Zc ) must
be chosen carefully. All transmission lines considered in this book are assumed to be passive, meaning
that the real parts r and g of z and y are nonnegative. We then choose the square root in (1.9) such
that
Re(γ) = α ≥ 0 (1.13)
The solution for α which obeys this constraint can be obtained by squaring (1.9) and separating into
real and imaginary parts:
α2 − β 2 = Re(zy) = rg − xb
2αβ = Im(zy) = rb + gx
Eliminating β between these two equations gives a quadratic equation for α2 , whose nonnegative solution
gives α as: r
|zy| + Re(zy)
α= (1.14)
2
where p
|zy| = (r2 + x2 )(g2 + b2 )
and all square roots are taken to be nonnegative.

6
 If α as computed from (1.14) is positive, then the value of β is uniquely determined as:

rb + gx
β= (1.15)
2α
and may be either positive or negative (a wave for which β > 0 is called a direct wave, while one for
which β < 0 is called a backward wave). Under these conditions, it can be shown that the real part of
the characteristic impedance given by (1.8) is:

g|z| + r|y|
Re(Zc ) = (1.16)
2α|y|

and thus is always nonnegative. The characteristic impedance thus lies in the right half of the complex
plane: −π/2 ≤ φZ ≤ π/2. We will see below that this implies that a direct wave carries power in the
same direction as phase moves, while a backward wave carries power in the opposite direction to the
phase. The ordinary classical transmission line depicted in Fig. 1.3 supports only direct waves if l and
c are positive, but other types of transmission line can sometimes support backward waves.
If α = 0, then (1.13) does not specify the value of β uniquely. This case can only arise if two
conditions are both satisfied:

i) the line is lossless (r = g = 0), and

ii) xb > 0.

We then have β 2 = xb. In this case, we will determine the sign of β by requiring that the characteristic
impedance (which by (1.8) must be real) be positive, in agreement with the fact that Re(Zc ) ≥ 0 in the
case when α > 0. The result of this requirement is
√
β = + xb (lossless line, x > 0 and b > 0) (1.17)

or √
β = − xb (lossless line, x < 0 and b < 0) (1.18)
Again, β may be either positive or negative, and the wave is denoted as a direct or backward wave
accordingly.
Our choice for the sign of the square root determining γ is based on a desire to attach a physical
meaning to each of the terms in equations (1.7) representing waves on the transmission line. If α > 0,
the forward wave of (1.7) is one that decays exponentially in the positive z direction. Now, the complex
power flow in the z-direction at a point z on the transmission line (Fig. 1.6) is:
1
P̂ (z) = V (z)I ∗ (z) (1.19)
2
where ∗ denotes the complex conjugate. The time-average power flow Pav is the real part of this complex
power. Thus the time-average power in the forward wave (in the absence of the reverse wave) is

|V0+ |2 e−2αz
 
1
Pav (z) = Re (1.20)
2 Zc∗

and therefore if power flow is nonzero, this wave is also the one that carries energy in the positive z-
direction, because Re(Zc ) ≥ 0. If α = 0, the choice of (1.17) or (1.18) for the phase constant of the
forward wave also results in energy flow in the positive z-direction. Likewise, the reverse wave in (1.7)
transports energy (if at all) in the negative z-direction, and if it decays, does so in this direction as well.
In any event, the forward waves should move toward larger values of z from a source located at smaller
values of z, while the reverse wave moves toward smaller values of z from a source located at larger

7
 I(z)

+
V(z)
-

z
^
P(z)

Figure 1.6: Power flow on a classical transmission line.

values of z. This allows us to decide which wave should be present on a section of line which extends to
+∞ or to −∞ in the z-direction.
Let us next consider the flow of power along the line when both forward and reverse waves are
present. By (1.7) the complex power at z is

1  + 2 −2αz
− |V0− |2 e2αz + 2jIm [V0− (V0+ )∗ e2jβz ]

P̂ (z) = ∗
|V0 | e (1.21)
2Zc

When Zc is real,
|V0+ |2 e−2αz − |V0− |2 e2αz
Pav = (1.22)
2Zc
This states that the powers carried individually by the forward and reverse waves superpose to give the
total power flow with no interaction between them. On the other hand, if Zc is imaginary (φZ = ±π/2),
there can be time average power flow only if both waves are present simultaneously:
1
Pav = ∓Im [V0− (V0+ )∗ e2jβz ] (1.23)
|Zc |

and no such power superposition principle holds in this case. On a lossy line intermediate between these
two cases, some contribution from both terms will be present. The complex oscillatory power on the
other hand is simply
 1
P̂osc = V I/2 = (V0+ )2 e−2γz − (V0− )2 e2γz

(1.24)
2Zc
regardless of the values of Zc and γ, and contains no cross terms.
It is convenient to express the amplitudes V0± in another form, so as to make the power flow on the
line more explicit. Let us introduce new variables a0 and b0 , called respectively the forward and reverse
wave amplitudes at z = 0, through the equations

V0+ = NV a0 V0− = NV b0 (1.25)

where NV is a suitably chosen normalizing factor. The question now is: how do we best choose this
factor? Sometimes, an attempt is made to choose it so as to make |a0 |2 and |b0 |2 represent the time-
average power carried by the forward and reverse waves respectively. From the preceding discussion, it is
seen that this cannot be done in general: for example, if Zc is imaginary the time-average power flow may

8
be zero even if a0 6= 0. Moreover, attempting to specify the time-average power leaves the phase angle
of NV completely undetermined, and we would have to resort to arbitrary criteria in order to render it
unique. On the other hand, the physically less important complex oscillatory power is expressed in a
mathematically simpler form via the forward and reverse waves.
We will thus choose to normalize the forward and reverse modes by requiring that each carries a
complex oscillatory power of 1ej0 watt in its direction of energy transport (or attenuation) at the
position z = 0 when a0 = 1 or b0 = 1 as appropriate. We will find that in important special cases, this
normalization will still lead to a simple form for Pav . Thus, if a0 = 1 and b0 = 0, we have the normalized
forward mode. Its voltage has phase angle φZ /2 and its current has phase angle −φZ /2 at z = 0. The
normalized reverse mode (a0 = 0, b0 = 1) has the same voltage as the forward mode, but the opposite
current at z = 0. Setting P̂osc = 1 for the normalized forward mode, or P̂osc = −1 for the normalized
reverse mode gives p p
NV = 2Zc = 2|Zc |ejφZ /2 (1.26)
where φZ lies in the interval [−π/2, +π/2], as noted above.
From (1.24), then, a20 and −b20 represent respectively the complex oscillatory powers carried through
the position z = 0 in the +z-direction by the forward and reverse waves on the transmission line. With
this choice of normalization condition, equations (1.21) and (1.24) for the complex and oscillatory powers
at any z become
P̂ (z) = ejφZ |a0 |2 e−2αz − |b0 |2 e2αz + 2j Im (b0 a∗0 e2jβz )

(1.27)
and
P̂osc = a20 e−2γz − b20 e2γz (1.28)
respectively. When Zc is real, the time-average power from (1.27) reduces to

Pav (z) = |a0 |2 e−2αz − |b0 |2 e2αz (1.29)

which is analogous to (1.28).

We can see in (1.7) the basic concept of a mode. A mode in the present context is a combination of a
voltage and current (later, it will be a combination of electric and magnetic fields) which propagate along
z according to the common propagation factor exp(jωt − γz) and which maintain a constant relationship
between V and I. Thus in (1.7) we identify the forward mode (whose amplitude decreases with, or whose
energy moves towards, increasing z) as
r
+
p
−γz + 2
V (z) = 2Zc a0 e ; I (z) = a0 e−γz
Zc
and the reverse mode (whose amplitude decreases with or energy moves towards decreasing z) as
r
−
p
γz − 2
V (z) = 2Zc b0 e ; I (z) = − b0 eγz
Zc
In each case, the voltage and current associated with each mode vary with z according to the same
law, exp(∓γz), and the ratio V (z)/I(z) is a constant, ±Zc , independent of the location z. Moreover,
the overall amplitude factor (a0 or b0 ) common to V and I is arbitrary, and is only determined by the
conditions of excitation of the line.
The wave amplitude of a mode at an arbitrary position along a uniform transmission line is denoted
by a(z) for the forward mode, and by b(z) for the reverse mode, where

a(z) = a0 e−γz b(z) = b0 eγz (1.30)

It is also sometimes convenient to define normalized total voltages and currents, defined by

Ṽ (z) = a(z) + b(z) ˜ = a(z) − b(z)

I(z) (1.31)

9
 √
when we wish not to clutter our formulas with factors like 2Zc . Evidently, the actual voltage and
current will be given by r
p 2 ˜
V (z) = 2Zc Ṽ (z) I(z) = I(z) (1.32)
Zc
The wave amplitudes a(z) and b(z) are given in terms of the voltage and current by:
V (z) + Zc I(z) V (z) − Zc I(z)
a(z) = √ ; b(z) = √ (1.33)
8Zc 8Zc

1.2.1 Example: The conventional transmission line

The simplest lossless classical transmission line is one in which x = ωl and b = ωc, with l and c both
positive and independent of frequency. Then we have a propagating mode with α = 0,
√
β = ω lc (1.34)

and r
l
Zc = (1.35)
c
These formulas are familiar from introductory texts on transmission lines. From (1.11) and (1.12), we
see that
1
vp = vg = √ (1.36)
lc

1.2.2 Example: An unconventional transmission line

For a lossless line in which one (but not both) of x(z) or b(z) is negative1 ,
√
jx −xb
Zc = √ = (1.37)
−xb jb
√
where −xb is taken to be positive. Now Zc is seen to be positive imaginary if x > 0 and b < 0, and
negative imaginary if b > 0 and x < 0.
An example of this behavior is furnished by the coaxial cable which has its inner conductor interrupted
at frequent periodic intervals as shown in Fig. 1.7. The effect of this upon the distributed constant model
of this transmission line is the insertion of an additional distributed capacitance into the line, in the form
of an elastance2 s∆z in series with the series inductance l∆z for a small length ∆z of the transmission
line. Using the distributed constant model for this line, we see that the shunt admittance per unit length
remains
y = jb = jωc
while the series impedance in a length ∆z of the line is
1
z∆z = jωl∆z + 1


jω s∆z

or
s
z = jx = jωl +
jω
Observe that the series capacitance in a length ∆z of line is taken to be proportional to 1/∆z rather
than to ∆z. This is because as ∆z increases, the effect is of combining the gap capacitances in series,
1 As we will see below, this can happen when using the transmission line equations to model plane wave problems, for

sufficiently large incidence angles, or when modeling unconventional transmission lines.

2 Elastance S, sometimes called the potential coefficient and denoted p, is the inverse of capacitance C.

10
 ....

s∆z l∆z

c∆z

z z+∆z

Figure 1.7: Transmission line with periodically interrupted conductor.

which results in an overall capacitance inversely proportional to the number of gaps in ∆z, and therefore
to ∆z itself.
By the same procedure employed for conventional transmission lines, we compute the propagation
constant to be p
√ ω 2 − ωc2
γ = zy = j (1.38)
vp∞
and the characteristic impedance to be
r
ωc2
r
z
Zc = = Zc∞ 1− (1.39)
y ω2

where
1
vp∞ = √ (1.40)
lc
is the phase velocity of the line in the limit as ω → ∞ (i. e., what we would get if the gaps were not
present),
r
l
Zc∞ = (1.41)
c
is the characteristic impedance we have without gaps (or as ω → ∞), and
r
s
ωc = (1.42)
l

is a cutoff frequency.

11
 For ω < ωc , the line does not support a propagating mode, but instead one which decays exponen-
tially, because γ = α is real for ω < ωc : p
ωc2 − ω 2
α= (1.43)
vp∞
In this frequency range where the mode is cutoff, the characteristic impedance is purely imaginary; by
(1.37),
r
ωc2
Zc = −jZc∞ −1 (1.44)
ω2
According to (1.23), this means that both a forward and reverse mode must be present in order for
time-average power to flow along the line. A section of such a line functions as a high-pass filter.
Another feature of note is that, even in the propagating regime ω > ωc , the phase and group velocities
are functions of frequency, with
vp∞
vp = p (1.45)
1 − ωc2 /ω 2
and p
vg = vp∞ 1 − ωc2 /ω 2 (1.46)
We see that the phase velocity is always greater than that of the gapless line, while the group velocity
is always smaller. The phenomenon of vp depending on ω (so that vp 6= vg ) is called dispersion, and
means that transient signals launched on a dispersive line have each of their frequency components phase
shifted by a different amount due to propagation along the line. Re-assembling the frequencies at the
load end of the line results in a distorted pulse. This distortion is studied in more detail in chapter 4.

1.3 Reflected Waves, the Smith Chart and Chain Parameters

In analysis of transmission-line problems, we define the total impedance (or line impedance) at any point
z along the line as
V (z) 1 + ρ(z)
Z(z) = = Zc (1.47)
I(z) 1 − ρ(z)
where for a uniform line,
V0− 2γz
ρ(z) = e (1.48)
V0+
is the position-dependent reflection coefficient on the line, if V0+ and V0− are the constants appearing
in (1.7). The reflection coefficient can also be expressed in terms of the wave amplitude description:
ρ(z) = b(z)/a(z).
The inverse relation to (1.47) allows the reflection coefficient to be computed from the line impedance:

Z(z) − Zc
ρ(z) = (1.49)
Z(z) + Zc

The normalized line impedance, defined as

Z̄(z) = Z(z)/Zc (1.50)

is free from explicit dependence on the characteristic impedance, and can easily be related to ρ(z):

1 + ρ(z) Z̄(z) − 1
Z̄(z) = ; ρ(z) = (1.51)
1 − ρ(z) Z̄(z) + 1

12
 Another way of expressing the amount of reflected wave at a given point on a transmission line is in
terms of the (voltage) standing wave ratio [(V)SWR], defined as

1 + |ρ(z)|
VSWR = (1.52)
1 − |ρ(z)|

Since |ρ(z)| ≤ 1 for a transmission line with real characteristic impedance (if no gain or sources are
present), it is clear that the VSWR is real and 1 ≤ VSWR ≤ ∞ in this case. In the absence of
attenuation on the line (α = 0), the VSWR is the same everywhere on a uniform, sourcefree section of
line.
A common way of displaying graphically the relative amount of reverse to forward mode content on
a line is by means of the Smith chart (Figure 1.8). The Smith chart represents the unit circle of the

The Complete Smith Chart

Black Magic Design

0.12 0.13
0.11 0.14
0.38 0.37 0.15
0.1 0.39 0.36
90
0.4 100 80 0.35 0.1
9
0.0 6

45
50
1 110 40 70 0.3
0.4 _ 4

1.0
0.9

1.2
0.1
55

8
0.8
0.0 X=1 35
7

1.4
0.3
.42 0 60
0.7

0 12 3
60

Yc )

1.6
7 jB/ 0.1
0.0 E (+ 30 8
0.6

3 NC 0.3
TA

1.8
0.4 0.2 2
EP 50
SC
65

0
13 SU

2.0
E
IV
0.5

0.
60

T 25

19
CI
0.

0.
44

PA

31
0.

CA
70

R
,O 0.4
c)
0

40
14
5

0.2
4
0.
0.0

/Z
5

20

0.3
jX
0.4

(+
T

3.0
75

EN

0.6
N
PO
4

0.2
0.0

M
0

0.2
6

1
30
0.3
15
0.4

CO

9
0.8 _ 15
E

4.0
80

NC

R=1
R

TA
TO

1.0

0.22
AC
RA

0.28
0.47

5.0
RE
E

1.0
GEN

160

VE

0.2

20
85

10
CTI
A RD

8
0.
U

0.23
IND
TOW

0.27
0.48

ANG
90

ANG
0.6
GTHS

LE OF
10

LE OF
170

0.1
WAVELEN

0.4

TRANSM

0.24
0.26
0.49

REFLECTION
20
0.2

ISSION COEFFICIENT IN
50
0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1.2

1.4

1.6

1.8

2.0

3.0

4.0

5.0

10

20

50

0.25

0.25
± 180
0.0

0.0

COEFFICIENT IN
RESISTANCE COMPONENT (R/Zc), OR CONDUCTANCE COMPONENT (G/Yc) 50
0.2
D
RD LOA

20

0.24
0.49

0.26
0.4
-170

DEGR
0.1

DEGR
TOWA

10

EES
EES
0.6
-90

0.23
THS

c)

0.27
0.48

jB/Y
ENG

8
0.
E (-

-10
VEL
-160
-85

-20

0.2
C

1.0
AN
A

0.22

5.0
W

PT

0.28
0.47

1.0
CE
US
-80

4.0
ES

0.8 -15
IV

0.2
.04
0

-30

0.3
CT
-15
6

0.2
1
0
0.4

DU

9
IN

0.6
-75

3.0
,O )
5

Zc

0.2

-20
0.0

X/
5

4
0.
0.3
0.4

(- j
-4
4
-1

T 0.4
EN
-70

N
PO
0.
06

19
0.

M
CO -25
0.
44

-65 0.5

3
0.

E
1
2.0

30 NC -5
-1 TA 0
7 AC 0.1
1.8

RE 0.2 8
0.0
IVE
0.6

0.3
3 IT -30
0.4 PAC 2
1.6

CA
-60

0 -60 0.1
0.7

8 -12
0.0 _ 7
1.4

-35 0.3
0.8

0.4
2
X = -1 3
1.2
-55

0.9

1.0

9 -70 0.1
0.0 -110 0
-4
6
0
-5

0.3
-4

1
0.4 0.1 -100 -80 0.15 4
-90
0.11 0.14 0.35
0.4 0.12 0.13
0.39 0.36
0.38 0.37
O (C dB O ]
F
. C K SS [ SS C [dB

RADIALLY SCALED PARAMETERS

EF
.
N P L L EN

P)

TOWARD LOAD TOWARD GENERATOR

SW d S [d EFF , E o
RT R FL.

A W. L. W. TT

EF O ]
P T.
SM EA O O

100 40 20 10 5 4 3 2.5 2 1.8 1.6 1.4 1.2 1.1 1 15 10 7 5 4 3 2 1

F, NS
N FL CO

TR S. RF S. A
R S B] , P r I
.L .
O CO EFF

40 30 20 15 10 8 6 5 4 3 2 1 1 1 1.1 1.2 1.3 1.4 1.6 1.8 2 3 4 5 10 20

R

8
S

0 1 2 3 4 5 6 7 8 9 10 12 14 20 30 0 0.1 0.2 0.4 0.6 0.8 1 1.5 2 3 4 5 6 10 15

8

I
or
E

1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.05 0.01 0 0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.5 3 4 5 10
F,
8

EF
O

1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1 0.99 0.95 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
.C
SM

CENTER
N
A
TR

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2

ORIGIN

Figure 1.8: The Smith Chart, with R̄ = 1, X̄ = 1 and X̄ = −1 circles emphasized.

13
complex ρ-plane, |ρ| ≤ 1. What makes it distinctive, however, is that it is calibrated in the real and
imaginary parts of the normalized impedance Z̄ = R̄ + j X̄ instead of real and imaginary parts of ρ. As
the location of the observation point z along the line moves, the locus of ρ(z) moves in a circle or a spiral
on the Smith chart.
Although individual determination of the amplitude constants V0+ and V0− in (1.7) requires knowledge
of the sources on the line, and will be deferred until the next chapter, we can determine the ratio V0− /V0+
which results when a lumped impedance element is connected to the line. Consider, for example, the
situation of Figure 1.9, where a load impedance ZL is connected at z = d on a transmission line whose
characteristic impedance is Zc and whose propagation coefficient is γ. At z = d, we must have, by

Z(0) (γ, Zc) ZL

z=0 z=d

Figure 1.9: Section of transmission line terminated with a load impedance.

Kirchhoff’s circuit laws and Ohm’s law:

V (d)
ZL = = Z(d)
I(d)
or,
Z̄(d) = ZL /Zc
By (1.51), we see that
ZL /Zc − 1 ZL − Zc
ρ(d) = = (1.53)
ZL /Zc + 1 ZL + Zc
On the other hand, by (1.48),

V0−
 
ZL − Zc
ρ(0) = = e−2γd ρ(d) = e−2γd (1.54)
V0+ ZL + Zc

and therefore the input impedance presented at the terminals at z = 0 is, by (1.51),
1 + ρ(0)
Z(0) = Zc Z̄(0) = Zc
1 − ρ(0)
ZL cosh γd + Zc sinh γd
= Zc (1.55)
ZL sinh γd + Zc cosh γd
A more powerful means of analyzing sections of uniform transmission line is to use a relationship
between voltage and current at one end and those at the other end. Suppose a section of transmission
line lies in the interval z1 < z < z2 . Regardless of what is connected to either end of the line, the
relations (1.7) can be used to express V (z1 ) and I(z1 ) in terms of V (z2 ) and I(z2 ). We can in fact write

V (z1 ) = A(z1 , z2 )V (z2 ) + B(z1 , z2 )I(z2 )

I(z1 ) = C(z1 , z2 )V (z2 ) + D(z1 , z2 )I(z2 ) (1.56)

14
which are true regardless of the values of the two amplitudes V0± in (1.7) if
A(z1 , z2 ) = cosh γ(z2 − z1 )
B(z1 , z2 ) = Zc sinh γ(z2 − z1 )
1
C(z1 , z2 ) = sinh γ(z2 − z1 )
Zc
D(z1 , z2 ) = cosh γ(z2 − z1 ) (1.57)
The quantities A(z1 , z2 ), B(z1 , z2 ), C(z1 , z2 ) and D(z1 , z2 ) are examples of elements of a chain matrix,
which is discussed more generally in section 2.1.1. In particular, we note here the properties
A(z1 , z2 )D(z1 , z2 ) − B(z1 , z2 )C(z1 , z2 ) = 1 (1.58)
and
A(z1 , z2 ) = D(z2 , z1 )
B(z1 , z2 ) = −B(z2, z1 )
C(z1 , z2 ) = −C(z2 , z1 )
D(z1 , z2 ) = A(z2 , z1 ) (1.59)
which turn out to be true in general, even for cascaded sections of line, or nonuniform lines as considered
in section 1.5.
To see how (1.56) can be applied, let the load impedance ZL = Z(d) be connected at z = d as in
Fig. 1.9. Then put z1 = 0 and z2 = d in (1.56), divide the two equations and obtain:
A(0, d)V (d) + B(0, d)I(d) A(0, d)Z(d) + B(0, d)
Z(0) = = (1.60)
C(0, d)V (d) + D(0, d)I(d) C(0, d)Z(d) + D(0, d)
which we readily see reduces to (1.55). Another useful result is the expression of the voltage across the
load impedance in terms of the voltage at z = 0. From (1.56), we have
V (d) V (d) Z(d) ZL
= = = (1.61)
V (0) A(0, d)V (d) + B(0, d)I(d) A(0, d)Z(d) + B(0, d) ZL cosh γd + Zc sinh γd
Other such relations can be quickly obtained in a similar way using the chain relations (1.56) for a
uniform classical transmission line.

1.3.1 Example: Reflection and transmission at a section of mismatched line

To illustrate the foregoing ideas, consider the example of a wave incident from the left on a transmission
line of characteristic impedance Zc1 and propagation constant γ1 to a section of different transmission
line of length d, whose characteristic impedance is Zc2 and propagation constant is γ2 , as shown in
Fig. 1.10. From (1.60) or (1.55), the input impedance seen by an incident voltage wave V1+ e−γ1 z in
z < 0 is
Zc1 + Zc2 tanh γ2 d
Z(0) = Zc2 (1.62)
Zc1 tanh γ2 d + Zc2
so that the reflection coefficient at z = 0 is
2 2
(Zc2 − Zc1 ) sinh γ2 d
ρ = ρ(0) = 2 2 (1.63)
(Zc2 + Zc1 ) sinh γ2 d + 2Zc1 Zc2 cosh γ2 d
We can similarly define a (voltage) transmission coefficient τ at z = d as
V (d) V (d) 2Zc1 Zc2
τ≡ + = (1 + ρ) = 2 2 (1.64)
V1 V (0) (Zc2 + Zc1 ) sinh γ2 d + 2Zc1 Zc2 cosh γ2 d

15
 (γ1, Zc1) ρ (γ2, Zc2) τ (γ1, Zc1)

z=0 z=d

Figure 1.10: Section of transmission line inserted in a different line.

Note that while ρ is uniquely defined as the ratio of reflected to incident waves at the same position on the
same transmission line, the transmission coefficient may have many definitions, because the transmitted
wave in question may be at a different position on the line, or on a different transmission line altogether.
It is therefore defined specifically for any particular application.
Let us consider two special cases in detail. In both cases, the characteristic impedance Zc1 will be
assumed to be real. In Case A, we will also assume that Zc2 is real, while γ2 = jβ2 is imaginary. The
central transmission line is thus lossless, and supports a propagating mode. We then have
2 2
j(Zc2 − Zc1 ) sin β2 d
ρ= 2 2 (1.65)
j(Zc2 + Zc1 ) sin β2 d + 2Zc1 Zc2 cos β2 d

and
2Zc1 Zc2
τ= 2 2 (1.66)
j(Zc2 + Zc1 ) sin β2 d + 2Zc1 Zc2 cos β2 d
As d varies, |ρ| varies between 0 (when d is an integer multiple of λg2 on the central section of line) and
a maximum of
2 2
Zc2 − Zc1
2 2
Zc2 + Zc1
when d is an odd multiple of a quarter wavelength. The value of |τ | likewise varies between a maximum
of 1 and a minimum of
2 2
Zc2 − Zc1
2 2
Zc2 + Zc1
at these same values of d.
For Case B, we suppose that the central section of line is lossless but cutoff, so that Zc2 = jXc2 is
imaginary, while γ2 = α2 is real, as is the case for ω < ωc in Example 1.2.2. From (1.63) and (1.64) we
now have
2 2
(Xc2 + Zc1 ) sinh α2 d
ρ=− 2 2 (1.67)
(Zc1 − Xc2 ) sinh α2 d + 2jZc1 Xc2 cosh α2 d
and
2jZc1 Xc2
τ= 2 − X 2 ) sinh α d + 2jZ X cosh α d (1.68)
(Zc1 c2 2 c1 c2 2

In contrast with Case A, the behavior of |ρ| and |τ | is now monotonic with d. Indeed, if α2 d ≫ 1, then
|ρ| ≃ 1, while
4Zc1 Xc2 −α2 d
|τ | ≃ 2 2 e
Zc1 + Xc2
which, although small, is not exactly zero. This nonzero transmitted wave is the result of what is called
“tunneling,” analogous to the similar phenomenon in quantum mechanics.

16
1.4 Transmission Lines as Models for Plane Wave Propagation
Although we derived the telegrapher’s equations from a distributed network model for the behavior of
two parallel conductors of appreciable length, they are also capable of modeling a wide variety of other
physical processes involving wave propagation. We will see in later chapters that the propagation of any
waveguide mode can be described in this way. In this section we will illustrate the analogy between the
classical transmission line and the propagation of plane waves in layered media.
Consider the case of a time-harmonic plane wave propagating in a lossless homogeneous medium
whose permittivity is ǫ and whose permeability is µ (in z < 0). In z > 0 is a lossless one-dimensionally
layered medium in z > 0 as shown in Fig. 1.11. For now, we will think of this region as having piecewise

Ey Hy

θ θ µ µ θ θ
ε ε

z=0
µ (z) x
µ (z)
ε (z) ε (z)

z
(a) (b)

Figure 1.11: Obliquely incident plane waves onto a plane inhomogeneous medium: (a) perpendicular
(TE) polarization; (b) parallel (TM) polarization.

constant permeability µ(z) and permittivity ǫ(z), but in the end, what we do below will turn out to be
valid for arbitrary variations of the material parameters with z (even in z < 0).
The fields of the incident wave in z < 0 contain a factor ej(ωt−kx x) , where kx = k sin θ, θ is the angle
√
of incidence to the z-axis, and k = ω µǫ is the wavenumber in the region z < 0. We further assume
that the incident fields have no dependence on y. The nature of the geometry implies that the total
field anywhere will possess both these properties as well. The exponential factor e−jkx x is present in the
fields for all values of z, with the same value of kx (otherwise continuity of tangential E and H could not
hold). We use this fact to define an angle of propagation θ(z) at each point in z > 0 from the relation

kx = k sin θ = k(z) sin θ(z) = constant (1.69)

p
where the local wavenumber of the medium is defined as k(z) = ω µ(z)ǫ(z). Eqn. (1.69) is a generalized
form of Snell’s Law for a layered medium.
We thus put ∂/∂y ≡ 0 and suppress the common exponential factor above from all the fields. With
these assumptions, the Maxwell equations (F.3) in a source-free region (Jext = 0) reduce to:

∂Ey ∂Hy
− = −jωµHx ; − = jωǫEx
∂z ∂z
∂Ex ∂Hx
+ jkx Ez = −jωµHy ; + jkx Hz = jωǫEy
∂z ∂z
−jkx Ey = −jωµHz ; −jkx Hy = jωǫEz (1.70)

It will be observed that this set of six equations in six scalar unknowns is decoupled : the only nonzero
field components might be Ey , Hx and Hz ; or Hy , Ex and Ez . The first of these situations is called

17
 Trans. line Plane wave (TE) Plane Wave (TM)

V (z) Ey (z) Ex (z)

I(z) −Hx (z) Hy (z)
2
kx
x(z) ωµ(z) ωµ(z) − ωǫ(z) = ωµ(z) cos2 θ(z)
2
kx
b(z) ωǫ(z) − ωµ(z) = ωǫ(z) cos2 θ(z) ωǫ(z)
ζ(z)
Zc (z) cos θ(z) ζ(z) cos θ(z)
p p
γ(z) jω µ(z)ǫ(z) cos θ(z) jω µ(z)ǫ(z) cos θ(z)

Table 1.1: Equivalent variables for transmission line and plane wave problems.

perpendicular polarization (that is, the E-field is perpendicular to the plane of incidence determined
by the direction of propagation and the normal to the surface of the medium; this polarization is also
sometimes called TE, meaning the field is Transverse Electric to these directions, or s, from senkrecht —
the German word for perpendicular), while the second is called parallel (or TM or p) polarization. A
general plane wave independent of y would be made up of a superposition of the two polarizations.
If Hz or Ez respectively is eliminated from the set of three scalar equations that result from Maxwell’s
equations in either of these polarizations, we have
k2
 
∂Ey ∂Hx
− = −jωµHx ; = jω ǫ − 2x Ey (1.71)
∂z ∂z ω µ
kx2
 
∂Hy ∂Ex
− = jωǫEx ; = −jω µ − 2 Hy (1.72)
∂z ∂z ω ǫ
Eqns. (1.71) and (1.72) are identical with the telegrapher’s equations:
dV dI
= −jx(z)I(z); = −jb(z)V (z) (1.73)
dz dz
if the correspondences for V , I, b and x given in Table 1.1 are used. More general problems involving
losses will result in the yet more general form of the telegrapher’s equations:
dV dI
= −z(z)I(z); = −y(z)V (z) (1.74)
dz dz
where z(z) and y(z) can be fairly arbitrary functions of z. This form is more general than (1.4) in that
it is true for nonuniform transmission lines, i. e., lines whose parameters may vary along the z-direction.

The implication of this equivalence is that we may solve reflection and transmission problems for
plane waves using only the telegrapher’s equations and any methods developed to solve transmission line
problems. For example, if kx = 0 in a homogeneous region, we have that the most general solution is a
pair of plane waves traveling in the ±z-directions:
E + −jkz E − jkz
V (z) = E + e−jkz + E − ejkz ; I(z) = e − e
ζ ζ
p
analogous to the transmission-line solutions of (1.7). Here E ± are complex constants and ζ = µ/ǫ is
the wave impedance of the medium. For arbitrary values of kx in either polarization, the general solution
is a pair of obliquely traveling plane waves:
E + −jkz z E − jkz z
V (z) = E + e−jkz z + E − ejkz z ; I(z) = e − e (1.75)
Zc Zc

18
where the phase constant in the z-direction is
√ p
kz = xb = k 2 − kx2 = k cos θ (1.76)

and the equivalent characteristic impedance

x kz
Zc = =
kz b
is given in Table 1.1 for the two polarizations.
When using the transmission line equations to model plane wave problems (or when modeling trans-
mission lines with unusual distributed parameters—see problem p1-3 and example 1.2.2), x(z) or b(z)
may become negative for sufficiently large incidence angles as a result of cos θ(z) becoming imaginary.
The signs of the square roots taken when calculating γ = jkz and Zc must then be chosen carefully. The
situation is similar to that of example 1.2.2. We choose the square root for γ as discussed in connection
with (1.9), so that β is positive if γ = jβ is pure imaginary, and γ = α > 0 if it is real (in which case
cos θ becomes negative imaginary). The characteristic impedance should be calculated from (1.37). This
point is illustrated by the example in the next subsection.

1.4.1 Example: Reflection and transmission at an interface

From what has been said above, the problem of plane wave reflection at the plane interface between
two media can be reduced to that of two transmission lines connected at the point z = 0. If the line
occupying z < 0 has characteristic impedance Zc1 , while that of the line in z > 0 is Zc2 , the reflection
coefficient of a wave incident on line 1 is easily obtained by recognizing that the semi-infinite line 2, on
which no incoming wave is present, is equivalent to a lumped load impedance ZL = Zc2 connected to
line 1 at z = 0. From (1.53), the (voltage) reflection coefficient for the incident wave on line 1 is:
Zc2 − Zc1
ρ= (1.77)
Zc2 + Zc1
The transmitted wave on line 2 must have a voltage at z = 0 equal to the total voltage on line 1 at this
same point, by Kirchhoff’s law. Thus,

V2 (0) = V1 (0) = (1 + ρ)V1inc

so that the (voltage) transmission coefficient τ is

V2 (0) 2Zc2
τ= =1+ρ= (1.78)
V1inc Zc2 + Zc1
It is evident that the reflection and transmission coefficients are functions only of the relative character-
istic impedances of the two lines, and will involve their propagation coefficients only indirectly (through
the angles of incidence and transmission).
Now let us use this model to treat TE-polarized plane waves. Let µ1 and ǫ1 be the parameters of the
region z < 0 from which the incident wave comes, and µ2 and ǫ2 be the parameters of the region z > 0.
Since kx is the same in both regions to ensure continuity of the fields for all x, we have Snell’s law:

kx = k2 sin θ2 = k1 sin θ1 (1.79)

where θ1 is the angle of incidence in region 1, and θ2 is the angle of propagation of the transmitted wave
in region 2. Then from (1.77), (1.78) and Table 1.1, we obtain for ρ and τ the electric-field reflection
and transmission coefficients given by the classical Fresnel formulas:
ζ2 cos θ1 − ζ1 cos θ2
ρ = (1.80)
TE ζ2 cos θ1 + ζ1 cos θ2

19
 2ζ2 cos θ1
τ = (1.81)
TE ζ2 cos θ1 + ζ1 cos θ2
Note that because there is only one component of electric field in this polarization, the voltage reflec-
tion and transmission coefficients on the transmission-line analog problem are identical to the Fresnel
coefficients ρ and τ which are ratios of the electric field.
TE TE
In the TM polarization, V is the analog of only one component (Ex ) of the total electric field, whereas
the Fresnel coefficients ρ and τ are ratios of total E-fields. In general, therefore, ρ and τ will
TM TM TM TM
not be the same as the ρ and τ obtained from the equivalent transmission line problem, except when the
incidence angle θ1 = 0. The reader is asked to fill in the details of the determination of the TM Fresnel
coefficients in problem p1-6.
If transmission line 2 has, say, x2 > 0 but b2 < 0, then its characteristic impedance is positive
imaginary, and the reflected wave on line 1 acquires a phase shift only, with no reduction in amplitude:

ρ = ejχ (1.82)

where r r
−1 x 1 b2 x 2 b1
χ = 2 tan − = π − 2 tan−1 − (1.83)
x 2 b1 x 1 b2
In particular, when modeling TE plane wave reflection when kx = k2 sin θ2 = k1 sin θ1 > k2 , we have
from Snell’s law that sin θ2 > 1, and hence cos θ2 must be negative imaginary (θ2 is a complex angle).
Then
ρ = ejχTE (θ) (1.84)
TE
p q
where (since cos θ2 = −j sin2 θ2 − 1 = −j k12 sin2 θ1 − k22 /k2 )
 
ζ2 k2 cos θ 1
χ (θ) = π − 2 tan−1  q  (1.85)
TE ζ1 k 2 sin2 θ − k 2
1 1 2

Thus, the magnitude |ρ | of the reflection coefficient becomes equal to 1, and we have the phenomenon
TE
of total reflection, which serves as the basis surface wave guidance in dielectric waveguides.

1.5 Nonuniform Transmission Lines and the WKB Approxima-

tion
We often encounter nonuniform transmission lines, either as actual transmission lines or as models of
mathematically analogous wave propagation problems. When the variation of parameters is discontinu-
ous, we simply connect terminals of sections of different lines together, and enforce continuity of current
and voltage as boundary conditions. When the line parameters vary continuously with z, however, we
can no longer employ the simple exponential solutions for forward and reverse going waves that are
obtained for the uniform line.
In most cases, the transmission-line equations cannot be solved exactly for arbitrary continuous
variations of z and y. One can take the approach of breaking the line into a cascaded connection of
a large number of line sections whose line parameters are constant within each section. Although the
problem could now be handled by ordinary techniques of transmission-line analysis, it leads to equations
which are unwieldy to solve. If the changes in line parameters with z are gradual enough, it is possible to
obtain an explicit approximate solution to the nonuniform transmission line problem that is considerably
simpler. This approximation is known as the WKB (Wentzel-Kramers-Brillouin) approximation.

20
 x(z) b(z)

z z
0 d 0 d
Zci , γi

... ∆zi ...

z0=0 zi-1 zi zN

Figure 1.12: Piecewise constant approximation to a nonuniform transmission line.

1.5.1 Lowest-order WKB approximation

To derive this approximation to lowest order, let us consider a lossless nonuniform transmission line
lying between 0 < z < d and terminated with a load impedance ZL at the point z = d (it is not difficult
to extend the analysis to lossy lines). Let us temporarily replace this line by an approximate one having
piecewise uniform values of x and b as shown in Figure 1.12. In each uniform section (section i will
be designated as the interval zi−1 < z < zi ), the characteristic impedance and propagation coefficient
have the constant values Zci and γi respectively. If the characteristic impedance Zc of this line varies
sufficiently slowly from one section (i, say) to the next (i+1), then a +z-traveling voltage wave in section
i:
V + (z) = Vi+ e−γi z (1.86)

upon encountering the junction point zi to the next section i + 1 will undergo nearly zero reflection, and
be transmitted with a voltage transmission coefficient that is nearly 1. More precisely, by (1.77) and
(1.78), we have
r r r
p 1+ρ p Zc,i+1 Zc,i+1
τ = 1 + ρ = 1 − ρ2 = 1 − ρ2 ≃ (1.87)
1−ρ Zci Zci
where we have neglected terms of the order of the very small quantity ρ2 . Hence the forward voltage
wave at the beginning of section i + 1 will be approximately
r
Zc,i+1 + −
V +
(zi+ ) ≃ V (zi ) (1.88)
Zci

In terms of wave amplitudes, this is simply

a(zi+ ) ≃ a(zi− ) (1.89)

In other words, a is continuous in this approximation.

21
 Applying this result to each of the junctions between transmission line sections, we get the wave
amplitude at a junction zN in terms of that at z0 = 0:
N
!
X
−
a(zN ) ≃ a(0+ ) exp − γi ∆zi (1.90)
i=1

In terms of the voltage, we have

N
r !
+ − + + ZcN X
V (zN ) ≃ V (0 ) exp − γi ∆zi (1.91)
Zc1 i=1

where ∆zi = di = zi − zi−1 is the length of section i.

If we pass to the limit as all the ∆zi ’s go to zero and the number of sections N goes to infinity in
such a way that the position zN = z on the nonuniform line segment remains constant, then the sum in
the exponent (which is a Riemann sum) passes over into an integral, and we get

a(z) ≃ a(0)e−Q(z) (1.92)

or s
Zc (z) −Q(z)
V + (z) ≃ V + (0) e (1.93)
Zc (0)
where Z z
Q(z) = γ(z ′ ) dz ′ (1.94)
0

γ(z) is the position-dependent propagation coefficient corresponding to the z-dependent values of the
line parameters x and b: p
γ(z) = jβ(z) = j x(z)b(z)
and Zc (z) is the position-dependent characteristic impedance
s
x(z)
Zc (z) =
b(z)

The formulas to be derived can easily be extended to lossy lines or nonpropagating modes by using the
formulas p
γ(z) = z(z)y(z)
and s
z(z)
Zc =
y(z)
for the more general case. As was done for the uniform line, we choose the square roots so that Zc is
positive real if γ is pure imaginary, and otherwise Re(γ) > 0. We write

Q(z) = Q1 (z) + jQ2 (z) (1.95)

where Z z
Q1 (z) = α(z ′ ) dz ′ (1.96)
0
and Z z
Q2 (z) = β(z ′ ) dz ′ (1.97)
0

22
The term Q2 (z) is called a phase integral, because it represents the accumulated phase reduction of
the voltage wave on the nonuniform line. Finally, since the reflection at each incremental section has
been assumed to be small, it must be true that the line impedance is tracking with the characteristic
impedance,
V + (z)
Z(z) = + ≃ Zc (z)
I (z)
Therefore, the current of this +z-traveling wave is given by
V + (0)
I + (z) ≃ p e−Q(z) (1.98)
Zc (0)Zc (z)
Analogous equations can be obtained for the voltage and current of a −z-traveling wave:

b(z) ≃ b(0)e+Q(z) (1.99)

s
− − Zc (z) +Q(z)
V (z) ≃ V (0) e (1.100)
Zc (0)
V − (0)
I − (z) ≃ − p e+Q(z) (1.101)
Zc (0)Zc (z)
We then see by examining the resulting equations that the input impedance at the point z = 0 of an
inhomogeneous section of transmission line between z = 0 and z = d loaded by an impedance ZL is
given by
ZL cosh Q(d) + Zc (d) sinh Q(d)
Z(0) ≃ Zc (0) (1.102)
Zc (d) cosh Q(d) + ZL sinh Q(d)
which corresponds to a reflection coefficient of
Z(0) − Zc (0)
ρ(0) = ≃ ρWKB (0) ≡ ρL e−2Q(d) (1.103)
Z(0) + Zc (0)
at z = 0, where
ZL − Zc (d)
ρL = (1.104)
ZL + Zc (d)
is the reflection coefficient of ZL at the load point z = d. These results can be used in place of (1.55) or
(1.54) when the section of line is nonuniform. Note that in this approximation, no reflection is introduced
by the nonuniformity of the line; if the load is matched at the far end z = d of the line [ZL = Zc (d)],
then it will be matched at the near end z = 0 as well (ρ(0) ≃ 0).
More generally, a section of nonuniform line can be represented in the lowest WKB approximation
by the chain parameter description (1.56), where now the chain parameters are given by the expressions
s
Zc (z1 )
A(z1 , z2 ) = cosh[Q(z2 ) − Q(z1 )]
Zc (z2 )
p
B(z1 , z2 ) = Zc (z1 )Zc (z2 ) sinh[Q(z2 ) − Q(z1 )]
1
C(z1 , z2 ) = p sinh[Q(z2 ) − Q(z1 )]
Zc (z1 )Zc (z2 )
s
Zc (z2 )
D(z1 , z2 ) = cosh[Q(z2 ) − Q(z1 )] (1.105)
Zc (z1 )

which reduces to (1.57) for the case of a section of uniform line. The proof is left to the reader as an
exercise. Equation (1.105) can be used to prove (1.102), for example.

23
1.5.2 First-order (Rayleigh-Bremmer) WKB approximation
If the load impedance ZL in (1.102) is matched to the local characteristic impedance Zc (d) at the end of
the nonuniform line section, we get Z(0) = Zc (0) independent of the electrical length ∆ of the section, so
to this order of approximation this section would be ideally matched to a uniform line of characteristic
impedance Zc = Z(0) connected to the left of this section in z < 0. However, this result is not exactly
true, as we have assumed that the small incremental reflections at each of the junctions between piecewise
constant line sections in Fig. 1.12 can be neglected. A better approximation is obtained if we account
for this reflection as follows.
We temporarily assume that the load reflection coefficient ρL = 0. At the point zi , the forward wave
is reflected with an incremental reflection coefficient
Zc,i+1 − Zci Z ′ (zi )
∆ρ|zi = ≃ ∆zi c
Zc,i+1 + Zci 2Zc (zi )
Seen from the input point z = 0, we must include a round-trip factor of phase or attenuation for the
wave, so the reflection coefficient appears there as
Zc′ (zi )
∆ρ|z=0 ≃ e−2Q(zi ) ∆zi
2Zc (zi )
Summing up all the contributions from these incremental reflections, and passing from a sum to an
integral as we let the number of piecewise constant sections of line become infinite as before, we have
Z d ′
−2Q(z) Zc (z)
ρ(0) ≃ ρ+B (0) ≡ e dz (1.106)
0 2Zc (z)

where ρ+ B (0) is the Rayleigh-Bremmer reflection coefficient, named for the scientist who first devised
this approximation. The notation ρ+ B (0) indicates that this is the cumulative reflection coefficient of an
initially forward wave, evaluated at z = 0, due to the incremental reflections from nonuniformity of the
characteristic impedance (and not reflection from the load impedance at the end of the line). It can be
shown by more detailed study of this approximation that it is valid whenever
 Zc′ (z) 
 
 4γ(z)Zc(z)  ≪ 1
 

which in turn implies that the reflection coefficient resulting from the accumulation of these incremental
reflections is small.
If the load reflection coefficient is not zero, there will be an additional contribution to the reverse
wave at z = 0 coming from the reverse wave that starts at z = d and passes through the length of
nonuniform line according to the lowest order WKB approximation of the previous subsection. The
total reverse wave amplitude at z = 0 is therefore:

b(0) ≃ ρ+
B (0)a(0) + e
−Q(d)
b(d) (1.107)

where the wave amplitudes a(z) and b(z) are defined in (1.33). In a similar manner, the wave emerging
from the z = d end of the line is given by

a(d) ≃ ρ−
B (d)b(d) + e
−Q(d)
a(0) (1.108)

where
d
Zc′ (z)
Z
ρ−
B (d) ≡ −e −2Q(d)
e2Q(z) dz (1.109)
0 2Zc (z)
is the Rayleigh-Bremmer reflection coefficient of a reverse (−z traveling) wave seen at z = d, which is
derived in a similar way to (1.106). Expressing the wave amplitudes in terms of voltages and currents,

24
and rearranging the resulting equations in the form of the chain relations (1.56), we find that the chain
parameters of a nonuniform line in the Rayleigh-Bremmer (first-order) approximation are:
s
1 Zc (0) n Q(d)  o
1 + ρ+ − −Q(d)
 
A(0, d) = e B (0) 1 − ρB (d) + e
2 Zc (d)
1p n o
Zc (0)Zc (d) eQ(d) 1 + ρ+ − −Q(d)
  
B(0, d) = B (0) 1 + ρB (d) − e
2
1 1 n o
eQ(d) 1 − ρ+ − −Q(d)
  
C(0, d) = p B (0) 1 − ρB (d) − e
2 Zc (0)Zc (d)
s
1 Zc (d) n Q(d)  o
1 − ρ+ − −Q(d)
 
D(0, d) = e B (0) 1 + ρB (d) + e (1.110)
2 Zc (0)

Once again, the algebraic details of this derivation are left to the reader as an exercise.

Example
As an example, suppose the propagation coefficient on the nonuniform section γ = jβ is constant, while
the characteristic impedance
Zc = Zc0 e(z/d) ln(ZL /Zc0 )
varies exponentially between the value Zc0 (equal to the characteristic impedance of a uniform line
connected to the left of the terminals z = 0) and the load impedance ZL . Substitution into (1.106) gives

ln(ZL /Zc0 ) d −2jβz

 
−jβd sin βd ZL
Z
ρ(0) ≃ ρ+ B (0) = e dz = e ln (1.111)
2d 0 2βd Zc0
According to the general criterion for validity of the WKB approximation, for slow enough transitions
that βd > 4 ln(ZL /Zc0 ) (say), this result should be accurate, and the largest magnitude attained by ρ(0)
under these conditions is no greater than 0.125, for any (large enough) βd.

1.5.3 Turning points

When the rate of change of the characteristic impedance of a nonuniform transmission line in z becomes
too rapid, the approximation that reflections at successive interfaces between piecewise uniform line
sections are nearly zero breaks down. This will occur, for example, at points for which c(z) or l(z)
has a simple zero—values of z at which this occurs are said to be turning points. The analysis must
be modified near turning points, and though the rigorous details are beyond the scope of this text, a
heuristic analysis can be made to account for some of their effects.
Suppose for definiteness that we have a capacitive turning point at z = zt . That is, x(zt ) 6= 0,
b(zt ) = 0 and b′ (zt ) 6= 0. We will further assume that b is a decreasing function of z, so that in fact
b′ (zt ) < 0. If b is otherwise still a slowly varying function of z, then the lowest order WKB approximation
will break down only in a relatively small neighborhood of the turning point: zt − δz < z < zt + δz. We
now replace b(z) in this neighborhood by the piecewise constant expression

b̃(z) = b(zt − δz) ≃ −δzb′ (zt )

for zt − δz < z < zt , and

b̃(z) = b(zt + δz) ≃ δzb′ (zt )
for zt < z < zt + δz. Thus, b̃ has a small discontinuity at the turning point, but joins continuously with
b(z) outside this neighborhood (see Fig. 1.13).
The ordinary WKB method can be applied to the entire range of z now, except at the turning point,
where we can obtain the reflection coefficient at the discontinuity by usual methods (and to lowest

25
 ~
b(z)
b(z)
zt - δz zt + δz z
zt

Figure 1.13: Modified value of b(z) near a turning point zt .

order, this is the only contribution to the overall reflection coefficient). A wave incident from the left
(z < 0) will thus propagate from z = 0 to z = zt and back, accumulating a round-trip phase factor of
exp[−2Q(zt )], and will additionally encounter total reflection at zt , with a reflection coefficient of
q q
x(zt ) x(zt )
j −δzb ′ (z ) − −δzb′ (zt )
= ejπ/2
t
ρ≃ q q
x(zt ) x(zt )
j −δzb′ (zt ) + −δzb′ (zt )

The voltage exponentially decays beyond the turning point, and if a second turning point is not located
too close to the first, we may ignore any penetration of the wave beyond the first one, and regard it as
if it were isolated.
Thus, in this case, a capacitive turning point located at zt in an inhomogeneous section of line
beginning at z = 0 produces an input impedance seen at z = 0 < zt of:

1 + e−2Q(zt )+jπ/2 j cosh Q(zt ) + sinh Q(zt )

Z(0) = Zc (0) −2Q(z )+jπ/2
= Zc (0) (1.112)
1−e t cosh Q(zt ) + j sinh Q(zt )

It is as if the effect of the portion of transmission line beyond the turning point were that of a lumped
load reactance of normalized impedance3 ZL /Zc = j placed at the turning point, with the line section
in z > zt removed. This case corresponds to the TE polarization of plane waves in Table 1.1 if the
permittivity ǫ is a function of position, while µ remains constant (the usual circumstance).
A similar derivation for the case of an inductive turning point (c(zt ) 6= 0, l(zt ) = 0 and l′ (zt ) < 0)
gives a voltage reflection coefficient of
ρ = exp(−jπ/2)
at the turning point, and thus an input impedance seen at z = 0 < zt of:

1 + e−2Q(zt )−jπ/2 −j cosh Q(zt ) + sinh Q(zt )

Z(0) = Zc (0) = Zc (0) (1.113)
1 − e−2Q(zt )−jπ/2 cosh Q(zt ) − j sinh Q(zt )

because the phase of the characteristic impedance just beyond the turning point is now −π/2 rather
than +π/2 as for the capacitive turning point. This case corresponds to the TM polarization of plane
waves in Table 1.1 if the permittivity ǫ is a function of position, while µ remains constant.

3 Of course, Zc (zt ) = ∞, so we must understand this idea in a limiting sense.

26
Example
As an application, we consider the reflection of an obliquely incident TE-polarized plane wave from a
half-space with constant material parameters ǫ1 and µ0 in z < 0 onto a region in z > 0 with linearly
decreasing permittivity ǫ = ǫ1 (1 − z/d), where d is a length scale which indicates the rate of variation
of ǫ with z.4 With kc = k1 , this problem is equivalent to a nonuniform transmission line problem with
l(z) = µ0 and c(z) = ǫ1 (cos2 θ − z/d). There is a turning point at zt = d cos2 θ, and so we have
zt 3/2
2 z 2
Z p
Q(zt ) = jω l(z)c(z) dz = jk1 t1/2 = jk1 d cos3 θ
0 3 d 3

Thus, the impedance seen at z = 0 is

1 + sin 2ϑ
Z(0) = jζ1
cos 2ϑ

µ0 /ǫ1 and ϑ = 23 k1 d cos3 θ, and the resulting reflection coefficient there is

p
where ζ1 =
  
−1 cos ϑ
ρ(0) = exp 2j tan
1 + sin ϑ

Perhaps the most noticeable difference between the behavior of this reflection coefficient and that of a
wave totally reflected from an abrupt dielectric interface is that as θ → π/2, the phase here varies as
cos3 θ, whereas that of the abrupt interface goes more rapidly (as cos θ). This has implications for the
use of this effect in dielectric waveguide operation.

4 Of course, this permittivity becomes negative for z > d, but we need to assume only that this linear variation holds

up to the turning point zt < d, after which the permittivity could, for example, stabilize to a constant without affecting
the result we are about to derive.

27
1.6 Notes and References
The reader will find it useful from time to time to refer to a good introductory text in electromagnetics
for such topics as Maxwell’s equations, elements of vector analysis, elementary transmission-line theory,
and so on. The following are recommended:

Javid and Brown (1963);

Johnk (1975);
Magid (1972);
Portis (1978).

but many others are also suitable for one or more of these topics.
Examples of natural waveguides in terrestrial environments are discussed in

Bremmer (1949);
Budden (1961);
Delogne (1982).

The transmission-line equations (with negligible series inductance) have also been found to provide a
good model for the description of electrical signals on neurons; see:

Tuckwell (1988).

Acoustic phenomena are also frequently modeled by the classical transmission-line equations. See, for
example,

Morse and Ingard (1968), chapter 9;

Fletcher and Rossing (1991).

The latter book shows how a wide variety of musical instruments can be described in terms of trans-
mission lines. The reader familiar with electric circuits and transmission lines will find the terminology
used in these books quite familiar. Interesting examples in the area of active and nonlinear transmission
lines (topics not covered in this book) are given in

Scott (1970);
Wong (1993);
Remoissenet (1996).

What is perhaps the earliest published account of an experimental demonstration of electrical trans-
mission line phenomena is given in:

W. Watson, “An account of the Experiments made by several Gentlemen of the Royal Society,
in order to discover whether or no the electrical Power, when the Conductors thereof were
not supported by Electrics per se would be sensible at great Distances: With an Inquiry
concerning the respective Velocities of Electricity and Sound: To which is added an Ap-
pendix, containing some further Inquiries into the Nature and Properties of Electricity,”
Phil. Trans. Roy. Soc. London, vol. 45, pp. 49-91 (1748).
W. Watson, “An account of the Experiments made by some Gentlemen of the Royal Society,
in order to measure the absolute Velocity of Electricity,” Phil. Trans. Roy. Soc. London,
vol. 45, pp. 491-496 (1748).

The description of the sometimes reluctant participants in these experiments is quite amusing in places,
especially given that no one seems to have been permanently harmed in the process. The telegraphers’
equations were first derived by Lord Kelvin more than a century later:

28
 W. Thomson (Lord Kelvin), “On the theory of the electric telegraph,” Proc. Roy. Soc.
London, vol. 7, pp. 382-399 (1855) [also in his Mathematical and Physical Papers, vol. 2.
Cambridge, UK: University Press, 1884, pp. 61-76].

although in this original work, the series inductance of the line was neglected. A few years later, Kirchhoff
would include the inductance for a special case, but Heaviside appears to have been the first to present
the complete equations for the general case:

O. Heaviside, “On the extra current,” Phil. Mag., ser. 5, vol. 2, pp. 135-145 (1876) [also in
his Electrical Papers, vol. 1. New York: Chelsea, 1970, pp. 53-61].

The classical transmission line with unusual series or shunt parameters can be used as a model for
TE or TM modes in hollow metallic waveguides:

Milton and Schwinger (2006), chapter 13.

(see also chapter 8). It also serves as a model for voltage and current in distributed transformer windings
that have significant capacitance between adjacent turns:

Bewley (1963), chapters 15-16.

Rüdenberg (1968), chapters 19-21.

The representation of plane wave reflection and transmission problems using equivalent transmission-
line networks is discussed in

Solimeno, et al. (1986), chapter 3, section 7.

More detailed treatments of the WKB method and propagation in inhomogeneous media are to be
found in many sources, for example:

Brekhovskikh (1960), section 16;

Sodha and Ghatak (1977);
Marcuse (1982), chapter 11;
Snyder and Love (1983), chapter 35;
Solimeno, et al. (1986), chapter 3.

The approach used here was first used by Rayleigh

Lord Rayleigh, “On the propagation of waves through a stratified medium, with special
reference to the question of reflection,” Proc. Roy. Soc. London A, vol. 86, pp. 207-226
(1912).

in a special case, but is usually attributed to Bremmer:

H. Bremmer, “The W.K.B. approximation as the first term of a geometric-optical series,”

Commun. Pure Appl. Math., vol. 4, pp. 105-115 (1951).

where higher order approximations are also derived. The phase shift due to reflection at a turning point
is usually treated by recourse to matched asymptotic expansions and Airy functions, as for instance in:

Brekhovskikh, loc. cit.;

Marcuse (1982), section 11.5.

The (less rigorous) approach we take is due to

29
 E. Persico, “Dimostrazione elementare del metodo di Wentzel e Brillouin,” Nuov. Cim., vol.
15, pp. 133-138 (1938).
S. Kaliski, “The WKB method in the ‘classical’ approach to quantum mechanics,” Bull.
Acad. Polonaise Sci., ser. Sci. Tech., vol. 18, pp. 553-559 (1970).
G. B. Hocker and W. K. Burns, “Modes in diffused optical waveguides of arbitrary index
profile,” IEEE J. Quantum Electron., vol. 11, pp. 270-276 (1975).
D. Marcuse, “Elementary derivation of the phase shift at a caustic,” Appl. Opt., vol. 15, pp.
2949-2950 (1976).
F. S. Crawford, “Heuristic derivation of the ‘WKB penetration’ phase constant for bound
states,” Amer. J. Phys., vol. 56, pp. 374-375 (1988).
In any case, knowledge of the phase shift by itself does not repair the defect of the WKB solution that
the approximate voltage and current can become singular at the turning point. There have been some
modifications of the WKB method which do not suffer from this drawback, and yet avoid the use of Airy
functions; see in particular:
S. N. Stolyarov and Yu. A. Filatov, “Reflection of electromagnetic waves from inhomogeneous
layers,” Radio Eng. Electron. Phys., vol. 28, no. 12, pp. 24-28 (1983).
The modified formulas in this paper also remain valid in other situations where ordinary WKB fails,
notably when Zc (z) has a step function behavior. The exact solution of the exponentially tapered
transmission line problem can be found in
Slater (1942), pp. 75-78.
Reich et al. (1953), sect. 4-6.
for comparison with the WKB approximation given in this chapter. The WKB solution can be used
to design optimal wideband low-reflection transitions between widely differing impedances. See, for
example,
R. W. Klopfenstein, “A transmission line taper of improved design,” Proc. IRE, vol. 44, pp.
31-35 (1956).
D. Kajfez and J. O. Prewitt, “Correction to ‘A transmission line taper of improved design’,”
IEEE Trans. Micr. Theory Tech., vol. 21, p. 364 (1973).
as well as
Kuznetsov and Stratonovich (1964), chap. 4.

1.7 Problems
p1-1 A length d of lossless transmission line (Zc1 , β1 ) is connected between a real load impedance
ZL and a second lossless line (Zc2 , β2 ) as shown. Show that the most general conditions
needed in order for no reflected wave to exist on the second line are that either
p (2n + 1)π
Zc1 = ZL Zc2 and β1 d = , n = 0, 1, 2, ...
2
or Zc2 = ZL , with

(2n)π
β1 d = , n = 0, 1, 2, ... if Zc1 6= Zc2
2
β1 d = arbitrary if Zc1 = Zc2

In the first case, we have the so-called quarter-wave transformer.

30
 (β2, Zc2) (β1, Zc1) ZL

z=0 z=d

p1-2 Show how to use a single short-circuited transmission-line stub of length d2 to provide a
matched (Zin = Zc ) input impedance to a transmission line section loaded with an arbitrary
impedance ZL . Specifically, given an arbitrary complex value of ZL , and real values of β and
Zc , give formulas (not a Smith-chart algorithm) for d1 and d2 for the matching arrangement
in the figure, which is known as a single-stub tuner. Assume that all transmission lines are
lossless (α = 0 and Zc is real).

d1

Zin (β, Zc) ZL

(β, Zc)
d2

p1-3 A wire helix located parallel to a ground plane as shown in part (a) of the figure can serve as
a transmission line. If the wire and ground plane are lossless, and losses in the intervening
medium can be neglected, then we can put r = g = 0. However, in addition to the series
inductance l∆z and shunt capacitance c∆z shown in the distributed circuit model of Fig. 1.3,
there are additional effects which should be taken into account. There is a capacitive effect
between adjacent windings of the helix, which can be modeled as an elastance s∆z in parallel
with l∆z as shown in part (b) of the figure.

31
 wire helix

s∆z
s∆z

l∆z

c∆z
∆z
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

z z +∆ z

(a) (b)

(a) Give a physical argument for why the additional capacitance term should have the
elastance form s∆z rather than a capacitance form cw ∆z (as is the case for the shunt
capacitance). In other words, as ∆z → 0, why should this additional capacitance go to
infinity rather than to zero?
(b) Using the distributed circuit model for this line, calculate the characteristic impedance
Zc and complex propagation coefficient γ.
(c) Explain the difference in the behavior of Zc and γ as functions of ω for this line from
that of the conventional line described by (1.8) and (1.9).

p1-4 Use equation (E.19) to prove that, on a passive transmission line which is terminated in a
load impedance ZL with real part greater than or equal to zero, the line impedance Z(z) will
have nonnegative real part for all z.

p1-5 Consider a lossy transmission line whose characteristic impedance has a phase angle φZ with
−π/2 < φZ < π/2.

(a) Using the result of the previous problem, prove that the set of possible values of the
reflection coefficient ρ(z) lies in the interior of the circle

|ρ(z) + j tan φZ | ≤ sec φZ

When φZ = 0, this is the interior of the unit circle as commonly displayed on the Smith
chart. Show that for φZ 6= 0, we can have |ρ| > 1 under appropriate circumstances.
(b) Show that the time-average power flow obtained from (1.21) at the point z = 0 is given
by
Pav = Pav,inc 1 − |ρ|2 − 2Im(ρ) tan φZ
 

where ρ = ρ(0), and if the mode is normalized,

Pav,inc = |a0 |2 cos φZ

Show that Pav ≥ 0, and thus that the condition |ρ| > 1 shown to be possible in part (a)
does not violate conservation of energy.

p1-6 By the method used to obtain (1.80) and (1.81), find the (electric field) Fresnel reflection
and transmission coefficients ρ and τ for an obliquely incident, parallel-polarized plane
TM TM

32
 wave at the interface between two media. Note: As normally defined, these coefficients are
the ratios of total electric fields (which in this polarization have both x and z components).
The transmission-line analogy of Table 1.1 produces only ratios of Ex in this polarization—
be sure to convert your answer to conform to the usual definitions, and label your figure to
indicate directions of positive electric field. For the case of total reflection (when µr1 ǫr1 >
(kx /k0 )2 > µr2 ǫr2 or in other words k12 > k12 sin2 θ1 > k22 ), express the reflection coefficient
in the form

ρ = ejχTM
TM

and give an expression for χ (θ).

TM

p1-7 Consider a lossless nonuniform transmission line with constant β but characteristic impedance
which varies as
    
z ZL A d
Zc (z) = Zc0 exp ln + z(z − d) z −
d Zc0 d3 2

where A is a real, dimensionless constant. This impedance, like the example of the exponential
line treated in the notes, varies continuously between the values of Zc0 at z = 0 and some
real load impedance ZL to be connected at z = d. The additional term allows the slope of
Zc (z) to be adjusted by choosing the value of A appropriately. We will investigate whether
this can achieve further reduction of the reflection coefficient at the input of the nonuniform
line section.

(a) Use equation (1.106) to obtain an expression for the reflection coefficient ρ(0) at the
input when the line is loaded with the impedance ZL at z = d.

(b) Use whatever means (analytical calculation or numerical computation on a computer)

you wish to determine the value of A which minimizes the magnitude of the second
highest maximum of |ρ(0)| as a function of βd (the largest maximum occurs at βd = 0,
and cannot be reduced in size). Use the value ZL = 4Zc0 for the load impedance. Plot
your result for |ρ(0)| vs. βd, along with comparisons to the result obtained in the notes
for the exponential line, and to the value for the case when ZL directly terminates the
line at z = 0, with no nonuniform line present.

p1-8 A length d of the transmission line described in section 1.2.2 is terminated with a real load
impedance ZL = RL . At the other end of the line, a voltage generator of strength VG is
connected. The parameters VG , RL , s, l and c are all assumed to be real and independent
of frequency. Make a rough sketch of the ratio |VL /VG | vs. ω, where VL is the voltage across
RL . You don’t have to make precise calculations, just indicate what the ratio does near
ω = 0, ω = ωc and as ω → ∞.

p1-9 Derive equation (1.16).

p1-10 Consider a classical transmission line whose distributed-parameter model is as shown in the
figure.

33
 s∆z

l∆z
c∆z

rm∆z

z z +∆ z

This is like the line of problem p1-3, except that there is an additional term in the shunt
impedance: a reluctance per unit length rm (reluctance is the inverse of inductance, and has
units of inverse Henries, also known as Sturgeons(!)). Obtain expressions for the propagation
constant and characteristic impedance of this line. If we assume that sc > rm l, determine
whether γ and Zc are real or imaginary in various ranges of the frequency, and specify their
algebraic signs in each case.

p1-11 Consider a classical transmission line whose distributed-parameter model is as shown in the
figure.

m∆z m∆z

l∆z l∆z l∆z

c∆z c∆z
... ...

z z+∆z

This differs from the classical distributed-network model of a transmission line (Fig. 1.3)
by the inclusion of a mutual inductance m per unit length between nearest-neighbor series
inductances of the circuit model. Derive the telegrapher’s equations that apply to this trans-
mission line (be sure to observe the “dot convention” for the mutual inductances). Show
that it is equivalent to a line of Fig. 1.3 if the series inductance per unit length l is replaced
by an equivalent value leq , and give an expression for leq .

p1-12 Consider a classical transmission line whose distributed-parameter model is as shown in the
figure.

34
 s1∆z s2∆z

l∆z
c∆z

z z +∆z

This is like the line of example 1.2.2, except that there is an additional term parallel to the
series inductance: an elastance per unit length s2 . Define the frequencies
r r
s1 s2 s2
ωc1 ≡ ωc2 ≡
l(s1 + s2 ) l

Make a qualitative plot of the series reactance per unit length x versus frequency, noting
in particular what happens when ω is near 0, ωc1 , ωc2 and ∞. Obtain expressions for the
propagation constant and characteristic impedance of this line. Determine whether γ and Zc
are real or imaginary in the various ranges of the frequency, and specify their algebraic signs
in each case. What kind of filter does this line resemble?

p1-13 Prove (1.105). Then use it to obtain (in the lowest-order WKB approximation) an expression
for the ratio V (d)/V (0) of output to input voltage for a section of nonuniform transmission
line terminated at z = d with a load impedance ZL .

p1-14 Consider a nonuniform transmission line on which both the characteristic impedance and
the propagation constant vary with position in 0 < z < d. Suppose that γ = jβ, with
 
1 z
β(z) = β0 +
2 d

and
z2
   
1 z ZL
Zc (z) = Zc0 exp + 2 ln
2 d d Zc0
where β0 , Zc0 and ZL are positive constants. Plot these two functions vs. z/d, using values
such that ZL /Zc0 = 4, and compare them to the corresponding functions from the example
of section 1.5.2 (taking β = β0 for that case). Obtain an expression for the Rayleigh-
Bremmer reflection coefficient ρ+ B (0) of this section of line. How does it compare to that of
the exponentially tapered line from the example of section 1.5.2?

p1-15 Prove (1.110).

p1-16 If a load impedance ZL is connected to the end z = d of a nonuniform transmission line,

such that the reflection coefficient at this end is
ZL − Zc (d)
ρL = 6= 0
ZL + Zc (d)

35
 show that the overall reflection coefficient from the input end z = 0 is given in the Rayleigh-
Bremmer approximation by

ρWKB (0)
ρ(0) ≃ ρ+
B (0) +
1 − ρL ρ−
B (d)

Give a physical interpretation of this result.

p1-17 Let the values r(> 0), g(> 0) and b of a classical transmission line be given. For what
value of x does the attenuation constant α attain its minimum possible value? What is this
minimum value, and what are the corresponding values of β and Zc ?
p1-18 A layer of material with permittivity ǫ and permeability µ lies between the planes z = 0
and z = d. Outside this layer (in z < 0 and z > d) is free space. Using the transmission line
equivalence derived in section 1.4, find the total reflection coefficient of a TE-polarized plane
wave incident at an angle θ to the z-axis from the surface z = 0 of this layer.
p1-19 Derive equations (1.56) and (1.57).
p1-20 Consider a lossless nonuniform transmission line with constant γ = jβ but characteristic
impedance that varies as
    
z ZL 2πz
Zc (z) = Zc0 exp ln + A sin
d Zc0 d

where A is a real, dimensionless constant. This impedance, like the example of the exponential
line treated in the notes, varies continuously between the values of Zc0 at z = 0 and some
real load impedance ZL to be connected at z = d. The additional term allows the slope of
Zc (z) within the interval 0 < z < d to be adjusted by choosing the value of A appropriately.
We will investigate whether this can achieve further reduction of the reflection coefficient at
the input of the nonuniform line section.

(a) Use equation (1.106) to obtain an expression for the Rayleigh-Bremmer reflection coef-
ficient ρ+
B (0) at the input when the line is loaded with the impedance ZL at z = d.
(b) Choose a small positive value of A, and using this choice of A plot ρ+
 
B (0) vs. βd for

0 < βd < 10. What is the effect of the modified variation of Zc in this case?
(c) Choose a small negative value of A, and using this choice of A plot ρ+
 
B (0) vs. βd for

0 < βd < 10. What is the effect of the modified variation of Zc in this case?
(d) By varying the values of A used in parts (b) and (c) above, can you in some sense
optimize the reflection coefficient? Note that if your choice of A causes ρ+B (0) > 1

for some frequencies, then the Rayleigh-Bremmer approximation is no longer valid, and
this choice of A is not permitted.

36
Chapter 2

MULTIPORT NETWORK
THEORY: MATRIX
DESCRIPTIONS

the long arm of coincidence

makes many radio connections
—don marquis,
“archy on the radio”
from the lives and times of archy and mehitabel

Transmission lines are used to interconnect circuit elements in far more complicated ways than those
indicated in chapter 1. Analysis and design of such networks is made more systematic by the introduction
of the concepts of multiport networks as used in lumped-element circuit theory. There are, however,
certain aspects of this formalism that are special to the inclusion of transmission lines in the network.

2.1 Sourceless Linear Two-Port Networks

Networks composed of lumped elements and sections of transmission line are often viewed in terms
of their connection points to the outside world. These terminals are nodes of the network, at which
connections to external circuits may be made. Not all nodes of a circuit are used as terminals, only
those to which external connections are to be made. We will consider networks with N + 1 terminals,
where N ≥ 1. At each terminal there is a current flow into the network, and the sum of all terminal
currents for a finite network must be zero by Kirchhoff’s current law.
Often it is convenient to consider networks whose terminals appear in pairs called ports, such that
the total current into the network at each port is zero (that is, any current entering the network through
one terminal of a port must leave through the other terminal). One terminal of each port is identified as
positive for voltage reference. Voltage across the port terminals with this convention is called the port
voltage, and the current going into the network through the positive port terminal is called the port
current. Unless provided for by external circuit connections or other considerations, it is not permissible
to measure a voltage between terminals belonging to two different ports (see the discussion of the classical
transmission line in section 1.2). One such possible external connection arises in the identification of an
(N + 1)-terminal network as an N -port network. To do this, we choose one of the network terminals and
designate it as “ground” or “reference”. Each of the other terminals then forms a port together with a
terminal connected to the ground terminal. Because each port shares a common terminal connection, it
is possible in this case to talk of voltage between any of the port terminals.

37
 I1

+
V1
– I2

+
I1 V2
–

I2

Figure 2.1: The linear, sourceless two-port network.

2.1.1 Impedance, admittance and chain matrices; image parameters

Consider a two-port network containing only linear lumped elements and/or segments of linear classical
transmission line, but no impressed sources (Figure 2.1). A linear relationship must therefore exist
among the terminal variables (the port voltages and currents):

V1 = Z11 I1 + Z12 I2 (2.1)

V2 = Z21 I1 + Z22 I2 (2.2)

or, in more compact matrix form:

[V ] = [Z][I] (2.3)
where  
V1
[V ] =
V2
 
I1
[I] =
I2
 
Z11 Z12
[Z] =
Z21 Z22
A network is said to be reciprocal (see section E.2.1 of Appendix E) if for any two states a and b of
the network (that is, for the port voltages and currents resulting from any two sets of external sources
connected to the ports of the network), we have

(V1a I1b − V1b I1a ) = (−V2a I2b + V2b I2a ) (2.4)

Eqn. (2.4) can be expressed in compact matrix-vector notation as:

[I b ]T [V a ] = [I a ]T [V b ] (2.5)
T
where the superscript denotes the transpose of a matrix or vector, e. g.,
 
Z11 Z21
[I]T = [I1 , I2 ] ; [Z]T =
Z12 Z22

38
 Z11 - Z12 Z22 - Z12

port 1 Z12 port 2

Figure 2.2: Equivalent T-network for a reciprocal two-port.

-Y12

port 1 Y11 +Y12 Y22 +Y12 port 2

Figure 2.3: Equivalent Π-network for a reciprocal two-port.

But since (2.3) holds for any state of this junction, (2.5) implies

[I b ]T [Z][I a ] = [I a ]T [Z][I b ]
 aT T
= [I ] [Z][I b ]
= [I b ]T [Z]T [I a ] (2.6)

since a scalar quantity is its own transpose. Since (2.6) must hold for any [I a ] and [I b ], we conclude that

[Z] = [Z]T (2.7)

In other words, [Z] is a symmetric matrix for a reciprocal network, which means that

Z12 = Z21 (2.8)

This reciprocity law holds only for certain kinds of two-port networks. Later, we will see how to charac-
terize the property of reciprocity in terms of the electromagnetic material parameters of the network. In
Appendix E, however, we show how the reciprocity property follows from the properties of the lumped
elements and transmission lines which constitute the network.
We might also characterize the junction in terms of an admittance matrix [Y ]:

[I] = [Y ][V ] (2.9)

Clearly [Y ] = [Z]−1 when the inverse exists, and it is evident that [Y ] is also a symmetric matrix when
the network is reciprocal.
As is well known from circuit theory, many networks with two ports can produce the same port
responses as a network with a given impedance matrix [Z]. The equivalent T-network for a reciprocal two-
port network which arises from a 3-terminal network is shown in Fig. 2.2. Also commonly used in this
case is the equivalent Π-network, which is most easily displayed in terms of the elements of the admittance
matrix (Fig. 2.3). Note that the impedances and admittances appearing in these equivalent circuits may
not each be physically realizable; some, for example, could have negative real parts. Nevertheless,
the equivalent circuit still serves a useful purpose in the analysis of transmission-line/lumped-element
circuits.

39
 1:1 1:1 1:1 1:1

(a) (b)

Figure 2.4: Generalized T- and Π-network equivalent circuits for reciprocal sourceless two-ports.

The T- and Π-networks given above are suitable if only legitimate port connections are allowed to be
made. That is, if external connections made to ports 1 and 2 are such that the current into one terminal
of each port is always equal to the current leaving the other terminal of the port, these equivalent circuits
will suffice. However, if the external connection could result in other current conditions (for example,
an impedance is connected between the positive terminals of ports 1 and 2, or there is a possibility that
the “ground” terminals of ports 1 and 2 might support a voltage difference), we must insert 1:1 ideal
transformers at each port to allow for this. Thus, we have the more general equivalent circuits shown in
Fig. 2.4.
If the network is not reciprocal, we cannot use the T or Π equivalent networks. In such cases, it
may be useful to use an equivalent network containing dependent voltage or current sources as shown in
Fig. 2.5. Such networks can be applied, for instance, in modeling a circuit using software such as SPICE.
Some idealized circuit elements do not have impedance or admittance matrices because one or more
of their elements is infinite. An example of this is the ideal transformer (Fig. 2.6). This device is
characterized by the relation
    
V1 1/n 0 V2
= (2.10)
I1 0 n −I2

where n is the turns ratio of the transformer. Relation (2.10) is a special case of the use of the so-called
chain matrix or ABCD-matrix of which the general form is
    
V1 A B V2
= (2.11)
I1 C D −I2

The sign of I2 is reversed in this expression because the usual application of the chain matrix is in
cascaded two-port networks (e. g., Fig. 2.7) in which −I2 of an output side becomes +I1′ of the next
input side, and the effect of the cascaded circuit is simply to multiply the corresponding chain matrices.
If we wish to utilize the chain matrix in the “reverse” direction, we have
   −1  
V2 A B V1
= (2.12)
−I2 C D I1

where
 −1  
A B 1 D −B
= (2.13)
C D AD − BC −C A

Comparing (2.11) with (2.2), we can obtain the chain parameters in terms of the impedance param-

40
 I1 I2

+ +
Z11 Z22

V1 V2
+ +
Z12I2 Z21I1
– –
– –

(a)

I1 I2

+ +
Y21V1
V1 Y11 Y22 V2
Y12V2
– –

(b)

Figure 2.5: Equivalent networks with dependent sources for nonreciprocal two-ports: (a) in terms of
Z-parameters, (b) in terms of Y -parameters.

I1 I2
+ +
(port 1) V1 V2 (port 2)
– –
1:n

Figure 2.6: The ideal transformer

41
 I1 I2 I1' I2'

A B A' B'

C D C' D'

Figure 2.7: Cascaded networks in chain matrix description.

(Z0, γl)

1:n 1:n

Figure 2.8: Equivalent circuit for reciprocal two-port based on chain parameters.

eters:
Z11
A =
Z21
Z22 Z11
B = − Z12
Z21
1
C =
Z21
Z22
D = (2.14)
Z21
From (2.14) we can verify that the determinant of the chain matrix is
Z12
AD − BC = (2.15)
Z21
which is equal to 1 if the network is reciprocal. For the special case of the ideal transformer, it can be
seen from (2.10)-(2.14) that the impedance matrix would need to have Z21 → ∞, in such a way that
Z22 /Z21 → n, Z11 /Z21 → 1/n and Z12 /Z21 → 1.
For reciprocal networks, one of the most natural equivalent networks expressed in terms of the chain
parameters is shown in Fig. 2.8. The network is a cascade connection of two ideal transformers and a
length of transmission line. The transmission line parameters are:
√
r
B
Z0 = ; γl = cosh−1 AD
C
while the transformer turns ratios are given by:
 1/4
D
n=
A

42
 Finally, occasional use is made of the image parameters of a two-port network. Image parameters
are used in the classical design of filters, and in the design of long transmission circuits in which an
impedance match must be maintained. The image impedances ZI1 and ZI2 are defined by the two
following requirements. First, if ZI2 is connected to port 2, the impedance seen looking into port 1
is ZI1 ; conversely, if ZI1 is connected to port 1, the impedance seen looking into port 2 is ZI2 . The
remaining image parameters for a two-port are the image transfer constants Γ12 and Γ21 . If port 2 is
terminated with ZI2 , then we put
−V2 I2
e−2Γ12 = −
V1 I1
and analogously for Γ21 (which equals Γ12 for a reciprocal network). The equivalent circuit of Fig. 2.8
can also be related to the image parameters, with
p
Z0 = ZI1 ZI2 ; γl = Γ12

and
 1/4
ZI2
n=
ZI1

2.1.2 The scattering matrix and related representations

For two-port networks, we will adopt the following conventions:

(i) A transmission line is assumed to be connected to each of the ports, with separate z-axes
for each line defined such that the +z-direction points into the two-port network.
(ii) The characteristic impedance of the transmission line connected to port i is denoted Zc,i .
(iii) We define a forward (incident) wave amplitude ai and a reverse (reflected) wave ampli-
tude bi at each port as [cf. (1.30)-(1.33)]:

Vi + Zc,i Ii
ai = p
8Zc,i
Vi − Zc,i Ii
bi = p (2.16)
8Zc,i

−1/2
if Zc,i 6= 0. The factor involving Zc,i is introduced into (2.16) so that whenever Zc,i is
real, the time-average power 12 Re(Vi Ii∗ ) entering port i is equal to (|ai |2 − |bi |2 ).
(iv) We define a characteristic impedance matrix [Zc ] as the diagonal matrix whose nonzero
elements are the characteristic impedances of the transmission lines associated with ports
1 and 2:  
 Zc,1 0 
[Zc ] = 


 (2.17)
0 Zc,2

It is particularly simple to calculate arbitrary powers of such a matrix, viz.:

 
q
 Zc,1 0
[Zc ]q = 



 (2.18)
q
0 Zc,2

for any real number q.

43
 In situations where we are concerned with standing wave ratios in the system, a sort of generalization
of the reflection coefficient known as the scattering matrix [S] is the most convenient way to describe
the junction. Introducing the incident and reflected wave amplitudes (a1 , a2 ) and (b1 , b2 ) from V1 , V2 ,
I1 and I2 according to (2.16), we define the [S]-matrix so that

[b] = [S][a] (2.19)

for any possible [a] and [b], where

     
 a1   b1   S11 S12 
[a] = 


 ; [b] = 


 ; [S] = 


 (2.20)
a2 b2 S21 S22

From (2.16), we have that

1
[a] = [2Zc ]−1/2 {[V ] + [Zc ][I]}
2
1
[b] = [2Zc ]−1/2 {[V ] − [Zc ][I]} (2.21)
2
Since relations (2.19), (2.21) and (2.3) must hold for all possible [I], we find that
−1
[S] = [Zc ]−1/2 {[Z] − [Zc ]} {[Z] + [Zc ]} [Zc ]1/2 (2.22)

Note that if the characteristic impedance at both ports is the same (Zc1 = Zc2 ), the matrix [Zc ] is
proportional to the identity matrix  
 1 0 
[1] = 


 (2.23)
0 1

and in this case (2.22) simplifies to

−1
[S] = {[Z] − [Zc ]} {[Z] + [Zc ]} (2.24)

It is sometimes convenient to use a normalized impedance matrix [Z̄] analogous to the scalar nor-
malized impedance Z̄ encountered in classical transmission-line theory. There are many ways in which
we could define such a matrix ([Zc ]−1 [Z], for example), but only one simple definition yields a matrix
which is symmetric if [Z] is symmetric:

[Z̄] = [Zc ]−1/2 [Z][Zc ]−1/2 (2.25)

and
[Z] = [Zc ]1/2 [Z̄][Zc ]1/2 (2.26)
(which reduce to [Z] = Zc [Z̄] if Zc1 = Zc2 ) so that for a reciprocal two-port:

[Z̄]T = [Z̄] (2.27)

Likewise, we can define the normalized admittance matrix

[Ȳ ] = [Z̄]−1 = [Zc ]1/2 [Y ][Zc ]1/2 (2.28)

which is also symmetric for reciprocal networks.

44
 Figure 2.9: Cascaded networks in T -matrix description.

Using the normalized impedance matrix, we have from (2.22):

  −1
[S] = [Z̄] − [1] [Z̄] + [1] (2.29)

Note the similarity of (2.29) to the relationship between the reflection coefficient and the impedance on
a classical transmission line (1.51). Rearranging (2.29) to solve for [Z̄], we have

−1
[Z̄] = {[1] − [S]} {[1] + [S]} (2.30)

This formula, too, is analogous to the corresponding expression for impedance in terms of reflection
coefficient on a transmission line.
Since [Z̄] is symmetric for a reciprocal network, it is readily seen that [S] is also symmetric in this
case:
−1 
[S]T

= [Z̄] + [1] [Z̄] − [1]
 −1
= [1] − 2 [Z̄] + [1]
  −1
= [Z̄] − [1] [Z̄] + [1]
= [S] (2.31)

i. e., S12 = S21 .

Another matrix which is useful for representing two-ports in cascaded connection is the (wave-
amplitude) transmission matrix (sometimes called the transfer matrix or [T ]-matrix). For this,
   
 b2   a1 
  = [T ]   (2.32)
   
a2 b1

Because the reflected wave from one junction becomes the incident wave at the next, an overall [T ]-matrix
for cascaded networks (Fig. 2.9) is obtained by matrix multiplication:
   
b′2  a1 
 = [T ′ ][T ] 
 

  

 (2.33)
′
a2 b1

The elements of [T ] are most easily related to those of [S], by comparing (2.32) to (2.19):

45
 S11 S22
T11 = S21 − (2.34)
S12
S22
T12 = (2.35)
S12
S11
T21 = − (2.36)
S12
1
T22 = (2.37)
S12

For the determinant of [T ], we have

S21
det[T ] ≡ ∆T = (2.38)
S12

which is 1 if reciprocity holds.

A matrix related to [T ] which is sometimes encountered is the [R]-matrix, for which
   
 b1   a2 
  = [R]   (2.39)
   
a1 b2

A comparison of (2.32) and (2.39), taking into account (2.38), shows that

T11
R11 = (2.40)
∆T
T21
R12 = − (2.41)
∆T
T12
R21 = − (2.42)
∆T
T22
R22 = (2.43)
∆T

This matrix, too, finds application in cascading problems.

2.1.3 Example: Equivalent network for a length of classical transmission line

Consider the two-port consisting of a length l of classical transmission line whose characteristic impedance
is Z0 and propagation coefficient γ = jβ, as shown in Fig. 2.10(a). The characteristic impedance
associated with both ports is Zc1 = Zc2 = Z0 . The scattering matrix for this circuit is readily obtained,
since waves pass through it without reflection:

b2 = e−jβl a1 (2.44)
b1 = e−jβl a2 (2.45)

and therefore
 
−jβl
 0 e 
[S] = 


 (2.46)
−jβl
e 0

46
 l
Zs Zs

(Z0 , γ ) Z12

(a) (b)
-Y12

Yp Yp

(c)
Figure 2.10: (a) A finite section of classical transmission line; (b) Equivalent T-network; (c) Equivalent
Π-network.

The normalized impedance matrix [Z̄] is then obtained from (2.30) as:
−1
[Z̄] = {[1] − [S]} {[1] + [S]}
  
−jβl −jβl
1  1 e  1 e 
=   
1 − e−2jβl   
e−jβl 1 e−jβl 1
 
 −j cot βl −j csc βl 
= 


 (2.47)
−j csc βl −j cot βl

Using  
 Z0 0 
[Zc ] = 



0 Z0

to denormalize [Z̄], we have  

 −jZ0 cot βl −jZ0 csc βl 
[Z] = 


 (2.48)
−jZ0 csc βl −jZ0 cot βl
Based upon the equivalent T-network of Fig. 2.2, we can assert that the length of transmission line can
be represented by the equivalent lumped circuit of Fig. 2.10(b), where
βl
Zs = jZ0 tan
2
Z12 = −jZ0 csc βl (2.49)

47
 Z

Figure 2.11: Series and shunt connections of impedance as two-port networks.

In a similar way, we find that the Π-equivalent network of Fig. 2.10(c) has parameter values:
j βl
Yp = tan
Z0 2
j
−Y12 = − csc βl (2.50)
Z0
Finally, use of (2.14) yields the chain parameters
A = cos βl
B = jZ0 sin βl
j
C = sin βl
Z0
D = cos βl (2.51)
which is a special case of (1.57) for lossless transmission lines.

2.1.4 Example: Use of chain matrices

In Figure 2.11, we show two simple two-ports consisting of a single impedance Z connected in series or a
single admittance Y connected in shunt between the ports. By inspection, the chain matrix of the series
impedance is seen to be    
 A B   1 Z 
 =  (2.52)
   
C D 0 1
and that of the shunt admittance is
   
 A B   1 0 
 =  (2.53)
   
C D Y 1

If a load impedance ZL is connected to port 2 of a sourceless two-port, it is readily shown that the
input impedance looking into port 1 is
AZL + B
Zin,1 = (2.54)
CZL + D
Likewise, the voltage at port 2 is expressed in terms of the voltage at port 1 as:
ZL
V2 = V1 (2.55)
AZL + B
Note the similarity to (1.55) and (1.61), which are the special cases of (2.54) and (2.55) when the two-port
is a segment of transmission line.

48
 1 +
+
V1 = V10
–
0
λ/2

V2

– 0
V20
+ –
2

Figure 2.12: Half-wave transmission-line loop as 1:2 transformer.

2.1.5 Example: Realization of a 1:2 ideal transformer

Ideal transformers are often useful circuit elements, providing as they do impedance transformation over
a wide frequency range. Unfortunately, an actual transformer will fail to produce the desired behavior,
increasingly so as frequency increases. This is due to losses in the transformer’s magnetic core, and to
leakage and parasitic (inductive, capacitive or resistive) effects which cannot be neglected as is the case
at lower frequencies. Thus, it is important to be able to accomplish this circuit function in another way
that is free of these drawbacks, even if wide bandwidth must be sacrificed.
One way of doing this is to use a length l (= λ/2 at the design frequency) of transmission line, one
terminal from each end connected together at terminal 0 as shown in Figure 2.12. We will view this
network as a two-port, at first taking terminals 1 and 0 to form port 1, and terminals 2 and 0 to form port
2, with terminals 1 and 2 taken to be the positive reference nodes and the corresponding port voltages
denoted as V10 and V20 respectively. From (2.51), we have A = D = −1 and B = C = 0, independent
of the characteristic impedance of the transmission line. From (2.55), we find that V10 = −V20 : since
l = λ/2, the voltage at one end of this loop will be the negative of that at the other end.
Now let us change our point of view so that the second port is now taken to be the terminal pair
1-2, with terminal 1 being the positive reference as shown. Port 1 remains defined as before, so that
V1 = V10 . On the other hand, at the new port 2 we have V2 = V10 − V20 = 2V1 , and a simple calculation
shows that the current at port 1 is twice that at port 2, meaning that this device behaves as a 1:2
ideal transformer, but only at the design frequency f0 . It deviates from ideal performance away from
that frequency, but the lack of the drawbacks of low-frequency transformer implementations makes it a
popular method to accomplish this circuit function at RF frequencies and above.

2.1.6 Example: A two-terminal antenna as a two-port

Wire antennas are used from very low frequencies up to UHF and beyond for transmission as well as
reception. They typically operate in a balanced mode with respect to the “ground” (i. e., the environ-
ment) as shown in Fig. 2.13(a). In this mode, the current at each of the antenna terminals is the same,
though oppositely directed. In contrast to the monopole mode of operation shown in Fig. 2.13(b), the
balanced mode of operation is relatively insensitive to the presence of other conductors (especially the
earth) in the vicinity of the antenna. An equivalent circuit for this antenna incorporating both modes of
operation is shown in Fig. 2.13(c). What is normally referred to as the antenna impedance is given by
Za = Za1 + Za2 , while the “unbalanced” antenna impedance Zu represents the impedance of interaction
between the antenna and ground. In other words, the antenna must generally be regarded as a two-port

49
 Im Im

Ib Ib A Za1 Za2
A B B

A B Zu

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
(ground)
(a) (b) (c)

Figure 2.13: Center-fed dipole antenna: (a) balanced mode; (b) monopole mode; (c) equivalent circuit.

B
Transmitter
A

(a)

B
Transmitter Za1 Za2
A

ZE Zu

(b)

Figure 2.14: (a) Antenna driven by coaxial line; (b) Equivalent circuit seen by TEM mode of coax.

rather than a one-port network, and the parameters of this two-port depend on the relationship of the
antenna to its environment in a way that is not easily modeled or predicted, especially if the monopole
mode of operation is produced.
If a shielded transmission line such as a circular coaxial cable is used to feed the antenna (in order
to avoid interference from stray fields), there is a possibility that the antenna will be driven partially in
the undesirable monopole mode. This is because currents flowing on the exterior of the coaxial shield
interact with the external environment in a complicated way, and present an equivalent impedance ZE
connected from the shield side of the coax to the ground (there is also a generally different impedance
ZEB seen between the inner conductor of the coax and ground, but in practice this is so large as to
be negligible by comparison). The driven antenna of Fig. 2.14(a) thus presents the equivalent circuit of
Fig. 2.14(b). Unless the partial antenna impedances Za1 and Za2 are small compared to Zu and ZE , we
readily see that the excitation of the antenna is significantly unbalanced: current will flow through the
impedance Zu representing the monopole mode of the antenna. A means of compensating for this effect
must be found, and a popular one is the balun, which is described in the next section.

50
 I1

+
V
- 1 IN

...
I1 VN +
-

IN
Vn
...
In + - In

Figure 2.15: An N -port network.

2.2 Multiport Networks

Networks with three or more ports (Fig. 2.15) can also be characterized by the [Z], [Y ] or [S] matrices.
Their definitions are direct generalizations from the two-port case: eqns. (2.3), (2.9) and (2.19), where
now [V ], [I], [b] and [a] are N -element column vectors in the case of an N -port network. Again, [Z], [Y ]
and [S] are symmetric if the network is reciprocal. In addition, eqns. (2.22)-(2.30) relating [Z] to [S]
are still valid. A generalization of the equivalent Π-network of Figure 2.3 holds for sourceless reciprocal
N -ports, and is shown in Figure 2.16.

2.2.1 Example: Analysis of baluns

Many applications in RF and microwave circuit design call for a balanced current flow or voltage balance
at two terminals of a network. Driving a two-terminal antenna in a balanced manner as described in
section 2.1.6 above is one such example; as a second example, mixers often employ balanced circuit
topologies in order to minimize noise and spurious signal generation. Due to the physical environment,
however, the two terminals in question cannot usually be considered as a single port. There is a third
“terminal” present (the “ground”) which together with the other two terminals forms a two-port, rather
than a one-port network from the original pair of terminals. A balun (bal anced to unbalanced converter)
is a network designed to accomplish the task of guaranteeing the balance of voltage or current at a pair
of terminals in a circuit. More specifically, a balun is a three-port network designed so that a source
connected to port 1 always results in balanced currents or voltages at ports 2 and 3 (that is, I2 = −I3
or V2 = −V3 ), regardless of the properties of the two-port connected to ports 2 and 3, or of any other
external connections. It is usually desirable for the balun to be lossless, and convenient (though not
necessary) for it to be reciprocal.
We will examine first the example of a current balun, used primarily to ensure the balance of currents
in antennas. Consider the case of a two-port antenna as considered before, driven through a balun as
shown in Fig. 2.17. Port 1 of the balun consists of terminals 1 and 0, port 2 of terminals 2 and 0, and port
3 of terminals 3 and 0. We demand that no current flow through the “stray” impedance Zst = ZE + Zu ,
regardless of its value, and those of Za1 and Za2 . We assume for simplicity that the balun is reciprocal.
If the balun has an admittance matrix [Y ], then from the current balance condition I2 + I3 = 0 we find

0 = I2 + I3 = (Y12 + Y13 )V1 + (Y22 + Y23 )V2 + (Y33 + Y23 )V3 (2.56)

For this to happen regardless of the values of the antenna impedances (and therefore of the values of V1 ,

51
 -Y12
port 1 port 2
Y11 +Y12 + ... Y1N Y21 +Y22 + ... Y2N

.
-Y2n .
-Y1n .
port n
Yn1 +Yn2 + ... YnN

Figure 2.16: Equivalent circuit for a sourceless reciprocal N -port network in terms of its Y -parameters
(not all elements shown).

I1 I2 Za1
+ 1 2
V1
- 0 Balun 3
I3 Za2

Zst

Figure 2.17: Two-port load driven through a balun network.

52
 1

-Y12 Y12
Y22
Y11 2 3

Y12 -Y12

Figure 2.18: The Boucherot bridge (or lattice) balun.

V2 and V3 ), we must have

Y13 = −Y12 and Y22 = Y33 = −Y23 (2.57)
A lumped element realization of a current balun can be obtained by applying conditions (2.57) to
the equivalent circuit of Fig. 2.16. The resulting circuit can be drawn as shown in Figure 2.18. The
characteristic bridge structure comprising two elements of Y12 and two of −Y12 is called a Boucherot
bridge. Of course, if the balun is a passive structure, we must have Y12 = jB. In a lossless structure, ±B
would be, for example, a pair of inductors and a pair of capacitors. They could also be realized by short-
circuited or open-circuited transmission-line stubs. A wide variety of lumped-element or transmission-line
baluns can be realized in this way.
The elements Y11 and Y22 do not affect the balun operation, and can be used to achieve impedance
matching. Because a lossless, reciprocal three-port network cannot be matched at all ports (i. e., cannot
have S11 = S22 = S33 = 0; see section 2.3), we must expect Y11 and Y22 to have nonzero real parts if the
input and output ports are to be matched.
Another type of balun is the voltage balun, often used in, for example, balanced mixer circuits.
One realization of the voltage balun uses a 1:2 ideal transformer as shown in Fig. 2.19. The two load
impedances ZL1 and ZL2 require to have identical voltages appear across them, even if ZL1 6= ZL2 . An
additional shunt susceptance Bs connected across the secondary transformer winding does not affect the
balancing action of the circuit nor introduce losses. The multiport circuit formed by terminals 1-0, 2-0
and 3-0 does not possess an impedance matrix, but does enforce V2 = −V3 = V1 : the balun doubles the
voltage from port 1 to that appearing across terminals 2 and 3 while maintaining the balanced voltage
condition.
The total balanced-mode impedance ZL = ZL1 + ZL2 of the load is transformed by the voltage balun
into an impedance of
ZL /4
Zin,1 =
1 + jBs ZL
seen looking into port 1. If Zin,1 is to be real (= Rin,1 ), then Bs must be chosen to impedance-match
the balanced-mode load impedance ZL = RL + jXL :
XL
Bs =
|ZL |2
whence
|ZL |2
Rin,1 =
4RL

53
 1
2
ZL1

Zin,1
Ys = jBs

ZL2
1:2 3
0

Figure 2.19: Balun using ideal transformer.

The balun thus performs an impedance transformation in addition to its primary function. If XL = 0
and thus Bs = 0, the transformation ratio is 4:1. A 300 Ω load impedance, for example, can be matched
to a 75 Ω coaxial transmission line using this type of balun. These are widely used in television antenna
applications.
At frequencies above a few megahertz, the performance of a traditional low-frequency transformer
will begin to degrade, and in practice we must use instead a loop of transmission line (such as coaxial
cable) to achieve this purpose as described in section 2.1.5. The resulting circuit is shown in Fig. 2.20.
Naturally, field fringing effects at the coax junctions which have been neglected here will modify these
results to some extent. In chapter 11 we will see how such effects are included by the use of appropriate
equivalent circuits. We may then use them to refine this simplified circuit model of the balun.

2.2.2 Example: Ring and hybrid networks

A hybrid network is a four-port network in which, when matched loads are placed at each port, an
incident wave at one port is split in some ratio and delivered as outgoing waves at two of the remaining
ports, while there are no outgoing waves at the incident port or the last remaining (“isolated”) port.
Hybrid networks have applications as duplexers: three-port networks which transfer all power from
one port to a second port, and all power incident at the second port to a third.1 To achieve this
functionality in a lossless three-port requires a nonreciprocal network (called a circulator); however, a
reciprocal four-port hybrid can be used for this purpose if the fourth port is terminated in a matched
load. Hybrid networks are also used in power combiners and repeater amplifier circuits to avoid feedback
which results in oscillation. The most general analysis of a hybrid network is quite complicated. We will
simplify matters by considering a special case.
One way of realizing a lossless hybrid network is a ring network as shown in Figure 2.21. The
admittances YA , YB and YD are imaginary. The analysis of this network can be simplified by the use of
even and odd excitations. We start by assuming a symmetric (even) excitation: V1 = V4 and V2 = V3 .
The symmetry of the network implies that no current passes through the symmetry plane, and thus the
bottom half of the network can be removed and open circuits left at the symmetry plane positions. The
result is a cascade of three two-port networks consisting of: (i) YA in shunt, (ii) YB in series and (iii) YA
in shunt again. The chain matrices of these three two-ports (from eqns. (2.52) and (2.53)) are multiplied

1 A duplexer is different than a diplexer, which is a three-port network that takes a signal incident at one port, and

transfers that signal to a second port over a certain frequency range while delivering the signal to a third port at all other
frequencies. A VHF-UHF splitter for television cables is an example of a diplexer. A device that splits a signal into more
than two frequency ranges is called a multiplexer.

54
 xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx 0
xxxxxxxxxxx
xxxxxxxxxxxxxxxxx
xxxxxxxxxxx x
xxxxxxxxxx
xxxxxxxxxxx
xxxxxxxx
xxxxxxxxxx
xxxxxxxxxxx x
x
xxxxxxxx
xxxxxxxxxx
xxxxxxxx
xxxxxxxxxx
x x
x
xxxxxxxx
xxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxx
xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x 1
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxx
xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx 2
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x ZL1
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
x
x
x
ZL2
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx x
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxx 3
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxx
xxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxx

Figure 2.20: Coaxial loop balun.

YA YB YA
1 2
YD
Symmetry
YD
Plane
3
4
YA YB YA

Figure 2.21: Lumped-element hybrid network.

55
together to get an overall chain matrix:
       
1 YA 1
 A B   1 0  1 YB  1 0   1+ YB YB 
 =   =  (2.58)
         
YA YA
C D YA 1 0 1 YA 1 YA 2 + YB 1+ YB

Terminating one port of this network with the characteristic impedance Zc = 1/Yc and using (2.58) and
(2.54), we find the input impedance at the other port to be

YA + YB + Yc
Zin,1 = (2.59)
YA (YA + YB + Yc ) + YB (YA + Yc )

from which we can obtain the even-mode voltage reflection and transmission coefficients ρe and τe . In
particular, we have
Zin,1 − Zc Yc2 − YA2 − 2YA YB
ρe = = 2 (2.60)
Zin,1 + Zc YA + Yc2 + 2YA YB + 2YB Yc + 2YA Yc

In a similar manner, we can use odd excitation, V1 = −V4 and V2 = −V3 , to arrive at odd mode
reflection and transmission coefficients ρo and τo . The equivalent circuit is the same as that of even
mode excitation, except that the shunt admittances YA are replaced by YA + 2YD .
We demand that the scattering parameters of this reciprocal network obey S11 = S22 = S33 = S44 = 0
and S23 = S14 = 0 in order to provide the desired hybrid behavior described above. This requires that
ρe = ρo = 0 if the same characteristic impedance reference Zc is used at all four ports. If ρe = 0, we
must have
Yc2 = YA (YA + 2YB ) (2.61)

while if ρo = 0, we need
Yc2 = (YA + 2YD )(YA + 2YB + 2YD ) (2.62)

Taken together, (2.61) and (2.62) lead either to the constraints

YA + YB + YD = 0; Yc2 = −YB2 + YD2 (2.63)

or to
YD = 0; Yc2 = YA (YA + 2YB ) (2.64)

In the case of (2.64), the circuits on opposite sides of the symmetry plane are completely decoupled,
and we arrive at S31 = S42 = 0, a trivial situation. We thus consider only the case (2.63), for which the
fraction of the power incident at port 1 that is delivered to port 2 (sometimes called the coupling factor)
can then be shown to be
Y2
|S21 |2 = 2 c 2 (2.65)
Yc − YD

(which is < 1 for a lossless network since Yc is real while YD is imaginary). Thus, a hybrid network
can be designed to provide any desired degree of power division between the coupled ports by a suitable
choice of the admittance elements in the circuit.
It is not necessary to use lumped elements to realize this hybrid network. A set of transmission lines
can be connected into a ring network, and replacing them with their equivalent Π-networks reduces the
ring to the lumped element form of Figure 2.21. At higher frequencies this is almost always preferable
to the use of lumped elements. Details will be left as an exercise.

56
2.3 Power and energy relations
An impedance Z, connected for example as a load on a classical transmission line, can be considered as
a passive one-port network with possibly many lumped circuit elements or even sections of transmission
line internal to it. Suppose that such a network consisting of lumped elements only (all capacitances
and conductances are located in tree branches, while all resistances and inductances are in the links) is
connected to the end of a transmission line at the point z = z1 . From the general properties of such
networks as outlined in Appendix E, we can deduce the following relation between the complex power
delivered to the port and the internal stored energies and dissipated powers in the network (see (E.12)):
V1 I1∗
= pL,av + 2jω (wM,av − wE,av )
2
where V1 = V (z1 ) and I1 = I(z1 ) are the port voltage and current, while
1
[Iint ]† [R][Iint ] + [Vint ]† [G][Vint ]

pL,av =
2
is the time-average power lost to the circuit conductances and resistances, and
1 1
wE,av = [Vint ]† [C][Vint ]; wM,av = [Iint ]† [L][Iint ]
4 4
are the time-average electric and magnetic stored energies in the circuit capacitances and inductances
respectively. Here the column vector [Iint ] denotes the internal link currents and the column vector [Vint ]
denotes the internal tree voltages of the one-port network to distinguish them from the port voltages and
currents. The matrices [R], [L], [G] and [C] are constituents of the branch admittance matrix [G]+ jω[C]
and link impedance matrix [R] + jω[L] of Appendix E.
If Z = Z11 = V1 /I1 is the impedance seen looking into the one-port, then
2
Z11 = (pL,av + 2jωWexcess ) (2.66)
|I1 |2
where Wexcess = wM,av − wE,av is the excess of the time-average stored magnetic energy over the time-
average stored electric energy. Note that it is purely a matter of whether wM,av > wE,av or wM,av < wE,av
that determines whether the imaginary part X11 of Z11 (i. e., the reactance) of the one-port is positive
(inductive) or negative (capacitive). Thus, any equivalent circuit that we construct for the one-port, while
it may result in different total electric and magnetic stored energies, must exhibit the same difference
between them for a given port voltage or current.
As with one-port circuits, the capacitive or inductive nature of two-port sourceless networks is de-
pendent upon whether predominantly electric or magnetic energy is stored in the junction. It is more
convenient to express the general relations for this in terms of the scattering matrix, rather than [Z] or
[Y ]. We start by applying the expression (E.12) for complex power flow to a network as in the steps
leading up to (2.66). In terms of port voltages and currents, the complex power delivered to the network
through its two ports is now
1 1
(V1 I1∗ + V2 I2∗ ) = [I]† [V ]
2 2
† ∗ T
where [I] = [I ] is the Hermitian conjugate of [I]. In place of (2.66) we now have
1 †
[I] [V ] = P + 2jωWexcess (2.67)
2
where P is the sum of the powers dissipated in resistances or conductances of the network, while Wexcess
is the excess of time-average stored magnetic energy over stored electric energy in the volume V . The
network is said to be passive if P ≥ 0. Using the wave amplitudes [a] and [b] defined in (2.21), we have
{[a] − [b]}† {[a] + [b]} = P + 2jωWexcess (2.68)

57
Introducing the scattering matrix [S] from (2.19), we have from (2.68):

[a]† [1] + [S] − [S]† − [S]† [S] [a] = P + 2jωWexcess


(2.69)

Separating real and imaginary parts, we get

[a]† {[1] − [S]† [S]}[a] = P (2.70)

and
[a]† {[S] − [S]† }[a] = 2jωWexcess (2.71)
For a passive network, (2.70) implies

[a]† [S]† [S][a] ≤ [a]† [a] (2.72)

for any possible value of [a]; we write this condition more succinctly as

[S]† [S] ≤ [1] (2.73)

If a two-port is lossless, then P = 0 in (2.70), and we must have

[S]† [S] = [1] (2.74)

A matrix which obeys (2.74) is said to be unitary. A consequence of (2.74) and relation (2.29) is that

[Z̄] = −[Z̄]† (2.75)

For a reciprocal network, the matrix [Z̄] is symmetric, and eqns. (2.27) and (2.75) imply that all the
elements of the normalized impedance matrix are imaginary (reactive). If the characteristic impedance
matrix is purely real, this conclusion holds for the elements of [Z] as well. This is only natural for a
network (and transmission lines) with no losses present. On the other hand, putting ai = 0 for all i 6= n,
but an 6= 0 in (2.71), we get
ω
Im(Snn ) = Wexcess (2.76)
|an |2
We see that Im(Snn ) is has the same sign as Wexcess .
Relations (2.70) and (2.71) are true for any choice of [a], but for general lossy two-port networks they
do not yield simple relations between elements of [S] and the excess energy Wexcess . An exception to
this is the case of scattering from a single shunt impedance element Z12 connected across a transmission
line, in which case the elements of [Z] are all equal: Z11 = Z12 = Z22 . The scattering matrix for such a
shunt impedance can be shown from (2.29) to have the form
 
1  −Z0 2Z12 
[S] =   (2.77)
2Z12 + Z0  
2Z12 −Z0

if both ports are referred to the same characteristic impedance Z0 . It is clear from (2.77) that Im(S11 ) is
negative or positive accordingly as Im(Z12 ) is negative or positive (so long as Z0 is real). From (2.76), we
see that for this simple shunt element, Im(Z12 ) is positive (inductive) if Wexcess is positive and negative
(capacitive) if Wexcess is negative. In a similar way, for the case of scattering from a single admittance
element −Y12 connected in series in a transmission line (cf. Fig. 2.3), the scattering matrix is
 
1  1 −2Z0 Y12 
[S] =   (2.78)
1 − 2Z0 Y12  
−2Z0 Y12 1

58
Similarly to the previous example, we see that Im(Y12 ) has the same sign as Wexcess .
For a lossless N -port network, relations (2.74), (2.75) and (2.76) are still true. The symmetry of a
[Z], [Y ] or [S] matrix for a reciprocal N -port network means that only N (N + 1)/2 of the N 2 complex
matrix elements are independent of each other. For a lossless network, eqn. (2.74) yields a further
N (N + 1)/2 real, independent equations which the elements of [S] must satisfy. In this way, the number
of parameters we must determine in order to describe completely the behavior of a multiport network
can be considerably reduced. Further reduction is possible if physical symmetries of the network can be
exploited.
As a further example of the power/energy relations, consider the restrictions placed upon a three-port
network that is lossless, reciprocal and matched. The corresponding conditions are (2.74), (2.31) and
S11 = S22 = S33 = 0 respectively. The latter two conditions result in
 
 0 S12 S13 
 
 
[S] = 
 S12 0 S23 
 (2.79)
 
 
S13 S23 0

The diagonal terms of (2.74) yield the equations

1 = |S12 |2 + |S13 |2 = |S12 |2 + |S23 |2 = |S13 |2 + |S23 |2 (2.80)

so we have
1
|S12 | = |S13 | = |S23 | = √ (2.81)
2
But from the off-diagonal terms of (2.74), we have
∗ ∗ ∗
S13 S23 = S12 S23 = S12 S13 =0 (2.82)

which is clearly contradictory to (2.81). Thus, a three-port network that is lossless, reciprocal and
matched is not possible.

59
2.4 Notes and References
Basic network theory and applications for RF and microwave circuits are treated in
Montgomery, Dicke and Purcell (1965);
Collin (1966);
Kerns and Beatty (1967);
Helszajn (1994);
Maas (1998);
Pozar (2005).
Both Kerns and Beatty, as well as Pozar, give tables presenting conversions among the various matrix
representations for multiport networks, of which we have only given a few examples here. More advanced
aspects of network theory are given in
Carlin and Giordano (1964);
Newcomb (1966);
T. Nishide and A. Matsumoto, “Multiport image parameter theory,” Recent Developments
in Microwave Multiport Networks [Monograph Series, Research Institute of Applied Elec-
tricity no. 18]. Sapporo, Japan: Research Institute of Applied Electricity, 1970, pp.
103-183.
Various specific types of multiport networks involving transmission lines (including filters) are discussed
in
Matsumoto (1970);
Karakash (1950);
Matthaei, Young and Jones (1964).
For more information on baluns, consult
King (1965), pp. 220-224;
Dworsky (1979), pp. 215-221;
Reich, Ordnung, Krauss and Skalnik (1953), pp. 204-212;
Weeks (1968);
Balanis (1982), pp. 365-368;
Uchida (1967), section 5.3;
Matsumoto (1970), section X.2;
Sevick (1990);
and
S. Frankel, “Reactance networks for coupling between unbalanced and balanced circuits,”
Proc. IRE, vol. 29, pp. 486-493 (1941).
N. Nagai and A. Matsumoto, “Application of distributed-constant network theory to balun
transformers,” Electron. Commun. Japan vol. 50, no. 5, pp. 114-121 (1967).
N. Nagai, “Theory and experiments on multiwire multiports,” Recent Developments in Mi-
crowave Multiport Networks [Monograph Series, Research Institute of Applied Electricity
no. 18]. Sapporo, Japan: Research Institute of Applied Electricity, 1970, pp. 1-68.
T. Nishide, N. Nagai and A. Matsumoto, “Balance-to-unbalance transformers,” Recent De-
velopments in Microwave Multiport Networks [Monograph Series, Research Institute of
Applied Electricity no. 18]. Sapporo, Japan: Research Institute of Applied Electricity,
1970, pp. 69-85.
N. Nagai, T. Nishide and A. Matsumoto, “Multiwire-line baluns and unbaluns,” Bull. Res.
Inst. Appl. Electricity, vol. 22, pp. 24-51 (1970).

60
 Hu Shuhao, “The balun family,” Microwave J., vol. 30, no. 9, pp. 227-229 (1987).
R. Sturdivant, “Balun designs for wireless, ... mixers, amplifiers and antennas,” Appl. Mi-
crowave, vol. 5, no. 3, pp. 34-44 (1993).
V. Trifunović and B. Jokanović, “Review of printed Marchand and double Y baluns: Char-
acteristics and applications,” IEEE Trans. Micr. Theory Tech., vol. 42, pp. 1454-1462
(1994).

Hybrid networks are treated in

Matthaei, Young and Jones (1964), chap. 13.

Sazonov, Gridin and Mishustin (1982), section 2.10.

and

A. Alford, “High frequency bridge circuits and high frequency repeaters,” U. S. Patent no.
2,147,809, Feb. 21, 1939.
L. Young, “Branch guide directional couplers,” Proceedings of the National Electronics Con-
ference, vol. 12, 1-3 October 1956, Chicago, pp. 723-732.
J. Reed and G. J. Wheeler, “A method of analysis of symmetrical four-port networks,” IRE
Trans. Micr. Theory Tech., vol. 4, pp. 246-252 (1956).
C. Y. Ho, “Some results on the design of N (2, 3, 4)-way branch line type power splitters,”
IEEE 1976 Region V Conference, 14-16 April 1976, Austin, TX, pp. 45-50.
R. G. Manton, “Hybrid networks and their uses in radio-frequency circuits,” Radio Electron.
Eng., vol. 54, pp. 473-489 (1984).
V. F. Fusco and S. B. D. O’Caireallain, “Lumped element hybrid networks for GaAs MMICs,”
Micr. Opt. Technol. Lett., vol. 2, pp. 19-23 (1989).

The paper by Reed and Wheeler and the book by Sazonov et al. describe the even and odd excitation
technique for the analysis of symmetrical multiport networks.

2.5 Problems

p2-1 A complex load impedance, when connected to a lossless transmission line, produces a nonzero
reflection coefficient ρL . Find the most general conditions which must be satisfied by a
passive, lossless, reciprocal two-port network so that when it is connected in cascade with
the load, the transmission line now has a reflection coefficient of zero. Your solution will
indicate that some degree of freedom is possible in the choice of matching network. Give a
particular example of such a matching network, either by specifying its equivalent circuit or
giving its scattering matrix in terms of ρL .

p2-2 Two classical transmission lines of different characteristic impedances and propagation coef-
ficients are connected at z = 0 as shown. Choosing the terminal planes A1 at z = 0− and A2
at z = 0+ , obtain the scattering matrix [S] and the impedance matrix [Z] for this junction,
associating the characteristic impedance Zc1 with port 1 and Zc2 with port 2.

61
p2-3 In the circuit of problem p2-2, a shunt capacitor is added to the circuit at z = 0. Find [S]
and [Z] for this modified circuit.

p2-4 Re-examine the quarter-wave transformer of problem p1-1 if we now account for the effects
of fringing fields at the junctions between line sections of different characteristic impedance.
Suppose that the equivalent circuit is as given below, assume that ωC1,2 Zc1,2,3 ≪ 1, but
is not zero, and find what new values of Zc2 and d must be used in order to preserve the
impedance match condition of the ideal matching circuit?

ρ=0 ρ=0

Zc1 C1 Zc2 C2 Zc3

p2-5 In the circuit of problem p2-2, a series inductor is added to the circuit at z = 0. Find the
[S]-matrix for this circuit.

62
 I1 L I2

+ +
(Zc1, β1) V1 V2 (Zc2, β2)
– –

A1 A2

p2-6 In certain situations, there may arise the need for an “unbalun”—a three-port network for
which I2 = I3 regardless of any external circuits connected to it. If the unbalun is to be
lossless and reciprocal, derive the additional constraints on the Y -parameters necessary to
achieve the desired behavior. Obtain a lumped element circuit that performs as an unbalun.

p2-7 Design a reciprocal three-port network that will force the relation b3 = Ab2 for a given
complex constant A, regardless of any external circuit connections. That is, find a matrix
[Y ], [Z] or [S] that ensures this behavior, and find a lumped element equivalent circuit that
corresponds to this matrix. Determine whether there are any restrictions on the value of A
if the network is to be passive, or if it is to be lossless.

p2-8 From (2.22) or (2.29), find the conditions which must be satisfied by the elements of the
[Z] matrix of a source-free two-port network in order for S11 = 0 to be true. Based on this
condition, and the equivalent circuit of Fig. 2.2, construct a lossless matching network based
on lossless lumped elements (i. e., inductors and capacitors) that will give zero reflection on
a transmission line of characteristic impedance Zc1 connected to port 1, if a matched load of
ZL = Zc2 is connected to port 2. Is it possible to simultaneously match in the other direction
(that is, using the same network, obtain zero reflection on a line of characteristic impedance
Zc2 connected to port 2 with a load impedance of Zc1 connected to port 1)? Assume that
Zc1 and Zc2 are given real quantities.

p2-9 (a) Consider a nonuniform classical transmission line in the lowest order WKB approxima-
tion, as described in section 1.5.1. Show that the scattering matrix for this line in this
approximation is given by
 
−Q(d)
 0 e 
[S] = 



e−Q(d) 0

where the port characteristic impedances are chosen as Zc1 = Zc (0+ ) and Zc2 = Zc (d− ).
Use this result to obtain the [Z] matrix and the chain matrix of this line in the lowest-
order WKB approximation.
(b) Consider this same line in the first-order WKB approximation described in section 1.5.2.
Obtain the [S] matrix, the [Z] matrix and the chain matrix to this higher accuracy.

63
p2-10 A voltage balun is a three-port network for which V3 = −V2 , regardless of external connec-
tions to the network (that is, regardless of the values of I1 , I2 and I3 ). Derive the conditions
which must be satisfied by the [Z] matrix of such a network. Find one example of a circuit
that obeys these conditions (use the result of problem p2-11).
p2-11 Show that a sourceless linear reciprocal three-port network possessing an impedance matrix
representation can be represented by the equivalent network shown below.

Z11+Z12+Z13
– V3 +
+
-Z13 Z13+Z23+Z33

V1
–
-Z12 -Z23
V2
–
+

Z12+Z22+Z23

p2-12 The branch-line hybrid is a four-port device with lengths of transmission line connected
between adjacent ports in the ring arrangement shown in the figure below.

βA, ZcA
1 2

βB, ZcB dB βB, ZcB

Symmetry
Plane
βA, ZcA
3
4
dA

(a) Use the equivalent Π-network for each transmission-line section given in Fig. 2.10(c) to
reduce this network to the lumped-element hybrid shown in Fig. 2.21. Express YA , YB
and YD in terms of the transmission-line parameters ZcA , ZcB , βA dA and βB dB .
(b) Write the constraints (2.63) in terms of the transmission-line parameters. If a 3 dB
hybrid is desired (|S21 |2 = 0.5), express the condition resulting from (2.65) in terms of
the transmission-line parameters.
(c) Obtain one possible design for a 3 dB branch-line hybrid satisfying the constraints found
in part (b). [Hint: dA = λA /4 and dB = λB /4 results in a well-known version of this
device.]

64
p2-13 Repeat problem p2-12, but do parts (b) and (c) for the case of an arbitrary value of |S21 |.
Assume that dA = λA /4.

65
66
Chapter 3

CLASSICAL TRANSMISSION
LINES: EXCITATION, COUPLING
AND RESONATORS

. . . I have been carrying on a fight with Monopolated

Light & Power for some time now. I use their service
and pay them nothing at all, and they don’t know it.
Oh, they suspect that power is being drained off, but
they don’t know where. All they know is that
according to the master meter back there in their
power station a hell of a lot of free current is
disappearing somewhere into the jungle of Harlem.
The joke, of course, is that I don’t live in Harlem but
in a border area.
—Ralph Ellison,
Invisible Man

3.1 Induced and Impressed Currents and Voltages; Equivalent

Circuits
It is well to be clear about terminology when we talk about sources, either in analysis of circuits or
of classical transmission lines. Let us state what we mean by the term “externally generated” sources.
In circuit theory, we make use of ideal voltage and current sources to simplify analysis and conceptual
understanding of the subject. These sources are regarded as capable of sustaining voltages or currents
we can “control”, whereas other voltages and currents at different places in the circuit (or even the
current through a voltage generator, for example) are subject to conditions that prevail in the rest of
the circuit. Of course, no such ideal sources can be achieved in practice because the circuit into which
a real source is connected will always affect it (load it down) to some degree. No real voltage source
produces constant voltage regardless of what load is connected to it (especially if that load is a short
circuit!). Nevertheless, we find the concept of an ideal generator a very useful one in the analysis of
networks. We say that such a generator is an impressed source. Currents or voltages which disappear
when the impressed sources are turned off are referred to as induced currents or voltages.

67
3.1.1 Thévenin and Norton equivalent circuits
It is often the case in circuit analysis that a network can be partitioned into two sections, one of which
can be replaced by a different subnetwork without affecting the voltages or currents in the other. The
two alternative subnetworks are said to be equivalent to each other as far as their effect on the second
subnetwork is concerned. Examples of such equivalence are Thévenin and Norton equivalent circuits.
Let us pursue this idea a little further than is usually done in introductory expositions of circuit theory,
at the same time providing a connection with the ideas of impressed and induced sources.
Consider an active multiport network (that is, one which may contain internal impressed sources
in addition to passive components such as resistors, capacitors, etc.). If the network is linear and the
impedance or admittance matrix representation exists when the internal sources are absent, it is possible
to relate the port voltages and currents in at least one of the following forms:

[V ] = [Z][I] + [V ext ] (3.1)

or
[I] = [Y ][V ] − [I ext ] (3.2)
where the impressed sources are related by [V ext ] = −[Z][I ext ]. We observe that if the internal sources
are deactivated (i. e., current sources replaced by open circuits and voltage sources by short circuits),
then the terms [V ext ] and [I ext ] must disappear from these expressions, and we recover the impedance
and admittance matrix descriptions of the resulting sourceless multiport. On the other hand, [V ext ]
represents the voltages appearing at the ports when the ports are open-circuited ([I] = 0) and −[I ext ]
the currents that flow when the ports are short-circuited. Thus, (3.1) and (3.2) are, in essence, statements
of generalized Thévenin and Norton equivalence theorems, respectively. That is, any such network can be
replaced by an equivalent circuit consisting of a sourceless multiport (characterized by the matrix [Z] or
[Y ]), with additional ideal impressed voltage sources in series with each port, or ideal impressed current
sources in parallel, as shown in Fig. 3.1. The voltages and currents induced in any circuitry connected

ext
I1 V1

+ –
+V ext
1 VN IN
–
– +
VN +
–

...
(a)
I1

+ IN
V1 ext
I1
– +
ext
IN VN
–

...
(b)

Figure 3.1: Thévenin (a) and Norton (b) equivalent circuits for an active multiport.

68
external to the multiport will remain the same (though of course internal voltages and currents will
generally be different).

3.1.2 Huygens (wave) sources

Other kinds of equivalence principles can be stated as well. Suppose (Fig. 3.2(a)) that some source
network drives a passive load ZL as shown. A current I and a voltage V are induced at the terminals of

+
V ZL (load) (a)
–

Huygens source

V
– +
I I +
Zs ZL V (b)
–

Figure 3.2: (a) Original network; (b) Equivalent impressed Huygens source (V , I) with arbitrary passive
impedance Zs .

the load by this generator, although they are not independent (since the load forces the relation V /I = ZL
to hold). Now let the sources within the source network be turned off, and impressed generators for V
and I be inserted as shown in Fig. 3.2(b). Elementary circuit theory shows that the same voltage and
current (V , I) appear at ZL , while the generator impedance Zs to the left has neither voltage across it
nor current through it, and therefore may have any value at all (not necessarily the Thévenin equivalent
impedance of the original source network). The presence of the element Zs is thus now irrelevant to the
voltage and current being produced at ZL . The combination of the sources V and I is called a Huygens
or wave equivalent source, its characteristic property being that it produces a response to the right (at
the load), but none to the left in what remains of the original network.1
Once V and I have been made impressed sources, however, we can adjust Zs to have any convenient
value without affecting the current or voltage at the load. If we set Zs = 0, the current generator is now
shorted out and can be removed (that is, its removal makes no difference to the current induced in ZL ,
which is now assured by the voltage generator). As far as ZL is concerned, then, the circuit of Fig. 3.2(a)
is completely equivalent to that of Fig. 3.3(a). In this case, the current flow I is an induced one, and
passes through the voltage source rather than the current source. The voltage V remains an impressed
one, and the load continues to “see” the same conditions it has throughout all the modifications of the
source network on the left. Conversely, we could make Zs equal to ∞ (an open circuit) and remove the
voltage generator, leaving the situation of Fig. 3.3(b). In this case the voltage is induced, the current is
impressed, and the load conditions continue to be the same.
This equivalence principle is an application of what is known in circuit theory as the substitution
theorem. Unlike the Thévenin and Norton theorems, the equivalent sources used now depend on values
1 Note that the direction of the current source is opposite to that used for the Norton equivalent sources in the previous

subsection. This is due to the conventional direction for positive current flow at the port of a network.

69
 V
- +
+
I ZL V (a)
-

I + (b)
I ZL V
-

Figure 3.3: (a) Zs made equal to zero, impressed V , induced I; (b) Zs made equal to infinity, impressed
I, induced V .

of induced currents or voltages in the original circuit. Since these are dependent on the nature of the
external circuit (here the load impedance), the resulting equivalent circuit is not the same if the external
circuit is changed. The implications of this in circuit theory are perhaps not as profound as are those
of the analogous equivalence principles in field theory, to be examined in chapter 10. Nevertheless, this
discussion shows that there are many equivalent impressed source arrangements capable of inducing
the same voltages and currents in a given portion of a circuit—including one that contains sections of
classical transmission line.
One such important application is the excitation of a wave in a single direction only on a transmission
line. Consider the Huygens source connected in an infinite transmission line as shown in Fig. 3.4. If

Huygens source

V0 a
– +
Zc Zc +
I0 V(z)
–

Figure 3.4: Huygens source in a transmission line.

the voltage source and current source are chosen such that the relation V0 = Zc I0 holds, then there will
be a forward wave produced to the right of the source (since the semi-infinite section of line presents
an impedance of Zc ), but nothing is produced to the left of the source. By (1.33), the forward wave
amplitude at a point just to the right of the wave source will be
V0
a = aext = √
2Zc

70
and we regard aext as the strength of this wave source. Accordingly, we will represent such a source
schematically as in Fig. 3.5. If we had chosen V0 = −Zc I0 instead, the wave source would have produced

Huygens source

a
aext
Zc Zc +
V(z)
–

Figure 3.5: Schematic representation of a Huygens source in a transmission line.

nothing to the right, but a reverse wave amplitude

V0
b = bext = − √
2Zc
to its left. We denote this source schematically as shown in Fig. 3.6.

Huygens source
b

bext +
Zc Zc V(z)
–

Figure 3.6: Schematic representation of a reverse Huygens source in a transmission line.

Now let us place both [V ext ] and [I ext ] at the ports of a network. Instead of (3.1) or (3.2), we now
have
[V ] + [V ext ] = [Z] [I] + [I ext ]


(3.3)
Defining the impressed wave sources [aext ] and [bext ] to be [compare (2.21)]:
1
[aext ] = − [2Zc ]−1/2 [V ext ] + [Zc ][I ext ]

2
1
[b ] = − [2Zc ]−1/2 [V ext ] − [Zc ][I ext ]
ext

(3.4)
2
let us also√suppose that [V ext ] = −[Zc ][I ext ] so that [aext ] = 0. Then [V ext ] = [2Zc ]1/2 [bext ] and
[I ext ] = − 2[Zc ]−1/2 [bext ] combine to form a wave source which launches waves only away from the
multiport as shown in Fig. 3.7, and represent the equivalent effect of sources inside the original multiport
without introducing spurious incident waves traveling toward it. The resulting relation between [a] and
[b] is then:
[b] = [S][a] + [bext ] (3.5)

71
 ext
b1
a1 ext
bN
b1 aN
bN

...

Figure 3.7: Wave sources for an active multiport.

3.2 Excitation of a Classical Transmission Line

Consider the excitation of a uniform classical transmission line in terms of the distributed constant
model of section 1.2. In place of the purely sourceless network model for an incremental length ∆z of
the line as shown in Fig. 1.3, we now suppose there are impressed series voltage generators and shunt
current generators distributed in a continuous manner along the line. Our network model is as shown
in Fig. 3.8 (note that shunt voltage sources and series current sources are not used, as they would force
the line voltage and current to specified values, rendering the problem trivial). Following the derivation

I(z)
z∆z ∆Vext I(z+∆z)
- +
+ +
V(z) y∆z V(z+∆z)
∆Iext
- -

z z+∆z

Figure 3.8: Distributed source excitation of a classical transmission line.

of Section 1.2, we can apply Kirchhoff’s laws to Fig. 3.8, and take the limit as ∆z → 0. We obtain the
telegrapher’s equations with distributed sources:

dV (z)
= −zI(z) + Eext (z)
dz
dI(z)
= −yV (z) + Hext (z) (3.6)
dz
where
∆V ext
Eext (z) = lim
∆z→0 ∆z
∆I ext
Hext (z) = lim (3.7)
∆z→0 ∆z

are the series voltage and shunt current sources per unit length impressed at each point along the line.
We give them a notation suggestive of field quantities because their units (volt/m and amp/m) are those

72
of electric and magnetic fields respectively. However, no physical interpretation as actual electric or
magnetic fields is necessarily implied here.
Now, from here on, we assume that the transmission line is uniform (z and y are constant). Further,
the impressed sources Eext and Hext are assumed to be zero outside some interval z1 < z < z2 . Therefore,
for z ≥ z2 , we expect only a forward wave to be present:
p
V (z) = Zc I(z) = 2Zc a0 e−γz (z ≥ z2 ) (3.8)

To the left of the sources, we have only a reverse wave:

p
V (z) = −Zc I(z) = 2Zc b0 eγz (z ≤ z1 ) (3.9)

In other words, V and I are totally produced by the localized sources, and there are no incoming waves
from infinity.
We can solve (3.6) by the method of variation of parameters. We introduce the variable wave
amplitudes a0 (z) and b0 (z) by generalization from the constant values defined in chapter 1 in terms of
V (z) and I(z):
p
V (z) = [a0 (z)e−γz + b0 (z)eγz ] 2Zc
r
−γz γz 2
I(z) = [a0 (z)e − b0 (z)e ] (3.10)
Zc
or,

1 γz V (z) + Zc I(z)
a0 (z) = e √
2 2Zc
1 −γz V (z) − Zc I(z)
b0 (z) = e √ (3.11)
2 2Zc

Differential equations for a0 and b0 are obtained by substituting (3.10) into (3.6), and using (1.8) and
(1.9). We obtain

1 γz Eext (z) + Zc Hext (z)

a′0 (z) = e √
2 2Zc
1 −γz Eext (z) − Zc Hext (z)
b′0 (z) = e √ (3.12)
2 2Zc

These can be readily integrated to obtain a0 and b0 , if appropriate initial conditions are available.
The boundary conditions are found from (3.8) and (3.9). We should have


a0 (z) ≡ 0 for z ≤ z1 

(3.13)

b0 (z) ≡ 0 for z ≥ z2 


Then, integrating (3.12) yields

z
1 Eext (z ′ ) + Zc Hext (z ′ ) γz′ ′
Z
a0 (z) = √ e dz (z > z1 ) (3.14)
2 z1 2Zc

and z2
1 Eext (z ′ ) − Zc Hext (z ′ ) −γz′ ′
Z
b0 (z) = − √ e dz (z < z2 ) (3.15)
2 z 2Zc

73
 |a0(z)|
or
|b0(z)|
|a02|

|a0(z)|

|b01|
|b0(z)|

z1 z2 z

Figure 3.9: Forward and reverse mode amplitudes as functions of z.

Note that since Eext and Hext are zero for z ≤ z1 and z ≥ z2 , a0 (z) becomes a constant a02 for z ≥ z2
and b0 (z) the constant b01 for z ≤ z1 , as stated earlier in (3.8) and (3.9). In fact,

1 z2 Eext (z ′ ) + Zc Hext (z ′ ) γz′ ′

Z
a02 = √ e dz
2 z1 2Zc
1 z2 Eext (z ′ ) − Zc Hext (z ′ ) −γz′ ′
Z
b01 = − √ e dz (3.16)
2 z1 2Zc

Thus, the forward-mode amplitude gradually changes from zero (at z1 ) and becomes constant after
passing z2 . The opposite occurs for a reverse-mode amplitude, as shown in Fig. 3.9. The constancy of
a0 (z) and b0 (z) in source-free regions is also immediately obvious from (3.12). We see that (3.16) are
the distributed analog of Huygens sources.
For a line with no attenuation (γ = jβ), the amplitude of the forward wave a02 can be maximized
relative to that of the reverse wave b01 in a variety of ways. The most obvious is to use both distributed
voltage and current sources such that Eext (z) = Zc Hext (z). If only one type of source (Eext (z), say)
is available, a02 can still be enhanced by matching the source’s phase progression in z with that of the
desired wave, e−jβz :
Eext (z) = (const)e−jβz
The integral for a02 in (3.16) increases linearly as the source region is increased, since the integrand
is constant. The integral for b01 on the other hand, has a highly oscillatory integrand, which upon
integration tends to cancel itself resulting in a much lower magnitude than for a02 . The reverse choice
of source phasing can be used to enhance the excitation of the reverse wave.

3.2.1 Lumped sources

Ordinary lumped voltage and current sources can be modeled in the language of the distributed sources
Eext and Hext . However, we need to employ a Dirac delta-function to do so. Suppose a series voltage
source V0 and shunt current source I0 are connected to the line at a point z = z0 as shown in Fig. 3.10.
From Kirchhoff’s laws at z0 , the currents and voltages on either side of z0 will be related by

74
 V0
– +
+ +
Zc V(z0–) I0 V(z+0) Zc z
– –

z = z0

Figure 3.10: Lumped sources at z = z0 .

V (z0+ ) − V (z0− ) = V0 (3.17)

I(z0+ ) − I(z0− ) = I0 (3.18)

On the other hand, if we integrate both sides of (3.6) from z0− to z0+ , we have
Z z0+
V (z0+ ) − V (z0− ) = Eext (z) dz
z0−
Z z0+
I(z0+ ) − I(z0− ) = Hext (z) dz (3.19)
z0−

since I and V suffer (at worst) step discontinuities across z0 . Comparing (3.19) with (3.17) and (3.18),
we find that the lumped generators can be expressed as

Eext (z) = V0 δ(z − z0 )

Hext (z) = I0 δ(z − z0 ) (3.20)

Thus, the description of eqn. (3.6) is capable of describing lumped excitations as well as distributed
ones.
As discussed in the previous section, we sometimes find it convenient to use the concept of equivalent
sources in circuit or transmission line problems. If we are not interested in the behavior of V (z) and
I(z) within the source region z1 < z < z2 , then it is possible to replace the actual sources with other,
equivalent ones inside the source region which produce the same V and I exterior to it. One possibility
would be the use of a Thévenin or Norton equivalent circuit for a section of line excited by distributed
sources. Another is to replace the sources by lumped sources V0 and I0 located at z0 within the source
region as shown in Fig. 3.10.
If z0 is a point in the interval [z1 , z2 ], let us require that V (z) and I(z), or what is the same thing,
a0 (z) and b0 (z), be identical outside of [z1 , z2 ] whether the original or the equivalent sources are used.
Now for the original sources, a0 (z) and b0 (z) are given by (3.13) or by the constants (3.16) as appropriate.
With the equivalent sources (3.20), eqn. (3.13) remains true, while (3.16) becomes

1 γz0 V0 + Zc I0
a02 = e √
2 2Zc
1 −γz0 V0 − Zc I0
b01 = − e √ (3.21)
2 2Zc

Thus, an equivalent circuit for Fig. 3.10 can be constructed using wave sources, as shown in Fig. 3.11.

75
 Zc b0 a0 Zc

z = z0

Figure 3.11: Lumped wave sources at z = z0 .

For (3.21) to be the same as (3.16), the equivalent sources must be given by
Z z2
V0 = [Eext (z) cosh γ(z − z0 ) + Zc Hext (z) sinh γ(z − z0 )] dz
z1
Z z2
Eext (z)
I0 = [Hext (z) cosh γ(z − z0 ) + sinh γ(z − z0 )] dz (3.22)
z1 Zc

If the source region is electrically small, |γ(z2 − z1 )| ≪ 1, then cosh ≃ 1 and sinh ≃ 0, so that
Z z2
V0 ≃ Eext (z) dz
z
Z 1z2
I0 ≃ Hext (z) dz (3.23)
z1

which is consistent with the delta-function (lumped-source) idealization (3.20) pictured in Fig. 3.10. This
concept of equivalent source representation turns out to have important application in the treatment of
coupled lines as well as for the excitation itself.

3.3 Multiconductor Transmission Lines

When two or more parallel transmission lines are placed in proximity to each other (such as adjacent
printed circuit transmission lines on a common substrate, or the conductors in a “ribbon cable” inter-
connecting circuit boards in a computer), coupling will arise between the lines. This may either be a
desired effect, which can be used to achieve actions such as directional coupling, or undesired, in which
case the phenomenon of crosstalk between the transmission line circuits arises. In this section, we give an
overview of the behavior of coupled transmission lines, leaving specific configurations to a later chapter.

3.3.1 Matrix telegrapher’s equations

Consider two coupled lines as shown in Fig. 3.12. One of the conductors is regarded as the “ground”
conductor, and the voltages V (1) and V (2) at any point z along the system are measured from each of
the other conductors to this reference. Positive currents I (1) and I (2) are defined on each conductor in
the axial direction z. Note that the conductor number is denoted here by a superscript, rather than a
subscript. We will at first limit ourselves to the lossless case in order to focus on the essential concepts.
If conductors 1 and 2, in addition to their shunt capacitances (c11 or c22 ) and series inductances (l11
or l22 ) per unit length, also exhibit a mutual inductance l12 and mutual capacitance c12 per unit length
as coupling between them (see Fig. 3.13), then by a similar procedure as in section 1.2, we can obtain

76
 I2

+
V2
I1 –

+
V1
– z

Figure 3.12: Two coupled transmission lines.

l22 ∆z
V(2)(z) + +
V(2)(z+∆z)
Mutual Inductance
c12 ∆z (c22-c12) ∆z
= l12 ∆z
l11∆z

+ +
V(1)(z) (c11-c12) ∆z V(1)(z+∆z)
- -

Figure 3.13: Distributed network model of coupled transmission lines.

the telegraphers’ equations for this multiconductor transmission line:

dV (1) (z)
= −jωl11 I (1) (z) − jωl12 I (2) (z)
dz
dV (2) (z)
= −jωl12 I (1) (z) − jωl22 I (2) (z)
dz
dI (1) (z)
= −jωc11 V (1) (z) + jωc12 V (2) (z)
dz
dI (2) (z)
= jωc12 V (1) (z) − jωc22 V (2) (z) (3.24)
dz
Notice that we have assumed reciprocity in putting c21 = c12 and l21 = l12 . Note too that the equivalent
circuit for the capacitive part of a length ∆z of this multiconductor line consists of a capacitance
(c11 − c12 )∆z from conductor 1 to ground, a capacitance (c22 − c12 )∆z from conductor 2 to ground, and
a capacitance c12 ∆z from conductor 1 to conductor 2.

77
 Equations (3.24) can be written in the matrix-vector form:

d[V (z)]
= −jω[l][I(z)]
dz
d[I(z)]
= −jω[c][V (z)] (3.25)
dz
where [l] and [c] are given by
 
 l11 l12 
[l] = 


 (3.26)
l12 l22
 
 c11 −c12 
[c] = 


 (3.27)
−c12 c22

and the voltage and current column vectors are:

 
(1)
 V (z) 
[V (z)] = 



(2)
V (z)

 
(1)
 I (z) 
[I(z)] = 



(2)
I (z)

In case there are losses or unconventional aspects to the multiconductor line, we replace jω[l] by [z] and
jω[c] by [y] in (3.25):

d[V (z)]
= −[z][I(z)]
dz
d[I(z)]
= −[y][V (z)] (3.28)
dz
From here on, we will treat the general case unless otherwise stated.
The flow of energy along a multiconductor line is described by a superposition of the powers on each
of the conductor/ground pairs. Thus the complex power P̂ (z) passing the point z is

1  (1)  1
P̂ (z) = V (z)I (1)∗ (z) + V (2) (z)I (2)∗ (z) = [I(z)]† [V (z)] (3.29)
2 2
and the complex oscillatory power is

1  (1)  1
P̂osc (z) = V (z)I (1) (z) + V (2) (z)I (2) (z) = [I(z)]T [V (z)] (3.30)
2 2

Although we have explicitly considered only the case of two conductors with ground, all of what
follows can be applied to the case of a line with N conductors plus ground, simply by considering the
vectors to be N × 1 column vectors and the matrices to be N × N square matrices.

78
3.3.2 Modes of multiconductor lines
Suppose we now search for mode solutions of (3.28):

[V ] = [Vm ]e−γm z ; [I] = [Im ]e−γm z (3.31)

where a subscript is used to indicate the mode index m. Then the system of equations

γm [Vm ] = [z][Im ]; γm [Im ] = [y][Vm ] (3.32)

2
must have a nontrivial solution. If we eliminate [Im ] from (3.32), we see that γm must be an eigenvalue
of the matrix [z][y]; likewise if we eliminate [Vm ], we see that it must also be an eigenvalue of [y][z].
Either [Vm ] or [Im ] respectively is the corresponding eigenvector. If γm 6= 0 (for γm = 0, we have a
degenerate case that requires special treatment, which we will not consider here), we may from (3.32)
obtain [Im ] if [Vm ] is known, or vice versa:
1 1
[Im ] = [y][Vm ]; [Vm ] = [z][Im ] (3.33)
γm γm

For every mode with eigenvalue γm that satisfies (3.32), there will thus also be a corresponding reverse
mode with γ−m = −γm for which we can choose

[V−m ] = [Vm ]; [I−m ] = −[Im ] (3.34)

We adopt the convention that modes denoted by positive values of m are forward modes (as defined at
the end of this subsection, while those with negative m are reverse modes.
The eigenvectors can be normalized as we have done for the single transmission line by requiring that
the complex oscillatory power in a forward mode equals 1:
1
[Vm ]T [Im ] = 1 (m > 0) (3.35)
2
Using (3.33), this normalization condition can also be expressed in either of the forms:

[Im ]T [z][Im ] = [Vm ]T [y][Vm ] = 2γm (m > 0) (3.36)

These eigenvectors can also be shown to obey so-called orthogonality properties. An orthogonality
property can be obtained by premultiplying the first of (3.32) by [In ]T , the second by [Vn ]T , generating
two further equations by interchanging m and n, and combining so as to cancel the terms on the right
sides. The result is
(γm − γn ) [Vm ]T [In ] + [Vn ]T [Im ] = 0


If γm 6= γn , then we have the orthogonality property

[Vm ]T [In ] + [Vn ]T [Im ] = 0 (3.37)

If γm = γn even though m 6= n, it is possible to use Gram-Schmidt orthogonalization procedures (see

section D.2) to ensure that (3.37) holds in that case as well. We will assume from here on that this has
been done. Putting m → −m in (3.37), adding the result to (3.37) itself and using (3.34), we obtain the
somewhat simpler version of orthogonality

[Vm ]T [In ] = 0 (m 6= ±n) (3.38)

The form of eqn. (3.38) suggests the reason that this property is called orthogonality. If two vectors A
and B in space obey the relation A · B = 0, they are perpendicular (or orthogonal) to each other. For
column vectors, the analog of the dot product is the multiplication of the transpose of one by the other

79
as on the left side of (3.38). Additional orthogonality-type properties can be obtained by using (3.33)
in conjunction with (3.38). If γm 6= 0 and γn 6= 0, we get

[Im ]T [z][In ] = 0 (m 6= ±n) (3.39)

and
[Vm ]T [y][Vn ] = 0 (m 6= ±n) (3.40)
The orthogonality and normalization properties (together known as orthonormalization) can be con-
veniently expressed as a single equation if we introduce the matrices
   
(1) (1) (1) (1)
 V1 V2   I1 I2 
   
[MV ] = [V1 ][V2 ] = 

;
 [M I ] = [I1 ][I2 ] = 


 (3.41)
(2) (2) (2) (2)
V1 V2 I1 I2

The normalization property (3.35) and the orthogonality property (3.38) imply that

[MV ]T [MI ] = 2[1] (3.42)

or in other words,
1 1
[MV ]−1 = [MI ]T ; [MI ]−1 = [MV ]T (3.43)
2 2
Complex power flow in a single mode (m) is described in a similar fashion. At z = 0,

1 1 1
P̂m = [Im ]† [Vm ] = [Im ]† [z][Im ] = ∗ [Vm ]† [y]† [Vm ] (3.44)
2 2γm 2γm

and thus the time-average power flow in this mode is

†
   
1 †

1 † [z] 1 † [y]
P̂av = Re [Im ] [Vm ] = Re [Im ] [Im ] = Re [Vm ] ∗ [Vm ] (3.45)
2 2 γm 2 γm

which must be nonnegative for a forward mode. If the attenuation constant αm differs from zero, we
can show that the sign of the propagation constant γm of the forward wave must be chosen such that
αm > 0 in order to achieve P̂av ≥ 0 (see problem p3-12).
A propagating mode (γm = jβm , with βm real) can occur only in the special case when the mul-
ticonductor line is lossless ([z] = j[x] and [y] = j[b], with [x] and [b] real). In this situation, (3.32)
becomes
βm [Vm ] = [x][Im ]; βm [Im ] = [b][Vm ] (3.46)
We see from this that it must then be possible to choose both [Vm ] and [Im ] as real vectors in this case,
and therefore the matrices [MV ] and [MI ] will be real as well. When considering whether a propagating
mode with no attenuation on lossless coupled transmission lines is a forward wave, we choose γm = jβm
such that Re(P̂m ) ≥ 0. The complex power (3.44) now becomes

1 1 1
P̂m = [Im ]T [Vm ] = [Im ]T [x][Im ] = [Vm ]T [b][Vm ] (3.47)
2 2βm 2βm

which is real. The sign of βm is thus chosen so as to make

1 1
[Im ]T [x][Im ] = [Vm ]T [b][Vm ] > 0 (3.48)
2βm 2βm

80
3.3.3 Total voltages and currents on a multiconductor line
Because of the orthogonality and linear independence of the eigenvectors of the modes, we can represent
arbitrary voltages and currents on the multiconductor line as superpositions of mode eigenvectors. For
example, if only forward modes are known to be present, then we write

[V + (z)] = a1 [V1 ]e−γ1 z + a2 [V2 ]e−γ2 z ; [I + (z)] = a1 [I1 ]e−γ1 z + a2 [I2 ]e−γ2 z (3.49)

where a1 and a2 are complex constants (actually modal wave amplitudes) which are to be determined.
To make the notation more compact, we create the column vector
 
 a1 
[a] = 


 (3.50)
a2

and the diagonal matrix  

−γ1 z
 e 0 
[E(z)] = 


 (3.51)
−γ2 z
0 e

Now (3.49) can be written more compactly as

[V + (z)] = [MV ][E(z)][a]; [I + (z)] = [MI ][E(z)][a] (3.52)

where [MV ] and [MI ] are the matrices defined in (3.41). By (3.52) we see that for any [V + ] and [I + ],
we have
[V + (z)] = [Zc ][I + (z)] (3.53)
where the matrix characteristic impedance [Zc ] is defined by

1
[Zc ] = [MV ][MI ]−1 = [MV ][MV ]T (3.54)
2
using (3.43). This matrix represents the [Z]-matrix of the two-port termination which must be used at
the end of a multiconductor line section if no reflected waves are to exist. Setting z = 0 in (3.52) allows
us to determine [a]:
[a] = [MV ]−1 [V + (0)] = [MI ]−1 [I + (0)] (3.55)
so we can rewrite the forward-going (direct) voltage (3.52) as

[V + (z)] = [MV ][E(z)][MV ]−1 [V + (0)]; [I + (z)] = [MI ][E(z)][MI ]−1 [I + (0)] (3.56)

We can similarly express the reverse voltage waves as

[V − (z)] = [MV ][E(−z)][MV ]−1 [V − (0)]; [I − (z)] = [MI ][E(−z)][MI ]−1 [I − (0)] (3.57)

These reverse waves obey

[V − (z)] = [MV ][E(−z)][b]; [I − (z)] = −[MI ][E(−z)][b] (3.58)

(where [b] is the vector of reverse wave amplitudes) and

[V − (z)] = −[Zc ][I − (z)] (3.59)

81
Combining the forward and reverse waves and using (3.53) and (3.59) gives the total voltage and current
on the coupled lines:

[V (z)] = [V + (z)] + [V − (z)] = [Zc ] [I + (z)] − [I − (z)]

(3.60)

and
[I(z)] = [I + (z)] + [I − (z)] (3.61)
If only forward modes (3.49) are present, the complex power carried along the multiconductor line
is found from (3.29) to be
1 + 1 n ∗ † −γ1∗ z ∗
o
[I (z)]† [V + (z)] = + a∗2 [I2 ]† e−γ2 z a1 [V1 ]e−γ1 z + a2 [V2 ]e−γ2 z

P̂ (z) = a1 [I1 ] e
2 2
1 2 † −2α1 z
= |a1 | [I1 ] [V1 ]e + |a2 |2 [I2 ]† [V2 ]e−2α2 z (3.62)
2 o
+a1 a∗2 [I2 ]† [V1 ]e−(α1 +α2 )z−j(β1 −β2 )z + a2 a∗1 [I1 ]† [V2 ]e−(α1 +α2 )z−j(β2 −β1 )z

which is significantly more complicated than the expression for power carried by a single mode. In
addition to the first two terms, which represent power carried by each mode in the absence of the other,
there are “cross-powers” represented by the third and fourth terms of (3.62), which indicate that the
powers carried by individual modes do not obey superposition in general.
However, in the special case of a propagating mode on a lossless line, we saw in the previous section
that the eigenvectors [Im ] are real, so that [Im ]† = [Im ]T . Using the normalization and orthogonality
properties (3.35) and (3.38), we can then reduce (3.62) to the simpler form
P̂ (z) = |a1 |2 + |a2 |2 (3.63)

Thus, one consequence of orthogonality of propagating modes on lossless multiconductor lines is that
the total complex power flow in the forward modes is equal to the sum of the powers carried by the
individual modes. This is one reason that so much importance is attached to the study of modes on
transmission lines and waveguides.

3.3.4 Reflection on a multiconductor line

Reflection on a multiconductor line is treated in a manner analogous to that on a single transmission
line. The matrix voltage reflection coefficient [ρ(z)] at an arbitrary position is defined by

[V − (z)] = [ρ(z)][V + (z)] (3.64)

which must hold for any combination of incident waves [V + (z)]. Using (3.53) and (3.59) we also find
that
[I − (z)] = −[Zc ]−1 [ρ(z)][Zc ][I + (z)] (3.65)
If the matrix line impedance is defined (again, by analogy with the case of a single transmission line)
by:
[V (z)] = [Z(z)][I(z)] (3.66)
for any possible [V (z)] and [I(z)], then

[ρ(z)] = [Zc ] ([Z(z)] + [Zc ])−1 ([Z(z)] − [Zc ]) [Zc ]−1 (3.67)

in analogy with (1.49).

If an arbitrary two-port termination characterized by the load impedance matrix [ZL ] is connected at
z = d, the voltage and current at the load are related by [V (d)] = [ZL ][I(d)], and we have [Z(d)] = [ZL ],
so that
−1
[ρ(d)] = [Zc ] ([ZL ] + [Zc ]) ([ZL ] − [Zc ]) [Zc ]−1 (3.68)

82
generalizing the result (1.53) for a single transmission line. Note that merely connecting a shunt load
impedance from each conductor to ground at the end of the line is not in general sufficient to suppress
all reflected waves: a full two-port network whose impedance matrix representation is [ZL ] = [Zc ] must
be connected to the end of the coupled lines. This point is explored further in problem p3-7.

3.3.5 Excitation of a multiconductor line

(1) (2)
Suppose that a pair of ideal voltage sources Vg and Vg are connected between conductors 1 and 2
respectively and ground at the end z = 0 of a semi-infinite two-conductor line located in z > 0. Only
direct modes are excited on the coupled lines, since there is no reflection. Defining the column vector
 
(1)
 Vg 
[Vg ] = 



(2)
Vg

and observing that [V + (0)] = [Vg ], we have from (3.56) that

1
[V + (z)] = [MV ][E(z)][MV ]−1 [Vg ] = [MV ][E(z)][MI ]T [Vg ] (3.69)
2
or
1 n (1) −γ1 z (1)
o
V (1)+ (z) = V1 e [I1 ]T [Vg ] + V2 e−γ2 z [I2 ]T [Vg ]
2
1 n (2) −γ1 z (2)
o
V (2)+ (z) = V1 e [I1 ]T [Vg ] + V2 e−γ2 z [I2 ]T [Vg ] (3.70)
2
(2)
Note that even if a voltage source is connected to only one of the conductors (Vg = 0, say), in general
non-zero voltages will appear on both conductors due to the coupling. This “cross-excitation” is known as
crosstalk, and is often undesirable, especially in high-speed digital circuits. An example of this behavior
is considered in problem p3-6.

3.3.6 Example: The symmetric three-conductor line

Consider the case when the two ungrounded conductors are situated identically relative to the ground
conductor. Then c11 = c22 and l11 = l22 . Here, the solution is considerably simpler than in the general
case, and can be written out explicitly as follows. We introduce the common notations
c12 l12
KC = ; KL =
c11 l11
for the so-called capacitive and inductive coupling coefficients, respectively.
Because of the symmetry of this configuration, its modes will be either even or odd with respect to
conductors 1 and 2. For the even mode,
   
(1) (1)
 Ve   Ie 
[Ve ] = 

;
 [Ie ] = 


 (3.71)
(1) (1)
Ve Ie

while for the odd mode    

(1) (1)
Vo Io
[I (o) ] = 
   
[Vo ] = 

;
 

 (3.72)
(1) (1)
−Vo −Io

83
From (3.33) we have
ω(c11 − c12 )
Ie(1) = Ve(1) (3.73)
βe
and
ω(c11 + c12 )
Io(1) = Vo(1) (3.74)
βo
The propagation coefficients βe,o for these modes are obtained by substitution of (3.71)-(3.74) into (3.32):
βe2 = β11
2
(1 − KC )(1 + KL ); βo2 = β11
2
(1 + KC )(1 − KL ) (3.75)
where β11 is the propagation coefficient in the absence of coupling:
2 2
β11 = β22 = ω 2 l11 c11 (3.76)
We speak of the even and odd mode scalar characteristic impedances defined by
(1) r (1) r
Ve (0) 1 + KL Vo (0) 1 − KL
Zce = (1) = Zc11 ; Zco = (1) = Zc11 (3.77)
Ie 1 − K C Io 1 + KC
(0)
where Zc11 is the characteristic impedance of one of the uncoupled lines:
r
(0) l11
Zc11 = (3.78)
c11
To get the matrix characteristic impedance for this case, we first use the normalization condition (3.35)

1 2
Zco Zco

(Zce - Zco) /2

Figure 3.14: Matched load impedance T-network for symmetric three-conductor line.
(1) (1)
or (3.36) to obtain the values of Ve and Vo :
q
(1)
Ve,o = Zc(e,o) (3.79)
We can now construct  
√ √
 Zce Zco 
[MV ] = 
 √
 (3.80)
√ 
Zce − Zco
and using this in (3.54) we get
 
1 Zce + Zco Zce − Zco 
[Zc ] =   (3.81)
2 
Zce − Zco Zce + Zco

The two-port load impedance network corresponding to this matrix characteristic impedance is shown
in Fig. 3.14. Unless Zce = Zco , termination of each conductor at the end of this line by a separate
impedance is not sufficient to prevent the appearance of reflected waves.

84
3.4 Resonators
A resonator (roughly defined) is a device capable of storing a large amount of energy during oscillations
in which relatively little of this energy is dissipated. We will consider here a fairly general class of RLC
networks as resonators, and afterwards illustrate by means of some examples.
Consider a circuit made up only of resistors, capacitors, inductors and ideal voltage and current
generators, such that we may choose a tree of the network with: (i) all tree branches consisting only
of resistors, capacitors and/or current sources in parallel, and (ii) all links consisting only of resistors,
inductors and voltage sources in series. Such a tree is called a proper tree. A section of such a circuit
showing a typical tree branch and typical link is presented in Fig. 3.15. From (E.1)-(E.2), the circuit

Ik

– +
Lk +
Rk Gr Jr
Uk
























Vr
Link k Cr
–





















Tree branch r

Figure 3.15: Portion of a resonator circuit showing tree branch r and link k.

equations for this network can be written in the form

[D][I] = ([G] + jω[C])[V ] − [J] (3.82)

−[D]T [V ] = ([R] + jω[L])[I] + [U ] (3.83)

where [D] is the matrix described in Appendix E that specifies the connections in the network, [I] is
a column vector of link currents, [V ] is a column vector of tree branch voltages, [J] is a column vector
of the independent current sources in the tree branches, and [U ] is a column vector of the independent
voltage sources in the links. The matrices [G], [C], [R] and [L] are diagonal matrices whose entries
are the values of tree branch conductance, tree branch capacitance, link resistance and link inductance,
respectively.
We begin by examining the free oscillations or modes of this network: that is, the possible nontrivial
solutions that can exist in the absence of excitation (when the voltage sources are replaced by short
circuits [U ] = 0 and the current sources by open circuits [J] = 0). We can expect this to happen only
for certain (generally complex) special values of the angular frequency ω called natural frequencies. We
denote these by Ωm = ωm + jsm , and the corresponding voltages [V (m) ] and currents [I (m) ] satisfy

[D][I (m) ] = ([G] + jΩm [C])[V (m) ] (3.84)

−[D]T [V (m) ] = ([R] + jΩm [L])[I (m) ] (3.85)

Taking [V (n) ]T times (3.84) and [I (n) ]T times (3.85) give, respectively:

[V (n) ]T [D][I (m) ] = [V (n) ]T ([G] + jΩm [C])[V (m) ]

85
 −[I (n) ]T [D]T [V (m) ] = [I (n) ]T ([R] + jΩm [L])[I (m) ]
If the indices m and n are interchanged in the second of these, and the two equations added, we obtain

0 = [V (n) ]T ([G] + jΩm [C])[V (m) ] + [I (m) ]T ([R] + jΩn [L])[I (n) ] (3.86)
If m and n are again interchanged in (3.86), and the result subtracted from (3.86) itself, we arrive at
the following identity:
n o
(Ωn − Ωm ) [I (m) ]T [L][I (n) ] − [V (m) ]T [C][V (n) ] = 0 (3.87)

If Ωn 6= Ωm , we have the orthogonality property:

n o
[I (m) ]T [L][I (n) ] − [V (m) ]T [C][V (n) ] = 0 (3.88)

analogous to (3.38) for multiconductor transmission lines. If there are linearly independent voltage
and current vectors satisfying (3.84) and (3.85) that have Ωn = Ωm , then these modes are said to be
degenerate. Such cases are rare (but do include the case of critical damping in the language of circuit
analysis). In any event, the degeneracy can always be removed by a small change in one or more of the
component values, so we will only consider the case of nondegenerate modes from here on.
If Ωm is a natural frequency of the resonator, then from (3.84) and (3.85) it can be seen that
−Ω∗m ≡ Ω−m will also be a natural frequency, corresponding to the voltages and currents

[V (−m) ] = [V (m) ]∗ ; [I (−m) ] = [I (m) ]∗ (3.89)

If ωm = 0, so that Ωm = jsm is purely imaginary, the mode is said to be overdamped, and it is possible to
choose the amplitudes of all mode voltages and currents to be real. In this case, we adopt the convention
that ±m refer to the same mode. If ωm 6= 0, the orthogonality property (3.88) together with (3.89)
implies that
[I (m) ]† [L][I (m) ] = [V (m) ]† [C][V (m) ]
i. e., the time-average stored electric and magnetic energies in the circuit for this mode are equal. In
this case, the distinct pair of modes Ω±m is said to be underdamped (usually the case of most interest).
Putting n = −m in (3.86) above, and using (3.89), we have, after some rearrangement:

[I (m) ]† [R][I (m) ] + [V (m) ]† [G][V (m) ]

sm = (3.90)
[I (m) ]† [L][I (m) ] + [V (m) ]† [C][V (m) ]
In other words, the inverse time constant or decay constant for a mode is equal to one-half the ratio of
the time-average dissipated power in the network to the time-average stored energy in the network.

3.4.1 Internal excitation of a resonator

Let us return now to the case where independent sources excite the resonator. Since the modes are
orthogonal as shown above, it can be shown that they are linearly independent and any voltage and
current vectors can be expressed as a linear combination of them:
X X
[V ] = dm [V (m) ]; [I] = dm [I (m) ] (3.91)
m m

where dm are complex constants to be determined. But this can be done by manipulations similar to
those that led to the orthogonality property between modes. First we left multiply (3.82) and (3.83) by
[V (n) ]T and [I (n) ]T respectively:
[V (n) ]T [D][I] = [V (n) ]T ([G] + jω[C])[V ] − [V (n) ]T [J]
(n) T T
−[I ] [D] [V ] = [I (n) ]T ([R] + jω[L])[I] + [I (n) ]T [U ]

86
Similarly, we left multiply (3.84) and (3.85) (with m → n) by [V ]T and [I]T respectively:

[V ]T [D][I (n) ] = [V ]T ([G] + jΩn [C])[V (n) ]

−[I]T [D]T [V (n) ] = [I]T ([R] + jΩn [L])[I (n) ]

The first and fourth of these equations are then subtracted from the sum of the second and third, giving
n o
0 = j(ω − Ωn ) [I (n) ]T [L][I] − [V (n) ]T [C][V ] + [I (n) ]T [U ] + [V (n) ]T [J] (3.92)

We now substitute the expansion (3.91) into (3.92) and use the orthogonality property of the modes,
which allows us to solve for dn :

j [V (n) ]T [J] + [I (n) ]T [U ]

dn =
ω − Ωn [I (n) ]T [L][I (n) ] − [V (n) ]T [C][V (n) ]

Thus the total voltage and current vectors are given by

X j[V (m) ] [V (m) ]T [J] + [I (m) ]T [U ]
[V ] = (3.93)
m
ω − Ωm [I ] [L][I (m) ] − [V (m) ]T [C][V (m) ]
(m) T

X j[I (m) ] [V (m) ]T [J] + [I (m) ]T [U ]

[I] = (3.94)
m
ω − Ωm [I (m) ]T [L][I (m) ] − [V (m) ]T [C][V (m) ]

Note that this result is independent of any arbitrary amplitude we may have chosen for the mode voltages
and currents themselves.
We see that any single mode’s contribution to these sums is resonant. That is, its magnitude relative
to its maximum value (which occurs at ω = ωm —the resonant frequency of the mode) varies with
frequency as
sm
p
(ω − ωm )2 + s2m
√
The half-power (or 1/ 2-amplitude) points of this response occur when |ω − ωm | = sm , corresponding
to a half-power bandwidth of 2sm . The quality factor or Q of the mode is then defined as the inverse of
the relative bandwidth:
|ωm |
Qm = (3.95)
2sm
and provides a measure of how sharp this resonance is in the frequency response of the circuit. The Q
of an overdamped mode is evidently 0.
Equations (3.93) and (3.94) also show us another way in which the excitation of a particular mode
may be enhanced relative to the others. We can choose the excitation [U ] and [J] to be orthogonal to
all but one of the modes as follows:

[U ] = [L][I (n) ]; [J] = [C][V (n) ]

The orthogonality property causes all the dm to be zero except dn .

If only one independent source is active in the resonator, we can use the foregoing to express the
input impedance or admittance seen by this source. In the case when only the current source Jr in
Fig. 3.15 is active, it sees an impedance by (3.93) of
(m)
Vr X j(Vr )2
Zin = = (3.96)
Jr m
(ω − Ωm )([I (m) ]T [L][I (m) ] − [V (m) ]T [C][V (m) ])

87
Likewise, when only the voltage source Uk is active, it sees an admittance by (3.94) of
(m)
Ik X j(Ik )2
Yin = = (3.97)
Uk m
(ω − Ωm )([I ] [L][I (m) ] − [V (m) ]T [C][V (m) ])
(m) T

Formally, eqn. (3.96) is identical to the impedance of a series string of RLCG modules (each made up of
the parallel combination of a lossy inductor—an R and L in series—and a lossy capacitor—a G and C
in parallel) as shown in Fig. 3.16(a). This is a so-called Foster canonical form of the network. Likewise,

G1 G2

Zin C1 C2 ...

L1 R1 L2 R2

(a)

L1 L2

Yin
R1 R2 ...

G1 G2
C1 C2

(b)

Figure 3.16: Equivalent circuits of lumped element resonant circuit seen by: (a) soldering-iron entry of
current source into tree branch; (b) pliers entry of voltage source into link.

eqn. (3.97) is formally identical to the admittance of a parallel combination of modules each made up of
the series combination of a lossy inductor and lossy capacitor as shown in Fig. 3.16(b). In either case,
each module corresponds to the contribution from a single pair of underdamped modes of the resonator,
or to that of a single overdamped mode (in which case the module degenerates into a GC or RL unit
only).
We said that this equivalence was formal because it is possible that the resulting equivalent circuit
might not be physically realizable, because negative values of resistance, inductance and so on would be

88
required. For low-loss networks or modes with high enough Q, however, it can be shown that all values
of R, L, G and C in these equivalent circuits are positive. We can, in any event, use the equivalent
circuit as we would an actual one for most practical purposes. Note also that the component values for
the two cases (Zin and Yin ) are in general unrelated, depending as they do not only on the mode but
also on the point and type of excitation of the network.
At frequencies near the isolated resonant frequency of a high-Q mode, one term of the equivalent
circuit (or one term of the series (3.96) or (3.97)) will dominate all the others, and the network behaves
as a simple four-element resonant circuit. For the link voltage source, we have
Zmax
Zin ≃ (3.98)
1 + 2jQm (ω − ωm )/ω|m|

where
(m)
(Vr )2
Zmax = −
sm ([I (m) ]T [L][I (m) ] − [V (m) ]T [C][V (m) ])
while for the tree branch current source,
Ymax
Yin ≃ (3.99)
1 + 2jQm (ω − ωm )/ω|m|

where
(m)
(Ik )2
Ymax = −
sm ([I (m) ]T [L][I (m) ] − [V (m) ]T [C][V (m) ])
This allows for considerable simplification of many resonator circuit analyses.

3.4.2 External excitation of a resonator

Usually a resonator does not stand by itself, but is connected to an external circuit in order to make a
component such as an oscillator, a filter, etc. Instead of ideal current or voltage generators placed within
the resonator, we thus need to consider a real generator (with a nonzero equivalent impedance) inserted
into the resonator network. Because of the Thévenin and Norton equivalence theorems, it is sufficient
to consider either an ideal voltage source with a series impedance, or an ideal current source with a
shunt admittance as the external excitation. In the former case, the excitation is inserted by what is
known as a pliers entry to the resonator: a branch of the resonator circuit is broken, and the Thévenin
equivalent circuit is inserted in series with this branch. In the latter case, the excitation is connected by
a soldering-iron entry: the Norton equivalent circuit of the external generator is connected across a pair
of existing nodes of the resonator circuit. If the external generator is to avoid making a large impact on
the properties of the resonator, the generator impedance should be low for a pliers entry, and high (i. e.,
low admittance) for a soldering-iron entry.
Consider the RLGC circuit shown in Fig. 3.17. If the voltage source VG and its associated impedance
ZG (which have been inserted using a pliers entry) is removed from the circuit by replacing them with
a short circuit, we have the elementary component of the equivalent circuit in Fig. 3.16(a), representing
a single mode of a general resonant circuit driven by a current source. If on the other hand the current
source IG and its associated admittance YG (which are connected by a soldering-iron entry) is removed
from the circuit, we have the elementary component of the equivalent circuit in Fig. 3.16(b) driven by
a voltage source. There is one tree branch and one link in this circuit: the tree branch voltage is that
across the capacitor V1 = VC , and the link current is that through the inductor I1 = IL .
The resonant modes satisfy (3.84) and (3.85):
(m) (m)
I1 = (G + jΩm C)V1
(m) (m)
−V1 = (R + jΩm L)I1

89
 IL
R
L
ZG +
+
VG VC C G YG IG
- -

Figure 3.17: An RLGC circuit.

which are easily solved to give a pair of modes. If

 2
1 1 G R
> −
LC 4 C L

then the modes form an underdamped pair with

s  2  
1 1 G R j G R
Ω±1 = ±ω1 + js1 = ± − − + + (3.100)
LC 4 C L 2 C L

while if  2
1 1 G R
< −
LC 4 C L
the modes are overdamped, with the imaginary resonant frequencies
s  2  
1 G R 1 j G R
Ω1,2 = ±j − − + +
4 C L LC 2 C L

For the underdamped modes, the quality factor is

ω1
Q= G
(3.101)
C +R L
√
For high-Q modes, ω1 ≃ 1/ LC. In this case, the maximum admittance seen by the voltage generator
VG in the absence of the generator impedance ZG is, from (3.99),
1
Ymax ≃ (3.102)
R + LG
C

and the maximum impedance seen by the current generator IG in the absence of YG is by (3.98)
1
Zmax ≃ (3.103)
G + CR
L

The insertion of the actual voltage generator VG and ZG , aside from inducing voltages and currents
in the resonator, also perturbs its natural frequencies. For, even if the ideal source VG were zero, the
added impedance ZG to the RL link would result in more resistance: R → R + RG , and a change in the
link reactance from ωL to ωL + XG . If the mode has a high Q, so that |Ωm L| ≫ R and |Ωm C| ≫ G,
and the generator impedance does not vary rapidly near the unperturbed resonant frequencies, then the

90
 (Zc , γ )
Z in

d
Figure 3.18: Transmission-line resonator.

effect of the generator impedance on the natural modes of the circuit is to change R → R + RG as above
and L → L + XG /|ωm |. This changes the real parts ω±1 of (3.100) by a small amount, due mostly to
XG . The imaginary part s1 may be more significantly changed, since now
 
1 G R + RG
s1 = +
2 C L

In fact, the quality factor has been changed from the unloaded Q (or QU ) of (3.101) to the loaded Q or
QL , which is found from
1 1 1
= + (3.104)
QL QU Qe
where Qe is called the external Q, and is related to the additional resistance introduced into the network
by the external generator:
1 RG /L
= (3.105)
Qe ω1
We often speak of the coupling factor κ = QU /Qe between the inserted generator and the resonator, so
that
1 1+κ
= (3.106)
QL QU
Maximum power transfer from the inserted generator to the resonator occurs when RG = 1/Ymax, or
from (3.102), (3.105) and (3.106),
Qe = QU or κ = 1 (3.107)
This is known as critical coupling.

3.4.3 Example: Short-circuited section of lossy transmission line

Consider the example of a length d of lossy transmission line short-circuited at one end as shown in
Fig. 3.18. The input impedance seen at the other end is

Zin = Zc tanh γd (3.108)

where Zc and γ are given by (1.8), (1.9) and (1.5), where r, l, c and g are assumed to be frequency-
independent. For resonant modes,√ the resonant frequencies must obey Zin = 0, or γd = jmπ, for any
integer m. If m = 0, the factors g + jωc in Zin will cancel, leaving the special case r + jωl = 0, or the
purely damped complex resonant frequency
r
Ω0 = j (3.109)
l

91
For m 6= 0, we have the complex resonant frequencies

Ω±m = ±ωm + jsm (3.110)

where s 2
mπ 1  g r 2
ωm = √ − − (3.111)
d lc 4 c l
and
1 g r
sm = − (3.112)
2 c l
The unloaded Q of these modes (if ωm ≫ sm ) is
mπ
QU = √ (3.113)
2sm d lc
which is larger for each higher mode number since all sm ’s are equal.

92
3.5 Notes and References
The most familiar circuit equivalence theorem is undoubtedly Thévenin’s theorem. A few texts, for
example:
Carlin and Giordano (1964), sects. 3.7 and 3.8.
present generalizations of Thévenin’s theorem to multiport networks. The generalization is in fact
contained in a paper by Helmholtz which predates Thévenin’s paper by 30 years:
H. L. F. von Helmholtz, “Ueber einige Gesetze der Vertheilung elektrischer Ströme in körper-
lichen Leitern mit Anwendung auf die thierisch-elektrischen Versuche,” Ann. Phys.
Chem., vol. 89, pp. 211-233, 353-377 (1853) [also in H. L. F. von Helmholtz, Wis-
senschaftliche Abhandlungen, vol. 1. Leipzig: Johann Ambrosius Barth, 1882, pp. 475-
519].
A summary of Helmholtz’ result can be found in:
Koenigsberger (1965), pp. 99-103.
Other sources for the generalized version of this theorem are:
H. Ataka, “On an extension of Thévenin’s theorem,” Phil. Mag., ser. 7, vol. 25, pp. 663-666
(1938).
L. Sartre, “Sur une généralisation du théorème de Thévenin et son application au calcul
des courants, dans le cas d’une charge dyssymétrique de réseaux polyphasés,” Rev. Gén.
Élec., vol. 61, pp. 238-240 (1952).
L. A. Zadeh, “On passive and active networks and generalized Norton’s and Thévenin’s
theorems,” Proc. IRE, vol. 44, p. 378 (1956).
L. A. Zadeh, “Multipole analysis of active networks,” IRE Trans. Circ. Theory, vol. 4, pp.
97-105 (1957).
M. A. Shakirov, “A multidimensional equivalent source theorem,” Radioelectron. Commun.
Syst., vol. 21, no. 5, pp. 24-28 (1978).
Related to the multiport Thévenin theorem is a theorem on maximum power transfer from a multiport
network containing sources. See:
C. A. Desoer, “The maximum power transfer theorem for n-ports,” IEEE Trans. Circ.
Theory, vol. 20, pp. 328-330 (1973).
A. T. de Hoop, “The N -port receiving antenna and its equivalent electrical network,” Philips
Res. Repts., vol. 30 (suppl.), pp. 302*-315* (1975).
The relation of the Thévenin theorem to the S-matrix representation is discussed in
D. Kajfez and D. R. Wilton, “Network representation of receiving multiport antennas,” Arch.
Elek. Übertragungstech., vol. 30, pp. 450-454 (1976).
Huygens or wave equivalent sources for circuits and transmission lines are discussed in
Haus and Adler (1959), chapter 5.
and also in
H.-J. Butterweck, “Die Ersatzwellenquelle,” Arch. Elek. Übertrag., vol. 14, pp. 367-372
(1960).
K. Hartmann, “Noise characterization of linear circuits,” IEEE Trans. Circ. Syst., vol. 23,
pp. 581-590 (1976).

93
 V. N. Dikyj and E. F. Zaytsev, “Calculation of noise in multi-terminal networks with specified
wave parameters,” Radio Eng. Electron. Phys., vol. 22, no. 12, pp. 80-85 (1977).
R. Pauli and W. Konig, “Wave sources (linear network design and analysis),” Arch. Elek.
Übertragungstech., vol. 41, pp. 365-368 (1987).
The substitution theorem (which is the analog of Love’s equivalence principle and Schelkunoff’s
induction theorem for electromagnetic fields; see chapter 9) is given in:
Desoer and Kuh (1969), chapter 16.
The relationship to their electromagnetic counterparts is given in
S. A. Schelkunoff, “On diffraction and radiation of electromagnetic waves,” Phys. Rev., vol.
56, pp. 308-316 (1939).
R. F. Harrington, “Field equivalence theorems and their circuit analogues,” Elec. Eng., vol.
73, pp. 923-927 (1954).
I. V. Lindell, “Huygens’ principle in electromagnetics,” IEE Proc. pt. A, vol. 143, pp.
103-105 (1996).
A discussion of the telegrapher’s equations for the source region can be found in
King (1965), chapter 1.
Johnson (1965), sect. 4.19;
Mashkovtsev, Tsibizov and Emelin (1966), chapters 1 and 4;
Felsen and Marcuvitz (1973), chapter 2;
Frankel (1977), chapters 4 and 10.
and also in
C. S. Lindquist, “Transmission line response to independent distributed sources,” Proc.
IEEE, vol. 56, pp. 1353-1354 (1968).
The related question of non-radiating sources in a transmission line was considered by
A. Sihvola, G. Kristensson, and I. V. Lindell, “Nonradiating sources in time-domain transmission-
line theory,” IEEE Trans. Micr. Theory Tech., vol. 45, pp. 2155-2159 (1997).
The mode concept in RLC networks is explored in:
C. A. Desoer, “Modes in linear circuits,” IRE Trans. Circ. Theory, vol. 7, pp. 211-223
(1960).
Zadeh and Desoer (1963), chapter 5.
R. R. Parker, “Normal modes in RLC networks,” Proc. IEEE, vol. 57, pp. 39-44 (1969).
Modes in networks containing both lumped elements and classical transmission lines are investigated in:
W. H. Ingram, “Electrical oscillations in a non-uniform transmission line,” Univ. Washington
Publ. Math., vol. 2, no. 1, pp. 17-38 (1930).
Adler, Chu and Fano (1960), chapter 6.
The notion of the quality factor can be traced back to the early part of the twentieth century, and
has undergone generalizations and modifications such that many different (and sometimes inconsistent)
definitions can be found in the literature. For the history of Q, see:
E. I. Green, “The story of Q,” Amer. Scientist, vol. 43, pp. 584-594 (1955).
P. B. Fellgett, “Origins of the usage of ‘Q’,” J. IERE, vol. 56, pp. 45-46 (1986).
K. L. Smith, “ ‘Q’,” Electron. Wireless World, vol. 92, no. 1605, pp. 51-53 (1986).

94
It is surprising how difficult it has proven to give precise definitions to the notions of “Q” and reso-
nant frequency. The “official” definition of Q given in the IEEE Standard Dictionary of Electrical and
Electronics Terms is
ωWmax
Q=
PL
where ω is the frequency of the time-harmonic source(s), Wmax is the maximum (it is not stated whether
this is a maximum with respect to time or to frequency) stored energy and PL is the time-average power
dissipated in the resonator. The frequency is usually taken to be a resonance frequency, which is itself
a somewhat ambiguous concept. A resonant frequency is defined “officially” to be the frequency of a
source or sources driving a system which results in a locally maximum response (i. e., some induced
voltage or current) from the system.
It is readily seen from consideration of simple examples that in general these maxima may occur at
different frequencies if different responses are considered. Moreover, one can construct examples to show
that the definition given above does not give a value of Q consistent with the bandwidth definition of
this quantity. This seems to be due to the fact that, in systems containing transmission line sections,
there can be an energy stored in these sections merely until it is dissipated in the load impedances which
terminate these sections. By varying the length of these sections, the stored energy is varied, but the
power dissipated is not changed proportionately, so different values of Q are obtainable for a network
with the same terminal response. To get around this problem, the notion of observable energies in the
definition of Q for systems of infinite extent was introduced by
D. R. Rhodes, “ Observable stored energies of electromagnetic systems,” J. Franklin Inst.,
vol. 302, pp. 225-237 (1976).
but the end result would seem to be that the bandwidth definition of Q is to be regarded as fundamental,
and agrees with the energy definition only in certain cases or with modifications.
If, as we have done, one makes use of the mode concept in defining Q, the definitions become precise.
Only when more than one mode is present does ambiguity set in, because then the Q’s of the individual
modes do not combine in any simple way to form an unambiguous Q for the network as a whole.
However, since near resonant frequencies one mode is often excited with a much larger amplitude than
all the others, use of the Q and resonant frequency for that mode will be a good approximation to those
of the entire circuit obtained from the “official” definitions (the more so the larger Q is).
The transition from resonators made up purely from finite lumped-element circuits or circuits con-
taining finite lengths of transmission line to ones of infinite extent can result in a much more complicated
spectrum of modes which we do not investigate here. See, for example,
Gross and Braga (1961).
A. E. Karbowiak, “Solution of the excitation problem in resonant dissipative and inhomoge-
neous structures,” Alta Frequenza, vol. 38 (supplement), pp. 109-115 (1969).
The issue of whether the expansion of the input impedance or admittance of a resonator as a series of
terms due to individual modes will correspond to a physically realizable equivalent circuit is addressed
in
Schelkunoff (1943), chapter 5.
Guillemin (1957), chapter 9.
Montgomery, Dicke and Purcell (1965), chapter 7.
as well as
S. A. Schelkunoff, “Representation of impedance functions in terms of resonant frequencies,”
Proc. IRE, vol. 32, pp. 83-90 (1944).
M. K. Zinn, “Network representation of transcendental impedance functions,” Bell Syst.
Tech. J., vol. 31, pp. 378-404 (1952).

95
 F. M. Reza, “RLC canonic forms,” J. Appl. Phys., vol. 25, pp. 297-301 (1954).
E. V. Zelyakh and A. V. Kisel’, “Canonical configurations of two-terminal networks consisting
of impedors of two types,” Telecommun. Radio Eng., vol. 19/20, no. 7, pp. 67-73 (1965).
E. V. Zelyakh and A. V. Kisel’, “Equivalent-circuit theory,” Telecommun. Radio Eng., vol.
22/23, no. 1, pp. 108-113 (1968).

The classic paper on multiconductor transmission lines is

J. R. Carson and R. S. Hoyt, “Propagation of periodic currents over a system of parallel
wires,” Bell Syst. Tech. J., vol. 6, pp. 495-545 (1927).

This paper still rewards the reader who is willing to seek it out. More recent treatments of the subject
are found in

R. Levy, “Directional couplers,” in Advances in Microwaves, vol. 1 (L. Young, ed.). New
York: Academic Press, 1966, pp. 115-209.
Uchida (1967);
Matick (1969), chapters 7 and 8;
Gupta and Singh (1974), chapter 5;
Gupta, Garg and Bahl (1979), chapter 8;
Frankel (1977), chapters 3 and 7;
Edwards (1981), chapter 6 and appendix A.
Paul (1992), chapter 10.
Paul (1994).

A good collection of important papers on this topic is to be found in

Young (1972).
Other relevant papers are

T. Hirota, “A contribution to the theory of two parallel lines,” Nippon Elec. Commun. Eng.,
no. 10, pp. 147-151 (1938).
J. Brown, “Propagation in coupled transmission line systems,” Quart. J. Mech. Appl. Math.,
vol. 11, pp. 235-243 (1958).
E. G. Vlostovskii, “Theory of coupled transmission lines,” Telecommun. Radio Eng., pt. 2,
vol. 22, no. 4, pp. 87-93 (1967).
M. M. Krage and G. I. Haddad, “Characteristics of coupled microstrip transmission lines—I:
Coupled-mode formulation of inhomogeneous lines,” IEEE Trans. Micr. Theory Tech.,
vol. 18, pp. 217-222 (1970).
J. C. Isaacs and N. A. Strakhov, “Crosstalk in uniformly coupled transmission lines,” Bell
Syst. Tech. J., vol. 52, pp. 101-115 (1973).
K. D. Marx, “Propagation modes, equivalent circuits, and characteristic terminations for
multiconductor transmission lines with inhomogeneous dielectrics,” IEEE Trans. Micr.
Theory Tech., vol. 21, pp. 450-457 (1973).
C. R. Paul, “On uniform multimode transmission lines,” IEEE Trans. Micr. Theory Tech.,
vol. 21, pp. 556-558 (1973).
V. K. Tripathi, “Asymmetric coupled transmission lines in an inhomogeneous medium,”
IEEE Trans. Micr. Theory Tech., vol. 23, pp. 734-739 (1975).
C. R. Paul, “Useful matrix chain parameter identities for the analysis of multiconductor
transmission lines,” IEEE Trans. Micr. Theory Tech., vol. 23, pp. 756-760 (1975).
L. N. Deryugin, O. A. Kurdyumov and V. F. Terichev, “Operating mechanism of directional
couplers on two lines with distributed electromagnetic coupling,” Radiophys. Quantum
Electron., vol. 21, pp. 208-209 (1978).

96
 J. O. Scanlon, “Theory of microwave coupled-line networks,” Proc. IEEE, vol. 68, pp.
209-231 (1980).
D. E. Bockelman and W. R. Eisenstadt, “Combined differential and common-mode scattering
parameters: Theory and simulation,” IEEE Trans. Micr. Theory Tech., vol. 43, pp.
1530-1539 (1995).
C. R. Paul, “Decoupling the multiconductor transmission line equations,” IEEE Trans. Micr.
Theory Tech., vol. 44, pp. 1429-1440 (1996).

Applications of multiconductor transmission lines are numerous. In addition to directional couplers,

they can be used to realize baluns, hybrid networks, power combiners and splitters and filters. A sample
of the vast literature on the subject is

A. Matsumoto and N. Nagai, “Synthesis of multiports with multiwire sections,” Developments

in Multiwire Networks [Monograph Series, Research Institute of Applied Electricity no.
12]. Sapporo, Japan: Research Institute of Applied Electricity, 1964, pp. 1-13.
N. Nagai, “Theory and experiments on multiwire multiports,” Recent Developments in Mi-
crowave Multiport Networks [Monograph Series, Research Institute of Applied Electricity
no. 18]. Sapporo, Japan: Research Institute of Applied Electricity, 1970, pp. 1-68.
K. Watanabe, N. Nagai and K. Hatori, “Study on power dividers,” Recent Developments in
Microwave Multiport Networks [Monograph Series, Research Institute of Applied Elec-
tricity no. 18]. Sapporo, Japan: Research Institute of Applied Electricity, 1970, pp.
87-101.

Active and nonlinear coupled transmission lines can be used as models of distributed amplifiers; see

C. S. Lindquist, “Uniform transmission line response to independent distributed sources,”

Proc. IEEE, vol. 56, pp. 1740-1741 (1968).
C. S. Lindquist, “Active transmission lines—Characteristic impedances and propagation
functions,” Proc. IEEE, vol. 57, pp. 1422-1423 (1969).

and

Wong (1993).

3.6 Problems

p3-1 Show that a uniform semi-infinite section of transmission line to the left of the point z = 0
on which an incident forward voltage wave V i (z) is present can be replaced by the Thévenin
equivalent circuit shown below, where Zc is the characteristic impedance of the line.

97
 Zc

V i (z) 2V i (0) +
–
(Zc)

z=0
z=0

p3-2 A section of uniform transmission line occupies the interval [z1 , z2 ], and is excited by the
impressed distributed sources Eext (z) and Hext (z) within this region. Find the two-port
Thévenin and Norton equivalent circuits for this transmission line section, with the ports
chosen at the ends z = z1 and z = z2 .
p3-3 Let the distributed sources for a uniform lossless classical transmission line be Hext (z) ≡ 0
and
Eext (z) = e−jβs z
for z > 0, and Eext (z) = 0 for z < 0, where βs is some real constant. Determine expressions
for the wave amplitudes a(z) and b(z) produced by this source. Graph |a(z)| and |b(z)| vs.
z for the cases βs = 0, βs = 0.99β and βs = β, where β is the propagation coefficient of the
transmission line. For purposes of these graphs, choose β and Zc numerically equal to 1.
p3-4 Show that the equivalent sources (3.22) when used in (3.21) produce the same amplitudes a0
and b0 as do the actual sources given in (3.16).
p3-5 Consider a resonator formed by a length l of lossless classical transmission line whose char-
acteristic impedance is Zc and whose propagation coefficient is β = ω/vp . Both Zc and vp
are real and independent of frequency. As the limit of lumped element circuits with indef-
initely large numbers of components, this resonator in principle has an infinite number of
tree branches and links, and therefore an infinite number of possible entry points. We will
consider only one: the terminals at one end.

(a) What are the complex natural frequencies Ωm for the modes of this resonator if the ends
of the line are open-circuited? What are the (unloaded) Q’s of these modes? What are
the natural distributions of voltage and current along the line for each mode (the analogs
of [V (m) ] and [I (m) ] in the lumped-element case)?
(b) Find the impedance Zin of the resonator as seen at one end of the transmission line
resonator (the other end being left open-circuited). Near an unloaded resonant frequency
of the resonator, find an approximate expression for Zin in the form of a pair of terms
Am A−m
+
ω − Ωm ω − Ω−m
where A±m are complex constants, and express this impedance in the form of a parallel
LC module representing the special case of a module from Fig. 2.21(a) with Rm and

98
 Gm equal to zero. Hence, sketch an equivalent circuit for the resonator in the form of
Fig. 2.21(a).
(c) A current generator consisting of an ideal current source IG in parallel with a gener-
ator admittance of YG is connected to one end of the transmission line resonator as
a soldering-iron entry point. Suppose that |YG Zc | ≪ 1, so that the resonator is only
slightly perturbed by the connection of the generator. Near the resonant frequency of
one of the modes, the circuit seen by the ideal current source is that of YG in parallel
with the LC module for this mode only (why?). Find the perturbed natural frequency
for this mode caused by the loading of the resonator by YG , and find the loaded Q for
the mode. Express your answers first in terms of the equivalent parameters in the LC
module, and then directly in terms of the parameters of the transmission line section.
(d) Instead of the current generator, suppose that a feed transmission line (with the same
characteristic impedance Zc as the resonator) is connected to the resonator through a
small (ωC0 Zc ≪ 1) series capacitance C0 as shown.

V+(z)
C0

z=0

If an incident voltage wave V+ is present on the feed line, what is the Norton equivalent
circuit for the feed line/C0 combination? Use your result to find an expression for the
reflection coefficient ρ on the feed line, in the vicinity of one of the resonant modes.
Sketch the a plot of |ρ| versus frequency in this range. You may use any numerical
values for Zc , vp , etc., that you want, provided that they meet the criteria stated above.

p3-6 The coupled-line example of section 3.3.6 is terminated at the far end (z = d) by a two-
port network whose impedance matrix is equal to the characteristic impedance matrix of
the coupled lines. If KC 6= KL , and the initial voltages at the near end (z = 0) are set
to V (1) (0) = VG by a generator, and V (2) (0) = 0 by a short circuit to ground, find an
expression for the voltages V (1) (d) and V (2) (d) at the far end of the coupled lines. Under
what conditions can V (1) (d) = 0?
p3-7 Consider a pair of lossless coupled transmission lines as in section 3.3.6. The inductive and
capacitive coupling coefficients are equal: KC = KL ≡ K. A length l of these coupled lines
forms a four-port network: port 1 is on line 1 at z = 0, port 2 on line 2 at z = 0, port 3 on
line 1 at z = l, and port 4 on line 2 at z = l. Assume that uncoupled transmission lines of
(0) √
characteristic impedance Zc = Zce Zco are connected to each port.
(a) Find the reflection coefficient matrix [ρ] at the load end z = l.
(b) Use the result of part (a) to find the scattering matrix of this four-port network. [Hint:
Make use of the symmetries of this structure to simplify calculations.]

99
 (c) By examining the dependence of S11 , S21 , S31 and S41 on βl, comment on the possible
practical function of this four-port network.

p3-8 Let the distributed sources for a uniform lossless classical transmission line be Hext (z) ≡ 0
and
Eext (z) = e−jβs z

for 0 < z < z0 , and Eext (z) = 0 elsewhere, where βs is some real constant. Determine
expressions for the wave amplitudes a(z) and b(z) produced by this source. Graph |a(z)| and
|b(z)| vs. z for the cases βs = 0, βs = 0.9β and βs = β, where β is the propagation coefficient
of the transmission line. For purposes of these graphs, choose β and Zc numerically equal to
1, and z0 = 10π.

p3-9 Consider the transmission line of problem p1-3, coupled to a straight wire coaxial to and
inside the helix as shown.

straight wire

wire helix

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

In (3.26), we should replace ω[l] by the matrix [x], where

 
ωl11
 1−ω 2 /ωc2 0 
[x] = 



0 ωl22

while [b] = ω[c] and [c] is given as (3.27) as usual. Find the values of the complex propagation
constants γ1,2 for this system of coupled transmission lines. Note that in this example, we
are not guaranteed that γ will be pure imaginary, or even real. Plot the values of γ1,2 (both
real and imaginary parts) versus ω/ωc for 0 < ω/ωc < 4 if ωc = 2π × 106 sec−1 ,

1
√ = 3 × 108 m/sec
l22 c22

l11 c11
=3
l22 c22
and
l11 c212
= 0.2
l22 c222

100
p3-10 On a uniform classical transmission line, some distributed external voltage Eext (z) and
current Hext (z) sources are located between z1 and z2 . Suppose that
dF
Eext (z) = + zG
dz
and
dG
Hext (z) = + yF
dz
for some functions F (z) and G(z) which vanish outside the interval z1 < z < z2 . Show that
these sources are non-radiating, in the sense that V (z) and I(z) on the transmission line are
identically equal to zero for z < z1 and for z > z2 . For the case where

F = F0 sin βz; G = G0 sin βz

and F0 and G0 are constants, plot the functions |V (z)| and |I(z)| versus z between z1 = 0 and
z2 = π/β (you may assume any convenient values for the parameters to make your plots).
p3-11 Consider the transmission line of problem p1-3, but containing two coupled helical conduc-
tors as well as the ground conductor. In (3.28), use [z] = j[x], where
 
ωl11
 1−ω 2 /ωc2 0 
[x] = 



ωl22
0 1−ω 2 /ωc2

while [y] = jω[c] and [c] is given as in (3.27) as usual.

Find the values of the complex propagation constants γ1,2 for this system of coupled trans-
mission lines. Note that in this example, we are not guaranteed that γ will be pure imagi-
nary, or even real. Plot the values of γ1,2 (both real and imaginary parts) versus ω/ωc for
0 < ω/ωc < 4, assuming that: ωc = 2π × 106 sec−1 , l11 = l22 ,
1
√ = 3 × 108 m/sec
l11 c11
c11
= 1.2
c22
and
c12
= 0.2
c22
p3-12 Consider a multiconductor transmission line that is lossy in general:

Re [I]† [z][I] ≥ 0


for any vector [I]

Re [V ]† [y][V ] ≥ 0


for any vector [V ]
If a mode [Vm ], [Im ], γm of this line obeys (3.32) with Re(γm ) = αm 6= 0, show that the
time-average power
†
   
1 †

1 † [z] 1 † [y]
P̂av = Re [Im ] [Vm ] = Re [Im ] [Im ] = Re [Vm ] ∗ [Vm ]
2 2 γm 2 γm

carried by this mode is positive if αm > 0 [this is the analog of the positive value of (1.16)
for a single transmission line].

101
102
Chapter 4

PULSE PROPAGATION AND

DISTORTION

And Time can crumble all, but cannot touch

the book that burns, faster than we can read.
—Michael Roberts,
The World’s End

4.1 Introduction
There is, of course, no such thing as a continuous wave (CW) time-harmonic signal which lasts from
time infinitely past to infinitely in the future. What we deal with in practice (whether in a circuit, on a
transmission line or in a waveguide) are signals which are of finite duration in time. That is, the signal
has a beginning and an end. It is of interest to determine how such real signals propagate and distort
when passing through such systems.

4.2 The Fourier Transform and the Analytic Signal

A wide class of functions f (t) of time t can be described in terms of a frequency spectrum F (ω) via the
Fourier transform (see Appendix A):
Z ∞
1
f (t) = F (ω)ejωt dω (4.1)
2π −∞
Z ∞
F (ω) = f (t)e−jωt dt (4.2)
−∞

If the complex conjugate of a second function g(t) is multiplied by f and integrated over all times t, we
have what is known as Parseval’s relation:
Z ∞ Z ∞
∗ 1
f (t)g (t) dt = F (ω)G∗ (ω) dω (4.3)
−∞ 2π −∞

A signal f (t) (it can be a voltage, a current, an electric or magnetic field, etc.) must be a real
function of time t. Because it is real, a constraint is imposed on its Fourier transform F (ω) by virtue of
(4.2):
F (−ω) = F ∗ (ω) (4.4)

103
for real ω, where ∗ denotes the complex conjugate. As a result, the information about F (ω) at negative
frequencies is redundant, and (4.1) can be written as an integral over positive frequencies only:

f (t) = Re[fA (t)] (4.5)

where ∞
1
Z
fA (t) ≡ F (ω)ejωt dω = |fA (t)|ejφA (t) (4.6)
π 0

is known as the analytic signal corresponding to f (t), a complex function whose magnitude is |fA (t)|
and whose phase is φA (t). Comparing its definition (4.6) with (4.1), we see that the analytic signal
has a Fourier transform FA (ω) which is zero for negative frequencies and equal to 2F (ω) for positive
frequencies.
The energy or power associated with a signal f (t) is proportional to the square of the signal, f 2 (t).
Indeed, applying the signal to a square-law detector such as a diode produces an output signal propor-
tional to this energy or power. In terms of the analytic signal, some algebraic manipulation applied to
the square of (4.5) gives
1 1
f 2 (t) = |fA (t)|2 + Re[fA2 (t)] (4.7)
2 2
Under some conditions (for example, the case of a narrowband signal considered in section 4.2.2 below),
the signal f 2 (t) can be passed through a filter which will eliminate the second term of (4.7):1
1
f 2 (t) filtered = |fA (t)|2
 
(4.8)
2
We see that in such a case the magnitude of the analytic signal is closely related to the output of a filtered
detector (a so-called envelope detector ). Such a detector is used to demodulate amplitude-modulated
(AM) signals.
If the signal is differentiated with respect to t (this can be accomplished by applying the signal
voltage to a series capacitor C followed by a shunt resistor R, provided that the RC time constant is
small compared to times characteristic of the variation of the signal) before being detected and filtered,
the output will be
 2
 ′ 2 1 1 d 1
[f (t)] filtered = |fA′ (t)|2 = |fA (t)| + |φ′A (t)|2 |fA (t)|2 (4.9)
2 2 dt 2

Thus, an output dependent on the phase (or at least its time derivative) of the analytic signal can also
be readily produced by such a slope detector circuit. If |fA (t)| is constant, then the output of this slope
detector is proportional to |φ′A (t)|2 alone. This is the principle of demodulation of frequency-modulated
(FM) signals.
The analytic signal is a useful tool for handling modulated signals. For example, let r(t) be a
given complex function of time (not necessarily an analytic signal itself), with Fourier transform R(ω).
Consider the signal
f (t) = Re r(t)ejω0 t
 
(4.10)
where ω0 is a nonnegative frequency called the carrier frequency, and r(t) is called the baseband signal.
The Fourier spectrum of f (t) is readily computed:
1
F (ω) = [R(ω − ω0 ) + R∗ (−ω − ω0 )] (4.11)
2

1 This is analogous to taking the time-average value of the power or energy of a time-harmonic signal in a circuit (see

Appendix E.3).

104
 |F(ω)|2

ω01 ω02 ω

Figure 4.1: Signal spectrum concentrated at two different frequencies.

(note that (4.11) obeys (4.4) as it must). Then using the Fourier frequency integral representation for
r(t) in terms of R(ω), we find from (4.6) that the analytic signal corresponding to f (t) is
Z ∞
1
fA (t) = [R(ω − ω0 ) + R∗ (−ω − ω0 )] ejωt dω (4.12)
2π 0

We next ask whether it is always possible to find a baseband envelope function r(t) for any given
arbitrary signal f (t). The answer is yes, but the resultant r(t) is not unique. A natural choice for it
would be
r(t) = e−jω0 t fA (t) (4.13)
but we are left to determine what ω0 should be. A potential approach would be to choose ω0 to be
where “most” of the spectrum F (ω) of the signal is concentrated. For example, this might be
R∞
ω|F (ω)|2 dω
ω0 = R0 ∞ (4.14)
0
|F (ω)|2 dω

But this might not be very representative of the actual signal. If, for example, the spectrum F (ω) were
concentrated around two widely separated frequencies ω01 and ω02 as shown in Fig. 4.1, eqn. (4.14)
would predict a value of ω0 somewhere between ω01 and ω02 , where almost no spectral content of the
signal is located. In such a case, it is preferable to split the signal as f (t) = f1 (t) + f2 (t), such that
the spectrum of each term is concentrated around a single frequency. Thus, a fairly general form for an
arbitrary signal might be "N #
X
jω0i t
f (t) = Re ri (t)e (4.15)
i=1

for suitably chosen values of ω0i . However the ω0i are selected, we observe that the spectrum of the
analytic signal corresponding to each term on the right side of (4.15) will be:
 
∗
 
 2Fi (ω) = Ri (ω − ω0i ) + Ri (−ω − ω0i ) for ω > 0 
 
FAi (ω) = (4.16)
 

 0 for ω < 0 


4.2.1 Energy moments and pulse widths of signals

Equation (4.14) is an example of a “moment” of a pulse, which is a way of characterizing the time or
frequency behavior of a signal of general form. Although exact or numerical evaluation of the Fourier

105
transform of a pulse can give us detailed information, it is often sufficient to know only a certain few
parameters of the pulse in order to understand how it is distorted after passing through a network or
propagation channel.
Two of the most important features of a pulse are the “center” and “duration”, which can be defined
in terms of its energy moments. The definition of the energy moments of a pulse depends on the notion
of a weighted average of a function with respect to the energy in the pulse. Thus, if x(t) is a function of
time (not necessarily real), we define its average value with respect to the energy in a pulse f (t) to be:
R∞
x(t)|f (t)|2 dt
hxif ≡ −∞R∞
2
(4.17)
−∞ |f (t)| dt

provided the integrals exist. Similarly, if X(ω) is the Fourier transform of a function x(t) of time, then
its average value with respect to the energy in the spectrum F (ω) of the pulse f (t) is defined as
R∞
X(ω)|F (ω)|2 dω
hXiF ≡ −∞ R∞
2
(4.18)
−∞ |F (ω)| dω

It is to be noted that the energies of the time-domain and frequency-domain versions of a pulse are
related by taking f = g in Parseval’s theorem (4.3):
Z ∞ Z ∞
2 1
|f (t)| dt = |F (ω)|2 dω (4.19)
−∞ 2π −∞

Evidently, ω0 as given by (4.14) is equal to hωiF .

The nth-order temporal moments of a pulse f (t) are defined as htn if . The first-order temporal
moment is called the center tc (f ) of the pulse:

tc (f ) ≡ htif (4.20)

This definition is analogous to that of the expected value of a quantity in quantum mechanics or the mean
value of a random variable in probability theory. Now, since (4.2) implies that the Fourier transform of
tf (t) is jdF (ω)/dω, we have from (4.3) that the center of a pulse can also be expressed by the ratio:
R∞ dF (ω) ∗
j −∞ dω
F (ω) dω
tc (f ) = R∞ (4.21)
−∞
|F (ω)|2 dω

If F (ω) 6= 0, and we express it in terms of amplitude and continuous phase F (ω) = |F (ω)|e−jφF (ω) , we
can express tc (f ) in terms of a spectral average value:

1 dF (ω)
tc (f ) = −hIm iF = hφ′F (ω)iF (4.22)
F (ω) dω

where
1 dF (ω)
−Im = φ′F (ω) = tg,F (ω) (4.23)
F (ω) dω
represents an intrinsic group delay time of the portion of the spectrum at the frequency ω (see (4.32)
below). The pulse center is thus a spectral average of this intrinsic group delay time.
We can by similar means define a pulse duration or pulse width for an arbitrary signal f (t). We do
this by using the second-order moment of the signal, namely
R∞
[t − tc (f )]2 |f (t)|2 dt
Td (f ) ≡ −∞ R ∞
2
2
(4.24)
−∞ |f (t)| dt

106
so that Td (f ) itself will be called the pulse duration. It will be seen that the quantity Td2 is analogous to
the variance of a random variable. The integrals in the numerator of (4.24) can be expressed as integrals
over the spectrum F (ω), as was done for tc in (4.21). We find:

Td2 (f ) = h[tg,F (ω)]2 iF − t2c (f ) (4.25)

We should note that in spite of the apparent generality of these results, the moment description
may not always be useful in practice. For example, if the “pulse” consists of two separate disturbances
widely separated in time (analogous to the separation of frequencies in the case of Fig. 4.1), the center
of the pulse defined by (4.20) will be located somewhere between the two disturbances, where there is
little signal at all. Likewise, the duration (4.24) of this signal will be much wider than that of either
of the individual disturbances which constitute the signal. It is best to resolve a general signal into a
superposition of pulses (similar to (4.15)), each of which is identifiably a single “bump,” and then to
apply the method of pulse moments to each term separately.
As the foregoing example illustrates, definitions of pulse center and duration based on the energy
moments are not always the most suitable ones in practice. This is particularly true of digital signals, for
which the energy distribution of a pulse is not as important as the voltage level relative to the threshold
values for certain logic gates. For these signals, concepts such as the rise and fall times of a signal are at
least as important as the duration, and more so than the center. They are also best defined using not
|f (t)|2 but f (t) itself or perhaps the envelope r(t). We will employ a particular case of this idea later
when developing the quasi-Gaussian approximation for evaluating pulse distortion.

4.2.2 Band-limited signals

As we have seen, not every complex function r(t) can be multiplied by a single factor ejω0 t to yield
an analytic signal. One important class of envelope functions which can is that for which R(ω) is
bandlimited; that is,
R(ω) = 0 for |ω| > ωb
where ωb > 0 denotes the bandwidth of r(t). Then if ω0 > ωb , we have R∗ (−ω − ω0 ) = 0 for ω > 0, and
the analytic signal (4.12) simplifies to
fA (t) = r(t)ejω0 t (4.26)
This example is thus analogous to the reconstruction of a sinusoidal function of time from a time-
independent phasor quantity by multiplying by ejωt and taking the real part. The function r(t) serves
as a complex envelope to modulate the carrier ejω0 t . If r(t) = |r(t)|ejφr (t) , then the signal (4.10) is said
to be an amplitude-modulated signal if φr (t) is constant, while if |r(t)| is constant the signal is said to
be frequency- or phase-modulated. In general, both kinds of modulation may be present simultaneously.
The energy (4.7) in a band-limited signal is thus given by

1 1
f 2 (t) = |r(t)|2 + Re[r2 (t)e2jω0 t ] (4.27)
2 2
The Fourier spectrum of each term on the right side of (4.27) can be found as follows. For the first term,
we write
Z ∞Z ∞
1
|r(t)|2 = R(ω1 )R∗ (ω2 )ej(ω1 −ω2 )t dω1 dω2
4π 2 −∞ −∞
Z ∞Z ∞
1
= R(ω1 )R∗ (ω1 − ω)ejωt dω1 dω (4.28)
4π 2 −∞ −∞
Z ∞ Z ∞ 
1 1
= R(ω1 )R∗ (ω1 − ω) dω1 ejωt dω
2π −∞ 2π −∞

107
where the change of variable ω = ω1 − ω2 was made. By comparison of (4.28) to (4.1), we see that the
Fourier transform of |r(t)|2 is Z ∞
1
R(ω1 )R∗ (ω1 − ω) dω1
2π −∞
from which we conclude that if R(ω) ≡ 0 for |ω| > ωb , then the spectrum of |r(t)|2 vanishes for |ω| > 2ωb .
In a similar fashion, we find that the spectrum of the second term on the right side of (4.27) lies in the
range 2(ω0 − ωb ) < ω < 2(ω0 + ωb ). Therefore if 2ωb < 2(ω0 − ωb ) (that is, ω0 > 2ωb ), then the spectra
of the terms on the right side of (4.27) do not overlap, and the second can be removed in principle by
passing f 2 (t) through a low-pass filter. A process like this is the basis of most practical demodulation
schemes.

4.3 Narrowband Pulse Distortion by a Linear Time-Invariant

Network
The basic ideas of pulse propagation on a transmission line or waveguide can be found in the analysis
of the simpler problem of predicting the distortion of a pulse in a linear, time-invariant but possibly
dispersive network. Consider a network for which we define two signals (each of which may be a voltage,
a current, a field component, or any linear combination of such quantities): the input f (t) and the
output w(t). Because the network is linear and time-invariant, the Fourier transforms of the input and
output are related by some complex transfer function H(ω):

W (ω) = H(ω)F (ω) (4.29)

The transfer function depends on the properties of the network, but not on the input or output signal.
It will prove convenient in what follows to write the transfer function in an exponential form:

H(ω) = A(ω)e−jΘ(ω) (4.30)

where A(ω) and Θ(ω) are real, continuous functions of frequency. Here A(ω) is called the amplitude of
the transfer function, while Θ(ω) is the phase delay of the transfer function. Note that this representation
is not necessarily the usual polar decomposition of the complex function H(ω), because in that expression
the amplitude is always nonnegative. In the version used here, A(ω) may be negative, so that if it passes
through a zero Θ(ω) may remain continuous. The DC response H(0) must be real, and for definiteness
we define Θ(0) = 0, so that A(0) = H(0), and define the functions A(ω) and Θ(ω) at other frequencies
by analytic continuation from ω = 0.

4.3.1 Quasistationary approximation

Suppose now that F (ω) is narrowband, in the sense that it is significantly different from zero only near a
nonnegative carrier frequency ω0 (and of course, near −ω0 as well, but since we will be using the analytic
signal negative frequencies will not need to be considered). Near ω0 , the phase of the transfer function
has a predominantly linear dependence of its phase on frequency, while the attenuation can be regarded
as approximately constant. We will emphasize these features of the transfer function by introducing a
new function Θd (the distortion part of Θ) according to

Θ(ω) ≡ ω0 tp0 + (ω − ω0 )tg0 + Θd (ω) (4.31)

The first and second terms come from the first two terms of the Taylor series expansion of Θ(ω) about
ω = ω0 . Here we have abbreviated the values of Θ and its derivatives at ω0 as

Θ0 ≡ Θ(ω0 ) ≡ ω0 tp0
Θ′0 ≡ Θ′ (ω0 ) ≡ tg0 (4.32)

108
and so on. The real quantity tp0 is called the phase delay time and tg0 the group delay time of the
transfer function at the frequency ω0 . We use a similar notation for A and its derivatives at ω0 . The
definition of Θd is such that it and its first derivative with respect to ω at ω0 are equal to zero. This
assures that Θd (ω) is in some sense much more slowly varying than Θ(ω) as a function of frequency near
ω0 .
The analytic signal of the output is now:

1 ∞
Z
wA (t) = H(ω)F (ω)ejωt dω (4.33)
π 0

If F (ω) is given by (4.11) and corresponds to an analytic signal of the form (4.26), then

1
F (ω) = R(ω − ω0 )
2
and we can express (4.33) as
∞
1 jω0 τp
Z
wA (t) = e Hd (ω0 + u)R(u)ejuτg du (4.34)
2π −∞

where
τp ≡ t − tp0 (4.35)
is the phase-delayed time,
τg ≡ t − tg0 (4.36)
is the group-delayed time, and we have made the change of integration variable ω = ω0 + u. The function

Hd (ω) ≡ A(ω)e−jΘd (ω) (4.37)

is the portion of the transfer function responsible for distorting the envelope: A(ω) distorts the amplitude,
while Θd (ω) distorts the phase.
Since the most significant contributions to the integral in (4.34) come from frequencies close to ω0 ,
it is appropriate to expand Hd into a Taylor series about this frequency, and integrate term by term.
Using the envelope representation (4.26) for fA , the result is
∞
1 jω0 τp X Hdn ∞ n
Z
wA (t) = e u R(u)ejuτg du
2π n=0
n! −∞
∞  n
X Hdn d
= ejω0 τp −j r(τg )
n=0
n! dτg
∞
X Hdn dn r(τg )
= ejω0 τp (−j)n (4.38)
n=0
n! dτgn

in which we have abbreviated the derivatives of Hd at ω0 as:

dn Hd (ω) 

Hdn = (4.39)
dω n ω=ω0

In terms of the quantities A0 , A′0 , A′′0 , Θ′′0 , etc., the first few coefficients (4.39) are given by:

Hd0 = A0 (4.40)

Hd1 = A′0 (4.41)

109
 Hd2 = A′′0 − jA0 Θ′′0 (4.42)
Equation (4.38) is called the quasi-stationary expansion for the distortion of a signal by a linear time-
invariant system.
Using these results, we can write out the first three terms of the expansion (4.38) as:
 
1
wA (t) ≃ ejω0 τp A0 r(τg ) − jA′0 r′ (τg ) − [A′′0 − jA0 Θ′′0 ] r′′ (τg ) (4.43)
2

The simplest approximation is obtained by keeping only the first term in curly brackets on the right side
of (4.43):
wA (t) ≃ A0 ejω0 τp r(τg ) (4.44)
To this order of precision, the output is described by an analytic signal whose phase is delayed by the
phase delay time tp0 of the network at the carrier frequency, but whose envelope is delayed by the group
delay time tg0 . An additional change of amplitude by the factor A0 also takes place, but no distortion of
the shape of the signal occurs. Despite the relative shift between phase and envelope, when the output
signal is applied to an amplitude detector, the result will be a scaled and time-delayed version of what
we get when the same is done to the input. The other terms in the curly brackets of (4.43) cause a
distortion of the shape of the input pulse. The kind and amount of distortion depends on the form of the
input pulse, as well as on the parameters A′0 , A′′0 and Θ′′0 which are characteristics of the linear system.

4.3.2 Example: AM Gaussian pulse (quasistationary approximation)

Consider an input signal which has a Gaussian amplitude envelope:
2
/2T 2
r(t) = e−t (4.45)

where the positive, real constant T is the characteristic duration of the pulse (for |t| > 2T , the amplitude
is less than 14% of its maximum value at t = 0, and the energy is less than 2% of its maximum value).2
If ω0 T ≫ 1, then r(t)ejω0 t is to a high degree of accuracy an analytic signal. The Gaussian pulse closely
approximates the output of many practical sources, notably lasers, so its study is of importance.
We will suppose for simplicity that the transfer function has constant amplitude, so that A′0 = A′′0 = 0,
and further assume that A(ω) = A0 = 1. Then the analytic signal of the output is given by (4.43) as
" !#
jω0 τp j∆0 τg2 2 2
wA (t) ≃ e 1+ 2
−1 e−τg /2T (4.46)
2 T

where
Θ′′0
∆0 = (4.47)
T2
is called the quadratic dispersion parameter of this pulse due to its passage through the linear system.
Examination of this expression reveals that two things have happened to the signal:
h  2 i
τ
1. The output phase is ω0 τp + tan−1 ∆20 Tg2 − 1 . This means that the carrier frequency ω0 has
been “chirped”: the time-dependent phase factor exp(jω0 t) has changed to a nonlinear function
of time, a kind of frequency modulation. This will be important if an FM detector is applied to
the signal.

2 Note that a calculation shows that if f is the Gaussian pulse corresponding to (4.45), then the pulse duration as given

by (4.24) is Td (f ) = T .

110
 2. The shape of the amplitude envelope of the pulse has changed. The output power envelope |wA (t)|2
(which is obtained as a result of AM detection) is
 !2 
2 2
τg

2 2 ∆0
|wA (t)|2 = e−τg /T 1 + −1 
2 T2

Note that the output maximum (1 + ∆20 /4) is larger than that of the input pulse, in violation of
conservation of energy. This is because wA itself has only been computed accurate to terms of
order ∆0 , and so can only be expected to obey energy conservation up to that order of accuracy.
More terms of (4.43) would need to be kept in order to alleviate this difficulty.

The major effect of nonlinear dependence of the phase of the transfer function on frequency has been
to change the duration of the pulse to a new value Tout . But because of the failure of this approximation
to obey energy conservation as noted above, it is not easy to characterize succinctly the value of Tout .
Since information is often sent through networks as a series of pulses, it is clear that adjacent output
pulses will have to be separated by at least Tout in order to prevent pulse overlap and loss of data. The
network thus imposes a maximum data rate of about 1/Tout pulses per second for this type of input
pulse. We therefore need a more accurate means of evaluating Tout .

4.3.3 Quasi-Gaussian approximation

Because the output of many narrowband pulse sources is approximately Gaussian, it is worthwhile to
investigate the distortion of such pulses by a method which makes use of the particular properties of
such pulses. Let us consider again the envelope (4.45), but now allow for both phase and amplitude
variations of this envelope by considering that T may be complex. In order to have a realistic pulse
with finite energy, this envelope must decay as t → ±∞. Thus, we require that Re(1/T 2 ) > 0, or
equivalently |Re(T )| > |Im(T )|. From the table of Fourier transforms in Appendix A, we find that the
Fourier transform of (4.45) is
√ 2 2
R(ω) = T 2πe−ω T /2 (4.48)
Now if r(t) is a more general pulse, but is still narrowband with a single maximum of its spectral
amplitude |R(ω)| occurring at ω = 0, it is reasonable to assume that its Fourier transform can be
approximated for small enough |ω| by a function of the form
√ 2 2
R(ω) ≃ R0 T0 2πe−ω T0 /2 (4.49)

where R0 and T0 are complex constants. To match the value of the spectrum R(ω) exactly at ω = 0, we
set R0 equal to
R(0)
R0 = √ (4.50)
T0 2π
Matching the second derivative of the spectrum at ω = 0 allows determination of T :
R∞ 2
R′′ (0) t r(t) dt
2
T0 = − = R−∞
∞ (4.51)
R(0) −∞ r(t) dt

in which the second equality follows from differentiating the Fourier integral directly. Note that, in
addition to |Re(T0 )| > |Im(T0 )|, the condition R(0) 6= 0 is also required for (4.49) to be a valid approxi-
mation. Narrowband pulses whose spectrum vanishes at ω = 0 can be handled by a modification of this
procedure, which we leave as an exercise for the reader.
Now let us re-examine the procedure which led to the quasistationary approximation. In order
to obtain a different approximation which more closely obeys energy conservation, we will make an

111
approximation to exp[−jΘd (ω0 + u)] which has a magnitude of one (in contrast to the Taylor series
expansion, which does not):
2 ′′
e−jΘd (ω0 +u) ≃ e−ju Θ0 /2 (4.52)
while the amplitude of the transfer function is treated in the same way as before:
1
A(ω0 + u) ≃ A0 + uA′0 + u2 A′′0 (4.53)
2
Then the analytic signal of the output pulse corresponding to the input signal envelope r(t), as approx-
imated by (4.49), is approximately
√
R0 T0 2πejω0 τp ∞ juτg 1
Z
2 2 ′′
wA (t) ≃ e [A0 + uA′0 + u2 A′′0 ]e−u (T0 +jΘ0 )/2 du (4.54)
2π −∞ 2
By defining a new integration variable
s
T02 + jΘ′′0
v=u
T02
we can express (4.54) in terms of a “stretched” version of r and its derivatives:
s " s ! s s !
jω0 τp T02 T02 ′ T02 T02
wA (t) ≃ e 2 ′′ A0 r τg 2 ′′ − jA0 r′ τg
T0 + jΘ0 T0 + jΘ0 T0 + jΘ′′0
2 T0 + jΘ′′0
2
s !#
A′′0 T02 ′′ T02
− r τg (4.55)
2 T02 + jΘ′′0 T02 + jΘ′′0

Equation (4.55) is called the quasi-Gaussian approximation of the pulse distortion. If A′0 = A′′0 = 0,
then to this order, the pulse is stretched in the time coordinate and scaled in amplitude, but in some
sense still retains the essential features of shape.
The interpretation of the complex stretching factor in (4.55) is not obvious in general, and we must
examine particular pulse shapes to determine its significance. We will do this for the case of the AM
Gaussian pulse in the next subsection.

4.3.4 Example: AM Gaussian pulse (quasi-Gaussian approximation)

We return to the example of the pulse envelope (4.45) (with T taken to be real). Assuming again a
constant amplitude transfer function with A0 = 1, (4.55) gives
s s ! r
T 2 T 2 1 2 2
jω0 τp jω0 τp
wA (t) ≃ e r τg =e e−τg /[2T (1+j∆0 )] (4.56)
T 2 + jΘ′′0 T 2 + jΘ′′0 1 + j∆0

where ∆0 is the quadratic dispersion parameter defined in (4.47). In order to re-examine the phase and
amplitude of this signal, we express the “stretched” term in the exponent as:
τg2 τg2 1 − j∆0
− = −
2T 2 (1 + j∆0 ) 2T 2 1 + ∆20
which allows us to observe that:
1. The output phase is
1 τg2 ∆0
tan−1 ∆0 +
ω0 τp −
2 2T 2 1 + ∆20
Again we see that a “chirping” of the signal has occurred, in this approximation as a quadratic
function of time. This amounts to a frequency modulation that is linear in time.

112
 2. The output power envelope |wA (t)|2 is now
T −τg2 /Tout
2
|wA (t)|2 = e (4.57)
Tout
where we can now easily identify the width of the output pulse as
 ′′ 2
2 2 2 2 Θ0
Tout = T (1 + ∆0 ) = T + (4.58)
T

We note that the duration Tout of the output pulse has been increased by comparison with that of
the input pulse T . As a function of T , Tout becomes large both for small T (that is, very narrow input
pulses are severely broadened) and for large T (where, of course, we started out with a broad pulse).
For a given phase characteristic Θ′′0 p
of the transfer function, there
√ is anp
optimum value of T for which
Tout will be minimum: T = Topt = |Θ′′0 |, giving Tout,opt = 2Topt = 2|Θ′′0 |, the narrowest pulse of
this form that can emerge from the network.
Since information is often sent through networks as a sequence of such pulses, adjacent pulses have
to be separated in time by at least Tout in order to prevent overlap
p and loss of data. The network itself
thus imposes a maximum data rate of about 1/Tout,opt = 1/ 2|Θ′′0 | Gaussian pulses per second.

4.3.5 Change of pulse center and duration by a lossless network

The quasi-stationary or quasi-Gaussian expansion is useful in practice if only a few terms of the expansion
provide an accurate prediction of the distortion of the pulse. This in turn requires not only that the
pulse be narrowband, but also that the pulse shape is sufficiently simple that easy analytical expressions
can be computed, as is the case for the Gaussian pulse, for example. To predict the distortion of more
general pulses, we must either resort to numerical calculation of the relevant Fourier integral (4.34) or
content ourselves with less complete information about the output pulse which will nevertheless be useful
in the design of communication systems. In this subsection, we compute the time delay of the pulse
center and change in pulse duration of an arbitrary signal due to passage through a lossless network.
We do this using the definitions (4.21) and (4.24) based on the energy moments.
If an arbitrary pulse is input to a lossless linear time-invariant system whose transfer function is
H(ω) = exp[−jΘ(ω)], then the total phase of the output spectrum W (ω) will be −φF (ω) − Θ(ω), while
|W (ω)| = |F (ω)|. Thus, the center of the output pulse w can be written as
tc (w) = hφ′F (ω) + Θ′ (ω)iF = tc (f ) + tg,av (f ) (4.59)
where
tg,av (f ) ≡ htg (ω)iF (4.60)
is the average value of the group delay time tg (ω) = Θ′ (ω), weighted by the spectral energy |F (ω)|2 at
each frequency ω. In other words, the center of the output pulse is delayed from the center of the input
pulse by the average group delay time tg,av (f ). The result (4.59) is exact for a pulse of arbitrary shape
(narrowband or not), provided the system transfer function is lossless. If the pulse is narrowband about
a carrier frequency ω0 , we have tg,av (f ) ≃ tg0 as is obtained in the quasistationary and quasi-Gaussian
approximations.
The duration of the output signal is given by
Td2 (w) = Td2 (f ) + 2C(tg,F , tg ) + h[tg (ω)]2 iF − htg (ω)i2F (4.61)
where C(tg,F , tg ) is a sort of covariance function between the group delay time of the input signal and
that of the transfer function:
C(tg,F , tg ) ≡ htg (ω)tg,F (ω)iF − htg (ω)iF htg,F (ω)iF (4.62)
which may be either positive or negative.

113
4.4 Propagation and Distortion of Pulses; Group Velocity
The results of the previous section can be directly applied to the propagation of pulses on a transmission
line. Suppose that a forward voltage wave is launched on a semi-infinite classical transmission line (in
z > 0) by connection of an ideal voltage source vG (t) = v(0, t) at the terminals z = 0. This is taken to
be the input signal. If the output signal is chosen to be v(z, t), the voltage at a point z and time t on
the line, then the transfer function will be

H(ω) = e−γ(ω)z = e−α(ω)z−jβ(ω)z

as is appropriate for a forward-traveling wave. The results of the previous section apply, with A0 = e−α0 z ,
Θ0 = ω0 tp0 = ω0 z/vp0 , A′0 = −α′0 zA0 , Θ′0 = tg = β0′ z = z/vg0 , etc. The subscripts 0 denote quantities
to be evaluated at ω = ω0 , while vp and vg denote the phase and group velocities respectively, defined
in (1.11) and (1.12). If only the first term in the curly brackets of (4.43) is retained, we have for the
analytic signal of the output pulse

vA (z, t) = e−α0 z ejω0 (t−z/vp0 ) r(t − z/vg0 ) (4.63)

We see from this result that the group velocity vg0 at the center or carrier frequency of the pulse is the
velocity at which the envelope function r(t) travels in this approximation, while the phase of the carrier
is traveling at the phase velocity vp0 .
Let us emphasize that (4.63), along with the concepts of group velocity and phase velocity, makes
sense only if the approximations leading to this result are valid. That is, the neglected terms in (4.43)
or (4.55) must be small. As z increases, terms such as Θ′′0 = β0′′ z will become significant, and distortion
of the input pulse will begin to be evident. Thus β ′′ (ω0 ) in particular is often used as a measure of what
is called the dispersive property of the transmission line. Of course, the other derivatives of α and β will
also influence pulse distortion, and must be accounted for in more accurate predictions of this effect.
When the results (4.56)-(4.58) for a Gaussian input pulse are applied to propagation along a lossless
transmission line, we find that the characteristic duration of the pulse changes from T to
q
T 2 + (β0′′ z)2 /T 2 ≡ Tout (z) (4.64)

Initially, the pulse duration is nearly unchanged, but as z increases, the duration increases. This will
put an upper limit on the distance a pulse may travel in a dispersive transmission line before overlap
with adjacent pulses occurs. At this point, a repeater capable of reshaping the pulse would have to be
inserted into the transmission line if resolvable pulse propagation over longer distances is desired.

4.4.1 Propagation of arbitrary pulses

The energy moment description of pulse distortion can be applied to the propagation of arbitrary pulses
along a transmission line. We replace
z
tg (ω) →
vg (ω)
and we find from (4.59) that the center of the pulse (now a function of z) is given by
z
tc (w) = tc (f ) + (4.65)
vg,av (f )

where
1 1
=h iF (4.66)
vg,av (f ) vg (ω)
is an average group velocity for the signal on the line, and is the velocity with which the center of the
pulse travels. This is true for any signal f , but if f is a narrowband signal with spectrum concentrated

114
around ω0 , we see that the average group velocity becomes approximately vg0 , in agreement with our
earlier result (4.63).
The pulse duration, from (4.61), is obtained from:
( 2 )
2 1 1 1
Td,out (z) = Td2 (f ) + 2zC(tg,F , ) + z 2 h iF − h i2 (4.67)
vg vg (ω) vg (ω) F

Thus, the square of the pulse duration is a quadratic function of z (a result we obtained approximately
for a narrowband pulse in (4.64)). Eventually, the duration increases linearly with z if z is large enough.
Note that if C(tg,F , v1g ) < 0, the pulse duration will initially decrease to some minimum value before
beginning to increase again. This condition, which involves selecting an appropriate phase characteristic
of the input pulse spectrum relative to the frequency dependence of the group velocity, can be used to
counteract the natural tendency of pulses to spread and overlap with adjacent ones in a pulse stream.
The technique is called chirping, because it involves a frequency modulation of the input pulse’s carrier
frequency.

4.5 Multimode Broadening of Narrowband Pulses

In recent years, various kinds of multimode waveguides (most notably in the optical frequency band)
have been used in practical communication systems. At optical frequencies especially, the relative
deterioration in bandwidth this entails is more than offset by the higher carrier frequency as compared
to a microwave system. In this section, we will examine the effect of multimode propagation on the
temporal distortion of a pulse.
In contrast with single-mode waveguide systems, the signal detector in a multimode waveguide is
usually not deeply concerned with the relative phases or other details of the portions of the signal
appearing in each mode. The modes are excited in large numbers, and somewhat indiscriminately, so
the detection process obtains only information about the sum of the powers contained in the modes.
The primary distorting effect is that each mode travels with a different group velocity from the others,
and the result seen at the detector is a considerably broadened version of that produced by the source.
Pictorially, the situation is as shown in Fig. 4.2. Denote by |gAm (z, t)|2 the detected output pulse present
in the mth mode of the waveguide. We assume at first that an equal amount of energy is imparted to
each mode at z = 0 as shown in Fig. 4.2(a). As z increases, in the simplest approximation the pulse in
each mode will travel undistorted but delayed by the group delay tgm = z/vgm of the mode (Fig. 4.2(b)).
The spread in the pulse centers of the various modes is then the difference between the largest and
smallest mode group delays:
 
1 1
Td = (tgm )max − (tgm )min = z −
vg,min vg,max

This is called the delay difference between the slowest and fastest modes, often expressed as delay
difference per unit length, Td /z. In many waveguides, the group delays per unit length of the individual
modes are closely and roughly evenly spaced. Natural filtering effects of the detector tend to smooth out
the ripple which would otherwise be expected when contributions from all modes are added together.
Thus, the response of the detector at the output of a multimode waveguide will appear as in Fig. 4.2(c).
If the original pulse width at z = 0 is small compared to Td , the detected pulse at the output will have
a half-width of about Td /2.
This broadening of the pulse would appear to be much more severe than that which results from
dispersion of a single mode. However, there are several factors which alleviate the multimode distortion
effects to some extent. For one thing, some modes are usually excited with larger amplitudes than other
modes, and the amplitudes often vary slowly with mode number m. Also, the attenuation constant αm
of the mth mode is frequently larger for the modes that are excited poorly, so the amplitudes of these

115
 |gAm(0,t)|2
all modes

t
(a)

|gAm(z,t)|2 m=1 m=2 m=M

• • • •

Td t
(b)
Σ |gAm(z,t)|2

t
(c)

Figure 4.2: (a) Mode intensities at z = 0; (b) mode intensities at z > 0; (c) total intensity at detector
for z > 0.

116
 |gAm(z,t)|2 m=1 m=2 m=M

• • • •

t
(a)

Σ |gAm(z,t)|2

t
(b)

Figure 4.3: Nonuniform mode amplitudes: (a) mode intensities at z > 0; (b) total intensity at detector
for z > 0.

modes are reduced even further when z > 0. In this case, we end up with a situation like that of Fig. 4.3,
with the reduction in some mode amplitudes serving to reduce the width of the detected pulse. Local
irregularities which cause inter-coupling between the modes can also reduce the width of the detected
pulse. The price one pays for the reduction in pulse width is the power that is lost in mode attenuation.

4.6 Wavefront Distortion

Modern communication systems make frequent use of information encoded in digital form, a series of
rectangular (or nearly so) pulses in the time domain. Digital signals are inherently wideband, as very
high frequency constituents are needed to reproduce the fast rise and fall times of such pulses. It is thus
important to be able to predict the distortion of step-like functions of time when passed through a linear
system.
For this purpose, it is more suitable to use the Laplace transform than the Fourier transform to
analyze transient behavior. Let us assume that all functions of time to be considered will be identically
zero for times t < 0. In other words, all functions will contain the Heaviside unit step function, defined
as

θ(t) = 1 (t > 0)
= 0 (t < 0) (4.68)

A rectangular voltage pulse of strength V0 and duration T , beginning at t = 0, can then be written as

v(t) = V0 [θ(t) − θ(t − T )] (4.69)

Other digital pulses can be expressed in similar fashion.

117
 The Laplace transform F (s) of a function f (t) is defined as
Z ∞
F (s) = f (t)e−st dt (4.70)
0

We see that it is similar to the Fourier transform (4.2) for functions which vanish when t < 0, if we
make the replacement of the real frequency ω by a complex frequency s = jω. If s is permitted to
have a positive real part, the integral in (4.70) will converge even for functions f which do not decay to
zero as t → ∞. Although there is an inversion formula for the Laplace transform analogous to (4.1) to
reconstruct f (t) from a knowledge of F (s), it requires integration in the complex s-plane and as such is
beyond the scope of this text. We will be content to reconstruct time-domain functions by reference to
a table of Laplace transforms, of which a few of the most relevant to our needs are given in Appendix A.
An approximate technique for calculating the distortion of a step function input to a linear system
whose transfer function is H(s) is based on Heaviside’s first expansion theorem, which is as follows.
Suppose the Laplace transform of a function f (t) can be expanded as a series in inverse powers of the
complex frequency s:
F1 F2
F (s) = F0 + + 2 + ... (4.71)
s s
as s → ∞, where F0 , F1 , . . . , are real constants. We identify the inverse transform of each term of this
series from Table A.2, and add the results to form f (t):
∞
X ti
f (t) = F0 δ(t) + θ(t) Fi+1 (4.72)
i=0
i!

As a simple example, consider a unit step function f (t) = θ(t) which is a voltage applied across a
series combination of a resistor R and a capacitor C. The output w(t) is to be the voltage across the
capacitor. Elementary circuit analysis shows that the transfer function will be
1
H(s) = (4.73)
1 + sRC
Then
1 1 1
W (s) = H(s)F (s) = H(s) = − + ... (4.74)
s RCs2 (RC)2 s3
Comparing to the expansion (4.71), we can identify the expansion coefficients for W as W0 = W1 = 0,
and
1
Wi = (−1)i (i ≥ 2) (4.75)
(RC)i−1
Using (4.72), we obtain
∞ i
ti

X
i+1 1 h i
w(t) = θ(t) (−1) = θ(t) 1 − e−t/RC (4.76)
i=1
RC i!

which is, of course, a well-known result showing how the initially sharp input step pulse is distorted by
the network.
The point to be recognized here is that even in more complicated examples where the series (4.72)
cannot be identified in simple closed form, we may compute a finite number of the initial terms of
the series and use them to approximate the distortion of the initial part of the pulse for small values
of t (the so-called wavefront ). In systems containing transmission lines or other delay elements, the
transfer function may have a large s expansion more complicated than (4.71), containing an additional
exponential factor:  
Hf 1 Hf 2
H(s) = e−sTf Hf 0 + + 2 + ... (4.77)
s s

118
where the constant Tf is the limiting value of the phase and group delay times as complex frequency
approaches infinity. When the input is again a unit step function, the transform of the output is
 
−sTf Hf 0 Hf 1 Hf 2
W (s) = e + 2 + 3 + ... (4.78)
s s s

and from (4.72) and the time-shift rule from Table A.3, we get
∞
X (t − Tf )i
w(t) = θ(t − Tf ) Hf i (4.79)
i=0
i!

which expresses the distortion of the wavefront after it is delayed by Tf . The initial output response is
often quite small relative to its value at later times. This initial part of the response is called a forerunner
or precursor to distinguish it from the main part of the output signal. Under certain conditions the
forerunner signal can be comparable in magnitude to the main signal, in which case the differences
in time delay can once again lead to undesirable interference in the digital information and bit error
conditions. The result is a reduction of attainable bit transmission rates in the system.

4.7 Moment Matching and AWE

While the wavefront (early-time) expansion (4.79) is accurate in predicting the initial shape of the output
pulse, as time increases it is necessary to include a larger and larger number of terms in the expansion
in order to maintain this accuracy. This makes (4.79) an inefficient means of forecasting the distortion
of the entire pulse. The trouble is that while the transfer function has been well approximated for large
values of s, no provision has been made for accuracy for small s, where information about the pulse
shape at later times is contained.
To handle this aspect of pulse behavior, let h(t) be the impulse response of the system [that is, the
output w(t) when the input f (t) = δ(t) is a delta function]. Then we write H(s) as the Laplace transform
of h(t), and expand the factor e−st as a power series in s to obtain the so-called moment expansion of
H(s):
Z ∞
H(s) = h(t)e−st dt
0
∞
(−s)n ∞
X Z
= h(t)tn dt
n=0
n! 0
∞
X
= mn s n (4.80)
n=0

where ∞
(−1)n
Z
mn = h(t)tn dt (4.81)
n! 0

are called the moments of h(t). Equation (4.80) is an expansion of H(s) about s = 0, complementing
the expansion (4.77) about s = ∞.
An approximation for the output of the system valid for late times t is obtained if we choose an
approximation Hap (s) to H(s) which matches that of H(s) itself as s → 0, and use it to compute W (s)
and thus w(t). To be useful, such an approximation should have an easily computed inverse Laplace
transform. We choose the form
q
" #
X R i
Hap (s) = e−sTf Hf 0 + (4.82)
i=1
s − pi

119
where Tf and Hf 0 have the same meaning as in (4.77), while the pi are a set of poles in the left half
of the complex s-plane, and Ri are the residues of these poles. It is straightforward to see that the
approximate impulse response corresponding to (4.82) will be
q
X
hap (t) = Hf 0 δ(t − Tf ) + θ(t − Tf ) Ri epi (t−Tf ) (4.83)
i=1

The approximate response wap (t) to a unit step function input f (t) = θ(t) will similarly be
q q
" ! #
X Ri X Ri pi (t−Tf )
wap (t) = θ(t − Tf ) Hf 0 − + e (4.84)
p
i=1 i i=1 i
p

We determine the 2q unknown constants Ri and pi in (4.82) by expanding Hap (s) about s = 0 and
s = ∞. We first obtain l equations by matching the terms proportional to s−1 , s−2 , . . . s−l as s → ∞.
Since the behavior of the s0 term is already matched, we get no new information from it. If l = 0, we
do not match any terms in negative powers of s. As s → ∞,
∞ q
" #
−sTf
X 1 X n
Hap (s) = e Hf 0 + Ri pi (4.85)
n=0
sn+1 i=1

so our possible matching conditions there are:

q
X
s−1 : Hf 1 = Ri (4.86)
i=1
Xq
s−2 : Hf 2 = Ri pi (4.87)
i=1

and so on up to the terms in s−l .

We obtain the remaining (2q − l) conditions necessary to determine all the constants by matching
the terms proportional to s0 , s1 , s2 , . . . s2q−l−1 as s → 0. In this limit (remembering that we must also
expand the term e−sTf in a power series in s),
q q q
! " ! #
X Ri X Ri X Ri
Hap (s) = Hf 0 − + s −Tf Hf 0 − − + ··· (4.88)
i=1 i
p i=1 i
p p2
i=1 i

so our possible matching conditions are:

q
X Ri
s0 : m 0 = Hf 0 − (4.89)
i=1
pi
q
X Ri
s1 : m1 = −Tf m0 − (4.90)
i=1
p2i

and so on up to the terms in s2q−l−1 .

This technique (especially when we choose l = 0), is called asymptotic waveform evaluation (AWE),
and has been widely applied to simplify the calculation of time responses of complicated systems. In
the simplest situation (q = 1), we may solve these equations explicitly (this is left as an exercise). If
q = 2, we already have to do some clever algebra to get the explicit solution, and for higher orders
of approximation these equations must be solved numerically to obtain the approximate response. We

120
will outline here the explicit calculation for the case q = l = 2 to show what is involved. Let us write
(4.86)-(4.87) in the matrix form  
 1 1 
  [R] = [Hf ] (4.91)
 
p1 p2

where    
 R1   Hf 1 
[R] = 

;
 [Hf ] = 



R2 Hf 2

Likewise, we express (4.89)-(4.90) as

 
1 1
 p1 p2 
  [R] = [M ] (4.92)
 
1 1
p21 p22

where    
   M1 
Hf 0 − m 0
[M ] = 

≡
 


−m0 Tf − m1 M2

Solving (4.92) for [R] and substituting into (4.91) we get

 
 p1 + p2 −p1 p2 
  [M ] = [Hf ] (4.93)
 
p21 + p1 p2 + p22 −p1 p2 (p1 + p2 )

Introducing the new unknown variables r1 = p1 + p2 and r2 = p1 p2 , we can after a little algebra solve
the system of equations (4.93) as

M 2 Hf 2 − M 1 Hf 1 M1 Hf 2 − Hf21
r1 = ; r2 = (4.94)
M2 Hf 1 − M12 M2 Hf 1 − M12

The poles are then obtained as solutions of a quadratic equation:

p
r1 ± r12 − 4r2
p1,2 = (4.95)
2
and finally the residues are obtained from the solution of (4.91):

Hf 2 − p 2 Hf 1 p 1 Hf 1 − Hf 2
R1 = ; R2 = (4.96)
p1 − p2 p1 − p2

As an example, consider the transfer function

√
s2 +ωc2
H(s) = e−Tf (4.97)

with an input taken to be the rectangular pulse given by (4.69). This transfer function can represent,
for example, the effect of a transmission line with a low-frequency cutoff of ωc and high-frequency time

121
delay Tf . The exact step function response of this system can be found from entries 4 and 8 of Table A.2
in Appendix A:  q  

Z t J1 ω c u 2 − T 2
f
wstep (t) = θ(t − Tf ) 1 − ωc Tf q du (4.98)
Tf u2 − Tf2

where J1 is a Bessel function (see Appendix B). The response to the rectangular input pulse (4.69) is
then
w(t) = V0 [wstep (t) − wstep (t − T )] (4.99)
and this is plotted in Figure 4.4. Note how there is considerable “sag” and some ringing in the output

w(t)
1

Exact
AWE: q=1, l=1
0.5 AWE: q=2, l=2

0
0 0.5 1 1.5 2
t, µsec
–0.5

–1

Figure 4.4: Response of the linear system (4.91) to a rectangular step input (4.69): V0 = 1, ωc = 4π ×106
sec−1 , Tf = 10−7 sec, T = 10−6 sec.

pulse after the initial arrival at t = Tf . Such deviations from ideal pulse shapes can lead to false triggering
of logic gates, or failure to trigger when expected, increasing the bit error rate of digital systems. The
first- (q = 1, l = 1) and second-order (q = 2, l = 2) AWE approximations to the output response are
also shown in Figure 4.4. We see that the second-order AWE gives a certain reproduction of the ringing,
but only for the first minimum, while the first-order approximation does not reproduce the ringing at
all. It is therefore important to use an AWE of sufficiently high order if accurate results are desired.

122
4.8 Notes and References
General treatments of modulation and detection are to be found in

Rowe (1965), chapters 3 and 4;

Thomas (1969), chapter 7.
Stremler (1990), chapters 5 and 6.

Relatively few books treat pulse propagation in much detail. See, for example,

Jackson (1975), chapter 7;

Javid and Brown (1963), chapter 13.

A nice overview, and English translations of the classic papers of Sommerfeld and Brillouin on the
subject are in:

Brillouin (1960).

This book deals mainly with a detailed study of the precursor problem, as do the following (but for
propagation channels with various dispersion laws):

Carslaw and Jaeger (1963), chapter 9;

Kuznetsov and Stratonovich (1964).

A more detailed and precise treatment of the Sommerfeld-Brillouin problem can be found in

Oughstun and Sherman (1994).

Detailed comparison of precursors to group-delayed main signals is made in

H. Hosono and T. Hosono, “Superluminal group velocities in passive media,” IEICE Trans.
Electron., vol. E87-C, pp. 1578-1585 (2004).

General treatments of pulse dispersion can be found in:

Vainshtein and Vakman (1983);

Vakman (1998);
Agrawal (2001), chapter 3.

as well as the papers:

L. A. Vainshtein, “Propagation of pulses,” Sov. Phys. Uspekhi, vol. 19, pp. 189-205 (1976).
D. E. Vakman and L. A. Vainshtein, “Amplitude, phase, frequency—fundamental concepts
of oscillation theory,” Sov. Phys. Uspekhi, vol. 20, pp. 1002-1016 (1977).

For more information on analytic signals, see:

Franks (1969), section 4.4;

Vainshtein and Vakman (1983);
Vakman (1998).

The quasi-stationary approximation (of which there are many variations) is usually attributed to
Carson and Fry in their treatment of FM distortion, but was actually anticipated over 50 years earlier
by Gouy. Many subsequent developments have been made, and we list only a small sample of the
literature here:

123
 M. Gouy, “Sur la propagation des ondes lumineuses, eu égard à la dispersion,” J. Math.
Pures Appl., ser. 3, vol. 8, pp. 335-356 (1882).
J. R. Carson and T. C. Fry, “Variable frequency electric circuit theory with application to
the theory of frequency-modulation,” Bell Syst. Tech. J., vol. 16, pp. 513-540 (1937).
F. L. H. M. Stumpers, “Distortion of frequency-modulated signals in electrical networks,”
Communication News, vol. 9, pp. 82-92 (1948).
J W. Head and C. G. Mayo, “The response of a network to a frequency-modulated input
voltage,” Proc. IEE (London), part C, vol. 105, pp. 509-512 (1958).
Baghdady (1961), chapter 19.
M. L. Liou, “Noise in an FM system due to an imperfect linear transducer,” Bell Syst. Tech.
J., vol. 45, pp. 1537-1561 (1966).
D. Anderson, J. Askne and M. Lisak, “Wave packets in an absorptive and strongly dispersive
medium,” Phys. Rev. A, vol. 12, pp. 1546-1552 (1975).

A collection of the classic papers on distortion of FM signals is:

Klapper (1970).
The quasi-Gaussian approximation is an extension of a method of treating the broadening of a
Gaussian pulse using a second-order Taylor series expansion of the propagation constant. This method
has been treated in many papers, of which a sample follows:

C. G. B. Garrett and D. E. McCumber, “Propagation of a Gaussian light pulse through an

anomalous dispersion medium,” Phys. Rev. A, vol. 1, pp. 305-313 (1970).
F. P. Kapron and D. B. Keck, “Pulse transmission through a dielectric optical waveguide,”
Applied Optics, vol. 10, pp. 1519-1523 (1971).
E. F. Kuester and D. C. Chang, “Single-mode pulse dispersion in optical waveguides,” IEEE
Trans. Micr. Theory Tech., vol. 23, pp. 882-887 (1975).
D. Marcuse, “Pulse distortion in single-mode fibers,” Applied Optics, vol. 19, pp. 1653-1660
(1980).

Third-order dispersion terms can be accounted for using Airy functions:

J. R. Wait, “Distortion of pulsed signals when the group delay is a nonlinear function of
frequency,” Proc. IEEE, vol. 58, pp. 1292-1294 (1970).
The analysis of the evolution of arbitrary pulses in terms of their moments is due to:
S. H. Moss, “Pulse distortion,” Proc. IEE (London), part IV, vol. 98, pp. 37-42 (1951).
L. C. Baird, “Moments of a wave packet,” Amer. J. Phys., vol. 40, pp. 327-329 (1972).
D. G. Anderson and J. I. H. Askne, “Wave packets in strongly dispersive media,” Proc.
IEEE, vol. 62, pp. 1518-1523 (1974).
D. Anderson, J. Askne and M. Lisak, “ The velocity of wave packets in dispersive and slightly
absorptive media,” Proc. IEEE, vol. 63, pp. 715-717 (1975).
H. M. Bradford, “Propagation and spreading of a pulse or wave packet,” Amer. J. Phys.,
vol. 44, pp. 1058-1063 (1976).
H. M. Bradford, “Propagation of a step in the amplitude or envelope of a pulse or wave
packet,” Amer. J. Phys., vol. 47, pp. 688-694 (1979).
G. Bonnet, “Au delà d’une vitesse de groupe: vitesse d’onde et vitesse de signal,” Ann.
Télécommun., vol. 38, pp. 345-366 and 471-487 (1983).
D. E. Vakman, “Evolution of the parameters of a pulse propagating with dispersion and
attenuation,” Sov. J. Commun. Technol. Electron., vol. 31, no. 7, pp. 100-105 (1986).
D. Anderson and M. Lisak, “Propagation characteristics of frequency-chirped super-Gaussian
optical pulses,” Opt. Lett., vol. 11, pp. 569-571 (1986).

124
 D. Anderson and M. Lisak, “Analytic study of pulse broadening in dispersive optical fibers,”
Phys. Rev. A, vol. 35, pp. 184-187 (1987).
Lin Guo-Cheng, Sui Cheng-Hua and Lin Quang, “Non-Gaussian pulse propagation and pulse
quality factor using intensity moment method,” Chin. Phys. Lett., vol. 16, pp. 415-417
(1999).

whose treatment we have followed. Various definitions of pulse width and rise time are considered in

Brown (1963), chapter 5.

A good introduction to multimode pulse distortion can be found in:

D. Gloge, “Dispersion in weakly guiding fibers,” Applied Optics, vol. 10, pp. 2442-2445
(1971).
D. Gloge, A. R. Tynes, M. A. Duguay and J. W. Hansen, “Picosecond pulse distortion in
optical fibers,” IEEE J. Quantum Electron., vol. 8, pp. 217-221 (1972).

Effects of mode coupling and attenuation on multimode pulse propagation are treated in

D. Marcuse, “Pulse propagation in multimode dielectric waveguides,” Bell Syst. Tech. J.,
vol. 51, pp. 1199-1232 (1972).
S. Geckeler, “Pulse broadening in optical fibers with mode mixing,” Appl. Opt., vol. 18, pp.
2192-2198 (1979).

The discussion of wavefronts and precursors is based on:

E. O. Schulz-DuBois, “Sommerfeld pre- and post-cursors in the context of waveguide tran-

sients,” IEEE Trans. Micr. Theory Tech., vol. 18, pp. 455-460 (1970).
A. Karlsson and S. Rikte, “Time-domain theory of forerunners,” J. Opt. Soc. Amer. A, vol.
15, pp. 487-502 (1998).
P. M. Jordan and A. Puri, “Digital signal propagation in dispersive media,” J. Appl. Phys.,
vol. 85, pp. 1273-1282 (1999).

For Heaviside’s first expansion theorem, see

van der Pol and Bremmer (1987).

The AWE was first introduced in

L. T. Pillage and R. A. Rohrer, “Asymptotic waveform evaluation for timing analysis,” IEEE
Trans. Computer-Aided Des., vol. 9, pp. 352-366 (1990).

Some of the subsequent developments of the technique can be found in

Chiprout and Nakhla (1994);

Pillage, Rohrer and Visweswariah (1995).

Finally, here are several papers on off-the-beaten-track subjects related to pulse distortion:

A. M. Bel’skii and A. P. Khapalyuk, “Propagation of a space-limited pulse through an

isotropic medium,” J. Appl. Spectroscopy, vol. 17, pp. 947-951 (1972).
I. A. Man’kin, “Waveguide excitation by nonstationary extraneous currents,” Radio Eng.
Electron. Phys., vol. 24, no. 4, pp. 16-23 (1979).
F. Mainardi, “Energy velocity for hyperbolic dispersive waves,” Wave Motion, vol. 9, pp.
201-208 (1987).

125
4.9 Problems
p4-1 Consider a classical transmission line whose line parameters l, c, r and g are all independent
of frequency. Find the unique condition relating the values of these line parameters which
guarantees that both α and vp will be independent of frequency. Under this condition, find
the voltage v(z, t) of a forward wave on the line if the voltage at z = 0 is some function
v(0, t) = f (t).

p4-2 A pulse transmission system consists of a generator, a destination (load) and a number N
of independent transmission lines connected between the generator and the destination. The
generator sends out an identical stream of narrowband pulses of center frequency ω0 onto each
transmission line. Each line is of identical length d, but has a different group velocity (vg,i
for the ith transmission line) and phase velocity (vp,i ) at ω0 . The destination is assumed to
combine all the arriving voltage pulses by adding them together, without reflection back along
the transmission lines. If the pulse stream from the generator is a sequence of rectangular
pulses (that is, the envelopes are rectangular) of height Vp , duration (“on time”) Tp and
time between pulses (“off time”) To , give an expression for the maximum length d of the
transmission lines for which pulses arriving at the destination do not overlap with adjacent
ones arriving from any of the connected transmission lines. Assume that the transmission
lines are labeled such that line i = 1 has the lowest group velocity vg,1 of them all, and as i
increases, so does the group velocity, until we reach line i = N , for which vg,N is the largest
of all.

p4-3 If the AM Gaussian signal described by (4.45) is applied to a system with nonuniform loss
or gain [so that A(ω) is not constant in frequency], find an expression for the distortion of
the signal using the first three terms of the quasi-stationary expansion. Compare this result
to that of the quasi-Gaussian approximation. How is the distortion of the signal changed by
the fact that A′0 , A′′0 6= 0?

p4-4 Consider a Gaussian pulse that is frequency-modulated in addition to being amplitude-

modulated, with
t2
 
r(t) = exp − 2 + jM (t)ω0 t
2T
where M (t) = t/Tm is a linear frequency modulation function with a modulation rate 1/Tm,
and Tm is a constant. Using the quasi-Gaussian approximation, calculate the distortion of
the amplitude envelope of this pulse due to propagation for a distance z along a lossless
transmission line with γ(ω) = jβ(ω). Show that for certain values of Tm the pulse amplitude
duration may initially get smaller before eventually increasing as it does for a pure AM
Gaussian pulse.

p4-5 A gradual form of the unit step function has the form

1
r(t) =
1 + e−2t/TR

where the constant TR is a measure of the rise time of the step. Use the quasi-stationary
approximation (up to the first three terms) to find an expression for the output w(t) of a
lossless (A0 = 1, A′0 = A′′0 = 0) linear time-invariant network when the input is a narrowband
signal with envelope r(t) and carrier frequency ω0 . Plot the square of the magnitude of the
analytic signal wA versus τ /TR for Θ′′0 /TR2 = 0.1, 1, and 10. Use any suitable value for ω0 .
Repeat this procedure using the quasi-Gaussian approximation (4.55).

126
p4-6 If an input pulse f (t) is applied to a lossy linear time-invariant network, show that the center
of the output pulse w(t) is related to that of the input pulse by

tc (w) = tc (f ) + tg,av (p)

where p(t) is a function whose Fourier transform is F (ω)A(ω), F is the Fourier transform of
f , and A is the real magnitude of the network transfer function.
p4-7 A capacitor C is connected to the end of a lossless transmission line with characteristic
impedance Zc . Both C and Zc are real and independent of frequency.
(a) Obtain an expression for the frequency-dependent reflection coefficient ρ(ω) of this load.
(b) Consider the reflection coefficient to be a transfer function (i. e., the input is the incident
voltage and the output is the reflected voltage). Obtain expressions for the phase delay
time tp0 and group delay time tg0 at a frequency ω0 . If ω0 = 2πf0 , where f0 = 100
MHz, Zc = 50 Ω and C = 100 pF, compute these two delay times.
(c) Given the same parameter values as in part (b), what is the smallest possible duration
of a reflected Gaussian pulse from this load?
p4-8 Use the quasi-Gaussian approximation to study the distortion by a linear network of a “super-
Gaussian” pulse whose envelope is
4 4
e−t /2T
where T is a positive real constant. Assume that A′0 = A′′0 = 0. Give expressions for the
output signal’s phase and power envelope. How does the width of the output pulse compare
to that of the input pulse? What is the narrowest pulse of this form that can emerge from
this network?
p4-9 (a) Obtain explicit expressions for the constants R1 and p1 that appear in the first-order
(q = 1, l = 1) AWE approximation (4.84) for the output response of a linear system
to a unit step-function input. These expressions should be in terms of the appropriate
properties of the transfer function H(s).
(b) Apply the result of part (a) to calculate the response of the linear system whose transfer
function is (4.91) to a rectangular input pulse. Use the parameters V0 = 1, ωc = 2π×106
sec−1 , Tf = 10−7 sec and T = 10−6 sec. Plot your result and compare to the exact and
second-order AWE responses as in Fig. 4.4.

127
128
Chapter 5

HOLLOW METALLIC
WAVEGUIDES

Bottomless vales and boundless floods,

And chasms, and caves, and Titan woods,
With forms that no man can discover
—Edgar Allan Poe,
“Dream-Land”

In this chapter, we begin our study of the electromagnetic analysis of the modes of various specific
types of waveguide. We start with what were historically the earliest waveguides to find wide application
at microwave frequencies: hollow metallic tubes. Hollow rectangular waveguides (cf. Figure 1.1(a)) are
still among the most commonly used waveguides for microwave frequencies. Even as high as 60 GHz,
they remain the medium of choice for many applications. However, as frequency increases even further,
their smaller size can make them hard to fabricate (the accurate machining of the corners and polishing of
the metal surfaces being some of the problems). Moreover, other properties of these guides, such as wall
losses, etc., are not always ideal for some applications, and hollow waveguides of other cross-sectional
shapes are used instead. Although generally used at medium microwave frequencies, hollow metallic
waveguides can be employed at frequencies as low as 300 MHz or as high several Terahertz (THz).
In this chapter, we will consider the general properties of hollow metallic waveguides, with simply-
connected (one-piece) boundaries. First we study the simple, though highly idealized, case of a parallel-
plate waveguide, which exhibits many of the important properties common to all metallic guides, and
then we proceed to consider waveguides of arbitrary cross-section. Two methods for analyzing special
cross-sectional shapes will be discussed: separation of variables (which will be used to study the case of
the circular cross-section), and the ray method (which will be used to study the properties of rectangular
and triangular shapes in this chapter, and extended in Chapter 7 to deal with a wider variety of guiding
structures).

5.1 The Parallel-Plate Waveguide

Even though a given structure may fit our definition of a waveguide as a structure of infinite length in the
z-direction whose material properties and geometry are independent of z, it may not actually guide field
energy in this direction unless appropriate physical mechanisms are built into it to confine or localize
the waves in the cross-section. This may be done by metallic walls or dielectric boundaries, but in any
case, the shape and size of these boundaries will play a critical role in determining the characteristics of
the guided modes of the structure. For waveguides of arbitrary structure, we can appeal to Maxwell’s

129
 x
(σ → ∞)
x=a xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

(ε, µ)

x=0 xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
z
(σ → ∞)

Figure 5.1: The parallel-plate waveguide.

equations and associated boundary conditions to answer the question of whether guided waves will exist.
Indeed, this is the only rigorous way to address the problem, and the framework for doing so will be
outlined in the next section.
On the other hand, there are many important concepts in the theory of guided waves that have their
basis in fairly simple ideas from plane wave solutions of Maxwell’s equations.1 Let us illustrate this for
the case of perhaps the simplest of all waveguides, the parallel-plate waveguide shown in Figure 5.1. This
guide consists of two perfectly conducting planes, infinite in the y- and z-directions, located at x = 0
and x = a, and separated by a homogeneous medium of electrical properties µ and ǫ. Consider the plane
wave
E0
E = ux a0 E0 e−jkz ; H = uy a0 e−jkz
ζ
where a0 is an arbitrary (constant) wave amplitude and E0 is a reference field amplitude.2 This field is
not perturbed by the presence of the conducting walls, so long as the surface charge and current densities
necessary to terminate the normal E and tangential H fields on the walls are present also. Hence, if we
identify γ = jk and
E0
E = ux E0 ; H = uy (5.1)
ζ
we have found a source-free field solution for this waveguide of the form:

E(x, y, z) = a0 e−γz E(x, y)

H(x, y, z) = a0 e−γz H(x, y) (5.2)

A field of such a form is called a mode of the waveguide.

Plane waves propagating in directions other than the z-direction will be perturbed by the presence
of the walls: what happens, of course, is that they are reflected. For example, consider the effect of an
obliquely incident plane wave whose electric field is polarized perpendicular to the plane of incidence as
shown in Figure 5.2. For an incidence angle of θ with respect to the normal to the wall, let

Einc = uy aE E1 e−jk(−x cos θ+z sin θ) (5.3)

In order to satisfy the boundary condition Ey = 0 at the wall, a second (reflected) plane wave must be
present, propagating away from the wall at the same angle θ to the normal as shown in Figure 5.2. The

1 For a brief summary of Maxwell’s equations and the properties of their solutions which will be needed in this book,

refer to Appendix F.
2 Chosen, for example, so that when a = 1, unit power is carried by the wave.
0

130
 x

Einc Er
Hr
Hinc
θ θ
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
x=0 xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx z

Figure 5.2: Reflection of a plane wave by a perfectly conducting plane (perpendicular polarization).

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
x=a xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

2 4
1 θ θ 3

x = 0 xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx z

Figure 5.3: Successive ray reflections in a parallel-plate waveguide.

resulting total electric field is

n o
E = uy aE E1 e−jk(−x cos θ+z sin θ) + E2 e−jk(x cos θ+z sin θ) (5.4)

Of course, we know that the reflection coefficient at the wall is −1, so that E2 = −E1 , but let us not use
that knowledge for the time being. A wave incident on the upper wall of the guide similarly produces
a reflected wave propagating downward. Accounting for all possible reflections of these plane waves as
in Figure 5.3, we see that all of the wave vectors of these plane waves must be in one of two directions,
k = k1 = k(ux cos θ + uz sin θ) or k = k2 = k(−ux cos θ + uz sin θ), corresponding to the waves labeled
1 or 2 in Figure 5.3. Thus, the general form of a field with this polarization must be that of (5.4) above
(waves 3 and 4 of Figure 5.3 have the same form as waves 1 and 2 respectively, and do not need to be
included since an appropriate adjustment of the values of E1 and E2 could be used to account for them,
and we assume this has been done). We will, though somewhat inaccurately, refer to the directions of
propagation of these plane waves as rays. However, it should always be kept in mind that the field is not
localized to the schematized arrows representing these rays in the figures, but is rather made up of plane
waves which extend to infinity in the y and z directions and are uniform in any direction perpendicular
to their directions of propagation.
The allowed values of the angle θ are found by enforcing the boundary conditions Ey = 0 at x = 0
and x = a. From x = 0 we get E1 = −E2 , while at x = a we find

E1 ejka cos θ = −E2 e−jka cos θ

131
Thus, the two conditions together imply that

mπ
  mπ 2 1/2
cos θ = ; sin θ = 1 − (5.5)
ka ka

and
mπx −j[k2 −( mπ 2 1/2
E = uy a E
m (2jE1 ) sin e a ) ] z
(5.6)
a
for m = 1, 2, 3, .... By identifying
mπ 2 1/2
γm = jk sin θ = j[k 2 − ( ) ] (5.7)
a
and
mπx
E = uy (2jE1 ) sin (5.8)
a
we have clearly obtained another set of modes for this waveguide.
So long as the wavenumber k is greater than the “cutoff” wavenumber kc,m = mπ/a, we can interpret
the field of such a mode as the result of plane waves bouncing back and forth between the plates at the an-
gle θ = cos−1 (mπ/ka). Below cutoff (k < kc ), the angle θ becomes complex, and n we must interpret (5.6)
1/2 o
as two inhomogeneous plane waves, attenuating with the coordinate z as exp − (mπ/a)2 − k 2

z .
√
There is thus a critical or cutoff frequency ωc,m = kc,m / µǫ such that when ω > ωc,m , the TEm mode
propagates (γm = jβm ) , while when ω < ωc,m , the mode attenuates in the z-direction (γm = αm ). The
cutoff phenomenon is common to modes of many different waveguide structures.
We can construct another set of modes for this structure by considering parallel-polarized plane
waves. It is most convenient in this case to deal with the magnetic field, which for a wave incident at
an angle θ is
Hinc = uy aH H1 e−jk(−x cos θ+z sin θ) (5.9)
where H1 is the amplitude of the magnetic field. The requirement that the tangential electric field
should vanish at the perfectly conducting boundary is equivalent to requiring that ∂Hy /∂x be zero
there. Following basically the same analysis as in the previous case, we find another set of allowable
mode fields with
mπx −j[k2 −( mπ )2 ]1/2 z
H = uy a Hm (2H1 ) cos e a ; m = 0, 1, 2, ... (5.10)
a
which are yet a further set of modes of the structure. Note that the m = 0 mode of (5.10) is, in fact,
none other than the special mode (5.1) obtained above. The expression for γm is the same as in (5.7),
while the field distribution is different, viz.:
mπx
H = uy (2H1 ) cos (5.11)
a
We have displayed here the forward modes of this waveguide; a similar procedure can be used to
obtain the reverse modes. A number of other observations can be made about the properties of parallel-
plate waveguide modes:

(i) Whether a mode propagates with no attenuation (γ = jβ) or decays exponentially (γ = α) depends
upon the frequency ω, the plate separation a, and the mode number m. For a given frequency
(and hence wavenumber k), there will only be a finite number of propagating modes.

(ii) The factor e−γm z which determines the change in the magnitude or phase of the mode field with
z is different for each mode of the waveguide. If a spatial field distribution resulting from the
superposition of several modes is present at z = 0, this distribution is distorted as it propagates
to cross-sections at other values of z.

132
(iii) For a propagating mode of this structure, since β ≤ k, the phase velocity vp given by equation
(1.11) will always be greater than or equal to the speed of light in the medium filling the guide.
The group velocity vg is always less than or equal to that of a plane wave in the filling medium.
(iv) The phase velocity vp varies with ω and hence differs from the group velocity vg for all but the
TEM mode. For the other modes, a signal containing more than one frequency component will be
dispersed as it propagates along the guide, as discussed in the previous section.
The similarities between each of the modes of the parallel-plate waveguide and those of the classical
transmission line are striking. In both cases, an amplitude constant am is present which cannot be
determined without information about the sources. Moreover, the propagation factor of exp(−γz) is
present, and forward and reverse modes exist in both cases. This close parallel between the classical
transmission line and the modes of a waveguide will be exploited later on in Chapters 10-11 to describe
waveguide excitation and waveguide networks in terms of equivalent transmission-line circuits.

5.2 Waveguide Modes in General

5.2.1 Modes
Outside of a region containing impressed sources, i.e., any place where the current density Jext is zero, a
general waveguide can be expected to support mode fields, which vary with the axial coordinate z and
time t as exp(jωt − γz), and have the form (5.2) analogous to the mode solutions (1.7) for a classical
transmission line. For a given value of ω, not all values of γ will correspond to nontrivial field solutions.
For example, only the values given by (1.9) yield nonzero mode voltages and currents on a classical
transmission line. Moreover, as we have seen, the modes of a parallel-plate waveguide have only the
allowed values of propagation coefficient:
r
mπ 2
γ = ±γm = ± − ω 2 µǫ
a
where m is a nonnegative integer and µ and ǫ are the permeability and permittivity of the medium
filling the guide. The values γm are known as characteristic values or eigenvalues of the modes of this
waveguide. This case illustrates a common situation which occurs when the waveguide is completely
made up of lossless media: the allowed values of γ = α + jβ are either pure imaginary (corresponding to
the so-called propagating modes of the guide) or pure real numbers (corresponding to evanescent modes).
The modal field, while it may differ in magnitude and phase, will not change its characteristic
distribution E(x, y) or H(x, y) from one cross-section to another. The field distributions E and H are
vector functions of the cross-sectional coordinates (x, y) or (ρ, φ). Note that this does not necessarily
mean that Ez = 0 or Hz = 0; it only means that none of the components of E or H depends on z.
It may happen (especially in simple geometries) that for certain modes Ez or Hz will vanish identically.
It has become customary to use the following terminology to classify modes:
1. Ez = 0, Hz = 0: TEM (transverse electromagnetic) mode.
2. Ez = 0, Hz 6= 0: TE (transverse electric) mode; sometimes called an H-mode.
3. Ez 6= 0, Hz = 0: TM (transverse magnetic) mode; sometimes called an E-mode.
4. Ez 6= 0, Hz 6= 0: Hybrid mode; the terminologies HE-mode and EH-mode are sometimes used;
many other terms are also used in specific contexts.
In general, a waveguide mode will be hybrid; TEM, TE and TM modes exist only under certain condi-
tions. We have already seen that the parallel-plate waveguide can support TEM (eqn. (5.1)), TE (5.8)
and TM (5.11) modes. We will see in Chapter 5 that any metallic waveguide filled with a homogeneous

133
medium supports only TE and TM modes if its boundary wall is simply-connected (i.e., is all of one
piece). TEM modes can exist on multi-conductor structures such as those of Figures 1.1(b) and (c).
In Chapters 6, 7 and 8, we will study structures which support hybrid modes, such as the microstrip
transmission line and the dielectric optical waveguide.

5.2.2 Reduced Maxwell equations for modes

Mode fields must satisfy the source-free version of Maxwell’s equations (F.3), namely:

∇×E = −jωµH
∇×H = jωǫE (5.12)

Substituting the form (5.2) of a modal field into (5.12) gives us the following general equations for E
and H in cartesian coordinates:

 
 
 u uy uz 
 x
 
 
 ∂
 ∂x
∂
−γ  = −jωµH
∂y
 
 
 E Ey Ez 
 x
 
 
 u uy uz 
 x
 
 
 ∂
 ∂x
∂
−γ  = jωǫE (5.13)
∂y
 
 
 H Hy Hz 
 x

We can split the left side of each of equations (5.13) into two terms, introducing a convenient short-
hand in the process:

     
     
 u uy uz   ux uy uz   ux uy uz 
 x
     
     
 ∂
 ∂x
∂
∂y −γ  =  ∂x
∂ ∂
∂y 0  +  0 0 −γ  = ∇T × E − γuz × E (5.14)
     
     
 E Ey Ez   Ex Ey Ez   Ex Ey Ez 
 x

where we have introduced the subscript “T ” to denote the transverse (xy or ρφ) part of a vector or

134
operator. Thus, for instance,

ET = ux E x + uy E y
= uρ E ρ + uφ E φ
∂Φ ∂Φ
∇T Φ = ux + uy
∂x ∂y
∂Φ 1 ∂Φ
= uρ + uφ
∂ρ ρ ∂φ
1 ∂ 1 ∂Aφ
∇T · A = (ρAρ ) + (5.15)
ρ ∂ρ ρ ∂φ
∂Ax ∂Ay
= +
∂x ∂y
 
∂Az ∂Az ∂Ay ∂Ax
∇T × A = ux − uy + uz −
∂y ∂x ∂x ∂y
 
1 ∂Az ∂Az 1 ∂ 1 ∂Aρ
= uρ − uφ + uz (ρAφ ) −
ρ ∂φ ∂ρ ρ ∂ρ ρ ∂φ
∂
Hence, ∇T is the ∇ operator with ∂z set to zero whenever it appears. Note that ∇T × A does have a
z-component and does involve Az as well. Using this notation, (5.13) can be written as

∇T × E − γuz × E = −jωµH
∇T × H − γuz × H = jωǫE (5.16)

Yet another form for these equations is obtained if we realize that the transverse part of ∇T × A can
be written
(∇T × u)T = ∇T × (uz Az ) = −uz × ∇T Az
We can then partially split each of (5.16) into transverse and longitudinal parts:

−uz × ∇T Ez − γuz × ET = −jωµHT (5.17)

∇T × ET = −jωµuz Hz (5.18)
−uz × ∇T Hz − γuz × HT = jωǫET (5.19)
∇T × HT = jωǫuz Ez (5.20)
Furthermore, if we take the cross product of uz with eqn. (5.17), we get

∇T Ez + γET = −jωµuz × HT (5.21)

because uz × (uz × AT ) = −AT for any transverse vector AT . Using this equation and equation (5.19),
we can solve for all of the transverse fields in terms of Ez and Hz :
1
ET = − [−jωµuz × ∇T Hz + γ∇T Ez ] (5.22)
ω 2 µǫ + γ 2

1
HT = − [jωǫuz × ∇T Ez + γ∇T Hz ] (5.23)
ω 2 µǫ + γ 2
In other words, the components Ez and Hz are special in that if we know them we can find all other
field components from (5.22) and (5.23). It is the special nature of Ez and Hz which led to the
TEM/TE/TM/Hybrid mode classification scheme outlined earlier in this section.

135
5.2.3 Forward and reverse modes
In a waveguide made up of linear isotropic material (the situation to which we limit ourselves in this text),
there corresponds to each forward mode that propagates (or attenuates) in the +z-direction another
(reverse) mode that propagates (or attenuates) in the −z-direction. Let us adopt the convention that a
forward mode will have a propagation coefficient γm with either a positive real part or (if Re(γm ) = 0) a
positive real complex power flow in the +z direction. Let Em , Hm be the fields of such a forward mode,
where m(> 0) is an index designating the mode in question (in certain situations, m may stand for a
double index, as will be seen for example in the case of TEmn or TMmn modes of a hollow rectangular
waveguide).
To each forward mode there is a corresponding reverse mode (with propagation coefficient γ−m =
−γm ) which we will designate with an index m < 0. It is clear upon examination of (5.17)–(5.20) that
its fields can be found by reversing the signs of the longitudinal component of E and the transverse
components of H:
E−mT = EmT ; E−mz = −Emz (5.24)
H−mT = −HmT ; H−mz = Hmz (5.25)
Of course, we could equally well have reversed the signs of ET and Hz , but (5.25) is the commonly used
convention which will be followed here.

5.2.4 Normalized mode fields, power and stored energy

The fields of a waveguide mode contain an undetermined amplitude constant which can be related to
the complex power (or alternatively to the complex oscillatory power) carried by the mode. As with the
case of the classical transmission line, it is convenient (when possible) to deal with mode fields that are
normalized (have the field amplitudes adjusted) so that unit power is carried by the mode through its
cross section.3 Although we could try to impose the condition that the complex power of a normalized
mode is 1, this leaves the phase angle of the field amplitude arbitrary and is not even possible in general,
as was seen in chapter 1 for the classical transmission line and as we will illustrate below in general.
Let us consider the case when the media from which the waveguide is constructed are lossless (ǫ
and µ are real and any boundary walls are perfectly conducting). Denote the cross-sectional area of
the waveguide by S0 , and the boundary contour of this cross section by C0 . In problem p5-17 it is
shown that, for any mode carrying nonzero complex power, it must be possible to choose the arbitrary
amplitude of the mode fields such that ET is real, so long as γ is either pure real (a cutoff mode, in
which case HT is imaginary) or pure imaginary (a propagating mode, in which case HT is real). The
fact that ET is real implies that
 Z ∗
1 1
Z
∗ ∗
P̂m ≡ Em × Hm · uz dS = Em × Hm · uz dS ≡ P̂m,osc
2 S0 2 S0
For a propagating mode, the phase of P̂m is either 0 (in which case, we call the mode a forward mode,
in analogy with the treatment for the classical transmission line in chapter 1) or π (in which case, we
call the mode a reverse mode). For a cutoff mode the phase of P̂m is ±π/2 (see eqns. (K.29) and (K.24)
of Appendix K). We have

1
Z real if m is a propagating mode
P̂m = (EmT × H∗mT ) · uz dS = (5.26)
2 S0
imaginary if m is a cutoff mode
3 In certain exceptional cases, the power carried by a mode may be zero, and it thus cannot be normalized in this way.

In these cases, there arise so-called complex modes (see Appendix K), which can have propagation coefficients that are
neither pure real nor pure imaginary even for a lossless waveguide. Although of some practical importance, the study of
these modes is beyond the scope of this book, and the reader is referred to the references at the end of Appendix K for
further information.

136
This remains the case when ET is multiplied by an arbitrary complex constant, although clearly the
phase of P̂m,osc will be altered by such a multiplication.
In the more general situation of a lossy waveguide, a forward mode is one whose fields contain the
factor e−γm z , and αm = Re(γm ) > 0, while a reverse mode is one whose fields vary as eγm z under the
same constraint on αm . As remarked in Appendix F, the magnitudes of the complex power P̂m and
complex oscillatory power P̂m,osc are not necessarily identical for modes of a lossy waveguide. The only
case of a lossy waveguiding structure where |P̂m | = |P̂m,osc | is the classical transmission line (and a
few waveguide equivalents to it), where the power is expressed in terms of a single voltage and a single
current. Since it is the complex oscillatory power rather than P̂m that is closely connected with the
orthogonality property to be derived in chapter 9, it will prove more convenient to use the former rather
than the latter to normalize the mode fields of an arbitrary waveguide. This will uniquely determine the
fields to within a plus or minus sign.
In analogy with the normalization imposed on the classical transmission line modes in chapter 1, we
thus demand that the complex amplitude of any mode fields will hereafter be understood to have been
constrained by the following normalization condition:
1
Z
P̂m,osc = Em × Hm · uz dS
2 S0
= 1 (forward mode)
= −1 (reverse mode) (5.27)
A forward mode field of arbitrary amplitude will have the form (5.2), and thus a2m represents the complex
oscillatory power associated with the mode.
When normalization is imposed on a mode of a lossless waveguide, the complex power of the mode
will retain the same phase angle as discussed above, but will have unit amplitude; the phase angle of
P̂m,osc will now be zero or 180◦ . If real, this number will be +1 for m > 0 by the normalization condition.
If imaginary, it will be ±j. The normalization condition for modes in a lossless waveguide can thus be
written in the form
1
Z
P̂m = EmT × H∗mT · uz dS = 1 γm = jβm (5.28)
2 S0
= jum γm = αm
where um = +1 if the mode stores more magnetic than electric energy per unit length, and um = −1 if
the mode stores more electric than magnetic energy per unit length (cf. (K.24)). When the waveguide
is not lossless, no simple statement can be made about the complex power of a mode.
It is worth mentioning that the oscillatory power of a waveguide mode can also be expressed in terms
of the oscillatory stored energy per unit length of the guide. From (K.5) we have that
jω
P̂m,osc = Um,osc (5.29)
γm
where Um,osc is the complex oscillatory energy stored in the mode field per unit length of the waveguide:
1
Z
Um,osc = (ǫEm · Em + µHm · Hm ) dS (5.30)
4 S0
If the mode is purely propagating (γm = jβm ), then (5.29) becomes
P̂m,osc = vp Um,osc (5.31)
where vp is the phase velocity of the mode. In some cases, the energy form of computing the normalization
is more convenient than the direct computation of the power. Furthermore, by (K.4), the oscillatory
energy can also be written in either of two alternative forms (K.8) which can help to simplify calculations
even more.

137
5.2.5 Example: The parallel-plate waveguide
When the normalization condition is imposed on the fields of the parallel-plate waveguide treated earlier
in the chapter, the constants E0 , E1 and H1 are determined. Here, the “cross-section” of the waveguide
is just the interval from x = 0 to x = a, and our so-called powers are in fact powers per unit width in
the y-direction, since this two-dimensional guide has infinite extent in y. The TEM mode whose fields
are given by (5.1) is subjected to:
a
1 aE02
Z
E × H · uz dx = =1
2 0 2ζ
or r
2ζ
E0 = (5.32)
a
and we have finally r r
2ζ 2
E = ux ; H = uy (5.33)
a aζ
The transverse E field of a TE mode is given by (5.8); we can use (5.18) and (5.23) to obtain Hz and
HT . We arrive at the result r
ζTEm
E1 = −j
a
where
Ey jωµ
ζTEm = − =
Hx γm
is the so-called intrinsic wave impedance for the TEm mode of the parallel-plate waveguide. Thence,
r
ζTEm mπx
E = uy 2 sin (5.34)
a a
r r
2jmπ ζTEm mπx 2 ζTEm mπx
H = uz cos − ux sin
ωµa a a ζTEm a a

Likewise the transverse H field of a TM mode is given by (5.11); we use (5.20) and (5.22) to retrieve
the remaining field components. We get in a similar fashion
s
1
H1 =
ζTMm a

where
Ex γm
ζTMm = =
Hy jωǫ
is the so-called intrinsic wave impedance for the TMm mode of the parallel-plate waveguide, so that
s
1 mπx
H = uy 2 cos (5.35)
ζTMm a a
s s
2jmπ 1 mπx 1 mπx
E = uz sin + ux 2ζTMm cos
ωǫa ζTMm a a ζTMm a a

if m 6= 0. If m = 0, the TM0 mode is equivalent to the TEM mode from above, and (5.32)-(5.33) should
be used.

138
5.3 “Hollow” Metallic Waveguides of Arbitrary Cross-Section
Consider the waveguide illustrated in Figure 5.4. A perfectly conducting metallic cylinder of cross-section
S0 (and boundary C0 ) is oriented along the z-axis as shown. The cylinder is filled with a homogeneous

Figure 5.4: Hollow metallic waveguide.

medium of electrical parameters ǫ and µ (as mentioned in chapter 1, we can easily allow this medium to
be lossy by replacing ǫ by the complex dielectric constant ǫ̂). Most often this medium will be air, and
we refer to this type of waveguide as a hollow metallic waveguide.
From (5.18), (5.20), (5.22) and (5.23), a mode solution of Maxwell’s equations for this guide must
satisfy:
uz · ∇T × ET = −jωµHz (5.36)

uz · ∇T × HT = jωǫEz (5.37)
1
ET = [jωµuz × ∇T Hz − γ∇T Ez ] (5.38)
kc2
1
HT = [−jωǫuz × ∇T Ez − γ∇T Hz ] (5.39)
kc2
where we have introduced the notation
kc2 = ω 2 µǫ + γ 2 (5.40)
and kc is called the cutoff wavenumber for reasons which will become apparent later.

139
 Equations for Ez and Hz can be obtained by substituting (5.38) into (5.36) and (5.39) into (5.37).
Since this is a homogeneous waveguide (ǫ, µ and hence also kc , are not functions of position), we obtain
the equations
∇2T + kc2 Ez = 0


(5.41)

∇2T + kc2 Hz = 0


(5.42)

because of the vector identity ∇T × ∇T f ≡ 0 for any suitable function f (see eqn. (D.9) of Appendix D).
Here ∇2T is the transverse Laplacian operator

∇T 2 f ≡ ∇T · ∇T f
∂2f ∂2f
= 2
+ 2 (5.43)
∂x ∂y
1 ∂2f
 
1 ∂ ∂f
= ρ + 2 2
ρ ∂ρ ∂ρ ρ ∂φ

Naturally ∇2T differs from the ordinary Laplacian ∇2 only in the absence of the z-derivative term ∂ 2 /∂z 2 .
We see that, as you may already have seen in the study of hollow rectangular waveguides, separate
equations are obtained for Ez and Hz from Maxwell’s equations, and the remaining field components are
obtained at once from (5.38) and (5.39).
To specify the problem of finding Ez and Hz completely, we must find the appropriate boundary
conditions satisfied by these quantities on the wall of the guide. Since it is tangential to the wall, which
is a perfect conductor, Ez must vanish there:

E z |C 0 = 0 (5.44)

However, Hz cannot be dealt with so immediately, since we can only relate it to a surface electric wall
current, itself not yet known. We can obtain a second boundary condition by forcing the other tangential
component of E to vanish on the wall. Denoting the inward normal from the wall as the unit vector un ,
the vector uℓ ≡ uz × un will be in the direction of this remaining tangential component (see Figure 5.5),
so from (5.38), we require

Figure 5.5: The transverse cross-section.

140
 E ℓ |C 0 = uℓ · ET |C0
1
= [jωµun · ∇T Hz − γuℓ · ∇T Ez ]C0 (5.45)
kc2

But we know that 

∂Ez 
uℓ · ∇T Ez |C0 = =0
∂ℓ C0

because the condition Ez |C0 = 0 does not change as we move along the curve C0 . Hence (5.45) becomes

∂Hz 
=0 (5.46)
∂n C0

where ∂/∂n indicates the derivative in the direction normal to the wall at any point on the wall (i.e.,
∂f /∂n = un · ∇f ).
The fact that equations (5.41) and (5.44) for Ez and (5.42) and (5.46) for Hz are completely separate
means that either Ez or Hz can be independently set equal to zero, and the modes of the hollow waveguide
can all be chosen either pure TE or pure TM. Since, as we will see in Chapter 10, the TE and TM modes
which are propagating along a lossless hollow waveguide carry power independently of each other, it is
customary to choose the modes of these waveguides in these polarizations rather than in some other
fashion.

5.4 General Properties of TE and TM Modes

5.4.1 TM modes
The properties of the TM modes can be found from (5.38) and (5.39) by setting Hz ≡ 0. Outside of
being rotated by 90◦ from each other, the transverse E and H fields differ only by a proportionality
constant, viz.,

ET = −ζTM uz × HT
γ
= − 2 ∇T Ez (5.47)
kc

The constant of proportionality ζTM is known as the intrinsic wave impedance for a given TM mode,
and is equal to
γ
ζTM = (5.48)
jωǫ

5.4.2 TE modes
Similarly, for TE modes we have the relations

ET = −ζTE uz × HT
jωµ
= uz × ∇T Hz (5.49)
kc2

where the TE mode wave impedance is

jωµ
ζTE = (5.50)
γ
Of course, neither ζTE nor ζTM is known until the propagation coefficient γ is determined.

141
5.4.3 The eigenvalues
Because of (5.40), kc2 can be considered as the eigenvalue for hollow guides, and γ determined from it.
The problem thus reduces to solving the equation

(∇2T + kc2 )Ψ = 0 (5.51)

subject to
Ψ|C0 = 0 (5.52)
when Ψ = Ez , or 
∂Ψ 
=0 (5.53)
∂n C0
2
when Ψ = Hz , and to simultaneously finding the permitted set of eigenvalues kc,m ; m = 1, 2, .... For a
specific waveguide (e.g., rectangular), the mode index m may actually refer to a double index mn, as in
the TEmn or TMmn modes.
Note that nowhere does ǫ or µ enter into the problem at this point, and it is not until we wish to
find ET and HT from (5.47) or (5.49), or γ from
q
γ±m = ± kc,m2 − ω 2 µǫ (m > 0) (5.54)

where Re(γ) > 0 (or Im(γ) > 0 if Re(γ) = 0) that the electrical properties of the filling come in.
We will state here (without proof) some properties of the solution of (5.51) subject to (5.52) or (5.53):

(a) There is a countably infinite, discrete set of solutions (eigenfunctions) Ψm , m = 1, 2, ..., corre-
2
sponding to the eigenvalues kc,m in either the TM or TE case (m = 0 is used to designate a special
type of TE mode as discussed below).
2 2
(b) Writing κm = kc,m for TE modes, and λm = kc,m for TM modes, all κm and λm are real and
positive when C0 is simply connected (except for κ0 = 0), so that we can always assume that they
are ordered in increasing value:

0 = κ0 < κ1 ≤ κ2 ≤ ...
0 < λ1 ≤ λ2 ≤ λ3 ≤ ...

The lowest TE “mode” corresponding to κ0 = 0 for a hollow guide with a simply-connected

boundary is just a constant:
Ψ = Hz = K0 = constant
This mode (whose only nonzero field component is Hz and which can only exist at zero frequency)
is an exceptional case which can often be ignored in applications. It is examined in detail in
Appendix I.
Thus, if the medium filling the guide is lossless (µ, ǫ are real), each eigenvalue has a corresponding
cutoff frequency

ωc,m = 2πfc,m
kc,m
= √ (5.55)
µǫ

above which the mode is a propagating mode whose propagation coefficient is imaginary:
√ q
γm = jβm = j µǫ ω 2 − ωc,m
2 (ω > ωc,m ) (5.56)

142
 and below which the mode is evanescent or cut off :
√ q 2
γm = αm = µǫ ωc,m − ω2 (ω < ωc,m ) (5.57)

The terminology of cut-off wavenumber for kc,m is now clear from (5.55).
The TE1 mode is often called the dominant or fundamental mode of this waveguide, because

κ1 < λ1 (5.58)

and thus it is the mode with lowest cutoff frequency which is capable of transporting energy.

(c) One now can also give expressions for the guide wavelength, phase velocity and group velocity for
modes with index ±m which are above cutoff (ω > ωc,m ). These follow from (1.10)-(1.12):

2π
λg,m = √ q
µǫ ω 2 − ωc,m
2

1
vp,m = √ q
2 /ω 2
µǫ 1 − ωc,m
q
2 /ω 2
1 − ωc,m
vg,m = √
µǫ

5.4.4 Power flow and stored energy

The complex power flow through the cross section S0 of the guide is of interest in applications as well
as from a conceptual standpoint. It is given by
1
Z
P̂ = E × H∗ · uz dS
2 S0

From (5.47) and (5.49), we can write this as

ζm
Z
P̂ = HT · H∗T dS (5.59)
2 S0

where ζm denotes the intrinsic wave impedance

 
 
 ζTE
 (TE modes) 

ζm = (5.60)
 
 ζTM
 (TM modes) 


of the mode. When the mode is above cutoff, the impedance is real, and since the integral in (5.59)
is real, this means there is net, nonzero time-average power flow in the z-direction for a mode in these
conditions. On the other hand, when a single mode decaying in the +z-direction exists on the guide, the
wave impedance is purely imaginary, and no net time-average power flow takes place. Instead, energy is
stored in the fields of the mode, as in a capacitor or inductor.
The oscillatory power is similarly given by
ζm
Z
P̂osc = HT · HT dS (5.61)
2 S0

143
 Figure 5.6: Cross-section of circular waveguide.

and this relation can be used to normalize the fields of a mode using (5.27). This equation could also
be obtained with the help of (K.8).
It can be shown that cutoff TE modes store predominantly magnetic energy (they have an extra
component of H) while cutoff TM modes store predominantly electric energy (they have an extra com-
ponent of E) in a hollow metallic waveguide. This has consequences for the equivalent circuits of certain
types of discontinuity in such waveguides (see chapter 11). Details of the proof are left to the reader
(problem p5-22).

5.5 Circular Waveguides (Method of Separation of Variables)

For any hollow metallic guide, our task is now to solve (5.51)-(5.53) for a particular waveguide shape.
We will first illustrate the method of separation of variables and consider the circular waveguide whose
cross-section is illustrated in Figure 5.6.
A perfectly conducting metal tube of radius a is assigned cartesian and cylindrical coordinates as
shown. We first seek the TM modes for this waveguide. In the method of separation of variables, we
assume that a solution Ez of (5.41) can be written as the product:

Ez (ρ, φ) = R(ρ)Φ(φ) (5.62)

where R is a function of ρ only, and Φ is a function of φ only. We insert this into (5.41) to get
Φ ′ 1
ΦR′′ + R + 2 RΦ′′ + kc2 RΦ = 0 (5.63)
ρ ρ
where primes denote differentiation with respect to the argument (either ρ or φ as the case may be). If
we divide this result by RΦ/ρ2 and rearrange a bit, we get
R′′ R′ Φ′′
ρ2 +ρ + ρ2 kc2 = − (5.64)
R R Φ
But this equation states that a function of ρ only (the left side) is equal to a function of φ only (the right
side). Thus, both sides must be a constant. With a certain amount of hindsight, we call this constant
ν 2 . Making the right side of (5.64) equal to ν 2 , we have

Φ′′ + ν 2 Φ = 0

whose solution is well-known:

Φ(φ) = C1 cos νφ + C2 sin νφ (5.65)

144
Here, C1 and C2 are some yet unknown complex constants to be determined later.
Equating the left side of (5.64) to ν 2 , we have, after rearranging,

ρ2 R′′ + ρR′ + (kc2 ρ2 − ν 2 )R = 0 (5.66)

This can be cleaned up slightly by defining a new variable w = kc ρ, so that in terms of w, (5.66) becomes

d2 R dR
w2 2
+w + (w2 − ν 2 )R = 0 (5.67)
dw dw
Equation (5.67) is Bessel’s differential equation for R as a function of w. As shown in Appendix B, the
general solution of (5.67) is
R = C3 Jν (w) + C4 Yν (w) (5.68)
where Jν is the Bessel function of the first kind, and Yν that of the second kind, of order ν and argument
w. Once again we acquire two constants C3 and C4 which must be determined.
From (5.62), (5.65) and (5.68), then, our solution for Ez is

Ez (ρ, φ) = [C1 cos νφ + C2 sin νφ][C3 Jν (kc ρ) + C4 Yν (kc ρ)] (5.69)

It remains for us now to determine the unknown constants C1 through C4 , the separation constant ν,
and the eigenvalues kc2 by enforcing the boundary conditions at the waveguide wall together with certain
other “common sense” restrictions on the fields. The following constraints are applied:
(a) From Appendix B, it can be seen that Yν (kc ρ) blows up at ρ = 0:

|Yν (kc ρ)| → ∞ as ρ → 0

However, there is no physical reason for the fields to blow up at the center of the guide.4 We
conclude, therefore, that C4 = 0. Equation (5.69) now becomes

Ez (ρ, φ) = [Cc cos νφ + Cs sin νφ]Jν (kc ρ) (5.70)

where we have combined the three remaining amplitude constants into two: Cc = C1 C3 and
Cs = C2 C3 .
(b) In cylindrical coordinates, the angle φ can vary continuously over the interval [0, 2π] and beyond,
as the wave equation holds continuously for increasing φ. But a given point described by φ = φ0
is the same (for a given value of ρ) as a point with φ = φ0 + 2π. We must expect however that Ez
should be single-valued as a function of φ, because the same solution of the form (5.70) must still
be valid for both values of φ. Thus,

Ez (ρ, φ) = Ez (ρ, 2π + φ)

for all φ. This is guaranteed only if ν in the terms cos νφ and sin νφ is an integer, so we set ν = l
with l = 0, 1, 2, .... Equation (5.70) then becomes

Ez (ρ, φ) = [Cc cos lφ + Cs sin lφ]Jl (kc ρ); l = 0, 1, 2, ... (5.71)

Setting ν = l has fixed the indices of the Bessel functions as well. Negative values of l are
unnecessary, since cos(−lφ) = cos lφ, and J−l , Y−l can be related to Jl , Yl (Appendix B), so that
no new functions result.
4 Infinite fields are in fact allowed under certain conditions provided the total field energy in a finite volume near

the edge is finite. It can

p be shown that the stored energy of a field for which the longitudinal E-field behaves as Ez ∝
Yν (kc ρ) cos νφ exp(−j k 2 − kc2 z) near ρ = 0, is infinite.

145
 (c) Finally, we require that the tangential field Ez be zero at the guide wall ρ = a (cf. (5.44)). For
this requirement to be true for all angles φ, we must have
Jl (kc a) = 0 (5.72)
if we discard the trivial solution Cc = Cs = 0. From Appendix B, the Bessel functions Jl are zero
only at certain values jlm of their argument.5 The subscript m denotes the ordering of the root in
increasing sequence; thus
jl1 < jl2 < jl3 < ...
while l denotes the index of the Bessel function involved. For example, Table B.1 of Appendix B
indicates the first three zeroes of J0 as j01 ≃ 2.405, j02 ≃ 5.520, j03 ≃ 8.654. From (5.72), we then
conclude that
kc a = jlm
or
kc,TMlm = jlm /a; m = 1, 2, 3, ... (5.73)
Substituting this result into (5.70), we get a general solution which is a linear combination of the
two possible solutions  
  
 Elmc cos lφ 
  jlm

Ez (ρ, φ) = Jl ρ (5.74)
  a
 Elms sin lφ 
 
where we have renamed the remaining arbitrary constants Cc and Cs as mode amplitudes Elmc
and Elms . These constants are determined by normalizing the oscillatory power of the mode to 1
(see below).
We can now obtain the transverse field components by substituting (5.74) into (5.47):
 
 
aγlm  Elmc cos lφ  ′ jlm
   
Eρ = − J ρ (5.75)
jlm   l a
 Elms sin lφ 
 
 
  
lγlm a2

 Elmc sin lφ  jlm

Eφ = 2 Jl ρ (5.76)
jlm ρ   a
 −Elms cos lφ 
 
with HT obtained from
uz × ET
HT = (5.77)
ζTM
The modes we have just described are known as the TMlm modes of the circular waveguide. Indeed,
there are two possible polarizations: TMlmc and TMlms (except for l = 0; the TM0ms modes, of course,
do not exist, as their fields vanish identically). Thus, for example, while the TM01 mode is invariant
under rotations, the TM11 mode can be changed into a different polarization by rotating it through
90◦ . Hence, the designation TMlm for l > 0 actually refers to a pair of degenerate 6 modes TMlmc and
TMlms having orthogonal polarizations. The same kind of thing happens in a square waveguide with
the TE10 /TE01 modes, etc.
Figure 5.7 shows the relative locations of the cutoff frequencies for the circular waveguide. For TM
5 No
√
confusion should arise between j = −1 and jlm .
6 Two or more modes are said to be degenerate if their propagation coefficients are identical. Ordinarily this is
undesirable since very little irregularity is needed to convert from one of these modes to another, and launching and
receiving schemes are generally designed to deal only with one specific mode. In the presence of such a degeneracy,
some type of mode filter must be devised if control over the polarization direction in the waveguide is to be maintained.
However, on occasion there are devices whose design specifically utilizes the presence of degenerate modes, as in the case
of Example 5.5.2.

146
 TE01
TE11 TE21 TE31

fc /fc,TE
0 1 2 3 11

TM01 TM11

Figure 5.7: Relative cutoff frequencies for circular waveguide modes.

Figure 5.8: Field distributions of TM01 and TM11s circular waveguide modes.

modes, the first few are:

0.383
TM01 : fc,TM01 = √
a µǫ

0.609
TM11 : fc,TM11 = √
a µǫ

Transverse field distributions for the TM modes can be constructed with the aid of equations (5.75)-
(5.77) and Appendix J and are sketched in Figure 5.8. According to the usual conventions, solid lines
denote the E-field, and dashed lines the H-field. It might be noted that the field of the TM11 mode
bears a striking resemblance to that of the TEM mode on a two-wire transmission line (Figure 1.1(c))
and this resemblance could be used as a means of exciting this particular mode, the principles of which
are presented in Chapter 10.
The TE modes are investigated in a manner almost identical to that for the TM modes. The
difference is that Hz is being solved for instead of Ez (cf. (5.42)), and in step (c), we must enforce

147
boundary condition (5.46) on Hz . We have
 
 
 Hlmc cos lφ
 

Hz (ρ, φ) = Jl (kc ρ) (5.78)
 
 Hlms sin lφ 
 

and (5.46) implies

Jl ′ (kc a) = 0 (5.79)

From Appendix B, the roots of Jl′ (x) are simple, and can be denoted jlm
′
. A short table of them is given
in Appendix B. We then have
′
kc,TElm = jlm /a (5.80)

or
′
jlm
fc,TElm = √ (5.81)
2πa µǫ

These modes (again degenerate pairs for l > 0) are called TElm modes. The first several cutoff frequencies
are
0.293
TE11 : fc,TE11 = √
a µǫ

0.486
TE21 : fc,TE21 = √
a µǫ

0.609
TE01 : fc,TE01 = √
a µǫ
′
From Tables B.1 and B.2, we can see that the TE11 mode (j11 = 1.841) actually has a lower cutoff
frequency than the TE01 or TM01 modes, and is in fact the dominant or fundamental mode of this
waveguide (Figure 5.7). Once again the transverse fields (of the TElmc modes) are obtained from (5.78)
and (5.49):
jωµla2
 ′ 
j
Eρ = ′2 Hlmc sin lφJl lm ρ (5.82)
jlm ρ a

′
 
jωµa jlm
Eφ = ′ Hlmc cos lφJl′ ρ (5.83)
jlm a

uz × ET
HT = (5.84)
ζTE
Transverse field patterns for the TE11c , TE21s and TE01 modes are shown in Figure 5.9.
The transverse fields are only a part of the story of any mode, of course, and it is desirable to be
able to visualize the longitudinal part of the field as well. Since for lossless guides the longitudinal fields
are in phase quadrature with the transverse ones, we cannot simply take the phasor quantities and use
them to plot the field lines. What is normally done is to take a “snapshot” in time of the time-varying
field, and represent it either as a three-dimensional plot, or in a side view cross section, as we do for
the TE11 mode in Fig. 5.10. If we could let time increase in such a figure, we would see the entire field
structure for a propagating mode moving in the direction of positive z. In addition, there are surface
charge and current densities on the waveguide walls that likewise move along with the field distribution.

148
 Figure 5.9: Field distribution of TE11c , TE21s and TE01 circular waveguide modes.

Figure 5.10: Top view of instantaneous field distribution for TE11c circular waveguide mode.

149
 xxxx
xxxx
xxxx
xxxx
rotates fixed
resistive sheet
(a) (b)

Figure 5.11: A circular waveguide attenuator: (a) side view; (b) end view.

5.5.1 Normalized mode fields

The amplitude constants Elms , Elmc , Hlms and Hlmc in (5.74) and (5.78) are determined from the
normalization condition (5.27). The calculations are simplified somewhat by the results of problem p5-14.
Thus, for TM modes, we need
2kc2
Z
Ez2 dS =
S0 jωǫγ
but (5.74) gives after some calculations
jlm
2πa2
Z Z
Ez2 dS = Elm
2
2 ∆ Jl2 (u)u du (5.85)
S0 jlm l 0

where ∆l is the so-called Neumann factor :

 
 
 1
 l=0 

∆l = (5.86)
 
 2
 l=
6 0 


From (B.34), (B.35) and (B.38), we find that the integral can be evaluated explicitly, and we have

πa2 ′2
Z
Ez2 dS = Elm
2
J (jlm ) (5.87)
S0 ∆l l

Thus, we find from (5.27) that

√
kc 2∆l
Elm = √ (5.88)
aJl′ (jlm ) jωǫγπ
and an arbitrary choice of sign for Elm has been made since no convention is in common use for this.
Similarly we find for TE modes that
√
kc 2∆l
Hlm = p
′ ) jωµγπ(1 − l 2 /j ′2 )
(5.89)
aJl (jlm lm

5.5.2 Example: A variable attenuator

A circular waveguide attenuator can be constructed as shown in Figure 5.11. Tapered transitions (not
shown) from rectangular cross-section waveguides are used at either end together with fixed thin sheets of

150
lossy material placed perpendicular to the electric field to assure the exclusive presence of one polarization
of the TE11 mode (say, TE11c ).
To a good approximation, if the thickness of this resistive sheet is small compared to a skin depth,
it can be described by the boundary condition that Etan is continuous across the sheet (whose thickness
is now approximated by zero), and
Etan = RS JS (5.90)
where JS is the surface current in the sheet and RS is the so-called surface resistance of the sheet.
Eqn. (5.90) is a sort of surface version of Ohm’s law. When the sheet is inserted horizontally and thus
lies perpendicular to the electric field lines, little disturbance to the mode field by the sheet takes place.
As would happen if the sheet were a perfect conductor, surface charges and currents appear on the sheet
to properly terminate the electric and magnetic fields. The sheet is so thin that little loss is incurred in
this way.
A rotatable thin sheet of lossy material occupies a diameter of the cross-section of the central portion
of circular waveguide as shown. When it is horizontal, the wave coming from one end of the attenuator
passes through with little disturbance. On the other hand, when the sheet is oriented vertically, it
represents a severe perturbation to the TE11c mode, because it is parallel to the electric field lines. The
modified mode which results has quite large currents JS induced in the sheet, and suffers a good deal of
attenuation as a result. By rotating the sheet between the horizontal and vertical positions, a variable
amount of attenuation can be realized.

5.6 Rectangular Waveguides (Ray Description of the Modes)

Only a few special cross-sectional shapes permit exact determination of the mode fields and eigenvalues
of a hollow waveguide by separation of variables. It is desirable to have some general principles from
which we can deduce some approximate characteristics of modes for any shape of waveguide, and gain
some physical insight into the guiding mechanism as well. For this purpose, we will consider another
method by which some types of hollow metallic waveguides can be exactly analyzed: the ray approach
which was used for the parallel-plate waveguide in section 5.1.
Since the z-dependent factor of a mode field is always exp(−γz), we only need actually concern
ourselves with projections of rays in the transverse (xy) plane of a waveguide. A general plane wave
whose z-dependence is exp(−γz) will have an electric field of the form:

E = ue Eo e−γz e−jkc (x cos θ+y sin θ) (5.91)

where ue is a (constant) unit vector, Eo a constant field amplitude, and θ the angle of the projection of
the ray in the transverse plane. A ray solution of

(∇2T + kc2 )E = 0 (5.92)

will then be made up of reduced or projected plane waves of the form:

E = ue Eo e−jkc (x cos θ+y sin θ) (5.93)

and the H-field will similarly be a sum of terms of the form

H = uh Ho e−jkc (x cos θ+y sin θ) (5.94)

When we speak of rays in the remainder of this section, then, we will actually mean projections of rays
in the xy-plane.
Consider the hollow rectangular metallic waveguide of sides a and d whose cross-section is shown in
Fig. 5.12. For (let us say) the TE modes of such a guide, it is convenient to deal with the field component
Hz (the analysis of the parallel-plate guide could also have been carried out in terms of Hz ). A plane

151
 Figure 5.12: Cross-section of hollow rectangular waveguide

Figure 5.13: Transverse ray directions in a rectangular waveguide.

wave
1 propagating at an angle θ between the x-axis and the projection of the ray in the xy-plane will
have
Hzinc = H1 e−jkc (x cos θ−y sin θ) (5.95)
Upon reflecting the projection of this ray in the waveguide walls, it is evident from Figure 5.13 that
three other ray directions are generated inside the waveguide. Thus, the general solution for Hz must
be

Hz = H1 e−jkc (x cos θ−y sin θ)

+H2 e−jkc (x cos θ+y sin θ)
+H3 e−jkc (−x cos θ+y sin θ) (5.96)
+H4 e−jkc (−x cos θ−y sin θ)

The amplitude Hi , i = 1, 2, 3, 4, of each plane wave corresponds to the ray

i in Figure 5.13.
To determine acceptable values for kc and θ, we need to apply the boundary condition ∂Hz /∂n = 0
on each part of the wall in turn. For example, at the wall y = 0, ∂/∂n = ∂/∂y and we have

∂ n
0 = H1 e−jkc (x cos θ−y sin θ) + H2 e−jkc (x cos θ+y sin θ)
∂y
o
+H3 e−jkc (−x cos θ+y sin θ) + H4 e−jkc (−x cos θ−y sin θ)
y=0
−jkc x cos θ jkc x cos θ

= jkc sin θ (H1 − H2 )e − (H3 − H4 )e

This condition must hold identically for all values of x, which can only happen if the terms dependent
on e−jkc x cos θ and ejkc x cos θ each vanish separately. That is, we can examine separately the reflections
1 → 2 , which gives H1 = H2 , and 4 → 3 , which gives H3 = H4 .

152
 Likewise, from the reflections at x = 0,
3 →
2 and
4 →
1 we get

H3 = H2 and H4 = H1
From the wall x = a:
H3 ejkc a cos θ = H2 e−jkc a cos θ
H4 ejkc a cos θ = H1 e−jkc a cos θ
and finally, from the wall y = d:
H4 ejkc d sin θ = H3 e−jkc d sin θ
H1 ejkc d sin θ = H2 e−jkc d sin θ
From these conditions it is not difficult to show that
H1 = H2 = H3 = H4
kc a cos θ = mπ; m = 0, 1, 2, ... (5.97)
kc d sin θ = nπ; n = 0, 1, 2, ...
Hence, the solution (5.96) in terms of four bouncing plane waves can be recombined to yield
 mπx   nπy 
Hz = 4H1 cos cos ;
a d
 1/2
mπ 2  nπ 2
kc,mn = + ; (m, n not both zero ) (5.98)
a d
 2 1/2 1/2
− ω 2 µǫ = j ω 2 µǫ − kc,mn
2

γm,n = kc,mn
which is, of course, the well-known result for TE modes of a rectangular waveguide. The remaining
transverse fields could now be obtained from (5.49), if desired. Figure 5.14 shows the relative locations
of the cutoff frequencies for a rectangular waveguide with a = 2d. Comparison with the distribution of

TE
TE01 11
TE20
TE10

fc /fc,TE
0 1 2 3 10

TM11

Figure 5.14: Relative cutoff frequencies for rectangular waveguide modes (a = 2d).

cutoff frequencies for the circular waveguide shown in Fig. 5.7 shows that the larger aspect ratio of the
rectangular waveguide has the effect of shifting the cutoff frequencies of the lowest TM mode and the
next highest TE mode higher relative to that of the fundamental TE mode.
Note that the second and third of equations (5.97) represent what are commonly called transverse
resonance conditions: the requirement that any of the plane waves comprising the mode be in phase
with all re-reflected plane waves propagating in the same direction (cf. rays 1 and 7 in Figure 5.13).
This method can also be used for the exact solution of the modes of triangular waveguides in the cases
of vertex angles: (a) 45◦ -45◦ -90◦ ; (b) 60◦ -60◦ -60◦ ; (c) 30◦ -60◦ -90◦. Details are left as an exercise.

153
5.6.1 Normalized mode fields
Whether the TE and TM modes of the hollow rectangular metallic waveguide can be found by the ray
method as above, or by the method of separation of variables, as is done in many introductory texts in
electromagnetics, the fields of a forward TEmn mode are found to be:

mπx nπy 
Hz = Hmn cos a cos d 






jωµ nπ mπx nπy 
Ex = k2 d Hmn cos a sin d 


c 


jωµ mπ mπx
Ey = − k2 a Hmn sin a cos d  nπy (5.99)
c 



Hx = −Ey /ζTE 







Hy = Ex /ζTE 


where 
jωµ 
ζTE = γmn






(5.100)
p
γmn = kc2 − ω 2 µǫ 



mπ 2 nπ 2 
kc2 2

= kc,mn =( a ) +( d ) 
and m and n range through 0, 1, 2, · · · , but are not both zero. The amplitude Hmn is now determined
from (5.27):
1 b a
Z Z
1= E × H · uz dx dy (5.101)
2 0 0
Upon carrying out the integral in (5.101), we find that
√
jkc 2∆m ∆n
Hmn = √ (5.102)
−jωµγmn ad
where ∆n is the Neumann factor defined in (5.86). As noted in previous examples, Hmn is only deter-
mined up to a ± sign. The choice is arbitrary, and as no standard convention exists for these modes, we
have made our selection in eqn. (5.102) so as to make Ey for the normalized TE10 mode, for example,
point in the +y direction.
Likewise, the TMmn forward mode fields are given by

mπx nπy 
Ez = Emn sin a sin d 






γmn mπ mπx nπy 
Ex = − k2 a Emn cos a sin d  

c 


γmn nπ mπx
Ey = − k2 d Emn sin a cos d  nπy (5.103)
c 



Hx = −Ey /ζTM 







Hy = Ex /ζTM 


where γmn and kc2 are the same as in (5.100), while

γmn
ζTM = (5.104)
jωǫ

154
and m and n range through 1, 2, · · · . The normalization condition (5.27) gives
√
2 2jkc
Emn = √ (5.105)
−jωǫγmn ad
where an arbitrary selection of the ± sign has once again been made.

5.7 Dielectric Losses in Hollow Waveguides

Losses occur in real waveguides due to the imperfect conductivity of the walls and also because of the
presence of losses in the dielectric filling the guide. The latter are easily accounted for by the foregoing
theory simply by allowing ǫ → ǫ̂ = ǫ′ − jǫ′′ to be complex. Instead of (5.56) or (5.57), we find that γm
is now neither purely real nor purely imaginary; from (5.54) we have
q
γm ≡ αm + jβm = kc,m 2 − ω 2 µǫ′ + jω 2 µǫ′′ (5.106)

Instead of a sharp transition from cutoff to propagating mode, there is always some attenuation present,
since solving for αm and βm gives
 q 1/2
2
kc,m − ω 2 µǫ′ + (kc,m
2 − ω 2 µǫ′ )2 + (ω 2 µǫ′′ )2
αm =   (5.107)
2

 q 1/2
2
−(kc,m − ω 2 µǫ′ ) + 2
(kc,m − ω 2 µǫ′ )2 + (ω 2 µǫ′′ )2
βm =   (5.108)
2

It can be seen that the frequency

kc,m
ω̃c,m = √ ′ (5.109)
µǫ
serves as a kind of de facto cutoff frequency, and the behavior of αm and βm resembles that of Figure 5.15,
with a smooth transition from attenuating to propagating conditions. The attenuation coefficient αm
never becomes exactly zero, but attains a minimum value at
s
2
ωmin = ω̃c,m (5.110)
1 + (ǫ′′ /ǫ′ )2

and subsequently increases with frequency if ǫ′ and ǫ′′ are substantially independent of frequency. Even-
tually, the attenuation rate gets very large,
ωǫ′′
r
µ
αm → p ′
as ω → ∞ (5.111)
′′ ′
[2 + 2 1 + (ǫ /ǫ ) ]2 1/2 ǫ
although the rate of increase is fairly slow if ǫ′′ ≪ ǫ′ (as will be the case for a practical guide). If ǫ′′
and/or ǫ′ depends on frequency, the details of the behavior of αm are rather more involved.
If ǫ′′ is small enough that
ω 2 µǫ′′ ≪ |ω 2 µǫ′ − kc,m
2
|
which, more than just ǫ′′ ≪ ǫ′ , also requires the mode to be sufficiently far from the de facto cutoff
frequency, then
q
αm ≃ 2
kc,m − ω 2 µǫ′ if ω < ω̃c,m
ω 2 µǫ′′
≃ q if ω > ω̃c,m (5.112)
2 ω 2 µǫ′ − kc,m
2

155
 Figure 5.15: Real and imaginary parts of γm due to dielectric losses in hollow waveguide.

ω 2 µǫ′′
βm ≃ q if ω < ω̃c,m
2
2 kc,m − ω 2 µǫ′
q
≃ ω 2 µǫ′ − kc,m
2 if ω > ω̃c,m (5.113)

The treatment of wall losses in waveguides is less straightforward, and we will delay our study of
them until Chapter 9 when the appropriate tools for the investigation are at our disposal.

156
5.8 Notes and References
For general treatments of waveguide modes, see, for example:

Johnk (1975), chapters 8 and 9;

Johnson (1965), chapter 4;
Ramo, Whinnery and Van Duzer (1984), chapter 8;
Mrozowski (1997);
Rozzi and Mongiardo (1997);

and also:

R.B. Adler, “Waves on inhomogeneous cylindrical structures,” Proc. IRE, vol. 40, pp. 339-
348 (1952).

Use of metallic waveguides at frequencies of several Terahertz is described in:

G. Gallot, S. P. Jamison, R. W. McGowan and D. Grischkowsky, “Terahertz waveguides,” J.

Opt. Soc. Amer. B, vol. 17, pp. 851-863 (2000).

In certain exceptional cases, waveguide modes may not be normalizable because either P̂ or P̂osc (or
both) is zero. This can happen in a lossless waveguide for a so-called complex mode, where γ = α + jβ,
and neither α nor β is zero. Consideration of such modes is somewhat delicate, and the interested reader
is referred to:

A. M. Belyantsev and A. V. Gaponov, “Waves with complex propagation coefficients in

coupled transmission lines without energy dissipation,” Radio Eng. Electron. Phys., vol.
9, pp. 980-988 (1964).
A. Magos, “Singular frequencies in wave guides with inhomogeneous dielectric,” Int. J. Appl.
Electromag. Mater., vol. 4, pp. 189-196 (1994).

and

Mrozowski (1997)

for more information.

The material on the general properties of uniformly filled waveguides is quite standard; see, for
example:

Johnson (1965), Chapter 4;

Ramo, Whinnery and van Duzer (1984), Chapter 8;
Johnk (1988), Chapter 8;
Harrington (1961), Chapter 7;
Collin (1960), Chapter 5;
King, Mimno and Wing (1945), Chapter 3.
Milton and Schwinger (2006).

Proofs for the properties of solutions to (5.51)/(5.52) or (5.51)/(5.53) which are given in section 5.4.3
can be found in, e.g.,

Jones (1964), sections 4.6-4.12 and 5.1-5.2;

but see also

Courant and Hilbert (1953), Chapters 5 and 6;

Garabedian (1964), Chapter 11.

157
In the latter reference, the reader will find the following interesting estimates for the lowest TM and
nontrivial TE eigenvalues of a hollow waveguide of arbitrary shape:
′ 2
κ1 ≤ π(j11 ) /A ≃ 10.65/A

λ1 ≥ π(j01 )2 /A ≃ 18.17/A
where A is the cross-sectional area of the waveguide. It is also known that if κ1 is degenerate, it is at
most doubly so. See:
N. S. Nadirashvili, “On the multiplicity of eigenvalues of the Neumann problem,” Sov. Math.
Dokl., vol. 33, pp. 281-282 (1986).
A useful summary of methods for computing the modes of various kinds of metallic waveguides, along
with an extensive catalog of the then-known results from the literature is given in:
F. L. Ng, “Tabulation of methods for the numerical solution of the hollow waveguide prob-
lem,” IEEE Trans. Micr. Theory Tech., vol. 22, pp. 322-329 (1974).
Information on the use of ray diagrams and transverse resonance methods can be found in:
J.B. Keller and S.I. Rubinow, “Asymptotic solution of eigenvalue problems”, Annals of
Physics, vol. 9, pp. 24-75 (1960).
Continuous rather than discrete sets of rays can be used as the basis of approximate numerical solutions
for waveguide modes. See:
H. Bateman, “On the numerical solution of linear integral equations,” Proc. Roy. Soc.
London A, vol. 100, pp. 441-449 (1922).
N. I. Platonov, “Application of plane wave expansions to the method of partial regions,” Sov.
J. Commun. Technol. Electron., vol. 37, no. 14, pp. 20-28 (1992).

5.9 Problems
p5-1 Prove the statement following equation (5.9) that requiring the tangential electrical field to
vanish at the perfect conductor is equivalent to requiring ∂Hy /∂x to vanish, for the case of
parallel polarization.
p5-2 Derive expression (5.10) using the repeated reflection of plane waves produced by (5.9).
p5-3 Show that
∇ × [E(x, y)e−γz ] = (∇T × E − γuz × E)e−γz

p5-4 Show that for TEM, TE or TM modes of an arbitrary waveguide, the E and H fields are
locally perpendicular, that is, E · H = 0. Will this be true in general for hybrid modes?
p5-5 Give details of the derivation of (5.34) and (5.35).
p5-6 Obtain the TM mode Ez field and eigenvalues for the hollow semi-circular waveguide shown
below, using separation of variables.

158
p5-7 Obtain eqns. (5.99) for the fields of TE rectangular waveguide modes by using the method
of separation of variables. Likewise obtain (5.103) for the fields of the TM modes.
p5-8 Use the ray method to derive the eigenvalues and the longitudinal electric field Ez for the TM
modes of a hollow rectangular waveguide. Show that these are identical to those obtained
using separation of variables.
p5-9 Use the ray method to derive the TM mode eigenvalues for the right isosceles triangular
waveguide (45◦ − 45◦ − 90◦ ) shown below:

p5-10 Repeat Problem p5-9 for the TE mode eigenvalues.

p5-11 The transverse electric fields of a TEmn mode of a rectangular hollow waveguide are given
by (5.99). Using the method of Appendix J, show that the ET -field lines for this mode are
given by  mπx   nπy 
cos cos = K = constant
a d
with different field lines generated by different values of K. Use this expression to plot the
E-lines for the TE11 (rectangular) mode.
p5-12 Compare the field patterns for TE and TM modes of hollow circular waveguides given in
Figures 5.8 and 5.9 with those for TE and TM modes of a rectangular guide shown below:

It seems clear that the TM01 (circular) mode is closely related to the TM11 (rectangular)
mode, and that one would change continuously into the other if the boundary wall were
gradually changed from a circular to a rectangular cross-section. It is likewise clear that a
similar correspondence exists between the TE11 (circular) and the TE10 (rectangular) modes.
Identify the rectangular waveguide modes which correspond to the following modes:

159
 (a) TM11 (circular)
(b) TE01 (circular)
(Hint for part (b): Consider the special case of a square cross-section.)
p5-13 Show that, if ǫ̂ = ǫ′ − jσ/ω, and ǫ′ and σ are independent of frequency, and ω > ω̃c,m ,
then αm due to dielectric losses is a monotonic function of ω, increasing if ω̃c,m < σ/2ǫ′ ,
and decreasing if ω̃c,m > σ/2ǫ′ . The only exception to this is if ω̃c,m = σ/2ǫ′ , when αm is
independent of frequency. Here ω̃c,m is given by (5.109).
p5-14 (a) Show that
jωǫγ
Z
P̂osc = ∇T Ez · ∇T Ez dS
2kc4 S0

for a TM mode, and

jωµγ
Z
P̂osc = ∇T Hz · ∇T Hz dS
2kc4 S0

for a TE mode of a hollow metallic waveguide.

(b) Verify the identity

∇T · (Ez ∇T Ez ) = Ez ∇2T Ez + ∇T Ez · ∇T Ez

Use this result together with that of part (a) to prove that

jωǫγ
Z
P̂osc = E 2 dS
2kc2 S0 z

for a TM mode. Likewise, show that

jωµγ
Z
P̂osc = Hz2 dS
2kc2 S0

for a TE mode.
p5-15 Show that a TM mode of a waveguide possessing an Ez field that behaves as Yl (kc ρ) cos lφ
for ρ near 0 violates the condition that stored energy in a small cylinder about ρ = 0 must
be finite.
p5-16 Show that the complex oscillatory power carried by a TE or TM mode of a hollow metallic
waveguide at the cutoff frequency is either zero or ∞ (if the fields have finite amplitude),
and thus that it is impossible to normalize a such a mode at cutoff.
p5-17 If µ and ǫ of a medium filling a waveguide bounded by perfectly conducting walls are real
functions of x and y, we know from Appendix K that γ 2 is real for modes of this waveguide
for which the complex power P̂ 6= 0. Use equations (5.17)-(5.20) to show that the transverse
electric field ET of the mode satisfies the differential equation
 
1 1
∇T ∇T · (ǫET ) − µ∇T × ( ∇T × ET ) + (ω 2 µǫ + γ 2 )ET = 0
ǫ µ

along with appropriate boundary conditions. Hence, show that ET may always be chosen to
be real before normalization is imposed.
p5-18 A circular metallic waveguide of radius a is fitted with a perfectly conducting fin located in
φ = 0 as shown.

160
 a
φ=0

Obtain expressions for the TE mode cutoff wavenumbers kc for this guide. Give the numerical
value of kc a for the fundamental mode.

p5-19 An X-band rectangular metallic waveguide (a = 1 in., d = 0.5 in.) is uniformly filled with
a lossy dielectric for which µ = µ0 , ǫ′ = 2.25ǫ0 and ǫ′′ = 10−2 ǫ0 . These material properties
are independent of frequency. Find the frequency fmin at which the attenuation coefficient α
of the dominant TE10 mode of this waveguide is a minimum, and calculate the value of this
attenuation coefficient. In spite of the fact that standard waveguide dimensions are given in
inches, express all your results in MKS units.

p5-20 Find the Hz field, and the cutoff wavenumbers kc of the TE modes of a quarter-circle
air-filled metallic waveguide of radius a. The boundary walls are at φ = 0, φ = π/2 and
ρ = a.

p5-21 Show that the E lines of a TE mode obey the equations Hz (x, y(x)) = K, where K is a
constant which takes on different values for different field lines. Show that the H lines of a
TM mode obey the equations Ez (x, y(x)) = K. Use the method of Appendix J.

p5-22 Show that TE modes in a closed waveguide below their cutoff frequency have a greater
magnetic than electric time-average stored energy per unit length (UM0 > UE0 ), while TM
modes on such a guide have UE0 > UM0 . Here the energy densities are as defined in (K.25)
and (K.26). [Hint: Consider the results of problem p5-14, modified to give the complex
power P̂ instead of P̂osc .]

p5-23 Using the ray method, obtain an expression for all the cutoff wavenumbers kc of the TE
modes of the hollow metallic waveguide whose wall is in the form of an equilateral triangle
of side 2a shown below.

161
 y

a√3

-a a x

Use the coordinate axes shown in the figure. Make sure you include all possible ray directions
in your expression for the Hz field.

162
Chapter 6

SURFACE WAVE MODES: BASIC

OPTICAL WAVEGUIDES

Ineluctable modality of the visible: at least that if no

more . . .
—James Joyce,
Ulysses

6.1 Introduction
A metal which is a nearly perfect reflector all the way up to microwave frequencies begins to behave
differently as frequency increases beyond a value of 50 ∼ 100 GHz (the millimeter-wave band). What
constitutes a “smooth” metal surface at 1 GHz might be unacceptably rough (on the scale of a wave-
length) at 100 GHz. Moreover, whereas the metal may be a “good” conductor (ǫ′′ /ǫ′ = σ/ωǫ ≫ 1) at
microwave frequencies, it may exhibit completely different behavior at higher frequencies. For example,
silver at λ = 1 µm has a complex relative dielectric constant of

ǫ̂/ǫ0 = ǫr ′ − jǫ′′r ≃ −25 − j2 (6.1)

or, in terms of the complex refractive index n̂ = (ǫ̂/ǫ0 )1/2 (which is more commonly used at optical
frequencies)1
n̂ ≃ 0.2 − j5 (6.2)
The negative value of ǫ′r indicates very different electrical behavior from that at lower frequencies. In
fact, like many metals at optical frequencies, silver behaves like a plasma (a medium containing free
charges) which can possess permittivities with negative real parts like those of (6.1).
It is clear, then, that the changing material parameters of metals will cause traditional metallic
waveguide structures to perform differently at higher frequencies than as described in chapter 5 (this
is also true of the TEM and quasi-TEM structures which we will study in chapter 8. Moreover, the
fabrication of a traditional single-mode waveguide at such high frequencies may be quite difficult: the
dimensions required for a single-mode metallic rectangular waveguide at λ = 1 µm, for instance, would be
0.5 µm < a = 2b < 1µm. As sources and devices become available at such high frequencies (millimeter-
wave, infrared, visible, ultraviolet and even X-ray frequencies) we need to have practical waveguides
which have low losses to take over the role that traditional waveguides play at lower frequencies.
1 As with ǫ, we will usually write n̂ as simply n, with the understanding that it can be made complex if desired in order

to describe a lossy medium.

163
 Figure 6.1: Total internal reflection at a dielectric interface.

Before we start searching for new structures, however, we should first re-examine the basic physics
of wave guidance. This will provide us with some principles by which useful waveguiding structures can
be designed.
It will be recalled from Chapter 5 that the propagating modes of metal-boundary waveguides can
be interpreted in terms of propagating plane waves which are repeatedly reflected from the conducting
walls. Similarly, one may interpret TEM modes (chapter 8) as basically axially propagating plane waves
whose fields have been “warped” in order to satisfy boundary conditions at the two conductors which
guide the wave, but which have the same propagation coefficient and wave impedance as the plane wave.
The mechanism of successive reflections does not require metallic boundaries. It can also take place
with dielectric interfaces under appropriate conditions. A plane wave incident on a dielectric interface
as shown in Figure 6.1 may be totally reflected if the angle of incidence is greater than the so-called
critical angle. This critical angle exists only if ǫ1 > ǫ2 , but suggests a variety of possible waveguiding
structures using this total reflection phenomenon. Figure 6.2 illustrates a variety of guides of this type.
Figure 6.2(a) depicts the step-index optical fiber, which is perhaps the simplest realistic guide of this
type. Figure 6.2(b) illustrates a channel guide embedded in a substrate, as is encountered in what has
come to be called integrated optics. The rib guide of Figure 6.2(c) operates on slightly more complicated
principles, as we will see in Chapter 7. The image guide shown in Figure 6.2(d) is commonly used at
millimeter-wave frequencies, since its ground plane provides desirable partial shielding of the waveguide
circuitry. All these waveguides are practical because smooth interfaces between dielectrics can be made,
and materials are available in the mm-wave band and higher which have acceptably low losses. This
type of guide is especially ideal in the optical region.
Indeed, it is even possible to induce the change in permittivity by external means, using the electro-
optic or Pockels effect. This effect appears in certain types of materials (LiNbO3 or GaAs, for instance),
and manifests itself as a change in the index of refraction seen by an optical frequency field due to the
presence of a much lower frequency (e. g., microwave) electric field in the material. Thus, a waveguide
can be created where none was present before by impressing a DC voltage across a pair of electrodes on a
piece of electro-optic material as shown in Fig. 6.3(a). Another application of these materials is to change
the properties of an existing waveguide by impressing a variable voltage across a pair of electrodes near
the guide region. This can achieve a modulation of the phase of a signal in the waveguide, and offers the
potential of utilizing the very high bandwidth that optical carrier frequencies offer (Fig. 6.3(b)). Further
information can be found in the references cited at the end of the chapter.

6.2 Dielectric Slab Waveguides—The Ray Description

In order not to obscure the essentials of the problem, it is traditional to treat first the simple, though
somewhat unrealistic, dielectric slab waveguide of Figure 6.4. Consider a dielectric slab of thickness
a and permittivity ǫ1 situated between two semi-infinite cladding regions of permittivity ǫ2 as shown
in Figure 6.4. Let us not yet make any restrictions on ǫ1 and ǫ2 . The slab and cladding extend to
infinity in both y- and z-directions, and, to obtain a purely two-dimensional problem, we seek solutions

164
Figure 6.2: (a) Optical fiber; (b) Channel (integrated optical) guide; (c) Rib guide; (d) Image guide.

165
Figure 6.3: (a) Electro-optically induced waveguide; (b) Electro-optic modulator.

Figure 6.4: The dielectric slab (ǫ1 > ǫ2 ).

166
 Figure 6.5: Plane-wave reflection at a dielectric interface—TE polarization.

independent of y:
∂
=0
∂y
This structure is then analogous to the parallel-plate metallic waveguide considered in section 5.1. All
media are assumed to have the same permeability µ = µ0 .
In order to treat this structure using the ray method, we consider the plane-wave reflection problem
for a dielectric interface in some detail (note the change in coordinate system from the treatment in
section 1.4.1, which now aligns the z-direction with the direction of propagation of the mode). For a
TE-polarized incident wave

Einc = uy E0 e−jk0 n1 (−x cos θ+z sin θ) (x > 0) (6.3)

as shown in Figure 6.5, there will arise a reflected wave

E r = uy ρ E0 e−jk0 n1 (x cos θ+z sin θ) (x > 0) (6.4)

TE

and, in the second dielectric, a transmitted wave

E t = uy τ E0 e−jk0 n2 (−x cos θt +z sin θt ) (x < 0) (6.5)

TE

where E0 is the amplitude of the incident wave, ρ the electric-field reflection coefficient of the wave,
TE
and τ is the electric-field transmission coefficient into the second region.
TE
If both regions are assumed to be nonmagnetic, the Fresnel reflection coefficient ρ from (1.80)
TE
takes the form:
n1 cos θ − n2 cos θt
ρ = (6.6)
TE n1 cos θ + n2 cos θt
where θt is related to θ by Snell’s law:

n2 sin θt = n1 sin θ (6.7)

167
If n1 > n2 , there is a critical angle θ = θc given by
 
n2
θc = sin−1 (6.8)
n1
such that when θ > θc , the transmission angle θt becomes complex in such a way that sin θt > 1 and cos θt
is negative imaginary. Under these conditions, we have from (1.84)-(1.85) that |ρ | = 1; specifically,
TE
jχTE (θ)
ρ =e (θ > θc ) (6.9)
TE

where
q   
n21 sin2 θ − n22
" #
−1  −1 n 1 cos θ −1 n 1 cos θ
χ (θ) = 2 tan  = π − 2 tan  q  = π − 2 sin p
TE n1 cos θ n2 sin2 θ − n2 n21 − n22
1 2
(6.10)
Hence we have the phenomenon of total internal reflection when n1 > n2 ; all energy present in the
incident wave is carried back away from the interface in the reflected wave. The transmitted wave is
evanescent ; the x-dependent factor behaves as
( p )
sin2 θ − sin2 θc
exp −k0 n2 |x|
sin θc

and decays exponentially as x becomes more negative. This wave carries no energy away from the
interface but merely stores it temporarily in the vicinity of x = 0 before it is carried away by the
reflected wave.
If we compare equation (6.9) to the reflection coefficient of this wave at a perfect conductor, ejπ
(obtained from (6.9) and (6.10) by letting n2 become complex and taking |n2 | → ∞), we see that in our
case there is an additional phase shift after reflection as compared to the perfectly-conducting boundary.
It is as if the reflection had actually occurred at a perfectly conducting boundary located at a distance
dTE (θ) somewhat below the position of the dielectric interface. The additional phase shift is now due to
further propagation through the same material as medium 1 (see Figure 6.6):
π−χ
 
1 TE
dTE (θ) =
k0 n1 cos θ 2
" #
1 −1 n1 cos θ
= sin p (6.11)
k0 n1 cos θ n21 − n22
The field Ey at x = −dTE would be zero if such a perfect conductor were actually present. This shift
in the apparent location of the reflecting boundary also can be observed for reflection of a light beam of
finite width, in which case it is called the Goos-Hänchen shift, after the researchers who first detected
the effect experimentally. In the limit of grazing incidence (θ ≃ π/2), dTE approaches the value
1
dTE ≃ p (6.12)
k0 n21 − n22
which is independent of incidence angle. We show how dTE varies with θ between grazing and critical
angles in Figure 6.7. If cos θ < 0.25 cos θc , the error in using the approximate value in (6.12) is less than
1%.
We can now apply the same procedure as in the case of the parallel-plate waveguide (Chapter 5) to
obtain a transverse resonance condition for the TE modes of this dielectric slab. If in 0 < x < a, the
field is assumed to be a superposition of upward and downward traveling plane waves
Ey = E1 e+jk0 n1 x cos θ + E2 e−jk0 n1 x cos θ e−jk0 n1 z sin θ


(6.13)

168
 x

Actual interface location

n1
x=0 n1 z

Location of equivalent perfect concutor

Figure 6.6: Goos-Hänchen shift modeled as reflection from displaced perfect conductor–TE polarization.

as shown in Fig. 6.8, then the reflection conditions at x = 0 and x = a are:

E2 = ρ E1
TE
E1 ejk0 n1 a cos θ = ρ E2 e−jk0 n1 a cos θ (6.14)
TE

or,
exp[2jk0 n1 a cos θ] = exp[2jχ (θ)] (6.15)
TE

Hence, the TE mode transverse resonances for the dielectric slab occur when θ satisfies the equation

k0 n1 a cos θ = χ (θ) + mπ ; m = 0, 1, 2, ... (6.16)

TE

Each value of m corresponds to a mode of the slab, which we designate as the TEm mode.
In view of (6.11), we can also write equation (6.16) as

k0 n1 (a + 2dTE ) cos θ = (m + 1)π (6.17)

Comparing this with (5.5), we can see that the TE modes of the dielectric slab can be thought of as
equivalent to those of a parallel-plate waveguide of an equivalent width aeq = a + 2dTE (Figure 6.9).
The equivalent width depends not only on the contrast in refractive indices n1 and n2 , and on the
wavenumber k0 , but also on the angle θ at which the plane waves are incident at the interfaces. If the
incidence angle is sufficiently close to grazing, we can see from (6.12) that the equivalent parallel-plate
waveguide will have the same equivalent width

2
aeq ≃ a + p
k0 n21 − n22

for any mode at a given frequency of operation.

169
 1.6

1.5

1.4

dTE(θ)/dTE(θ)
1.3

1.2

1.1

1
0 0.2 0.4 0.6 0.8 1

cos θ/cos θc

Figure 6.7: Dependence of normalized dTE on angle of incidence.

x
n2
x=a

2
θ θ n1
1

x=0
z
n2

Figure 6.8: Successive ray reflections in a dielectric slab waveguide.

6.3 Dielectric Slab Waveguides—Field Analysis

While the ray description of slab waveguide modes yields much in the way of physical insight, it is rather
limited in terms of the class of waveguides which it can treat. In this section, we will rederive the TE
mode eigenvalue equation and fields from a more general viewpoint.
On setting ∂/∂y = 0 in equations (5.17)-(5.20), we get two separate groups of equations:

γEy = −jωµHx
∂Ey
= −jωµHz (6.18)
∂x
∂Hz
− − γHx = jωǫEy
∂x

170
 dTE
n1
x=a

aeq n1 Equivalent perfect conductors

x=0 n1
dTE

Figure 6.9: Equivalent parallel-plate waveguide for TE modes.

γHy = jωǫEx
∂Hy
= jωǫEz (6.19)
∂x
∂Ez
− − γEx = −jωµHy
∂x
The field components (Ey , Hx , Hz ) are completely decoupled from (Hy , Ex , Ez ) as far as (6.18) and (6.19)
are concerned, and the boundary conditions do not couple them either. We can thus examine each set
of components separately.
The fields satisfying (6.18) are TE-polarized: by eliminating Hx and Hz among the three equations,
we obtain a reduced wave equation for Ey :
 2 
∂ 2 2
+ ω µǫ + γ Ey = 0 (6.20)
∂x2
Of course, the x-derivatives are really ordinary derivatives, and we anticipate that γ = jβ for useful
propagating modes. Moreover, ǫ depends on position, equaling ǫ1 for 0 < x < a, and ǫ2 for x < 0 or
x > a. We thus rewrite (6.20) as
 2 
d 2
+ h 1 Ey = 0; (0 < x < a) (6.21)
dx2
 2 
d 2
− p 2 Ey = 0; (x > a or x < 0) (6.22)
dx2
where
h21 = ω 2 µǫ1 − β 2 = k12 − β 2 (6.23)
p22 = β 2 − ω 2 µǫ2 = β 2 − k22 (6.24)
and for definiteness, we specify that p2 and h1 be taken with positive real parts or purely positive
imaginary. Note that h1 and β are related to the ray angle θ of section 6.2 by
β = k1 sin θ
h1 = k1 cos θ (6.25)
In the cladding, the solutions have the general form
Ey = Ae−p2 x + Be+p2 x ; (x > a)
= Ce−p2 x + De+p2 x ; (x < 0) (6.26)

171
Now, the solutions we seek are guided modes; if the fields are finite, we demand that they carry a finite
amount of power (per unit width in the y-direction in this two-dimensional waveguide). A field solution
that does not decay at infinity is not guided in this sense because the integral that gives the total power
is divergent (it extends to x = ±∞). Such a field would be impossible to excite with an actual source of
finite power. Thus we eliminate the growing solutions by setting B = C = 0;

Ey = Ae−p2 x (x > a)
= De+p2 x (x < 0) (6.27)

Within the slab, the general solution of (6.21) is equally well known:

Ey = E cos(h1 x) + F sin(h1 x) (0 < x < a) (6.28)

The remaining unknown constants A, D, E, F and β (which is, of course, embedded in h1 and p2 ) must
be found by enforcing the boundary conditions at x = 0 and x = a.
At x = 0, Ey and Hz (the tangential field components) must be continuous. Continuity of Ey gives,
from (6.27) and (6.28),
Ey |x=0 = D = E (6.29)
On the other hand, continuity of Hz implies continuity of dEy /dx [cf. (6.18)]; at x = 0 this gives

dEy 
= p 2 D = h1 F (6.30)
dx x=0

At x = a, continuity of Ey gives

Ey |x=a = Ae−p2 a
= E cos h1 a + F sin h1 a (6.31)

Using (6.29)–(6.31), we can eliminate all the unknown amplitude constants except D, and reduce
(6.27) and (6.28) to

Ey = Dep2 x (x < 0)
 
p2
= D cos h1 x + sin h1 x (0 < x < a) (6.32)
h1
 
−p2 (x−a) p2
= De cos h1 a + sin h1 a (x > a)
h1

As always, we are left with one remaining indeterminate amplitude constant (in the absence of a specified
source), but our one remaining boundary condition will give us a condition on the propagation coefficient
β. Continuity of dEy /dx at x = a gives
 
p2
−p2 D cos h1 a + sin h1 a = D [−h1 sin h1 a + p2 cos h1 a]
h1
or
2p2 h1
tan(h1 a) = (6.33)
h21 − p22
This eigenvalue equation determines the allowed values of propagation coefficient β (recall that h1 and
p2 are both specified functions of β from (6.23) and (6.24)).
In the notation of this section, the previously obtained eigenvalue equation (6.16) has the form
 
p2
h1 a = mπ + 2 tan−1 (6.34)
h1

172
or, alternatively,  
h1
h1 a = (m + 1)π − 2 tan−1 (6.35)
p2
Other forms of the eigenvalue equation are
2p2 h1
sin(h1 a) = (−1)m (6.36)
k12 − k22

h21 − p22
cos(h1 a) = (−1)m (6.37)
k12 − k22
It is left as an exercise for the reader to demonstrate that these are all equivalent to (6.33). We will
address the solution of the eigenvalue equation in the next section; let us now look briefly at a few ways
of expressing the field distribution.
Expression (6.32) for the field in the core (0 < x < a) can be rewritten in the form
sin[h1 (x + dTE )]
Ey = D (6.38)
sin(h1 dTE )
where, from (6.25), we have
! !
1 h1 1 h1 1 p2
dTE = tan−1 ( ) = sin−1 p = cos−1 p (6.39)
h1 p2 h1 k12 − k22 h1 k12 − k22

With D as an arbitrary amplitude, we recognize this as a TE mode of the equivalent parallel-plate

waveguide of Figure 6.9.
Because of the symmetry of the slab waveguide, its mode fields are often referred to the center of the
slab rather than to one of its boundaries as we have done here. To obtain a symmetric form, we write
(6.38) as
D n a a
Ey = cos[h1 ( + dTE )] sin[h1 (x − )]
sin(h1 dTE ) 2 2
a a o
+ sin[h1 ( + dTE )] cos[h1 (x − )] (6.40)
2 2
From (6.35) and (6.39), we have h1 ( a2 + dTE ) = (m + 1)π/2, so that the first term in (6.40) vanishes
when m is even, and the second when m is odd. The field distributions Ey of the TEm modes thus share
the parity of their mode index:

(−1)m/2 D a
Ey = cos[h1 (x − )] (m even)
sin(h1 dTE ) 2
(−1)(m+1)/2 D a
= sin[h1 (x − )] (m odd) (6.41)
sin(h1 dTE ) 2
Using (6.36) and (6.37), we find that the field in x > a from (6.32) can be rewritten as

Ey = (−1)m De−p2 (x−a) (x > a) (6.42)

Comparing this to the first line of (6.32), we see that the external fields are also even or odd according to
the value of m. The form of the eigenvalue equation usually encountered when the modes are considered
in this way is
h1 a
tan( ) = p2 /h1 (m even)
2
= −h1 /p2 (m odd) (6.43)

173
which can be obtained directly from (6.33) using the half-angle trigonometric identity.
The field amplitude D is determined by imposing the normalization condition (5.27). It is shown in
problem p6-3 that

1 ∞ 1 ∞ βaeff 2 ω(ǫ1 − ǫ2 )
Z Z
E × H · uz dx = − Ey Hx dx = D (6.44)
2 −∞ 2 −∞ 4 h21

where aeff is the effective width of the slab, defined (for a given mode at a specific frequency) as

aeff = a + 2/p2 (6.45)

The effective width is the physical width of the core plus a so-called penetration depth, 1/p2 , (the distance
from the dielectric interface at which the field in the cladding has decayed to 1/e ≃ 37% of its value
at the interface) added on to each side of the core. This effective width turns out to be a more useful
measure of the concentration of energy near the core than the equivalent width aeq = a + 2dTE defined
earlier. Note that in this example, as for all two-dimensional waveguides with ∂/∂y = 0 (e. g., the
parallel-plate waveguide in chapter 5), we must carry out the normalization in (5.27) using an integral
over x only, since the cross-section is uniform and infinite in the y-direction. Equating (6.44), which is
the two-dimensional equivalent of (5.27), to 1, we have

2h1 a ωµ
r
D= (6.46)
V βaeff

where V is defined by (6.47) below.

6.4 Dielectric Slab Waveguides—The Eigenvalue Equation for

TE Modes
The eigenvalue equation in any of its forms (6.33)–(6.35) or (6.43) cannot be solved for β explicitly;
such an equation is known as a transcendental equation. Either graphical, numerical, or approximate
methods must therefore be used to determine the propagation coefficients. Although a computer program
which will find the roots of the eigenvalue equation can be constructed, the task is not altogether a
straightforward one. The iteration method outlined in Appendix C is one possibility, but it requires
rearranging the eigenvalue equation into a suitably convergent form. An alternate approach to the
solution of the eigenvalue equation is a graphical method. from which much insight can be obtained.
For this purpose it is most convenient to deal with equations (6.43). Introducing a normalized frequency
V as
q
V = k12 − k22 a
q
= n21 − n22 k0 a (6.47)
p
= (p2 a)2 + (h1 a)2

where n1 and n2 are the core and cladding refractive indices respectively, equations (6.43) become, after
some rearrangement,
p
V 2 − (h1 a)2 = p2 a = h1 a tan(h1 a/2) ≡ Fe (h1 a) (m even)
= −h1 a cot(h1 a/2) ≡ Fo (h1 a) (m odd) (6.48)

Figure 6.10 gives plots of the left and right sides of equations (6.48). The circle of Figure 6.10(c) will
be expanded and/or contracted as the value of V is varied. For values of h1 a > V , the square root in

174
 p
Figure 6.10: (a) Plot of Fe (h1 a) vs. h1 a; (b) Plot of Fo (h1 a) vs. h1 a; (c) Plot of V 2 − (h1 a)2 .

175
 Figure 6.11: Graphical solution of (6.47) for V < π.

(6.48), which is equal to p2 a, becomes imaginary, and no such solutions are allowed because they do not
decrease away from the slab (cf. (6.32)). Similarly, imaginary values of h1 are not allowed because in
that event the left side of (6.48) would be positive real, while the right side was negative real, and no
root would exist.
By superposing the graphs of Figure 6.10(a) or (b) with Figure 6.10(c), we locate solutions of (6.43)
at the points of intersection of the curves. For V < π, the situation is as shown in Figure 6.11. Only
one intersection occurs—for 0 < h1 a ≤ π—and the corresponding mode of propagation (the dominant
TE mode) is called the TE0 mode. This mode has zero cutoff frequency, i.e., like a TEM or quasi-TEM
mode, it always propagates, and in this case, has a propagation coefficient β near that of a plane wave in
the cladding, k2 . When V is increased beyond π, a second intersection—the TE1 mode—occurs between
π ≤ h1 a ≤ 2π as indicated in Figure 6.12. Evidently, a new mode solution appears each time V passes
through mπ, where m is a positive integer. Thus, the frequency
mc
fc,m = p (6.49)
2a n21 − n22

is the cutoff frequency for the TEm mode. It further follows from (6.34) and (6.35) that for this mode

mπ (m + 1)π
< h1 < .
a a
Actually, the term “cutoff” could be considered slightly misleading here. In a closed metallic wave-
guide, a mode passes from one which propagates to one which attenuates in z at the cutoff frequency.
In a slab waveguide (or indeed, any open waveguide—whose “boundary” is at infinity) the mode simply
ceases to exist when the frequency decreases to that of mode cutoff. As can be seen from (6.46), the

176
 Figure 6.12: Graphical solution of (6.47) for π < V < 2π.

normalized mode amplitude D then becomes zero, because otherwise the mode field would be carrying
infinite energy. Thus, at any finite frequency, only a finite number of guided modes can propagate. In
this respect, dielectric waveguides differ quite substantially from their hollow metallic counterparts.
Another difference between these modes and those of hollow metallic waveguides is that the field
patterns of these modes depend on frequency as does the propagation coefficient. Near cutoff, p2 a is
very small, and h1 a is very close to its lower limit, mπ. The result is a mode whose fields extend a
large distance beyond the core-cladding boundaries as shown in Figure 6.13(a). Thus, for example, the
fundamental TE0 mode could not be used at arbitrarily low frequencies (even though it has, in principle,
zero cutoff frequency) because of the large field spread in this range (Figure 6.13(a)). The effective width
aeff of the mode can be used as a gauge of the field spread when deciding whether a surface wave mode
can be used in a given situation. Since in an actual optical waveguide the cladding does not extend to
infinity, it is important to keep the value of aeff well below the dimension of the cladding region.
Far from cutoff, on the other hand, h1 a approaches the limiting value of (m + 1)π, and for large V
we must also have large p2 a. The result of this is that very little field penetrates into the cladding region
(aeff ≃ aeq ≃ a). In particular, the boundary shift (6.39) is very small. The fields very nearly are zero
at the core boundaries, as shown in Figure 6.13(b).
This dependence of the field pattern on V (and thus, for fixed frequency, on a) is the basis of an
end-fire antenna design. If we start with a wide (V ≫ 1) slab waveguide on which the TE0 mode is
propagating, the slab width can then be gradually tapered by reducing a until the field spreads over a
much wider region. By this point the slab is very narrow and can be allowed to disappear altogether.
At this point, the field of the surface wave radiates into the uniform space in the manner of an aperture
antenna, and produces a very narrow beam because of the wide extent of the aperture field.
We can use the information assembled above together with equations (6.32) and (6.41) to construct
field plots for the first few modes as shown in Figure 6.14. It is seen that the mode index m refers to the
number of zeroes of Ey within the core region. The field decays exponentially with transverse distance
away from the slab, and is thus localized within the core and near the surfaces separating core from
cladding. For this reason, a mode with such a field pattern is often called a surface-wave mode.

177
 Ey Ey

x x
0 a 0 a

(a) (b)

Figure 6.13: TE0 fields: (a) near cutoff; (b) far from cutoff.

This kind of mode is also called a slow-wave mode, because its phase velocity is smaller than that
of a plane wave in the surrounding (cladding) region. A uniformly filled metallic waveguide mode by
contrast has a larger phase velocity than that of a plane wave in the filling medium, and could thus be
referred to as a fast-wave mode. A fast-wave surface-wave mode is an impossibility, as the fields in the
cladding medium would not decay with distance away from the core.
We can make a final practical observation at this point. Whereas the dimension of a parallel-plate
metallic guide must be such that
λ0
k0 a < π or a<
2
in order to support only a single mode of propagation (TEM in this case), for the slab we need only
V < π, or
λ0
q
n21 − n22 k0 a < π or a< p 2
2 n1 − n22
for only a single TE mode to propagate (as we will see a bit later, a TM mode can also propagate along
the slab under these conditions). Thus if we make the dielectric contrast small (which is quite natural
if core and cladding are to be made of similar materials), a considerably larger guide width a can be
tolerated than in the metallic waveguide case. For example, suppose that the slab is made from a glass
whose refractive
p index is n2 = 1.5, and the core is doped slightly to increase this value by 1%. Then
the quantity n21 − n22 (which is sometimes referred to as the numerical aperture or N A of the guide) is
about 0.21. This means that at a wavelength of λ = 1µm (f = 3 × 1014 Hz = 300 Terahertz (THz)!), the
dimension a can be as large as 2.4 µm, nearly 5 times as large as the corresponding metallic waveguide
for single-mode operation.
Even though the eigenvalue equation is not solvable for β explicitly, we can use (6.35) and (6.39) to
put it in a form which is solved for V in terms of hdTE :

V sin h1 dTE = (m + 1)π − 2h1 dTE (6.50)

It is thus quite easy to plot a graph of h1 dTE vs. V , but it is more customary instead to plot a confinement
parameter b:
 p a 2
2
b = cos2 h1 dTE = (6.51)
V

178
 Ey Ey

(TE ) (TE )
0 1

x x
0 a 0 a
Ey
(TE )
2

0 a x

Figure 6.14: Ey field plots for TE0 , TE1 and TE2 modes.

The confinement parameter is zero at cutoff (when the field is not well-confined to the vicinity of the
core) and approaches unity far from cutoff (when the field is highly confined near the core). In view of
(6.51), it is convenient to rewrite (6.50) in terms of V and b:
√ √
V 1 − b = (m + 1)π − 2 cos−1 b (6.52)

or √ √
V 1 − b = mπ + 2 sin−1 b (6.53)
One also sometimes sees plots of the effective refractive index of a mode:

neff ≡ β/k0 (6.54)

but since b can be expressed in terms of neff :

n2eff − n22
b= (6.55)
n21 − n22
and since a plot of b vs. V is a universal curve for TE modes (cf. (6.52) or (6.53)) independent of a, n1 ,
n2 and f , it is preferable to plot b and to obtain neff from it. Figure 6.15 shows a universal plot of b vs.
V for the first several TE modes. Note that the inequality 0 ≤ b < 1 implies that n2 ≤ neff < n1 for all
modes of the slab.

6.4.1 Near-cutoff and far-from-cutoff approximations

In parameter ranges both near and far from cutoff, it is possible to obtain explicit (though approximate)
expressions for b (and thereby, for the related quantities β, neff , p or h) as a function of V . For example,

179
 1

0.8

0.6

b
0.4

m=0
m=1
0.2
m=2
m=3
m=4
m=5
0
0 5 10 15 20 25 30
V

Figure 6.15: Confinement parameter b for TEm modes of a dielectric slab.

near cutoff, b → 0 and V → mπ for the TEm mode. We put b = sin2 ξ in (6.53) to get

V cos ξ = mπ + 2ξ (6.56)

The graphical meaning of ξ is shown in Figures 6.11 and 6.12. For small values of ξ, we can put
cos ξ ≃ 1 − ξ 2 /2 in (6.56), and solve the resulting quadratic equation for ξ. The resulting expression for
b is:
[4 + 2V (V − mπ)]1/2 − 2
 
2
b ≃ sin (near cutoff) (6.57)
V
√
Far from cutoff, on the other hand, b → 1 and V 1 − b → (m + 1)π; for this limit it is more
convenient to start from (6.52). We have that the angle φ = π/2 − ξ is small in this case; if we write
(6.52) as
V sin φ = (m + 1)π − 2φ (6.58)
we can approximate φ by sin φ. The resulting equation gives
 2
(m + 1)π
b≃1− (far from cutoff) (6.59)
V +2

Approximations (6.57) and (6.59) are plotted in Figure 6.16 along with the exact values of b for the
TE0 and TE1 modes. Note that (6.59) can be written in terms of β as
 2
(m + 1)π
β 2 ≃ k12 −   (6.60)
a + √ 22 2
k0 n1 −n2

180
 1

Exact TE0
0.8 Near cutoff
Far from cutoff

0.6 TE1
b
0.4

Exact
0.2 Near cutoff
Far from cutoff

0
0 2 4 6 8 10
V

Figure 6.16: Exact, near-cutoff and far-from-cutoff approximations for the confinement parameter of the
TE0 and TE1 dielectric slab waveguide modes.

The quantity
 
 (m + 1)π 
a + √ 22
k0 n1 −n22

is the cutoff wavenumber p k̃c of the TMm+1 mode of a metallic parallel-plate waveguide whose plate
separation is a + 2/k0 n21 − n22 , which is the grazing-incidence limit of the effective width aeq shown in
Fig. 6.9 (cf. (6.12)), and also of the effective width aeff defined in (6.45).

6.5 TM Modes of the Dielectric Slab

The TM modes can be analyzed in almost precisely the same way, either from ray description (cf. prob-
lem p6-1) or by proceeding from equations (6.19) to obtain expressions for Hy . While Hy must be
continuous at the dielectric interfaces (it is a tangential field component), the boundary condition that
∂H
tangential E (Ez in this case) be continuous now forces 1ǫ ∂xy to be continuous at the interfaces, i. e.,
 
1 dHy  1 dHy 
= (6.61)
ǫ2 dx x=0− ǫ1 dx x=0+

The resulting eigenvalue or mode equation is somewhat different from (6.33); we can express it in the
form:
2p2 h1
tan(h1 a) = 2 2 2 (6.62)
h1 (n2 /n1 ) − p22 (n21 /n22 )

181
or ( √ )
√ (n1 /n2 ) b
V 1 − b = mπ + 2 sin−1 p (6.63)
(n1 /n2 )2 b + n22 (1 − b)/n21
Although this latter equation is slightly different than (6.53), the general properties of its solutions are
similar. Details can be found in any of the standard references at the end of this chapter. Note that
(6.63) is again solved for V as a function of b, but does not generate universal curves of b vs. V , because
the additional parameter n1 /n2 appears in the equation. In the same way as for TE modes, we can
obtain approximate formulas for b as a function of V which are good both near and far from cutoff
(though they are somewhat more complicated in form; see problem p6-7).

6.6 The Grounded Dielectric Slab

A structure related to the dielectric slab is the grounded dielectric slab shown in Figure 6.17. The
slab lies on a conducting ground plane, and its thickness is denoted by d. This configuration is also an

ε2
x=d
ε1

x = 0 xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Ground plane

Figure 6.17: The grounded dielectric slab.

important building block in the analysis of more elaborate quasi-optical waveguides. Its solutions are
easily obtained from those already obtained for the ungrounded slab.
Consider, for example, the TE modes of the ungrounded slab. The odd TE modes, according to
(6.41), have Ey = 0 at the center of the slab (x = a/2) by virtue of their symmetry. But this means
that a perfectly conducting plane could be inserted at the center of the slab without disturbing any of
the fields, because the boundary conditions would be properly satisfied. Thus, the TE modes of the
grounded slab of thickness d have identical fields (above the ground plane) and identical propagation
coefficients with those of the odd TE modes of an ungrounded slab of twice the thickness (a = 2d). In
particular, its eigenvalue equation, by (6.43), is
h1
tan(h1 d) = − (6.64)
p2
and is solved by the same methods as before.
On the other hand, the analysis of the TM modes of the ungrounded slab shows that Ez = 0 at the
center of the slab for the even TM modes. Again, we can insert a conducting plane in the center of
the slab without affecting these modes, and thereby obtain the TM modes of the grounded slab. The
resulting eigenvalue equation for these TM modes is
ǫ 1 p2
tan(h1 d) = (6.65)
ǫ 2 h1

182
 Figure 6.18: The circular optical fiber.

Because only the odd TE modes and even TM modes of the ungrounded slab are present, the only
mode
p of the grounded slab with no cutoff frequency is the TM0 mode. The TE1 mode will not exist if
k12 − k22 d < π/2.

6.7 Circular Optical Fibers

We turn now to the rather more realistic geometry of Figure 6.18, the circular optical fiber. The core
region, ρ ≤ a, has a permittivity ǫ1 which is larger than that of the cladding (ρ ≥ a), ǫ2 . As with the
slab, we will assume that µ = µ0 everywhere. For this waveguide, we cannot necessarily assume that
only TE or TM modes will exist; we must assume that the modes will be hybrid modes, taking Ez 6= 0
and Hz 6= 0, and dealing with all six components of Maxwell’s equations in (5.17)-(5.20). Naturally,
circular cylindrical coordinates are most convenient here, and we use them to express the transverse
fields in terms of Ez and Hz by means of (5.22) and (5.23).
Let us again introduce the notations
γ = jβ
q
p2 = β 2 − k22
q
h1 = k12 − β 2
which were used in the analysis of the slab, and thus write (5.22) and (5.23) as
 
1 ∂Ez jωµ ∂Hz
Eρ = 2 −jβ − (6.66)
h1 ∂ρ ρ ∂φ
 
1 jβ ∂Hz ∂Ez
Hφ = 2 − − jωǫ1 (6.67)
h1 ρ ∂φ ∂ρ
 
1 jβ ∂Ez ∂Hz
Eφ = 2 − + jωµ (6.68)
h1 ρ ∂φ ∂ρ
 
1 ∂Hz jωǫ1 ∂Ez
Hρ = 2 −jβ + (6.69)
h1 ∂ρ ρ ∂φ
when ρ < a, and as  
1 ∂Ez jωµ ∂Hz
Eρ = jβ + (6.70)
p22 ∂ρ ρ ∂φ

183
  
1 jβ ∂Hz ∂Ez
Hφ = + jωǫ 2 (6.71)
p22 ρ ∂φ ∂ρ
 
1 jβ ∂Ez ∂Hz
Eφ = 2 − jωµ (6.72)
p2 ρ ∂φ ∂ρ
 
1 ∂Hz jωǫ2 ∂Ez
Hρ = 2 jβ − (6.73)
p2 ∂ρ ρ ∂φ
when ρ > a. These expressions should be compared with equations (5.38)-(5.39) for the homogeneously-
filled metallic guide.
Substitution of (6.66)-(6.69) or (6.70)-(6.73) into (5.18) and (5.20) gives us reduced wave equations
for Ez and Hz :

(∇2T + h21 )Ez = 0

(∇2T + h21 )Hz = 0 (6.74)

when ρ < a, and

(∇2T − p22 )Ez = 0

(∇2T − p22 )Hz = 0 (6.75)

when ρ > a. The transverse Laplacian ∇2T is given (in cylindrical coordinates) by

1 ∂ ∂f 1 ∂2f
∇2T f ≡ ρ + 2 2 (6.76)
ρ ∂ρ ∂ρ ρ ∂φ
(cf. eqn. (5.43)).
Equations (6.74) and (6.75) are solved by separation of variables, in a manner similar to that used
for the circular metallic guide in Chapter 5. In light of our experience with that problem, we expect the
solution for Ez to take the form
 
 
 cos lφ 
 
Ez (ρ, φ) = C1 Fl (ρ) ; l = 0, 1, 2, ... (6.77)
 
 sin lφ 
 

the choice of cos lφ corresponding to a “c” polarization and that of sin lφ corresponding to an “s”
polarization. We have included an arbitrary constant C1 in this solution. The function Fl (ρ), which is
yet to be determined, must be continuous at ρ = a. In order to be compatible with eqns. (6.67)-(6.68)
and (6.71)-(6.72) for the tangential fields Eφ and Hφ when we enforce continuity at ρ = a, it will be
necessary to choose a solution for Hz in the form
 
 
 sin lφ 
 
Hz (ρ, φ) = C2 Gl (ρ) (6.78)
 
 − cos lφ 
 

where C2 is an arbitrary constant and sin lφ and − cos lφ correspond to the “c” and “s” polarizations,
respectively. The function Gl (ρ) must yet be determined, but will be continuous at ρ = a.
Upon substituting (6.77) and (6.78) into (6.74) and (6.75), we find that Fl and Gl must satisfy the
equations      
′′ ′
     
 Fl 
   Fl 
   Fl 
 
2 2 2 2
ρ +ρ + (h1 ρ − l ) =0 (ρ < a) (6.79)
 G′′l   G′l 
     
     Gl 
 

184
or      
 Fl′′  Fl′
     
 
  
  Fl
 

ρ2 +ρ − (p22 ρ2 + l2 ) =0 (ρ > a) (6.80)
 G′′l   G′l 
     
     Gl 
 

As always, we use primes to denote differentiation with respect to the argument (ρ). For ρ < a, we
recognize that eqn. (6.79) is identical with (5.66), which we have already encountered in the study of
metallic circular waveguides, except that kc → h and ν → l. Since Ez and Hz must be finite at ρ = 0,
we can put

Fl (ρ) = Jl (h1 ρ)
Gl (ρ) = Jl (h1 ρ) (6.81)

in the core region, there already being arbitrary amplitude constants present in (6.77) and (6.78).
Equation (6.80), however, is slightly different; if we make the change of variable w = p2 ρ, we have
     
 d2 Fl   dFl   F 
dw 2 dw l
w2 +w − (w2 + l2 ) =0 (ρ > a) (6.82)
 d2 G2l   dGl   Gl 
dw dw

From Appendix B, eqn. (B.23) ff., we see that the general solution to (6.82) is a linear combination of
the modified Bessel functions Il (w) and Kl (w). But as ρ → ∞, we require that the fields decay as we
move away from the core (for a surface wave). This is contrary to the behavior of Il (w) summarized in
Appendix B, and so we have

Fl (ρ) = C3 Kl (p2 ρ)
Gl (ρ) = C4 Kl (p2 ρ) (ρ > a) (6.83)

where the constants C3 and C4 can be obtained by requiring that the functions Fl (ρ) and Gl (ρ) be
continuous at ρ = a as noted above. Comparing (6.81) with (6.83), we see that

Jl (h1 a)
C3 = C4 = (6.84)
Kl (p2 a)

and so Fl (ρ) and Gl (ρ) are actually the same function:

Fl (ρ) = Gl (ρ) = Jl (h1 ρ) (ρ < a)

Jl (h1 a)
= Kl (p2 ρ) (ρ > a) (6.85)
Kl (p2 a)
There now remain two amplitude constants C1 and C2 to determine in addition to the eigenvalue β,
and two remaining boundary conditions—that Eφ and Hφ be continuous at ρ = a. Substituting (6.77)
and (6.78) into (6.67)-(6.68) and (6.71)-(6.72), we obtain for these two boundary conditions that
   
1 jβ ′ − 1 jβ ′ +
C1 lFl (a) + jωµC2 Fl (a ) = − C1 lFl (a) − jωµC 2 Fl (a ) (6.86)
h21 a p22 a
   
1 jβ ′ − 1 jβ ′ +
− C2 lFl (a) − jωǫ1 C1 Fl (a ) = 2 C2 lFl (a) + jωǫ2 C1 Fl (a ) (6.87)
h21 a p2 a
Here, Fl′ (a± ) refers to the derivatives of Fl (ρ) just outside and inside the core boundary, respectively:

Kl′ (p2 a)
Fl′ (a+ ) = p2 Jl (h1 a) (6.88)
Kl (p2 a)

185
 Mode Type HE EH
i1/2 i1/2
∆l 2 ∆l 2
h h
− ∆2l 2 2
− ∆2l + l2 B 2



Eigenvalue equation Θl = − 2 +l B Θl = + 2
q q
C2 ǫ1 lB C1 µ lB
C1 and C2 C1 = µ
 1/2
C2 =− ǫ1
 1/2
∆l 2
 2
∆l ∆l ∆l
2 + 2 +l2 B 2 2 + 2 +l2 B 2
q q
C2 ǫ1 C1 µ
(0 ≤ C1 < µ) (0 ≥ C2 >− ǫ1 )

Table 6.1: Eigenvalue equations and C1 /C2 ratios for HE and EH modes.

Fl′ (a− ) = h1 Jl′ (h1 a) (6.89)

For the following manipulations, some simplification is gained by introducing the notations

Fl′ (a− ) Fl′ (a+ )

Θl = +
h21 Fl (a) p22 Fl (a)
Jl′ (h1 a) Kl′ (p2 a)
= + (6.90)
h1 Jl (h1 a) p2 Kl (p2 a)

n22 − n21 Fl′ (a+ )

∆l =
n21 p22 Fl (a)
n2 − n21 Kl′ (p2 a)
2
= (6.91)
n21 p2 Kl (p2 a)
 
β 1 1
B= + 2 (6.92)
k1 a h21 p2
(note that ∆l > 0 whenever p2 is real and positive and that B > 0 if both p2 and h1 are real). Some
rearrangement of (6.86) and (6.87) gives
r
µ
lBC1 = − Θl C2
ǫ1
ǫ1
r
lBC2 = − [Θl + ∆l ]C1 (6.93)
µ

which are simultaneous equations to be satisfied by C1 and C2 . If they are not both to be zero, we must
have
l2 B 2 = Θl (Θl + ∆l ) (6.94)
which determines the allowed values of β for the circular fiber.
Solving (6.94) for Θl gives two possible equations for β:
" 2 #1/2
∆l ∆l 2 2
Θl = − ± +l B (6.95)
2 2

Each of these corresponds to a particular ratio of C1 /C2 , either positive or negative, whose value is used
to classify the modes of the fiber as HE or EH, according to the entries in Table 6.1.
The HE/EH nomenclature for modes of the circular fiber originated from the first applications of
this structure at microwave frequencies, when a large core/cladding permittivity contrast (ǫ1 ≫ ǫ2 ) was
usually encountered. In this limit, the transverse fields of the HE modes are found to closely resemble
those of the TE (or H) modes of the circular metallic waveguide. Similarly, fields of EH modes resemble

186
those of TM (or E) modes in this limit. As it turns out, the second letter of the hybrid mode designation
refers to the dominant longitudinal field component. But neither of these features is apparent in the case
when the core and cladding refractive indices are close to each other, when only a partial resemblance
to hollow waveguide modes can be distinguished. Some field plots for the various modes of the fiber in
such a situation will be presented in the next subsection, which should prove more useful for determining
what it is that distinguishes them from a physical point of view.

6.7.1 The weakly guiding fiber

For many practical applications in optical waveguides, the inequality
(n21 − n22 )/n22 ≪ 1 (6.96)
will be satisfied. This is known as the “small contrast” or (rather misleadingly) “weakly-guiding”
condition. In this situation we also have p2 /k2 ≪ 1 and h1 /k2 ≪ 1 (since p22 + h21 = k12 − k22 ). For
a TE mode on a slab waveguide, from (6.18) we see that the maximum value of |Hz | will be much
smaller than that of |Hx | in this case, so the weakly-guiding limit produces a mode whose fields are
“almost” TEM. This also holds true for the TM modes of the slab, and for modes of other types of
dielectric waveguide as well. It is worth noting that for slab waveguides, (6.62) reduces to (6.33) and
(6.63) to (6.53) in the weakly guided limit. The TEm and TMm modes of the slab are thus nearly
degenerate in this limit. A similar thing occurs for other types of dielectric waveguide.
For the step-index circular fiber, the analysis of exact mode fields is quite cumbersome, and consid-
erable simplification results in the applicable formulas if this approximation is invoked. We will thus
forego the exact analysis in favor of this simpler treatment.
If (6.96) holds, then we can expect that ∆l ≃ 0 so that the eigenvalue equations in Table 6.1 for the
EH and HE modes can be further simplified. By (B.34) and (B.36) they can be written as
Jl−1 (h1 a) Kl−1 (p2 a)
= (HE modes) (6.97)
h1 aJl (h1 a) p2 aKl (p2 a)
Jl+1 (h1 a) Kl+1 (p2 a)
+ =0 (EH modes) (6.98)
h1 aJl (h1 a) p2 aKl (p2 a)
The l = 0 modes are special cases. The HE0 modes are TM modes (C2 = 0 by Table 6.1), while the
EH0 modes are TE (C1 = 0). Their eigenvalue equations in the weakly guiding limit are given by (6.97)
and (6.98) with l = 0, although for TM modes it is more common to use (B.34)-(B.37) to replace J−1
and K−1 by J1 and K1 :
J1 (h1 a) K1 (p2 a)
+ =0 (TM modes) (6.99)
h1 aJ0 (h1 a) p2 aK0 (p2 a)
The derivation of expressions for the transverse field components in the weakly guiding approximation
is left as an exercise (problems p6-9 and p6-10).
When analyzing either of the characteristic equations (6.97) or (6.98), it is again convenient to deal
in the dimensionless parameters b and neff given by (6.54) and (6.55). The normalized frequency of a
fiber is defined as q q
V = k12 − k22 a = n21 − n22 k0 a (6.100)
In these terms, the eigenvalue equations are
√ √
Jl−1 (V 1 − b) Kl−1 (V b)
√ √ = √ √ (HE or TM modes) (6.101)
V 1 − bJl (V 1 − b) V bKl (V b)
√ √
Jl+1 (V 1 − b) Kl+1 (V b)
√ √ =− √ √ (EH or TE modes) (6.102)
V 1 − bJl (V 1 − b) V bK(V b)

187
Like the eigenvalue equations for the slab, eqns. (6.101) and (6.102) do not yield explicit solutions for
β (or b) in terms of V . In fact, we cannot even find V as a function of b as we did in eqns. (6.52) and
(6.53). Graphical solutions similar to those for the slab can be pursued; we note that the behavior of
the left sides resembles that of tan(h1 a)/h1 a, while the right sides vary essentially only algebraically and
nonperiodically as functions of h1 a. We shall not pursue the graphical solutions here (see the references
at the end of the chapter), but will instead analyze the behavior of the solutions to (6.101) and (6.102)
in the ranges both near and far from cutoff.
Far from cutoff, we have V ≫ 1; in this case b → 1. From the large argument behavior of the
modified Bessel functions (eqn. (B.33)), we have
√
Kl±1 (V b) 1 1
√ √ ∼ √ ∼ →0 (6.103)
V bKl (V b) V b V
√ √
as V b → ∞. Hence by (6.101) and (6.102), the quantity V 1 − b approaches a constant in this limit;
either a root jl−1,m of the Bessel function
√ Jl−1 for the HElm mode, or a root jl+1,m of Jl+1 for the
2
EHlm mode.
√ Thus, the quantity
√ [V 1 − b − jl±1,m ] is small far from cutoff, and the Bessel functions
Jl±1 (V 1 − b) and Jl (V 1 − b) can be expanded as Taylor series:
√ √ ′
Jl±1 (V 1 − b) ≃ Jl±1 (jl±1,m ) + (V 1 − b − jl±1,m )Jl±1 (jl±1,m )
√
= ±(V 1 − b − jl±1,m )Jl (jl±1,m ) (6.104)
and √
Jl (V 1 − b) ≃ Jl (jl−1,m ) (6.105)
The second line of (6.104) follows from eqns. (B.34) and (B.35) and the fact that Jl±1 (jl±1,m ) = 0.
Hence, √ √
Jl±1 (V 1 − b) V 1 − b − jl±1,m
√ √ ≃± √ (6.106)
V 1 − bJl (V 1 − b) V 1−b
√
Setting this equal to 1/V by (6.103), and solving for V 1 − b, we get
√ jl±1,m
V 1−b≃ (far from cutoff) (6.107)
1 + 1/V
Equation (6.107) should be compared with the corresponding form (6.59) for the TE modes of the slab.
These two situations have in common that h1 —their transverse wavenumber—approaches the transverse
wavenumber kc for a TM mode of the corresponding hollow metallic structure. Also, the distributions
of the transverse fields Ex , Ey , Hx and Hy of the fiber modes (in the core region) approach the Ez field
distributions of the TM modes of the metallic guide (cf. problems p6-9 and p6-10). Again we see that
a surface wave mode far from cutoff tends to have nearly all of its fields concentrated within the core,
almost as if an actual metal wall were present just outside the core-cladding boundary.
It is rather more difficult to obtain closed-form approximations for b in the range near cutoff. We will
content ourselves here with a determination of the cutoff frequencies Vc , and obtain an approximation
for b only for the fundamental mode. Since b → 0 at cutoff, we have that
√
Jl±1 V 1 − b) Jl±1 (V )
√ √ → (6.108)
V 1 − bJl (V 1 − b) V Jl (V )
as b → 0. From (B.26) and (B.27) of Appendix B, we have that
√
Kl+1 (V b) 1
√ √ → − √ (l = 0)
V bKl (V b) V b[ln( 2 b ) + γE ]
2 V

2l
→ (l ≥ 1) (6.109)
V 2b
2 The roots j−1,m are the same as j1 , m for the case of HE0m (TMm ) modes.

188
while
√
Kl−1 (V b) 1
√ √ → − √ (l = 0)
V bKl (V b) V 2 b[ln( V b
) + γE ]
2
√
b V
→ −[ln( ) + γE ] (l = 1) (6.110)
2
1
→ (l ≥ 2)
2(l − 1)

as b → 0, where γE = 0.5772... is Euler’s constant. By applying (6.108) and (6.109) or (6.110) to (6.101)
or (6.102), we find that for TEm or TMm modes (l = 0), cutoff (b = 0) occurs at V = Vc , where

Vc = j0m (TEm or TMm modes) (6.111)

For EHlm modes (l ≥ 1), we have

Vc = jlm (EHlm modes, l ≥ 1) (6.112)

and for HElm modes (l ≥ 2), we find

Vc = jl−2,m (HElm modes, l ≥ 2) (6.113)

where (B.34) and (B.35) were used to help obtain (6.113), and jlm denotes the mth root of the Bessel
function Jl , not including zero. Only the HE1m class of modes allows a zero cutoff frequency; from
(6.108) and (6.110) we have (as b → 0):

4 −2
h i
J (V )
γE + V J0 (V )
b≃ e 1 (6.114)
V2
Hence we have

Vc = 0 (HE11 mode)
= j1,m−1 (HE1m modes) (6.115)

Thus, the HE11 mode can (in principle) propagate at all frequencies, and is the fundamental mode of
the fiber. The locations of the various cutoff frequencies can be seen in Figure 6.19, where b vs. V is
plotted in the weakly guiding limit for the lowest-order modes. Note the very flat part of the HE11 curve
for about 0 ≤ V < 0.7. Even though in principle this is a guided mode (surface wave) the value of b
in this range is so small (cf. eqn. (6.114)) that the fields extend for a very large distance away from
the core without much decay. This is characteristic of all three-dimensional optical waveguides with
homogeneous cladding, and differs markedly from the behavior of the slab shown in Figure 6.15. This
means that if the fiber is to be operated in the “single-mode” regime (actually a two-mode range if both
polarizations of the HE11 mode are accounted for), less confinement is possible (b ≃ 0.54) than in the
case of the slab (b ≃ 0.7).
Figure 6.19 also illustrates the fact that in this approximation the propagation coefficients for EHlm
are identical to those for HEl+2,m modes. This can be shown directly from (6.101) and (6.102), which
reduce to the same equation in this case. Thus, in the weakly-guiding limit, the HEl+2,m and EHlm
modes are approximately degenerate. Despite the approximate nature of the degeneracy, it may be
quite a long propagation distance before phase differences (due to the slight differences in β) show up,
so that in practice the effect can be similar to what would be seen if exact degeneracy were present. It
is common practice to combine these nearly degenerate modes in such a way that the resulting fields
are linearly polarized, that is, of all the transverse field components only (Ex , Hy ) or (Ey , Hx ) remain.
These combinations are commonly (though inaccurately) referred to as LP modes, with associated indices

189
 Figure 6.19: b vs. V for weakly-guided fiber modes.

Figure 6.20: Exact ( ); near cutoff ( . ) and far from cutoff (— — —)

calculations of b vs. V for the HE11 mode of the circular fiber.

190
 Figure 6.21: E-field for the HE11 mode.

l + 1, m. The LP0m and LP1m combinations arise as special cases; LP0m are combinations of the two
polarizations of HE1m modes, while the LP1m are combinations of HE2m , TEm and TMm modes.
Figure 6.20 shows a detailed comparison of the exact dispersion curve (b vs. V ) for the HE11 mode
with the approximate values predicted by (6.107) for large V and by (6.114) for small V . Though
less accurate than our corresponding approximations for the slab, these approximations are more than
adequate for many design purposes.
Obtaining transverse field plots for the fiber modes is made quite easy in the weakly-guiding limit,
since according to the results of problems p6-9 and p6-10, all transverse field components have the same
dependence on the radial coordinate. From eqn (J.2) of Appendix J, and the field expressions given in
those problems, we have the field-line equation
dρ
= − cot lφ (“c” polarization)
ρdφ
for HElm modes, while for EHlm modes,
dρ
= + cot lφ (“c” polarization)
ρdφ
These are easily solved and give

ρ = C[sin lφ]−1/l (HE modes)

or
ρ = C[sin lφ]1/l (EH modes)
By taking different values of the constant C, we generate a family of curves corresponding to the mode
fields. According to the convention described in Appendix J, the density of field lines in the core can be
obtained from the field amplitudes. According to problems p6-9 and p6-10, these are

|ET | ∝ Al (ρ) ∝ Jl−1 (h1 ρ) (HE modes)

|ET | ∝ Bl (ρ) ∝ Jl+1 (h1 ρ) (EH modes) (6.116)

For weakly-guided modes, the transverse magnetic field lines are identical to the ET -lines, except that
they are rotated by an angle of π/2l about the origin.
Figure 6.21 displays the E-field lines for the HE11 mode. For this case, the field lines obey

ρ sin φ = const

that is, y = const. This is therefore a linearly-polarized field, and resembles a plane wave. Sketches
of HE21 , HE31 , EH11 and EH21 mode patterns are given in Figure 6.22. It can be seen that the main

191
Figure 6.22: E-lines for some higher-order fiber modes: (a) HE21 ; (b) HE31 ; (c) EH11 ; (d) EH21 .

Figure 6.23: E ( ) and H (— — —) lines for (a) TE1 mode and (b) TM1 mode.

192
 Figure 6.24: Mode intensity patterns for (a) HE11 and (b) TE1 , TM1 and HE21 modes.

difference between HE and EH modes is that the field lines of HE modes tend outward and away from
the center of the core, while for EH modes, the lines loop around and tend back towards the center.
Figure 6.23 shows the fields for the TE1 and TM1 modes. The sense in which a TE mode is a special case
of EH modes, and a TM mode a special case of HE modes can be seen by comparing these with those
of Figure 4.21, and the special position occupied by the HE11 mode is quite evident from Figure 6.21.
In many (though by no means all) applications, the vector character of an optical field is irrelevant,
and only the intensity distribution is of interest (i.e., |ET |2 ). In the weak-guidance limit, these are
circularly symmetric, being given by (6.116) for all modes (including TE and TM). For higher-order
modes, the pattern is generally one or more rings, while for the HE11 mode it is simply a single circular
spot centered in the core (Figure 6.24). The intensity patterns of the LP mode combinations are generally
more complex, because of interference occurring between the fields of different modes.

6.8 Step-Index Fibers of Other Cross-Sections

Optical fibers come in many shapes and varieties besides the circular step-index fiber. None but the
circular geometry can be solved exactly in closed form, but computer numerical analyses have been done
for many other types of fiber.
When the fiber is weakly guiding and a given mode on it not too close to cutoff, we can use the ideas of
penetration depth and shift of the reflection plane discussed in section 6.2 to obtain good approximate
formulas for the propagation coefficient. We have already seen in eqn. (6.60) how the propagation
coefficient of a slab waveguide mode can be related to that of a parallel-plate metallic guide of a larger,
“effective” width. A similar result can be obtained from eqn. (6.107) using the propagation coefficients
of the TM modes of a hollow circular metallic waveguide of slightly larger radius. These expressions
have the form
β 2 = k12 − k̃c2 (6.117)
where k̃c is the cutoff wavenumber of a TM mode of a metallic circular guide of the same shape but
−1/2
larger radius, a + k0−1 (n21 − n22 ) :

jl±1,m
k̃c = −1/2
(6.118)
a+ k0−1 (n21 − n22 )

The increase in the guide radius is simply the near-grazing limit (6.12) of the apparent shift dTE of the
reflecting surface. Now the limit of grazing incidence θ → π/2 corresponds to β = k0 n1 sin θ → k1 , which
means that the mode is far from cutoff, as indeed was the case for eqn. (6.107).

193
 Figure 6.25: Effective cross-section for a step-index dielectric waveguide of arbitrary core shape.

This principle can be extended to step-index fibers of any core shape, provided that we know the
cutoff wavenumbers kc for the TM modes of a correspondingly shaped hollow metallic guide. We will
again obtain the result (6.117), but now k̃c will denote the cutoff wavenumber for a TM mode of a
hollow waveguide of a certain effective cross-section as shown in Figure 6.25. The effective cross-section
is obtained by outwardly extending the core-cladding boundary at each point in the normal direction
−1/2
by a distance equal to the grazing-angle shift k0−1 (n21 − n22 ) . This approximate method greatly
extends the number of fibers that we can analyze, including rectangular, triangular and more. Other
approximate techniques for analyzing the properties of optical waveguides do exist. Several which are
based on concepts of ray theory and transverse resonance will be described in the next chapter.

6.9 The Mode Spectrum of Lossless Open Waveguides

6.9.1 The radiation modes
All of the waveguides we have treated in this chapter have only a finite number of discrete modes at
any given frequency of operation. This is true of any “open” waveguide, including those such as open
microstrip, slotline, two-wire line, coplanar lines, etc. that will be treated in chapter 8. In all these
cases, there is no such thing as higher-order modes which are simply “cutoff”. Instead, open waveguides
possess a set of unguided “radiation modes”.
It can be instructive to think of these modes as limits of modes of a closed waveguide whose metallic
boundary wall recedes to infinity in the limit. Consider a dielectric slab waveguide enclosed by metallic
parallel plates as shown in Fig. 6.26. Although we will not do so in detail here, it is possible to set up
expressions for the fields in the various regions of this structure, and by enforcing boundary conditions
obtain a transcendental equation for the allowed values of propagation coefficient γ. It turns out that
for any lossless closed waveguide, these allowed values are either pure real or pure imaginary for modes
which carry nonzero complex oscillatory power, and lie at positions in the complex γ-plane as shown in
Fig. 6.27(a). As the distances a1 and a2 of the metallic plates from the slab increase, the propagation
coefficients of most of the modes of this waveguide become more and more closely spaced, much the
same as do the modes of a parallel-plate metallic waveguide with homogeneous filling (Fig. 6.27(b). In
the limit of a1 → ∞ and a2 → ∞, these modes form a continuous spectrum of radiation modes that
have a qualitatively different nature than do the discrete, proper modes we have studied so far in this
chapter.
Radiation modes describe fields which are not trapped by the guiding structure, but carry power or
store energy in regions not near to the guiding structure. An example of such a field would be a plane
wave in a slab waveguide whose incidence angle in the slab region was not greater than the critical angle,
and hence produced transmitted plane waves in the cladding traveling away from the core (Figure 6.28).
Unlike trapped (surface wave) modes, there is no restriction on the angle θ made by these rays with

194
 Figure 6.26: Dielectric slab waveguide bounded by metallic planes.

Figure 6.27: Location of propagation coefficients for modes of a shielded dielectric slab waveguide: (a)
a1 and a2 moderate; (b) a1 and a2 large.

195
 Figure 6.28: Untrapped (radiation) mode of a dielectric slab.

respect to the normal to the walls. Thus we have the continuous distribution of allowed values of γ
as noted above, rather than the isolated values associated with discrete modes. For the dielectric slab,
these values are:

γ = jβ ; 0 ≤ β ≤ k2
γ = α ; 0≤α<∞

for “forward-traveling” continuous spectrum modes.

A single radiation mode (corresponding to a single value of γ of the continuous spectrum) cannot
exist by itself if sources of finite energy content are used for excitation (a plane wave is uniform out
to infinity in directions transverse to that of its propagation, and carries an infinite amount of energy).
Only a superposition of such radiation modes over all possible values of γ can represent the field of such
a realistic source.
Although we will not be concerned with specific applications of radiation modes in this text, the
reader should be aware of their existence, and will be reminded of them from time to time in subsequent
chapters. More detailed treatments can be found in the references at the end of the chapter.

6.9.2 Location of mode propagation coefficients

There are some principles that we can lay out that describe where the propagation coefficients of modes
must be in the complex plane for a fairly general class of waveguides. For example, a qualitative analysis
of the permitted range of β for bound (surface) modes of open waveguides of general type can be carried
out. From equation (K.66) of Appendix K, we know that
p
β ≤ ω (µǫ)max (6.119)

for any mode on a lossless waveguide in which µ > 0 and ǫ > 0 everywhere, where (µǫ)max represents
the largest value of the product µǫ over the whole guide cross-section. Furthermore, equality can be
achieved in (6.119) only if the fields are identically zero except where µǫ = (µǫ)max , which will not occur
in an inhomogeneous guide (such as a dielectric slab, for example).
On the other hand, we obtain lower bounds to β by demanding that β be greater than the wavenum-
bers of plane waves or other wave species that can propagate out to infinity in the transverse direction.
Only this guarantees exponentially decaying fields as we approach infinity in any transverse direction. For
example, any waveguide of finite cross-section embedded in an infinite homogeneous region of wavenum-
√
ber k2 = ω µ2 ǫ2 (e. g., the dielectric slab and fiber waveguides) must have β > k2 in order that its
exterior far-field (the so-called “space wave”) field decay in the transverse direction. Another example
is the channel guide embedded in a substrate (regarded as extending to infinity) as shown in Fig. 6.2(b).
For this case we require β < k1 , but β > k2 and β > k3 for a bound mode and hence ǫ1 must be the
largest of ǫ1 , ǫ2 , and ǫ3 for this to happen.

196
 When the surrounding medium of an open waveguide contains dielectric layers that extend to infinity,
then not only plane waves but possibly also surface wave species may be capable of propagation away
from the guiding region in the transverse direction. An example of this is given by the dielectric rib
waveguide of Fig. 6.2(c). For this guide, β must not only be greater than the plane wavenumber k2 of
the surrounding medium, but also greater than any of the surface wave propagation coefficients βs along
the slab itself.
Note that in all cases the upper bound on β is found from local properties of the permittivity
distribution (i. e., where is the maximum value of µǫ attained?), whereas the lower bound is due to the
global properties of the guide (i. e., what wave species can propagate away from the local guiding region
out to infinity?).

197
6.10 Notes and References
Properties of metals at optical frequencies are discussed in

Born and Wolf (1975), chapter 13.

The electro-optic effect and its applications are treated in

Haus (1984), chapters 12 and 13;

R. C. Alferness, “Waveguide electro-optic modulators,” IEEE Trans. Micr. Theory Tech.,
vol. 30, pp. 1121-1137 (1982).

A nice overview of fabrication techniques for optical waveguides is given by

Lee (1986), chapter 7.

Good introductions to the theory of optical dielectric waveguides are:

Kapany and Burke (1972);

Marcuse (1974);
Arnaud (1976);
Adams (1981);
Marcuse (1982), chapters 8-12;
Mickelson (1993).

The derivation of the surface wave modes of a dielectric slab is to be found (in various versions) in any
of these books. Figure 6.15 is adapted from

Arnaud (1976), chapter 5.

Approximations similar to (6.57) and (6.59), along with further studies along these lines, were ob-
tained by

J.F. Lotspeich, “Explicit general eigenvalue solutions for dielectric slab waveguides”, Applied
Optics, vol. 14, pp. 327-335 (1975).
M. Miyagi and S. Nishida, “An approximate formula for describing dispersion properties of
optical dielectric slab and fiber waveguides”, J. Opt. Soc. Amer., vol. 69, pp. 291-293
(1979).

A nice collection of reprints on the general subject of slab waveguides is:

Marcuse (1973).

The exact analysis of the modes of a circular step-index fiber can be found in

Marcuse (1982), chapter 8;

Arnaud (1976), chapter 5;
Kapany and Burke (1972);
Unger (1977), chapter 4;
Adams (1981), chapter 7.

Elements of all of these have been incorporated into the analysis of weakly-guiding fiber modes presented
here. The interested reader should also consult the collection of reprints

Gloge (1976).

This contains two important articles:

198
 D. Gloge, ‘Weakly guiding fibers,” Appl. Optics, vol. 10, pp. 2252-2258 (1971).
from which Fig. 6.19 is taken and

A.W. Snyder, “Asymptotic expressions for eigenfunctions and eigenvalues of a dielectric or

optical waveguide”, IEEE Trans. Micr. Theory Tech., vol. 17, pp. 1130-1138 (1969).
A classic paper not included in Gloge’s collection is

E. Snitzer, “Cylindrical dielectric waveguide modes”, J. Opt. Soc. Amer., vol. 51, pp.
491-498 (1961).

Snitzer tells how to construct the field plots illustrated in Figs. 6.21-6.23. His method was used to
construct plots of the first three HE and EH mode groups in
Vzyatyshev (1970).

Other useful approximate formulas for propagation coefficients of fiber modes are given in Gloge (1971),
loc. cit. above, but these are of limited accuracy. More accurate (though more complicated) approxi-
mations are given in Miyagi and Nishida (1979), loc. cit. above.
Fibers of arbitrary cross-section have been treated in the weakly-guiding limit in

D.L.A. Tjaden, “First-order correction to ’weak-guidance’ approximation in fibre optics the-

ory”, Phillips J. Res., vol. 33, pp. 103-112 (1978).
A.W. Snyder and W.R. Young, “Modes of optical waveguides”, J. Opt. Soc. Amer., vol. 68,
pp. 297-309 (1978).
Numerical solutions for various shapes of the fiber core are given in

C. Yeh, K. Ha, S.B. Dong and W.P. Brown, “Single-mode optical waveguides”, Applied
Optics, vol. 18, pp. 1490-1504 (1979).

The effective cross-section idea comes from

E.F. Kuester, “Propagation coefficients for linearly polarized modes of arbitrarily-shaped

optical fibers, or dielectric waveguides”, Optics Letters, vol. 8, pp. 192-194 (1983).

and is extended to fibers which are not weakly guiding in

E.F. Kuester, “The effective cross-section method for dielectric waveguides in or on a sub-
strate”, Radio Science, vol. 19, pp. 1239-1244 (1984).

Discussions of the radiation mode spectrum of optical and other open waveguides can be found in:
Shevchenko (1971);
Marcuse (1974);
Marcuse (1982), chapters 8-10;
Vassallo (1991).

6.11 Problems
p6-1 Using the ray method of section 6.2, and the result of problem p1-6, determine the core
field Hy and the eigenvalue equation to be solved for β for the TM modes of the dielectric
slab waveguide shown in Figure 6.4. Note: It can be somewhat more convenient to use the
current (i. e., magnetic-field) reflection coefficient rather than the Fresnel coefficient for this
problem, since the only component of the magnetic field is Hy .

199
p6-2 Using the method of section 6.3, obtain expressions for the field Hy (both in the core and
in the cladding) of the TM modes of the dielectric slab waveguide shown in Figure 6.4,
along with the eigenvalue equation for determining the allowed values of β (Note: Be careful
to apply the correct boundary conditions). Illustrate graphically the difference between the
TE-mode and TM-mode eigenvalue equations.
p6-3 Show that the complex power
1 ∞
1
Z Z
P̂ = E × H∗ · uz dxdy
2 y=0 x=−∞

carried through a unit width (in y) of the total cross-section of a dielectric slab waveguide
for a TE mode can be expressed in terms of the effective width aeff = a + 2/p2 as follows:

βaeff ω(ǫ1 − ǫ2 )
P̂ = |D|2
4 h21

where D is the amplitude factor from equation (6.32). Analogously, show that the complex
oscillatory power is given by

1 1
Z ∞
βaeff 2 ω(ǫ1 − ǫ2 )
Z
P̂osc = E × H · uz dxdy = D
2 y=0 x=−∞ 4 h21

[Hint : How can the trig functions sin(ha) and cos(ha) be eliminated from your expression?].
p6-4 Show that the various forms (6.33), (6.34), (6.35), (6.36), (6.37) and (6.43) for the eigenvalue
equation of the TE modes of the dielectric slab are equivalent to each other.
p6-5 Consider the asymmetric slab waveguide pictured below.
(a) Give an expression for Ey in both cladding regions as well as the core region, and derive
the eigenvalue equation for the TE modes of this guide using the method of section 6.3.
(b) Show that if n2 6= n3 , then all TE modes (even the TE0 ) have a non-zero cutoff frequency,
and give an expression for this frequency.

Note: In the interest of standardized notation, put

p22 = β 2 − k02 n22

p23 = β 2 − k02 n23

V22 = (n21 − n22 )k02 a2
V32 = (n21 − n23 )k02 a2
Also, since no loss of generality results by doing so, assume that n3 > n2 .

200
p6-6 Repeat problem p6-5 for the TM modes of the asymmetric slab waveguide.

p6-7 Show for the TMm modes of the dielectric slab [which obey (6.63)] that
 2 
2 n2 V − mπ
b ≃ sin
n21 2

as b → 0 (V → mπ), and
 2
(m + 1)π
b ≃ 1−
V + 2n22 /n21
as b → 1 (V → ∞).

p6-8 Consider the asymmetric slab waveguide pictured in problem p6-5.

(a) Using the method of section 6.2, give an expression for Ey in the core region, and derive
the eigenvalue equation for the TE modes of this guide.
(b) Show that if n3 > n2 , then all TE modes (even the TE0 ) have a non-zero cutoff frequency,
and give an expression for this frequency.

Use the standardized notation suggested.

p6-9 (a) For the HElm modes of a weakly guiding step-index circular fiber, show that (for l 6= 0),
the transverse field components are given approximately by
 
cos lφ
Eρ = −jC1 Al (ρ)
sin lφ
 
sin lφ
Eφ = jC1 Al (ρ)
− cos lφ
 
jβ sin lφ
Hρ = − C1 Al (ρ)
ωµ − cos lφ
 
jβ cos lφ
Hφ = − C1 Al (ρ)
ωµ sin lφ
q
µ
where C1 ≃ ǫ1 C2 and

β
h1 Jl−1 (h1 ρ) (ρ < a)

Al (ρ) =
β Jl (h1 a)
p2 Kl (p2 a) Kl−1 (p2 ρ) (ρ > a)


(b) For the special case l = 0 of the HE modes (the TM modes) show that the transverse
fields are
Eρ = −jC1 A0 (ρ)
Eφ = 0
Hρ = 0
jβ
Hφ = − C1 A0 (ρ)
ωµ

201
p6-10 (a) For the EHlm modes of a weakly-guiding step-index circular fiber, show that (for l 6= 0),
the transverse field components are given approximately by
 
cos lφ
Eρ = jC1 Bl (ρ)
sin lφ
 
sin lφ
Eφ = jC1 Bl (ρ)
− cos lφ
 
jβ sin lφ
Hρ = − C1 Bl (ρ)
ωµ − cos lφ
 
jβ cos lφ
Hφ = C1 Bl (ρ)
ωµ sin lφ
q
where C1 ≃ − ǫµ1 C2 and

β
h1 Jl+1 (h1 ρ) (ρ < a)

Bl (ρ) =
Jl (h1 a)
 − pβ2 K l (p2 a)
Kl+1 (p2 ρ) (ρ > a)

(b) For the special case l = 0 of the EH modes (the TE modes), show that the transverse
fields are:
Eρ = 0
jωµ
Eφ = C2 B0 (ρ)
β
Hρ = −jC2 B0 (ρ)
Hφ = 0

p6-11 A circular step-index optical fiber has a core of index n1 = 2.21, and a finite cladding of
index n2 = 2.20 which extends only out to the radius c. The fiber is to be operated at a
(free-space) wavelength of λ0 = 10.6µm.

(a) What is the maximum value of a for which the fiber will support only the fundamental
modes?
(b) If a is chosen to be the maximum value found in part (a), what value of c must be
chosen to assure that the fields at ρ = c have decayed to 0.1% of their values at ρ = a?
p
p6-12 Use the effective cross-section method of section 6.8 to plot b vs. V = k0 a n21 − n22 for the
fundamental modes of a weakly-guiding rectangular-core optical fiber of dimensions a × d, if
a = 2d.

202
p6-13 Show that choosing Ez = C1 Fl (ρ) cos lφ and Hz = C2 Gl (ρ) cos lφ for the same mode leads
to boundary conditions for Eφ and Hφ that cannot be satisfied if l 6= 0.
p6-14 Using the results of problems p6-9 and p6-10, show that the transverse mode fields of
appropriate linear combinations of HEl+2,m and EHlm modes are linearly polarized in the
core, and give approximate expressions for them.
p6-15 A dielectric strip waveguide is shown below, with n1 > n3 > 1.

Can this structure support bound waveguide modes if

(a) n1 > n2 ?
(b) n2 > n1 > n3 ?
(c) n3 > n1 ?
Explain your answers in each case.
p6-16 A symmetric slab dielectric waveguide is to be made from materials such that n2 = 3.5
and n1 = 1.03n2 . What is the largest thickness a of the core layer possible without the
existence of higher order (TE1 , etc., or TM1 , etc.) modes, if the operating frequency range
is to cover the range of free-space wavelengths 0.8 µm ≤ λ0 ≤ 1.5 µm? Suppose that one of
the cladding regions is replaced by a half-space of air (n3 = 1). Is it possible now to find a
core layer thickness a such that only the lowest order TE and TM modes exist throughout
this frequency range? If so, provide a suitable value of a.
p6-17 Prove that the time-average power flow through any plane x = constant is zero for a TE
surface-wave mode on a dielectric slab waveguide.
p6-18 Using the result of Problem p6-6, compute and plot the normalized propagation constant

n2eff − n23
b=
n21 − n23

versus the normalized frequency

q
V3 = k0 a n21 − n23

for the values n1 = 1.51, n2 = 1.0 and n3 = 1.50. Cover the range 0 < V3 < 20, and plot
results for all TM modes which exist in that range. Compare these results with ones obtained
for the symmetric slab for the case n1 = 1.51 and n2 = n3 = 1.50.

203
204
Chapter 7

TRANSVERSE RESONANCE IN
GUIDED WAVE STRUCTURES

The upper air burst into life!

And a hundred fire-flags sheen,
To and fro they were hurried about!
And to and fro, and in and out,
The wan stars danced between.
—Samuel Taylor Coleridge,
The Rime of the Ancient Mariner

Only a very limited class of waveguides permits exact solutions for their mode fields and propagation
coefficients. In fact, we have examined most of them in Chapters 5 and 6 (a few others which support
TEM modes will be considered in chapter 8). Other important waveguides (microstrip, for example) do
not yield to any exact solution and must be handled either with numerical methods or by approximation.
While numerical techniques are of substantial importance, they are beyond the scope of this book, in
which we hope to examine more of the fundamental physics involved in guided waves.
In this chapter we will examine several types of waveguides using a very important analysis technique
called the transverse resonance method. The method can be viewed as an extension of the ray methods
treated in Chapter 5. In section 7.1, it is introduced as an exact technique for the analysis of multilayer
slab waveguides, whereas subsequently it will be extended as an approximate method capable of treating
many types of waveguide not otherwise capable of being analyzed by other than numerical means. In
specific applications other names for essentially the same approach (e.g., the effective index method in
optical waveguides) are sometimes used. The technique will be introduced and developed by means of
examples.

7.1 Multilayer Slab Waveguides

7.1.1 TE modes
It is of considerable interest to characterize the behavior of slab waveguides which consist of more
than one central guiding layer. The layering can be employed to achieve a variety of desired physical
effects. For example, thin layers of semiconducting materials with greatly differing bandgaps can be
combined into so-called multiple quantum well (MQW) structures. These structures have properties
which are readily exploited to make devices such as intensity modulators and optical switches. Although
the methods of chapter 6 could, in principle, be used to analyze the waveguiding properties of such
multilayered slabs, the algebra involved quickly becomes much too cumbersome to handle, and is even

205
Figure 7.1: (a) Multilayered slab; (b) equivalent transverse transmission line network for TE modes; (c)
decomposition into left-looking and right-looking sections.

awkward to implement in a computer program. We seek to obtain the waveguide modes by simpler
means, if possible.
Consider then the multilayer slab structure of Fig. 7.1(a). The TE mode fields thus satisfy (6.20)
again, where now ǫ = ǫ(x) = ǫ0 n2 (x). The solution within the ith layer xi−1 < x < xi , where n(x) = ni ,
is given as
Ey = Ãi e−pi x + B̃i e+pi x (7.1)
where q
pi = β 2 − k02 n2i = jhi (7.2)
√
and Ãi = Ai 2Zc and similarly for B̃i . By (6.18), the Hz field in this layer must be
1 ∂Ey pi h i
Hz = − = Ãi e−pi x − B̃i e+pi x (7.3)
jωµ ∂x jωµ
As in section 1.4, we compare (7.1) and (7.3) with expressions (1.7) for the voltage and current on a

206
classical transmission line. It can be seen that the behavior of the TE mode field in the transverse (x)
direction is given by that of an equivalent transmission line described in section 1.2, if the following
replacements are made:

z → x
γ → pi
V → Ey (7.4)
I → Hz
jωµ ωµ
Zc → =
pi hi
In the cladding region x < x0 = 0, we have only the wave B0 ep0 x , while in x > xN −1 = a, we have only
AN e−pN x .
The apparatus of section 1.2 can now be carried over for use in the solution of this multilayered
waveguide. The multilayered slab is replaced by a network of equivalent transmission line sections as
shown in Fig. 7.1(b). We are seeking mode solutions, so no sources are present in the transmission
line network. We wish to know under what conditions non-zero voltages and currents may exist in the
network (i. e., nonzero fields in the slab).
Let us choose any arbitrary point x = xA in this network, and temporarily disconnect the right and
left sides of the network at this point as shown in Fig. 7.1(c). The impedance seen “looking right” into
the right-hand section is
→
+ V (x+A)
Z (xA ) = Z(xA ) = (7.5)
I(x+A)

which is the ordinary line impedance at x = x+

A as defined in (1.47). This value is easily calculated by
→
starting at x = xN −1 with Z (xN −1 ) = ZcN (since ρ(x) = 0 in region N ), and repeatedly transferring
this value backwards to xN −2 , xN −3 and so on, using (1.55):
→
→ Z (xi+1 ) cosh pi+1 di+1 + Zc,i+1 sinh pi+1 di+1
Z (xi ) = Zc,i+1 → (7.6)
Zc,i+1 cosh pi+1 di+1 + Z (xi+1 ) sinh pi+1 di+1

where di+1 = xi+1 − xi . In the last step to xA , if xA does not coincide with one of the layer boundaries
xi , we must replace di+1 by xi+1 − xA in (7.6), as well as xi by xA .
Likewise, let us “look left” into the left-hand section. Here we see

←
− V (x−
A)
Z (xA ) = −Z(xA ) = − (7.7)
I(x−A)

and the minus sign appears because I(x) for a transmission line is always defined as positive when flowing
← →
toward increasing x. Now, Z is calculated in precisely the same way as Z , starting now at x = x0 with
←
Z (x0 ) = Zc0 and working towards xA . Specifically,
←
← Z (xi ) cosh pi+1 di+1 + Zc,i+1 sinh pi+1 di+1
Z (xi+1 ) = Zc,i+1 ← (7.8)
Zc,i+1 cosh pi+1 di+1 + Z (xi ) sinh pi+1 di+1

Reconnecting the lines at xA once again, we require that Kirchhoff’s current law I(x− +
A ) = I(xA ) and
− +
voltage law V (xA ) = V (xA ) be satisfied in order for a nontrivial source-free solution to be possible. By
(7.5) and (7.7) we see that this means that the transverse resonance condition
← →
Z (xA )+ Z (xA ) = 0 (7.9)

207
must be satisfied.
Only for certain values of β (the eigenvalues) can this occur, and so we have yet a third approach
to obtaining the characteristic equation of, and physical insight into, such dielectric slab waveguides.
Note that (7.9) does not describe an impedance match condition, since in fact both forward and reverse
waves exist in all regions (line sections) except 0 and N , and the reflection coefficient ρ(x) is thus not
zero. Rather, (7.9) is simply a concise statement of Kirchhoff’s laws for compatibility of voltage and
current at the connection point xA when the two halves of the transmission line network are connected
together. This condition is similar to the condition of oscillation for a one-port oscillator. Only when
this compatibility occurs is the network capable of supporting nonzero source-free voltages and currents.
It is an easy example to demonstrate that this algorithm gives the correct eigenvalue equation for
←
the TE modes of the step-index slab. Choosing xA = 0 in the slab of Fig. 6.4, we have Z (0) = jωµ0 /p2 ,
while
→ jωµ0 h1 cos h1 a + p2 sin h1 a
Z (0) =
h1 p2 cos h1 a − h1 sin h1 a
so that
h1 cos h1 a + p2 sin h1 a h1
+ =0
p2 cos h1 a − h1 sin h1 a p2
which is equivalent to (6.33). We could also have chosen the matching point to be at xA = a/2, in
which case the impedances looking right and looking left are identical by virtue of the symmetry of the
structure:
→ a ← a jωµ0 h1 cos h21 a + p2 sin h21 a
Z ( ) =Z ( ) =
2 2 h1 p2 cos h21 a − h1 sin h21 a
The temptation is to conclude that imposition of the transverse resonance condition implies that these
two impedances must equal zero. This is indeed one possibility, but not the only one. We have chosen
a matching point at a plane of symmetry of the structure, and general symmetry considerations tell us
that all fields of a mode of this structure must either be even or odd functions about this plane. Because
of (7.3), if Ey is even, then Hz is odd, and vice versa. Setting both left- and right-looking transverse
impedances equal to zero is tantamount to choosing the modes where Ey vanishes at the symmetry
plane; those modes for which the other symmetry applies will correspond to infinite values of left- and
right-looking impedances. Equivalently, we can view our procedure in terms of admittances, for which
the latter case requires zero values at the symmetry plane. The result of imposing transverse resonance
at xA = a/2 for the dielectric slab is then
h1 a h1 a
h1 cos + p2 sin =0
2 2
or
h1 a h1 a
p2 cos − h1 sin =0
2 2
which are equivalent to (6.43).

7.1.2 TM modes
The TM modes of the multilayered slab waveguide in Fig. 7.1(a) can also be analyzed using the transverse
resonance method. Instead of (7.4), we must make the correspondences
z → x
γ → pi
V → Ez (7.10)
I → −Hy
pi hi
Zc → =
jωǫi ωǫi

208
 Figure 7.2: (a) Graded-index profile; (b) Piecewise constant (staircase) index approximation.

Other more complicated multilayered slabs can also be readily analyzed by this method. It is espe-
cially suitable for computer implementation, which is left to the reader in certain of the problems at the
end of the chapter.

7.2 Graded-Index Profiles

Many methods of fabrication for optical waveguides do not result in simple abruptly changing refractive
index profiles such as we have studied in chapter 6 or in the previous section. Instead of such step-index
distributions as in Figure 7.1(a), we more often have graded-index profiles as shown in Figure 7.2(a).
The rays corresponding to the field in a step-index guide abruptly change direction upon reflection by
the boundary, while we might imagine that those of the graded-index guide do so continuously at each
point within the guide. We can visualize the process approximately by imagining the continuous index
profile of Figure 7.2(a) to be approximated by the piecewise constant one of Figure 7.2(b). A ray will
be refracted a little bit at each jump in index until total reflection occurs, reversing the direction of
transverse ray travel.
In most cases, the field equations cannot be solved exactly for continuously graded index profiles, and
the piecewise constant index approximation of Figure 7.2(b), though it can be handled by the transverse
resonance technique, leads to equations which are unwieldy to solve. If the change in refractive index
with x is gradual enough, it is possible to use the WKB approximation of section 1.5 to obtain an
approximate transverse-resonance solution to the problem.
From the WKB approximation to the solution of the equivalent nonuniform transverse transmission
line, the right-looking impedance at x = xi of an inhomogeneous section of transmission line between
x = xi and xi+1 is given by
→ R xi+1 R xi+1
→ Z (xi+1 ) cosh
xi
p(x) dx + Zc (x−
i+1 ) sinh xi
p(x) dx
Z (xi ) = Zc (x+
i ) R xi+1 → R xi+1 (7.11)
−
Zc (xi+1 ) cosh xi p(x) dx+ Z (xi+1 ) sinh xi p(x) dx

This can be used in place of (7.6) when the layer is inhomogeneous. When the rays in a layer approach
the critical angle for one of its interfaces, the ordinary WKB approximation breaks down. This will
occur whenever p(x) nears zero—these values of x are turning points. The analysis must be modified
near turning points as indicated in section 1.5, and the result (for TE waves) is that a turning point
appears to totally reflect a propagating (p(x) is pure imaginary) voltage wave with a reflection coefficient
of ejπ/2 . The voltage and current decay exponentially beyond the turning point, and if a second turning
point is not located too close to the first, we may ignore any penetration of the wave beyond the first
turning point, and regard it as if it were isolated. In this case, a turning point located at xt (so that

209
 εr(x)

n12

n22

n02
0
2 1 0 1 2 3 4 5
x/a

Figure 7.3: Diffused slab waveguide permittivity profile.

p(xt ) = 0) in an inhomogeneous section of line with p(x) imaginary for x < xt produces the right-looking
impedance at xi < xt of:
Rx Rx
→
+
j cosh xit p(x) dx + sinh xit p(x) dx
Z (xi ) = Zc (xi ) Rx Rx (TE polarization) (7.12)
cosh xit p(x) dx + j sinh xit p(x) dx

This, too, can be used in place of (7.6) in the transverse resonance analysis of continuously graded slab
waveguides. For TM waves, the relation corresponding to (7.12) is
Rx Rx
→
+
−j cosh xit p(x) dx + sinh xit p(x) dx
Z (xi ) = Zc (xi ) Rx Rx (TM polarization) (7.13)
cosh xit p(x) dx − j sinh xit p(x) dx

7.2.1 Example: A diffused slab waveguide

Consider a slab waveguide with an exponential permittivity profile
ǫr (x) = n20 (x < 0)
= n2 + (n21 − n22 )e−x/a
2
(x > 0)
as shown in Fig. 7.3, where a now represents a characteristic depth of the variation of ǫ into the substrate
region x > 0, while n1 is the maximum and n2 the minimum value of refractive index in the substrate.
This type of profile can result from in-diffusion of metals into previously homogeneous substrates. The
WKB method will be used here to find the eigenvalue equation for the TE surface wave modes of this
waveguide.
For the substrate region, we can write
q q p
h = h(x) = k02 ǫr (x) − β 2 = k0 n21 − n22 e−x/a − b (7.14)

210
 1

0.8

0.6

b
0.4

Diffused slab WG

0.2 Step-index slab WG

0
0 5 10 15 20
V1

Figure 7.4: Dispersion curve for TE0 mode of diffused slab waveguide with n0 = 1, n2 = 4.0 and
n1 = 4.1, with TE0 mode of step-index slab waveguide of thickness a for comparison.

where again
n2eff − n22
b=
n21 − n22
and
neff = β/k0
The turning point xt is located at xt = a ln(1/b). The phase integral is evaluated by elementary functions:
Z xt Z xt
√ √ √
p dx = j h dx = j2V1 [ 1 − b − b cos−1 b] (7.15)
0 0

where q
V1 = k0 a n21 − n22
and so using (7.12) we can set up the transverse resonance condition for this waveguide:

√ √ √ i
p p
h n2 − n2eff + n2eff − n20
tan 2V1 ( 1 − b − b cos−1 b) = p 12 p (7.16)
n1 − n2eff − n2eff − n20

This equation predicts neff with high accuracy. A typical dispersion curve for this case is shown in
Fig. 7.4. Comparing with results for a step-index slab waveguide with the same cladding materials but
having a constant index n1 over the region 0 < x < a before dropping to the constant value n2 for
x > a, we see that both modes have a nonzero cutoff frequency, but the step-index guide has b rising

211
 Figure 7.5: Hollow metallic guides with stepped-profile cross-sections.

much more rapidly to its high-frequency limit. This is because the diffused guide’s spatial region of field
concentration (where neff > n(x)) becomes smaller and smaller as frequency increases, while that of the
step-index guide remains fixed at 0 < x < a.

7.3 Loaded Rectangular Waveguides

The principles of transverse resonance can be extended to other kinds of complicated waveguides if
we consider reflections and transmissions of more general types of constituent waves than simple plane
waves. If, for example, we wish to examine stepped cross-section metallic waveguides such as those
shown in Fig. 7.5, we must begin by looking at the species of wave that can exist in a uniform segment
of the structure. Since such a uniform segment is a parallel plate waveguide, we know from chapter 1
that a TEM wave as well as TE and TM higher order waves can be present. Suppose that a parallel
plate region has a height d as shown in Fig. 7.6(a). Now all these waves could be present, propagating
(or attenuating) in any direction in the xz-plane. If kd < π, the higher order waves are cutoff, and will
quickly decay to negligible levels as we move away from the places at which they originate. If a TEM
wave is incident on a junction between two parallel plate regions as shown in Fig. 7.6(b), we expect
that there will be a reflected as well as transmitted TEM wave on the respective sides of the junction as
indicated. In order to maintain consistency of the phase variation in the z-direction, the reflection and
transmission angles must be the same as the incident angle, if the same medium is present on both sides
of the junction. However, these waves along with the incident wave are not generally enough to satisfy
the boundary conditions for the total field at the junction itself; it follows that higher order cutoff waves
must also be generated in the vicinity of the junction. As will be seen in chapter 11, we often wish only
to consider the field due to the fundamental TEM waves, and this is essentially equal to the total field
if we do not need accurate values of the field near the junction. All we require is that we be able to
calculate reflection and transmission coefficients of the TEM wave from the junction, and this has been
done for many types of junctions as indicated in the references at the end of the chapter.
Therefore, consider a hollow metallic waveguide whose cross-section can be partitioned by one or
more vertical lines into several sections of parallel-plate waveguides of various heights, as shown in
Fig. 7.5. Within each section, we express the field as a pair of TEM (with respect to the ray directions)
parallel-plate waves propagating at some angle with respect to the z-axis (we would, of course need to
supplement these near the junction regions by the cutoff wave fields as discussed above). The factor
exp(−γz) is common to all the fields, and we need deal only with the projected behavior of these waves
in the xy-plane.
Thus, for example, in the adjacent parallel-plate sections shown in Fig. 7.7, we write

Ez = E1 e−jkc x + E2 e+jkc x x < x0 and 0 < y < d1 (7.17)

and
Ez = E3 e−jkc x + E4 e+jkc x x > x0 and 0 < y < d2 (7.18)

212
Figure 7.6: (a) Parallel plate region of height d; (b) Junction between two parallel plate regions.

Figure 7.7: Projected TEM parallel-plate waves in adjacent parts of a cross-section.

213
 Figure 7.8: Reflection and transmission coefficients at a typical junction.

where kc2 = ω 2 µǫ + γ 2 = k 2 + γ 2 . At such a junction between parallel-plate sections, each of the

incident waves (E1 and E4 ) will partially reflect back into the same section and partially transmit into
the adjacent section. We denote the reflection coefficients of the waves incident from left or right as ρA
and ρB , respectively. The transmission coefficient from left to right is called τA , and that from right to
left is called τB . The situation is illustrated schematically in Fig. 7.8.
From all this, there follow two conditions on the waves (7.17) and (7.18) at the junction; namely,

E2 ejkc x0 = ρA (E1 e−jkc x0 ) + τB (E4 ejkc x0 ) (7.19)

and
E3 e−jkc x0 = ρB (E4 ejkc x0 ) + τA (E1 e−jkc x0 ). (7.20)
Note how we must be careful to use the correct values of the incident and reflected waves in (7.19)
and (7.20). In what follows, we will consider the parallel-plate region to be described by an equivalent
transverse transmission line for which (a filling medium of relative permittivity ǫr is assumed):

z → x
√
p 

γ → jkc = j k02 ǫr − β 2 = jk0 ǫr cos θ 


V → Ey TEM waves (7.21)
I → Hz



Zc → ωµ


kc

A simple example of the foregoing is furnished by the example of a rectangular metallic waveguide
of Fig. 5.12. While we already have a full solution for this waveguide by other means, it is instructive to
see it done by transverse resonance. The equivalent transverse transmission line network is a length a
of transmission line short-circuited at both ends. Breaking this circuit at (say) x = 0, and applying the
transverse resonance condition, we obtain tan kc a = 0, or kc a = mπ; m = 1, 2, . . . . These are the TEm0
modes. Note that no other modes of this waveguide are found by this method, because only TEM waves
have been used as building blocks for the fields.
Next, consider the effect of loading the rectangular waveguide with a nonmagnetic dielectric region
of permittivity ǫ occupying 0 < x < h, while the rest of the waveguide is air-filled. The equivalent
transverse network is shown in Figure 7.9. The characteristic impedance and cutoff wavenumber are
different for the two sections of transmission line:
ωµ0 p
Zc1 = ; kc1 = k 2 − β 2
kc1
and
ωµ0
q
Zc2 = ; kc2 = k02 − β 2
kc2

214
 h

Zc1 Zc2
kc1 kc2

a
Figure 7.9: Equivalent transverse transmission line network for dielectric-loaded rectangular waveguide.

where k and k0 are the wavenumbers of the dielectric and air regions respectively. We set up the
transverse resonance equation as

Zc1 tan kc1 h + Zc2 tan[kc2 (a − h)] = 0 (7.22)

or
tan kc1 h tan[kc2 (a − h)]
+ =0 (7.23)
kc1 kc2
As is the case for the dielectric slab waveguide, this is a transcendental equation for the phase coefficient
β, and must in general be solved numerically. This is left as an exercise for the reader.

7.3.1 Ridged, grooved and finned rectangular metallic waveguides

Exact explicit expressions for the reflection and transmission coefficients at a junction involving a di-
aphragm or a change in height do not exist in general. However, under certain conditions (usually that
the cutoff frequency is not too large—kc d ≤ π) approximate equivalent transverse networks that include
lumped circuit elements can be used to generate approximate values for these coefficients. A number of
useful equivalent circuits for TEM waves in parallel-plate discontinuities are shown in Table 7.1. They
refer to an equivalent transverse transmission line described by (7.21). Approximations for other dis-
continuities can be obtained by a number of methods, notably the mode-matching technique described
in section 11.2. In this way, we can reduce the problem to one of transverse resonance on an equivalent
transverse transmission line circuit appropriate to the waveguide under consideration.
We illustrate the use of Table 7.1 on the example of a symmetrically finned rectangular waveguide
shown in Fig. 7.10(a). The equivalent transmission line circuit for the transverse resonance method is
shown in Fig. 7.10(b). The side walls at x = 0 and x = a are special cases of junction 7.1.1 as g → 0,
i. e. short circuits, whereas the gap in the center at x = a/2 is described by a shunt susceptance B as in
the general form of 7.1.1. We set up the transverse resonance condition for this guide at xA = a2 − , so
that the shunt susceptance jB will appear in parallel with the line admittance seen looking to the right
at x = x+ A . Thus, the transverse resonance condition becomes

tan kc2a (2 − BZc tan k2c a )

=0 (7.24)
1 − BZc tan k2c a

215
Junction Parameters/Equivalent
 Circuit
2kc d πg
7.1.1 Symmetrical Window BZc = π ln csc 2d (kc d ≤ π)

g d (Zc , k c ) jB (Zc , k c )

A A
 
2kc d πg
7.1.2 Symmetrical Diaphragm BZc = π ln csc d (kc d ≤ π)

g
d (Zc , k c ) jB (Zc , k c )
g
A A
 
2kc d πg
7.1.3 Nonsymmetrical Window or BZc = π ln csc 2d (kc d ≤ π)
Diaphragm

d/2 (Zc , k c ) jB (Zc , k c )

g/2
A A
h 2 2
   i
1+N 2 1+N 2
7.1.4 Symmetrical Step in Height BZc = kcπd2 (1−N
2N 2
)
ln 1−N 2 + 2 ln 2N ;
p
N = d2 /d1 ≤ 1; (kc d1 ≤ π)

N:1
d1 d2 ( Zc , kc ) jB ( Zc , kc )

A
A
h 2 2
   i
1+N 2 1+N 2
7.1.5 Nonsymmetrical Step in Height BZc = kcπd2 (1−N
2N 2
)
ln 1−N 2 + 2 ln 2N ;
p
N = d2 /d1 ≤ 1; (kc d1 ≤ π)

N:1
d1/2 ( Zc , kc ) ( Zc , kc )
d2/2 jB
A A

Table 7.1: Transverse Equivalent Networks for TEM Waves at Parallel Plate Junctions.

216
 y

g (jkc, Zc) jB (jkc, Zc)

(d-g)/2
a
_ a x
0 x=0 a
_ a
2 2

(a) (b)

Figure 7.10: (a) Symmetrically finned rectangular waveguide; (b) equivalent transverse transmission line
network.

where, by 7.1.1 of Table 7.1,

2kc d  πg 
BZc = ln csc (7.25)
π 2d
One set of solutions to (7.24) has tan kc2a = 0 or kc a = 2mπ, m = 1, 2, 3, . . . . These are the TE2m,0
modes of the guide of width a with no fins at all. The Ey field for such modes is already zero at x = a/2
and is thus unaffected by the insertion of the fins. On the other hand, the TE2m+1,0 modes of the large
guide are substantially perturbed by the presence of the fins and correspond to the other set of solutions
to (7.24) which obey
kc a 2 π
tan = = (7.26)
kc d ln csc πg


2 BZc 2d
or " #
−1 π
kc a = 2mπ + 2 tan (7.27)
kc d ln csc πg


2d

This is a transcendental equation for kc which must be solved numerically or by approximation. If

m 6= 0, a useful approximation is obtained by setting kc = 2mπ/a in the right side of (7.27), giving
" #
−1 a
kc a = 2mπ + 2 tan (7.28)
2md ln csc πg


2d

while for m = 0 only numerical solution is sufficiently accurate. We plot results for the first three
modes (along with the approximation (7.28) for the TE30 mode) in Fig. 7.11 and corresponding E-field
distributions are shown in Fig. 7.12. For the TE2m+1,0 modes, we see that the cutoff frequencies are
reduced by comparison with the unfinned guide of width 2a as g/d decreases. In a sense, the standing-
wave distribution is allowed to “crawl down” the fins so that a larger transverse wavelength can “fit into”
the guide than is possible without the fins. Subject to verification that the next higher-order modes of
the finned waveguide do not undergo similar reductions in cutoff frequency, it can be seen that this guide
is capable of propagating only the dominant TE10 mode over a wider bandwidth than is the ordinary
rectangular guide.
At this point, it is well to summarize the limitations of this approach. Numerical calculations of kc a
for the TE10 mode of the finned guide are graphically indistinguishable from those computed from (7.27)
and displayed in Fig. 7.11. This accuracy may not be obtained in other situations. In order to assure
reasonable results, we should keep the following in mind:

217
 10

6 TE10

kca TE20
TE30
4
TE30 (approximate)

0
0 0.2 0.4 0.6 0.8 1
g/d

Figure 7.11: Normalized cutoff frequency kc a versus normalized gap width g/d for symmetrically finned
rectangular waveguide TE10 , TE20 , and TE30 modes; a = 2d.

(a) Only TEM parallel-plate waves have been used. If kc is large enough, other higher-order parallel-
plate waves may have to be used to adequately represent the field in each section.
(b) The method has been shown here only for TE modes. For TM modes, a similar analysis treating
Ez must be done, which again may require the use of higher-order parallel-plate waves to represent the
fields in each section.
(c) If kc d is too large (where d is the height of a section), the approximate reflection and transmission
coefficients used in Table 7.1 are no longer accurate. Greater precision can be obtained at the expense
of more complicated formulas.
(d) The representations (7.17) and (7.18) for the fields are not valid too close to a junction. This
is because higher-order, cutoff parallel-plate waves must be present in order to provide continuity of all
tangential fields at the junction (see Chapter 11). When using this method to determine kc , we must
be sure that the horizontal distance separating two junctions is sufficiently large that these cutoff waves
produced at one junction have decayed to a negligible value when they reach the second one. In the
example treated above, this would mean roughly that a/d ≥ 1/2.
Restrictions resulting from points (c) and (d) above can only be relaxed at the cost of considerable
complexity in the analysis, if at all. However, other kinds of modes can be analyzed if we use not TEM
waves, but other kinds of transversely propagating waves in order to represent the fields. In order to do
this, we must in each case set up an appropriate transmission line equivalence and build up a table of
equivalent circuit representations for discontinuities analogous to Table 7.1. We give an example of this
for dielectric waveguides later in the chapter. A further example of a waveguide which can be analyzed
on the basis of TEM waves is that of quasi-optical microstrip which is treated in the next section. Other
examples are given as problems at the end of the chapter.

218
Figure 7.12: Field plots for TE10 , TE20 , and TE30 modes of symmetrically finned rectangular waveguide.

Figure 7.13: Quasi-optical microstrip (cross-section).

7.4 Quasi-Optical Microstrip

Consider the waveguide shown in Fig. 7.13. Structurally it is identical to the microstrip line we will
study in section 8.6, but it is operated in a high-frequency range such that k0 w ≥ 1 and so operates in a
rather different way from the quasi-TEM microstrip mode. We therefore name it quasi-optical microstrip
and will analyze it using the transverse resonance method.
Once again we assume that the field in the parallel-plate region (0 < y < d, |x| < w/2) underneath the
strip (but away from the edges) consists essentially only of a pair of (y-polarized) TEM waves reflecting
from the edges of the strip as shown in Fig. 7.14. Our total electric field Ey would then be
√ √
Ey = E1 e−jk0 ǫr x cos θ
+ E2 e+jk0 ǫr x cos θ
(7.29)
As with the examples of section 7.3, the edges of the strip at x = ±w/2 are not simply plane reflecting
walls and, in fact, will scatter an incident plane wave in a much more complicated fashion than as a
single reflected plane wave. It is thus not possible to apply an exact boundary condition to (7.29) at the
edges to find the relation between E1 and E2 . Nevertheless, there will exist some electric field reflection
coefficient for the TEM wave from a single such edge for an incidence angle of θ, which we will denote
ρ(θ) = ejχ(θ) (7.30)
We can, however, make some general observations about the wave scattering process at the edge,
and their implications about the behavior of this open waveguide. Besides a reflected TEM wave under
the strip, there is also a field penetration beyond the edge, like the transmitted field into the second
medium in the dielectric interface problem of Fig. 6.5. The field in air will satisfy the equation
 
E
(∇2T + kT2 ) =0 (7.31)
H
where
kT2 = k02 − β 2 = −p2 (7.32)

219
 Figure 7.14: Quasi-optical microstrip (top view and ray diagram).

We call this portion of the field a space wave, because it is not localized to the slab. As long as β > k0 ,
kT2 is negative (i.e., kT is imaginary and p is real), and the solution of (7.31), like (6.5) for the interface
problem when θ > θc , decays exponentially away from the edge. But when β < k0 , kT2 is positive and
kT is real. This means that the solution of (7.31) will be a wave propagating away from  the edge and
whose energy is taken from the incident wave. For this reason only partial reflection, ρ(θ) < 1, is to
be expected.
−1/2
We thus conclude that β > k0 (or sin θ > ǫr ) is necessary if a guided wave without radiation loss is
to result from these repeated reflections at the edge. In fact, not only waves propagating into the air from
the edge might be present, but also surface waves along the grounded dielectric slab, beyond the edge.
Since the propagation factor exp(−γz) = exp(−jβz) is present, the surface waves whose propagation
coefficients are βsm (along some direction in the xz-plane), m = 0, 1, 2, . . . , will have a field structure
proportional to √ 2 2
e−psm y e−j βsm −β |x|
in the air region y > d beyond the edges of the strip, and to

sin(hsm y)
 √ 2 2
e−j βsm −β |x|
cos(hsm y)

in the dielectric, where psm and hsm are the values of h and p corresponding to the surface wave solutions
of (6.64) or (6.65). Evidently, if β < βsm , part of the field will radiate away from the edge as a surface
wave. For β > βsm , on the other hand, this field decays exponentially in the x-direction.
We thus have a set of qualitative criteria which constrain the possible values of β for proper (bound)
modes of planar quasi-optical structures. We first identify the types of wave which might carry energy
away from the guiding region. In the foregoing example, the space wave (which obeys (7.31)) and the

220
 Figure 7.15: Equivalent transverse transmission-line network for quasi-optical microstrip.

surface waves serve this role. A bound mode must then have a value of β or greater than a lower limit
for each of these wave types (k0 for a space wave and βsm for a surface wave). On the other hand,
Appendix K gives an upper limit of p
β ≤ ω (µǫ)max (7.33)
where (µǫ)max is the maximum value of µǫ over the waveguide cross-section. Thus, for our quasi-optical
microstrip we have that β must be restricted to the range
√
βs,TM0 ≤ β ≤ ω µ0 ǫ0 ǫr (7.34)
where βs,TM0 is the propagation coefficient for the TM0 surface wave on the grounded dielectric slab—the
highest value for any of the surface waves which may exist on the slab.
√
If the substrate is electrically thin (k0 ǫr d ≪ 1), it has been found that the phase χ(θ) of the
reflection coefficient can be accurately approximated by
χ(θ) ≃
  
2k0 d
q
2 2
1/2
(1 − ǫr sin θ) ln(k0 d ǫr sin θ − 1) + γE − 1
πǫr cos θ
 
2 1 − ǫr
− ǫr cos θ ln 2π − (ǫr − 1)Q (7.35)
1 + ǫr
In (7.35), γE is used to denote Euler’s constant (0.5772...), and Q is the function defined in (B.61).
Applying the reflection conditions at x = ±w/2, we obtain
√ √
ejχ(θ) E1 e−jk0 w√ǫr cos θ/2 = E2 ejk0 w√ǫr cos θ/2

(7.36)
ejχ(θ) E2 e−jk0 w ǫr cos θ/2 = E1 ejk0 w ǫr cos θ/2
whose solution yields E2 = ±E1 and
√
k0 w ǫr cos θ − χ(θ) = mπ m = 0, 1, 2, . . . (7.37)
As in the previous sections, we might also have obtained this result using the transverse resonance
of an equivalent transverse transmission line. The near-total reflection implied by (7.30) and (7.35) is
equivalent to a lumped load admittance YL connected at either end as shown in 7.15. In this circuit the
classical transmission-line variables are given by the replacements

z → x
√ 

γ → jh ≡ jk0 ǫr cos θ  

V → Ey (7.38)
I → Hz 


Zc → ωµ ζ

h = cos θ ,
0 

221
 √
and the lumped admittance YL for k0 ǫr d ≪ 1 is given by

1 1−ρ j χ
YL = =− tan (7.39)
Zc 1 + ρ Zc 2
jk0 d p2 h2
  1 − ǫ 
r
≃ [ln pd + γ − 1] + 2 ln 2π + (ǫr − 1)Q
πζ0 k02 k0 1 + ǫr
√
where β = k0 ǫr sin θ is the yet-to-be-determined
p propagation coefficient of the microstrip mode (not of
the equivalent transverse line), h = k 2 ǫ − β 2 is the transverse wavenumber of the TEM wave under
p 0 r
the strip, and p = β 2 − k02 . With the impedance looking left from x = −w/2 equal to 1/YL , we have
the transverse resonance condition
(1/YL ) cos hw + jZc sin hw 1
Zc + = 0, (7.40)
Zc cos hw + j(1/YL ) sin hw YL

which can be shown to be equivalent to (7.37).

Equation (7.37) is a transcendental equation which, like those encountered in chapter 6, must be
solved numerically or by analytic approximation. Since d/w is small under the conditions we have
imposed, we can accomplish the latter as follows. We substitute (7.35) in (7.37), expressing Q in terms
of  
1 − ǫr
rǫ = r (7.41)
1 + ǫr
where r(x) is a special function defined in eqn. (B.60) of Appendix B. From (B.62), we have
s
0.5160 0.0788
rǫ ≃ 0.4052 + + (7.42)
ǫr ǫ2r

Then, using ǫeff = ǫr sin2 θ instead of θ itself as the unknown, we have for the m = 0 mode

2d h  2πe1−γE i e1−γe
 
2d h   i
ǫeff 1 + ln √ = ǫr + ln √ + ǫr ln(2πrǫ ) (7.43)
πw k0 d ǫeff − 1 πw k0 drǫ ǫeff − 1

Now, as d/w → 0, the solution to this equation approaches ǫeff = ǫr . The appearance of ǫeff inside the
logarithm (which is why we still have a transcendental equation) provides only a weak dependence on
ǫeff . For purposes of approximation, we can with acceptable accuracy replace ǫeff (there only) by ǫr (this
is the first step of an iterative solution process, as described in Appendix C), thus obtaining an explicit
approximation for ǫeff : h i
2d e1−γ E


ǫr + πw ln k dr √ + ǫ r ln(2πrǫ )
ǫ ǫr −1
ǫeff ∼
0
= 2d 1−γE

2πe√
(7.44)
1 + πw ln k d ǫ −1
0 r

1
This formula is valid so long as k0 d ≪ 1 but k0 w ≥ 2. A typical example of the dispersion relation
(7.44) is shown in Fig. 7.16.
−1/2
A point worth noting about equation (7.35) is that, if sin θ > ǫr , χ(θ) is purely real if ǫr is real.
−1/2
But, if sin θ < ǫr , we have

πj
q q
ln(k0 d ǫr sin2 θ − 1) = + ln(k0 d 1 − ǫr sin2 θ),
2
  √
and we observe that ρ(θ) < 1 and the reflection is less than total. In the former case, β = k0 ǫr sin θ >
k0 , while in the latter case β < k0 . As discussed above, total reflection does not occur in the latter case
because of the propagating (rather than exponentially decaying) space wave in the air region beyond

222
 Figure 7.16: Dispersion of the fundamental mode on quasi-optical microstrip.

the edge of the strip. In the approximation k0 d ≪ 1 which we have invoked in (7.35), the TM0 mode
wavenumber is approximated by (see section 6.6 and problem p6-7):
2
βs,TM  ǫ − 1 2
r
0
≃ 1 + k02 d2
k02 ǫr
≃ 1 (7.45)

so that the left side of (7.34) becomes simply β ≥ k0 and is identical with the constraint that arises from
the space wave.

7.5 Dielectric Rib Waveguide: The Effective Index Method

Consider the dielectric rib waveguide shown in Fig. 7.17. A section of dielectric slab of thickness ai
and width w is located between two semi-infinite slabs of somewhat smaller thickness ae . (In reality, of
course, these outer slabs would merely be of large but finite extent.) Let us assume that
q
k0 ai n21 − n22 < π,

so that only the TE0 and TM0 surface waves may exist on either of the slab sections.
Suppose we try to model certain modes of this guide such that, in the central region, 0 < x < w,
the field can be expressed as the repeated reflection of TE0 surface waves between the discontinuities
at x = 0 and x = w. If βi is the surface wave propagation coefficient of the TE0 wave (as found from
(6.33) with a = ai ) and fi,TE (y) is the characteristic field distribution of the field component Ez of this

223
 Figure 7.17: Dielectric rib waveguide.

surface wave in the vertical direction (see (6.32) and (6.41)):

fi,TE (y) = cos hi1 y (|y| < ai /2)

hi1 ai −pi2 (|y|−ai /2)
= cos e (|y| > ai /2) (7.46)
2
p p
where hi1 = k12 − βi2 and pi2 = βi2 − k22 , then we write
√ 2 2 √ 2 2 
Ez = E1 e−j βi −β x + E2 e+j βi −β x fi,TE (y)

(7.47)

Here E1 and E2 are amplitude constants to be determined, while β is the (yet unknown) overall propa-
gation coefficient of the dielectric rib waveguide mode.
If we introduce the propagation factor exp(−jβz) into (7.47), we note that it consists of two surface
waves, propagating not in the z-direction nor in the x-direction purely, but at some angle in the xz-plane,
as suggested in Fig. 7.18. Indeed, the angle θ obeys

β = βi sin θ
q
βi2 − β 2 = βi cos θ (7.48)

and it is clear that a component of E in the x-direction must be present as well, due to the oblique
propagation of these surface waves. We do not, however, require it for the present analysis, and we do
have that Ey ≡ 0 in the approximation we are making.
In the outer slab regions the fields,
p which are still associated with the same propagation factor
exp(−jβz), will behave as exp − j βe2 − β 2 |x| , where βe is the TE0 surface wave propagation coeffi-
cient for these outer slabs. If the mode of the rib waveguide is to be bound to the vicinity of the central
region, we will have to have β > βe .
In order to set up a transverse resonance analysis of this waveguide, we need to have a supply of
information on equivalent circuit representations for TE0 slab surface wave discontinuities analogous to
Table 7.1. The transmission line analogy is summarized as

z → xp 
γ → j βi2 − β 2 TE0 slab surface waves. (7.49)
V : Ez = V (x)fi,TE (y)


where we omit specification of I and Zc , since only one value of Zc appears in the class of structures we
will deal with here. Appropriate equivalent networks are given in Table 7.2.

224
 Figure 7.18: TE0 surface waves reflecting from rib edge.

Junction Parameters/Equivalent
√ 2 2 2 Circuit
βi −β βe
7.2.1 Symmetrical Step in Height BZc = √ 2 2 β 2 (0 6= ae < ai )
β −βe i

√
βi2 −β 2 k22
7.2.2 Truncated Slab BZc = p̃ βi2
; where
p̃2 = 2 2 2

β − k2 + hi1 (4 + 8/pi2 ai )

Table 7.2: Transverse Equivalent Networks for TE0 Surface Waves at Dielectric Slab Junctions.

225
 We are now in a position to analyze the so-called E x modes of the rib waveguide (there being no
y-component of E in the constituent TE0 surface waves and approximately none overall). p The transverse
equivalent transmission-line network is again as in Fig. 7.15, except that h is replaced by βi2 − β 2 , and
YL is given by jB from entry 7.2.1 in Table 7.2. Hence, in analogy with (7.40) we obtain the following
eigenvalue equation for β:
2Zc B
q
tan βi2 − β 2 w = 2 2
Zc B − 1
or
2hsi pse
tan hsi w = 2 2 2 (7.50)
hsi (βe /βi ) − p2se (βi2 /βe2 )
where we denote
q
hsi = βi2 − β 2
p
pse = β 2 − βe2 (7.51)

The E y -modes of rib waveguides can similarly be analyzed as the transverse resonances of TM0 slab
modes. We let fi,TM (y) denote the characteristic field distribution of the Ey component of the TM0 wave
on the central slab, and now we let βi and βe denote the surface wave propagation coefficients of the
TM0 waves on the central and outer slabs, respectively. We write
√ 2 2 √ 2 2 
Ey = E1 e−j βi −β x + E2 e+j βi −β x fi,TM (y)

(7.52)

where E1 , E2 , and β are to be determined by transverse resonance. Our transmission-line analogy is


z → xp 
γ → j βi2 − β 2 TM0 slab surface waves. (7.53)
V : Ey = V (x)fi,TM (y)


Some useful equivalent networks for this case are given in Table 7.3. The setup and solution of the
eigenvalue equation for the E y modes of the rib guide as well as for other types of guides can now be
done just as before. The details are left as an exercise for the reader.
Note the close similarity between (7.50) and the TM-mode eigenvalue equation (6.62) for a dielectric
slab. Indeed, (7.50) would result if we analyzed the TM modes of a dielectric slab whose core refractive
index were βi /k0 = n1eff , the TE0 mode effective index of the central slab region in Fig. 7.17, while
the cladding index was βe /k0 = n2eff , the TE0 mode effective index for the outer slabs in Fig. 7.17.
It must be kept in mind that we must first solve equation (6.33) to obtain βi and βe before we can
proceed to solve (7.50) for β itself. The observed analogy between (7.50) and (6.62) has led to the
term “effective index method” for the application of transverse resonance techniques to this kind of
waveguide. But such a close analogy as this results only because of the specific form of the reflection
coefficient of the constituent TE0 wave at the rib boundary (i.e. of B) as given in 7.2.1 of Table 7.2.
However, this expression is merely an approximation which has proven sufficiently accurate in practice
to justify its use. In fact, the other entry in Table 7.2 and those in Table 7.3 do not result in such simple
interpretations of effective-index slab structures, and the name “effective-index” is really a misnomer in
such cases. Strictly, the “effective-index method” refers to a means of approximating B in Tables 7.2
and 7.3 and then to a straightforward application of transverse resonance techniques based upon these
approximations.

226
 Junction Parameters/Equivalent
√ 2 2 2 Circuit
β −βe βi
7.3.1 Symmetrical Step in Height BZc = √ 2 2 β 2 ; (0 6= ae < ai )
βi −β e

p̃ βi /ω
7.3.2 Truncated Slab BZc = √ ; where p̃2 = β 2 − k22 +
βi2 −β 2 dβi /dω
h2i1 (h2i1 −2pi2 /ai )(n22 /n21 )+2pi2 /ai +p2i2 (n21 /n22 )
4 h2i1 (1+2/pi2 ai )(n22 /n21 )+2pi2 /ai +p2i2 (n21 /n22 )

Table 7.3: Transverse Equivalent Networks for TM0 Surface Waves at Dielectric Slab Junctions.

227
7.6 Notes and References
Discussions of the transverse-resonance/equivalent transmission-line model for multilayer slabs are to be
found in
Felsen and Marcuvitz (1973), section 2.4;
Solimeno, et al. (1986), chapter 3.
Multiple quantum well structures are discussed in
T. H. Wood, “Multiple quantum well (MQW) waveguide modulators,” J. Lightwave Technol.,
vol. 6, pp. 743-757 (1988).
H. Yakamoto, M. Asada and Y. Suematsu, “Theory of refractive index variation in quantum
well structure and related intersectional optical switch,” J. Lightwave Technol., vol. 6,
pp. 1831-1840 (1988).
Applications of the WKB method to inhomogeneous dielectric slab waveguides (including the expo-
nential diffused profile example) are given in:
A. Gedeon, “Comparison between rigorous theory and WKB-analysis of modes in graded-
index waveguides,” Opt. Commun., vol. 12, pp. 329-332 (1974).
G. B. Hocker and W. K. Burns, “Modes in diffused optical waveguides of arbitrary index
profile,” IEEE J. Quantum Electron., vol. 11, pp. 270-276 (1975).
J. Janta and J. Čtyroký, “On the accuracy of WKB analysis of TE and TM modes in planar
graded-index waveguides,” Opt. Commun., vol. 25, pp. 49-52 (1978).
V. Ramaswamy and R. K. Lagu, “Numerical field solution for an arbitrary asymmetrical
graded-index planar waveguide,” J. Lightwave Technol., vol. 1, pp. 408-417 (1983).
Perhaps the earliest application of approximate transverse resonance methods was in
S.B. Cohn, “Properties of ridge waveguide,” Proc. IRE, vol. 35, pp. 783-788 (1947).
Other applications to metallic waveguides are to be found in
L.O. Goldstone and A.A. Oliner, “Leaky-wave antennas I. Rectangular Waveguides,” IRE
Trans. Ant. Prop., vol. 7, pp. 307–319 (1959).
J.R. Pyle, “The cutoff wavelength of the TE10 mode in ridged rectangular waveguide of any
aspect ratio,” IEEE Trans. Micr. Theory Tech., vol. 14, pp. 175–183 (1966).
P.J.B. Clarricoats and P.E. Green, “Waveguide structures for double-beam leaky-wave an-
tennas,” Proc. IEE (London), vol. 114, pp. 604–610 (1967).
E. V. Orleanskaya, “Broadband device for rotating the plane of polarization,” Telecommun.
Radio Eng., vol. 22/23, no. 8, pp. 75-80 (1968).
E. V. Orleanskaya, “Application of the methods of network theory to problems of waveguides
partially filled with dielectric,” Telecommun. Radio Eng., vol. 24/25, no. 3, pp. 86-91
(1970).
A.A. Oliner and P. Lampariello, “The dominant mode properties of open groove guide: An
improved solution,” IEEE Trans. Micr. Theory Tech., vol. 33, pp. 755–764 (1985).
The equivalent networks in Table 7.1 are obtained from
Marcuvitz (1965), chapter 5.
Details of the analysis of quasi-optical microstrip can be found in
D.C. Chang and E.F. Kuester, “Total and partial reflection from the end of a parallel-plate
waveguide with an extended dielectric slab,” Radio Science, vol. 16, pp. 1-13 (1981).
E.F. Kuester, R.T. Johnk and D.C. Chang, “The thin-substrate approximation for reflection
from the end of a slab-loaded parallel-plate waveguide with application to microstrip
patch antennas,” IEEE Trans. Ant. Prop., vol. 30, pp. 910-917 (1982).

228
The “exact” (numerical) solution given in Fig. 7.16 is due to
R.H. Jansen, “High speed computation of single and coupled microstrip parameters including
dispersion, high-order modes, loss and finite strip thickness,” IEEE Trans. Micr. Theory
Tech., vol. 26, pp. 75-82 (1978).
The original paper on effective-index methods was
R. M. Knox and P. P. Toulios, “Integrated circuits for the millimeter through optical fre-
quency range,” in Proc. Symp. Submillimeter Waves, Brooklyn: Polytechnic Press, 1970,
pp. 497-510,
and since than a number of refinements and extensions have been made. See
W. V. McLevige, T. Itoh and R. Mittra, “New waveguide structures for millimeter-wave
and optical integrated circuits,” IEEE Trans. Micr. Theory Tech., vol. 23, pp. 788-794
(1975).
S.-T. Peng and A. A. Oliner, “Guidance and leakage properties of a class of open dielectric
waveguides: Part I—Mathematical formulations,” IEEE Trans. Micr. Theory Tech.,
vol. 29, pp. 843-855 (1981).
A. A. Oliner, S.-T. Peng, T.-I. Hsu and A. Sanchez, “Guidance and leakage properties of a
class of open dielectric waveguides: Part II—New physical effects,” IEEE Trans. Micr.
Theory Tech., vol. 29, pp. 855-869 (1981).
F. P. Payne, “A new theory of rectangular optical waveguides,” Opt. Quantum Electron.,
vol. 14, pp. 525-537 (1982).
S. Zhendong and L. Jiang, “Effective dielectric parameter method—a new method for ana-
lyzing propagation characteristics of dielectric waveguides,” Micr. Opt. Technol. Lett.,
vol. 2, pp. 427-431 (1989).
A nice summary of these methods is to be found in
Adams (1981), chapter 6.
We follow here the treatments of
Vassallo (1985), tome 2, chapter 7,
Koshiba (1992), chapter 5.
from which most of the entries of Tables 7.2 and 7.3 are obtained, except for entry 7.3.1 which is due to
unpublished work by D.C. Chang and L.H. Xiong.
The papers of Oliner, Peng et al. in particular discuss the effects of considering transformations
between TE and TM modes during the reflection process at the edge of a dielectric rib. It may happen
that neglecting this coupling predicts a perfectly bound mode (consisting of TE surface waves only),
while the presence of a small but nonzero constituent of TM surface waves actually causes the mode to
be leaky, in the manner discussed in Section 7.4. Such qualitative considerations should always be borne
in mind when applying approximate transverse resonance methods.

7.7 Problems
p7-1 A multilayer slab waveguide consists of an inner core of index n1 , two coating layers of index
n3 , and a cladding of index n2 with

n1 > n2 > n3

as indicated.

229
 (a) Obtain the eigenvalue equation for the TE modes of this waveguide.
p
(b) Write a computer program to calculate neff vs. V = k0 a1 n21 − n22 for the TE0 mode.
Plot your numerical results for the case when n1 = 2.0, n2 = 1.9 and n3 = 1.5, with
a2 /a1 = 0.1, 0.3 or 1.0 successively.

p7-2 Derive the transverse resonance equation for the TM modes of the slab waveguide with
exponential permittivity profile described in section 7.2. Plot b vs. V1 for the fundamental
TM mode using the values n1 = 5, n2 = 4.9 and n0 = 1. Use the result of problem p7-4.

p7-3 A slab waveguide has a refractive index profile of

n2 (x) = n22 (|x| > a)

x2
= n22 + (1 − 2 )(n21 − n22 ) (|x| < a)
a
and µ = µ0 everywhere.

Use the WKB method to calculate and plot the values of

n2eff − n22
b=
n21 − n22

230
 p
vs. normalized frequency V1 ≡ k0 a n21 − n22 for the fundamental TE mode of this guide.
You may ignore the penetration of the fields beyond the turning points (when is this a good
approximation?).

p7-4 Show that the TM modes of a multilayered slab waveguide can be analyzed by the transverse
resonance method if we use the correspondences (7.10).

p7-5 Show that equation (7.40) is equivalent to equation (7.37).

p7-6 Using the methods of Sections 7.3 and 7.4, construct an eigenvalue equation for determining
the propagation coefficients of E y -polarized modes of the ridged quasi-optical microstrip
shown.

These modes are formed by the transverse resonance of the appropriate TEM parallel-plate
waves.

p7-7 Using a TEM wave description as in Section 7.3, construct an eigenvalue equation for de-
termining the propagation coefficients of the lowest-order TE mode (predominantly E y -
polarized) of the symmetric double-ridge metallic waveguide shown.

Solve this equation numerically for kc w1 for the dominant mode if w1 = w2 = d2 and
d1 /d2 = 0.5. Sketch what you believe the transverse electric field distribution for this mode
looks like. What advantages does this guide have over an ordinary metallic rectangular
waveguide of width 3w1 and height w1 ?
p
p7-8 Plot dispersion curves of b = (β 2 /k02 − n22 ) (n21 − n22 ) versus V = k0 w n21 − n22 for the lowest-

p p
order solution of (7.50) if n21 = 2.25, n22 = 1, k0 a1 n21 − n22 = 1.0, and k0 aq 2
2 n1 − n2 =
2

0.95. Cover the range 0 ≤ V ≤ 100. Also display the value of Veff ≡ k0 w n21,eff − n22,eff
along the horizontal axis for comparison with V .

p7-9 A rectangular metallic waveguide whose height d = a/2 is half the width a is fitted with a
pair of fins whose height is d/2 each as shown below.

231
 a/3

d/2 a/3
d=a/2
d/2

Use the transverse resonance method to obtain an equation for determining the eigenvalues
kc for the modes of this waveguide that are formed from TEM parallel plate waves. Use
the software of your choice to determine the values of kc a for the first three such modes,
in order of increasing cutoff wavenumber, and compare these values with those of an empty
rectangular waveguide.
p7-10 For the partially filled waveguide of Figure 7.9, let a = 0.9 in, d = 0.4 in, h = 0.1 in, and
let the relative permittivity of the dielectric layer be ǫr = 10. Find the value of the cutoff
frequency fc for the fundamental mode of this waveguide (it occurs when γ = α + jβ = 0).
Numerically solve the transcendental equation (7.23) for β, covering at least the range of
frequencies from fc to 3fc . Compare these results with those for an unloaded (air-filled)
rectangular waveguide of the same overall dimensions.
p7-11 A slab waveguide has a refractive index profile of

n21 − n22
n2 (x) = n22 +
cosh2 (x/a)

and µ = µ0 everywhere. Use the WKB method to calculate and plot the values of

n2eff − n22
b=
n21 − n22
p
vs. normalized frequency V1 ≡ k0 a n21 − n22 for the fundamental TE mode of this guide.
You may ignore the penetration of the fields beyond the turning points (when is this a good
approximation?).
[Hints: The integral
Z r
1
− b du
cosh2 u
maypbe usefully transformed using the successive changes of variable cosh u = 1/y and
y = sin2 θ + b cos2 θ. Then you may find the integral
π/2
dθ π
Z
2 2
= √ ; b>0
0 sin θ + b cos θ 2 b
of some use in this problem.]

232
Chapter 8

TEM AND QUASI-TEM MODES:

BASIC PLANAR TRANSMISSION
LINES

There exist a class of men trained in soul-craft whom

they call straighteners, as nearly as I can translate a
word which literally means “one who bends back the
crooked”. . . . the straighteners have gone so far as to
give names from the hypothetical language (as taught
at the Colleges of Unreason), to all known forms of
mental indisposition, and to classify them according
to a system of their own, which, though I could not
understand it, seemed to work well in practice . . .
—Samuel Butler,
Erewhon

8.1 Introduction
In solid-state circuits operating at microwave (or higher) frequencies, especially if low power levels are
involved, it is convenient to be able to fabricate intricate transmission-line networks by the same processes
(photolithography, etc.) which have proved so successful in conventional circuit design, and have made
printed circuits and integrated circuits commonplace. The use of transmission lines involving essentially
only planar (strip) conductors reduces manufacturing problems in many situations. In this chapter, we
will examine the quasistatic mode properties of such striplines, and discuss two important classes of
modes in the study of such transmission lines: TEM and quasi-TEM modes.
Striplines which are filled with an electrically homogeneous medium (µ and ǫ are independent of x
and y) belong to the TEM class of waveguide structures, examples of which are shown in Figure 8.1.
Also belonging to this class are the coaxial and two-wire transmission lines shown in Figures 1.1(b) and
(c). Common to all TEM waveguides is the presence of two or more separate conducting boundaries
which allows a nontrivial TEM mode to exist.
As will be noted from these examples, some types of transmission lines have cross sections which
are (at least conceptually) of infinite extent, while others are partially or completely shielded by one
of the conducting boundaries. Generally, unshielded versions of planar transmission lines are suitable
for use at frequencies up to about 30 or 40 GHz. Above this range, the tendency of planar circuits
towards unwanted radiation dictates that shields be employed. A potential drawback to a shield is that

233
Figure 8.1: TEM-type stripline structures: (a) open stripline; (b) shielded stripline; (c) balanced
stripline.

it involves more metal surface area for the waveguide structure, which can result in increased conductor
loss if a significant surface current flow is present there. In practice, shielded lines are used well into the
millimeter wave band, up to 140 GHz.
Perhaps the simplest of all striplines is a thin, perfectly-conducting strip located above and parallel
to a conducting ground plane of infinite extent. This idealized transmission line is sometimes referred
to as open, unshielded, or unbalanced stripline (Figure 8.1(a)). A variation on this is the completely
shielded line of Figure 8.1(b), called shielded stripline or rectangular coaxial transmission line (RCTL),
which is used in measurement applications to form a TEM cell. Another version is the partially-shielded
structure of Figure 8.1(c), which is called triplate, semi-open or balanced stripline, and has two infinitely
extended ground planes.
Striplines using air dielectrics must have their center conductors supported at the ends of a section
of line, and this may not be adequate if close dimensional tolerances must be maintained. A homoge-
neous dielectric filling might be provided for the purpose of supporting the strip, but covering the strip
complicates the fabrication process, makes external circuit connections more difficult, and the presence
of losses in the dielectric can lead to unwanted attenuation on the line. Commonly, then, we see the use
of striplines partially filled with dielectric material, so that support is provided, and yet manufacture is
easy (the strips are simply deposited on the surface of the dielectric) and the total amount of potentially
lossy dielectric is reduced. Strips deposited directly on the surface of a dielectric layer thus facilitate the
manufacture of microwave integrated circuits as mentioned before.
As we will see, such structures of inhomogeneous cross-sectional properties (that is, where ǫ = ǫ(x, y)
and/or µ = µ(x, y) are functions of transverse position) cannot generally support pure TEM modes,
except in the trivial case of zero frequency. However, there can exist modes which are (in a sense to
be made more precise later on) “close” to TEM, and are called quasi-TEM modes. In contrast to
TEM modes, quasi-TEM modes are dispersive (phase velocity depends on frequency) and admit only
an approximate description in terms of voltage and current. The quasi-TEM nature of these modes is
generally present only when the characteristic transverse dimensions of the structure are small compared
to a wavelength. Once frequency becomes sufficiently large, they behave more like a TE or TM mode of
a hollow waveguide, or like a surface wave on a dielectric waveguide (cf. chapter 6).
Several quasi-TEM stripline structures are depicted in Figure 8.2. Figure 8.2(a) is known as mi-
crostrip, or more specifically, open microstrip; it uses a dielectric substrate such as alumina to support
the strip conductor above the ground plane. Figure 8.2(b) illustrates shielded microstrip, while Fig-
ures 8.2(c) and (d) show two variants of elevated microstrip, one shielded and one open. When the strip
is located beneath the substrate but over the ground plane, we have an example of inverted microstrip

234
Figure 8.2: Quasi-TEM stripline structures: (a) microstrip; (b) shielded microstrip; (c) elevated mi-
crostrip; (d) inverted microstrip.

235
 Figure 8.3: (a) Slotline; (b) coplanar waveguide; (c) bilateral slotline (open fin-line).

as shown in Figure 8.2(d).

While striplines are generally (but not always) transmission lines where the dielectric, when present,
is located between the two line conductors, other planar structures are used as well. Figure 8.3 shows
some coplanar lines, which have in common that all line conductors are deposited on the same side of the
substrate. In this class are the basic slotline of Figure 8.3(a) and the coplanar waveguide of Figure 8.3(b).
A bilateral slotline is shown in Fig. 8.3(c); as discussed in section 8.10, it has a tendency to radiative
leakage of energy and is normally surrounded by a shield (Fig. 8.4). This is an example of a finline
which is basically a shielded slotline. Although finlines do not possess two or more separate conducting
boundaries as do the other quasi-TEM structures considered in this chapter, they do share many of the
same characteristics and are usually associated with them rather than with closed metallic waveguides.
Each of these structures has its own advantages and weaknesses which tailor it to a particular type of

Figure 8.4: Bilateral finline.

236
 xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
Cg
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
("ground")
xxxxxxxxxxxxxxx

xxxxxxxxxxxx
xxxxxxxxxxxx
C1
("positive")

un ul xxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx
S0 C
un ul
ul un

µ, ε y

z x

Figure 8.5: Cross-section of TEM waveguide.

application.
We will begin this chapter by discussing the general properties of TEM modes, and then proceed to
discuss the characteristics of some quasi-TEM modes. The chapter concludes with some applications to
specific types of planar transmission-lines.

8.2 TEM Modes

Let us begin by considering a waveguide with a doubly-connected (two-piece) perfectly-conducting walls
as shown in Figure 8.5. If the guide is homogeneously filled, then the conclusions reached in chapter 5
apply to this guide as well: it supports an infinite number of TE and TM modes; each mode has some
positive cutoff frequency, etc. Let us suppose for now that our waveguide is inhomogeneous, which means
that ǫ(x, y) and/or µ(x, y) are nonconstant functions of the transverse variables. There will still be modes
of such waveguides analogous to the TE and TM modes of uniformly-filled metallic waveguides, in that
they possess positive cutoff frequencies, although now the modes are generally hybrid, as were the modes
of the dielectric waveguides studied in chapter 6.
This doubly-connected waveguide, as we will see, is also capable of propagating a mode with zero
cutoff frequency. An important particular situation when this occurs is that of a TEM (transverse
electromagnetic) mode. We saw in section 5.1 the example of a parallel-plate metallic waveguide, which
supports a TEM mode. This was basically the confinement between two parallel plates of a plane wave,
which is also an example of a TEM wave. We devote this section to a study of the general properties of
TEM modes.
For a TEM mode, Ez and Hz will vanish. Thus, if a TEM mode is possible, equations (5.22) and
(5.23) are not going to be of much use to us. Thus we return to (5.17)-(5.20). Setting Ez = 0 and
Hz = 0, we get
−γuz × ET = −jωµHT (8.1)

237
 ∇T × ET = 0 (8.2)
−γuz × HT = jωǫET (8.3)
∇T × HT = 0 (8.4)
Equations (8.1) and (8.3) tell us that unless ET and HT are both zero (the trivial case), we must have

γ jωǫ
=
jωµ γ
or,
γ = ±jk (8.5)
where
√
k = ω µǫ (8.6)
is the wavenumber of a plane wave in the medium filling the transmission line. We can conclude that
unless µǫ = constant, there can be no pure TEM mode on a transmission line. Notice, however, that
this does not necessarily require µ and ǫ individually to be constant. We could imagine (with the use
of magnetic materials such as ferrites) that the waveguide cross-section is filled with different media for
which µǫ is always the same.
Then, (8.1) and (8.3) become
uz × ET
HT = ± (8.7)
ζTEM (x, y)
with s
µ(x, y)
ζTEM = = ζ(x, y) (8.8)
ǫ(x, y)

where ζ(x, y) is the wave impedance [µ(x, y)/ǫ(x, y)]1/2 of a plane wave in the filling medium. The ±
signs in (8.7) correspond to those in (8.5). Hereafter, we will confine ourselves to forward modes only,
and take + signs in (8.5) and (8.7).
If the transverse divergence is taken of eqns. (8.1) and (8.3), and use made of identity (D.8) and
eqns. (8.2) and (8.4), our TEM mode field equations decouple into independent ones for ET and HT (if
ω 6= 0):
∇T · (ǫET ) = 0 ; ∇T × ET = 0 (8.9)
∇T · (µHT ) = 0 ; ∇T × HT = 0 (8.10)
Equations (8.9) and (8.10) are in fact the two-dimensional (∂/∂z ≡ 0) source-free, static field equations
for ET and HT . The boundary condition for (8.9) is that the tangential field component Eℓ = ul · ET
(cf. Figure 8.5) vanishes at the walls; for (8.10), that the normal component Hn = un · HT vanishes
there.
It must be possible to express ET as the transverse gradient of a scalar potential1 :

ET = −∇T Φ(x, y) (8.11)

Substituting (8.11) into the first of (8.9), we get

∇T · [ǫ∇T Φ(x, y)] = 0 (8.12)

or, if ǫ is independent of x and y,

∇2T Φ = 0 (8.13)

1 To avoid the problematical introduction of a “script Φ”, we simply use capital Φ for the present purpose. We will

later use a script Q to avoid similarly a “script ρ”. No conflict should arise from this usage.

238
which is Laplace’s equation for the potential function Φ(x, y). Hence, if as in Chapter 2 we denote the
boundary wall by the contour C0 (which is made up of two parts, C1 and Cg as shown in Figure 8.5),
then (8.11) and the boundary condition on E implies that ∂Φ/∂ℓ = 0 on Cg , or in other words,
Φ|Cg = constant (8.14)

Were the second conductor absent, the uniqueness theorem for Laplace’s equation (or the more general
equation (8.12)) tells us that Φ would be constant throughout the whole cross-section of the waveguide,
and thus ET (and with it HT ) would vanish identically. Hence, there can be no TEM mode in a waveguide
with a simply-connected boundary.
When two or more conducting boundaries are present, however, we can have Φ assume different
constant values on each separate portion of the boundary, and nontrivial solutions for ET and HT can
be found. By convention, we take C1 to be the “positive” conductor with respect to Cg (which is
sometimes referred to as “ground” and taken to have potential zero). For the two-conductor case shown
in Figure 8.5, we thus put
Φ|C1 = Φ1 = constant
Φ|Cg = 0 (8.15)
with Φ1 6= 0. Once we have solved for Φ throughout the entire cross-section of the waveguide, equations
(8.5)-(8.8) and (8.11), in principle, tell us all we need to know about this mode.
Since TEM structures, often referred to as transmission lines, are usually connected between localized
circuits, it is useful to associate some circuit concepts with the TEM mode. In this way, they can be
related to the classical transmission line described in section 1.2, and described by the same equations.
Because of the boundary conditions on normal E and tangential H at the conducting surfaces C1 and
Cg , surface charge and current densities will be present on both conductors:
ǫun · E|C1 = ρS1
un × H|C1 = JS1 (8.16)

ǫun · E|Cg = ρSg

un × H|Cg = JSg (8.17)
where ρSi and JSi are the surface charge and current densities on conductor Ci , for i = 1 or g. Of
course, these surface densities both depend on the coordinate z as e−jkz , so we can write
ρSi = QSi (ℓ)e−jkz
JSi = JSi (ℓ)e−jkz (8.18)
where QSi and JSi depend only on the coordinate ℓ which is the arc length along either C1 or Cg .
Furthermore, since Hz = 0, there can be no transverse component of the surface current, and JSi is
thus totally z-directed:
JSi = uz Jzi (ℓ) (8.19)
The electrostatic behavior of the configuration of Figure 8.5 can be summarized by the total charge
per unit length (or linear charge density) on the conductor C1 :
I
Ql1 = QS1 (ℓ) dℓ (8.20)
C1

and the potential difference Φ1 between C1 and Cg :

Z 1
Φ1 = − ET · dℓ (8.21)
g

239
 xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
Cg
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx

xxxxxxxxxxxx
xxxxxxxxxxxx
C1

1 BT
ul
ET
un
g

z x

Figure 8.6: Contour for calculating potential and magnetic flux between C1 and C2 .

Because ET is the gradient of a scalar function, the path of integration in (8.21) can be any contour
between a point 1 on C1 and a point g on Cg (see Figure 8.6). Likewise, because of the first of (8.9)
and the two-dimensional divergence theorem, the path of integration in (8.20) may be taken to be any
closed contour C lying outside of C1 but inside of Cg (see Figure 8.5).
The total charge per unit length on Cg is simply the negative of that on C1 , as we can show using
(8.16), (8.17), (8.20), the two-dimensional divergence theorem, and (8.9):
I
Qlg = QSg dℓ
Cg
I
= ǫun · ET dℓ
Cg
I Z
= − ǫun · ET dℓ − ∇T · (ǫET ) dS (8.22)
C1 S0
= −Ql1

The electrostatic characteristics of the line are succinctly described by defining the distributed ca-
pacitance c per unit length as
c ≡ Ql1 /Φ1 (8.23)
As stated above, we can write the linear charge density as
I
Ql1 = ǫEn dℓ (8.24)
C

for an arbitrary contour C between C1 and Cg , keeping in mind that n represents the normal away from
C. Thus, if we know the solution ET to (8.9) for some transmission line configuration, (8.21), (8.24)
and (8.23) will allow us to determine the capacitance c. It is to be emphasized that these electrostatic

240
characteristics are determined completely by the function ǫ(x, y) and the geometry of the line, and not
by the magnetic properties µ(x, y) at all.
The magnetostatic problem (8.10) is characterized by the current flowing in conductor 1 (that we
will denote by I1 ) and the magnetic flux per unit length Ψl which passes between the two conductors of
the line. Referring to Figure 8.6, we find the flux by integrating BT · un along any path connecting C1
to Cg (for instance, we may use the same one used before in (8.21) to evaluate the voltage Φ1 ), where
un here is the unit vector normal to this path and to uz as well. We obtain
Z 1
Ψl = BT · un dℓ (8.25)
g

The current I1 flowing in the +z direction on conductor C1 is given by integrating the current density
over C1 :
I
I1 = Jz1 (ℓ) dℓ
IC1
= HT · dℓ (8.26)
C1

As with the charge density, the path of integration in (8.26) can be taken to be any closed contour C
between C1 and Cg . It is readily shown by methods analogous to those used in (8.22) (problem p8 - 1)
that an equal but opposite current must flow in the other conductor Cg :
I
Ig = HT · dℓ
Cg
= −I1 (8.27)

The magnetostatic characteristics of the line are then succinctly described in terms of the distributed
inductance l per unit length of the line, defined as

l ≡ Ψl /I1 (8.28)

We emphasize that the magnetostatic characteristics of the TEM mode are determined only by the
magnetic properties µ(x, y) and geometry of the line.
We now establish the connection between the voltage and current associated with this mode, which
will turn out in fact to be the telegrapher’s equations as found in chapter 1 for the classical transmission
line. Indeed, we can take the voltage of forward mode of the classical transmission line corresponding to
this TEM mode to be
V (z) = Φ1 e−jkz (8.29)
while the current of this mode is
I(z) = I1 e−jkz (8.30)
Take the dot product of (8.3) with un on C1 , and integrate over C1 . From (8.24) and (8.26), the result
can be written as:
−γI1 = −jωQl1 (8.31)
which is a statement of conservation of surface charge on C1 . Likewise, taking the dot product of (8.1)
with un on the path between points g and 1 in Fig. 8.6 and integrating along the path, the result is
(using (8.21) and (8.25)):
−γΦ1 = −jωΨl (8.32)
Relating Ql1 to Φ1 using (8.23) and Ψl to I1 using (8.28), and noting that −γ corresponds to ∂/∂z
for this forward mode, we see that (8.31) and (8.32) are precisely the telegrapher’s equations studied in

241
chapter 1. In particular, having computed l and c for a given configuration of a uniform TEM mode
structure, we have that √
γ = jω lc (8.33)
and r
l
Zc = (8.34)
c
For the TEM mode, however, we have already seen that γ is determined by the properties of the
filling medium of the waveguide, eqn. (8.5). Comparison with (8.33) thus shows that the line parameters
are intimately related in this case:

lc = µǫ TEM modes only (8.35)

We see that the distributed inductance and distributed capacitance are not independent for TEM lines.
There are thus several alternative expressions for Zc in terms of l, c, and the product µǫ of the medium
parameters:
√
µǫ l
Zc = = √ TEM modes only (8.36)
c µǫ

8.2.1 Power, energy and normalization of TEM modes

The result of problem p8-2 shows that the complex power carried by a TEM mode can also be expressed
in terms of the voltage and current associated with the mode:

1
P̂ = Φ1 I1∗ (TEM modes only) (8.37)
2
and the oscillatory power is equal to

1
P̂osc = Φ1 I1 (TEM modes only)
2
Since for a forward TEM mode Φ1 = Zc I1 , it can be seen that the normalization condition (5.27) is
equivalent to setting the voltage and current to the particular values:
p p
Φ1 = 2Zc ; I1 = 2/Zc (8.38)

for a TEM mode.

The line parameters l and c of a TEM mode can alternatively be given in terms of energy, rather
than the forms presented above. For the electrostatic problem, we can write

Q2l1
I
= Φ1 Ql1 = Φ1 QS1 (ℓ) dℓ
c C1
I
= ΦǫET · un dℓ
C1 +Cg
Z
= − ∇T · (ΦǫET ) dS
S0
Z
= − ∇T Φ · ǫET dS
S0
Z
= ǫET2 dS
S0

242
where we have used (8.9) and (D.6), and ET2 = ET · ET . Hence as an alternative to (8.23), we could also
have defined c without explicit reference to the voltage Φ1 , but using only the linear charge density and
the energy per unit length2 stored in the electric field:
1 1
Z
= 2 ǫE 2 dS (8.39)
c Ql1 S0 T

Similarly, the line inductance can also be expressed without explicit reference to the flux:
1
Z
l= 2 µH2T dS (8.40)
I1 S0

where H2T = HT · HT . Since

µH2T = ǫET2 (8.41)
by (8.7), we have that
Q2l1
= lI12 (8.42)
c
Then by (8.31) we obtain (8.33) without the need to utilize information about the voltage or magnetic
flux. On the other hand, using (8.23) and (8.28), we could also obtain energy forms which depend on
Φ1 and Ψl , but not explicitly upon the linear charge density or the total current. These energy forms
have application to the calculation of wall losses using the incremental inductance rule (see section 9.5.2
and Appendix M), and forms of a similar type will prove crucial to the modeling of quasi-TEM and
pseudo-TEM modes later on.

8.2.2 Dielectric losses of TEM modes

It should be noted that if ǫ is complex, the quantity defined in (8.23) will also be complex, and represents
c − jg/ω instead of c. Likewise, a complex value of µ implies that (8.28) is equal to l − jr/ω. In the
case when µ and ǫ are constants, putting ǫ̂ = ǫ′ − jǫ′′ , we have from (8.5) and (8.6) that
p
γTEM = αTEM + jβTEM = jω µ(ǫ′ − jǫ′′ ) (8.43)

where "p #1/2

√ (ǫ′ )2 + (ǫ′′ )2 − ǫ′
αTEM =ω µ (8.44)
2
and "p #1/2
√ (ǫ′ )2 + (ǫ′′ )2 + ǫ′
βTEM =ω µ (8.45)
2
This, of course, is the same complex propagation coefficient we would get for a plane wave in the lossy
2
dielectric. It is identical to (5.106)-(5.108) in the special case kc,m = 0. Note that if ǫ′′ ≪ ǫ′ , (8.44) and
(8.45) become approximately
ωǫ′′ µ
r
αTEM ≃ (8.46)
2 ǫ′
s
µǫ′
βTEM ≃ k0 (8.47)
µ0 ǫ 0

2 Since there is no complex conjugate taken in the following expressions, the surface integrals below are proportional to

the oscillatory energy, rather than the time-average energy. For lossless transmission lines, these are essentially the same
thing.

243
 xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
Cg
b
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
a
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
C1
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
µ, ε
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Figure 8.7: The coaxial transmission line.

8.3 Some TEM Structures

In this section, we will give some examples of transmission lines supporting TEM waves, and formulas
for their characteristic impedances. In all these examples, µ and ǫ are taken to be independent of x and
y.

8.3.1 The coaxial transmission line

Figure 8.7 shows a familiar TEM structure, the coaxial transmission line. The field solutions for this
structure are well known:
Φ1
E = uρ (8.48)
ρ ln(b/a)
Φ1
H = uφ (8.49)
ζρ ln(b/a)

where ζ = (µ/ǫ)1/2 , and a and b are the inner and outer radii of the line, respectively. From (8.49),

2πΦ1
I1 =
ζ ln(b/a)

and hence,
ζ b
Zc = ln (8.50)
2π a
To normalize the mode field, we must determine the amplitude constant Φ1 which enters into these
equations. We have
2π b
1 1 Φ21 ρ dρ dφ
Z Z Z
E1 × H1 · az dS =
2 S0 2 0 a ζρ2 [ln(b/a)]2
πΦ21
= (8.51)
ζ ln(b/a)

244
 Figure 8.8: The two-wire transmission line.

According to (5.27), this must be equal to 1, and we find that

r
ζ p
Φ1 = ln(b/a) = 2Zc (8.52)
π

in accordance with (8.38). We could equally well have chosen Φ1 to be the negative of this value, but
this would not be in agreement with the conventional definition of voltage and current polarity for such
a transmission line (inner conductor is positive).

8.3.2 Two-wire transmission line

Another familiar example is the two-wire transmission line shown in Fig. 8.8. The wires are identical,
with radii a and their centers are separated by a distance b = 2h as shown. The fields of the TEM mode
here are also well-known:
√   
2Zc x − d/2 x + d/2
E = ux −
2 ln[(b + d)/2a] (x − d/2)2 + y 2 (x + d/2)2 + y 2
 
y y
+ uy − (8.53)
(x − d/2)2 + y 2 (x + d/2)2 + y 2

√   
2Zc x − d/2 x + d/2
H = uy −
2ζ ln[(b + d)/2a] (x − d/2)2 + y 2 (x + d/2)2 + y 2
 
y y
− ux 2 2
− (8.54)
(x − d/2) + y (x + d/2)2 + y 2

where p
d= b2 − 4a2 (8.55)
Integration of the magnetic field to obtain the current I results in the following expression for the
characteristic impedance:
ζ b+d ζ b
Zc = ln = cosh−1 (8.56)
π 2a π 2a

245
8.3.3 Open stripline
Another example where the region between the two conductors extends to infinity is the open stripline
of Figure 8.9. An infinitely thin metal strip of width w is suspended at a height h above a ground plane.

Figure 8.9: Open stripline.

Unlike the previous two examples, the potentials and fields for even this most simple of the striplines
cannot be written down in closed form. Various approximations can be made, or numerical solutions
attempted (see the references at the end of the chapter), but we will not pursue any detailed derivation
here. Let us rather present some results, and concentrate on their physical importance.
If the strip were wide compared to its height above the ground plane (w ≫ h), we would expect
ET to look as illustrated in Figure 8.10. For a given voltage Φ1 between the strip and ground plane,

Figure 8.10: Field lines for wide stripline.

the surface charge on the strip is mostly on the bottom side in this case, and nearly uniform except
very near the edges. Here, the field fringes, disturbing the uniformity which prevails near the center
of the strip, and causing there to be more surface charge near the edges than would be the case if the
field were assumed uniform and vertical there. The neglect of the fringing fields is commonly made in
elementary discussions of parallel-plate capacitors, and corresponds to the field shown in Figure 8.11.
The capacitance per unit length in this parallel-plate approximation is well-known:

Figure 8.11: Stripline fields, neglecting fringing.

w
cpp = ǫ (8.57)
h
However, even for very wide strips, this simple model can cause a significant error in the value of the
capacitance: nearly 30% when w/h = 10. Moreover, for narrow strips, the model completely breaks
down.
When the strip is narrow (w ≪ h), the field ET behaves as in Figure 8.12(a). It is quite similar

246
 Figure 8.12: Field lines for (a) narrow stripline and (b) wire above ground plane.

to that of a wire of radius a located at the same height h above a ground plane, Figure 8.12(b). The
capacitance of the latter is well-known (related by image theory to that of the two-wire line considered
above):
2πǫ
cwire = (8.58)
ln(2h/a)
for a ≪ h. If a circular wire were chosen with a radius such that the charge on it were the same as that
on the strip of Figure 8.12(a) (for the same voltage Φ1 on each), we could use (8.58) to approximate the
capacitance of this narrow strip. In fact, it has been found that a thin circular wire of equivalent radius:

aeq = w/4 (8.59)

does support the same charge as the strip for a given voltage under these conditions. The concept
of equivalence between thin conductors of various shapes and circular wires is widely encountered in
electromagnetics, and enables us to use design information for transmission lines or antennas made up
of circular wires for the analysis of structures using conductors of arbitrary cross-section. The concept
is applicable whenever the dimensions of the conductor cross-section are small by comparison with the
distance of the conductor from any other object. Equivalent radii for conductors of other cross-sections
are given in the references cited at the end of this chapter.
The following formula has been found to give c for the stripline of Figure 8.9 for any value of w/h,
with an error of less than 1%:
4πǫ
c≃   q  (8.60)
h h h 2 π 2
ln 1 + 32 w w + ( w ) + ( 8 )

This expression has the advantage of being “reversible”, i.e., it can be solved for the ratio w/h in terms
of c, allowing design of a stripline with a desired value of Zc :
p
w 4 π 2 + 4[exp(4πǫ/c) − 1]
=
h exp(4πǫ/c) − 1

8.4 Distributed-Circuit Model of Non-TEM Modes

The transmission lines illustrated in Figures 8.2 and 8.3 do not fall into the class of waveguides we
have analyzed in the previous sections, but have instead arbitrary inhomogeneous variations of ǫ in x
and y. Some types of planar (and, for that matter, nonplanar) transmission lines involve the use of

247
ferrite substrates which introduce variations of µ as well. We have seen that a TEM mode can exist on
structures of this type only if µǫ is constant throughout the cross-section.
A mode of a two-conductor line that is not TEM can nevertheless still be described in distributed-
circuit terms similar to those we used for the TEM mode, with several important differences.3 Let us
examine Maxwell’s equations (5.17)-(5.20) to see how they imply different behavior for the fields than
the satisfaction of (8.9) and (8.10) implied for TEM mode fields. For the moment, we will continue to
assume that the boundaries C1 and Cg are perfect conductors. Eqns. (5.18) and (5.20) provide the most
natural generalizations of the second parts (the curl equations) of (8.9) and (8.10) for non-TEM modes:

∇T × ET = −jωµuz Hz (8.61)

∇T × HT = jωǫuz Ez (8.62)
The algebraic relationship between ET and HT now involves the longitudinal field components. An
appropriate generalization of (8.1) and (8.3) is obtained from (5.21) and taking uz × (5.19):

∇T Hz = −γHT + jωǫuz × ET (8.63)

∇T Ez = −γET − jωµuz × HT (8.64)

Taking the transverse curl of (8.63) and (8.64), and then using (8.61)-(8.62) together with some vector
identities yields generalizations of the first parts (the divergence equations) of (8.9) and (8.10) for non-
TEM modes:
∇T · (ǫET ) = γǫEz ; ∇T · (µHT ) = γµHz (8.65)
but it should be kept in mind that these are consequences of (8.63)-(8.64), and not independent equations.
Refer to the cross-section shown in Figure 8.13. We again regard C1 as the “positive” conductor
of the line (with respect to the “ground” Cg ). In contrast to the TEM case, we cannot define either a
potential difference Φ1 or a magnetic flux per unit length Ψl between C1 and Cg in a unique fashion.
For, suppose P is some path between C1 and Cg (Figure 8.13). We can define a voltage ΦP associated
with that path as Z
ΦP = − ET · dℓ (8.66)
P
in analogy with the definition for a TEM mode.
If we choose a second (different) path P ′ between C1 and Cg , then a surface SP P ′ will be enclosed
by P , P ′ and segments of the conductors C1 and Cg as shown in Figure 8.14(a). If we integrate ET · dℓ
around the closed path bounding SP P ′ , use Stokes’ theorem, and note that ET · uℓ = 0 on C1 and Cg ,
we obtain Z
ΦP − ΦP ′ = uz · ∇T × ET dS (8.67)
SP P ′

or, from (8.61), Z

ΦP ′ = ΦP + jω µHz dS (8.68)
SP P ′

i. e., ΦP ′ differs from ΦP by jω times the longitudinal magnetic flux passing between P and P ′ . In an
analogous fashion, we can integrate the second of equations (8.65) over SP P ′ and use the two-dimensional
divergence theorem to obtain Z
ΨP ′ = ΨP + γ µHz dS (8.69)
SP P ′

where Z
ΨP = µHT · un dℓ (8.70)
P
3 The following development is relatively general, and can be adapted (as is done for example in problem p8-14) to

apply to virtually any mode of a waveguide with a metallic boundary—even a single-piece boundary.

248
 xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
Cg
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx

xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
C1
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx

1 xxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxx
C

ul xxxxxxxxxxx P
S0 un xxxxxxxxxxx
xxxxxxxxxxx

un ul
ul ul
un
un
µ(x,y), y
ε(x,y) g

z x

Figure 8.13: Cross-section of a non-TEM transmission line.

Cg
Cg

C1 C1

P' C
ul
P ul u SCC'
n
SPP' un
y
C'
z
x

(a) (b)

Figure 8.14: (a) Two paths P and P ′ , and the surface SP P ′ bounded by them. (b) Two contours C and
C ′ , and the surface SCC ′ bounded by them.

249
is the magnetic flux per unit length past P (which is likewise dependent on the path P ).
Thus, we fix some choice for the path P in order to define the voltage and flux unambiguously. These
can then be related in the same way as for TEM modes. We proceed by integrating (8.64) along P :
Z Z Z
∇T Ez · dℓ + γ ET · dℓ = −jω (uz × µHT ) · dℓ
P P
ZP
= −jω µHT · un dℓ (8.71)
P

where un is the unit vector normal to P . The first integral on the left side of (8.71) is equal to
Ez |C1 − Ez |Cg = 0. The second term, by (8.66), is equal to −ΦP , while the integral on the right side is
ΨP , so we have:
−γΦP = −jωΨP (8.72)
similar to (8.32) in the TEM case. This relation holds true no matter what path P is chosen.
Similarly, we can compute the conduction current I1 on C1 by (8.26) once again, and the linear free
charge density Ql1 by (8.24). Now, however, that because (8.9) and (8.10) no longer hold, the current
and charge are no longer the same if we use an arbitrary contour C between C1 and Cg to compute
them, in contrast to the situation for a TEM mode. In fact, let us denote
I
QlC ≡ ǫEn dℓ (8.73)
C

and I
IC = HT · dℓ (8.74)
C
Now consider any two possible closed contours C and C ′ such that C is outside of C1 but inside of C ′ ,
which in turn is inside of Cg (Figure 8.14(b)). Integrating the first of (8.65) over the portion of the
cross-section SCC ′ between C and C ′ , and using the two-dimensional divergence theorem gives
Z
QlC ′ = QlC + γ ǫEz dS (8.75)
SCC ′

Likewise, doing the same for the z-component of (8.62) and using Stokes’ theorem gives
Z
IC ′ = IC + jω ǫEz dS (8.76)
SCC ′

Thus, just as the voltage and flux can depend on the path for a non-TEM mode, the current and charge
lose something of their unambiguous character as well—they now depend on which contour is used to
define them. In particular, the current IC ′ differs from IC by the longitudinal displacement current
through the portion of the cross section between the two contours.
Finally, the charge and current on the ground conductor Cg are no longer necessarily the negatives
of those on C1 . We have Z
Qlg = −Ql1 − γ ǫEz dS (8.77)
S
while doing the same for the z-component of (8.62) and using Stokes’ theorem gives
Z
Ig = −I1 − jω ǫEz dS (8.78)
S

We then choose some suitable contour C and integrate (8.63) around it, giving
I I I
∇T Hz · dℓ + γ HT · dℓ = jω (uz × ǫET ) · dℓ
C C
IC
= jω ǫET · un dℓ (8.79)
C

250
The first integral vanishes because Hz is single-valued; the second is equal to IC by (8.74). The integral
on the right is QlC by (8.73), so that
−γIC = −jωQlC (8.80)
which reduces to (8.31) in the TEM case. Note that (8.80) holds for an arbitrary choice of the contour
C, although most frequently it is chosen to be C1 .
As mentioned above, a mode that is not TEM does not have fields which are found exactly from
(8.9) and (8.10), but from the full, coupled Maxwell’s equations (5.17)-(5.20) (or (8.61)-(8.64)). To make
(8.80) and (8.72) fully correspond to the telegrapher’s equations, we have some freedom as to how to
proceed. One way is to define inductance and capacitance per unit length in terms of oscillatory field
energies and the current and linear charge density on C, in analogy with (8.39) and (8.40):

1 1
Z
ǫET2 + µHz2 dS


= 2 (8.81)
cC (ω, γ) QlC S0

and
1
Z
µH2T + ǫEz2 dS


lC (ω, γ) = 2 (8.82)
IC S0

In these equations, we have used the form of oscillatory energy from (K.8) that most closely resembles
its form for the TEM case. It can be judged that there is a certain natural connection between the pairs
of fields (ET , Hz ) and (HT , Ez ) because of (8.61) and (8.62). If the longitudinal fields are small (as will
be the case for the quasi-TEM modes discussed in the next section), (8.81) and (8.82) reduce to (8.39)
and (8.40) from the TEM case. Since the two forms of oscillatory energy used to define lC and cC above
are equal: Z Z
ǫET2 + µHz2 dS = µH2T + ǫEz2 dS



S0 S0

then from (8.81), (8.82) and (8.80) can be obtained the result of classical transmission line theory:
p
γ = ±jω lC (ω, γ)cC (ω, γ) (8.83)

An alternative method based on the field energies for characterizing non-TEM modes is to use the
path-dependent voltage and magnetic flux together with the oscillatory energy to define path-dependent
line capacitance and inductance as follows:
1
Z
ǫET2 + µHz2 dS


cP (ω, γ) = 2 (8.84)
ΦP S 0

and
1 1
Z
µH2T + ǫEz2 dS


= 2 (8.85)
lP (ω, γ) ΨP S0

Again, we obtain from (8.84), (8.85) and (8.72) the result of classical transmission line theory, in the
form p
γ = ±jω lP (ω, γ)cP (ω, γ) (8.86)
Although in general cP (ω, γ) 6= cC (ω, γ) and lP (ω, γ) 6= lC (ω, γ), the values of γ obtained from (8.86)
and (8.83) must be the same.
Let us pause to note several things at this point. We will use the notations l(ω, γ) and c(ω, γ) to
refer either to the set lC (ω, γ) and cC (ω, γ) or lP (ω, γ) and cP (ω, γ) as might be applicable. Since our
mode is not a TEM mode, we cannot immediately calculate γ without knowing l(ω, γ) and c(ω, γ) since
(8.1) and (8.3), and therefore also (8.5) do not hold.4 As a matter of fact, equations (8.61) through
4 Equation (8.86) does imply though that the product l c must be independent of the path P just as (8.83) implies
P P
that the product lC cC must be independent of the contour C.

251
(8.86) might well apply to essentially any mode of this waveguide, or any other waveguide for which
appropriate definitions for ΦP and IC can be made. If a mode below cutoff were under consideration,
either l(ω, γ) or c(ω, γ) would have to be negative. Since l and c may depend on γ, equations (8.83) and
(8.86) are generally only implicit equations which must be solved to determine γ. An example of how
complicated this can make things will be seen in the example of the slotline, later on in this chapter.
If the conductors C1 and Cg are not perfect, but instead have a finite conductivity, the preceding
discussion has to change somewhat. In particular, even though we can still define a path-associated
voltage (8.66), the following derivation (8.67)-(8.72) is no longer true, because it assumed that the
tangential components of E were zero on the surfaces of the conductors. However, the derivation (8.73)-
(8.83) remains valid with only minor changes, as we discuss below. In addition to allowing imperfect
conductors, let us also assume that there may be losses in the region between the conductors. Both
conditions are accounted for by allowing the permittivity ǫ and/or the permeability µ to be complex
functions of position, including “inside” the conductors C1 and Cg . The derivation (8.73)-(8.83) remains
true if we extend the surface integrals to include also the cross-sections S1 and Sg of the conductors, and
if the line parameters are modified to include shunt conductance and series resistance per unit length:
1 1
Z
ǫET2 + µHz2 dS


= 2 (8.87)
jωcC (ω, γ) + gC (ω, γ) jωQlC S0 +S1 +Sg

and
jω
Z
µH2T + ǫEz2 dS


jωlC (ω, γ) + rC (ω, γ) = 2 (8.88)
IC S0 +S1 +Sg

whence the propagation constant obeys

p
γ = ± [jωlC (ω, γ) + rC (ω, γ)][jωcC (ω, γ) + gC (ω, γ)] (8.89)

Once again, the results agree formally with those of classical transmission line theory.

8.5 Quasi-TEM Modes

8.5.1 “Approximately” TEM modes and the quasi-TEM limit
We must admit that even for a two-conductor transmission line with an inhomogeneously filled cross
section, in the static limit (ω → 0, γ → 0), there will be nontrivial solutions ET and HT to the field
equations. These are the static, purely transverse fields of z-independent charge and longitudinal current
distributions on the conductors of the line, like those of true TEM modes. It seems reasonable to expect
that at “sufficiently low” frequencies some kind of “almost” TEM mode could exist, for which Ez and
Hz were “small” compared to the corresponding transverse fields.
To say more precisely what “small” frequency means, call L a typical length over which the field
varies significantly in the transverse direction (roughly speaking, this can be thought of as one of the
transverse dimensions of the waveguide, e. g., its diameter). Then operating on a field with the transverse
gradient, curl or divergence operators will produce a quantity which is of the order of magnitude of that
√
field divided by L. If γ can be thought of as being the same order of magnitude as k ≡ ω µǫ, then if
kL ≪ 1, (8.63)-(8.64) provides the estimates

Hz ∼ |HT |O(kL) + |ET /ζ|O(kL)

Ez ∼ |ET |O(kL) + |HT ζ|O(kL)

p
where ζ = µ/ǫ is the local wave impedance, and O(kL) denotes terms which are no larger in magnitude
than some constant times kL. The right sides of (8.61) and (8.62) are then O(k 2 L2 ) smaller than the left
sides, and can be set approximately equal to zero. We then revert to eqns. (8.9) and (8.10). Physically,

252
this quasi-TEM approximation means that the transverse dimensions of the structure must be small
compared to the wavelength of a plane wave in whatever material may constitute the waveguide structure.
The transverse fields will then approximately satisfy (8.9) and (8.10) with µ and ǫ now functions of x
and y, but not (8.1) or (8.3) any longer ((8.63)-(8.64) hold instead, and approximately determine Ez and
Hz ).
The most important application of the formalism of section 8.4 is to the quasi-TEM mode of the
transmission line which approaches the static, two-dimensional field configuration for small values of
kL. In many such situations, the difference between ΦP and ΦP ′ for any two paths P and P ′ becomes
negligible because of (8.68) and the smallness of ω and Hz , and ΦP approaches a unique frequency-
independent value Φ1 , obtained from the static field ET 0 satisfying the equations (8.9). Likewise, it
becomes possible to replace ΨP by the frequency-independent value Ψl , obtained from the static field
HT 0 that satisfies (8.10). As a result, lP and cP become independent of both ω and γ. In the same
way, the difference between IC and IC′ for any two contours C and C ′ becomes negligible under these
conditions because of (8.76). We can thus replace IC by the (approximately) frequency-independent
value I1 obtained from the magnetostatic field HT 0 , and analogously QlC → Ql1 . The definitions of
lC (ω, γ) and cC (ω, γ) become independent of C, and also approach constants as frequency approaches
zero. In this limit, we say that the mode is a quasi-TEM mode, and we can put

lC (ω, γ) ≃ lP (ω, γ) ≃ l
cC (ω, γ) ≃ cP (ω, γ) ≃ c (8.90)

where l and c are the static inductance and capacitance per unit length of line computed just as they
were in the TEM case, from (8.9), (8.10), (8.23) and (8.28), µ and ǫ now being regarded as functions
of position (x, y) in the cross section.5 In the case of a loss-free quasi-TEM mode, then, we have from
(8.83) or (8.86) that √
γ = jβ ≃ jω lc (8.91)
for a forward mode. We emphasize that β cannot now generally be expressed directly in terms of
particular values of µ and ǫ of the line in any simple way. It is, however, possible to show that the value
of the product lc will lie somewhere between the minimum and maximum values of the product µǫ over
the cross-section of the line (cf. Appendix K). The characteristic impedance is given in the quasi-TEM
regime by: p
Zc = l/c (8.92)
Here, Zc may be used in all the same ways as for TEM modes and thus the entire body of transmission-
line analysis may be applied here just as to a TEM line. It is important to remember, though, that only
(8.92), and none of the other alternative forms in (8.36) can be used to calculate Zc when the mode is
not pure TEM.
In the case when the conductors C1 and Cg are not perfect, or there are losses in the material filling
the waveguide cross-section, we include also the line parameters gC (ω, γ) and rC (ω, γ) as indicated at
the end of section 8.4. For sufficiently low frequencies, g and r become independent of C and of γ,
though not generally independent of ω (especially in the case of r, due to skin effect—see section 9.5).
If the conductivity of the conductors is high, ET inside them will be very small, while Ez , while also
small, will contribute to a significant longitudinal current density, which is the major contribution to r
for quasi-TEM modes. In this book, we will only evaluate these effects of loss by perturbation methods
(sections 9.4-9.6), but in some cases a full solution of Maxwell’s equations is necessary to accurately
model them.

5 Again, for certain configurations such as the slotline, l and c may fail to approach finite nonzero limits. In such cases,

it is necessary to refine our treatment somewhat to calculate quantities l(ω, γ) and c(ω, γ) which do not approach as static
limit. Such modes will be called pseudo-TEM modes.

253
 εrε0

xxxxxxxxxxxx
xxxxxxxxxxxx
xxxxxxxxxxxx
C1

ε0

xxxxxxxx
xxxxxxxx
xxxxxxxx
xxxxxxxx
xxxxxxxx
xxxxxxxx
xxxxxxxx
xxxxxxxx
Cg

Figure 8.15: Transmission line containing two homogeneous dielectric regions.

8.5.2 Quasi-TEM modes on lines having air and one other dielectric
Many important quasi-TEM lines have a dielectric which is piecewise constant. Indeed, the dielectric
constant is homogeneous (ǫ = ǫr ǫ0 ) within a substrate region, and the remainder of the cross section is
air (ǫ = ǫ0 ), as shown in Fig. 8.15. Suppose that the entire waveguide is made of nonmagnetic material
(µ = µ0 ). Let us emphasize the dependence of c on the relative permittivity ǫr of the “substrate” region
by writing:
c = C(ǫr ) (8.93)
Evidently, C(1) is the capacitance of the corresponding TEM mode structure obtained from the same
configuration of conductors C1 and Cg , but uniformly filled with air. The inductance l of the original
quasi-TEM line must be the same as that of the air-filled line, since ǫ has no effect on the quasi-TEM
inductance, which is completely determined by the solution of (8.10) which involves µ but not ǫ. Then,
from (8.35) we have
µ0 ǫ 0
l= (8.94)
C(1)
and (8.91) and (8.92) can be expressed in the forms
s
C(ǫr )
β ≃ k0 (8.95)
C(1)

and s
C(1)
Zc (ǫr ) = Zc (1) (8.96)
C(ǫr )
√
for quasi-TEM modes, where k0 = ω µ0 ǫ0 and
√
µ0 ǫ 0
Zc (1) =
C(1)

254
is the characteristic impedance of the corresponding air-filled line.
These expressions suggest the definition of an effective dielectric constant ǫeff for quasi-TEM modes
according to
C(ǫr )
ǫeff = (8.97)
C(1)
so that
√
β ≃ k0 ǫeff (8.98)
and
√
Zc (ǫr ) = Zc (1)/ ǫeff (8.99)
The physical significance of ǫeff is thus that a line homogeneously filled with a medium of relative
permittivity ǫeff and permeability µ0 would have the same β and Zc as given by (8.98) and (8.99). It
therefore suffices for us to obtain an expression for C(ǫr ) as a function of ǫr to completely characterize
the mode under quasi-TEM conditions.

8.6 Example: Open Microstrip

Consider the open microstrip whose cross-section is shown in Figure 8.16. It is identical to the open
stripline of Figure 8.9, except that the region 0 < y < h is occupied by a substrate of relative dielectric

µ0
ε0 y

w
µ0
h εrε0
x

Figure 8.16: Open microstrip.

constant ǫr . The region above the strip is taken to be air, and all regions are assumed to be non-magnetic.
The line is thus of the two-dielectric type studied in the previous section.
As was the case with open stripline, an exact formula for C(ǫr ) for microstrip cannot be given. It is
possible to give one that is valid for arbitrary values of w/h with an accuracy of 2% or better:

2πǫ0 (ǫr + 1)
Cmicr (ǫr ) ≃ ( " r  #) (8.100)
2 2  2
32h h 2 hr π(ǫ r +1)
ln 1 + w w rǫ + w
ǫ
+ 16ǫr

where by (7.42) s
0.5160 0.0788
rǫ ≃ 0.4052 + +
ǫr ǫ2r

In conjunction with (8.96) and (8.97), equation (8.100) can be used to calculate Zc and ǫeff . Since
the use of (8.100) in (8.96) results in a somewhat messy expression, the following formula is more often

255
used to find Zc :
  s 
r  2
ζ0 2  32h  h hrǫ ǫr + 1  π 2 
Zc ≃ ln 1 + rǫ + + (8.101)
4π ǫr + 1  w w w 2ǫr 8 

This formula is reversible: It can be solved for w/h in terms of Zc , and used to design a microstrip with
a desired characteristic impedance:
s
w 1 8π 2 (ǫr + 1)
= + 64Krǫ (8.102)
h K ǫr

where " r #
4πZc ǫr + 1
K = exp −1
ζ0 2

Some examination of these formulas shows that C(ǫr ) > C(1) when ǫr > 1. This is due to the fact
that a portion of the field now resides in a region of higher dielectric constant (the fields also rearrange
themselves slightly by comparison to the air-filled line, but this is a relatively smaller effect). It is thus
possible to achieve lower values of characteristic impedance (say, 50 Ω) using narrower strips than in the
case of an air-filled stripline by using an electrically dense (ǫr ∼ 10) substrate such as alumina. Typical
values of Zc and ǫeff are shown in Fig. 8.17.
The charge and current distributions are present on both the top and bottom sides of the strip, with
more density on the bottom side the wider the strip. The density is often given as a total density, adding
the densities from the top and bottom sides at a given point x. The total current density can be given
quite accurately by:
const w
Jz1 (x) ≃ q (|x| < ) (8.103)
cosh2 ( πw ) − cosh2 ( πx ) 2
8h 4h

where x = 0 is the center of the strip, and the constant is chosen to make the total current on the strip
equal to I1 . This current distribution is singular at the edges of the strip (x = ±w/2) because the strip
has been assumed to be infinitely thin, and to have perfectly sharp edges. Plots of (8.103) are given
in Figure 8.18 for a very wide and a very narrow strip. It is evident that for wide strips, the current
is very uniformly distributed (except near the edges), as would be expected from a “parallel-plate”
approximation that neglects fringing of the fields. The increase in current density near the strip edges
causes the attenuation of the mode due to the finite conductivity of the strip metal to be larger than if
the current were uniformly distributed. This point will be examined in more detail in section 9.5.
An accurate approximation to the charge distribution on open microstrip has been found to be:
const
QS1 (x) ≃ r     (|x| < w/2) (8.104)
cosh2 4(ǫπǫr +1)h
rw
− cosh2 2(ǫπǫ rx
r +1)h

where the constant is now chosen to make the total charge per unit length equal to QS1 . For wide strips
and large values of ǫr , the charge distribution will be somewhat flatter than the current distribution
(Figure 8.19). This demonstrates (from the principle of charge conservation) that at least some transverse
current density Jx1 (x) must also be present on the strip for this quasi-TEM mode.

256
 200

εr = 9.8
150

Zc, Ω
100

50

0
0 0.5 1 1.5 2
w/h

(a)

7.5

εr = 9.8

7
εeff

6.5

5.5
0 0.5 1 1.5 2
w/h

(b)

Figure 8.17: Quasi-TEM values for (a) characteristic impedance and (b) effective permittivity of mi-
crostrip.

257
 4

w/h = 0.04
3 w/h = 10
Jz1(x)
Jz1(0)
2

0
1 0.5 0 0.5 1
2x/w

Figure 8.18: Normalized total surface current density Jz1 (x) on open microstrip with µ = µ0 .

Jz1(x) QS1(x)
3 Jz1(0) QS1(0)

0
1 0.5 0 0.5 1
2x/w

Figure 8.19: Normalized charge and current distributions for open microstrip with ǫr = 10 and w/h = 10.

258
8.7 More Examples: Coplanar Lines
In this section, we present the quasi-TEM properties of two other commonly encountered planar trans-
mission lines. In all cases, the inductance l is given in terms of the expression for capacitance evaluated
at ǫr = 1, as in (8.94).

8.7.1 Open coplanar waveguide (CPW)

A further commonly used planar transmission line is the coplanar waveguide (CPW) shown in Fig-
ure 8.20. Its positive conductor is a strip on a dielectric substrate, like the microstrip, but unlike the
microstrip, its ground plane is in two parts and lies on the same side of the substrate as the strip. Thus,
CPW is formed from a substrate with a ground plane on one side into which two identical slots have

w s w

εr h

Figure 8.20: The open coplanar waveguide (CPW).

been cut, separated by a distance s. The mode of interest here has the two outer conductors at ground
potential, while the center conductor is the positive reference, resulting in the field distribution shown
in Figure 8.21. Coplanar waveguide combines the advantages of microstrip with additional ones: both

Figure 8.21: Transverse electric field distribution of the fundamental (quasi-TEM) mode in coplanar
waveguide.

series and shunt lumped elements can easily be connected to the line without disturbing the substrate.
As was the case for microstrip, exact expressions for the line parameters of CPW are not available.
The following formula for c has been found to track well with numerically calculated results for a wide
range of substrate thicknesses h, particularly if ǫr ≫ 1 (it is exact either for ǫr = 1 or as h → ∞):
 
K(k) ǫr − 1 K(k1 )
c = CCPW (ǫr ) = 4ǫ0 + (8.105)
K(k ′ ) 2 K(k1′ )
where K(k) is the complete elliptic integral of the first kind, and is discussed in Appendix B where
simple approximate formulas for it are given. The moduli k and k1 appearing in (8.105) are
s
k =
s + 2w
sinh(πs/4h)
k1 = (8.106)
sinh[π(s + 2w)/4h]

259
 √ p
while the complementary moduli are k ′ = 1 − k 2 and k1′ = 1 − k12 . As h → ∞, we have k1 → k, and
in this limit as well as ǫr → 1, the expression (8.105) becomes exact for the quasi-TEM case.
In this approximation, the dependence of c on ǫr is much simpler than was Cmicr (ǫr ). We have ǫeff
and Zc given by
ǫr − 1 K(k ′ ) K(k1 )
ǫeff = 1 + (8.107)
2 K(k) K(k1′ )
and
ζ0 K(k ′ )
Zc = √ (8.108)
4 ǫeff K(k)
based on (8.105). Some typical behavior of Zc and ǫeff vs. s/(s + 2w) is shown in Figure 8.22.

200

h/(s + 2w) = ∞
150
h/(s + 2w) = 0.25
h/(s + 2w) = 0.1
Zc, Ω

100

50
εr = 9.8

0
0 0.2 0.4 0.6 0.8 1
s/(s + 2w)

(a)

6
h/(s + 2w) = ∞

h/(s + 2w) = 0.25

εeff

3 h/(s + 2w) = 0.1

εr = 9.8
2
0 0.2 0.4 0.6 0.8 1
s/(s + 2w)
[h]
(b)

Figure 8.22: (a) Characteristic impedance of coplanar waveguide, and (b) effective dielectric constant of
coplanar waveguide or coplanar strip line.

Notice that since there are two ratios of length dimensions for us to adjust, we can specify both Zc
and ǫeff for a design with a given value of ǫr . This is in contrast to the cases of slotline or microstrip,

260
in which only one of these can be independently specified. Equations (B.57) and (B.59) of Appendix B
allow us to calculate the value of the modulus k given the value of the ratio of K(k ′ )/K(k). Thus,
we could proceed as follows. With Zc and ǫeff given, we find the value of K(k ′ )/K(k) and therefore k
from (8.108). The value of K(k1′ )/K(k1 ) and thus k1 is then found from (8.107). The value of s/w is
then found from the first of eqns. (8.106), and s/h or w/h from the second of these equations. In the
latter case, the equation is not explicitly solvable for the desired quantity, and some form of numerical
or approximate method (e. g., the iteration method outlined in Appendix C) must be used to obtain its
value.

8.7.2 Coplanar strip line (CPS)

A structure complementary to CPW which is sometimes used as a traveling-wave electrode structure for
electro-optic devices is the coplanar strip line (CPS) shown in Figure 8.23. Here again, both series and

w s w

εr h

Figure 8.23: Coplanar strip line (CPS).

shunt connections of lumped circuit elements may easily be made to this line.
Corresponding formulas to (8.105)-(8.108) for the CPW are also available for CPS. With the moduli
k and k1 again given by (8.106), we have
 
ǫr − 1 c0 K(k1 )
c = CCPS (ǫr ) = c0 1 + (8.109)
2 ǫ0 K(k1′ )
where c0 is the value of c when ǫr = 1:
K(k ′ )
c0 = CCPS (1) = ǫ0 (8.110)
K(k)
We can now give the formulas for ǫeff and Zc as
ǫr − 1 K(k ′ ) K(k1 )
ǫeff = 1 + (8.111)
2 K(k) K(k1′ )
(the same as for CPW), while
ζ0 K(k)
Zc = √ (8.112)
ǫeff K(k ′ )
Typical plots of Zc for CPS are shown in Fig. 8.24; reference can be made to Fig. 8.22(b) for values of
ǫeff .
Other design data and formulas for quasi-TEM planar transmission lines can be found in the refer-
ences at the end of the chapter.

8.8 Dispersion
When the frequency of operation increases beyond the quasi-TEM range, β begins to display a nonlinear
dependence on ω (that is ǫeff depends upon frequency). The effective dielectric constant ǫeff ≡ β 2 /k02

261
 400

h/(s + 2w) = ∞
h/(s + 2w) = 0.25
300
h/(s + 2w) = 0.1

Zc, Ω
200

100
εr = 9.8

0
0 0.2 0.4 0.6 0.8 1

s/(s + 2w)

Figure 8.24: Characteristic impedance of coplanar strip line.

can be shown to increase with frequency (see eqn. (K.71)). This phenomenon is called dispersion, and
is in fact an inherent property of non-TEM modes.6 The reason for this (in the case of microstrip,
for example) is that a larger fraction of the electric fields become concentrated in the substrate, hence
making the wave “see” more of the denser medium, and behave more like a plane wave would in that
medium. This is true quite generally for all waveguides, but the proof involves some subtlety and is left
as an exercise (problem p8-16). The nonlinear dependence of β on frequency causes distortion of pulses
transmitted along the line (see chapter 4), and so will impose a limit on the rate at which information
can be sent. It is thus important to be able to predict the variation of ǫeff with frequency.

8.8.1 Dispersion in microstrip

Actual computation or approximation of ǫeff as a function of ω can be carried out numerically. The
process can be rather sophisticated, but considerable headway has been made since about 1970. Typical
dispersion characteristics (ǫeff vs. frequency) for open microstrip are shown in Figure 8.25. It is clear
that for a given substrate permittivity ǫr and line aspect ratio w/h, a thinner substrate allows for lower
dispersion up to much higher frequencies.
There have also been efforts to provide empirical design formulas giving accurate approximations for
the dispersive behavior of open microstrip. Most accurate of existing formulas seems to be:
ǫr − ǫeff (0)
ǫeff (f ) = ǫr − (8.113)
1 + (f /fh )m
where ǫeff (0) is the zero-frequency limit of ǫeff , and
fT
fh = (8.114)
0.75 + (0.75 − 0.332/ǫ1.73
r )(w/h)
h q i
tan−1 ǫr ǫǫreff−ǫ(0)−1
eff (0)
fT = p (8.115)
2πh µ0 ǫ0 (ǫr − ǫeff (0))
6 This is evident from (K.69), wherein the inequality is strict whenever there are nonzero z-components of the field

present. When these field components are zero or negligible (as at low frequencies), dispersion will be absent, as in the
case of TEM or quasi-TEM modes. Dispersion also becomes negligible in the limit of very high frequency, since, as shown
by (K.66), ǫeff cannot increase beyond the upper limit of ǫr , in which case the z-components of the field once again become
small and the field tends to concentrate almost exclusively within the region of highest µǫ. We can say that a mode which
is essentially nondispersive at a given frequency is behaving as a plane wave, which shares these properties.

262
 9

8.5 εr = 9.7 h = 1.27 mm

εeff
7.5

7
h = 127 µm
6.5

6
0 5 10 15 20 25 30
f (GHz)

Figure 8.25: Effective dielectric constant as a function of frequency for open microstrip: µ = µ0 ; ǫr = 9.7;
w/h = 0.96.

m = m0 mc (8.116)
1 p
m0 = 1 + p + 0.32(1 + w/h)−3 (8.117)
1 + w/h

1.4
mc = 1+ [0.15 − 0.235 exp(−0.45f /fh)] (w/h ≤ 0.7)
1 + w/h
= 1 (w/h ≥ 0.7) (8.118)
The question of defining a characteristic impedance for a mode outside the quasi-TEM regime is
rather more problematical. Undoubtedly, one possible choice would be to set
s s
ΦP lC (ω, γ) lP (ω, γ)
ZV I = = = (8.119)
IC cP (ω, γ) cC (ω, γ)

as follows from (8.72), (8.80), (8.81)-(8.82) and (8.84)-(8.85), but this definition has a number of draw-
backs. First, it depends on the choice of the path P and contour C on which the voltage ΦP and current
IC are defined. Second, it uses a mixed choice of definitions for line capacitance and line inductance
(lC (ω, γ) and cP (ω, γ), for example), whose product does not directly result in the propagation coeffi-
cient. Finally, for modes in general, the complex power carried through the cross-section of the line is
not related to ΦP and IC as it is for TEM modes:
1 1
Z
P̂ = ET × H∗T · uz dS 6= (ΦP IC∗ ) (8.120)
2 S0 2
so, in general, we have
1
P̂ 6= ZV I |IC |2
2
Now this might suggest the definition
ZP I = 2P̂ /|IC |2 (8.121)

263
but in this case, what meaning can be attached to the “voltage”

ZP I IC ?

Generally, none. We are forced to admit that, except for quasi-TEM situations where voltage and
current are unambiguous concepts, the definition of Zc is fairly arbitrary, and is a matter of convenience
for analysis of concrete problems using the language of circuit theory. A model of equivalent transmission
lines is introduced in chapter 10 for any waveguide, and some value of Zc is needed to make the analogy
to a classical transmission line complete. We might find that a particular definition turns out to give
simpler equivalent circuits for waveguide discontinuities or excitation schemes than another. That must
be the eventual basis for choosing any definition for Zc , provided that we can be self-consistent in the
matter. However, it is only the connection with circuit theoretical concepts that requires Zc at all, and
in principle, we can bypass the need for it completely by dispensing with the notions of voltage and
current, and speaking only of the fields on the transmission line.

8.8.2 Dispersion in coplanar transmission lines

Dispersion likewise occurs in other quasi-TEM transmission lines. Details about these for several im-
portant cases can be found in the references cited at the end of the chapter. Here we present empirical
formulas for the dispersion of coplanar transmission lines. The formula for effective permittivity of either
CPW or CPS as a function of frequency is slightly different than (8.113):
√ p
p p ǫr − ǫeff (0)
ǫeff (f ) = ǫeff (0) + (8.122)
1 + (f /fh )m

in which the parameters are given by

m = −1.8 (8.123)
A0.56
fh = p (8.124)
4h µ0 ǫ0 (ǫr − 1)
h s i
A = exp u ln +v (8.125)
w
s h  s i2
u = 0.54 − 0.64 ln + 0.015 ln (8.126)
h h
s h  s i2
v = 0.43 − 0.86 ln + 0.54 ln (8.127)
h h

8.9 Pseudo-TEM Modes

Some waveguides which might appear at first glance to support a quasi-TEM mode turn out not to do
so, because of the failure of the line parameters l(ω, γ) and/or c(ω, γ) to approach a finite, nonzero limit
as ω → 0. The dominant modes of such waveguides often turn out to be sorts of hybrids between quasi-
TEM modes and surface waves or other waveguide mode types. Like surface-wave and closed waveguide
modes, these hybrids will be essentially dispersive at all frequencies. Nevertheless if certain important
dimensions of the guide cross section are small compared to a wavelength, it will still make sense to use
a formalism similar to that used for quasi-TEM modes for their description. We can call such modes
pseudo-TEM modes, several examples of which are shown in Figs. 8.26 and 8.27. In Fig. 8.28, we show
some structures whose fundamental mode is quasi-TEM at low frequencies, but becomes pseudo-TEM
when the frequency becomes large enough.
In Fig. 8.26, the waveguides have a small central conductor on which a current and linear charge
density can be defined fairly unambiguously, but where no unique voltage exists because the second

264
 ε, µ, σ
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
ε0xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxx Goubau xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
ε
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx Line xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Wire in Tunnel

ε0 ε0

ε ε
ε0 ε0
Single Conducting Strip Wire in Substrate
on Substrate

Figure 8.26: Waveguides supporting pseudo-TEM modes of quasi-TM type.

ε0

ε0 ε ε0 ε

ε0
Slotline

ε0

ε0
Antipodal Slotline

ε0 E Field

ε0

Finlines Higher-order CPW Mode

Figure 8.27: Waveguides supporting pseudo-TEM modes of quasi-TE type.

265
 ε0

ε0
ε0xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx Shielded
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
ε
xxxxxxxxxxxxxx Goubau Shielded Suspended Microstrip
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx Line

ε0

Wire over Grounded Substrate

Figure 8.28: Waveguides supporting modes that are quasi-TEM at low enough frequencies and pseudo-
TEM of quasi-TM type at sufficiently high frequencies.

conductor is electrically very far from the first or is nonexistent. The current on the central conductor
produces a predominantly TM field, and we call this type of pseudo-TEM mode quasi-TM. On the other
hand, in Fig. 8.27, the waveguides have a small gap between different parts of the conducting wall across
which a voltage and magnetic flux can be reasonably defined, but where no unique current or charge
density exists because there is only one piece of the conducting wall overall. The field of this kind of
mode is predominantly TE, and we call this type of pseudo-TEM mode quasi-TE. In the following we
consider one example of each type in some detail.

8.9.1 The Goubau line

The Goubau line is a perfectly conducting wire of radius a surrounded by a dielectric layer occupying
a < ρ < b (Fig. 8.29). This waveguide supports surface waves similar to those of the dielectric fiber
(cylinder) considered in section 6.7. Unlike the pure dielectric cylinder, however, there is (for some
modes at least) substantial current flow in the wire which can be connected to transmission lines as
a form of excitation and reception. This line can be used in antenna structures and as a microwave
transmission line that does not require a second (return) conductor.
Proceeding as in section 6.7, we consider the TM1 mode of this line. The azimuthally-independent
(∂/∂φ ≡ 0) electric field has the form Ez = C1 F0 (ρ), where

Y0 (h1 ρ)
F0 (ρ) = J0 (h1 ρ) − J0 (h1 a) (a < ρ < b)
Y0 (h1 a)
 
Y0 (h1 b) K0 (p2 ρ)
= J0 (h1 b) − J0 (h1 a) (ρ > b) (8.128)
Y0 (h1 a) K0 (p2 b)
p
Hφ and Eρ are obtained from (6.66) and (6.67) or (6.70) and (6.71). Here h1 = k02 ǫr − β 2 and
whilep
p2 = β 2 − k02 , and ǫr = ǫ/ǫ0 is the relative permittivity of the wire coating.
We now use these fields to compute cC and lC from (8.81) and (8.82). If the result is approximated

266
 2a

ε0xxxxxxxxxxxxx
xxxxxxxxxxxxx
xxxxxxxxxxxxx
xxxxxxxxxxxxx
xxxxxxxxxxxxx
xxxxxxxxxxxxx
xxxxxxxxxxxxx
xxxxxxxxxxxxx 2b
xxxxxxxxxxxxx
xxxxxxxxxxxxx
xxxxxxxxxxxxx ε
xxxxxxxxxxxxx
xxxxxxxxxxxxx
xxxxxxxxxxxxx

Figure 8.29: Goubau line.

for an electrically thin wire (h1 b ≪ 1 and p2 b ≪ 1), we arrive at

β2
 
µ0 2
l(ω, β) = ln γE − 2 (8.129)
2π e p2 a 2k0
 
1 1 1 b 2 1
= ln + ln γE − (8.130)
c(ω, β) 2πǫ0 ǫr a e p2 b 2
These are then inserted into (8.83) to get a transcendental equation for determining β:

β2 ln eγE2p2 a
= (8.131)
k02 1
ǫr ln ab + ln eγE2p2 b

Numerical
√ solutions of this equation are left to the reader as an exercise. A sample plot of ǫeff vs.
V = k0 b ǫr − 1 is shown in Fig. 8.30

8.9.2 The slot line

The slotline consists of a grounded substrate of thickness h the ground plane of which has been inter-
rupted by a slot of width w. Note that there is no conductor on the bottom of the substrate. The
slotline has the advantage that shunt elements can be easily connected across the line, a connection
which is quite cumbersome and destructive on microstrip. On the other hand, series connections are
troublesome because the current and charge densities on the conductors are spread out over a relatively
large distance in the transverse direction. Slotline has been employed mainly in short lengths for inter-
mediate transitions between other lines where impedance transformations or baluns are needed, or as
slot antennas.
If we attempt to carry out a DC calculation of l and c for this line, we find that l = 0 and c = ∞.
This is basically because any voltage difference between the conductors produces an infinite charge on
each of them, because they are both infinitely wide.7 The mode is thus a pseudo-TEM mode of quasi-TE
type. Choosing the path P for defining the voltage and magnetic flux of the slotline to be a straight
line across the slot, we compute lP (ω, β) and cP (ω, β) as described in section 8.4 for any mode of an
arbitrary waveguide.

7 If the conductors are very wide but finite, l is very small and c is very large, and because of the large dimensions

dispersion will begin to occur at much lower frequencies.

267
 2

1.8

1.6

εeff
1.4

1.2

1
0 0.2 0.4 0.6 0.8 1
V

Figure 8.30: Effective permittivity of a Goubau line: ǫr = 2.2, b/a = 2.5.

εr h

Figure 8.31: Slotline.

Unless w/h is impractically large (≥ 10), the value of cP (ω, β) can be expressed (accurate to a few
percent) in terms of the capacitance Cmicr (ǫr ) from (8.100) which describes the microstrip structure
having the same values of w and h:
!
2ǫ0 8 ǫ20 (ǫ2r − 1)
cP (ω, β) = Cslot (ǫr , ω, β) = ln p + (8.132)
π eγE w β 2 − k02 Cmicr (ǫr )

where γE refers to Euler’s constant (eγE = 1.78107 . . .), and k02 = ω 2 µ0 ǫ0 . The line inductance lP (ω, β)
is related to the line capacitance for ǫr = 1, just as is the case for quasi-TEM modes (eqn. (8.94)):
µ0 ǫ 0
lP (ω, β) = (8.133)
Cslot (1, ω, β)
Notice now that equation (8.83) for this line,
Cslot (ǫr , ω, β)
β 2 = ω 2 lP (ω, β)cP (ω, β) = k02 (8.134)
Cslot (1, ω, β)

268
is an implicit definition of β; we must solve for β and then define ǫeff for the slotline by

ǫeff = β 2 /k02 (8.135)

The solution of (8.134) can be carried out by graphical or numerical means. In particular, we may do
so by the iteration described in Appendix C, since the right side of (8.134) is a relatively slowly varying
function of the unknown β (it appears inside a logarithm). We choose an initial guess at β, call it β (0) ,
say, and substitute into the right side of (8.134). the resulting value is the square of a new estimate, β (1)
for the unknown, which can be substituted repeatedly in this way into the right side of (8.134). This
sequence of approximations converges rapidly to the value of β.
Once we know β, a version Zlc of the characteristic impedance can be computed from:
s √
lP (ω, β) √ ǫeff
Zc (ω) = = µ0 ǫ 0 (8.136)
cP (ω, β) Cslot (ǫr , ω, β)

Since the slotline possesses a fairly well-defined voltage, but an ill-defined current (especially at low
frequencies where the surface wave character of the mode becomes pronounced and the current den-
sity spreads out extensively over the conducting planes), the power-voltage definition of characteristic
impedance is also frequently used, and can differ considerably from Zlc .
Typical behaviors of ǫeff and Zlc vs. ω for a slotline are shown in Figure 8.32. We see that the effective
permittivity is rather sensitive to the ratio of slot width to substrate thickness, while the characteristic
impedance is much less dependent on this ratio, and is difficult to reduce below about 50 Ω in practical
situations.

269
 6
εr = 9.8
5
w/h = .5
εeff 4

w/h = 2
2

1
0 0.05 0.1 0.15 0.2
k 0w

(a)

150

w/h = 2

100
w/h = .5
Zc , Ω

50

εr = 9.8
0
0 0.05 0.1 0.15 0.2
k0w

(b)

Figure 8.32: (a) Effective dielectric constant for a slotline with ǫr = 9.8, w/h = 0.5 and 2 vs. normalized
frequency k0 w; (b) Zc for this slotline.

270
 Figure 8.33: (a) Open slotline; (b) slotline with ground plane.

8.10 Qualitative Analysis of Open Planar Waveguides

As we have done in section 6.9.2, as well as in section 7.4 for quasi-optical microstrip and in section 7.5
for dielectric rib waveguide, a qualitative analysis of the permitted range of β for bound (surface) modes
of open planar waveguides of general type can be carried out. Thus, as before, eqn. 6.119 is valid for the
bound modes of any open lossless waveguide. On the other hand, in an open planar guide, we obtain
lower bounds to β by identifying the species of wave which can propagate out to infinity and demand
that β be greater than the wavenumbers characterizing each of these species. Only this guarantees
exponentially decaying fields as we approach infinity in any transverse direction.8
Consider the examples of slotline, both with and without a ground plane as shown in Fig. 8.33. With
no ground plane, Fig. 8.33(a), both a space wave (wavenumber k0 ) and one or more surface waves of the
grounded slab (maximum wavenumber βs,TM0 ) can propagate to infinity. From this and from (6.119),
we find
√
βs,TM0 < β < k0 ǫr ,
which was also true of microstrip as shown in section 7.4. However, with a ground plane present
√
(Fig. 8.33(b)), there can now exist a TEM parallel-plate wave (wavenumber k0 ǫr ) going out to infinity
on this cross-section, and so we would have to have
√ √
k0 ǫr < β < k0 ǫr ,

which is impossible with strict inequalities. Hence, no bound modes can exist on slotline which has a
ground plane unless we can somehow arrange for the TEM parallel-plate wave constituent to be absent
from the total field.
Now, we ought to note that just because a particular wave species traveling to infinity might exist in
a waveguide cross-section does not necessarily mean that it will form a constituent of every mode which
that waveguide might support. Consider, for instance, the open symmetric finline shown in Fig. 8.34.
Due to the symmetry of this structure, it seems possible that a mode might exist having the electric
field lines shown in either Fig. 8.34(a) or(b). However, the fields of Fig. 8.34(b) are evidently settling
down to a TEM wave as we move into the parallel plate regions away from the gaps and so, according
to our earlier discussions, could not be a proper bound mode. On the other hand, the fields shown in
Fig. 8.34(a) have an opposite symmetry to those of Fig. 8.34(b) and can contain no TEM wave at all.
Therefore, for those modes of the finline which by symmetry contain no constituent TEM waves, there
may be bound modes which can exist such that
√
q
k02 ǫr − π 2 /d2 = β1 < β < k0 ǫr
8 We must qualify this statement for the case of a TEM mode on an open waveguide (the two-wire transmission line,

for example). Here, the propagation coefficient β is exactly equal to the wavenumber k0 of the space wave in the exterior
region. Thus, the field of a bound mode cannot have exponential decay in the transverse direction, but it is still possible
(as in the example cited) for the mode to decay algebraically at a sufficient rate that the mode carries finite power, and is
therefore a legitimate proper mode. The strict inequality β > k0 is therefore not necessary in such a case.

271
 Figure 8.34: Open symmetric finline: (a) proper mode; (b) improper mode.

Figure 8.35: Shielding and grounding schemes to eliminate leaky modes.

where β1 is the propagation coefficient of TE1 and TM1 parallel-plate waves.

Unfortunately, these kinds of modes can be rather unstable since small errors in dimensional toler-
ances can cause deviations from symmetry in an actual guide. This means in general that some TEM
wave will be present in practice, causing what would have been a purely bound mode to be “leaky.”
This will make it necessary even in such cases of apparent symmetry to use a shield or “grounding”
connections to assure the existence of the desired modes without leakage as in Fig. 8.35. Looked at from
one perspective, the shield serves the same purpose as a grounding bond: to assure that the upper and
lower plates are at the same potential, so that the TEM wave is suppressed. From another viewpoint,
no wave species can propagate to infinity in the transverse direction, and the only lower bound on β
is zero. In a case like this, however, any modes which do exist are not particularly well localized to
the neighborhood of the slot but have field distributions which tend to extend throughout the entire
cross-section within the shield. This loss of field localization may be a disadvantage in some applications.

272
8.11 Notes and References
The standard development of the theory of TEM modes can be found in
Johnk (1975), chapter 9;
Ramo, Whinnery and van Duzer (1984), chapter 8;
Johnson (1965), chapter 4;
Collin (1960), chapter 4.
Limiting expressions for stripline and microstrip capacitance as w/h → 0 or w/h → ∞ are to be
found in:
E.A. Shchapoval, “Capacitance, inductance and effective relative permittivity of microstrip
line”, Electron. Lett., vol. 11, pp. 225-226 (1975).
D.C. Chang and E.F. Kuester, “An analytic theory for narrow open microstrip”, Arch. Elek.
Übertragungstech., vol. 33, pp. 199-206 (1979).
E.F. Kuester and D.C. Chang, “Closed-form expressions for the current or charge distribution
on parallel strips or microstrip”, IEEE Trans. Micr. Theory Tech., vol. 28, pp. 254-259
(1980).
W.C. Chew and J.A. Kong, “Asymptotic formula for the capacitance of two oppositely
charged discs”, Math. Proc. Camb. Phil. Soc., vol. 89, pp. 373-384 (1981).
S.Y. Poh, W.C. Chew and J.A. Kong, “Approximate formulas for the line capacitance and
characteristic impedance of microstrip line”, IEEE Trans. Micr. Theory Tech., vol. 29,
pp. 135-142 (1981).
The unified approximations (8.60), (8.100) and (8.101) were first obtained by H.A. Wheeler:
H.A. Wheeler, “Transmission-line properties of a strip on a dielectric sheet on a plane”, IEEE
Trans. Micr. Theory Tech., vol. 25, pp. 631-647 (1977).
This paper includes formulas accounting for the nonzero thickness of the strip as well. Expression (7.41)
for rǫ is an improvement on that in Wheeler’s formula. See:
E.F. Kuester, “Accurate approximations for a function appearing in the analysis of mi-
crostrip”, IEEE Trans. Micr. Theory Tech., vol. 32, pp. 131-133 (1984).
Equivalent radii (sometimes also called external conformal radii or outer radii) for conductors of
various shapes are given in:
Pólya and Szegö (1951), pp. 251-254, 273;
Yu. Ya. Iossel’ and L. S. Perel’man, “Calculation of the capacitance of wires,” Elektrichestvo,
no. 9, pp. 38-43 (1986).
Approximate equivalent radii for some other conductor shapes are given in:
C.W.H. Su and J.P. German, “The equivalent radius of noncircular antennas”, Microwave
J., vol. 9, no. 4, pp. 64-67 (1966).
Some methods for obtaining stripline capacitance are given in
Collin (1960);
formulas and tables for line parameters of many types of stripline are given in:
Saad (1971), vol. 1, pp. 115-151;
Gunston (1972);
Hilberg (1979).

273
 The approach of section 8.4 to the definition of transmission-line parameters for non-TEM modes is
a modification of that given in:

J.R. Brews, “Transmission line models for lossy waveguide interconnections in VLSI”, IEEE
Trans. Electron Devices, vol. 33, pp., 1356-1365 (1986).
R. B. Marks and D. F. Williams, “A general waveguide circuit theory,” J. Res. Nat. Inst.
Standards Technol., vol. 97, pp. 533-562 (1992).

This method is distinguished by the use of energy and power-loss integrals over the entire cross-section
of the waveguide to define the line parameters. Other authors have used line integrals analogous to those
used for TEM modes in section 8.2; see

S. A. Schelkunoff, “Impedance concept in wave guides,” Quart. Appl. Math., vol. 2, pp.
1-15 (1944).
R. A. Waldron, “Characteristic impedances of waveguides,” Marconi Rev., vol. 30, pp. 125-
136 (1967).

Both methods yield the same results in the case of a quasi-TEM or pure TEM mode, but in other
cases give different values for the line voltage and current, and thus for the characteristic impedance.
Quasi-TEM modes are discussed from a number of viewpoints in:

C. Snow, “Alternating current distribution in cylindrical conductors,” Sci. Papers Bureau of

Standards, no. 509, pp. 277-338 (1925) [also in Proceedings of the Seventh International
Congress of Mathematicians, vol. II. Toronto: University of Toronto Press, 1928, pp.
157-218].
G.K. Grunberger, V. Keine and H.H. Meinke, “Longitudinal field components and frequency-
dependent phase velocity in the microstrip transmission line”, Electron. Lett., vol. 6, pp.
683-685 (1970).
A.F. dos Santos and J.P. Figanier, “The method of series expansion in the frequency domain
applied to multidielectric transmission lines”, IEEE Trans. Micr. Theory Tech., vol. 23,
pp. 753-756 (1975).
M. Aubourg, “Onde quasi-TEM des guides cylindriques blindes”, Ann. Telecomm., vol. 34,
pp. 45-51 (1979).
A. Magos, “Calculation of quasi-TEM waves by series expansion in powers of frequency”,
Periodica Polytechnica (Elec. Eng.), vol. 23, pp. 7-25 (1979).
Ye. N. Korshunova, A. N. Sivov and A. D. Shatrov, “Quasistatic theory of metal-magneto-
dielectric waveguide structures,” Sov. J. Commun. Technol. Electron., vol. 37, no. 7,
pp. 75-83 (1992).

More information on microstrip and other types of planar quasi-TEM lines is given in:

Gupta, Garg and Bahl (1979);

Gupta and Singh (1974), chapter 3;
R. Mittra and T. Itoh, “Analysis of microstrip transmission lines”, in Advances in Mi-
crowaves, vol. 8 (L. Young, ed.). New York: Academic Press, 1974, pp. 67-141;
Nefedov and Fialkovskii (1980);
Edwards (1981);
Hoffmann (1987);
Rozzi and Mongiardo (1997).

These references contain discussion and further references on dispersion, losses, effects of nonzero strip
thickness, and so on.
The coplanar waveguide was introduced by

274
 C.P. Wen, “Coplanar waveguide: A surface strip transmission line suitable for nonreciprocal
gyromagnetic device application”, IEEE Trans. Micr. Theory Tech., vol. 17, pp. 1087-
1090 (1969).
Formulas (8.105)-(8.108) for CPW are a special case of a result due to
E. S. Kochanov, “Capacitance of a planar strip line allowing for the dielectric substrate
thickness,” Telecommun. Radio Eng., vol. 29/30, no. 1, pp. 127-128 (1975).
Formulas (8.109)-(8.112) for CPS were given by
G. Ghione and C. Naldi, “Analytical formulas for coplanar lines in hybrid and monolithic
MICs”, Electron. Lett., vol. 20, pp. 179-181 (1984).
The method used to obtain the CPW and CPS parameters is a quite general one, capable of treating a
variety of two-dielectric lines; see:
E. S. Kochanov, “Computation of electric capacitance of wires in two-layer structures,”
Elektrichestvo, no. 7, pp. 81-84 (1977).
Kochanov also gave a CPS formula by this method. Although it is less accurate than that of Ghione
and Naldi, it contains the case of unequal strip widths:
E. S. Kochanov, “Parasitic capacitances in printed wiring of radio equipment,” Telecommun.
Radio Eng., vol. 21/22, no. 7, pp. 129-132 (1967).
Synthesis formulas analogous to (8.102) for the design of CPW’s with desired values of characteristic
impedance (but not desired values of ǫeff ) are given in:
T. Q. Deng, M. S. Leong and P. S. Kooi, “Accurate and simple closed-form formulas for
coplanar waveguide synthesis,” Electron. Lett., vol. 31, pp. 2017-2019 (1995).
and for CPS in:
T. Q. Deng, M. S. Leong, P. S. Kooi and T. S. Yeo, “Synthesis formulas for coplanar lines
in hybrid and monolithic MICs,” Electron. Lett., vol. 32, pp. 2253-2254 (1996).
Many other design formulas are given in
Wolff (2006), chapter 2.
Calculation of the dispersion of quasi-TEM lines generally requires quite an elaborate mathematical
analysis. See, for example,
G. Kowalski and R. Pregla, “Dispersion characteristics of single and coupled microstrips”,
Arch. Elek. Übertragungstech, vol. 26, pp. 276-280 (1972).
from which Figure 8.25 is adapted. See also Chang and Kuester (1979) above and
E.F. Kuester and D.C. Chang, “Theory of dispersion in microstrip of arbitrary width”, IEEE
Trans. Micr. Theory Tech., vol. 28, pp. 259-265 (1980).
For engineering applications, it is often best to make use of empirical closed-form results such as (8.113),
which is due to
M. Kobayashi, “A dispersion formula satisfying recent requirements in microstrip CAD,”
IEEE Trans. Micr. Theory Tech., vol. 36, pp. 1246-1250 (1988).
while (8.122) for CPW and CPS was given in:

275
 G. Hasnain, A. Dienes and J. R. Whinnery, “Dispersion of picosecond pulses in coplanar
transmission lines,” IEEE Trans. Micr. Theory Tech., vol. 34, pp. 738-741 (1986).

Other dispersion formulas for CPW and CPS can be found in:

M. Y. Frankel, S. Gupta, J. A. Valdmanis and G. A. Mourou, “Terahertz attenuation and

dispersion characteristics of coplanar transmission lines,” IEEE Trans. Micr. Theory
Tech., vol. 39, pp. 910-916 (1991).
S. Gevorgian, T. Martinsson, A. Deleniv, E. Kollberg and I. Vendik, “Simple and accurate
dispersion expression for the effective dielectric constant of coplanar waveguides,” IEE
Proc. pt. H, vol. 144, pp. 145-148 (1997).
A. K. Rastogi and S. Mishra, “Coplanar waveguide characterization with thick metal coat-
ing,” Int. J. Inf. Millimeter Waves, vol. 20, pp. 505-520 (1999).

The pseudo-TEM mode guided by a wire in a tunnel is discussed in:

E. F. Kuester and D. B. Seidel, “Low-frequency behavior of the propagation constant along

a thin wire in an arbitrarily shaped mine tunnel,” IEEE Trans. Micr. Theory Tech., vol.
27, pp. 736-741 (1979).

The results (8.132) and (8.134) for slotline are from unpublished work of Kuester and Chang, who
improved the results of:

L.A. Vainshtein, N.I. Lesik and V. Kondrat’ev, “Quasi-static theory of the dominant mode
in a slot line”, Radio Eng. Electron. Phys., vol. 22, no. 9, pp. 36-43 (1977).
Finally, finlines are comprehensively treated in

Bhat and Koul (1987).

A variety of open transmission lines based on metallized dielectric cylinders is discussed in:

Shestopalov (1997).
Discussions of issues surrounding whether a structure supports bound modes in certain ranges of β,
can be found in

A.-M.A. El-Sherbiny, “Exact analysis of shielded microstrip lines and bilateral fin lines,”
IEEE Trans. Micr. Theory Tech., vol. 29, pp. 669–675 (1981).
R.W. Jackson, “Considerations in the use of coplanar waveguide for millimeter-wave inte-
grated circuits,” IEEE Trans. Micr. Theory Tech.,, vol. 34, pp. 1450–1456 (1986).
H. Shigesawa, M. Tsuji and A.A. Oliner, “Conductor-backed slotline and coplanar waveguide:
Dangers and full-wave analysis,” 1988 IEEE MTT Symposium Digest, vol. 1, New York,
NY, pp. 199–202 (May 1988).
These papers also treat the related question of so-called “leaky” modes. Leaky modes display attenuation
along the direction of propagation due to radiation loss into one or more of the wave species that can
propagate in the transverse direction if β is less than their wavenumber. Although not proper bound
modes, (they are in fact superpositions of continuous spectrum waves concentrated near a particular
value of β), leaky wave modes can behave for many purposes as if they were.

8.12 Problems
p8-1 Show that the current in conductor Cg (the “ground”) of a TEM transmission line is equal
and opposite to the current flowing on C1 ; i. e., prove (8.27).

276
p8-2 The complex power carried by a mode on a waveguide is
1
Z
P̂ = E × H∗ · uz dS
2 S0
where S0 is the cross-section of the guide. When the mode is a TEM mode, show that
1
P̂ = (Φ1 I1∗ )
2
which is identical with the corresponding result (1.19) for the distributed constant model of
a classical transmission line. Here I1 and Φ1 are defined by (8.26) and (8.21) respectively.
Likewise show that the complex oscillatory power
1
Z
P̂osc = E × H · uz dS
2 S0
is equal to Φ1 I1 /2.
p8-3 Design a coaxial cable that will have a characteristic impedance of Zc = 72 Ω, using a
dielectric of polyethylene (ǫr = 2.26) and an inner conductor of AWG 20 gauge solid copper
wire (diameter = 0.8118 mm).
p8-4 A coaxial line is made up of an inner conductor of radius a, a dielectric layer of permittivity
ǫ1 in a < ρ < c, a dielectric layer of different permittivity ǫ2 in c < ρ < b, and an outer
conductor of radius b as shown.

Obtain the quasi-TEM line parameters l, c, Zc and β for this line. (Hint: cf. Johnk (1975),
example 4-4.)
p8-5 Repeat problem p8-4 for the case where ǫ(ρ) and µ(ρ) are arbitrary functions of the radial
coordinate ρ, but are independent of φ. Obtain expressions for the quasi-TEM line parameters
l, c, Zc and β for this line.
p8-6 Design microstrip lines according to each of the following specifications:
(a) Zc = 10 Ω; use a 50 mil (1.27 mm) alumina substrate (ǫr = 9.7).
(b) Zc = 300 Ω; use a 50 mil (1.27 mm) alumina substrate (ǫr = 9.7).
(c) Zc = 100 Ω; use a 2 mm thick Teflon substrate (ǫr = 2.1).

277
p8-7 Suppose a quasi-TEM line is formed with air and a homogeneous substrate with relative
permeability µr 6= 1, as well as relative permittivity ǫr 6= 1.
(a) Let the inductance per unit length of this line be given by the function L(µr ).
Show that
µ0 ǫ 0
L(µr ) =
C( µ1r )
Hence show that the propagation coefficient is
s
√ C(ǫr )
β = ω µ0 ǫ 0
C( µ1r )

and that the characteristic impedance is

ǫ0
Zc = ζ0 q
C(ǫr )C( µ1r )

(b) Are the values of β and Zc increased or decreased by an increase in the value
of µr ? Why?
(c) Define an effective permeability µeff in analogy with ǫeff , and express β and
Zc in terms of µeff and ǫeff . Can you give your definition a clear physical
interpretation?
p8-8 Plot the characteristic impedance Zc of a microstrip line versus w/h for several values of ǫr
(say, 1.0, 2.5 and 10), using (8.101).
p8-9 Some microstrip circuitry is laid out on a Teflon substrate (ǫr = 2.1, h = 2mm) and operates
at a frequency of f = 1 GHz. We need to design a matching section between two portions of
line, one of which has a characteristic impedance of 50 Ω, and the other 75 Ω as shown.

Assuming that junction fringing effects can be neglected, find the width w for each of the
three sections of microstrip, and the length d of the matching section.
p8-10 A substrate of yttrium iron garnet, or YIG (ǫr = 16, h = .055 in = 1.4 mm) is to be used in
the design of a 72 Ω slotline. If the operating frequency is f = 3 GHz, what is the required
slot width w? What will the corresponding value of ǫeff be? Note that a transcendental
equation has to be solved numerically for this problem.
p8-11 A CPW is to be fabricated on a substrate with ǫr = 6.2 and h = 5mm. If Zc is to be 75 Ω,
plot a graph showing w vs. s such that this value of characteristic impedance is achieved.
p8-12 A 100 Ω load impedance is connected to the end of a 50 Ω microstrip line as shown.

278
 The microstrip is fabricated on a 50 mil (1.27 mm) alumina substrate (ǫr = 9.7). An open-
ended length d2 of line with characteristic impedance Zcs is to be used as a single-stub tuner
to match the line to the load at f = 2 GHz. Find the lengths d2 needed to achieve this
match for the three cases Zcs = 20 Ω, 50 Ω and 100 Ω. In each case, use the solution with the
smallest value of d1 . Field fringing effects at the junction, the load, and the open end of the
stub are to be ignored.

p8-13 A coplanar stripline as in Fig. 8.23 is to have Zc = 50 Ω, while ǫeff = 4. If ǫr = 17,

obtain values of w/h and s/h such that these line parameters are achieved. A transcendental
equation has to be solved numerically to obtain the solution.

p8-14 Suppose that for a TE mode of a lossless hollow metallic waveguide with a simply connected
wall, we define a voltage ΦP as Z
ΦP = − ET · dℓ
P

where P is a path from a point 1 to a second point 2 on the waveguide wall as shown. Assume
that the path P is such that ΦP 6= 0, and that the mode is above cutoff (k > kc ).

xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
C0
xxxxxxxxxxxxxx
1 xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx
xxxxxxxxxxxxxx

ul
uz
un
xxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxx
P
xxxxxxxxxxxxxxxxx

(a) Show that for this mode

lP (ω, β) = µKH
and
ǫ β2
cP (ω, β) =
KH k 2

279
 where KH is a dimensionless parameter defined by:
R 2
(∂Hz /∂n) dℓ
KH ≡ P 2 R
kc S0 Hz2 dS

The results of problem p5-14 may be used in your solution without proving them.
(b) Obtain an expression for the “characteristic impedance”
s
lP (ω, β)
Zc,P ≡
cP (ω, β)

of this mode, in terms of the parameters appearing in part (a). Sketch a graph of Zc,P
vs. frequency, showing its general form.
(c) Show that the oscillatory power is given by

Φ2P
P̂osc =
2Zc,P
no matter what path P is chosen.
(d) Apply the results of parts (a) and (b) to the TE10 mode of a rectangular metallic
waveguide, choosing P to be a vertical line from y = 0 to y = d located at an arbitrary
position x = x0 along the x-axis. Find the quantities lP , cP and KH , and determine
Zc,P .

p8-15 Suppose that for a TM mode of a lossless hollow metallic waveguide with a simply connected
wall, we define a current IC as Z
IC = HT · dℓ
C
where C is a portion of the conducting wall between a point 1 and a second point 2 on the
waveguide wall as shown. Assume that C is such that IC 6= 0, and that the mode is above
cutoff (k > kc ).

Cxxxxxxxx
xxxxxxxx
xxxxxxxx
uz xxxxxxxx
xxxxxxxx
2 xxxxxxxx
xxxxxxxx
xxxxxxxx

1
ul un

xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
C0

280
 (a) Show that for this mode
cC (ω, β) = ǫKE
and
µ β2
lC (ω, β) =
KE k 2
where KE is a dimensionless parameter defined by:
R 2
C
(∂Ez /∂n) dℓ
KE ≡ R
kc2 S0 Ez2 dS

The results of problem p5-14 may be used in your solution without proving them.
(b) Obtain an expression for the “characteristic impedance”
s
lC (ω, β)
Zc,C ≡
cC (ω, β)

of this mode, in terms of the parameters appearing in part (a). Sketch a graph of Zc,C
vs. frequency, showing its general form.
(c) Show that the oscillatory power is given by

1
P̂osc = Zc,C IC2
2
no matter what segment C is chosen.
(d) Apply the results of parts (a) and (b) to the TM11 mode of a rectangular metallic
waveguide, choosing C to be the bottom wall of the waveguide: y = 0 from x = 0 (point
1) to x = a (point 2). Find the quantities lC , cC and KE , and determine Zc,C .

p8-16 Consider a propagating forward mode (γ = jβ, β > 0) on an arbitrary lossless waveguide
whose permeability is that of free space µ0 while the permittivity ǫ may be inhomogeneous (a
function of (x, y)) but not dispersive (ǫ is not a function of ω). The permittivity is assumed
to have a maximum value of ǫmax . Demonstrate the following properties as ω → ∞:
(∞) (∞)
(a) ǫeff → ǫeff , where ǫeff is a constant less than or equal to ǫmax /ǫ0 .
(b) vp → vg .
(c) In a certain sense, |Ez /ET | and |Hz /HT | approach zero.
(∞)
(d) In fact, ǫeff = ǫmax /ǫ0 .
(e) E and H approach zero at all points except where ǫ = ǫmax .

[Hint: Make use of the results of Appendix K, especially (K.34), (K.36), (K.40) and (K.71),
as well as (5.17) and (5.19).]

p8-17 Give upper and lower bounds for the propagation coefficients β of the CPW and CPS
coplanar structures considered in section 8.7. Can either of these support bound modes if
an infinite ground plane is placed on the bottom side of the substrate? If so, are any special
shielding or grounding arrangements required?

p8-18 A groove guide is a pair of metallic parallel-plates with a bulge or groove at the center as
shown in cross-section below.

281
 Could a bound mode exist on this waveguide? Explain your reasoning.
p8-19 An open single-strip line as shown below is essentially a microstrip without a ground plane.

Discuss whether a bound mode can exist on this structure, and explain your reasoning. If a
mode does exist, will it be of quasi-TEM or pseudo-TEM type?
p8-20 Consider an open stripline as in Fig. 8.9 with a strip width w = 1 mm. The dielectric
medium is nonmagnetic, with ǫ = 6ǫ0 . What values of the strip height h above the ground
plane would be necessary to achieve the following characteristic impedances: Zc = 10 Ω,
Zc = 30 Ω, Zc = 50 Ω, Zc = 100 Ω, Zc = 300 Ω and Zc = 500 Ω? Comment on the practical
feasibility of each of these transmission lines.
p8-21 It can be observed from Figure 8.22 that for given values of ǫr and h/(s + 2w), there is a
certain value of s/(s + 2w) for which ǫeff for a CPW becomes a minimum.
(a) If ǫr = 12 and h/(s + 2w) = 1, find the value of s/(s + 2w) at which the minimum of ǫeff
occurs, give its minimum value, and calculate the characteristic impedance Zc at this
minimum point.
(b) Explain physically why this minimum occurs in general.
p8-22 Provide details of the derivation of (8.75) and (8.76).

282
Chapter 9

ORTHOGONALITY, POWER
FLOW AND WAVEGUIDE
LOSSES

—Intelligence has established that the people

attacking us are the enemy.
—Do the enemy realize that you have this
information?
—Oh no, we got ’em fooled, they think that we’re the
enemy.
—Spike Milligan,
Battle of Spion Kop

9.1 Reciprocity and Reaction

One of the important tools in electromagnetic theory is the reciprocity theorem of Lorentz. We will
derive it in this section and follow up with a number of applications, including the orthogonality of
waveguide modes and the approximate calculation of waveguide losses. Further applications will be
made in the succeeding chapters.
When dealing with sources which are small or physically remote from the points where the field is
of interest, it is occasionally useful to consider the effect of fictitious magnetic currents1 M in addition
to that of the electric currents J. If such fictitious magnetic currents could exist, they would enter into
Maxwell’s equations as
∇ × E = −jωµH − M (9.1)
∇ × H = jωǫE + J (9.2)
In chapter 10, we will examine in more detail the ways in which magnetic currents can arise, specifically
in the context of field equivalence principles.

1 The use of M to denote fictitious magnetic currents is not universal and is in fact fairly recent. Many texts on

electromagnetics use Jm to denote this current, while M is used to signify the magnetization
 density of dipole moment per
unit volume. Since we will not use this latter quantity (which would be given by M (jωµ) in our notation ), the relatively
simple notation M will be followed here.

283
 Consider a set of sources Ja and Ma which produces the fields Ea , Ha in some linear isotropic
medium whose parameters are ǫ and µ (which may be functions of position). We have
∇ × Ea = −jωµHa − Ma

(9.3)
∇ × Ha = jωǫEa + Ja
The fields Ea , Ha produced by the currents Ja , Ma constitute a state “a” of our system. If we remove
Ja and Ma and replace them with a different set of sources Jb and Mb (but place them in the same
medium and boundary geometry), we observe a different field Eb , Hb which obeys
∇ × Eb = −jωµHb − Mb

(9.4)
∇ × Hb = jωǫEb + Jb
and constitutes a state “b”. Our next steps resemble the derivation of Poynting’s theorem for oscillatory
energy as done in Appendix F.
We take the dot product of the first of equations (9.3) with Hb and the second of equations (9.4)
with Ea and subtract the results. Using (D.8), we get
Hb · ∇ × Ea − Ea · ∇ × Hb
= ∇ · (Ea × Hb )
= −jωµHa · Hb − jωǫEa · Eb − Ma · Hb − Jb · Ea (9.5)
A second relation can be obtained by interchanging the superscripts a and b:
∇ · (Eb × Ha ) = −jωµHa · Hb − jωǫEa · Eb − Mb · Ha − Ja · Eb (9.6)
Finally, subtracting (9.6) from (9.5) gives
∇ · (Ea × Hb − Eb × Ha ) = −(Ea · Jb − Ha · Mb ) + (Eb · Ja − Hb · Ma ) (9.7)
This is the differential form of the Lorentz Reciprocity Theorem. We can obtain the integral form of the
theorem by integrating (9.7) over a volume V bounded by the closed surface S (Fig. 9.1) and applying
the divergence theorem:
I
(Ea × Hb − Eb × Ha ) · un dS
S
Z Z
= − (Ea · Jb − Ha · Mb ) dV + (Eb · Ja − Hb · Ma ) dV (9.8)
V V

Here un is the normal unit vector pointing outward from the surface S. In (9.8) we require only that
the medium parameters ǫ and µ be the same inside V for states “a” and “b.” Outside of V , the material
environments can be completely different.
If the volume V is extended to cover all the sources “a” and “b,” then the right side of (9.8) no longer
depends on the choice of V . We denote
Z
ha, bi ≡ (Ea · Jb − Ha · Mb ) dV
Vb
Z
hb, ai ≡ (Eb · Ja − Hb · Ma ) dV (9.9)
V a

in this case, where V contains all the “a” sources and V b contains all the “b” sources. The quantity
a

ha, bi is called the reaction of the field “a” on the sources “b”. From (9.8), then,
I
(Ea × Hb − Eb × Ha ) · un dS = −ha, bi + hb, ai (9.10)
S

284
 S
un

Figure 9.1: Volume and surface used for Lorentz reciprocity theorem.

where S is any surface completely enclosing both V a and V b , and the left side of (9.10) is independent
of the choice of S within these constraints. If for the moment we suppose that a perfectly conducting
surface Sc completely surrounds all the sources in the problem, then by choosing S to be the same as Sc
(see Fig. 9.2), we find that this constant must be zero:
I I
a b b a
(E × H − E × H ) · un dS = [un × Ea · Hb − un × Eb · Ha ] dS = 0
Sc Sc

because un × E = 0 on the surface of a perfectly conducting body. If we imagine unbounded space as

the limiting case where Sc recedes to infinity, we have shown that the surface integral is zero in this
general case as well.2 Hence, we conclude that
ha, bi = hb, ai (9.11)
when the sources are of finite extent.
A direct physical interpretation of the reaction is rather subtle in general. In the case when M = 0,
we find that the so-called self-reaction ha, ai is equal to
Z
ha, ai = Ea · Ja dV (9.12)
Va

and by the results of Appendix F, we can relate this to the oscillatory part of the power being dissipated
in or generated by the sources of the “a” field. In more general circumstances, however, when both
electric and magnetic currents are present, we can probably best think of ha, bi roughly as a measure of
how well correlated are the “a” fields with the “b” sources, or as a sort of “measured” value of the “a”
field with the “b” sources acting as a probe. We will expand on this idea below.
When some or all of the sources are confined to surfaces or lines, the reaction can be computed by
writing the volume current distributions in terms of appropriate Dirac delta functions. Let n be the
normal distance from a surface S b that contains the surface currents JbS and MbS , measured as positive
on the side of S b toward which the unit vector un points. Then the corresponding volume densities are
Jb = JbS δS b (n)
2 The actual rigorous proof is somewhat more complicated.

285
 Sc

un

Jb, Mb
Ja, Ma
Vb
Va
V

Figure 9.2: Two sets of sources in a region enclosed by a perfect conductor Sc .

Mb = MbS δS b (n)
where δS b (n) is the Dirac delta function concentrated on the surface S b . Then if a field Ea , Ha is
continuous at S b , the properties of the delta function give
Z Z
ha, bi ≡ (E · J − H · M ) dV = (Ea · JbS − Ha · MbS ) dS
a b a b
(9.13)
Vb Sb

Similar definitions apply to the case of line sources and point sources (dipoles).

9.1.1 Example
As a simple example of the application of the reciprocity theorem, consider the situation shown in
Fig. 9.3(a). On the surface S ≡ S a of a perfect conductor is placed a tangential electric surface current
JaS , resulting in the field Ea , Ha over all space. We can use the reciprocity theorem to show that this
field is identically zero. Let a “b” state be produced by either an electric Hertz dipole
Jb = jωpδ(x − x0 )δ(y − y0 )δ(z − z0 ) (9.14)
or a magnetic Hertz dipole
Mb = jωµmδ(x − x0 )δ(y − y0 )δ(z − z0 ) (9.15)
which we will use as a probe for the “a” state field (see Appendix F for the properties of Hertz dipoles).
We have either
ha, bi = jωp · Ea (x0 , y0 , z0 ) (9.16)

286
 S S

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
JS (induced)
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx H≠0
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxa xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
JS
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx ext
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
(x0, y0, z0) JS (impressed)
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
perfect
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxun perfect
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
conductor
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
conductor
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
(E=H=0)
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

(a) (b)

Figure 9.3: (a) A tangential impressed surface current on a perfect conductor; (b) Zero external field
produced by electric currents at the surface of a perfect conductor.

in the case of (9.14), or

ha, bi = −jωµm · Ha (x0 , y0 , z0 ) (9.17)
in the case of (9.15), so that ha, bi is a measure of some component of the “a” state field. But ha, bi = hb, ai
and, by (9.13), Z
hb, ai = Eb · JaS dS = 0 (9.18)
S

because Ebtan on the perfect conductor S is identically zero. By (9.16) and (9.17), this implies that
Ea , Ha are both zero at any point in space. We might say that the conductor “shorts” out the source
JaS and results in no net exterior field.
If we imagine a small space (which is reckoned eventually to approach zero in the limit) between an
impressed source JextS and the perfect conductor as shown in Fig. 9.3(b), we can gain a little further
insight into what is happening here. An equal and opposite induced current JS appears on the conductor
in a way similar to that predicted by image theory (even though the conductor is not a plane surface). The
resultant E field becomes zero everywhere as the surface S containing the impressed sources approaches
the conductor. The H-field inside the conductor and outside S also becomes zero, but the H-field in
the infinitesimal layer between them continues to support a nonzero H = JS × un = un × Jext S . In
circuit theory, this is analogous to an ideal current generator which has a short circuit connected across
it. Current flows around the loop, but currents and voltages outside this loop remain zero.
This result is used in an approximate way to reduce the electromagnetic interference (EMI) produced
by high-speed digital currents flowing on printed circuit boards (PCBs). If a large ground plane is placed
next to the plane(s) in which the current-carrying conducting traces lie (separated from the traces by
a thin dielectric substrate), then the fields produced outside the PCB are nearly zero, because of being
approximately short-circuited by the ground plane.

9.2 Orthogonality
The reciprocity theorem can be used to deduce a number of properties of waveguide modes, many of
which are presented in Appendix K. One of the most important such properties is the orthogonality

287
 Sz (@ z=z1)

∆S {

S0 (@ z=0)

Figure 9.4: Slice of waveguide between z = 0 and z = z1 .

between fields of different modes of the same waveguide. Orthogonality is a kind of statement that, in
a uniform section of waveguide, each mode behaves independently of all the others, in a sense to be
specified below. The property is used extensively in later chapters when discussing excitation, scattering
and coupling of waveguide modes.
Consider the application of the Lorentz reciprocity theorem to the fields of two modes of the same
waveguide. Take
Ea = Em e−γm z ; Ha = Hm e−γm z
Eb = En e−γn z ; Hb = Hn e−γn z (9.19)
and apply (9.8) to a source-free region of the waveguide. Specifically, let the volume V be a slice cut
out of the waveguide cross-section (which we can, as in Fig. 9.2, temporarily assume to be bounded by a
perfectly conducting wall and then, if desired, allow the wall to recede to infinity to allow for cases—such
as open transmission lines or optical waveguides—where the waveguide cross-section is unbounded) as
shown in Fig. 9.4. The surfaces at z = z1 (Sz ) and z = 0 (S0 ) make up the front and back parts of the
bounding surface S, while the lateral surface at the wall of the waveguide is denoted ∆S. Because the
region is assumed source-free, the right side of (9.8) will be zero. Since the waveguide wall is perfectly
conducting, we have on ∆S
E a × H b · un = un × E a · H b = 0
along with a similar expression with a ↔ b, so the contribution to the surface integral from ∆S vanishes.
Equation (9.8) now becomes
Z Z
−(γm +γn )z1
e (Em × Hn − En × Hm ) · uz dS − (Em × Hn − En × Hm ) · uz dS = 0 (9.20)
Sz S0

Now E and H depend only on x and y and not on z, so the integral over Sz is the same as the
integral over S0 , and thus
Z
[e−(γm +γn )z1 − 1] (Em × Hn − En × Hm ) · uz dS = 0 (9.21)
S0

288
or, since (9.21) must hold for any value of z1 ,
Z
(Em × Hn − En × Hm ) · uz dS = 0 (if γm 6= −γn ) (9.22)
S0

This is known as the orthogonality property between fields of modes having other than exactly opposite
propagation coefficients. It should be noted that only the transverse components of E and H enter
into the orthogonality relation (9.22). Sometimes orthogonality is expressed in a somewhat different
form than (9.22). If we let n → −n in (9.22) and apply (5.25), we obtain a second result which can be
combined with (9.22) to yield
Z
2
Em × Hn · uz dS = 0 (if γm 6= γn2 ) (9.23)
S0

For certain waveguides, we have seen that more than one mode can sometimes have the same prop-
agation coefficient γ. Modes with this property are said to be degenerate. In a lossless waveguide, if
they are not already orthogonal, it is always possible to form combinations of degenerate modes into a
new subset of modes so that (9.22) holds under the sole requirement that m 6= −n. Alternatively, we
could imagine a slight perturbation being introduced to the waveguide (a change in its wall shape, for
instance) which destroys the degeneracy in practice. In any event, we will always assume that (9.22)
holds for m 6= −n in all our applications.
Thus, with the normalization condition (5.27) taken into account, we finally have in place of (9.23)
1
Z
Em × Hn · uz dS = 0 (if m2 6= n2 )
2
S0
= 1 (if + m or − m = n > 0; forward mode) (9.24)
= −1 (if + m or − m = n < 0; reverse mode)

When continuous or radiation modes can exist on a waveguide structure, the orthogonality condition
takes a somewhat different form. If E(κ) and H(κ) are the fields of a continuous mode whose propagation
coefficient is γ(κ), and if γ(κ1 ) 6= γ(κ2 ) for κ1 6= κ2 , and γ(−κ) = −γ(κ), then we can normalize the
continuous mode fields such that
Z


E(κ1 ) × H(κ2 ) − E(κ2 ) × H(κ1 ) · uz dS = 4δ(κ1 + κ2 ) (9.25)
S0

where δ is the Dirac delta-function. Moreover, a discrete guided mode is always orthogonal to a contin-
uous mode: Z


E(κ) × Hm − Em × H(κ) · uz dS = 0 (9.26)
S0

9.2.1 Orthogonality of modes on lossless waveguides

Equation (9.24) can be put in a somewhat different form for most modes of a lossless waveguide. In
such a situation, if the propagation coefficient γm is either pure real or pure imaginary (i. e., the mode
is not a complex mode), then EmT can be chosen purely real before normalization is imposed (as shown
in problem p5-17). From the orthogonality property (9.24) we then have
Z Z ∗
∗
EmT × Hm′ T · uz dS = EmT × Hm′ T · uz dS =0 (9.27)
S0 S0

if m 6= ±m′ . Multiplication of the fields by a complex constant will obviously not change this result, so
it is true also for the normalized mode fields.

289
 am
sources sources
bm

z
z1 z2

Figure 9.5: Sources producing modes ±m.

We further have from section K.2 of Appendix K that γm is either pure real or pure imaginary
(corresponding to cutoff or propagating modes respectively) provided that the complex power P̂ carried
by the mode is not zero, and further that P̂m is either real or imaginary as found in (5.26). Then, by
(5.28), eqn. (9.24) for m > 0 and m′ > 0 can be written in the form

1
Z
EmT × H∗m′ T · uz dS = 0 m2 6= m′2
2 S0
= 1 m2 = m′2 , γm = jβm (9.28)
2 ′2
= jum m = m , γm = αm

where um = +1 if the mode stores more magnetic than electric energy per unit length, and um = −1 if
the mode stores more electric than magnetic energy per unit length (cf. (K.24)). This has implications
for power transport in waveguides as discussed in the next section.
A final note about the fields of a propagating mode on a lossless waveguide: if EmT and HmT are
chosen to be real, it is evident that Emz and Hmz will be pure imaginary. Thus, reverse mode fields can
be expressed as
∗
E−m = Em
H−m = −H∗m (9.29)

9.3 Power Flow in Lossless Waveguides

There is a close analogy which exists between any waveguide mode and the voltage and current on a
classical transmission line. Let us develop the notion of this equivalence here in a systematic way.
Suppose that we somehow set up sources on either side of a source-free section z1 ≤ z ≤ z2 of a
uniform waveguide such that only ±z-traveling modes with given indices ±m are excited, as shown in
Fig. 9.5. Then the field in this region will be given by:

E = am Em (x, y)e−γm z + bm E−m (x, y)eγm z

(9.30)
H = am Hm (x, y)e−γm z + bm H−m (x, y)eγm z

where am , bm are wave amplitudes associated with the ±m modes. We will see how to compute them
in chapter 10. Because of relations (5.25) between components of the ±m mode fields, we can obtain

290
separate expressions for the longitudinal and transverse components of the fields:

ET = EmT (x, y)[am e−γm z + bm eγm z ]


(9.31)
HT = HmT (x, y)[am e−γm z − bm eγm z ]

Ez = Emz (x, y)[am e−γm z − bm eγm z ]


(9.32)
Hz = Hmz (x, y)[am e−γm z + bm eγm z ]
The z-dependent factor in the expressions for ET and Hz is formally the same as the expression for the
z-dependent voltage on a classical transmission line, eqn. (1.7), if we identify am (bm ) as the forward-
(reverse-) traveling wave amplitude, and γm as the propagation coefficient. Within a constant factor
(the characteristic impedance), the z-dependent factor in the expressions for HT and Ez is formally
identical to that for the current on this same line.
Now let us assume that the transverse fields in a lossless waveguide are given as a sum of all possible
such ±z-directed mode pairs from (9.31):
X
ET = [am e−γm z + bm eγm z ]EmT (x, y)
m>0
X
HT = [am e−γm z − bm eγm z ]HmT (x, y) (9.33)
m>0

The total complex power being carried through the cross-section is then
1
Z
P̂ (z) = (ET × H∗T ) · uz dS
2 Sz
1 X X
= [am e−γm z + bm eγm z ] (9.34)
2 m>0 ′
m >0
Z
∗ ∗
×[a∗m′ e−γm′ z − b∗m′ eγm′ z ] (EmT × H∗m′ T ) · uz dS
S0

From the orthogonality condition (9.27), the double sum in (9.34) then reduces to a single sum of the
form:
1 X
Z
∗ ∗
P̂ (z) = [am e−γm z + bm eγm z ][a∗m e−γm z − b∗m eγm z ] (EmT × H∗mT ) · uz dS (9.35)
2 m>0 S0

It can be inferred from (9.35) that the complex power carried by a collection of modes existing in a
lossless waveguide is equal to the sum of the complex powers associated with each forward/reverse (±z-
directed) pair of modes individually. This is a generalization of the result (3.63) obtained for power
carried by propagating modes on lossless multiconductor transmission lines.
When we take the real part of (9.35) to obtain the time-average power carried through the guide, we
then obtain using (9.28) that
X
|am |2 − |bm |2

Pav (z) = Re P (z) =
m>0;γm =jβm
X
+ 2um Im(am b∗m ) (9.36)
m>0;γm =αm

This is a generalization of relation (1.22) for a lossless classical transmission line. The forward and
reverse propagating modes thus all carry power independently of each other. The cutoff modes can
carry power only in the presence of both forward and reverse attenuating modes; if only one type is
present, then am or bm is zero, and their contribution to the total power flow vanishes.

291
9.4 Dielectric Losses in Waveguides
For homogeneously filled waveguides, we have previously seen that accounting for the effect of dielectric
losses on the modes—especially on their propagation coefficients—is fairly straightforward. We have
only to make ǫ (or µ) appropriately complex, and use the formulas already derived for the modes. When
the waveguide is inhomogeneously filled, the problem becomes more difficult.
The effect of dielectric losses in the medium filling an arbitrary waveguide can be determined by an
approximation procedure if the losses in the medium are small. Suppose for simplicity that the filling
material is piecewise homogeneous, with relative permittivity ǫr taking on the values ǫr1 , ǫr2 , and so on,
in the subareas S1 , S2 , etc., of the waveguide cross section S0 (which may extend to infinity in the case
of an open waveguide). Depending on the context, we may define an effective permittivity or effective
index for a particular mode as
ǫeff = n2eff = β 2 /k02 (9.37)
and emphasize its functional dependence on the relative dielectric constants

ǫr1 = n21 ; ǫr2 = n22 ; .... (9.38)

of the various sections by writing

ǫeff = ǫeff (ǫr1 , ǫr2 , ...) (9.39)
The presence of small losses ǫ′′1 , ǫ′′2 , ..., in the materials causes
ǫ′′1,2
ǫr1,2 → ǫr1,2 − j
ǫ0
The complex propagation coefficient γ can be evaluated by taking the first-order terms in a Taylor series
expansion about real values of the ǫri :
X ǫ′′ ∂γ 

i
γ = α + jβ ≃ γ|ǫr1 ,ǫr2 ,...=real − j 
i
ǫ0 ∂ǫri ǫr1 ,ǫr2 ,...=real

The value of γ = jβ and its derivatives at real values of the filling material permittivities can be expressed
in terms of, say, ǫeff , by means of (9.37):
q
β|ǫr1 ,ǫr2 ,...=real = k0 ǫeff |ǫr1 ,ǫr2 ,...=real
 
∂β  k0 ∂ǫeff 
= q
∂ǫri ǫr1 ,ǫr2 ,...=real 2 ǫeff | ∂ǫri ǫr1 ,ǫr2 ,...=real
ǫr1 ,ǫr2 ,...=real

We now define filling factors q1 , q2 , etc., as


∂ǫeff 
qi = (9.40)
∂ǫri ǫr1 ,ǫr2 =...=real

Then to a first approximation, α and β are given by

β ≃ k0 neff
!
k0 X ǫ′′
α ≃ qi i (9.41)
2neff i
ǫ0

where neff refers to the value of neff when the ǫri are real. From (9.41), we can see that β is virtually
unaffected by small dielectric losses. The effect upon α of losses in different regions depends on the
magnitude of the filling factor for those regions.

292
 In Appendix K, we show how the filling factors can be expressed in terms of fractional parts of stored
energy or power carried in the various dielectrics of the waveguide, as well as the group velocity. This
physical interpretation is useful in gaining insight into the loss mechanisms of a waveguide. We have
that
∗
R
c2 /ǫri Si ǫEm · Em dS
qi = R
∗ dS
(9.42)
vp vg S0 ǫEm · Em
where c is the speed of light in vacuum, and vp and vg are the phase and group velocities of the mode of
the corresponding lossless guide. For a metallic waveguide uniformly filled with dielectric, we have from
(K.41) and (9.42) that q = 1, and (9.41) reduces to (5.112) and (5.113) of chapter 5.

9.4.1 Example: Dielectric loss for quasi-TEM modes

For a quasi-TEM mode on a structure containing only air and one other uniform dielectric region (the
substrate, with subscript i = 1 which will be dropped here), we have
r
ω µ0
α≃ qǫ′′ (9.43)
2 ǫ0 ǫeff
where
∂ǫeff
q= (9.44)
∂ǫr
and ǫeff is to be evaluated at (the real value of) ǫr . Equation (9.43) should be compared to the small
loss limit of the attenuation coefficient (8.46) of a TEM mode:
r
ω µ ′′
αTEM ≃ ǫ (9.45)
2 ǫ′
For microstrip for instance, the wider the line is, the closer to 1 the filling factor will be, and the larger
the attenuation coefficient for a given value of ǫ′′ . Qualitatively, however, dielectric losses for quasi-TEM
modes are very similar to those of pure TEM modes.

9.4.2 Example: Dielectric loss in step-index, weakly-guiding waveguides

Consider a step-index dielectric waveguide which is weakly-guiding. It is known that the waveguide can
be described completely by two normalized quantities: the normalized frequency
√
V = k0 d ǫr1 − ǫr2 (9.46)

where d is some convenient length parameter of the cross-section, ǫr1 is the relative permittivity of the
core and ǫr2 is that of the cladding; and the confinement parameter
ǫeff − ǫr2
b= (9.47)
ǫr1 − ǫr2
which implies the relation
ǫeff = ǫr2 + b(ǫr1 − ǫr2 ) (9.48)
In this case, b is a function of k0 , ǫr1 , ǫr2 and d only through the parameter V :

b = f (V ) (9.49)

Now, from (9.46) we have that partial derivatives of V with respect to ǫr1 , ǫr2 and even the frequency
ω are related to each other:
∂V ∂V k0 d
=− = √ (9.50)
∂ǫr1 ∂ǫr2 2 ǫr1 − ǫr2

293
and
∂V 2(ǫr1 − ǫr2 ) ∂V d√
= = ǫr1 − ǫr2 (9.51)
∂ω ω ∂ǫr1 c
where c = (µ0 ǫ0 )−1/2 is the velocity of light in vacuum.
Although we may not know an explicit form of f , we do know from (9.48) that
∂ǫeff ∂b ∂V
q1 = = b + (ǫr1 − ǫr2 ) = b + (ǫr1 − ǫr2 )f ′ (V ) (9.52)
∂ǫr1 ∂ǫr1 ∂ǫr1
and
∂ǫeff ∂b ∂V
q2 = = 1 − b + (ǫr1 − ǫr2 ) = 1 − b + (ǫr1 − ǫr2 )f ′ (V ) (9.53)
∂ǫr2 ∂ǫr2 ∂ǫr2
so it follows immediately from (9.50) that

q1 + q2 = 1 (9.54)

This relation holds only for weakly-guiding step-index dielectric waveguides.

There is also a relation between the filling factors (and thus the attenuation coefficient) and the
group velocity of a mode on a weakly guiding step-index guide. It is more usual to express this relation
in terms of the so-called group-index of the mode:
c dβ d dneff
ng = =c = (ωneff ) = neff + ω (9.55)
vg dω dω dω

Taking the derivative of (9.48) with respect to ω and using (9.49), we get
dneff ∂V
2neff = (ǫr1 − ǫr2 )f ′ (V ) (9.56)
dω ∂ω
Substituting (9.51), (9.52) and (9.55) into this equation, we obtain after some algebra:

neff ng − n22
q1 = (9.57)
n21 − n22

The comparison of q1 as given by (9.57) with b as given by (6.55) is interesting. Equations (9.57) and
(9.54) allow us to compute the dielectric losses for a weakly-guiding fiber completely from a dispersion
curve (β vs. ω) for each mode, if we can accurately read the slope at the desired frequency.

9.5 Wall Losses in Waveguides

The effect of wall losses due to finite conductivity of the boundary is not so easy to account for as that
of dielectric losses. If we were to let the walls be treated as a finitely conducting medium, the waveguide
would no longer be electrically homogeneous, and any simplifications which were possible in the case
of ideal walls (such as the existence of pure TE and TM modes) will be lost. Besides, if (as would
be reasonable to assume in practice) the walls are highly conducting, there should be very little field
penetration into the walls at all. In fact, recalling the behavior of a plane wave being refracted into a
highly conducting medium (Fig. 9.6), we might guess that the fields vary almost purely normally away
from the interface and rapidly decay in that same direction within a few skin depths δc , where
r
2
δc = (9.58)
ωµc σc
This characteristic behavior of fields within a good conductor makes it possible to formulate an
approximate boundary condition at its surface which eliminates the need to calculate fields within the

294
 n

(ε,µ)
|E|

(εc,µc,σc)

-n

Figure 9.6: Plane-wave penetration into a good conductor for (ǫc − jσc /ω)/ǫ ≪ 1.

conductor at all. Boundary conditions of this type are called impedance boundary conditions3 and take
the form
Etan = ZS un × Htan (9.59)
where un is the unit normal vector pointing away from the conductor at the interface, and Etan and
Htan are the components of the fields outside the conductor, tangential to the interface (Fig. 9.7). The
quantity ZS is known as the surface impedance associated with the given interface. It may be a function
of position along the surface.
In Appendix L, we show that, for a smooth, highly conducting surface,
µc
r
ZS = ζc = (9.60)
ǫc − jσc /ω

where ζc is the complex wave impedance for a plane wave in the conducting medium.4 If, as is usually
the case, σc /ωǫc >> 1, we have approximately
r
ωµc 1+j
ZS = (1 + j) = ≡ RS + jXS (9.61)
2σc δc σc
Unfortunately, even the use of the simpler boundary condition (9.59) is not enough to allow exact
solution of the mode fields for most cross-sectional shapes (even rectangular). Thus, we will develop here
3 Though (9.59) is superficially similar to the resistive sheet boundary condition (5.90), it differs in that H
tan is nonzero
only on one side (n > 0) of the boundary.
4 Many types of surface (rough, covered with dielectric layers, etc.) can also be characterized by surface impedances

ZS given by other expressions.

295
 un
u1 (ε, µ)
u2

(εc, µc, σc)

Etan = E1 u1 + E2 u2 (good conductor)
Htan = H1 u1 + H2 u2

Figure 9.7: Geometry for impedance boundary conditions (u1 and u2 are unit vectors tangent to the
interface).

an approximate method to determine the effect of wall losses on the propagation coefficient by assuming
that the fields are very nearly what they would be if the walls were perfectly conducting. To do this, we
apply the reciprocity theorem to the fields

Ea = Em0 e−γm0 z ; Ha = Hm0 e−γm0 z

Eb = E−m eγm z ; Hb = H−m eγm z (9.62)

where Em0 , Hm0 , and γm0 are associated with a mode of the perfect guide, and E−m , H−m , and −γm
with a corresponding mode of the guide with imperfect walls propagating (or attenuating) in the opposite
direction. We apply reciprocity to the volume of Fig. 9.4. The manipulations are very similar to those
in the proof of orthogonality, except that the integral over ∆S does not vanish because un × Em is not
zero on the guide wall but obeys (9.59). Since
Z Zz1 I
= dz dℓ,
∆S 0 C0

we obtain
H H
ZS Hm0(tan) · H−m(tan) dℓ ZS Hm0(tan) · H−m(tan) dℓ
C0 C0
γm − γmo = R =−R (9.63)
(Em0 × H−m − E−m × Hm0 ) · uz dS (Em0 × Hm + Em × Hm0 ) · uz dS
S0 S0

Though (9.63) is exact, we can do little with it as it stands since Em and Hm are not known. However,
we reason that, if ZS is small enough, we should have

E−m ≃ E−m0

H−m ≃ H−m0
so that H
ZS Hm0 · H−m0 dℓ
C0
γm − γm0 ≃ R (9.64)
−2 Em0 × Hm0 · uz dS
S0

Only the components of H tangential to the wall enter into the line integral in the numerator of (9.64);
these would be the scalar components Hm0l and Hm0z in the coordinate systems of Figures 5.5 or 8.5.

296
 If the medium filling the guide is lossless, the real part of (9.64) is interpretable as one half the power
lost into the walls per unit length divided by the power carried through the cross-section S0 of the guide.
Indeed, the real part of (9.64) gives the attenuation coefficient of the mode, and if the unperturbed guide
is lossless, then for a propagating mode (see (9.29)),
RS Hm0 · H∗m0 dℓ
H H
RS Hm0 · H−m0 dℓ
C0 C0
αm ≃ = R (9.65)
2 Em0 × H∗m0 · uz dS
R
−2 Em0 × Hm0 · uz dS
S S0

For a TM mode (and also, as a special case, for a TEM mode) of a homogeneously filled and otherwise
loss-free waveguide whose walls have uniform surface resistance, we have from (9.61), (9.65), (5.47), and
(5.48) that, for ω > ωc,m ,
|Hm0ℓ |2 dℓ
H
RS 1
αm ≃ R C0 (9.66)
Hm0T · H∗m0T dS
q
2ζ 2
1 − ωc,m /ω 2
S0

which is the attenuation (or additional attenuation) due to finite conductivity of the walls. Here ζ is the
wave impedance of the medium filling the guide, while RS is the surface resistance of the walls (eqn.
(9.61)).
For TE modes of homogeneously filled guides, a similar but more complicated expression follows from
(5.49) and (5.50) and problem p5-14:
|Hm0ℓ |2 dℓ |Hm0z |2 dℓ 
 H H 
2 2
ωc,m /ω

RS  q
2 /ω 2 R C0 C0

αm ≃ 1 − ωc,m ∗ +q R
2
(9.67)
2ζ 
 Hm0T · Hm0T dS 2 /ω 2
1 − ωc,m |Hm0z | dS  
S0 S0

The ratios of the line and surface integrals in (9.66) and (9.67) are independent of frequency, so it is
evident that the attenuation αTM of a TM mode due to wall losses goes as
√
αTM ∼ Am ω as ω → ∞
√ −3/2
αTE ∼ Bm ω + Cm ω as ω → ∞,
where Am , Bm , and Cm are mode-dependent constants. Clearly the attenuation of all modes increases
for high enough frequency except for TE modes (if any exist) for which Bm = 0—i.e., for which |Hm0l |2
integrates to zero around the boundary C0 . We will see below that the TE0m mode of the hollow circular
waveguide is one such, which (along with the fact that, unlike the TE11 mode, it is non-degenerate) makes
it attractive to use in practical applications, especially at millimeter-wave, submillimeter-wave and even
infrared frequencies.

9.5.1 Example: The TE01 mode of a hollow circular waveguide

Consider the TE0m modes of a hollow circular waveguide as studied in chapter 5. Its E and H fields
are given by (5.78) and (5.82)-(5.84) with n = 0. It will be noticed that the transverse component of H
which is tangential to the wall, Hl = Hφ , is identically zero. The first term in (9.67) thus vanishes, and
the second term is evaluated by substitution of the field expressions from chapter 5 into the remaining
term and carrying out the integrals. With the help of (B.34), (B.35) and (B.38) we find that
RS f 2 /f 2
αTE0m ≃ p c (9.68)
ζa 1 − fc2 /f 2

where fc = fc,TE0m is the cutoff frequency of the mode. The decay of this expression as frequency
increases can be seen.

297
Figure 9.8: Theoretical and measured values of attenuation coefficient αc due to wall losses for the TE01
mode of a circular metallic waveguide.

In Fig. 9.8, a plot of the theoretical (eqn. (9.68)) and measured values of the attenuation coefficient
of the TE01 mode are presented. The actual losses are seen to decrease with frequency less rapidly
than theoretically predicted; the deviation between theory and experiment is attributable to an increase
of ZS with frequency when the roughness of the conductor surface is taken into account. This effect
will eventually negate the inherent low-loss properties of this mode, but there will be a broad range of
frequencies for which very low attenuation can be achieved in practice.
The TE01 mode was extensively researched during the 1950’s and 1960’s as a practical communica-
tions channel. The Bell System put into operation the WT-4 millimeter waveguide system using this
mode. The system has the following parameters:
• Repeater spacing: 20 miles
• Bandwidth: 40-110 GHz
• Capacity: 60 2-way channels @ 282 M Bits/sec/channel = 240,000 2-way voice circuits.
In order to use the TE01 mode effectively, ways had to be found to suppress the other modes which can
propagate along with it: at the very least, the TE11 , TE21 and TM11 (each in both polarizations) and
the TM01 modes. Mode filters and converters can be devised for this purpose; for example, a wire helix
can be used to line the waveguide wall, which disturbs the TE01 mode very little, but tends to suppress
the other undesired modes.

9.5.2 The incremental inductance rule

In some circumstances, it is possible to express the wall loss in a different form which is easier to
compute than the ratio of field integrals (9.64). This is accomplished by the incremental inductance rule

298
of H. A. Wheeler, which we will derive below. Assume for present purposes that µ and ǫ do not change
(or change negligibly) with frequency. Then the results (K.4) and (K.20) from Appendix K show that
R
µHm0 · H−m0 dS
∂γm0 S0
= −j R (9.69)
∂ω Em0 × Hm0 · uz dS
S0

Using this expression in (9.64), we find that (again, for an otherwise lossless mode)

δn(ℓ)Bm0 · H∗m0 dℓ
H
(1 + j)ω ∂βm0 C0
γm − γm0 ≃ (9.70)
Bm0 · H∗m0 dS
R
2 ∂ω
S0

where
RS
δn(ℓ) = (9.71)
ωµ
Here RS , ǫ and µ may be functions of position. In the case where RS is given by (9.61), δn is equal to
µc δ c
δn(ℓ) = (9.72)
2µ
The integral around C0 in (9.70) could be interpreted approximately as a surface integral of Bmo ·H∗mo
over a strip of width δn(ℓ) just inside the conductor surface C0 . The change in propagation coefficient
due to wall skin effect is thus expressible in terms of the rate of change ∂βm0 /∂ω — that is, the group
velocity of the mode, and a ratio of magnetic energies in this small strip and in the entire waveguide
cross-section. Indeed, the factor
δn(ℓ)Bm0 · H∗m0 dℓ
H
2
c C0
qc ≡ (9.73)
Bm0 · H∗m0 dS
R
vp vg
S0

can be regarded as a sort of “magnetic filling factor” analogous to the dielectric filling factors (9.42) of
the previous section, and (9.70) can be written as
(1 + j)k0
γm − γm0 ≃ qc (9.74)
2neff
which is completely analogous to (9.41).
Now for quasi-TEM modes in the quasi-static regime, we show in Appendix M that the ratio of
integrals in (9.70) can be expressed in terms of the change δl of the inductance per unit length of the
line due to a small shift of the conductor boundaries C0 = C1 ∪ C2 normally into the conductors by the
amount δn. In fact, from (M.18), we have
(1 + j)ω ∂βm0 δl
γm − γm0 ≃ (9.75)
2 ∂ω l0
This, in its most general form, is the incremental inductance rule. If µ, µc and δc (and thus also δn) are
independent of position, we can write (9.75) in the form
(1 + j)ω ∂βm0 δn δl
γm − γm0 ≃ (9.76)
2 ∂ω l0 δn
where δl/δn is calculable by direct differentiation of lo with respect to the geometrical dimensions of the
line, without the need to perform any integrations or to know the values of the fields at all.

299
 √ √
If lo and co are independent of ω as ω → 0, then βmo ≃ ω lc and thus ∂βm0 /∂ω ≃ lo co , and from
(9.76) we have:
(1 + j)RS 1 δl
γm − γm0 ≃ (9.77)
2Zc0 µ δn
p
where Zc0 is the characteristic impedance l0 /c0 of the original lossless line. Of interest is the real part
of (9.77), which gives the attenuation coefficient of the mode due to wall conduction losses:

RS δl
αm ≃ (9.78)
2µZc0 δn

We can also cast the incremental inductance rule in terms of equivalent transmission line parameters:
the series impedance per unit length r+jωl and the shunt admittance per unit length g+jωc. According
to classical transmission-line theory, (equation (1.9)), we have
p
γm = (r + jωl)(g + jωc) (9.79)

Correspondingly, for the unperturbed (lossless) line we have

p
γm0 = jβm0 = jω l0 c0 (9.80)

If r/(ωl0 ), g/(ωc0 ), (l − l0 )/l0 and (c − c0 )/c0 are all small compared to one (the finite wall conductivity
does not change the line parameters much), then
 
l − l0 c − c0 r g
γm − γm0 ≃ jβm0 + + + (9.81)
2l0 2c0 2jωl0 2jωc0

Comparing this to equation (9.75), we find that we can obtain the same expression if we take g = 0 (if
the medium filling the guide is lossless), c to be the capacitance per unit length co of the original lossless
line, and
l ≃ l0 + δl (9.82)
and
r ≃ ωδl (9.83)
The inductance of the line with imperfect conductors is thus the same as that of a lossless line whose
walls are indented by a distance
µc δ c
δn = (9.84)
µ 2
and the additional inductance is often referred to as the internal inductance of the conductors of the
line. It is worth noting that, in the lossy line lc 6= µǫ; even if the unperturbed mode was pure TEM, the
perturbed mode will only be quasi-TEM.
A word of caution is in order regarding the application of (9.77) to strip-type transmission lines. It is
common practice to model the strips of such lines as infinitely thin, which does not cause any essential
error when computing l and c so long as the actual thickness of the strip is small compared to the
width of the strip and the substrate thickness. The infinitely thin strip, however, is an inadequate model
when considering conductor losses, because the strip thickness is not large compared to a skin depth,
and this violates the validity of our surface impedance concept. More than this, the inverse square-root
singularity in the current at the edges of an infinitely thin strip (cf. Fig. 8.18) causes the line integral in
the numerator of (9.63) to diverge. In order to compute conductor loss effects for thin strip conductors,
a way must be found to correctly account for the influence of the nonzero thickness of the strip as well
as the detailed shape of the strip edge. Methods to do this are found in the references at the end of the
chapter.

300
9.5.3 Example: Conductor loss in coaxial line
Consider once again the case of a coaxial line as shown in Fig. 8.7. We have

√ µ b
lo = µǫ Zc0 = ln (9.85)
2π a
If both inner and outer conductors have the same conductivity σc and permeability µc , we can simply
compute
δl ∂l0 ∂l0 µ 1 1
= − = + (9.86)
δn ∂b ∂a 2π b a
Note that we place a minus sign in front of the ∂/∂a term in (9.86) because a recession of the inner
conductor means a decrease in a, whereas the corresponding motion of the outer conductor means an
increase in b. From (9.82) and (9.83), we have
   
µ b δ c µc  1 1 
l≃ ln + + (9.87)
2π a 2µ b a

ωδc µc  1 1 
r≃ + , (9.88)
4π b a
and by (9.78) the attenuation coefficient is

ωµc δc  1 1 
α ≃ +
4ζ ln(b/a) b a
1 1 1
≃ + .
2ζδc σc ln(b/a) b a

If α is plotted as a function of a for a fixed value of b, the result (in arbitrary units) is as shown in
Fig. 9.9. The attenuation is minimized when ∂α/∂a = 0, or

0 0.2b 0.4b 0.6b 0.8b b a

Figure 9.9: Variation of coaxial transmission line attenuation constant versus a for fixed b.

b 1
= b


a ln a −1

301
whose numerical solution gives an optimum outer to inner conductor radius ratio of (b/a)opt = 3.591.
For this value, the characteristic impedance has the value
 
1 ζ0 b 76.71 Ω
Zc,opt = √ ln = √
ǫr 2π a opt ǫr

If ǫr = 1, the optimum characteristic impedance is approximately 75 Ω, while for polyethylene (ǫr = 2.26),
the optimum value is close to 50 Ω.

9.6 Conclusion: Loss Effects on Orthogonality and Power Flow

The orthogonality properties (9.22) and (9.24) hold even for the modes of a lossy waveguide, whose
dielectric constant is complex or whose walls impose an impedance boundary condition. Only (9.28) is
restricted to the case of lossless waveguides. This does mean that the conclusions of section 9.3 on power
flow must be modified for a lossy waveguide, however, since the “power orthogonality” condition (9.28) is
no longer true. A lossy waveguide on which several modes exist transfers a power which generally consists
not merely of the sum of the powers in individual modes, but is a complicated expression involving cross
terms between all the various modes. In a waveguide with low losses, the conclusions arrived at for
lossless guides can be expected to hold true to a good approximation, but the limitations of our earlier
conclusions should be kept in mind.
Likewise, our derivation of the effect of wall losses was done initially without regard for whether
the corresponding guide with ideally conducting walls was lossless or not. Thus, (9.64) is valid for any
unperturbed guide, whether or not it has dielectric losses. This unperturbed guide’s dielectric losses can
be computed as in section 9.4, however, and from the form of (9.64) we see that in this approximation
the dielectric losses and wall losses have an effect on the attenuation coefficient of a mode that is simply
additive:
αm = αm,c + αm,d (9.89)
Any attenuation αm,d due to dielectric losses from the material filling the guide must be added to the
wall loss attenuation αm,c to give the total attenuation coefficient of the mode.

302
9.7 Notes and References
The discussion of reciprocity, reaction and orthogonality comes mostly from

Harrington (1961), chapters 3,7;

but also from

Collin(1960), pp. 27–29, 483–485;

Vainshtein (1988), chapter 14;

More general kinds of waveguides are treated in

Olyslager (1999).

Equation (9.43) for the dielectric loss in a microstrip is due to:

M.V. Schneider, “Dielectric loss in hybrid integrated circuits”, Proc. IEEE, vol. 57, pp.
1206-1207 (1969).

The reader should exercise caution in using formulas involving the filling factors q of planar lines which
are given in the literature, since not all authors use the same definition for q. Equation (9.43) is
completely accurate to first order in ǫ′′ , regardless of the geometry of the line involved. Some authors
quote the formula
ǫeff ≃ (1 − q) + qǫr (9.90)
implying
ǫeff − 1
q≃ (9.91)
ǫr − 1
for the filling factor. This would be exactly true if q were independent of ǫr —that is, if ǫeff were a linear
function of ǫr . Another widely used formula for the filling factor of microstrip is

1
q≃ [1 + (1 + 10h/w)−1/2 ] (9.92)
2
but the accuracy of this formula is not quite as good as that of (9.91), as is checked in the solution of
problem p9-9(b). Expression (9.91) can also be used with fairly good accuracy for CPW and CPS, but
not necessarily for other planar lines, especially the slotline.
The connection of dielectric losses with the group velocity for weakly-guiding fibers seems to have
been first noted by

D. Gloge, “Propagation effects in optical fibers”, IEEE Trans. Micr. Theory Tech., vol. 23,
pp. 106-120 (1975).

for the case of the circular fiber; this paper is also reprinted in Gloge (1976). The approximate expression
q1 ≃ V 2 /(V 2 + 2) for q1 of the TE0 mode of a dielectric slab was given by:

D. Botez, “Analytical approximation of the radiation confinement factor for the TE0 mode of
a double heterojunction laser”, IEEE J. Quantum Electron., vol. 14, pp. 230-232 (1978).

who called it the radiation confinement factor. The expression was shown to be accurate for small V as
well as for large V , and was used by

K.-L. Chen and S. Wang, “An approximate expression for the effective refractive index in
symmetric DH lasers”, IEEE J. Quantum Electron., vol. 19, pp. 1354-1356 (1983).

303
to obtain the approximate relation
2
b≃1− ln(1 + V 2 /2)
V2
for the TE0 mode for any value of V .
For the discussion of wall losses, see
Johnson (1965), pp. 153–162;
Collin (1960), pp. 182–189.
Information on the Bell WT-4 waveguide system can be found in:
Martin (1969), Chapters 8 and 9;
Martin (1971), Chapter 14.
Figure 9.8 is based on experimental data given in:
S. Hatano and F. Nihei, “Measurement of surface resistance in oversized circular waveguide
at millimeter wavelengths,” IEEE Trans. Micr. Theory Tech., vol. 24, pp. 886-887
(1976).
The incremental inductance rule was first given in
H.A. Wheeler, “Formulas for the skin effect,” Proc. IRE, vol. 30, pp. 412–424 (1942).
For a rigorous proof, see
V. Alessandrini, et al., “The skin effect in multiconductor systems,” Int. J. Electron., vol.
40, pp. 57–63 (1976).
One way of computing conductor losses for planar transmission lines is to use formulas for the
inductance per unit length for strips of finite thickness, and then apply the incremental inductance rule.
Such formulas are more complicated than those given in chapter 8, and for microstrip are given in
M.V. Schneider, “Microstrip lines for microwave integrated circuits,” Bell Syst. Tech. J., vol.
48, pp. 1422–1444 (1969).
H.A. Wheeler, “Transmission-line properties of a strip on a dielectric sheet on a plane,” IEEE
Trans. Micr. Theory Tech., vol. 25, pp. 631–647 (1977).
See also
Gupta, Garg and Chadha (1981), chapter 3.
An alternative to this approach which requires only information about the properties of the line with
infinitely thin strip conductors was developed independently in:
L. Lewin, “A method of avoiding the edge current divergence in perturbation loss calcula-
tions,” IEEE Trans. Micr. Theory Tech., vol. 32, pp. 717-719 (1984).
L. A. Vainshtein and S. M. Zhurav, “Strong skin effect at the edges of metal plates,” Sov.
Tech. Phys. Lett., vol. 12, pp. 298-299 (1986).
and has been extended and refined by
S. M. Zhurav, “Edge loss in metal plates of rectangular cross section,” Sov. Tech. Phys.
Lett., vol. 13, pp. 147-148 (1987).
G. Ya. Slepyan, “More precise impedance conditions for calculating the heat losses in a thin
open metal screen,” Sov. J. Commun. Technol. Electron., vol. 33, no. 4, pp. 172-175
(1988).

304
 E. L. Barsotti, E. F. Kuester and J. M. Dunn, “Strip edge shape effects on conductor loss
calculations using the Lewin/Vainshtein method,” Electron. Lett., vol. 26, pp. 983-985
(1990).
E. L. Barsotti, E. F. Kuester and J. M. Dunn, “Effect of metallization edge shape on con-
ductor loss of open coplanar waveguide,” Micr. Opt. Technol. Lett., vol. 3, pp. 389-391
(1990).
C. L. Holloway and E. F. Kuester, “Edge shape effects and quasi-closed form expressions for
the conductor loss of microstrip lines,” Radio Science, vol. 29, pp. 539-559 (1994).
C. L. Holloway and E. F. Kuester, “A quasi-closed form expression for the conductor loss of
CPW lines, with an investigation of edge shape effects,” IEEE Trans. Microwave Theory
and Techniques, vol. 43, pp. 2695-2701 (1995).

9.8 Problems
p9-1 Let a “b” state be produced by an electric current loop situated along a closed contour C:
Jb = Iuℓ δC (x, y, z)
where uℓ is the unit vector tangent to the contour C at any point on the contour, and the
delta function δC is to be understood as concentrated at the loop in the following sense: for
any function f (x, y, z), we have
Z Z
f δC dV = f dℓ
V C
if the volume V completely contains the loop C. The “b” state possesses no magnetic current
sources. Let the “a” state field be produced by any sources at all, provided that Ma = 0 for
points at the loop.
(a) Express the reaction ha, bi as a line integral around C.
(b) Use Stokes’ theorem to re-express ha, bi as an integral over a surface S spanning C.
(c) Suppose this surface
R S contracts to zero and the current I on the loop goes to infinity in
such a way that I S un dS → m, where m is a constant vector, and un is a unit vector
normal to S. Show that ha, bi approaches an expression of the form (9.17) in this limit.
[Note: m is called the magnetic dipole moment of the loop.]
p9-2 For a pair of TE modes on a hollow waveguide, show that the properties
Z
∇T Hz1 · ∇T Hz2 dS = 0 (γ12 6= γ22 )
S0
Z
Hz1 Hz2 dS = 0 (γ12 6= γ22 )
S0

follow from orthogonality property (9.24).

p9-3 For a pair of TM modes on a hollow waveguide, show that the properties
Z
∇T Ez1 · ∇T Ez2 dS = 0 (γ12 =
6 γ22 )
S0
Z
Ez1 Ez2 dS = 0 (γ12 6= γ22 )
S0

follow from orthogonality property (9.24).

305
p9-4 Show that a TE mode is always orthogonal to a TM mode in the sense of (9.23) for a hollow
2
waveguide, regardless of whether or not γm = γn2 .
p9-5 In equations (5.75)–(5.77), we give expressions for ET and HT of the TM11c circular wave-
guide mode.
(a) Give explicit expressions for ET and HT for the TM11s mode similar to (5.75)–(5.77)
[cf. equation (5.74] for the Ez field of this mode).
(b) Show that (9.22) holds for m = +TM11c and n = −TM11s , even though γm = −γn for
these degenerate modes.
p9-6 It is known that an arbitrary time-harmonic electric field E can be expressed in terms of a
scalar potential Φ and a vector potential A as follows:
E = −∇Φ − jωA
Show that, in the absence of magnetic currents (M ≡ 0), the reaction ha, bi can also be
expressed as Z
ha, bi = −jω (Φa ρb + Aa · Jb ) dV
Vb

where V b is a volume which contains the “b” sources and on whose surface un · Jb = 0, while
ρb is the volume (electric) charge density corresponding to Jb .
p9-7 A source excites a rectangular waveguide all of whose modes are below the cutoff frequency.
One end of the waveguide opens out into empty space as shown, and a measurement shows
the presence of a nonzero radiated power coming from the end of the guide.

Explain qualitatively how this cutoff waveguide can deliver power from the source into space
as described.
p9-8 Suppose there exist two modes E1 , H1 and E2 , H2 in a waveguide which have the same
propagation coefficient γ1 = γ2 = γ. Suppose, too, that the mode fields E1 , H1 and E−2 ,
H−2 are not orthogonal to each other. That is,
Z
(E1 × H−2 − E−2 × H1 ) · uz dS 6= 0.
S0

Find the constant A in order that the combination fields

Ea = E1 + AE2
Ha = H1 + AH2
are orthogonal to E1 , H1 . In other words, give A in terms of E1 , H1 and E2 , H2 such that
Z
(E1 × H−a − E−a × H1 ) · uz dS = 0.
S0

306
p9-9 (a) Using (9.40) and the limiting forms of (8.100), show that the filling factor q for a
microstrip line is approximately 12 for narrow strips (w/h ≪ 1) and approximately 1 for
a very wide strip (w/h ≫ 1).
(b) Plot the filling factor q vs. w/h for several values of ǫr (say, 1.0, 2.5 and 10), using
(8.100). Compare these values with those obtained from (9.91) and (9.92), and deter-
mine which has better accuracy.

p9-10 Use Fig. 6.20 and the results of section 9.4 to find the attenuation coefficient α of the TE0
mode in a slab waveguide with a = 2µm, n1 = 2.01, n2 = 2.0, λ0 = 0.5µm, ǫ′′1 /ǫ0 = 0.001,
and ǫ′′2 /ǫ0 = 0.002.

p9-11 For a weakly-guiding step-index optical waveguide of arbitrary cross-sectional shape, show
that
V db
q1 = b +
2 dV
Plot q1 vs. V for the TE0 mode of a dielectric slab waveguide, and compare it to a plot of b
vs. V for this same mode.

p9-12 Obtain from (9.66) an expression for the attenuation coefficient of a TM11 (rectangular)
waveguide mode due to finite wall conductivity. The H-field of this mode is
n π πx πy π πx πy o
H = H11 ux sin cos − uy cos sin .
d a d a a d

p9-13 There are many cases in which we desire to calculate wall losses of a mode which is degenerate
(or nearly so) with another mode of the same waveguide. One special example of this is at or
near cutoff, where both the forward and reverse modes’ propagation coefficients approach 0.
In such cases, it works out that the approximation of E−m and H−m which led from (9.24)
to (9.25) is inadequate and a more refined approach is required.

(a) Suppose that the mode (γm0 , Em0 , Hm0 ) is nearly degenerate with the mode (γn0 , En0 ,
Hn0 ). Use the approximation

E−m ≃ AE−m0 + BE−n0

H−m ≃ AH−m0 + BH−n0

in (9.63) to obtain a relationship between γm and the yet unknown constants A and B.
(b) Repeat the derivation of part (a), using

Ea = En0 e−γn0 z ; Ha = Hn0 e−γn0 z

for the “a” state fields instead of the fields indicated in (9.62). Use the approximation
for E−m and H−m suggested in part (a) above to obtain a second relationship between
γm , A, and B. Eliminate A and B between this result and that of part (a).
(c) Now assume that (γn0 , En0 , Hn0 ) is the reverse-traveling mode corresponding to (γm0 ,
Em0 , Hm0 ). That is, suppose that

γn0 = −γm0 ; En0 = E−m0 ; Hn0 = H−m0

From the results of parts (a) and (b) above, show that
q
γm ≃ ± γm0 2 + 2(α + α )γ
z T m0 + 4αz αT

307
 where
2
H
ZS Hm0ℓ dℓ
C0
αz = R
2 Em0 × Hm0 · uz dS
S0
2
H
ZS Hm0z dℓ
C0
αT = − R .
2 Em0 × Hm0 · uz dS
S0

p9-14 Use the techniques of problem p9-13 to derive a similar type of perturbation expression for
dielectric losses for a waveguide mode which may be near a cutoff frequency.
p9-15 (a) Obtain the following “complex power”-type of reciprocity relation:

∇ · (Ea × Hb∗ + Eb∗ × Ha ) = −Eb∗ · Ja − Hb∗ · Ma − Ea · Jb∗ − Ha · Mb∗

valid where ǫ and µ are real quantities. If ǫ and µ are complex, we may still use this
result if we absorb the effects of their imaginary parts ǫ′′ and µ′′ into the sources as
Ja,b = ωǫ′′ Ea,b , and Ma,b = ωµ′′ Ha,b . We can even allow ǫ′′ and µ′′ to be different for
states a and b if their effect is included into the source terms in this fashion.
(b) Use the result of part (a) to obtain the power-type orthogonality relation
Z
(Em × H∗n + En∗ × Hm ) · uz dS = 0
S0

valid for a lossless waveguide if γm 6= −γn∗ .

(c) Use the result of part (b) to provide an independent proof of the first line of (9.28).
p9-16 Use the result of problem p9-15(a) and the fields

Ea = Em0 e−γm0 z ; Ha = Hm0 e−γm0 z

Eb = Em e−γm z ; Hb = Hm e−γm z
where γm0 = jβm0 is the propagation coefficient of a mode of a lossless waveguide whose
dielectric constant is ǫ′ (state a) and γm = αm + jβm is that of a waveguide whose dielectric
constant has the slightly lossy value of ǫ′ − jǫ′′ (state b) to obtain the approximation

ωǫ′′ Em0 · Em0

∗
R
dS
αm = γm0 + γm ≃ R S0
∗
∗
2 S0 Em0 × Hm0 · uz dS

for the attenuation coefficient of the lossy waveguide mode. Show that this result is compat-
ible with (9.41)-(9.42).
p9-17 An open, two-wire line whose conductors have radius a and are separated by a distance b
as shown is made from copper conductors (ǫc = ǫ0 , µc = µ0 , σc = 5.8 × 107 mhos/m).

If a = 1 mm and b = 2 cm, find the attenuation coefficient of the TEM mode of this line at a
frequency of f = 30 MHz.

308
p9-18 Let S(z) be a surface in the plane z = constant bounded by the closed curve C(z). The
shapes of S(z) and C(z) do not change as z is varied; only a shift of their location is performed.

C(z)

S(z)

un

Prove the following consequence of the reciprocity theorem:

−W (a, b; z) + W (b, a; z)
d
Z
= (Ea × Hb − Eb × Ha ) · uz dS
dz S(z)
I
+ (Ea × Hb − Eb × Ha ) · un dℓ
C(z)

where un is the outward unit normal vector to C(z) and

Z
W (a, b; z) = (Ea · Jb − Ha · Mb ) dS
S(z)

is a sort of reaction density per unit length in the z-direction. (Hint: Integrate this expression
from 0 to z and compare with equation (9.8).)

p9-19 Obtain from (9.67) a formula for the attenuation coefficient α due to finite wall conductivity
σ of the TE11 and TE01 modes of an air-filled circular metallic waveguide of radius a. The
unperturbed mode fields are given in (5.82)-(5.84).

p9-20 Use the results in section 8.7.1 to obtain a formula for the filling factor q of the quasi-TEM
mode of CPW. Plot q vs. s/(s + 2w) for the cases (s + 2w)/h = 0.2, 1, and 4.

p9-21 A rectangular metallic waveguide of width a and height d has top and bottom walls (y = 0
and y = d) made of one metal, whose surface impedance is ZS1 , and side walls (x = 0 and
x = a) made of another (ZS2 ). Obtain a formula for the attenuation constant due to wall
loss of the TE10 mode of such a waveguide. If the top and bottom walls are made of silver
(σ = 6.17 × 107 S/m) and the sidewalls are made of an alloy of copper and zinc for which
σ = 2.56 × 107 S/m, find the value of α for a WR-90 size waveguide (a = 0.9 in, d = 0.4 in)
operating at f = 10 GHz. Compare this to the case when the same waveguide has all walls
made of the copper-zinc alloy.

309
p9-22 Obtain an “energy orthogonality” property for modes on an arbitrary waveguide by pro-
ceeding as follows. In (9.5), let the a state be the field of mode m of the waveguide, and the
b state be mode n. Integrate over the slice of waveguide shown in Fig. 9.4. Show that this
results in the orthogonality relation
1
Z
(ǫEm · En + µHm · Hn ) dS = 0 (if m 6= n)
4 S0
γm
= (if m = n > 0; forward mode)
jω
γm
= − (if m = n < 0; reverse mode)
jω

Note the connection with (5.29)-(5.30).

In the same way, obtain the energy orthogonality relation corresponding to (9.28) for a lossless
waveguide.
p9-23 Consider the two-wire transmission line shown in Fig. 8.8. Using the incremental inductance
rule, obtain a formula for the attenuation constant α of the TEM mode of this line due to
finite conductivity of the wires. If the medium surrounding the wires is assumed to be air,
determine the value of a/b for which this attenuation is minimum (regard the separation b
between wire centers as fixed while a is varied). What is the characteristic impedance of the
line under this minimum attenuation condition?

310
Chapter 10

EXCITATION OF WAVEGUIDES

Beauty is like a train that ceaselessly roars out of the

Gare de Lyon and which I know will never leave,
which has not left. It consists of jolts and shocks,
many of which do not have much importance, but
which we know are destined to produce one Shock,
which does.
—André Breton,
Nadja

Now that we have seen several examples of waveguides and the types of modes they can support, it
is relevant to ask how we can go about setting up mode fields in a guide. This question is answered by
specifying the sources to be used to set up the fields. The current sources may be electric or magnetic;
they may be actual currents flowing in a thin wire or equivalent currents distributed over some surface.
In this chapter, we will use the orthogonality property of waveguide modes to determine the amplitudes
with which modes are excited by a given set of sources.

10.1 Impressed Sources versus Induced Currents for Electro-

magnetic Fields
Up to this point, we have been concerned only with solutions of Maxwell’s equations in regions where
no sources are present. Of course, some kind of externally generated currents and/or charges must
be supplied somewhere in order to produce our mode solutions with nonzero amplitude, just as such
impressed sources are required to energize a circuit. In a time-harmonic problem, it is enough to specify
the current density, since the charge density will then follow by the law of conservation of charge.
It is well to be clear about what we mean by the term “externally generated” currents. One way of
writing Maxwell’s equations in a material medium is (for time-harmonic fields)
)
∇× E = −jωB
(10.1)
∇ × µB0 = jωǫ0 E + J

However, we must understand that the current density J in (10.1) includes, in addition to currents we
can “control” (generators, antennas, etc.), other currents caused by the presence of the electromagnetic
field in the medium. These are the conduction currents

Jcond = σE (10.2)

311
 xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
Jext xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
JS = un × H
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
perfect
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx un
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
conductor
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Figure 10.1: Induced surface currents on a perfect conductor.

the polarization currents

Jp = jω(ǫ − ǫ0 )E (10.3)
and the magnetization currents   
1 1
Jm = ∇ × − B (10.4)
µ0 µ
On the surface of a perfect conductor as shown in Fig. 10.1, surface currents
B
JS = un × (10.5)
µ
will also result as a limiting form of (10.2). The current densities (10.2)–(10.5) are ones we cannot
“control” directly because they depend on the field E, H present in the medium or on the conductor
surface. This in turn must be solved for from (10.1), which requires that we know J, which depends on
E and H, which were unknown in the first place.
To avoid such a vicious circle, it is common practice to account for (10.2) and (10.3) by way of
material parameters µ, σ and ǫ and to rewrite (10.1) in the form

∇ × E = −jωµH
(10.6)
∇ × H = jωǫ̂E + Jext

where the H field is defined using the permeability µ:

B = µH

and
Jext = J − Jcond − Jp − Jm (10.7)
are called externally-impressed (or simply impressed) currents whose value we know or control, even
before we obtain a knowledge of E and H by solving Maxwell’s equations. Hereafter we will drop the
subscript “ext” on impressed electric or magnetic currents J or M except when necessary for clarity or
emphasis.
Of course, no such current sources can be achieved in practice because the environment in which we
place a real source will always affect it to some degree. In this sense, the concept of an impressed source
is very much like that of an ideal voltage or current generator in circuit theory. We will refer to currents
such as (10.2)–(10.5), which disappear when the impressed sources (and hence the resultant fields) are
turned off, as induced currents. An example illustrating the distinction between impressed currents and
induced currents will be studied in some detail in the next section in connection with Love’s equivalence
theorem.

312
10.2 Equivalence Principles
Equivalence principles analogous to those of circuit theory play an important role in electromagnetics,
and come in many varieties. For example, to see how magnetic currents might be useful, consider a field
E, H that is a solution of the “ordinary” Maxwell’s equations (F.3). Now consider a hypothetical field
E1 , H1 given by
J
E1 = E+
jωǫ
H1 = H (10.8)

(Note that this field is identical to E, H wherever J = 0, i. e., outside the region containing the sources.)
Evidently, E1 , H1 satisfy
 
J
∇ × E1 = −jωµH1 + ∇ ×
jωǫ
∇ × H1 = jωǫE1 (10.9)

That is (see (F.4)), these fields act as though they were produced by the equivalent, purely magnetic
current density  
J
Meq = −∇ × (10.10)
jωǫ
So long as J and our fictitious Meq are zero at a given observation point, the original field E, H is
precisely that which would be produced by Meq instead of by J. Often, the equivalent magnetic current
given by (10.10) proves to be simpler to describe than the original J. In such cases, it proves more
convenient to use Meq directly. This result is an equivalence principle. Such principles are often used in
the formulation of electromagnetic problems to simplify concepts or computations and to increase our
insight and understanding of the physics involved.
Next, consider a set of sources J and M located inside a closed surface S as shown in Fig. 10.2(a).
Let E, H be the fields produced by these sources. Let E1 , H1 be defined as
(
E at points exterior to S;
E1 =
0 at points interior to S
(
H at points exterior to S;
H1 =
0 at points interior to S.
Now E1 and H1 satisfy the source-free Maxwell equations both inside and outside of S but are discontin-
uous across S. If they are to represent a legitimate electromagnetic field, there must be surface currents
(along with appropriate surface charges) present on S to support this discontinuity. From section F.1 of
Appendix F, these surface currents must be

JS,eq = un × H|S +
MS,eq = E × u n |S + (10.11)

Hence, the electric or magnetic field produced by the sources J and M inside S is exactly the same as
that produced by the equivalent sources (10.11) on S, observed at any point outside S. This statement is
known as Love’s equivalence principle.
The fields E1 and H1 of the equivalent problem (Fig. 10.2(b)) are identically zero throughout the
whole interior of S. Hence, Love’s equivalence principle will be most useful to us when there is a surface
S enclosing the original sources, inside which we do not need to know the fields and on which we know,
at least approximately, the fields produced by the original sources, so that the equivalent sources (10.11)

313
 S
S
(x0, y0, z0)
(x0, y0, z0)
J, M

no sources
JS,eq

MS,eq
un
field = 0

interior of S same field and

sources as original
exterior of S

(a) (b)

Figure 10.2: (a) Original sources; (b) Equivalent sources for observation point (x0 , y0 , z0 ) outside of S.

can be treated as if they were impressed ones. In general, even though the original sources may be pure
electric currents, we must expect that tangential components of both E and H will exist on S, and so
both electric and magnetic equivalent sources will be required.
We can see that instead of choosing E1 and H1 to be zero inside of S, we could have made them
equal to any convenient functions. In this case, we would require generally different distributions of
surface current on S to support the field discontinuities, as well as volume source distributions J1 and
M1 to support the nonzero field:

J1 = ∇ × H1 − jωǫ̂E1
interior to S (10.12)
M1 = −∇ × E1 − jωµH1

A great degree of flexibility is thus possible in the use of equivalence principles. We will explore some
variations on this theme by way of an example.

10.2.1 Example: Impressed and induced sources at a perfectly conducting

surface
In Fig. 10.2(b), it can make no difference what medium is located within S (where the original sources
were), since in the equivalent problem the field is zero there. In particular, let us put a perfectly
conducting surface just inside of S. If we do this, we still have the impressed current MS = E × un
as before, but we shall see that the surface electric current on S can be taken either as the impressed
current Jext
S = JS = un × H that is present in Fig. 10.2(b) or the induced current JS = un × H that
results when only a magnetic current Mext S = MS is impressed at S.
Thus, the situation in Fig. 10.3(a) is obtained by placing a perfect conductor just inside S, which
already had zero fields in the situation of Fig. 10.2(b). Both Jext ext
S and MS are still impressed sources,
and the total field resulting from these sources is the superposition of the fields resulting from Jext
S and
Mext ext
S separately. We have already shown in section 9.1.1 that an impressed current JS on a perfectly
conducting surface produces no external fields. That result is similar to, though distinct from, image
theory (Appendix F), which is itself a form of equivalence principle. Thus it must be that whatever
fields are present in Fig. 10.3(a) must be produced by Mext S alone. However, the boundary conditions

314
 S S
xxxxxxxxxx
xxxxxxxxxx
xxx
un un
xxxxxxxxxxx
xxxxxxxxxx
xxx
xxxxxxxxxx xxxxxxxxxxx
xxx
xxxxxxxxxxx
xxx
xxxxxxxxxx
xxxxxxxxxx (x0, y0, z0) xxxxxxxxxxx
xxxxxxxxxxx (x0, y0, z0)
xxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx
ext xxxxxxxxxxx
xxxxxxxxxxx JS (induced)
xxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx
JS (impressed) xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
PEC: xxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
E, H = 0 xxxxxxxxxx
xxxxxxxxxx ext PEC: xxxxxxxxxxx
xxxxxxxxxxx ext
xxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx
MS (impressed) E, H = 0
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx MS
xxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
(impressed)
xxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxx ext xxxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx
H=JS ×un xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx H=0 xxxxxxxxxxx
xxxxxxxxxxx ext
xxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
xxxxxxxxxxx
H=JS ×un
xxxxxxxxxxx

(a) (b)

Figure 10.3: (a) Impressed electric currents on S producing zero field between S and perfect conductor;
(b) Induced surface currents on perfect conductor (produced by MextS on S alone) with nonzero H-field
between S and perfect conductor.

and uniqueness theorems for the electromagnetic field (Appendix F) still require that the discontinuity
in tangential H field from the right side of S to the interior of the conductor (where it is zero) must be
compensated by a surface electric current. Since this current is no longer impressed on S, it must be
induced in the surface of the perfect conductor, as shown in Fig. 10.3(b). That is, the surface current is
now a response rather than an excitation. By the uniqueness theorem for the fields, this induced current
JS must be the same as the equivalent impressed current JextS in Fig. 10.3(a). This result, known as the
Schelkunoff Induction Theorem, is the electromagnetic analogy of the situation in the circuit problem
of Fig. 3.2. The electric impressed sources, like the current generator I, have no effect in the region in
which we are interested.

10.2.2 Example: Thévenin and Norton types of equivalence theorems for

electromagnetic fields
Suppose that a thin perfectly electrically conducting (PEC) sheet is placed at the surface S in Fig. 10.2(a),
and that the sources J, M acting inside it produce the “short-circuit” surface current density JS,sc in
the sheet, as shown in Fig. 10.4(a). Then the field outside of S in the original problem is the same as
that produced by the equivalent electric surface currents JS,eq = −JS,sc as shown in Fig. 10.4(b) in the
exterior of S, without the PEC present and with the original sources deactivated. This is the analog of
the Norton equivalence theorem in circuit theory.
On the other hand, suppose that a sheet which is a perfect magnetic conductor (PMC) (that is, on
which un × H = 0) is placed at S, resulting in the “open-circuit” magnetic surface current density MS,oc
on the sheet, as shown in Fig. 10.5(a). The field exterior to S is now the same as that produced by
the equivalent magnetic surface current source MS,eq = −MS,oc located at S, without the PMC present
and with the original sources deactivated. This is the electromagnetic analog of Thévenin’s equivalence
theorem.
Note that, in contrast with the Love equivalence principle, the fields on the interior side of S are not
generally zero for the Norton and Thévenin equivalents, nor are they the same as in the original problem.
It is thus not generally justified to make changes in media (such as inserting conducting bodies) inside S

315
 S: PEC
S: no PEC

J, M JS,sc(induced) JS,eq = -JS,sc

no sources

field ≠ 0, but differs

from original

same field and

sources as original

(a) (b)

Figure 10.4: (a) Original currents inside S radiating with S “short-circuited” by a PEC; (b) Norton
equivalent surface electric currents radiating at S without PEC present.

S: PMC
S: no PMC

J, M MS,oc(induced) MS,eq = -MS,oc

no sources

field ≠ 0, but differs

from original

same field and

sources as original

(a) (b)

Figure 10.5: (a) Original currents inside S radiating with S “open-circuited” by a PMC; (b) Thévenin
equivalent surface electric currents radiating at S without PMC present.

316
after replacing the original sources by these equivalents, as it was for the case of Love equivalent sources.
The unchanged medium inside S is the analog of the Thévenin or Norton equivalent impedance from
circuit theory.

10.2.3 Example: Tangential magnetic and normal electric surface sources

Consider a surface concentration of normal electric current:

J = JSn un δ(n) (10.13)

According to (10.10), this source is equivalent to the (tangential) magnetic surface sources
   
JSn un δ(n) JSn
Meq = −∇ × = un × ∇t δ(n) (10.14)
jωǫ jωǫ

Now suppose that the electric field has been expressed in terms of scalar and vector potentials as:

E = −∇Φ − jωA

Suppose also that we have used the Love equivalence principle and the induction theorem as in sec-
tion 10.2.1 to arrive at an equivalent surface magnetic source

MS,eq = Et × un |n=0+ = un × ∇t Φ|n=0+ − jωAt × un |n=0+ (10.15)

in front of a PEC. The term due to the scalar potential has the same form as (10.14), and so it can be
further made equivalent to a surface distribution of normal electric currents. The final set of equivalent
sources is then
JSn,eq = jωǫΦ|n=0+ MS,eq = − jωAt × un |n=0+ (10.16)

10.2.4 Example: Large apertures and the Kirchhoff approximation

When an electromagnetic wave is incident at an aperture which is large in extent (compared to a
wavelength and to all other characteristic dimensions of the problem), it is sometimes possible to utilize
the Kirchhoff approximation to simplify the analysis. To be more specific, consider the situation shown
in Fig. 10.6. In addition to the large aperture, there may be obstacles such as the dielectric shown in the
figure which also affect the field. When making the Kirchhoff approximation, we assume that we can
neglect the backscattered field behind the aperture as compared to Einc , Hinc as far as computing the
aperture field E0 , H0 is concerned. Because E0 and H0 are being approximated, any fields we compute
in front of the aperture using Love’s equivalent sources will also be approximate. In fact, if we now fill in
the aperture with (say) a perfect electric conductor, use of approximate Kirchhoff equivalent sources will
in general lead to a somewhat different field computed outside the aperture. The discrepancy between
the two computed fields is one measure of the error we commit by making the Kirchhoff approximation.
We will illustrate the use of the Kirchhoff approximation in section 10.5.1.

10.2.5 Example: Equivalent dipoles for electrically small apertures

The usefulness of the equivalence principle in any of its variants depends upon the extent to which we
may know accurately the values of electric or magnetic fields on a given surface. For electrically large
apertures, a Kirchhoff-type approximation may be appropriate, wherein some simple incident field is
used instead of the exact field (incident plus scattered) at the surface containing the aperture. When
the aperture is small electrically, the fields near the aperture tend to be strongly dependent upon the
shape of the aperture as well as upon the incident field.

317
 dielectric body

incident wave Einc, Hinc

surface S:
E0 ≅ Einc; H0 ≅ Hinc

Figure 10.6: Kirchhoff approximation for a wave incident at a large aperture.

Small aperture coupling has many important applications in the design of resonators, directional
couplers and measurement devices. The analysis of small apertures is carried out using Bethe’s small-
aperture theory, in which the equivalent sources MS and JS at the aperture are replaced by an appro-
priately chosen set of Hertzian dipoles.
Consider the situation shown in Fig. 10.7. A perfectly conducting plane is located at n = 0, where
n is the normal distance from the plane as shown. The region n < 0 under the plane has material
parameters ǫ− and µ− , while above it (n > 0) the parameters are ǫ+ and µ+ . The plane itself is
described by the cartesian coordinates (or other arc-length coordinates) (x1 , x2 ) so that (x1 , x2 , n) form
a right-handed coordinate system. Some given impressed sources are present which produce what we
will call the “short-circuit” field Esc , Hsc in the presence of the plane but with no aperture cut into it
(that is, the aperture is metallized or “shorted out”).
When an electrically small aperture A is cut into the plane, the field produced by these impressed
sources is perturbed and is now equal to

Etot = Esc + EA

(10.17)
Htot = Hsc + HA

where EA and HA are due to the presence of the aperture. The qualitative behavior of these fields is
illustrated in Fig. 10.8. Fig. 10.8(a) shows a short-circuit electric field which is nonzero only on the left
side of the screen (n < 0). Esc is, of course, normal to the screen. After the aperture is cut into the wall,
the electric field fringes and penetrates through the screen, roughly as shown in Fig. 10.8(b). A small
tangential electric field is present in the aperture, and by the equivalence theorem we could express EA
and HA in n < 0 (or n > 0) by an appropriate surface magnetic current impressed at n = 0− (or n = 0+ ,
respectively) and with the aperture closed off by a perfect conductor.
This tangential aperture electric field, however, is rather hard to determine in simple form, and it is
more convenient to use a set of equivalent sources like those of section 10.2.3 in order to compute EA
and HA . Upon examining Fig. 10.8(b), we see that EA on either side of the screen closely resembles

318
 aperture
n>0: (ε+, µ+) n

A x2

x1

conducting screen S
n=0

n<0: (ε-, µ-)

Figure 10.7: Aperture in a conducting plane.

p- p+
Esc un

Etot EA
(n<0) (n>0)
S
(a) (b) (c)

m- m+
Hsc
un

HA
Htot

(n<0) (n>0)
S
(d) (e) (f)

Figure 10.8: Field fringing at a small aperture in a conducting wall (after Collin, 1966).

319
the electric field that would be produced by a Hertzian electric dipole (cf. Appendix F) normal to the
aperture, with the aperture closed off once again (“metallized”) behind it, as indicated in Fig. 10.8(c).
Likewise, a short-circuit magnetic field in n < 0 (Fig. 10.8(d)) will fringe as shown in Fig. 10.8(e) when
the aperture is cut. The field HA which results on either side is nearly the same as the magnetic field
of a Hertzian magnetic dipole with the aperture metallized behind it, as in Fig. 10.8(f).
To justify the use of dipoles as equivalent sources, we suppose that JSn,eq and MS,eq have been
obtained as in (10.16), with the potentials chosen so that Φ = 0 and At = 0 on the screen S, so that the
equivalent sources differ from zero only in the aperture A. If the linear dimensions of the aperture are
small compared to a wavelength, the potentials near the aperture can be found approximately from the
static field equations. Moreover, if our observation point is not too close to the aperture, the distributed
equivalent sources can be approximately lumped into dipoles concentrated at some point (x10 , x20 ) in
the middle of the aperture. At n = 0+ , we put

MS,eq ≃ jωµ+ m+ δ(x1 − x10 )δ(x2 − x20 ) (10.18)

JS,eq ≃ jωp+ δ(x1 − x10 )δ(x2 − x20 ) (10.19)

as given in equations (F.14) and (F.15). The dipole moments p+ and m+ are chosen so as to give the
same integrated values of JS,eq and MS,eq over the plane n = 0. Hence
Z Z
jωµ+ m+ = MS,eq dS = jω un × At dS (10.20)
A A

or
1
Z
m+ = un × At dS (10.21)
µ+ A
Similarly, Z
p+ = ǫ + un Φ dS (10.22)
A

We need dipoles on both sides of the closed-off aperture (at n = 0± ) in order to produce the field
E , HA in both n > 0 and n < 0. At n = 0+ , we place the dipoles p+ and m+ , while at n = 0− , we
A

place the dipoles p− and m− which are found from the values of Φ and At in the aperture similarly to
p+ and m+ . Because the potentials must be continuous through the aperture, the dipole strengths on
the two sides turn out to be related by
p− p+
=− (10.23)
ǫ− ǫ+
and
µ− m− = −µ+ m+ (10.24)
Most generally, the short-circuit field will exist on both sides of the screen. The strengths of the
equivalent dipoles depend on these short-circuit field values as follows:
p+ 2 +
= − αE [Dsc ]0n=0−
ǫ+ ǫ+ + ǫ−
2µ+ µ− ↔ 0+
µ+ m + = α M · [Hsc ]n=0− (10.25)
µ+ + µ−

while p− and m− are given by (10.23) and (10.24). The jumps [Dsc ] and [Hsc ] of the short-circuit field
across the screen are related to the surface charge and current densities which exist when the aperture
is metallized:
0+ 0+
ρS = un · [Dsc ]n=0− ; JS = un × [Hsc ]n=0−
The coefficient αE in (10.25) is known as the electric polarizability of the aperture, while the dyadic
↔
(or tensor) (essentially a 2 × 2 matrix here) α M is called the magnetic polarizability of the aperture.

320
Their dimensions are those of volume; in a sense they are effective volumes—volumes of equivalent
scatterers which can produce the same polarization and magnetization (approximately) as the dipoles
(10.23)-(10.25). The polarizabilities of various aperture shapes can be calculated from the corresponding
↔
static field problems. We give a brief list of some of these in Table 10.1. Values of αE and α M for other
apertures can be found in the references at the end of the chapter.

321
 ↔
Aperture shape αE αM
10.1.1 Circle of radius r0
2 4
3 r03 3 r03 (u1 u1 + u2 u2 )
x2

r0 x1

10.1.2 Ellipse with

πr13 (1 − e2c ) πr13 e2c

semiaxes r1 > r2 u1 u1
3E(ec ) 3 K(ec ) − E(ec ) 
(1 − e2c )u2 u2
+
E(ec ) − (1 − e2c )K(ec )
x2

r1
x1
r2
p
ec = 1 − r22 /r12 K is given by (B.46), E by (B.47)
10.1.3 Rectangle  
2 2
pw l 4 l 3 u1 u1 3
+ w −1
u2 u2
6 w2 + l 2 9 −1
sinh (l/w) sinh (w/l)
x2
w x1
l

10.1.4 Greek cross h p √ i

w2 l(2l − w) 2 l2 + w2 + l 2 3 2 3
4w (l + w l − w )(u1 u1 + u2 u2 )
2 2 9l w w l
12(l + 2w )
sinh−1 + sinh−1 − A
l l w
x2

l w x1
w
l √
where A = ln(1 + 2) ≃ 0.8814

Table 10.1: Polarizabilities for electrically small apertures in an infinite plane screen located at n = 0.
Expressions given for the circle and ellipse are exact; the remainder are the best available approximations.

322
 Figure 10.9: Plane wave normally incident at a small aperture in a planar conductor.

The dipoles (10.25), (10.23) and (10.24) succeed only approximately in determining EA and HA .
They are strictly valid only for apertures in infinite perfectly conducting planes and for static fields.
They do not accurately predict the perturbing field very close (within a few aperture diameters) to
the aperture. Nevertheless, we may use the Bethe theory for a more general surface provided that the
aperture is located at a place where the radius of curvature is large compared to the aperture size and
away from corners and bends in the surface.
We may use the theory for nonzero frequency provided that the largest linear dimension (i. e.,
diameter) of the aperture is small compared to a wavelength. The approximate nature of (10.25) becomes
apparent when conservation of energy is examined. As the reader may verify, e. g., for the simple example
of plane wave excitation studied below, the field EA , HA will generally not satisfy conservation of energy
exactly, the discrepancy tending to zero as the ratio of aperture diameter to wavelength goes to zero.
In some applications it may be desirable to remedy this defect (when calculating quality factors of
resonators, for example), and a means to do so is given in Appendix N.
To illustrate the use of (10.23)-(10.25), consider the situation of Fig. 10.9. A normally incident plane
wave in z < 0 is reflected by a perfectly conducting plane at z = 0, in which is cut a small aperture
centered at x = 0, y = 0. The media on both sides of the plane are supposed to be free space. With the
aperture closed off, the short-circuit field is the sum of the incident wave Ei , Hi and the reflected wave
Er , Hr :
Esc = Ei + Er = ux E0 (e−jk0 z − e+jk0 z ) (z < 0)

(10.26)
Hsc = Hi + Hr = uy (E0 /ζ0 )(e−jk0 z + e+jk0 z ) (z < 0)
0+
The short-circuit fields are equal to zero in z > 0. From these equations, [Dsc ]n=0− = 0 while
0+
[Hsc ]n=0− = −2uy E0 /ζ0 (10.27)
where we have taken n = z. By (10.25), therefore, there will be no electric dipoles, while the magnetic
dipoles
↔ 0+ E0
m+ = α M · [Hsc ]n=0− = −2 (ux αMxy + uy αMyy ) (at z = 0+ ) (10.28)
ζ0

323
 Type of mode Zm
10.2.1 TE mode of metallic wave- ζTE
guide
10.2.2 TM mode of metallic wave- ζTM
guide
10.2.3 TEM or quasi-TEM modes Zc

Table 10.2: Common choices for Zm for waveguide modes.

and
E0
m− = 2 (ux αMxy + uy αMyy ) (at z = 0− ) (10.29)
ζ0
radiate in the presence of the conducting plane with the aperture closed off. The fields EA and HA due
to the aperture are thus found by using image theory (a dipole of twice the strength radiates in infinite
free space) as described in Appendix F.
Other examples of the application of small aperture theory will be given later in this chapter.

10.3 Equivalent Transmission Lines—Normalization and Char-

acteristic Impedance
The analogy between waveguide modes and equivalent transmission lines introduced in section 9.3 is
incomplete in one respect: there is no definition for the characteristic impedance of an arbitrary wave-
guide mode. If a suitable definition for a characteristic impedance Zm for any mode of a waveguide
could be made, the discussion above could be extended to define a voltage and current associated with
the mode, just as is done for the TEM case. The analogy between each waveguide mode and a classical
transmission line suggested in eqns. (9.31) and (9.32) would then be complete. The trouble is, that
there is no obviously unique way to define Zm for a non-TEM mode.
Although any definition of Zm for a non-TEM mode is not unique, some constraints do exist. From
section K.2 of Appendix K, we have that γm P̂ for any normalizable mode of a lossless waveguide is
pure imaginary, and therefore if an expression analogous to (8.37) for complex power is to hold for our
equivalent transmission line of a mode, along with condition (9.28), we must have that Zm is real if
γm = jβm (a propagating mode), or imaginary if γm = αm (a cutoff mode). Note that this implies that
a single cutoff mode carries no time-average power in the z-direction, since the complex power will be
imaginary, as was shown in another way in section 9.3.
Other than this, the precise value of Zm is essentially unrestricted for non-TEM modes. There are,
however, certain choices for Zm which are traditionally made in specific cases, and these are shown in
Table 10.2.
Considering the field representations (9.31) and (9.32), let us make some choice for Zm , and take
it to be the characteristic impedance of an equivalent classical transmission line. We can thus rewrite
(9.31) and (9.32) as:

ET = EmT (x, y)Ṽm (z) 
HT = HmT (x, y)I˜m (z)


(10.30)
Ez = Emz (x, y)I˜m (z)  
Hz = Hmz (x, y)Ṽm (z)


where following section 1.2 for the classical transmission line,

Vm (z)
Ṽm (z) = √ = am e−γm z + bm eγm z (10.31)
2Zm

324
 Figure 10.10: Equivalent transmission-line representation of a waveguide.

r
Zm
I˜m (z) = Im (z) = am e−γm z − bm eγm z (10.32)
2
The constant factors involving Zm have been introduced into (10.32) so that, as suggested by (5.27), we
will have
1 1
Z
E × H · uz dS = Vm Im
2 S0 2
and when (9.28) holds as well, we will have

1 1
Z
E × H∗ · uz dS = Vm Im
∗
2 S0 2

as would be the case for TEM modes.

Thus, we have a complete, formal analogy between each set of forward/reverse (±z-traveling) modes
on a uniform section of waveguide, and an equivalent transmission line with parameters Zm and γm . The
entire waveguide is thus modeled by an infinite set of completely decoupled equivalent transmission lines,
as shown in Fig. 10.10. By differentiating (10.31) and (10.32), we can easily obtain a set of telegrapher’s
equations for these equivalent lines, in the form (cf. (1.4)):

dVm (z)
= −γm Zm Im (z) (10.33)
dz
dIm (z) γm
=− Vm (z) (10.34)
dz Zm
Clearly, we may identify γm Zm as a series impedance per unit length of the equivalent line, while
γm /Zm serves as a shunt admittance per unit length. The voltage Vm (z) and current Im (z) may almost
be regarded as quantities of physical significance, since the fields may be obtained from them by using
(10.30).

325
 Figure 10.11: Sources in a waveguide.

10.4 Expansion of the Field as a Sum of Modes

We are now in a position to calculate the excitation amplitudes of the modes in a general waveguide
in a manner analogous to that for the classical transmission line in section 3.1. Consider the situation
of Fig. 10.11. A collection of impressed current sources J and M is distributed in a waveguide over a
volume contained between the planes z = z1 and z = z2 . Outside the source region, i. e., in z ≤ z1 or
z ≥ z2 , it seems reasonable that any possible field can be expressed as a sum of mode fields (which are
themselves source-free solutions of Maxwell’s equations). If there are no sources at infinity, then to the
right of z2 we expect to have only forward modes which propagate or attenuate in the +z direction:
X 
E = am0 Em (x, y)e−γm z 

m>0
X (z > z2 ) (10.35)
H = am0 Hm (x, y)e−γm z 

m>0

The amplitudes of Hm are the same as those for Em since Maxwell’s equations would otherwise not be
satisfied. We sum over only positive m to indicate that only forward modes are present. Similarly, to
the left of z1 , the most general field should be a sum of reverse modes which propagate or attenuate in
the −z direction: X 
E = bm0 E−m (x, y)eγm z 


m>0
X (z < z1 ) (10.36)
H = bm0 H−m (x, y)eγm z 


m>0

What we will assume more generally here is the so-called completeness property of the modes to
describe the transverse components of the total field in a uniform waveguide with sources of bounded
extent (Fig. 10.11). This assumption is a generalization of our assumption (3.10) for the classical
transmission line, and of representation (10.30) for the transverse field components in terms of equivalent
transmission lines: X
ET = Ṽm (z)EmT (x, y) (10.37)
m>0

I˜m (z)HmT (x, y)

X
HT = (10.38)
m>0

where 
Vm (z)
Ṽm (z) = √ 2Zm
= am (z)e−γm z + bm (z)eγm z 
q (10.39)
˜
Im (z) = Zm
am (z)e−γm z − bm (z)eγm z 
2 Im (z) =

326
 Sb
Sf
un

un
un z
V sources

zb zf

Figure 10.12: Application of Lorentz reciprocity to a waveguide containing impressed sources.

From (10.35) and (10.36) it is clear that

am (z) ≡ 0 ; bm (z) ≡ bm0 for z ≤ z1 (10.40)

bm (z) ≡ 0 ; am (z) ≡ am0 for z ≥ z2 (10.41)

The completeness property (10.37)-(10.38) for the transverse fields is far from obvious, especially within
the source region. In cases where this property has been proved, the proof is quite difficult; we will not
attempt to describe such a proof here. There are, in fact, many waveguides for which the completeness
property has not been proved at all. Nevertheless, the property is routinely assumed to be true anyway
on the grounds that it is physically reasonable.
In cases like microstrip or dielectric waveguides, some or all of whose “sidewall” boundaries extend
to infinity in the transverse plane, the discrete modes of propagation are generally not complete by
themselves (as we have mentioned already in chapter 6), and must be supplemented by a spectrum of
continuous or radiation modes. Although such modes can be treated (at least formally) in a manner
similar to the discrete modes, the details are rather delicate. To get a feel for what is involved in
the extension to continuous modes, observe that in place of the sums (10.35) and (10.36), we have to
represent the fields as integrals over the continuous spectrum of radiation modes in addition to a sum
over the discrete modes, e. g.:
X
E = am0 Em (x, y)e−γm z
m>0
Z k2 Z ∞
−jβz
+ a0 (jβ)E(jβ; x, y)e dβ + a0 (α)E(α; x, y)e−αz dα
0 0

where E(γ; x, y) is the field of a radiation mode of propagation coefficient γ as discussed in sections 6.9
and 9.2, and a0 (γ) is the wave amplitude of that spectral component of the field. Further details of this
procedure can be found in the references at the end of the chapter.
We will now apply Lorentz’ reciprocity theorem to the volume V illustrated in Fig. 10.12. The surface
S bounding V consists of the back end plane Sb at zb and the front end plane Sf at zf , together with
the sidewall of the guide for zb < z < zf . Here zb and zf are two values of z chosen for convenience; we
will specify them below. For the “a” state, we take the actual fields produced in the guide by the given

327
sources; their transverse components are given by (10.37)-(10.38). For the second state, which we will
call the “n” state, we take the field of a single mode of the guide with index n:
En = En (x, y)e−γn z
n
H = Hn (x, y)e−γn z (10.42)
We insert these fields into (9.8). The surface integral over the sidewall vanishes since un × Ea =
un × En = 0 on this wall. Moreover, the “n” state sources are identically zero, and so we have
X  Z Z 
Ṽm (zf ) ˜
Em × Hn · uz dS − Im (zf ) En × Hm · uz dS e−γn zf
m>0 S0 S0
 Z  Z
− Ṽm (zb ) Em × Hn · uz dS − I˜m (zb ) En × Hm · uz dS e−γn zb
S0 S0
Z zf Z
−γn z
= e dz [En · J(x, y, z) − Hn · M(x, y, z)] dS (10.43)
zb S(z)

where S0 is a generic cross-section of the waveguide, while S(z) is the cross-section at the plane z.
Now by (9.24), the cross-section integrals on the left side of (10.43) will vanish by orthogonality
unless m = |n|. The sums in (10.43) reduce to single terms, and these nonzero cross-section integrals
are evaluated from the normalization condition (5.27). As a result, (10.43) becomes:
Z zf Z
e−γn z dz [En · J(x, y, z) − Hn · M(x, y, z)] dS
zb S(z)
= 4[b|n| (zf ) − b|n| (zb )] if n > 0
= −4[a|n| (zf ) − a|n| (zb )] if n < 0 (10.44)
where (10.39) has been used. Consider the case where zf = z2 for n > 0 and where zb = z1 for n < 0;
we have
b|n| (zf ) = a|n| (zb ) = 0
by (10.40) and (10.41). Then taking zb = z for n > 0 and zf = z for n < 0, we have for the functions
am (z) and bm (z):
am (z) =
(Z )
z
1
Z
γm z ′ ′ ′
− e [E−m · J(x, y, z ) − H−m · M(x, y, z )] dS dz ′
4 z1 S(z ′ )

(z > z1 ) (10.45)

bm (z) =
(Z )
z2
1
Z
−γm z ′ ′ ′
− e [Em · J(x, y, z ) − Hm · M(x, y, z )] dS dz ′
4 z S(z ′ )

(z < z2 ) (10.46)
Eqns. (10.45) and (10.46) are seen to be fully equivalent to (3.14) and (3.15) for the classical trans-
m m
mission line if we identify the equivalent distributed sources Eext (z) and Hext (z) for the ±m mode pair
as:
r
Zm
Z
m
Eext (z) = (Emz Jz − HmT · MT ) dS
2 S(z)
1
Z
m
Hext (z) = √ (Hmz Mz − EmT · JT ) dS (10.47)
2Zm S(z)

328
along with the identifications γ → γm and Zc → Zm . Inserting (10.45) and (10.46) into (10.39), and
thence into (10.37) and (10.38) gives us a complete solution for the transverse fields due to any impressed
sources J and M, in terms of all the modes of the waveguide. The description of each mode amplitude
is fully analogous to an equivalent classical transmission line.

10.4.1 Representation of the longitudinal field

The completeness property (10.37)-(10.38) which generally holds for transverse fields is not always true
for the z-components of E and H. This is related to the fact that Ez and Hz do not enter into the
orthogonality property (9.24) which is used to obtain the amplitudes in (10.37) and (10.38). We must
instead infer expressions for Ez and Hz from (10.37)-(10.38) and Maxwell’s equations. From the z-
components of the Maxwell curl equations (9.1) and (9.2), we obtain

Jz 1
Ez = − + uz · ∇T × HT (10.48)
jωǫ jωǫ
Mz 1
Hz = − − uz · ∇T × ET (10.49)
jωµ jωµ
We insert the expansions (10.37) and (10.38) for ET and HT into (10.48) and (10.49), and make use of
(5.18) and (5.20). Our result is:

Jz
I˜m (z)Emz (x, y)
X
Ez = − + (10.50)
jωǫ m>0

Mz X
Hz = − + Ṽm (z)Hmz (x, y) (10.51)
jωµ m>0

In other words, the z-components, in addition to containing a sum of mode fields as in (10.30), must
also contain terms corresponding to the z-components of the impressed currents. Outside of the source
region, these extra terms are naturally zero, and a sum of modes is sufficient to represent the entire
field, as we should expect. In some situations, however, especially when equivalent sources are used in
the formulation of a problem, the extra terms in (10.50) and (10.51) may be important.

10.4.2 Optimizing excitation of a given mode

Examination of the formulas for excitation of a waveguide reveals some guidelines for enhancing the
excitation of a particular mode relative to that of other possible modes. A given mode is excited
primarily in the +Z rather than −z direction using the methods discussed in section 3.2, together with
m m
eqns. (10.47). The source distributions J and M should be chosen to maximize Eext and/or Hext for
the desired mode while minimizing those for undesired modes.
As an example, consider the excitation of a rectangular metallic waveguide by a current probe—a
wire carrying a filamentary electric current I (Fig. 10.13). Examining (10.45)-(10.46) or (10.47), we see
that a current of given magnitude will have maximum effect in exciting a given mode if it is directed
parallel to the electric field and located at a position where the magnitude of that field is maximum. For
a TE10 mode of the rectangular guide, this would be a vertically oriented wire located at the center of
the guide cross section as shown in Figure 10.13(a). If a second wire carrying an equal but oppositely
directed current were similarly positioned in the guide at a distance l separated from the first in the
z-direction as shown in Fig. 10.13(b), then from (10.45), the total amplitude with which the mode is
excited would differ from that of the single wire excitation by a factor of (1 − ejβl ). This becomes zero
if l is an integer multiple of a guide wavelength λg , but is maximized if l is an odd number of half-guide
wavelengths. Since the configuration of Fig. 10.13(b) is equivalent to that of Fig. 10.13(c) (where the
original probe radiates a distance l/2 in front of a short-circuiting plane) by image theory, it can be seen

329
Figure 10.13: Current probe excitation of rectangular metallic waveguide: (a) Single probe; (b) Two
probes separated by a distance l; (c) Image equivalent of (b).

that a single wire can be made to launch the dominant mode of this guide quite efficiently by placing it
λg /4 in front of such a short circuit.
If the wire is the inner conductor of a coaxial line, we appear to have obtained a suitable means of
transition between the coax and the rectangular waveguide as shown in Fig. 10.14. In fact, positioning
the coaxial probe exactly λg /4 from the waveguide short circuit is not a good choice from the point
of view of an impedance match, but fortunately a small deviation of this position from λg /4 does not
change the mode excitation very much, and allows the reflection from this junction to be reduced by
adjusting it properly.
Similar considerations apply to excitation of other kinds of waveguide. Examples will be studied
later in the chapter.

10.5 Excitation by Given Arbitrary Aperture Fields

If the excitation currents are confined to a transverse plane (the cross-section z = 0, for instance), then
the volume integrals (10.45) and (10.46) are replaced by surface integrals. For instance,

1
Z
am (z) ≡ am0 =− (E−m · JS − H−m · MS ) dS (z > 0) (10.52)
4 S0

Suppose now that we take these to be the equivalent surface currents corresponding to fields produced
by sources radiating in z < 0. By the Love equivalence theorem,

JS = uz × H0 ; M S = E 0 × uz

330
 Figure 10.14: Coaxial-to-rectangular waveguide transition.

Figure 10.15: (a) Perfect conductor placed behind equivalent sources at z = 0; (b) Equivalent source
plus image radiating in infinite guide.

where E0 and H0 are the actual fields present at z = 0 (the so-called aperture field ). Then from (10.52),

1
Z
am0 = − (E0 × H−m − E−m × H0 ) · uz dS (10.53)
4 S0

Eqns. (10.53), (10.37)-(10.39) and (10.50)-(10.51) completely describe the field produced in z > 0.
The field produced by these equivalent currents is zero for z < 0 as discussed in section 10.2. We are
thus free to insert a perfectly conducting plane at z = 0− , just behind the equivalent sources, without
disturbing the field in z > 0 (Fig. 10.15(a)). From section 10.2, we know that the effect of JS is “shorted
out” by the perfect conductor, and we can consider the effect of MS only. But by image theory, the effect
of MS together with the conductor at z = 0 is the same as that of a source 2MS with the conductor
removed. Hence, we can also write (10.53) as

1
Z
am0 = − E0 × H−m · uz dS
2 S0
1
Z
= E0 × Hm · uz dS (10.54)
2 S0

This is none other than the Thévenin equivalent source of section 10.2.2.

331
 Figure 10.16: Side-wall aperture in a waveguide

In a similar way, we could short out the effect of MS by placing a magnetic wall at z = 0, and after
invoking image theory we would obtain
1
Z
am0 = Em × H0 · uz dS (10.55)
2 S0
which is the expression for excitation by the Norton equivalent sources of section 10.2.2. We thus need
only to know the transverse E or H field at z = 0 in order to completely determine the strength of each
mode in z > 0.
If E0 or H0 is chosen appropriately (say E0 = Em or H0 = Hm ), then am = 1, and all other mode
amplitudes vanish by orthogonality. Naturally, it may be difficult if not impossible to realize a given
desired aperture field in practice. Even so, this result gives a goal to strive for if maximum and exclusive
excitation of a particular mode is desired.
Suppose now that an aperture Sw is cut in the side-wall of a waveguide: a wall which in the absence
of the hole would be a perfect conductor. Introduce the right-handed coordinate system (n, ℓ, z) in the
aperture, where the unit vector un normal to the sidewall points into the interior of the waveguide, and
uz × un = uℓ . If a tangential electric field
E0 = uℓ E0ℓ + uz E0z (10.56)
is applied at Sw as shown in Fig. 10.16, then according to the Love equivalence principle of section 10.2,
this may be modeled by the equivalent magnetic surface current
MS = E0 × un |Sw = E0z uℓ − E0ℓ uz (10.57)
impressed at Sw . The aperture may be closed up with a perfectly conducting wall just behind these
currents, and the equivalent electric surface currents at Sw ignored because they produce no field in the
waveguide. From (10.47) we have
r
Zm
I
m
Eext (z) = − Hmℓ E0z dℓ (10.58)
2 C(z)∩Sw
1
I
m
Hext (z) = − √ Hmz E0ℓ dℓ (10.59)
2Zm C(z)∩Sw
where C(z) is the boundary of the waveguide at z, and the boundary integrals are taken over the
intersection of this contour with Sw .
In an open waveguide with no metallic sidewall, the “aperture” must be taken to extend for the whole
length of the guide, or at least over those places where the “aperture” fields are significantly different
from zero. Moreover, since we do not metallize the aperture when applying the equivalence principle
in this case, we will have to use both electric and magnetic equivalent surface sources over whatever
“aperture” we use.

332
 Figure 10.17: Excitation of a dielectric slab by an incident plane wave

10.5.1 Example: Plane wave excitation of a dielectric slab

Consider a semi-infinite dielectric slab waveguide illuminated by the normally incident plane wave

Einc = uy Ei e−jk0 z
Ei
Hinc = −ux e−jk0 z (10.60)
ζ0
from free space as shown in Fig. 10.17. If the resulting aperture field E0 , H0 in the plane z = 0 were
known, we could use any of eqns. (10.53), (10.54) or (10.55) to calculate the excitation amplitude of any
of the TE surface waves on the slab. If we can assume that E0 , H0 are approximately given by (10.60)
(this is the Kirchhoff approximation described in section 10.2.4), then our calculations canR proceed.
R
From (10.55), for example, along with eqns. (6.46), (6.32) and (6.43), we have (putting dS → dx)

1 ∞
Z 
inc

am0 ≃ − Ey Hx dx
2 −∞ z=0
1 Ei ∞
Z
= Ey dx (10.61)
2 ζ0 −∞
2Ei V
= √ (m even)
p2 ζ0 neff aeff h1 a

while am0 = 0 for m odd. From (10.54) on the other hand, we get

1 ∞ inc
Z 

am0 ≃ − Ey Hx dx
2 −∞ z=0
Z ∞ 
inc

= −neff Ey Hx dx (10.62)
−∞ z=0
√
2Ei neff V
= √ (m even)
p2 ζ0 aeff h1 a
and again am0 = 0 for m odd. Thus, the two approximations differ only by the factor neff , and our
result can be expected to be accurate if neff is close to 1. The result of (10.53) will be the arithmetic
mean of these two values, and in the absence of other evidence of accuracy might be chosen as the best
of the three approximations. This of course has to be checked by numerical calculations.
Naturally, a better excitation of the TE0 mode could be obtained by illuminating the end of the slab
with a field whose profile more closely matches that of the mode. A Gaussian beam is such a field: a

333
Figure 10.18: (a) Excitation of a slab waveguide by a prism coupler; (b) Plane wave inside prism and
equivalent currents.

well-collimated beam which is capable of propagating relatively long distances without significant loss of
amplitude due to diffraction (spreading of the energy in the direction transverse to that of propagation).
The mathematical details of the calculation are in this case rather more cumbersome.

10.5.2 Prism coupler

Another way of exciting the surface wave modes of a dielectric waveguide is from the side, as shown in
Fig. 10.18(a). A plane wave enters a dielectric prism at an angle such that total internal reflection would
occur at the bottom face of the prism if no waveguide were present underneath (Fig. 10.18(b)). The
sources and material of the prism can, by Love’s equivalence principle, be replaced by a uniform material
of the same properties as the cladding, and an equivalent surface electric and magnetic currents where
the bottom face of the prism was, and these can be used to calculate the excitation of the waveguide.
Since the field below the bottom of the prism is evanescent when no guide is present, we assume that
the field at the bottom of the prism is nearly the same even when the guide is present for purposes of
calculating these equivalent currents.
Though the detailed evaluation of the mode amplitudes will not be done here, we can note that the
phase variation of the plane wave field in the prism should be the same as that of the desired surface
wave mode (e−jβz ) in order to maximize that particular mode amplitude. The arguments are similar to
those given at the end of section 3.2.

10.5.3 Example: Excitation of microstrip by a slot in the ground plane

Consider next a microstrip line which is excited by a slot in the ground plane transverse to the strip,
and across which a voltage Vslot is applied. A side view of the situation is shown in Fig. 10.19. We will

334
 strip
y

h εr
s z
– V +
slot
ground plane

Figure 10.19: Slot excitation of microstrip

assume that the slot width g is so small electrically that the electric field in the slot can be reckoned as
quasistatic, giving this applied voltage a well-defined meaning. The quasistatic field is written as:
(|z| < g2 )

−uz E0z (z)
E0 |y=0 = (10.63)
0 (|z| > g2 )

where E0z (z) satisfies

Z g/2
E0z (z) dz = Vslot (10.64)
−g/2
m
For this “sidewall” aperture, we have un = uy , ul = −ux and hence Hml = −Hmx . By (10.59), Hext ≡ 0,
while by (10.58)
r Z ∞ 
m Zm 
Eext (z) = E0z (z) Hmx dx

2 −∞ 
y=0
 g
−E0z (z) (|z| < 2 )
≃ (10.65)
0 (|z| > g2 )
We have made use of the fact that for the quasistatic limit of the quasi-TEM mode,
Z ∞ 

Hmx dx = I
−∞ y=0
r
2
≃
Zm
(compare (8.38)). The minus sign in (10.65) reflects the fact that the voltage source has been inserted
into the ground conductor of this transmission line instead of the positive side. By (3.22) and (10.64),
we see as expected that the slot is equivalent to a single series-connected lumped voltage source as in
Fig. 3.10, with V0 = −Vslot .

10.6 Excitation by Electric and Magnetic Dipoles and Electri-

cally Small Apertures
An elementary electric dipole

J = jωpδ(x − x0 )δ(y − y0 )δ(z − z0 ) (10.66)

335
concentrated at the point (x0 , y0 , z0 ) produces, according to (10.41) and (10.45), the mode amplitude
jω
am0 = − p · E−m (x0 , y0 )eγm z0 (10.67)
4
or, from (10.47) and (3.23), the equivalent lumped sources
r
(m) Zm
V0 = jω Emz (x0 , y0 )pz
2
(m) 1
I0 = −jω √ EmT (x0 , y0 ) · pT (10.68)
2Zm
located at z = z0 . Similarly, the elementary magnetic dipole

M = jωµmδ(x − x0 )δ(y − y0 )δ(z − z0 ) (10.69)

produces the mode amplitude

jωµ
am0 = m · H−m (x0 , y0 )eγm z0 (10.70)
4
or the equivalent lumped sources
r
(m) Zm
V0 = −jωµ HmT (x0 , y0 ) · mT
2
(m) 1
I0 = jωµ √ Hmz (x0 , y0 )mz (10.71)
2Zm
These results can also be used in conjunction with the small aperture theory of section 10.2.5 to
obtain excitation coefficients for waveguides excited by electrically small apertures in their walls, as
shown in example 10.6.2 below.

10.6.1 Example: Dipole excitation of a coaxial cable

Let us consider the effect of an elementary electric dipole p = uρ Il/(jω) located at a radius ρ = ρ0 and
at z = 0 in the coaxial line described in section 8.3.1. According to (10.67), we have in this case

a10 1
= − Iluρ · E∓1 (ρ0 , φ0 )
b10 4
√
ζIl
= −p (10.72)
π ln(b/a)4ρ0

where the index m = 1 denotes the TEM mode.

The power carried away from the dipole by the TEM wave (in both directions) is

Prad = |a10 |2 + |b10 |2

ζI 2 l2
= (10.73)
8πρ20 ln (b/a)

If the dimensions of the coax are such that all higher-order modes are cutoff, any of these which may be
excited will carry no power since Zm for a cutoff mode is purely imaginary. In that case, we can define
a radiation resistance Rrad of the dipole in the coaxial line as
2
ζl2

2Prad l
Rrad ≡ 2
= 2 = Zc (10.74)
|I| 4πρ0 ln (b/a) 2ρ0

336
 aTE
incident TE10 10
(tramsmitted) y
d
y0
(reflected)
bTE
10
z
y=0
z=0

(side view)

x0 y
d
y0

(end view)

Figure 10.20: Diaphragm with small aperture in a rectangular waveguide.

This should be compared with the radiation resistance of a dipole in free space, which is

ζ0 k02 l2
Rrad = (10.75)
6π

Unlike (10.75), (10.74) does not vanish with frequency, and so this means of excitation is equally efficient
(or inefficient) at all frequencies.

10.6.2 Example: Diaphragm with a small aperture in a rectangular wave-

guide
Consider next the case of a rectangular waveguide with a perfectly conducting diaphragm at z = 0,
into which a small aperture centered at x0 and y0 has been cut (Fig. 10.20). We wish to compute the
amplitudes of the TE10 modes transmitted through and reflected back from this diaphragm when a TE10
mode of unit amplitude is incident from the left.
If there were no aperture present, the incident wave would be totally reflected at the diaphragm,
resulting in the short-circuit field

Esc = [E1 e−jβ1 z − E−1 ejβ1 z ] (no aperture)

sc
H = [H1 e−jβ1 z − H−1 ejβ1 z ] (10.76)

337
in z < 0. Here, E±1 ,H±1 and β1 are the normalized TE10 mode fields and propagation coefficient as
given in section 5.6.1. We can express the fields as
πx
E±1 = uy E10 sin (10.77)
 a 
E10 πx jπ πx
H±1 = ∓ux sin + uz cos
Z1 a aβ1 a

where by (5.102), r
jωµa Z1
E10 =− H10 = 2 (10.78)
π ad
and
ωµ
Z1 = ζTE10 =
β1
is the characteristic impedance of the mode.
The jump in the short-circuit field across (x0 , y0 , 0) at the diaphragm is thus
(x ,y ,0+ )
[Dsc ](x00 ,y00 ,0− ) = 0
(x ,y ,0 ) + 2E10 πx0
[Hsc ](x00 ,y00 ,0− ) = ux sin (10.79)
Z1 a
By section 10.2.5, the fields in z > 0 resulting from the aperture being opened up are approximately
equal to those produced by the magnetic dipole
↔ (x ,y ,0+ )
m+ = α M · [Hsc ](x00 ,y00 ,0− )
β1 πx0 ↔
= 2 E10 sin (α M · ux ) (10.80)
ωµ a

located at (x0 , y0 , 0+ ) with the aperture closed off. The fields in z < 0 are the sum of (i) the incident
wave and reflected wave as given by (10.76), and (ii) a field produced by the magnetic dipole

β1 πx0 ↔
m− = −2 E10 sin (α M · ux ) (10.81)
ωµ a

located at (x0 , y0 , 0− ) with the aperture closed off.

By image theory, the dipole (10.80) radiating in the presence of the diaphragm is (for z > 0) equivalent
to a dipole of twice the strength acting without the diaphragm (e. g., in the infinite waveguide). The
amplitude of the forward-traveling TE10 mode in z > 0 is therefore, by (10.70) and (10.80)

1
aTE10 ≃ jωµm+ · H−1 (x0 , y0 ) = j∆ (10.82)
2
where
4β1 πx0
∆= αM,xx sin2 (10.83)
ad a
and ↔
αM,xx = ux · α M · ux
is the xx-component of the tensor αM . The total amplitude of the −z traveling TE10 mode in z < 0 is,
similarly,
bTE10 ≃ −1 + j∆ (10.84)
which includes the reflected part of the field Esc , Hsc from (10.76).

338
 The careful reader will have detected a slight anomaly in this solution. If all modes except the
TE10 mode are cutoff, then only they are capable of transferring power in the infinitely long sections of
waveguide. Now, the power carried toward the diaphragm by the incident wave is 1, since the wave has
unit amplitude. Likewise, the reflected and transmitted waves carry the powers

|bTE10 |2

and
|aTE10 |2
away from the diaphragm. The total of these should equal the total power incident on the diaphragm,
i. e.,
|aTE10 |2 + |bTE10 |2 = 1 (10.85)
A glance at (10.82) and (10.84) shows that our solution gives

|aTE10 |2 + |bTE10 |2 = 1 + 2∆2 (10.86)

The problem is that small aperture theory

↔
as presented in section 10.2.5 and used here is correct
↔
only to first order in the small quantities α M and αE , and is, in fact, in error by terms of order (α M )2
and α2E . This is precisely what we observe in eqn. (10.86). We may fix this difficulty by using, instead
of m± as given by eqns. (10.80) and (10.81),
m±
m′± = (10.87)
1+δ
for the equivalent dipole of the aperture, where δ is a complex constant which has the same order of
magnitude as ∆, and is to be determined by the condition of conservation of complex power.
When this change is made, we find instead of (10.82) that

j∆
aTE10 = (10.88)
1+δ
and instead of (10.84) that
j∆
bTE10 = −1 + (10.89)
1+δ
Upon enforcing condition (10.85), we find that δ = δr + j∆, where the real part δr is not determined
by (10.85) (conservation of the real part of the complex power). This term must be obtained from other
considerations, as is shown in Appendix N. For now, we set δr = 0 so that we finally have the expressions
j∆
aTE10 =
1 + j∆
j∆
bTE10 = −1 + (10.90)
1 + j∆
which have been corrected to satisfy conservation of energy on the time average. We will see in chapter 11
that when δr is specified in this way, the simplest possible equivalent circuit for the diaphragm results.

339
10.7 Notes and References
Field equivalence principles are discussed in
Harrington (1961), chapter 3;
Collin (1960), chapter 1;
as well as
S. A. Schelkunoff, “Kirchhoff’s formula, its vector analogue, and other field equivalence
theorems,” Commun. Pure Appl. Math., vol. 4, pp. 43-59 (1951).
V. H. Rumsey, “Reaction concept in electromagnetic theory,” Phys. Rev., vol. 94, pp.
1483-1491 (1954); erratum; ibid., vol. 95, p. 1705 (1954).
V. H. Rumsey, “Some new forms of Huygens’ principle,” IRE Trans. Ant. Prop., vol. 7, pp.
S103-S116 (1959).
B. S. Svetov and V. P. Gubatenko, “Equivalence of systems of internal electric and magnetic
currents,” Sov. J. Commun. Technol. Electron., vol. 30, no. 7, pp. 67-72 (1985).
J. Appel-Hansen, “Comments on field equivalence principles,” IEEE Trans. Ant. Prop., vol.
35, pp. 242-244 (1987).
Schelkunoff’s paper above includes electromagnetic analogs of the Thévenin and Norton equivalence
theorems from circuit theory; they are close in spirit to the static field equivalence theorem given by
Helmholtz:
H. L. F. von Helmholtz, “Ueber einige Gesetze der Vertheilung elektrischer Ströme in körper-
lichen Leitern mit Anwendung auf die thierisch-elektrischen Versuche,” Ann. Phys.
Chem., vol. 89, pp. 211-233, 353-377 (1853) [also in H. L. F. von Helmholtz, Wis-
senschaftliche Abhandlungen, vol. 1. Leipzig: Johann Ambrosius Barth, 1882, pp. 475-
519].
a summary of which (in English) can be found in:
Koenigsberger (1965), pp. 99-103.
The general theory of small apertures and their polarizabilities is given in
Collin (1960), pp. 285-302.
Montgomery et al. (1965), pp. 176-179.
Mashkovtsev et al. (1966), chapter 4.
Collin (1966), pp. 190-197.
Sporleder and Unger (1979), pp. 47-55.
J. van Bladel, “Small-hole coupling of resonant cavities and waveguides,” Proc. IEE (Lon-
don), vol. 117, pp. 1098-1104 (1970).
J. Van Bladel, “Small Holes in a waveguide wall,” Proc. IEE (London), vol. 118, pp. 43-50
(1971).
Methods of computing the polarizabilities, numerical values and approximations for various aperture
shapes are given in
R. F. Fikhmanas and P. Sh. Fridberg, “Theory of diffraction at small apertures: Computation
of upper and lower bounds of polarizabilities,” Radio Eng. Electron. Phys., vol. 18, pp.
824–829 (1973).
R. de Meulenaere and J. Van Bladel, “Polarizability of some small apertures,” IEEE Trans.
Ant. Prop., vol. 25, pp. 198–205 (1977).
E. E. Okon and R. F. Harrington, “The polarizabilities of electrically small apertures of
arbitrary shape,” IEEE Trans. Electromag. Compat., vol. 23, pp. 359–366 (1981).

340
 E. Arvas and R. F. Harrington, “Computation of the magnetic polarizability of conducting
disks and the electric polarizability of apertures,” IEEE Trans. Ant. Prop., vol. 31, pp.
719–725 (1983).
V. I. Fabrikant, “Magnetic polarizability of small apertures: analytical approach,” J. Phys.
A: Math. Gen., vol. 20, pp. 323–338 (1987).
V. I. Fabrikant, “Electrical polarizability of small apertures: analytical approach,” Int. J.
Electron., vol. 62, pp. 533–545 (1987).
↔
The approximate entries for αE and α M in Table 10.1 are due to Fabrikant, who also assesses their
accuracy.
For “star-shaped” apertures with respect to an origin inside A, the inequality
1
Z
αE ≤ ρ dS (10.91)
π
A

where ρ is the distance from that origin, was obtained by

L. E. Payne, “Isoperimetric inequalities and their applications,” SIAM Review, vol. 9, pp.
453-488 (1967).
The further inequalities,
4A2 2  A 3/2
≤ αE ≤ (10.92)
3πP 3 π
and
4  A 3/2 αM,uu + αM,vv 4  P 3
≤ ≤ (10.93)
3 π 2 3 2π
where A is the area of the aperture and P its perimeter, were conjectured in
D. L. Jaggard and C. H. Papas, “On the application of symmetrization to the transmission of
electromagnetic waves through small convex apertures of arbitrary shape,” Appl. Phys.,
vol. 15, pp. 21–25 (1978).
Lastly, for an aperture of any shape we have
1 1 1
≤ + (10.94)
αE αM,uu αM,vv

if the axes u and v are chosen so that αM,uv = αM,vu = 0 (these are the so-called principal axes). This
inequality was proved in
K. S. H. Lee and L. K. Warne, “A general relation between electric polarizabilities of a plane
aperture,” Radio Sci., vol. 39, art. RS6007, 2004.
A detailed consideration of the validity of the completeness assumption is given for hollow metallic
waveguides in
Kurokawa (1969).
Milton and Schwinger (2006), chapter 10.
A treatment of completeness which covers waveguides with piecewise constant filling (e. g., microstrip)
is given in
Yu. G. Smirnov, “Use of operator-beam method in the problem of the natural wave modes
of a partially filled waveguide,” Sov. Phys. Dokl., vol. 35, pp. 430-431 (1990).
Mode expansions outside of the source region are treated in

341
 Collin (1960), pp. 200-214 and 483-485;
van Bladel (1964), chapter 13;
Johnson (1965), pp. 195-202.
A very lucid discussion on the choice of Zm for equivalent transmission lines (and its arbitrary nature)
is given in
S. A. Schelkunoff, “Impedance concept in wave guides,” Quart. Appl. Math. vol.2, pp. 1-15
(1944).
Other approaches to the choice of Zm for non-TEM modes are to be found in:
A. Magos, “Characteristic impedance of waveguides used in quasi-TEM mode,” Periodica
Polytechnica Elec. Eng. vol. 30, pp. 17-26 (1986).
J. R. Brews, “Characteristic impedance of microstrip lines,” IEEE Trans. Micr. Theory
Tech. vol. 35, pp. 30-34 (1987).
For the treatment of mode expansions within the source region, we have loosely followed Vainshtein’s
classic Russian text on electromagnetic waves,
Vainshtein (1988), chapter 14.
Some discussion of the expressions for the longitudinal field components (10.50)-(10.51) in the English
language is to be found in
G. S. Kino, “Mode properties of passive transmission systems,” in Electronic Waveguides.
Brooklyn: Polytechnic Press, 1958, pp. 269-281.
Felsen and Marcuvitz (1973), chapter 8.
R. E. Collin, “On the incompleteness of E and H modes in wave guides,” Canad. J. Phys.
vol.51, pp.1135-1140 (1973).
C. Vassallo, “On the expansion of axial field components in terms of normal modes in per-
turbed waveguides,” IEEE Trans. Micr. Theory Tech. vol. 23, pp. 264-265 (1975).
The extension of mode expansions to open waveguides (including the continuous spectrum) is found in
A. B. Manenkov, “The excitation of open uniform waveguides,” Radiophys. Quantum Elec-
tron. vol.13, pp. 578-586 (1970).
The prism coupler for dielectric waveguides is treated in
P. K. Tien, R. Ulrich and R. J. Martin, “Modes of propagating light waves in thin deposited
semiconductor films,” Appl. Phys. Lett., vol. 14, pp. 291-294 (1969).
The radiation resistance of dipoles in waveguides is discussed in
Johnson (1965), pp. 198-201;
Collin (1960), sect. 7.1.

10.8 Problems
p10-1 A small battery forces a voltage V to exist across a small gap of width g between two sections
of perfectly conducting wire as shown.

342
 (a) Sketch the electric field lines that will exist near the gap (explain!).
(b) What equivalent current source(s) JS and/or MS on the surface of the wire and/or
battery (ρ = a) can be used to model this battery? Sketch the directions of the current
lines.
(c) If what is connected to the far ends of the wire is unknown, is there a system of equivalent
sources which uses only (essentially) known sources?

p10-2 A quasi-TEM mode is propagating on a wide (w/h ≫ 1) microstrip line as shown, with an
amplitude am in the +z-direction.

A small square hole of side a is cut into the ground plane y = 0 centered at x = z = 0. Using
an approximation for the fields of this mode that neglects fringing at the edges of the strip,
find expression(s) for the moments of the equivalent dipole(s) radiating in the presence of
the closed-off aperture that, according to Bethe small-aperture theory, produce the fields EA
and HA caused by the aperture.
p10-3 At z = 0, a vertical strip carrying the impressed surface electric current density
I
JS = uy
w
is connected between the line and ground plane of a stripline as shown.

The strip has width w and height h. Find the externally impressed sources appearing in
eqns. (10.47) for this configuration, and draw an equivalent circuit for this source, if m
denotes the index of the fundamental TEM mode. Calculate the mode amplitudes am and
bm for this mode.

343
p10-4 Suppose a voltage source like that of problem p10-1 is placed in the inner conductor of a
coaxial line at z = 0 as shown.

Use an equivalent magnetic current representation for this source as found in problem p10-1
to find the equivalent circuit of this generator in the equivalent transmission line for the TEM
mode of this waveguide.

p10-5 Repeat problem p10-4 for the case of the TEM mode of a two-wire transmission line with
generator inserted in one of the conductors as shown.

p10-6 An electrically small circular coupling hole of radius r0 is located in the common wall of two
rectangular waveguides at z = 0 as shown.

d w
2r0
x

d
z

344
 If a TE10 mode (index = m) traveling in the +z-direction with amplitude aum exists in the
upper guide, what are the amplitudes alm and blm of the ±z-traveling TE10 modes produced
in the lower guide due to the presence of this hole?
p10-7 What are the total amplitudes of the ±z-traveling TE10 modes produced in the upper guide
in problem p10-6 due to the presence of the coupling hole? Is conservation of energy satisfied
by this solution? If not, can you suggest a modification by means of which power conservation
is satisfied?
p10-8 Rework example 10.6.2 for the case when the right-hand waveguide (in z > 0) has dimensions
a′ , d′ instead of a, d.
p10-9 Rework example 10.6.2 for the case when the right-hand waveguide (in z > 0) is uniformly
filled with a medium of relative permittivity ǫr and permeability µr .
p10-10 Rework example 10.5.1 for the case of an obliquely incident plane wave

Einc = uy Ei e−jk0 (z cos θ+x sin θ)

Ei −jk0 (z cos θ+x sin θ)
Hinc = (−ux cos θ + uz sin θ) e
ζ0
using the Kirchhoff approximation. If m denotes the TE0 mode, sketch the dependence of
am0 vs. θ for some reasonable value of V in the single-mode range.
p10-11 The inner conductor of a coaxial line is to serve as a probe for the excitation of a hollow
rectangular waveguide as shown.

(side view) d
2c
w
z
b

(top view)
a

a/2
z

345
 The coaxial line has inner radius c, outer radius b, and the inner conductor extends a distance
w down from the center of the top wall of the rectangular guide as shown. Assume that the
current I2 (z2 ) in the y-direction on the probe end is the remnant of a standing wave on the
coaxial line:
V+
I2 (z2 ) = −2j 2 sin β2 (y − w) (0 < y < w)
Z2
where V2+ is the voltage of the incident wave on the coaxial line (relative to the end of the
probe at y = w), and Z2 and β2 are the characteristic impedance and propagation constant of
the coaxial line. Assume further that the equivalent magnetic current due to the electric field
at the annular aperture in the top wall makes a negligible contribution to the excitation. Find
the amplitudes aTE10 and bTE10 of the TE10 modes excited by this probe in the rectangular
waveguide. Assume that c is sufficiently small that it can approximately be regarded as a
current filament.
p10-12 Consider a general TEM waveguide mode on the structure of Fig. 8.5. A current-carrying
wire is located on an arbitrary path between conductors C1 and Cg as shown.

xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
Cg
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx

xxxxxxxxxxxx
xxxxxxxxxxxx
xxxxxxxxxxxx
C1

ul
un

The corresponding impressed current density is

J = −uℓ Iδ(z)δ(n)

where ℓ and n are arc length and normal coordinates on the wire path, analogous to those
on the path “g-1” shown in Fig. 8.6.
(a) Show that the equivalent distributed sources (10.47) for the m = TEM mode of this
waveguide are
m
Eext (z) = 0
m
Hext (z) = −Iδ(z)
(b) Obtain expressions for the forward and reverse mode amplitudes am (z) and bm (z) of
(10.45) and (10.46) for this mode.
(c) Explain the negative signs in front of your answers for parts (a) and (b).

346
p10-13 A slotline of width w is excited by a transversely oriented strip of width s carrying a total
current I0 (distributed as JSx (z)) and situated on the opposite side of the substrate from the
slot as shown.

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
current-carrying
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
strip (on top)
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
dielectric substrate I
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
0
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
90º
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+ V (z )
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
z w
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
2 2
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
–
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
s
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
h (ε ) slotline (on bottom)
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
r
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

ground plane

Assume that s is small compared to a wavelength. Find the equivalent distributed sources
m m
Eext (z) and Hext (z) that drive the fundamental mode of the slotline under quasi-TEM con-
ditions. Show that these distributed sources are equivalent to a lumped shunt current source
I0 located at the point of crossover between the slotline and the strip. Be sure to indicate
the polarity of the sources; that of the voltage of the normalized slotline mode is indicated
in the figure.
p10-14 For the situation of problem p10-2, suppose that a plane wave with electric field amplitude
E0 at the center of the square aperture is incident from the air region below the ground plane
(y < 0) instead of having the quasi-TEM mode of the microstrip incident at the aperture.
If the plane wave is normally incident to the ground plane, what are the amplitudes of the
+z- and −z-traveling quasi-TEM microstrip modes excited by the equivalent dipoles of the
aperture? Is it possible, by having the plane wave be obliquely incident to the ground plane
and/or by changing the shape or orientation of the aperture, to suppress the excitation of
the −z-traveling quasi-TEM mode? If so, how?

347
348
Chapter 11

NETWORK THEORY FOR

GUIDED WAVES

She came naturally by her confused and groundless

fears, for her own mother lived the latter years of her
life in the horrible suspicion that electricity was
dripping invisibly all over the house. It leaked, she
contended, out of empty sockets if the wall switch had
been left on. She would go around screwing in bulbs,
and if they lighted up she would hastily and fearfully
turn off the wall switch and go back to her Pearson’s
or Everybody’s, happy in the satisfaction that she
had stopped not only a costly but a dangerous leakage.
Nothing could ever clear this up for her.
—James Thurber,
My Life and Hard Times

11.1 Introduction
Up to this point, we have (in accordance with the definition in Chapter 1) dealt only with waveguides
infinite in length and uniform along the guiding (z) direction. In real life, lengths of waveguide start
somewhere, must negotiate bends and other types of irregularity, and finally terminate. The exact field
solution for a typical waveguide run is usually hopelessly complex, and not worth the trouble. Much
useful insight for design and analysis can be obtained by using the formal equivalence between waveguide
modes and classical transmission lines which was outlined in the previous chapter.
For one thing, we now know that a modal representation of waveguide fields is possible not only in an
infinitely long waveguide (for which mode solutions were originally obtained), but for any finite section
of uniform guide as well. For, consider the complicated waveguide system of Figure 11.1(a). Whatever
fields are produced in the uniform section of waveguide between z1 and z2 by the sources outside this
section can also be thought of as produced by a set of equivalent surface currents located at z1 and z2 , as
indicated in Figure 11.1(b), which radiate into an infinitely long waveguide. By the equivalence theorem,
once the fields at z1 and z2 are specified, the fields throughout the uniform section can be found from
the mode expansion (10.37)-(10.38) and (10.50)-(10.51) of the previous chapter.
Secondly, since the mode representation has been shown to be valid here, it suffices to find only the
wave amplitudes am (z) and bm (z) (or equivalently the voltages Vm (z) and currents Im (z)) associated
with each mode at any plane z, in order to determine the total field there. But at a cross-sectional plane

349
Figure 11.1: (a) Complex waveguide system containing uniform section z1 ≤ z ≤ z2 ; (b) Equivalent
currents producing the same fields.

of a waveguide which adjoins a nonuniform region, a relationship among the wave amplitudes (or currents
and voltages) at this plane must exist which is determined by the properties of the nonuniform region
itself. Specifically, consider a waveguide feeding a source-free termination as shown in Figure 11.2(a).
We define the terminal plane A at some constant value z = zA within the uniform waveguide as shown.
The uniform waveguide section can be modeled (á la Chapter 10) as an infinite sequence of equivalent
transmission lines with associated wave amplitudes am (z) and bm (z). If the termination (that is, the
region beyond the terminal plane) contains only linear material in addition to containing no sources,
then by the linearity of Maxwell’s equations there must be a linear relationship between the am ’s and
bm ’s at z = zA of the following type:
   I I
 
b1 (zA ) SG,11 SG,12 ··· a1 (zA )
I
 b2 (zA )   SG,21 I
= SG,22 ···   a2 (zA ) 
 
(11.1)
.. .. .. ..

..
. . . . .

I I
where SG,ij are constants independent of am (zA ). In matrix notation this is simply [b(zA )] = [SG ][a(zA )],
I I
where the matrix [SG ] is made up of the elements SG,ij . Of course, this is like the scattering matrix
representation of chapter 2, but with a port for each mode, whether or not the mode is above cutoff:
I
[SG ] is called the Generalized Scattering Matrix (GSM) of the termination.
In terms of the equivalent transmission line model, eqn. (11.1) is nothing more than a mathematical
representation of a lumped network terminating the transmission lines. Thus we have an equivalent
circuit as shown in Figure 11.2(b). This representation is mathematically rigorous, since all modes
(whether propagating or not) are accounted for. However, once this waveguide is connected to other
networks (some with with sources, represented for example by a Thévenin equivalent network as shown
in Figure 11.3), an infinite set of linear equations will need to be solved to determine rigorously the
amplitudes of all the modes. In practice, however, we normally have only one (or a few) of the modes

350
 waveguide z termination

A: z=zA
(a)

mode 1

[SGI]
mode 2

•
•
•

A
(b)

Figure 11.2: (a) Waveguide junction; (b) Equivalent network.

V1II
- +
mode 1
[SGII] V2II [SGI]
- +
mode 2
•
•
•

Figure 11.3: Equivalent circuit of waveguide with two terminations.

351
 mode 1
[SGI]

Z2

Z3

•
•
•
A

(a)

mode 1 Z11 or S11

(b )

Figure 11.4: (a) Cutoff mode equivalent transmission lines replaced by their characteristic impedances;
(b) Equivalent load impedance for mode 1.

propagating, and all the rest are cutoff. Here, we will assume for simplicity that the only propagating
mode is mode 1. To give the equivalent transmission line concept practical value, we stipulate that no two
nonuniformities in a waveguide system are to occur so close together that a cutoff mode excited at one of
them has appreciable amplitude at the position of the other. Quantitatively, this means that |α2 l|, |α3 l|,
and so on, must be large compared to one, where l is the minimum distance between nonuniformities,
and α2 , α3 are the real parts of γ2 , γ3 , . . . .
This condition means that a terminal plane of a termination can always be chosen such that only
the amplitude of mode 1 is significant at that plane, simply by moving the terminal plane away from
the termination by a sufficient distance. The equivalent lines for modes 2, 3, etc., now appear effectively
semi-infinite looking from the equivalent network of the termination, since any amount of the cutoff mode
excited on one of these lines decays essentially to zero before encountering any discontinuity, and thus
the reflection from such a discontinuity can be ignored. These semi-infinite lines can thus be replaced by
lumped impedances equal to their characteristic impedances Z2 , Z3 , ... , and the transmission line for
I
mode 1 will simply see a reflection coefficient of S11 = SG,11 , and thus an equivalent scalar impedance

1 + S11
Z11 = Z1
1 − S11
as the entire effect of the termination (Figure 11.4). We will assume hereafter that the terminal planes
of a junction can always be chosen so that only propagating modes need be considered, and the resulting
equivalent network will have one port for each propagating mode of each waveguide connected to the
junction. With this understanding, the equivalent network of Figure 11.2(a) is thus the one-port network
consisting of the scalar impedance Z11 of Figure 11.4(b).
Two things should be emphasized at this point. First, despite the fact that we seem to be considering
only the fundamental mode in this model, the higher order modes do play a role in determining the

352
 Figure 11.5: Sign conventions for multiport networks.

values of Z11 and S11 . These modes store energy in a localized region near the junction, and in a lossless
waveguide contribute reactive effects not accounted for by the fundamental mode. We will have more to
say about this in sections 11.3-11.7. Second, Z11 has meaning only with respect to a specifically chosen
characteristic impedance Z1 associated with mode 1, and therefore with this port of terminal plane A.
To each port of any equivalent network, therefore, we must associate a particular value of characteristic
impedance—that of the waveguide mode which constitutes the given port.
In waveguide network theory, one assumes that a given nonuniformity in the waveguide can be
characterized by a lumped equivalent network in the manner we have just described, even though actual
calculation of the network parameters does require solution of a field problem, and calculations of this
type (even approximate ones) can be quite formidable. Although a comprehensive treatment of the
advanced numerical and analytical methods used to address these problems is beyond the scope of this
book (see the references at the end of this chapter), we will give a brief introduction to one important
technique in the next section, to give a flavor of what is involved. Once the network is found, however,
all of the relatively elementary methods of circuit theory and transmission line theory can be used to
analyze complicated waveguide junction problems.
For general multiport networks involving more than one waveguide, we find it useful to adopt the
following conventions, in addition to those from section 2.1.2 used for multiport networks containing
lumped elements and transmission lines:

(i) With respect to the junction, separate z-axes for each waveguide are defined such that
the +z-direction points into the junction, as shown in Figure 11.5.
(ii) For propagating modes on lossless waveguides, both the normalization of the mode fields
to unit power transfer (9.28) as well as the usual normalization condition (5.27) hold,
while Zm is taken to be real.
(iii) The index m will hereafter be used to denote not only a particular mode of a waveguide,
but also will designate which waveguide (i.e., which port of the network) the mode exists
on.
(iv) At the terminal plane Am (z = zAm ) of the mth port, we define a terminal voltage and
current as those of the associated waveguide mode:

Vm (zAm ) ≡ Vm (11.2)
Im (zAm ) ≡ Im (11.3)

We also define a forward (incident) wave amplitude am and a reverse (reflected) wave

353
 amplitude bm at this port as:

Vm + Zm Im Ṽm + I˜m
am = √ =
8Zm 2
Vm − Zm Im Ṽm − I˜m
bm = √ = (11.4)
8Zm 2

These amplitudes are related to the voltage amplitudes Am (z) and Bm (z) defined in
(10.39) by the relations:

am = am (zAm )e−γm zAm

bm = bm (zAm )eγm zAm (11.5)
−1/2
The factor involving Zm is introduced into (11.4) so that whenever Zm is real, the
expression 21 Re(Vm Im
∗
) = (|am |2 − |bm |2 ) represents the time-average power entering port
m, similar to what has been done in Chapters 1, 2 and 10. This is always true for
propagating modes on lossless waveguides as discussed in (ii) above. This normalization
is perhaps the most widely used one in the literature, but is not the only such possible.
It does simplify certain network properties to be derived later on.

11.2 The Mode Matching Method

Determination of the scattering parameters of a waveguide junction requires in general the solution of a
boundary-value problem for Maxwell’s equations, often by numerical methods that are outside the scope
of this course. However, one technique exists that is capable of either high accuracy as a numerical
method, or moderate accuracy from relatively simple formulas. This method is known as the mode-
matching method, and is based on the expansion of the total field as a sum of waveguide modes that are
subjected to appropriate boundary conditions at the junction.
Rather than describe the mode-matching technique in full generality, we will illustrate its use for
a particular class of waveguide junctions. Changes required for other structures are usually obvious,
but in any case can be found in the references at the end of the chapter. Consider a step discontinuity
between two semi-infinite waveguides A and B as shown in Fig. 11.6. The cross sections SA and SB

SB-SA


Waveguide A
SA  SB Waveguide B


z

z=0

Figure 11.6: Step discontinuity between two waveguides.

of the waveguides are arbitrary except that we assume (to avoid certain complications) that the cross
section of waveguide B completely contains that of waveguide A: SB ⊃ SA . We will also assume that

354
the boundaries of both waveguides are perfect conductors, as is the portion of SB at the junction plane
z = 0 that is not common with SA , which we denote SB − SA .
In waveguide A (z < 0), we represent the transverse fields by (10.37) and (10.38):
X
ET = ṼmA (z)EmAT (x, y) (11.6)
m>0

I˜mA (z)HmAT (x, y)

X
HT = (11.7)
m>0

where
ṼmA (z) = amA e−γmA z + bmA eγmA z (11.8)
I˜mA (z) = amA e−γmA z − bmA eγmA z (11.9)
In waveguide B (z > 0) on the other hand,
X
ET = ṼmB (z)EmBT (x, y) (11.10)
m>0

I˜mB (z)HmBT (x, y)

X
HT = (11.11)
m>0

where
ṼmB (z) = bmB e−γmB z + amB eγmB z (11.12)
I˜mB (z) = bmB e−γmB z − amB eγmB z (11.13)
Note the change of position of amB and bmB in formulas (11.12)-(11.13). This is due to the convention
that bmB represents a scattered wave, while amB is an incident wave at the junction. Relationships
between the amplitudes amA , bmA , amB and bmB (that is, the GSM) are to be determined by the
conditions:
(i) The tangential fields ET and HT are continuous across the surface SA at z = 0.
(ii) The tangential field ET is zero on the perfectly conducting surface SB − SA at z = 0.
To impose these constraints exactly would require an infinite number of equations in the infinite
number of wave amplitudes for all the modes of both waveguides. Thus, we content ourselves with an
approximation in which only a finite number of modes MA is retained in the series (11.6)-(11.7), and
MB terms in the series (11.10)-(11.11). Under this approximation, the conditions on ET are:
MB
X MA
X
Ṽm′ B (0)Em′ BT (x, y) = Ṽm′ A (0)Em′ AT (x, y) (on SA )
m′ =1 m′ =1
= 0 (on SB − SA ) (11.14)
We cannot exactly satisfy these conditions at all points of the junction surface with a finite number of
terms in the mode expansions. As a compromise, we enforce the conditions only in a weighted-average
sense, with conveniently chosen weighting functions used to carry out this scheme. The best choice
turns out to be taking HmBT for each of the modes of the larger waveguide, forming the product
(−) × HmBT · uz with both sides of (11.14), and integrating over SB .1 The orthogonality property (9.23)
can be used to eliminate all but one term from the left side of this result:
MA
X
ṼmB (0) = Cmm′ Ṽm′ A (0) (m = 1, . . . MB ) (11.15)
m′ =1
1 If we had used H
mAT and only integrated over SA , we would have failed to enforce the condition on SB − SA even
approximately.

355
where Cmm′ is an overlap integral defined by
1
Z
Cmm =′ Em′ AT × HmBT · uz dS (11.16)
2 SA

The values of Cmm′ are known (can be computed) if the mode fields of both waveguides are known.
Forming the column vectors  
 Ṽ1A (0) 
 
 .. 
[ṼA ] = 
 . 
 (11.17)
 
 
ṼMA ,A (0)
 
 Ṽ1B (0) 
 
 .. 
[ṼB ] = 
 . 
 (11.18)
 
 
ṼMB ,B (0)

and the matrix  

 C11 C12 ··· 
 

[C] =  .. 
(11.19)
 C21 . ··· 

 
 . ..
..

. CMB ,MA

we can express (11.15) in the form

[ṼB ] = [C][ṼA ] (11.20)
which is a kind of generalized Kirchhoff Voltage Law (KVL) for the mode voltages at this junction (more
precisely, (11.20) expresses a multiport ideal transformer relationship between the voltages [ṼA ] and
[ṼB ]).
In a similar fashion, we can impose an approximate continuity of the tangential magnetic field over
SA at z = 0:2
MB MA
I˜m′ B (z)Hm′ BT (x, y) = I˜m′ A (z)Hm′ AT (x, y)
X X
(on SA ) (11.21)
m′ =1 m′ =1

As a weighting procedure for this boundary condition, we form the product EmAT × (−) · uz with both
sides of (11.21), and integrate over SA . Using orthogonality of the modes in waveguide A, we arrive at

[I˜A ] = [C]T [I˜B ] (11.22)

which is the Kirchhoff Current Law (KCL) for the mode currents at the junction. It is noteworthy that
the matrix whose transpose appears in (11.22) is the same as the one appearing in (11.20), so (11.22)
represents by the same multiport ideal transformer as does (11.20); moreover, no new computations of
matrix elements are required in this step. The extent to which the values of the “turns ratios” Cmm′
differ from one is a measure of how much mismatch there is between the transverse fields of the two
modes involved. Two identical mode field patterns will result in a 1:1 turns ratio (Cmm′ = 1).

2 We do not enforce a boundary condition for H on S − S , because it is related to the surface current there, which
T B A
is unknown at this stage.

356
11.2.1 Generalized Scattering Matrix (GSM)
The incident wave amplitudes in waveguide A are [aA ] = ([ṼA ] + [I˜A ])/2, while the scattered wave
amplitudes are [bA ] = ([ṼA ] − [I˜A ])/2. Likewise in waveguide B, remembering the convention about
directions, the incident wave amplitudes are [aB ] = ([ṼB ] − [I˜B ])/2, while the scattered wave amplitudes
are [bB ] = ([ṼB ] + [I˜B ])/2. We now define the GSM [SG ] through
   
 [bA ]   [aA ] 
  = [SG ]   (11.23)
   
[bB ] [aB ]

where [SG ] in partitioned form is  

 [SAA ] [SAB ] 
[SG ] = 


 (11.24)
[SBA ] [SBB ]
so that
[bA ] = [SAA ][aA ] + [SAB ][aB ]
[bB ] = [SBA ][aA ] + [SBB ][aB ] (11.25)
Some algebra based on the relationships between wave amplitudes and voltages and currents, together
with the generalized KVL (11.20) and KCL (11.22) leads to
[bA ] = [aA ] + 2[C]T [aB ] − [C]T [C] ([aA ] + [bA ]) (11.26)
and
[bB ] = −[aB ] + 2[C][aA ] − [C][C]T ([bB ] − [aB ]) (11.27)
from which we infer that
−1
[SAA ] = [1] + [C]T [C] [1] − [C]T [C]


(11.28)

−1
[SAB ] = 2 [1] + [C]T [C] [C]T (11.29)

−1
[SBA ] = 2 [1] + [C][C]T [C] (11.30)
−1
[SBB ] = − [1] + [C][C]T [1] − [C][C]T



(11.31)
By inspection we can see that the block matrices [SAA ] and [SBB ] are symmetric. It can also be shown
that [SBA ] = [SAB ]T , so it follows that the GSM is symmetric, just as is the case with an ordinary
reciprocal scattering matrix. If only one mode (m = 1, say) is above cutoff on each side of the junction,
we can recover the ordinary scattering matrix for the junction by taking only the “11” entry from each
of the block matrices:  
 SAA,11 SAB,11 
[S] = 


 (11.32)
SBA,11 SBB,11
For (11.32) to give sufficient accuracy, large enough number of modes MA and MB on each side of the
junction must be used. In practice, if only one mode in each waveguide is above cutoff, usually no more
than about 20 modes are needed on each side in order for the mode-matching technique to give accurate
values for [S]. Information on convergence of the method and criteria for truncation of the mode set
can be found in the references cited at the end of the chapter. One feature that should be noted is that
an optimal truncation is such that the maximum propagation constants of the highest modes retained
on each side of the junction should be chosen to be approximately equal, in order that the transverse
variations of the corresponding mode fields possess approximately the same spatial resolution.

357
11.2.2 Example: Parallel-plate step in height
Consider the step discontinuity in an air-filled parallel-plate waveguide as shown in Fig. 11.7. We

bmB
x
amA aB Waveguide B
Waveguide A aA
bmA amB
x=0 z

z=0

Figure 11.7: Step in height for parallel-plate waveguide.

consider only the case of TM modes (all overlap integrals between TE and TM modes of this waveguide
are identically zero). The transverse fields of the modes have only one component each, and can be
written as  
mπx
EmBx = ζmB HmB cos (11.33)
aB
and  
mπx
HmBy = HmB cos (11.34)
aB
and similarly for waveguide A by letting B → A everywhere in the above. Note that by convention the
index m = 0 is used to denote the TEM mode of the parallel-plate guide. In this example, therefore, all
summations will start at m = 0 instead of m = 1. We have defined
γmB
ζmB = (11.35)
jωǫ0
s  2
mπ
γmB = j k02 − (11.36)
aB

q
2



aB ζ0 (m = 0)
HmB = (11.37)
√ 2



aB ζmB
(m 6= 0)

with analogous definitions having B → A.

The overlap integrals are given by

1 aA
Z
Cmm′ = HmBy Em′ Ax dx (11.38)
2 0
 h  i h  i 
 sin aA mπ m′ π m′ π
ζm′ A aB + aA sin aA mπ
aB − aA

= HmB Hm′ A mπ mπ′ + mπ mπ′ (11.39)
4  a + a a − a

B A B A

By direct calculation, we find that the first several elements of the [C] matrix are:
r
aA
C00 = (11.40)
aB

358
 C01 = 0 (11.41)
√ r s
2 aB ζ0 πaA
C10 = sin (11.42)
π aA ζ1B aB
and s
r
2 aA ζ1A aA aB πaA
C11 = 2 2 sin (11.43)
π aB ζ1B aB − aA aB
First, let us examine what happens if only one mode on either side of the junction is retained. In
this case [C] and all the block matrices [SAA ] etc. are 1 × 1 matrices, i. e., scalars. We have
2
1 − C00 aB − aA
SAA = −SBB = 2 = a +a (11.44)
1 + C00 B A
√
2C00 2 aB aA
SAB = SBA = 2 = (11.45)
1 + C00 aB + aA
But the TEM mode of the parallel-plate waveguide can be characterized by its transmission line param-
eters as follows. Suppose for a moment that the width of the waveguide in the y direction is not infinite,
but merely a very large value w. Then the voltage between the plates is Φ1 = Ex a and the current on
the positive plate is I1 = Hy w. The characteristic impedance is thus
Φ1 a
Zc = = ζ0 (11.46)
I1 w
and so our S matrix in this lowest order approximation can be written
 
√
ZcB −ZcA 2 ZcB ZcA
 ZcB +ZcA ZcB +ZcA 
[S] = 
 √

 (11.47)
2 ZcB ZcA ZcA −ZcB
ZcB +ZcA ZcB +ZcA

which is exactly what classical transmission line theory predicts. A more general version of this result
is obtained in problem p11-6.
Next, we try including two modes (m = 0 and m = 1) on each side. We take aB /aA = 2 and
k0 aB = 2π(0.339) for comparison to more accurate numerical results. The matrix elements are calculated
to be C00 = 0.707, C01 = 0, C10 = 0.432(1+j), and C11 = 0.48. A numerical evaluation of the generalized
scattering matrix predicts an ordinary S-matrix of
   
 SAA,00 SAB,00   0.281 − j0.259 0.906 − j0.183 
[S] = 

=
 


SBA,00 SBB,00 0.906 − j0.183 −0.36 − j0.13

The more accurate value of S11 = 0.281 − j0.259 shows a substantial change in phase, and some in
magnitude as well from the value of 0.333 in the single-mode approximation. Pace and Mittra (see
the references at the end of the chapter) present the numerical result of S22 = −0.365 − j0.142, which
compares with our previous single-mode result of −0.333, and our two-mode result of −0.36 − j0.13,
which we can see is already quite good, considering the small number of modes retained. From (2.30)
and (2.25), we obtain the impedance parameters for this junction of

Z11 = Z12 = Z21 = Z22 = −j12.41 Ω

where we have used the characteristic impedances (11.46) with w arbitrarily chosen to be 1 m. This has
an equivalent circuit of a single shunt capacitive reactance X12 = −12.41 Ω.

359
 Sw
z un
un

A
V

Figure 11.8: Non-radiating one-port termination.

11.3 One-Port Equivalent Networks

We can say a few general things about the impedance or admittance of the one-port termination of
Figure 11.2(a) when the waveguide is lossless. Let us first consider the case when the termination is
closed (occupies a bounded volume V surrounded by perfectly conducting walls (or, more generally, walls
that have a nonzero surface impedance ZS ). The termination is thus bounded by the terminal plane A,
and some sections of walls Sw as shown in Fig. 11.8.
Applying Poynting’s theorem to the complex Poynting vector 12 E × H∗ over the volume V of the
termination, we have
1
Z
E × H∗ · un dS = (11.48)
2 A
jω 1
Z Z
− [µH · H∗ − ǫ̂∗ E · E∗ ] dV − E × H∗ · un dS
2 V 2 Sw

However, the terminal plane A has been chosen so that only the propagating mode is present at that
plane, so the complex power through the plane A will be:
1 1
Z Z
E × H∗ · un dS = − V1 I1∗ E1T × H∗1T · uz dS
2 A 4 A
V1 I1∗
= − (11.49)
2
by (9.28), (10.37) and (10.38), if the mode is, as mentioned before, propagating. The surface integral
over Sw , on the other hand, is simply the complex power directed into the walls (if any). The real part
is the energy loss due to finite wall conductivity, for example:
1 1
Z Z
∗
Pc,r = Re E × H · un dS ; Pc,i = Im E × H∗ · un dS
2 Sw 2 Sw

while the imaginary part can be associated with stored energy in the walls (Appendix F). On the right
hand side of (11.48), we find the stored time-average electric and magnetic energies WE and WM interior

360
to V , as well as the (dielectric) power loss Pd dissipated in the material inside V :
1
Z
wE,av = Re ǫ̂E · E∗ dV (11.50)
4
Z V
1
wM,av = µH · H∗ dV (11.51)
4 V
ω
Z
Pd = − Im ǫ̂E · E∗ dV (11.52)
2 V

Thus, (11.48) becomes

V1 I1∗
= PL + j [2ω(wM,av − wE,av ) + Pc,i ] (11.53)
2
where PL = Pc,r + Pd is the total power dissipated in both filling medium and the walls. Since, however,
V1 = Z11 I1 , where Z11 is the equivalent impedance of the junction, we have
2
Z11 = {PL + 2jωWexcess} (11.54)
|I1 |2
where
Pc,i 1
Z
Wexcess = wM,av − wE,av + = Re (µH · H∗ − ǫ̂E · E∗ ) dV
2ω 4 V
is the excess of stored magnetic energy over stored electric energy (both in the volume V and in the walls).
The real part of Z11 is due to ohmic power losses in the dielectric and in the walls, as expected physically.
If Pc,i is negligible, the sign of the imaginary part X11 of Z11 (i. e., the reactance) is determined by
whether or not wM,av is greater than wE,av . If wM,av > wE,av , the reactance is positive (inductive
termination); otherwise, the reactance is negative (capacitive termination).
Now suppose that the one-port is capable of radiating energy as shown in Fig. 11.9. The volume
V of the termination now extends to infinity, and its bounding surface must include a portion Sr that
extends to infinity. The derivation leading to (11.53) and (11.54) is changed only by the addition of an
extra surface integral over Sr :
1
Z
Pr = E × H∗ · un dS
2 Sr
which is the (real) time average power loss due to radiation from the one-port, and the necessity of
carrying out the volume integrals in (11.52) for the stored energies over a volume of infinite extent.
The calculation of Pr involves only the far fields of the termination, and essentially expresses one
aspect of how the termination functions as an antenna, for better or worse. For radiating fields such as
these, wM,av and wE,av as given by (11.52) are always infinite; however, their difference Wexcess remains
finite as we extend Sr to infinity. It is also true that the far field of a radiating structure becomes locally
a plane wave, for which it is known that the electric and magnetic energy densities are equal, and so
the excess stored energy Wexcess that enters into the calculation of X11 is dependent only on the near
field of the termination, which is often quasi-static in nature. We may thus replace wM,av and wE,av
by the so-called “near-field” or “observable” parts of the stored energies for purposes of this calculation
(see the references at the end of the chapter for how this can be done in a unique fashion). Thus for
a radiating termination, (11.54) should be modified by augmenting the heat losses PL by the radiated
power loss Pr . Otherwise, the interpretation of this equation remains the same.

11.4 Examples of One-Port Terminations

We present here three examples of one-port terminations. Resonant cavities represent a case of spe-
cial interest which will be investigated separately in the next chapter. Further examples of one-port
terminations are to be found in the references at the end of the chapter.

361
 Sw

z un V
un

Sr (→ ∞)

Figure 11.9: Radiating one-port termination.

11.4.1 The open-ended two-wire line

A simple example of a one-port equivalent circuit is provided by the open-ended two-wire transmission
line shown in Figure 11.10(a). We can readily argue that the admittance of this open end must be
capacitive rather than inductive, based on the behavior of the fields near the end of the line. For,
exactly at the end, the wire current must be near zero, while the voltage has built up to near its
maximum value (Figure 11.10(b)). We should expect, consequently, that the H-field of the TEM mode
of this line is nearly zero at z = 0, while the E-field is relatively large. Higher-order modes, whether
cutoff modes or radiation modes, must be present in order to ensure continuity of E and H in the vicinity
of the open end, but the relative magnitudes of E and H should not change much in this region. We
thus have wE,av > wM,av , and the reactance or susceptance will be capacitive.
A more detailed study of the electric field near the ends of the wires shows a fringing as in Fig-
ure 11.10(c). Similarly to the fringing at the edges of a stripline (Figure 8.10), this extra field produces
an additional capacitance Cf at the end of the line, over and above that accounted for by

πǫ0
c= (b >> a) (11.55)
ln(b/a)

(the capacitance per unit length of the line). A solution of the static field problem of Figure 11.10(c)

362
Figure 11.10: (a) Open-ended two-wire transmission line; (b) approximate voltage and current distribu-
tions near the end; (c) static fringing field near the end; (d) aperture field and equivalent currents at
z = 0; (e) terminal admittance of open end at z = 0.

gives the value of

b c2
Cf = (b >> a) (11.56)
2π ǫ0
for this fringing capacitance.
More than this, an examination of the aperture field at z = 0 (Figure 11.10(d)) shows that a number
of magnetic current loops appear there to radiate into the half-space z > 0. As a result, a certain
amount of power is lost from the transmission line, and can be accounted for by adding an equivalent
radiation conductance Gr to the equivalent circuit at z = 0 (Figure 11.10(e)). The value of Gr is found
by calculating the power radiated from the end of the line using standard methods of antenna theory,
and is given by
k 2 b2 ζ0
Gr = 0 (b >> a); (k0 b << 1) (11.57)
8π Zc2
where Zc is the characteristic impedance of the two-wire line:
ζ0
Zc = (b >> a) (11.58)
π ln(b/a)
For a frequency of f = 60 Hz, and a = 0.5 mm, b = 1.0 cm (dimensions similar to those of household
power outlets), we have from (11.58) that
Zc = 359.5 Ω
and thus
Gr = 1.83 × 10−20 S
A voltage of 115 volts RMS (163 volts peak) thus radiates a power of
1
(Gr )1632 = 2.43 × 10−16 watts
2
from the open end. Hence, Thurber’s mother-in-law was right, however slightly.

363
11.4.2 A microstrip patch antenna
A more practical example of a one-port termination is the microstrip patch antenna shown in Fig-
ure 11.11. Here, we show the patch fed by a microstrip of characteristic impedance Zc at one of its

Figure 11.11: Rectangular microstrip patch antenna fed by a microstrip line.

edges—one of a number of possible feed methods. An analysis of this structure shows that the patch
presents a generally complex impedance Zin to the feed line. Normalized values Z̄in = Zin /Zc are
shown for a particular example in Figure 11.12. As the frequency increases from 1.9 GHz to 2.1 GHz,
we observe the impedance to pass through a resonance where Zin is purely real (about 1.93 GHz). A
matching network could be installed on the feed line to insure that almost all of the power incident from
the microstrip line is “dissipated” in Zin (in reality, of course, this power would be that radiated by the
patch).
The microstrip patch structure finds wide application because it is relatively compact and easy to
fabricate to close tolerances. Since the feed itself is a microstrip transmission line, we have a completely
integrated waveguide/antenna structure eminently suited for use in satellite communications systems.
Further information on microstrip patch antennas can be found in the references at the end of the
chapter.

11.4.3 Truncated dielectric slab waveguide

As illustrated in the first example above, truncated TEM or quasi-TEM lines tend to reflect nearly all
of the energy incident at the truncation. By contrast, surface waves on dielectric waveguides “want” to
continue propagation in the forward direction; the more so the smaller the contrast between core and
cladding permittivities. Fig. 11.13 shows some typical reflection coefficients for the TE0 surface wave
at a truncation of the slab. As the quantity 1 − |ρ|2 represents the power radiated from the end of the
guide, it can be seen that this structure is very suitable for use as an endfire antenna, as most of the
incident power is indeed radiated.

364
Figure 11.12: Normalized input impedance Zin /Zc for the microstrip patch antenna of Fig. 11.11; ǫr =
2.2, h = 1.0 mm, l = 6.0 cm, w = 5.0 cm, Zc = 59.9 Ω.

Figure 11.13: Reflection coefficient of TE0 mode at truncated dielectric slab waveguide.

365
 V un
un
A2
A1

Figure 11.14: Geometry for application of Lorentz reciprocity to a two-port junction.

11.5 Two-Port and Multiport Networks

A multiport junction of waveguides can be viewed as a multiport equivalent network, described by the
formalism of sections 2.1 and 2.2. Each port is connected to the transmission line which is equivalent to
one of the waveguide modes; thus, a two-port junction may arise from the connection of two single-mode
waveguides, or from the termination of a single guide supporting two propagating modes. The network
description will be the same in either case, although the parameters of the equivalent network can be
expected to be different.
By similar arguments to those used for one-port networks, we know that a linear relationship must
exist among the terminal variables of a multiport network. In matrix form as in chapter 2:

[V ] = [Z][I] (11.59)

Not all the elements of the impedance matrix [Z] are independent if the material filling the junction
obeys reciprocity. In fact, consider any two possible fields (Ea , Ha ) and (Eb , Hb ) which can exist in the
case of a two-port junction. To these there will correspond the terminal variables (V1a , V2a , I1a , I2a ) and
(V1b , V2b , I1b , I2b ) respectively. Applying Lorentz reciprocity to the volume V bounded by the junction
walls and terminal planes A1 and A2 (Fig. 11.14), we find from (9.8) that
Z Z
a b b a
(E × H − E × H ) · un dS = − (Ea × Hb − Eb × Ha ) · un dS (11.60)
A1 A2

because the surface integrals over the walls of the junction vanish, and because for a passive junction
there are no sources in V . Using (5.27), (10.37), (10.38) and the fact that un = −uz on each of the
terminal planes, we find that
(V1a I1b − V1b I1a ) = (−V2a I2b + V2b I2a ) (11.61)
Eqn. (2.4) can now be expressed in matrix-vector notation as:

[I b ]T [V a ] = [I a ]T [V b ] (11.62)

But since (2.3) holds for any state of this junction, we have as in chapter 2 that [Z] is a symmetric
matrix, provided the material inside the junction is such that the reciprocity theorem is valid. Indeed,
all the apparatus developed in chapter 2 to describe the behavior of multiport lumped-element or classical
transmission-line networks can be carried over without modification to the description of any waveguide
multiport junction.
As with one-port equivalent circuits, the capacitive or inductive nature of multiport networks is
dependent upon whether predominantly electric or magnetic energy is stored in the junction. It is more
convenient to express the general relations for this in terms of the scattering matrix, rather than [Z] or

366
[Y ]. We start by applying Poynting’s theorem to the volume occupied by the junction, in the manner of
eqn. (11.48). The complex power leaving the junction through the ports is now

1
− [I]† [V ]
2
In place of (11.53) we now have
1 †
[I] [V ] = P + 2jωWexcess (11.63)
2
where P = Pr +PL is the sum of the radiated and ohmic power losses from the junction (cf. eqns. (11.48)
and (11.52)), and once again Wexcess is the excess of stored magnetic energy over stored electric energy
in the volume V . Using the wave amplitudes [a] and [b] defined in (2.21), we have

{[a] − [b]}† {[a] + [b]} = P + 2jωWexcess (11.64)

Again introducing the scattering matrix [S] as in (2.19), we have from (11.64):

[a]† [1] + [S] − [S]† − [S]† [S] [a] = P + 2jωWexcess


(11.65)

Separating real and imaginary parts, we get (2.70) and (2.71) as in chapter 2, where the powers and
stored energies are now expressed in terms of field integrals instead of summations over circuit elements.
In the example of scattering from a “thin” obstacle in a hollow metallic waveguide, such an obstacle
can be represented by a two-port network, whose equivalent circuit is a single shunt impedance element
Z12 . In this case the scattering matrix takes the form of (2.77), if both ports are referred to the same
characteristic impedance Z0 . Therefore (2.76) holds, and we conclude, just as we did for a one-port
equivalent network, that for this simple shunt element, Im(Z12 ) is positive (inductive) if wM,av > wE,av
and negative (capacitive) if wM,av < wE,av .
Since cutoff TE modes store predominantly magnetic energy and cutoff TM modes store predomi-
nantly electric energy (problem p5-22) in a hollow metallic waveguide, if we can (by using considerations
of symmetry or some other feature of the junction) determine that only one kind (TE or TM) of cutoff
mode exists in the vicinity of a “thin” obstacle, then we can determine the sign of the shunt reactance
of the equivalent circuit. If a two-port is lossless, then P = 0 in (2.70), and we must have again that [S]
is unitary as in (2.74). This then implies (2.75).

11.6 Examples of Two-Port Networks

11.6.1 Transverse diaphragm with small aperture in a rectangular waveguide
A first example of a two-port junction is provided by a vertical conducting diaphragm in a rectangular
waveguide. The diaphragm is located at z = 0, with only an electrically small aperture centered at
(x0 , y0 ) allowing coupling through it. This is the structure we looked at in example 10.6.2 using small
aperture theory. We display it again in Fig. 11.15. We will take the terminal planes for both ports 1
and 2 to be at z = 0 (plane A) as shown. From our results (10.90) together with definition (2.19) for [S]
and (11.4) for [a] and [b], we conclude that

j∆
S12 = (11.66)
1 + j∆
S11 = −1 + S12 (11.67)

where, from (10.83) we have

4β1 πx0
∆= αM,xx sin2 (11.68)
ad a

367
 Figure 11.15: Diaphragm with small aperture in a rectangular metallic waveguide.

By reflection symmetry about z = 0, we have S22 = S11 and S12 = S21 (which also follows from
reciprocity).
Now, this [S]-matrix has the form of (2.77) appropriate to a “thin” obstacle, with

Z̄12 = Z12 /Z0 = j∆/2 (11.69)

Thus, Z̄12 is positive imaginary, and represents a pure shunt inductance across the equivalent transmis-
sion line at z = 0. According to (2.76), there is more magnetic energy stored near the diaphragm than
there is electric energy. We infer that primarily TE modes are present near the terminal plane.

11.6.2 Capacitive diaphragm; Partitioning of equivalent networks

It sometimes happens that an equivalent network can be partitioned into two or more sections, each
of which has its own significance as a building block in making up equivalent networks for totally new
junctions. We will illustrate this idea with the example of a capacitive diaphragm in a rectangular
metallic waveguide, as shown in Fig. 11.16(a). This “thin” obstacle presents a capacitive susceptance B

Figure 11.16: (a) Capacitive diaphragm in a rectangular metallic waveguide; (b) equivalent circuit.

to the equivalent transmission line as indicated in Fig. 11.16(b). However, the equivalent circuit could be
redrawn in an infinite number of ways. One such way is shown in Fig. 11.17, where the shunt susceptance
has been divided into two parts, and these parts further isolated using a pair of ideal transformers whose
turns ratio n is chosen to transfer the characteristic impedance of the waveguide to some other (but as
yet unspecified) value.
If we turn now to a different problem wherein the right side of the diaphragm opens onto free space
and radiates as an aperture antenna (Fig. 11.18(a)), we have fringing and radiation fields to the right
of the diaphragm, but the fields to the left of the diaphragm remain largely unaltered. The resulting
one-port could be modeled as a single lumped impedance Z11 , but it is more instructive to show the

368
 Figure 11.17: Alternate equivalent circuit for Fig. 11.16, with two “building-block” sections.

Figure 11.18: (a) Capacitively diaphragmmed waveguide radiating into free space; (b) equivalent circuit.

equivalent circuit as in Fig. 11.1(b). Here, the fields of cutoff modes to the left of the diaphragm show
up as one of the half-sections of Fig. 11.17, while the fringing susceptance Bf and radiation conductance
Gr appear to the right of this half-section, modeling the effect of the fields in this half-space. The ideal
transformer’s turns ratio n can be selected in such a way that B/2 depends only on properties of the
diaphragm in the waveguide and not on those of the half-space into which this aperture radiates, while
Bf and Gr do not depend on the properties of the fields inside the waveguide.
This same “radiation network” Gr + jBf could be connected to a different “half-network” if the
aperture were fed by some other waveguide. The interchangeability of component parts of an equivalent
circuit is contingent upon the coupling mechanism between the partitioned sections being sufficiently
weak (i. e., the aperture must be small enough). Nonetheless, this concept is often useful in the modeling
and design of various kinds of equivalent networks.

11.6.3 Right-angle bend in microstrip

It is necessary in microstrip and other planar circuits to change the direction of the line to accommodate
specific design goals (Fig. 11.19(a)). This causes a discontinuity whose effects must be modeled by an
appropriate two-port. Often the bend is chamfered as shown in Fig. 11.19(b) by trimming off a portion
of the sharp exterior corner in order to help minimize reflection from the bend. Further improvement can

369
 Figure 11.19: (a) Right-angle microstrip bend; (b) chamfered bend; (c) double bend.

be expected if we split the bend into two gentler ones as shown in Fig. 11.19(c). Some typical data for
the S-parameters of these bends are shown in Fig. 11.20. It will be observed that an almost reflectionless
bend can be obtained over a wide frequency range by means of chamfering or double-bending, whereas
the uncompensated bend’s performance is seriously degraded as frequency increases.

11.7 Examples of Multiport Networks

11.7.1 Use of physical symmetry–The E-plane and H-plane T–junctions in
a rectangular waveguide
Consider the E-plane and H-plane T-junctions in rectangular waveguides as shown in Fig. 11.21. All
the ports are assumed to have identical cross-sections and to support only the TE10 mode above cutoff.
Both are therefore 3-port networks.
In chapter 2, we saw how the symmetry of a network composed of lumped elements and transmission
lines could be used to simplify the derivation of its scattering parameters. The same kinds of symmetry
considerations can be applied to the electromagnetic field in order to find the S-parameters of waveguide
junctions. In the present case, from Fig. 11.22(a), we can see that the E-field in the E–plane junction
will fringe in such a way that S13 = −S23 . From Fig. 11.22(b) on the other hand, it is clear that
S13 = +S23 for the H-plane tee. From the physical symmetry of the junctions, it should also be evident
that S11 = S22 in both cases.
As a result, we conclude that the symmetric [S] matrix for the E-plane tee must have the form
 
 S11 S12 S13 
 
 
[S] = 
 S12 S11 −S13 
 (11.70)
 
 
S13 −S13 S33

and for the H-plane tee, we must have

 
 S11 S12 S13 
 
 
[S] = 
 S12 S11 S13 
 (11.71)
 
 
S13 S13 S33

370
Figure 11.20: Magnitudes of S11 and S21 for bends of Fig. 11.19: h = substrate thickness = 0.24 mm,
ǫr = 9.8.

371
 Figure 11.21: (a) E-plane T-junction; (b) H-plane T-junction.

Figure 11.22: Field-fringing in (a) E-plane tee and (b) H-plane tee junctions of rectangular waveguides.

372
 z1

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
microstrip (on top)
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
dielectric substrate
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
90º
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+ V (z )
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
z
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
2 2 2
–
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
+
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
slotline (on bottom)
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
–
V1(z1) ground plane

Figure 11.23: 90◦ microstrip-slotline transition.

In both cases, only 4 complex parameters remain to be determined. If the junctions are lossless in
addition, the following real relationships follow from the unitary property (2.74) of the [S] matrix for
the E-plane tee:

|S11 |2 + |S12 |2 + |S13 |2 = 1 (11.72)

2|S13 |2 + |S33 |2 = 1 (11.73)
∗ ∗
S11 S12 + S12 S11 − |S13 |2 = 0 (11.74)
∗ ∗ ∗
S11 S13 − S12 S13 + S13 S33 = 0 (11.75)

These constraints mean that only 4 independent real parameters would now need to be found to com-
pletely specify the behavior of this junction. Similar constraints can also be obtained for the H-plane
junction.

11.7.2 90◦ Microstrip-slotline crossover

In section 10.5.3 and problem p10-13, we considered the excitation of a microstrip line by a voltage-
excited slot in the ground plane, and that of a slotline by a transverse current-carrying strip on the
opposite side of the substrate. If the respective sources of excitation in these two problems are regarded
as a slotline and a microstrip carrying suitable voltage and current in the vicinity of the crossover point
z1 = z2 = 0 of a 90◦ microstrip-slotline transition (see Fig. 11.23), then we can use the results to obtain
an equivalent circuit for the crossover. From the results of the two excitation problems mentioned above,
we can extract the equivalent circuit of Fig. 11.24(a). From eqn. (2.10), we recognize that the behavior
of this equivalent circuit is the same as that of an ideal 1:1 transformer as connected in Fig. 11.24(b),
and thus this too is an equivalent circuit for the crossover, valid at least for low enough frequencies that
all phenomena involved can be regarded as quasistatic.
The elements of the scattering matrix can be determined by placing matched loads at various ports
and exciting the remaining ones in certain ways. For example, if Zm is the characteristic impedance of

373
 I1(z1) V2(0)
+ -
(Zm) microstrip

z1 = 0 z1

+ slotline
V2(z2) (Zs)
I1(0)
–

z2 = 0 z2

(a)

port 1 port 2 microstrip

z1 = 0 z1
1:1

slotline
port 3 port 4

z2 = 0 z2

(b)

Figure 11.24: (a) Dependent generator equivalent circuit for Fig. 11.23; (b) Ideal transformer equivalent
circuit.

374
 Figure 11.25: Equivalent network for Fig. 11.23 if ports 3 and 4 are matched.

the microstrip and Zs is that of the slotline, placing matched loads of Zs at ports 3 and 4 means that the
1:1 transformer in the microstrip line presents a lumped series impedance of Zs /2 in the line. Choosing
the terminal planes for ports 1 and 2 at z1 = 0 (and likewise for ports 3 and 4 at z2 = 0), we reduce to
the two-port problem as shown in Fig. 11.25. If, further, port 2 is connected to a matched load of Zm
and port 1 is driven with (say) a voltage of V1 = 1, then we can easily evaluate the resulting voltage V2
and currents I1 and I2 at ports 1 and 2. Wave amplitudes a1 , a2 , b1 and b2 are then found from (2.21),
and we find: 
b1  Zs
S11 = = (11.76)
a1 a2 =a3 =a4 =0 Zs + 4Zm

b2  4Zm
S21 = = (11.77)
a1 a2 =a3 =a4 =0 Zs + 4Zm
From the physical symmetry of the junction, we have S11 = S22 and S12 = S21 . The other elements of
[S] can be found in a similar way.

11.7.3 A microstrip coupler

As has been seen in section 3.3, the presence of distributed coupling (capacitive and/or inductive)
between two or more transmission lines can result in significant transfer of energy between the two lines.
The more detailed treatment of this coupling mechanism by analysis of the electromagnetic fields will
be done in detail in chapter 12. Without yet presenting this theoretical background, let us present here
some typical results for a microstrip coupler—the four-port shown in Fig. 11.26. The magnitudes and

Figure 11.26: Microstrip coupler.

phases of the scattering parameters Si1 are shown in Fig. 11.27. Notice how significant coupling in both
the “forward” (S31 ) as well as “reverse” (S41 ) directions occurs, depending on the frequency of operation.

375
Figure 11.27: S-parameters for a microstrip coupler: ǫr = 12.9; w = 0.074 mm; g = 0.074 mm; L = 2.00
mm; h = 0.10 mm.

376
At approximately 25 GHz, nearly all incident power at port 1 is delivered either to port 2 or port 3. The
amount of power as well as the ports to which it is delivered can be varied by adjusting the parameters
of the coupler.

11.7.4 Intersecting dielectric waveguides

In order to achieve significant densities of circuit elements in integrated optics, ways have to be found
to allow dielectric waveguides to cross each other without undesired crosstalk or radiation loss. One
way of doing this is simply to let two such guides intersect each other as shown in Fig. 11.28. This

Figure 11.28: Intersecting dielectric slab waveguides.

configuration results in coupling, not only between the surface waves of the two guides, but also between
the surface waves and radiation modes of the two guides. There is a complex interference pattern due
to the superposition of these effects, and the resulting structure exhibits radiated power loss, as well as
an S13 that goes to zero at a certain value of intersection angle between the guides (Fig. 11.29). The
existence of this so-called “magic” angle suggests a promising way of crossover for integrated optical or
millimeter wave dielectric waveguides, with relatively minor radiation losses.

377
Figure 11.29: (a) |S13 |2 , and (b) radiation loss 1 − |S1i |2 as a function of angle of intersection for
P
dielectric slab waveguides.

378
11.8 Notes and References
In a very short period of time, waveguide network theory has become well enough studied to be regarded
as classical. More details can be found in the treatments of

Altman (1964);
Collin (1966);
Ghose (1963);
Kerns and Beatty (1967);
Montgomery, Dicke and Purcell (1965).

Shorter treatments are available in

Harrington (1961), chapter 8;

Ramo, Whinnery and van Duzer (1984), chapter 11;
Johnson (1965), chapter 14;
van Bladel (1964), chapter 13.

Mode-matching techniques are covered in

Vassallo (1985), section 8.3;

Uher et al. (1993), section 2.1;
G. Conciauro et al. (2000), chapter 3

as well as in

J. R. Pace and R. Mittra, “Generalized scattering matrix analysis of waveguide discontinuity

problems,” in Proceedings of the Symposium on Quasi-Optics. Brooklyn, NY: Polytechnic
Press, 1964, pp. 177-197.
G. I. Veselov and V. M. Temnov, “The solution of some systems of equations in electrody-
namics and the ‘relative convergence’ phenomenon,” Radio Eng. Electron. Phys., vol.
26, no. 10, pp. 13-20 (1981).
Y. C. Shih, “The mode-matching method,” in Numerical Techniques for Microwave and
Millimeter-Wave Passive Structures (T. Itoh, ed.). New York: Wiley, 1989, chapter 9.
A. S. Il’inskii and E. Yu. Fomenko, “Investigation of infinite-dimensional systems of linear
algebraic equations of the second kind in waveguide diffraction problems,” Comp. Math.
Math. Phys., vol. 31, no. 3, pp 1-11 (1991).
G. V. Eleftheriades, A. S. Omar, L. P. B. Katehi and G. M. Rebeiz, “Some important prop-
erties of waveguide generalized scattering matrices in the context of the mode matching
technique,” IEEE Trans. Micr. Theory Tech., vol. 42, pp. 1896-1903 (1994).
A. Morini and T. Rozzi, “On the definition of the generalized scattering matrix of a lossless
multiport,” IEEE Trans. Micr. Theory Tech., vol. 49, pp. 160-165 (2001).
P. Russer, M. Mongiardo and L. B. Felsen, “Electromagnetic field representations and com-
putations in complex structures III: Network representations of the connection and sub-
domain circuits,” Int. J. Num. Model.: Elec. Netw., Dev. Fields, vol. 15, pp. 127-145
(2002).

The concept of multi-winding ideal transformers as circuit elements is discussed in the papers by Morini
and Rossi, and by Russer et al. above, as well as in

Newcomb (1966), section 1-2.

Other theoretical techniques for evaluating circuit parameters for waveguide and transmission-line
discontinuities are given in

379
 Montgomery, Dicke and Purcell (1965);
Collin (1960);
Ghose (1963);
Harrington (1961);
Johnson (1965);
van Bladel (1964);
Vainshtein (1969b);
Marcuvitz (1965);
Schwinger and Saxon (1968);
Jones (1964), chapter 5;
Mittra and Lee (1971);
Lewin (1975);
Katsenelenbaum et al. (1998);
Milton and Schwinger (2006), chapter 13.
and
F. E. Borgnis and C. H. Papas, “Electromagnetic waveguides and resonators,” in Handbuch
der Physik, vol. 16 (S. Flügge, ed.). Berlin: Springer-Verlag, 1958, pp. 285-422.
Although a network formalism may, in principle, be used to treat discontinuities in optical or other
dielectric waveguides just as for any other type of waveguide, impedance concepts are not generally
invoked in this context. The scattering matrix is, however, sometimes used, and analysis techniques for
discontinuities in such open waveguides are to be found in
Marcuse (1982), chapter 9;
Marcuse (1974);
Shevchenko (1971).
Discontinuities involving curvature and other changes in direction of the waveguide axis are given in
Lewin (1975), chapter 4;
Lewin, Chang and Kuester (1977);
Londergan, Carini and Murdock (1999)
The expression of the stored energies of radiating junctions in terms of “observable” or near-field
stored energies has been treated from differing points of view. See:
D. R. Rhodes, “ Observable stored energies of electromagnetic systems,” J. Franklin Inst.,
vol. 302, pp. 225-237 (1976).
D. R. Rhodes, “A reactance theorem,” Proc. Roy. Soc. London, vol. A353, pp. 1-10 (1977).
I. M. Polishchuk, “The Q-factor and energy center of antennas,” Radio Eng. Electron. Phys.,
vol. 28, no. 6, pp. 1-5 (1983).
Chapter 5 of
King (1965)
discusses equivalent circuits of many types of discontinuities in coaxial and two-wire lines. Equation
(11.56) can be found there, as well as in
Vainshtein (1969b)
where expression (11.57) is also given.
The results of Figs. 11.12, 11.20 and 11.27 are obtained from numerical results generated at the
University of Colorado, Boulder. The programs for analysis of microstrip circuit discontinuities were
developed by B. L. Brim, D. I. Wu, J. Zheng and D. C. Chang:

380
 J. X. Zheng and D. C. Chang, “Numerical modeling of chamfered bends and other microstrip
junctions of general shape in MMICs,” IEEE International Microwave Symposium Digest,
vol. 2, pp. 709-712 (1990).
D. I. Wu and D. C. Chang, “A review of the electromagnetic properties and the full-wave
analysis of the guiding structures in MMIC,” Proc. IEEE, vol. 79, pp. 1529-1537 (1991).
D. C. Chang and J. X. Zheng, “Electromagnetic modeling of passive circuit elements in
MMIC,” IEEE Trans. Micr. Theory Tech., vol. 40, pp. 1741-1747 (1992).
The open-ended dielectric waveguide is treated by many authors. The results given here were taken
from
S. Ray and R. Mittra, “Numerical analysis of open waveguide discontinuities,” Radio Science,
vol. 19, pp. 1289-1293 (1984).
See also
P. Kaczmarski and P. E. Lagasse, “Bidirectional beam propagation method,” Electron. Lett.,
vol. 24, pp. 675-676 (1988).
Equivalent circuits and their parameters for a variety of microstrip discontinuities are given in
Gupta, Garg and Bahl (1979), chapters 3 and 4.
Similar data for coplanar waveguides are given in
Wolff (2006), chapters 3-5.
A highly useful collection of formulas and graphs for junctions and discontinuities in hollow rectangular
and circular waveguides as well as coaxial transmission lines is found in:
Marcuvitz (1965).
For more on the microstrip-slotline crossover of section 11.7.2, see
J. B. Knorr, “Slot-line transitions,” IEEE Trans. Micr. Theory Tech. vol. 22, pp. 548-554
(1974).
H.-Y. Yang and N. G. Alexopoulos, “A dynamic model for microstrip-slotline transition and
related structures,” IEEE Trans. Micr. Theory Tech. vol. 36, pp. 286-293 (1988).
T. Uwano, T. Itoh and R. Sorrentino, “Characterization of microstrip-to-slotline transition
discontinuities by transverse resonance analysis,” Alta Frequenza vol. 57, pp. 183-191
(1988).
The crossover of two dielectric slab waveguides has been examined in
N. Agrawal, L. McCaughan and S. R. Seshadri, “A multiple scattering interaction analysis
of intersecting waveguides,” J. Appl. Phys., vol. 62, pp. 2187-2193 (1987).
L. McCaughan and N. Agrawal, “Novel physical effects in intersecting waveguides,” Appl.
Phys. Lett., vol. 51, pp. 1389-1391 (1987).
from which the data in Fig. 11.29 are taken.

11.9 Problems
p11-1 Find the remaining elements of the scattering matrix for the 4-port network representing
the microstrip-slotline transition described in example 11.7.2. The terminal planes for ports
1 and 2 are to be taken at z1 = 0, while those for ports 3 and 4 are to be taken at z2 = 0.

381
p11-2 Consider the magic-tee shown below, where all 4 ports are rectangular metallic waveguides
of identical cross section supporting only the TE10 mode above cutoff.

Using the properties of reciprocity, losslessness and physical symmetry of this junction, find
as many relationships as you can between the elements of the scattering matrix for this
junction. Look especially for any property which might explain the name “magic-tee” which
is used for this structure—are any ports isolated from each other?

p11-3 Suppose that port 2 of a microstrip-slotline 90◦ crossover is terminated in a short circuit,
and port 4 is terminated in an open circuit.

(a) Find the scattering matrix of the 2-port which results from the remaining ports
1 and 3.
(b) What further modification(s) to this structure are needed to realize a reflec-
tionless microstrip-to-slotline transition if Z1 6= Z2 ?
(c) Suggest ways to achieve the idealized short and open circuit terminations at
ports 2 and 4, in view of the fact that at the physical locations of these terminal
planes there are likely to be significant amounts of higher-order modes present.

p11-4 Consider the step-junction of two dielectric slab waveguides as shown.

x
ε2

amA bmB aB
z
Waveguide A aA ε1
bmA amB
Waveguide B
ε2

z=0

382
 Using only the TE0 mode on each waveguide, perform a simple mode-matching procedure to
obtain the S-parameters of this junction. Multiply both sides of the equation for continuity
of the tangential E-field by the modal H field of waveguide B, and the equation for tangential
H by the E field of waveguide A to obtain your results.

p11-5 Show that [SBA ] = [SAB ]−1 , where the matrices are given in (11.29) and (11.30).

p11-6 If m′ denotes a TEM mode on waveguide A, and m denotes a TEM mode on waveguide B,
shown that the overlap integral (11.16) for these two modes is equal to
r
ZcA
Cmm′ =
ZcB

no matter what the detailed structure of either waveguide, where ZcA,B are the characteristic
impedances of the TEM modes on guides A and B respectively.

p11-7 Consider a mode “m” of a uniformly filled metallic waveguide (it may be either TE or
TM). When the medium filling the waveguide has material parameters ǫ1 and µ1 , we call the
(1) (1) (1)
waveguide “guide 1” and denote the mode fields Em , Hm , the propagation constant γm ,
(1)
and the intrinsic wave impedance of the mode (5.60) as ζm . If the material filling the guide
has ǫ2 , µ2 , we call the guide “2” and replace the superscripts (1) in the previous notations
by (2) .

(a) Let mode m be incident from guide 1 to guide 2, the junction between the two guides
being a plane interface located at z = 0 as shown. Mode m in guide 1 is assumed to be
(1) (1)
above cutoff (γm = jβm ).

z=0

µ1, ε1 µ2, ε2

z
Show that only a reflected mode m in guide 1 and a transmitted mode in guide 2 are
produced, and that if the transverse electric and magnetic fields are given by
(1)
(1) (1) E
ET = (e−jβm z
+ ρejβm z ) qmT (z < 0)
(1)
ζm

(1)
(1) (1) H
HT = (e−jβm z
− ρejβm z ) q mT (z < 0)
(1)
ζm
(2) (2)
EmT e−jβm z
ET = τ q (z > 0)
(2)
ζm

383
 (2) (2)
H e−jβm z
HT = τ mT
q (z > 0)
(2)
ζm
then the reflection and transmission coefficients are given by
(2) (1) (2)
ζm − ζm 2ζm
ρ= (2) (1)
; τ= (2) (1)
ζm + ζm ζm + ζm

(b) Suppose a second interface back to the material of guide 1 is placed at z = l as shown.

z=0 z=l

µ1, ε1 µ2, ε2 µ1, ε1

z
Find expressions for the total ET and HT in 0 < z < l, and find the time-average power
passed in the positive z-direction for any z in this region. Give two separate expressions
for the power, depending on whether mode m in guide 2 is propagating or cutoff.

384
Chapter 12

COUPLED-MODE THEORY

When I get to the bottom I go back to the top of the

slide
Where I stop and I turn and I go for a ride
Till I get to the bottom and I see you again.
Yeah, yeah, yeah.
—John Lennon/Paul McCartney,
Helter Skelter

12.1 Field Analysis of Coupled Waveguides

The next part to be written.

12.2 Notes and References

Optical waveguide stuff, mostly.

385
386
Chapter 13

RESONANT ELEMENTS FOR

WAVEGUIDES

There is a tide in the affairs of men

Which, taken at the flood, leads on to fortune;
Omitted, all the voyage of their life
Is bound in shallows and in miseries.
On such a full sea are we now afloat;
And we must take the current when it serves,
Or lose our ventures.
—William Shakespeare,
Julius Caesar

13.1 Introduction
Another work in progress.

13.2 Notes and References

For various treatments of resonant elements in transmission line and waveguide circuits, see
Adler, Chu and Fano (1960), chapter 6;
Ghose (1963), chapter 8;
Altman (1964), chapter 5;
Montgomery, Dicke and Purcell (1965);
Collin (1966), chapter 7;
Slater (1969), chapter 4;
as well as many of the other references given at the end of chapter 11.

13.3 Problems
p13-1 There are always problems.

387
388
Appendices

That is one of the tricks

Which you must employ
If you want to be a cowboy
—Ralph McTell,
When I was a Cowboy

389
390
Appendix A

FOURIER AND LAPLACE

TRANSFORMS

In this Appendix, we collect some useful properties of Fourier and Laplace transforms. Further infor-
mation can be found in the references at the end of this Appendix. The notation θ(t) is used for the
Heaviside unit step function:
θ(t) = 1 (t > 0)
= 0 (t < 0) (A.1)
The gamma function Γ(ν) and the Bessel function Jν (z) are defined in Appendix B. The Dirac delta
function is denoted δ(t).

A.1 Fourier Transforms

Given a (generally complex) function f (t) with suitable properties, we define its Fourier transform F (ω)
as Z ∞
F (ω) = f (t)e−jωt dt (A.2)
−∞
The original function f (t) can be recovered from a knowledge of its Fourier transform through the inverse
transform relation Z ∞
1
f (t) = F (ω)ejωt dω (A.3)
2π −∞
For two functions f (t) and g(t) and their corresponding Fourier transforms, we have the Parseval relation:
Z ∞ Z ∞
1
f (t)g ∗ (t) dt = F (ω)G∗ (ω) dω (A.4)
−∞ 2π −∞
∗
where denotes the complex conjugate.

A.2 Table of Fourier Transforms

We present in Table A.1 a brief table of Fourier transforms that are needed in this book.
A word is in order about the derivation of the Fourier transform of the Gaussian function. It follows
from the identity Z ∞
2 √
e−x dx = π (A.5)
−∞

391
 R∞
Formula f (t) F (ω) = −∞
f (t)e−jωt dt

1 δ(t) 1

2 1 2πδ(ω)

1 π −|ω|T
3 t2 +T 2 Te

2
/T 2 √ 2 2
4 e−t T πe−ω T /4

Table A.1: Table of Fourier transforms.

which can be proved as follows. Denote the integral on the left side of (A.5) by I. Then the square of I
can be re-expressed using a change from rectangular to polar coordinates in the integrations:
Z ∞ Z ∞ Z 2π Z ∞ Z ∞
2 2 2 2
I2 = e−x dx e−y dy = e−ρ ρdρdφ = 2π e−ρ ρdρ (A.6)
−∞ −∞ 0 0 0

Changing the integration variable to u = ρ2 gives I 2 = π, or since I is obviously positive, (A.5). The
desired Fourier transform then follows by some strategic changes of integration variables.

A.3 Table of Laplace Transforms

We present in Table A.2 a brief table of Laplace transforms that are needed in this book. The Bessel

R∞
Formula f (t) F (s) = 0
f (t)e−st dt

1 δ(t) 1

1
2 θ(t) s

Γ(ν+1)
3 tν θ(t) (ν > −1) sν+1

1 −bs
4 θ(t − b) (b ≥ 0) se
 ν
5 Jν (at)θ(t) (ν > −1, a > 0) √ 1 √a
s2 +a2 s+ s2 +a2
√
√ e−b s +a 2 2
6 J0 (a t2 − b2 )θ(t − b) (b ≥ 0, a ≥ 0) √
s2 +a2
√ √
2 2
J1 (a t2 −b2 ) e−bs −e−b s +a
7 √
t2 −b2
θ(t − b) (a ≥ 0, b ≥ 0) ab
√ √
Rt 2 2
u2 −b2 ) e−bs −e−b s +a
8 ab b J1 (a
√
u2 −b2
du θ(t − b) (a ≥ 0, b ≥ 0) s

Table A.2: Table of Laplace transforms.

function Jν and the gamma function Γ appearing in this table are described in sections B.1 and B.2
respectively of Appendix B.

392
A.4 Table of Rules for the Laplace Transform
In Table A.3, we collect some general relations which allow the Laplace transform of one function to be
expressed in terms of the transform of another related one. Here, the quantities a, T , etc., are real positive
constants (unless otherwise noted), while n denotes a positive integer. The transform of a function
denoted by a lower-case letter is represented by the corresponding upper-case letter (g(t) → G(s), for
example). All functions of time are assumed implicitly to contain a factor θ(t). In this table, A and a

R∞
Formula f (t) F (s) = 0
f (t)e−st dt

1 g(t − T )θ(t − T ) e−sT G(s)

2 Re [Ae−at g(t)] Re [AG(s + a)]

1 s


3 g(at) aG a
n
4 tn g(t) (−1)n d ds
G(s)
n

dn g(t)
5 dtn sn − sn−1 f (0) − sn−2 f ′ (0) − . . . − f [n−1] (0)
Rt 1
6 0
g(u) du s G(s)
Rt
7 0 g1 (u)g2 (t − u) du G1 (s)G2 (s)

Table A.3: Table of rules for Laplace transforms.

are complex constants. As an example of the use of these rules, let g1 (u) = g(u) and g2 (u) = θ(u) in
Formula 7 of Table A.3. Then using Formula 2 of Table A.2, we obtain Formula 6 of Table A.3.

393
A.5 Notes and References
More extensive treatments of the Fourier transform are given in:

Davies (1978), chapters 7-11.

Titchmarsh (1986).
Extensive tables of the Fourier transform are given in:
Erdélyi (1954), chapters 1-3.
Good references for the properties and applications of the Laplace transform are:
Davies (1978), chapters 1-6.
van der Pol and Bremmer (1987).
More extensive tables of the Laplace transform are given in:
Erdélyi (1954), chapters 4 and 5.

394
Appendix B

SPECIAL FUNCTIONS

B.1 Bessel Functions

Bessel’s differential equation
d2 f df
x2 + x2 − ν 2 f = 0


2
+x (B.1)
dx dx
is a linear, second-order ordinary differential equation, and therefore will have two linearly independent
solutions. One of these can be obtained by power series methods, and is given by
f (x) = Jν (x)
∞
X (−1)m (x/2)2m+ν
= (B.2)
m=0
m!Γ(m + ν + 1)

where Γ is the gamma function or factorial function (see Section C.2). The function Jν (x) is known as
the Bessel function of the first kind of index ν (or order ν) and argument x.
Clearly J−ν (x) is also a solution of (B.1). However, when ν = n is an integer, it turns out that
J−n (x) is not an independent solution; in fact
J−n (x) = (−1)n Jn (x) (B.3)
So instead of using J−ν (x) as our second solution, we define the Bessel function of the second kind Yν (x)
as:
Jν (x) cos νπ − J−ν (x)
Yν (x) = (B.4)
sin νπ
Even though this expression becomes indeterminate (0/0) when ν → n, the limit exists and is still
independent of Jn (x). As a consequence of (B.3) and (B.4), we have also
Y−n (x) = (−1)n Yn (x) (B.5)
The Wronskian of the two types of Bessel functions satisfies the relation:
2
Jν (z)Yν′ (z) − Yν (z)Jν′ (z) = (B.6)
πz
Plots of J0 , J1 , Y0 , and Y1 are given in Figure B.1. Notice that these functions oscillate much like
sinusoidal functions, but do not have precisely regular periods and decay slowly as x increases. Also
note that Y0 and Y1 (and indeed all the Yν functions) have a singularity (i.e., “blow up”) at the origin
x = 0, while J0 , J1 , etc., are well-behaved there. In fact, by (B.2):
1  x ν
Jν (x) ≃ ; (ν 6= −1, −2, . . .) (B.7)
Γ(ν + 1) 2

395
 2

1.5
J0(x) Y0(x)

J1(x) Y1(x)
1

0.5

0 2 4 6 8 10
x

0.5

1.5

Figure B.1: Graphs of J0 (x), Y0 (x), J1 (x) and Y1 (x).

as x → 0, but  ν
Γ(ν) 2
Yν (x) ≃ − ; (ν > 0) (B.8)
π x
in the same limit. For the special cases of ν = 0 or 1,
2n x o
Y0 (x) ≃ ln( ) + γE
π 2
2
Y1 (x) ≃ − (B.9)
πx
where γE = .57721... is known as Euler’s constant, while
x
J0 (x) ≃ 1; J1 (x) ≃ (B.10)
2
as x → 0.
The zeroes of J0 and Y0 , J1 and Y1 , etc., interlace, like those of the sine and cosine, but unlike those
of the trigonometric functions, are not equally spaced. The positive zeroes of the Bessel function Jn (x)
are denoted by jnm , n denoting the index of the Bessel function, and m denoting the ordering of the
zero in increasing sequence. Equation (B.6) implies that the zeroes of Jn (x) are simple; that is, that
Jn′ (jnm ) 6= 0; since Yn (x) is finite for x > 0. The first several such zeroes are listed in Table B.1 to five
decimal places of accuracy. The first three zeroes of J0 (x) have also been indicated in Figure B.1. It is

396
also sometimes necessary to know the zeroes of Jn′ (x). We denote these by jnm ′
, so that Jn′ (jnm
′
) = 0.
′ ′
The first few values of jnm are given in Table B.2. Note that the roots of j1m are all smaller in magnitude
than those corresponding to other values of n. If |ν| ≥ 4, Re(ν) > 0 and m = 1, 2 or 3, an approximate

m n 0 1 2 3 4

1 2.40483 3.83171 5.13562 6.38016 7.58834

2 5.52008 7.01559 8.41724 9.76102 11.06471

3 8.65373 10.17347 11.61984 13.01520 14.37254

Table B.1: The first three nonzero roots jnm of the Bessel function Jn (x) for n = 0 to 4, to five decimal
places.

m n 0 1 2 3 4

1 3.83171 1.84118 3.05424 4.20119 5.31755

2 7.01559 5.33144 6.70613 8.01524 9.28240

3 10.17347 8.53632 9.96947 11.34592 12.68191

′
Table B.2: The first three nonzero roots jnm of Jn′ (x) for n = 0 to 4, to five decimal places.

formula for the mth nonzero root of the Bessel function Jν (x) is:
 ν 1/3 2  −1/3
3rm ν
jν,m ≃ ν + rm + (B.11)
2 20 2
Likewise, the mth nonzero root of the derivative Jν′ (x) is:
 ν 1/3  3r′2 1
 
ν −1/3
′ ′ m
jν,m ≃ ν + rm + − ′
(B.12)
2 20 10rm 2

The error for these approximations is no more than 1.5%, the worst case being ν = 4 and m = 3. The
′
accuracy is often considerably better. Values for the quantities rm and rm used in these expressions are
given in Table B.3.
Bessel functions of half-integer order can be expressed in closed form using elementary functions. For
example, r
2
J1/2 (x) = sin x (B.13)
πx
and r
2
Y1/2 (x) = − cos x (B.14)
πx
The zeros of (B.13) are obviously integer multiples of π, while those of (B.14)are (m + 12 )π. Zeroes of
′ ′
J1/2 (x) are not obtainable by elementary means; its lowest root is j1/2,1 = 1.16556 to five decimal places.

397
 ′
m rm rm

1 2.33810741 1.01879297

2 4.08794944 3.24819758

3 5.52055983 4.82009921

′
Table B.3: Values of rm and rm , to eight decimal places.

When the argument of the Bessel functions is large (|x| ≫ |ν|π ≥ 0), Jν and Yν can be approximated
by damped trigonometric functions:
r
2  νπ π
Jν (x) ∼ cos x − − (B.15)
πx 2 4
r
2  νπ π
Yν (x) ∼ sin x − − (B.16)
πx 2 4
Evidently, the zeroes of Jν for m ≫ ν are approximately
 
π 1
jνm ≃ 2m + ν − (B.17)
2 2

In this sense, Jν and Yν are analogous to the cosine and sine functions.
It is sometimes useful to define solutions of Bessel’s equation which are analogous to complex expo-
nential functions (i. e., e±jx ). These are the Hankel functions of the first and second kind:

Hν(1) (x) ≡ Jν (x) + jYν (x) (B.18)

Hν(2) (x) ≡ Jν (x) − jYν (x) (B.19)

(1) (2)
The behavior of Hν and Hν for large or small x is easily deduced from (B.9), (B.10), (B.15) and
(B.16). In particular, r
2 j(x− νπ − π )
Hν(1) (x) ∼ e 2 4 (B.20)
πx
r
2 −j(x− νπ − π )
Hν(2) (x) ∼ e 2 4 (B.21)
πx
for x >> νπ ≥ 0. From (B.4), (B.18) and (B.19), we have the relations
(1) (2)
H−ν (x) = ejνπ Hν(1) (x); H−ν (x) = e−jνπ Hν(2) (x) (B.22)

for expressing negative-index Hankel functions in terms of those with positive index.

B.1.1 Modified Bessel functions

Consider now the following variation of Bessel’s equation:

d2 g dg
y2 +y − (y 2 + ν 2 )g = 0 (B.23)
dy 2 dy

398
The change of variable x = jy transforms (B.23) into (B.1). Therefore, solutions of (B.23) can be
chosen from linear combinations of Jν (jy) and Yν (jy). It is always desirable (when possible) to deal
with functions which are real when the argument is real. In this case, such functions are the so-called
modified Bessel functions, Iν and Kν , defined by
νπ
Iν (y) ≡ e−j 2 Jν (jy) (B.24)

πj j νπ (1)
Kν (y) ≡ e 2 Hν (jy) (B.25)
2
The modified Bessel functions are analogous to the pure exponentials e±y . Plots of I0 , I1 , K0 and K1
are given in Figure B.2. Evidently, the Kν must blow up at y = 0. In fact, we have

1.5

I0(x) K0(x)
I1(x) K1(x)
0.5

0 0.5 1 1.5 2 2.5 3 3.5 4

x

Figure B.2: Graphs of I0 (y), I1 (y), K0 (y) and K1 (y).

y
K0 (y) ≃ −[ln( ) + γ] (B.26)
2

(l − 1)! 2 l
Kl (y) ≃ ( ); l≥1 (B.27)
2 y
as y → 0, while
I0 (y) ≃ 1 (B.28)

399
 (y/2)l
Il (y) ≃ ; l≥1 (B.29)
l!
in the same limit, where l is an integer. Finally, we have

I−n (y) = In (y) (B.30)

for any integer n, while

K−ν (y) = Kν (y) (B.31)
for any value of ν.
For y >> 1, Iν are essentially growing exponentials, while Kν are decaying ones:

ey
Iν (y) ∼ √ (B.32)
2πy

π −y
r
Kν (y) ∼ e (B.33)
2y
Neither Iν nor Kν has any positive zeroes (in y) for ν ≥ 0. Note that Kν (y) > 0 and Kν′ (y) < 0 for any
real y > 0.

B.1.2 Recurrence relations

It is possible to express derivatives of Bessel functions or modified Bessel functions in terms of other
orders of the same type of function. Thus, for example,
ν
Jν′ (x) = Jν−1 (x) − Jν (x) (B.34)
x
ν
Jν′ (x) = −Jν+1 (x) + Jν (x) (B.35)
x
ν
Kν′ (y) = −Kν−1 (y) − Kν (y) (B.36)
y
ν
Kν′ (y) = −Kν+1 (y) + Kν (y) (B.37)
y
Many other similar types of relations can also be obtained.

B.1.3 Integral formulas

A wealth of formulas for evaluating integrals containing Bessel functions are available in the literature.
We will quote only those which are needed in this book:

x2  2
Z
xZn2 (x) dx =

Zn (x) − Zn−1 (x)Zn+1 (x) (B.38)
2

y2  2
Z
yKn2 (y) dy =

Kn (y) − Kn−1 (y)Kn+1 (y) (B.39)
2
1 2 n 1 2 1
Z Z
Z (x) dx = Z (x) dx − [Z 2 (x) + Zn+1
2
(x)] (B.40)
x n+1 n+1 x n 2(n + 1) n
In these formulas, Zn (x) = AJn (x) + BYn (x) denotes an arbitrary linear combination of Jn (x) and
Yn (x).

400
B.2 The Gamma Function
The gamma function (or factorial function) is defined by the integral formula
Z ∞
Γ(z) = e−t tz−1 dt (B.41)
0

if z > 0. It can be defined for all values of z except for 0 and the negative integers, where it becomes
infinite. For positive integers, it becomes an ordinary factorial:

Γ(n + 1) = n!; n = 0, 1, 2, . . . (B.42)

For half-integer values of z, it is also possible to compute Γ(z); we have in particular,

∞ √
1
Z
2
Γ( ) = 2 e−u du = π (B.43)
2 0

Γ(x)

4 3 2 1 0 1 2 3 4
x

Figure B.3: The gamma function.

For any z, we have the properties

Γ(z + 1) = zΓ(z) (B.44)

and
π
Γ(z)Γ(1 − z) = (B.45)
sin πz

401
B.3 The Elliptic Integrals K and E
The complete elliptic integral of the first kind is defined by
π/2
dθ
Z
K(k) = p (B.46)
0 1 − k 2 sin2 θ

where k is called the modulus of K. The complete elliptic integral of the second kind is defined by
Z π/2 p
E(k) = 1 − k 2 sin2 θ dθ (B.47)
0

Graphs of these two functions are shown in Figure B.4. Both elliptic integrals can be computed using

K(k)
3
E(k)

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure B.4: The elliptic integrals K and E.

most standard mathematical software packages. However, it is important to be aware that some software
and literature uses the notation m = k 2 as the argument of the elliptic integrals. Be sure to check
carefully before using software or results found elsewhere.
The elliptic integrals obey a wide variety of identities. Of these we quote here only the following:
π
E(k)K(k ′ ) + E(k ′ )K(k) =
+ K(k)K(k ′ ) (B.48)
2
 
dK(k) 1 E(k)
= − K(k) (B.49)
dk k k ′2

402
 √
where k ′ is known as the complementary modulus k ′ ≡ 1 − k 2 .
One rapid method to compute the elliptic integrals is the method of the arithmetic-geometric mean.
Let
L0 = 1; M0 = k ′ (B.50)
Now generate the sequences Ln and Mn by
1 p
Ln+1 =
(Ln + Mn ); Mn+1 = Ln Mn (B.51)
2
These sequences converge to the same limit m(M0 , N0 ). Then
π
K(k) = (B.52)
2m(1, k ′ )
For values of k not extremely close to 1, only a few steps of (B.51) are needed to obtain an extremely
accurate value of K. The elliptic integral E(k) can be computed with the help of an additional sequence
Nn , determined by
1
N0 = k; Nn+1 = (Ln − Mn ) (B.53)
2
from which we then express E(k) as
∞
" #
X
n−1 2
E(k) = K(k) 1 − 2 Nn (B.54)
n=0

The infinite series converges rapidly for most values of k, and can be truncated after a sufficiently large
number of terms.
Often it is the ratio K(k)/K(k ′ ) rather than individual elliptic integrals which is encountered in
applications. There exist extremely accurate yet simple approximations for this ratio when k is real and
between 0 and 1:
√ !
K(k) 1 1+ k
≃ ln 2 √ for 12 ≤ k 2 ≤ 1
K(k ′ ) π 1− k
K(k) π 1
≃ √ ! for 0 ≤ k 2 ≤ 2 (B.55)
K(k ′ ) 1+ k′
ln 2 √
1 − k′
The maximum relative error in (B.55) is 3 × 10−6 .
Eqns. (B.55) are invertible to obtain k as an explicit function of the ratio K(k)/K(k ′ ). This is
commonly done using the so-called nome 1 q defined as
q = exp[−πK(k ′ )/K(k)] (B.56)
When 0 ≤ k 2 ≤ 1/2, we have 0 ≤ K(k)/K(k ′ ) ≤ 1, or 0 ≤ q ≤ 0.0432. In this case,
p
4 q(1 + 4q 2 ) K(k)
k≃ 2
; 0≤ ≤1 (B.57)
(1 + 2q) K(k ′ )
When 1/2 ≤ k 2 ≤ 1, we have 1 ≤ K(k)/K(k ′ ) ≤ ∞, or 0.0432 ≤ q ≤ 1. In this case we define a
complementary nome q ′ as
q ′ = exp[−πK(k)/K(k ′ )] (B.58)
so that 0 ≤ q ′ ≤ 0.0432. Then,
2
1 − 2q ′

K(k)
k≃ ; 1≤ ≤∞ (B.59)
1 + 2q ′ K(k ′ )
1 This should not be confused with the filling factor q defined in section 9.4, just as the modulus k should not be

confused with the wavenumber of a TEM mode or plane wave.

403
B.4 A Function Arising in Microstrip Analysis
A function that has no special name (it is related to the family of zeta functions) is

r(x) ≡ exp[Q(x)] (B.60)

where
∞  
X m+1
Q(x) ≡ xm ln (B.61)
m=1
m

When |x| < 1, which is the case if x = (1 − ǫr )/(1 + ǫr ) as above, a simple yet accurate approximation
for r(x) is p
r(x) ≃ 1 + 1.3471u + 0.3152u2 (B.62)
where u = x/(1 − x). Graphs of Q(x) and its approximation ln r(x) for −1 < x < 1 are shown in
Fig. B.5. This function arises in the analysis of microstrip structures.

Q(x)

ln[r(x)] 2

1 0.8 0.6 0.4 0.2 0 0.2 0.4 0.6 0.8 1 x

Figure B.5: The function Q(x) and its approximation.

B.5 Notes and References

For Bessel functions, see

Abramowitz and Stegun (1964), chapter 9;

N.N. Lebedev (1972), chapters 5 and 6.

The expansions (B.11) and (B.12) are given in

404
 F. W. J. Olver, “A further method for the evaluation of zeros of Bessel functions and some
new asymptotic expansions for zeros of functions of large order,” Proc. Camb. Phil. Soc.,
vol. 47, pp. 699-712 (1951).
along with further terms in the series which may be retained for more accuracy. More accurate (but also
more elaborate) expansions can be found in
F. W. J. Olver, “The asymptotic expansion of Bessel functions of large order,” Phil. Trans.
Roy. Soc. London A, vol. 247A, pp. 328-368 (1954).
′
The quantities −rm and −rm are the roots of the Airy function and its derivative, respectively.
For gamma functions, see
Abramowitz and Stegun (1964), chapter 6;
N.N. Lebedev (1972), chapter 1.
Properties of elliptic integrals can be found in
Byrd and Friedman (1971).
The approximation (B.55), among others, can be found in
W. Hilberg, ”From approximations to exact relations for characteristic impedances”, IEEE
Trans. Micro. Theory Tech., vol. 17, pp. 259-265 (1969).
Simple programs for computing these functions on a programmable pocket calculator are given in:
Henrici (1977), pp. 214-221, 241-267.
Ball (1978), pp. 210-221.
The function treated in section B.4 was studied in
E.F. Kuester, “Accurate approximations for a function appearing in the analysis of mi-
crostrip”, IEEE Trans. Micr. Theory Tech., vol. 32, pp. 131-133 (1984).

405
406
Appendix C

NUMERICAL SOLUTION OF
IMPLICIT EQUATIONS

Transcendental equations such as we have obtained for determining the propagation constants or other
line parameters of dielectric waveguides or planar transmission lines can be solved numerically on a
computer or programmable calculator by a variety of methods. Many commercial software packages
offer the capability of solving these kinds of equations. It can, however, be critical to arrange the
transcendental equation in a form which is stable for numerical solution procedures. Often, additional
insight into the solution can be gained by writing an algorithm especially to solve a particular problem.
We will briefly describe a few methods for finding roots (zeroes) of a real function F (x) of a real
variable x. Even though a user may not have programmed the root-finding algorithm, it is important
that properties of the algorithm be known so that the reliability of the results obtained can be judged.
It should especially be noted that no algorithm can guarantee to find all roots of a given function
unless more information about that function is available. Moreover, functions possessing singularities or
varying rapidly may make it difficult to find their roots, so rearranging the function into a different form
may sometimes be necessary, as will be illustrated for the iterative method below. Methods for finding
complex roots of complex functions are in general more complicated, and will not be treated here (see
the references for more information).

C.1 The Bisection Method

Perhaps the simplest numerical technique is the bisection method. Suppose we know that a root of F (x)
lies between the lower limit xl and the upper limit xu . We might know this if the values of F (xl ) and
F (xu ) have opposite signs, but this might not be the case if the function approaches infinity within the
interval (xl , xu ) (as for the function tan x at x = π/2, for example). We therefore restrict consideration
to functions that are bounded and continuous on the interval (xl , xu ). Now we consider the new point
xn = (xl + xu )/2 at the center of the interval. If F (xn ) has the same sign as F (xl ), then a root of F
must lie within the smaller interval (xn , xu ); if F (xn ) has the same sign as F (xu ), then a root of F must
lie within (xl , xn ). In either case, the region containing a root has been cut in half, and we can repeat
this process until the length of the interval containing a root has become small enough to determine the
root to a prespecified accuracy.

C.2 Newton’s Method

The bisection process is simple and reliable, but not particularly rapidly convergent. A more rapidly
convergent (but sometimes less reliable) is Newton’s method. Here, we do not need an interval known to

407
contain a root, but only an initial estimate x0 of the location of a root. Suppose that we can compute
not only the value of the function F at x0 , but also its derivative F ′ (x0 ). This can be done either
analytically if the form of the function is available, or by the finite difference approximation
F (x + δ) − F (x)
F ′ (x) ≃ (C.1)
δ
for some sufficiently small quantity δ (though not too small, so that we avoid problems with computer
roundoff error). Then the function F is approximated near x0 by a linear function obtained from the
first two terms of its Taylor series. This approximation is set equal to zero and solved for a new estimate
x1 of the root:
F (x0 )
x1 = x0 − ′ (C.2)
F (x0 )
This process is repeated until the approximation xn to the root is as accurate as required:
F (xn )
xn+1 = xn − ; n = 1, 2, . . . (C.3)
F ′ (xn )
Note that this will not work for multiple roots (where F (x) = F ′ (x) = 0), but when the process does
converge, it does so much faster than does the bisection method.

C.3 Iterative Methods

One method that is well-suited to the equations we encounter in guided-wave theory is the iterative
method, or method of successive substitution. The idea of this method is as follows. Suppose we have
a transcendental equation F (x) = 0 for an unknown quantity x, and that this equation can be written
in the form:
x = f (x) (C.4)
where f is some given function of x. If the rearrangement into the form (C.4) is such that a given variation
in x causes much less variation in f (x) (logarithms, arctangents, etc. are examples of such functions),
then we embark on the following succession of substitution. Choose a guess x0 for the solution. Then
compute the sequence

x1 = f (x0 )
x2 = f (x1 )
..
.
xn+1 = f (xn )

of improved approximate solutions until further iteration results in negligible change in xN +1 over xN .
Then xN +1 is the solution of (C.4) to desired accuracy. Note that it is crucial that f (x) varies less
rapidly in x than x itself. Otherwise the iteration procedure can run away rather than converge to the
desired answer.
This method can be used to generate tables, plots or individual values of the solutions of the tran-
scendental equations encountered in this text. As an example, consider two forms of the eigenvalue
equation for the even TE modes of a dielectric slab waveguide. The form (6.48) that was used for the
graphical solution can be written in the form:
q
2
h1 a = V 2 − [h1 a tan(h1 a/2)] (C.5)

This form is unsuitable for iterative solution, because the right side contains a tangent function of the
variable h1 a, which varies much more rapidly than h1 a itself—in fact, it becomes infinite when h1 a is

408
an odd multiple of π. This form of the equation is also unstable for numerical use because for some
values of h1 a the right side can become complex, whereas such solutions are not allowed from a physical
standpoint.
On the other hand, the form (6.53) of this eigenvalue equation (which is in fact applicable to both
even and odd modes) can be written as:
2
√

mπ 2
b=1− + sin−1 b (C.6)
V V

Here, the right side contains only an inverse trigonometric function of the unknown b, which varies more
slowly than b itself, especially when V is not small. This equation has the added advantage that one
can specify which mode m is desired, while (C.5) has more than one solution if V is large enough, and it
may be difficult to make a numerical algorithm select the solution for a particular mode. The reader is
encouraged to experiment with the iterative algorithm for different versions of the eigenvalue equation,
to see how efficient and accurate it can be.
This example contains an inverse trigonometric function; these and logarithms must be treated with
care in any root-finding technique. Both these kinds of function involve an arbitrary integer multiple of
π, and mathematical software is designed to choose a specific value for this integer during calculations.
This choice for these functions is usually called the principal branch of the function, and the program’s
documentation should be consulted for information on exactly what that choice is.

C.4 Notes and References

For information on finding complex roots of complex functions, see
Press et al. (1992), chapter 9;
Hamming (1986), chapter 5.
as well as
C. A. Barlow and E. L. Jones, “A method for the solution of roots of a nonlinear equation
and for solution of the general eigenvalue problem,” J. Assoc. Comput. Mach., vol. 13,
pp. 135-142 (1966).
B. Davies, “Locating the zeros of an analytic function,” J. Comp. Phys., vol. 66, pp. 36-49
(1986).

409
410
Appendix D

FORMULAS FROM VECTOR,

MATRIX AND TENSOR
ANALYSIS

D.1 Vectors
An ordered set of numbers is called a vector. Sometimes this is a set of similar physical quantities, such
as the voltages at the numbered nodes of a circuit. Other times, a vector may consist of exactly three
numbers which represent the components in some coordinate system of a directed physical quantity such
as current density, velocity, force, electric field, etc. These latter are often called Gibbsian vectors, and
are denoted by boldface letters in this book. Some formulas from the algebra and analysis of Gibbsian
vectors are given below.
We denote the unit vector in a given coordinate direction (say, v) by uv . Hence in cartesian coor-
dinates we have (ux , uy , uz ) and in cylindrical coordinates we have (uρ , uφ , uz ). A vector is generally
represented in boldface, as A. When a vector has no z-component we call it a transverse vector, and
denote it by AT , for example. The operator ∇ is defined as
∂ ∂ ∂
∇ = ux + uy + uz (D.1)
∂x ∂y ∂z

in cartesian coordinates, while

∂ ∂
∇T = ux + uy (D.2)
∂x ∂y
The following vector identities are often found useful; each holds equally well if ∇ is replaced by ∇T :

A · (B × C) = B · (C × A) = C · (A × B) (D.3)

A × (B × C) = (A · C)B − (A · B)C (D.4)

∇(f g) = f ∇g + g∇f (D.5)
∇ · (f A) = f ∇ · A + A · ∇f (D.6)
∇ × (f A) = f ∇ × A − A × ∇f (D.7)
∇ · (A × B) = B · ∇ × A − A · ∇ × B (D.8)
∇ × ∇f = 0 (D.9)

411
 ∇ · (∇ × A) = 0 (D.10)
If V is a volume bounded by a closed surface S and un is the outward unit normal to S, then we
have the divergence theorem Z I
(∇ · A) dV = un · A dS (D.11)
V S

and its variants Z I

(∇f ) dV = f un dS (D.12)
V S
Z I
(∇ × A) dV = (un × A) dS (D.13)
V S

If S is an open surface whose unit normal is un and whose closed boundary is the contour C, then we
have Stokes’ theorem Z I
(∇ × A) · un dS = A · dℓ (D.14)
S C

and its variant Z I

(un × ∇f ) dS = f dℓ (D.15)
S C

where the direction of the vector line element dℓ is related to un by the right-hand rule. If S is a surface
lying in a plane z = constant, C is its closed boundary, and un is an outward unit normal to C lying in
the plane of S, then Z I
(∇T · AT ) dS = AT · un dℓ (D.16)
S C
Z I
(∇T f ) dS = f un dℓ (D.17)
S C

where dℓ is the scalar arc length element on C. There also hold the following miscellaneous identities:

uz × (uz × AT ) = −AT (D.18)

∂A
uz × (∇ × A) = ∇Az − (D.19)
∂z
∇T × (uz × AT ) = uz (∇T · AT ) (D.20)

D.1.1 Vector differentiation in various coordinate systems

Here we present expressions for vector differential operations in the three major coordinate systems.

Rectangular (cartesian) coordinates

∂f ∂f ∂f
∇f = ux + uy + uz (D.21)
∂x ∂y ∂z

∂Ax ∂Ay ∂Az

∇·A = + + (D.22)
∂x ∂y ∂z
     
∂Az ∂Ay ∂Ax ∂Az ∂Ay ∂Ax
∇ × A = ux − + uy − + uz − (D.23)
∂y ∂z ∂z ∂x ∂x ∂y

412
Circular cylindrical coordinates

∂f 1 ∂f ∂f
∇f = uρ + uφ + uz (D.24)
∂ρ ρ ∂φ ∂z
1 ∂(ρAρ ) 1 ∂Aφ ∂Az
∇·A = + + (D.25)
ρ ∂ρ ρ ∂φ ∂z
     
1 ∂Az ∂Aφ ∂Aρ ∂Az 1 ∂(ρAφ ) 1 ∂Aρ
∇ × A = uρ − + uφ − + uz − (D.26)
ρ ∂φ ∂z ∂z ∂ρ ρ ∂ρ ρ ∂φ

Spherical coordinates

∂f 1 ∂f 1 ∂f
∇f = ur + uθ + uφ (D.27)
∂r r ∂θ r sin θ ∂φ

1 ∂(r2 Ar ) 1 ∂(sin θAθ ) 1 ∂Aφ

∇·A= 2
+ + (D.28)
r ∂r r sin θ ∂θ r sin θ ∂φ

     
1 ∂(sin θAφ ) ∂Aθ 1 1 ∂Ar ∂(rAφ ) 1 ∂(rAθ ) ∂Ar
∇×A = ur − +uθ − +uφ − (D.29)
r sin θ ∂θ ∂φ r sin θ ∂φ ∂r r ∂r ∂θ

D.2 Matrices
In linear algebra, one often has occasion to transform an ordered set of numbers into another; this kind
of transformation is called a matrix. Some formulas from matrix algebra are given below.
If [A] and [B] are square, N × N matrices, with elements Aik and Bik respectively, then [A]T (the
transpose of [A] is the square matrix whose elements are Aki , and the Hermitian conjugate [A]† of [A] is
the square matrix whose elements are A∗ki , where ∗ denotes the complex conjugate. The identity matrix
[1] is the identity matrix, defined by 1ii = 1, and 1ik = 0 for i 6= k. The inverse [A]−1 of a square matrix
is defined by
[A]−1 [A] = [A][A]−1 = [1] (D.30)
The inverse matrix does not always exist; if it does not, [A] is said to be singular. If a nonsingular matrix
is partitioned as follows:  
 [E] [F ] 
[A] = 



[G] [H]

where [E], [F ], [G] and [H] are suitable matrices of smaller size, then its inverse is expressed as:
 
 [E]−1 + [E]−1 [F ][M ]−1 [G][E]−1 −[E]−1 [F ][M ]−1 
[A]−1 = 


 (D.31)
−[M ]−1 [G][E]−1 [M ]−1

where
[M ] = [H] − [G][E]−1 [F ]
provided that [E] and [M ] are nonsingular. Repeated application of (D.31) allows the inverse of any
nonsingular square matrix to be computed.

413
 The following identities are useful:

([A][B])T = [B]T [A]T (D.32)

−1
([A][B]) = [B]−1 [A]−1 (D.33)
T †
A symmetric matrix is one for which [A] = [A]. A Hermitian matrix is one for which [A] = [A].
An N -element column vector [X] is constructed from a set of elements Xi ; i = 1, 2, . . . , N . Then its
transpose [X]T is an N -element row vector whose elements are Xi . Two N -element column vectors [X]
and [Y ] are said to be orthogonal if [Y ]T [X] = 0. If [X1 ], [X2 ], . . . , [XM ] (with M ≤ N ) are N -element
column vectors, we can always form a set of M linear combinations of them that are all orthogonal to
each other. One method for doing this is called Gram-Schmidt orthogonalization. The procedure is to
construct a new set of vectors [Y1 ], [Y2 ], . . . , [YM ] as follows. First, put

[Y1 ] = [X1 ] (D.34)

Next, form the vector

[Y2 ] = [X2 ] + a[X1 ] (D.35)
T
where the constant a is chosen to make [Y2 ] [Y1 ] = 0, i. e.,

[X2 ]T [X1 ]
a=− (D.36)
[X1 ]T [X1 ]

A similar procedure is used to construct each successive [Yi ]. For example, form the vector

[Y3 ] = [X3 ] + b[X2 ] + c[X1 ] (D.37)

and choose the constants b and c to make [Y3 ]T [Y1 ] = [Y3 ]T [Y2 ] = 0.

D.3 Dyadics or Tensors

When we deal with vectors in cartesian 3-space such as electromagnetic fields, currents, etc., it is common
practice to refer to linear transformations of them not as matrices, but as dyadics or second-rank tensors.
Such tensors are defined to be sums of one or more terms called dyads. A dyad is an association of two
vectors—not a scalar or vector product—which
↔
can operate on ordinary vectors according to the rules
we lay out below. We write a dyad D of two vectors A and B symbolically as
↔
D = AB (D.38)
↔
without the use of a dot or cross-product sign. We then define right and left dot products of D with a
third vector F as
↔
F·D = (F · A)B
↔
D · F = A(B · F) (D.39)
↔
In this sense, we can think of D as if it were the 3 × 3 matrix:
 
 Ax Bx Ax By Ax Bz 
 
 
D→
 Ay Bx Ay By Ay Bz 
 (D.40)
 
 
Az Bx Az By Az Bz

414
The identity dyadic or tensor is defined by
↔
I = ux ux + uy uy + uz uz (D.41)

An operation which is a little less commonly encountered with matrices is the cross product, but it can
readily be defined for dyads as:
↔
F×D = (F × A)B
↔
D × F = A(B × F) (D.42)

This definition is readily extended to include any tensor. Notice that the result of taking the cross
product of a tensor with a vector is once again a tensor.

D.4 Problems
pD-1 Prove that uz × (uz × AT ) = −AT for any transverse vector AT .

415
416
Appendix E

PROPERTIES OF SOLUTIONS TO
NETWORK AND
TRANSMISSION LINE
EQUATIONS

Since the classical transmission line is a limiting case of a lumped-element circuit, we may prove any
result for the case of an arbitrary circuit, and carry it over to the transmission-line case by taking
that limit. We will treat the properties of the network and classical transmission-line equations by
emphasizing their structural analogy with Maxwell’s equations. This allows us to see how theorems such
as reciprocity, power balance, equivalence and so on bear a close mathematical resemblance in all three
situations, and to recognize their essential similarity.

E.1 A Standard Form for Network Equations

Consider a linear network with n + 1 nodes, including a reference (or datum or ground ) node labeled
0. An example of such a network is shown in Fig. E.1. Let the network contain b branches connecting
the various nodes. On each branch, reference directions for current in the branch and voltage across the
branch are chosen as shown to obey the reference convention of circuit analysis.
We next define some terms from graph theory commonly used in circuit analysis. A tree of a graph
is a set of branches from the graph chosen so that (i) all nodes of the network are met by at least one
of the branches in the tree, and (ii) removal of any branch from the tree leaves at least one node not
connected to the tree. A graph of n + 1 nodes has trees with n tree branches. The remaining l = b − n
branches of the network are called links. A loop is a connected set of branches such that each node is
connected by exactly two of these branches. The fundamental loop of a link is the unique loop which
is formed from certain tree branches along with that link only. A cut-set is a set of branches chosen
so that (i) removal of all cut-set branches from the graph leaves two disconnected subgraphs remaining,
and (ii) removal of some but not all of the cut-set branches leaves a connected graph remaining. The
fundamental cut-set of a tree branch is the unique cut-set formed from certain links along with that tree
branch only.
Now choose any tree of the network, labeling the voltages across the tree branches Vr ; r = 1, . . . , n,
and the currents through the links Ik ; k = 1, . . . , l. Denote the link voltages by Uk ; k = 1, . . . , l and the
tree branch currents by Jr ; r = 1, . . . , n. Then Kirchhoff’s voltage law enforced around the fundamental

417
 +
Ik
Uk

Link

+ Vr -

Jr

Tree Branch

Figure E.1: Graph of example linear network with n = 3 and b = 5.

loop of each of the links in the circuit can be expressed in matrix form as

[U ] = −[D]T [V ]

Kirchhoff’s current law can be enforced on the fundamental cut-set for each of the tree branches: the
sum of the currents through all the cut-set branches (measured from one of the resulting disconnected
subgraphs to the other) must be zero. This is expressed in matrix form as

[J] = [D][I]

where [U ], [V ], [J] and [I] are column vectors constructed from the tree branch and link currents and
voltages described above, and the superscript T denotes the matrix transpose.
The matrix [D] is a submatrix of the fundamental cut-set matrix of the graph of the network, whose
elements are given by:
Drk = +1
if tree branch r is in the fundamental loop of link k, with the same reference direction as the loop is
traversed;
Drk = −1
if tree branch r is in the fundamental loop of link k, with the opposite reference direction as the loop is
traversed; and
Drk = 0
if tree branch r is not in the fundamental loop of link k.1
1 It can also be shown that D
rk = −1 if link k is in the fundamental cut-set of tree branch r, and passes between the
resulting pair of disconnected subgraphs with the same reference direction; Drk = +1 if link k is in the fundamental cut-set
of tree branch r, and passes between the resulting pair of disconnected subgraphs with the opposite reference direction;
and Drk = 0 if link k is not in the fundamental cut-set of tree branch r.

418
 Z
k
Y ext
r Jr
- ext
Uk
+

(a) (b)

Figure E.2: (a) Typical link and (b) typical tree branch, in the absence of mutual coupling.

To account for the constitutive properties of the elements making up the circuit, let us first assume
that the tree branches consist of admittances Yr in parallel with externally impressed current sources
(that is, ideal current sources) Jrext (Fig. E.2(a)). The links are assumed to consist of impedances Zk in
series with ideal voltage sources Ukext (Fig. E.2(b)). If no mutual coupling between elements is present,
this can always be done, by virtue of Thévenin’s and Norton’s equivalence theorems. Then Ohm’s law
for each element leads to the equations
[J] = [Y ][V ] + [J ext ]
[U ] = [Z][I] + [U ext ]
where [Y ] and [Z] are diagonal matrices formed by the element admittances and impedances indicated
above. Then, eliminating the vectors [U ] and [J] from the foregoing, we have
[D][I] = [Y ][V ] − [J ext ] (E.1)
−[D]T [V ] = [Z][I] − [U ext ] (E.2)
which is a matrix analog of the telegraphers’ equations and of Maxwell’s equations, with [D] playing the
role of the differential operator d(·)/dz or ∇ × (·).
When mutual couplings are present between branches of the network, we can account for them by
using dependent voltage or current sources in addition to the impressed sources already present above.
For the case of sources linearly dependent on other voltages or currents in the network, we can add a
column vector of dependent current sources to the tree branches:
[J dep ] = [Y ′ ][V ] + [T ][I]
and dependent voltage sources to the links:
[U dep ] = [Z ′ ][I] + [W ][V ]
so that
[D][I] = [Y ][V ] + [T ][I] − [J ext ] (E.3)
T ext
−[D] [V ] = [Z][I] + [W ][V ] − [U ] (E.4)
where [Y ] and [Z] are now general (not necessarily diagonal) square matrices, and [T ] and [W ] are
general rectangular matrices. The analogy of these equations to the Maxwell field equations is preserved
if the medium is regarded as bianisotropic (e. g., a magnetoelectric or Tellegen medium).

419
E.2 Reciprocity for Lumped Circuits and Classical Transmis-
sion Lines
E.2.1 Lumped networks
Consider two possible states of the network considered above. That is, consider the same network, with
two possible external excitations: [J a ] and [U a ] (state a), or [J b ] and [U b ] (state b).2 The link voltages
and tree branch currents will obey (E.3)-(E.4), with the appropriate source vectors present. We now
form the matrix products

[V a ]T [D][I b ] = [V a ]T [Y ][V b ] + [V a ]T [T ][I b ] − [V a ]T [J b ]

−[I b ]T [D]T [V a ] = [I b ]T [Z][I a ] + [I b ]T [W ][V a ] − [I b ]T [U a ]

These quantities are scalars, and the left sides of these expressions are the negatives of each other by
the rule for taking the transpose of a matrix product. Thus adding these equations gives

0 = [V a ]T [Y ][V b ] + [V a ]T [T ][I b ] − [V a ]T [J b ]
+[I b ]T [Z][I a ] + [I b ]T [W ][V a ] − [I b ]T [U a ]

A similar relation is obtained by interchanging the states a and b. If we subtract this from the original
one, we get

[V a ]T [Y ] − [Y ]T [V b ] + [V a ]T [T ] + [W ]T [I b ]
 

−[V b ]T [T ] + [W ]T [I a ] + [I b ]T [Z] − [Z]T [I a ]

 

−[V a ]T [J b ] + [V b ]T [J a ] − [I b ]T [U a ] + [I a ]T [U b ] = 0

If the matrices arising from the constitutive equations of the elements and dependent sources obey
the conditions that [Y ] and [Z] are symmetric, and [T ] = −[W ]T , then the first four terms in this
equation drop out, and we are left with the simpler form

[V a ]T [J b ] − [I a ]T [U b ] = [V b ]T [J a ] − [I b ]T [U a ]

In other words, under these conditions, the indicated combination of products of impressed voltages of
one state with induced currents of another and impressed currents of one with induced voltages of the
other does not change when the states are interchanged. This is a form of the reciprocity property for
networks. One special case for which it is true is that of a lumped element network with no mutual
coupling between branches of the network.
As an application, consider a linear source-free N -port network as in Fig. 2.15. Whatever network is
internal to this multiport, choose a tree of that network so that each port of the network is spanned by
a branch of the tree, with reference directions for the tree branch voltages chosen the same as for the
port voltages. Let the network be driven by ideal current sources J1 , J2 , . . . , JN connected to each port;
there are no voltage sources exciting this network. The current sources become the port currents I1 , I2 ,
. . . , IN (not to be confused with the internally defined link currents). The general reciprocity property
above gives us:
[V a ]T [I b ] = [V b ]T [I a ]
where [V ] and [I] are now the N × 1 column vectors representing port voltages and currents respectively.
If [Z] is now taken to represent the N × N impedance matrix for the N -port, we find that this [Z]
matrix must be symmetric if the internal network of the multiport satisfies the criteria given above for
reciprocity, and in particular if it is made up entirely of uncoupled lumped impedances or admittances.
2 We drop the superscript ext hereafter for brevity.

420
E.2.2 Transmission lines
Although as stated above we could simply make limiting arguments to derive the reciprocity theorem for
classical transmission lines, it is instructive to see the derivation done directly. The classical transmission
line equations (with nonuniform line parameters and distributed sources) are:
dV (z)
= −z(z)I(z) + Eext (z)
dz
dI(z)
= −y(z)V (z) + Hext (z) (E.5)
dz
Considering these equations for two possible states, a and b, arising from two separate excitations of the
line, we can obtain:
d
V a I b − V b I a = (V a Hext
b
− I a Eext
b
) − (V b Hext
a
− I b Eext
a


)
dz
If integrated between two points z1 and z2 on the line, this yields
Z z2
a b b a z2
 a b
(V Hext − I a Eext
b
) − (V b Hext
a
− I b Eext
a


V I − V I z1 = ) dz (E.6)
z1

These are the analogs of the lumped-circuit reciprocity properties of the previous subsection, and of the
Lorentz reciprocity properties for Maxwell’s equations derived in chapter 9.

E.3 Power Flow in Networks and Classical Transmission Lines

E.3.1 Lumped networks
The time-dependent power flow from all the independent sources into a lumped-element network is the
sum of the power flows from the individual sources:
X X X
p(t) = ps (t) ≡ uk (t)ik (t) + vr (t)jr (t)
s k r

where ps (t) = uk (t)ik (t) for the ideal voltage source uk in link k, and ik is the current flowing out of
its positive terminal, or ps (t) = vr (t)jr (t) for the ideal current source jr in tree branch r, and vr is the
voltage across its terminals.
When all sources are time-harmonic with the same angular frequency ω, the product of any pair of
resultant time-harmonic quantities (e. g., voltage and current, voltage squared, etc.) can be represented
in terms of the corresponding phasors (cf. (1.3)) in the following way:
p(t) = v(t)i(t)
1
= [V ejωt + V ∗ e−jωt ][Iejωt + I ∗ e−jωt ]
4
1
= Re[V I ∗ + V Ie2jωt ]
2
= pav + posc (t)
Thus, the power consists of a constant part, pav , which is the time-average power flow, and an oscillatory
part posc which has time average value zero, but in this case indicates the temporary exchange of energy
back and forth from the terminals where v and i are measured. It is customary for calculations of power
flow to define the complex power 3
1
P̂ = V I ∗ ≡ P + jQ (E.7)
2
3 It is conventional to define complex power using the complex conjugate of the current, rather than of the voltage

(which would have been equally possible, as the real part of P̂ is not affected).

421
where P is sometimes called the active power and Q the reactive power, and the complex oscillatory
power
1
P̂osc = V I (E.8)
2
so that the average and oscillatory powers can be written as
pav = Re(P̂ ) = P (E.9)
and
posc (t) = Re(P̂osc e2jωt ) (E.10)
The two complex power quantities are independent of t and are given solely by phasor voltage and
current quantities, so they can be used to completely characterize the power flow in a time-harmonic
lumped-element network. Note that the reactive power Q does not directly appear in the expression for
time-dependent power flow.
It should be noted that when the power is given in terms of a single voltage and current as in the
previous paragraph, the magnitudes of P̂ and P̂osc are equal; only their phases are different. However,
when a number of such terms is added together to compute a total power flow (as in what follows below),
we cannot expect |P̂ | = |P̂osc | any longer unless special conditions hold (such as the phases of all currents
being the same, for example). In general, the time-average and oscillatory parts of the power flow are
completely unrelated.
The total power flow is related to the energy storage in the network. The same techniques used
to derive the reciprocity theorem can be used to obtain this relationship. Consider the total complex
oscillatory power delivered by all the sources in a network:
X 1X 1X 1 T
[I] [U ] + [V ]T [J]

Posc = P̂s,osc = Uk Ik + Vr Jr =
s
2 2 r
2
k

Using the circuit equations (E.1) and (E.2) for a lumped element network with no mutual couplings (this
restriction can be relaxed), we have
1 T
[I] [Z][I] + [V ]T [Y ][V ] = PL,osc + 2jω (WM,osc + WE,osc )

Posc = (E.11)
2
where
1 T
[I] [R][I] + [V ]T [G][V ]

PL,osc =
2
is the complex oscillatory power dissipated in the circuit resistances and conductances, and
1 T 1 T
WE,osc = [V ] [C][V ]; WM,osc = [I] [L][I]
4 4
are the complex oscillatory electric and magnetic stored energies in the circuit capacitances and induc-
tances respectively.
Likewise, the complex power P̂ delivered to the circuit can be expressed as4
1 T ∗
[V ] [J] + [I]† [U ] = pL,av + 2jω (wM,av − wE,av )

P̂ = (E.12)
2
where
1 †
[I] [R][I] + [V ]† [G][V ]

pL,av =
2
is the time-average power lost to the circuit conductances and resistances, and
1 † 1 †
wE,av = [V ] [C][V ]; wM,av = [I] [L][I]
4 4
4 The superscript † denotes the Hermitian conjugate, or complex conjugate of the transpose, of a matrix.

422
are the time-average electric and magnetic stored energies in the circuit capacitances and inductances
respectively. Note that the reactive power Q delivered to the network is twice the angular frequency ω
times the difference between the average stored magnetic energy and average stored electric energy in
the circuit. It is thus positive if more magnetic energy is stored than electric, and negative otherwise.
In the special case of a lossless LC-network with only frequency-independent inductors in the links and
capacitors in the tree branches, excited by independent sources [J] and [U ] that are also independent of ω,
we can perform similar manipulations with the frequency derivatives [Vω ] = d[V ]/dω and [Iω ] = d[I]/dω
of [V ] and [I] and the matrix equations they satisfy. Leaving the details to the reader, we arrive at the
following identity:
j
wM,av + wE,av = − [Vω ]T [J]∗ + [Iω ]T [U ]∗


4
Now if we set [U ] = 0 and regard the current sources [J] as applied to the ports of an N -port network
whose N × N impedance matrix is [Z] as before, we find that

j 1
wM,av + wE,av = − [J]† [Zω ][J] = [J]† [Xω ][J]
4 4
since the impedance matrix for a lossless network is purely reactive: [Z] = j[X]. Similarly, driving the
ports of an N -port network with voltage sources [U ] only, we have

1 †
wM,av + wE,av = [U ] [Bω ][U ]
4
where the admittance matrix of the network is purely susceptive: [Y ] = j[B]. In both cases, the subscript
ω again denotes differentiation with respect to frequency. Thus the total time-average stored energy (in
a lossless network only) can be related to the frequency derivatives of the reactance or susceptance
matrices of a multiport. For a lossy network, no such general statement is possible.

E.3.2 Transmission lines

As with the reciprocity property, we will derive the power flow results for classical transmission lines
directly. The power carried towards positive z at the point z by the current and voltage on a classical
transmission line is p(z, t) = v(z, t)i(z, t). As with the lumped-element circuit, when v and i are time-
harmonic and expressed in terms of the phasors V (z) and I(z), we have

1
p(z, t) = [V (z)ejωt + V ∗ (z)e−jωt ][I(z)ejωt + I ∗ (z)e−jωt ]
4
1
= Re[V (z)I ∗ (z) + V (z)I(z)e2jωt ] (E.13)
2
= pav (z) + posc (z, t) (E.14)

Defining the complex power

1
P̂ (z) = V (z)I ∗ (z) (E.15)
2
and the complex oscillatory power
1
P̂osc (z) = V (z)I(z) (E.16)
2
at a position z along the line, the average and oscillatory powers can be written as

pav (z) = Re[P̂ (z)] (E.17)

and
posc (z, t) = Re[P̂osc (z)e2jωt ] (E.18)

423
The two complex power quantities depend only on z and are given solely by phasor voltage and current
quantities, and can be used to completely characterize the power flow in a classical transmission line.
The complex power quantities can be related to stored and dissipated energies on the line. Consider
a line on which the line parameters can possibly be functions of the position z on the line. Then the
telegrapher’s equations are again as in (E.5). Using these, we can obtain the identities:

d 1
P̂ (z) = − z|I|2 + y∗ |V |2

(E.19)
dz 2
d 1
P̂osc (z) = − zI 2 + yV 2

(E.20)
dz 2
The real part of (E.19) relates the change in power flow with respect to z to the time-average rate of
energy dissipation [r|I|2 + g|V |2 ]/2 per unit length along the line. The imaginary part of (E.19) has no
direct interpretation in terms of energy conservation (since it plays no part in the description of time
dependent power flow), but does relate the rate of change of the imaginary part of P̂ to the difference
between stored time-average inductive energy l|I|2 /4 and capacitive energy c|V |2 /4. Equation (E.20)
relates the oscillatory power flow to the oscillatory parts of the stored and dissipated energies.

E.3.3 Decibels (dB)

The power flow from a source to a passive load is a positive number. The unit for power is a Watt
(W), but often a relative unit of power, the decibel (dB) is used to express the ratio of two powers in a
convenient way:  
P1
PdB = 10 log
P2
and we say that P1 is PdB decibels above (or larger than) P2 . A single power level can be expressed in
the form  
P
PdB = 10 log , (E.21)
Pref
where P is the power we are measuring or calculating, and Pref is some given reference power level. At
microwave frequencies, very often this reference power level is 1 mW, and in this case the unit is called
a dBm:  
P
PdBm = 10 log (E.22)
1mW
many other reference power levels are in common use.
A positive number of decibels corresponds to a ratio greater than 1 (gain), while a negative number
represents a ratio less than 1 (loss). Decibels are convenient for two reasons: (1) they are easier to write
(for example, the range between +63 dB to -153 dB corresponds to 2 · 106 to a factor of 0.5 × 10−15 ),
and (2) adding decibels corresponds to multiplication, which is useful whenever there are several stages
cascaded in some system (for example, in a multistage amplifier, the gain can be found simply by adding
the individual gains in dB).
Since power is proportional to the square of voltage in a circuit or on a transmission line, we can
also use voltage ratios to obtain decibel levels for gain or loss, provided the same reference impedance
or characteristic impedance is used for both voltages being compared. The same is, incidentally, true of
currents. Thus, a voltage V1 is said to be
 
V1
PdB = 20 log
V2

decibels above the voltage V2 on a transmission line of identical characteristic impedance. It is important
to keep in mind that the decibel is always a measure of a power ratio; to say that voltages are so many

424
dB apart is somewhat an abuse of the terminology, although a common one. In an obvious way, we can
also define dB relative to some given voltage reference level, such as the dBµV: the number of decibels
above 1 microvolt. Although it is done less frequently, we can express ratios of currents, or electric or
magnetic fields, in dB as well.

E.4 Notes and References

Good sources for the treatment of circuit analysis, including modern aspects such as cut-set analysis,
are
Seshu and Balabanian (1959), chapters 3 and 4.
Carlin and Giordano (1964), chapter 1.
Balabanian and Bickart (1969), chapters 2-4.
Desoer and Kuh (1969), chapters 9-11.

and also
T. R. Bashkow, “The A matrix, new network description,” IRE Trans. Circ. Theory, vol.
4, pp. 117-119 (1957).
C.-W. Ho, A. E. Ruehli and P. A. Brennan, “The modified nodal approach to network
analysis,” IEEE Trans. Circ. Syst., vol. 22, pp. 504-509 (1975).
Some treatments emphasizing the algebraic-topological aspects of network analysis and the formal analo-
gies between the network equations and Maxwell’s equations (using differential forms) are given in
F. H. Branin, Jr., “The algebraic-topological basis for network analogies and the vector calcu-
lus,” in Proceedings of the Symposium on Generalized Networks (J. Fox, ed.). Brooklyn,
NY: Polytechnic Press (1966), pp. 453-491.
Strang (1986), sect. 2.3.
G. Strang, “A framework for equilibrium equations,” SIAM Review, vol. 30, pp. 283-297
(1988).
Bamberg and Sternberg (1990).

425
426
Appendix F

PROPERTIES OF SOLUTIONS TO
THE ELECTROMAGNETIC
FIELD EQUATIONS

F.1 Maxwell’s Equations and Boundary Conditions

The electromagnetic field in any region of space must satisfy Maxwell’s equations:
∂h
∇×e = −µ
∂t
∂e
∇×h = ǫ + σe + jext (F.1)
∂t
∇ · (ǫe) = ρext
∇ · (µh) = 0

where ǫ, σ, and µ are the electrical parameters (possibly position-dependent) characterizing a region of
space, e(x, y, z, t) is the time-dependent electric field, h(x, y, z, t) the magnetic field and jext (x, y, z, t) is
the volume density of impressed electric current in the region.1 The del operator ∇ is defined by
∂ ∂ ∂
∇ = ux + uy + uz (F.2)
∂x ∂y ∂z
in cartesian coordinates, where ux , uy , uz are unit vectors in the x, y and z directions, respectively.
Phasor fields E and H corresponding to time-harmonic solutions e and h of (F.1) (or equally well,
Fourier transforms of e and h with respect to t) are required to satisfy the time-harmonic Maxwell curl
equations

∇×E = −jωµH
∇×H = (jωǫ + σ)E + Jext (F.3)
= jωǫ̂E + Jext

which follow from (F.1); the divergence equations are redundant unless ω = 0.

1 Remember that lower-case letters denote real functions (fields, currents, etc.) of (x, y, z) and time t.

427
 If fictitious magnetic currents and charges are allowed, the time-harmonic Maxwell equations have
the form

∇×E = −jωµH − Mext

∇×H = (jωǫ + σ)E + Jext (F.4)
= jωǫ̂E + Jext

Standard arguments based on integrating Maxwell’s equations over “pillbox” volumes or thin strip
surfaces lead to boundary or jump conditions for the electromagnetic field across surfaces of discontinuity
between different materials or surfaces containing surface-concentrated current distributions. The most
general form of them applicable to time-harmonic fields with electric as well as magnetic sources is

JS,ext = un × (H1 − H2 )S
MS,ext = (E1 − E2 )S × un (F.5)

where S is the surface n = 0 separating regions 1 (in n > 0) and 2 (in n < 0), and un is the unit vector
normal to S pointing into region 1.

F.2 Plane Waves

The simplest solutions to the source-free (Jext = 0) time-harmonic Maxwell equations in an infinite
homogeneous medium are plane waves. These have the form

E = E0 e−jk·r ; H = H0 e−jk·r (F.6)

where r = ux x + uy y + uz z is the position vector, and k = ux kx + uy ky + uz kz is called the wave vector

of the plane wave, which points in the direction of propagation of the wave if its components are real.
By substitution of (F.6) into (F.3), we find that the fields are perpendicular to k:

k · E0 = 0; k · H0 = 0 (F.7)
√
that k is related to the wavenumber k ≡ ω µǫ by:

k · k = k2 (F.8)

and that E0 and H0 are mutually perpendicular (again if k is real) and related by:

k
× E0 = ζH0 (F.9)
k

where ζ ≡ (µ/ǫ)1/2 is the intrinsic wave impedance of the medium.

F.3 Poynting’s Theorem

Consider a region of space occupied by a possibly inhomogeneous but isotropic material (µ(r), ǫ(r), and
σ(r) may be functions of position but are scalars) with externally impressed volume electric current
density jext (r, t). Maxwell’s equations (F.1) must be satisfied by the resultant fields e and h. Using
them, we can obtain the following identity:
 
∂ 1 1
∇ · (e × h) = − ǫe · e + µh · h − σe · e − e · jext (F.10)
∂t 2 2

428
The term e·jext has the physical meaning of the rate of work being done (per unit volume) by the Lorentz
force of the electric field on the charges (per unit volume) whose motion makes up the external impressed
current density, or equivalently, the negative of the work done by the external impressed sources in
delivering energy to the fields. Similarly, the term σe · e represents the rate of energy dissipation per
unit volume due to conduction losses in the medium. If we interpret the quantity
1 1
w = wE + wM = ǫe · e + µh · h
2 2
as the stored electric and magnetic energy in the electric and magnetic fields (as is rigorously true for
static fields), then the term  
∂w ∂ 1 1
= ǫe · e + µh · h
∂t ∂t 2 2
represents the rate of increase of stored energy in the field, and finally, ∇ · (e × h) must be interpreted
as the energy outflow rate (per unit volume) from the surface surrounding that unit volume. The vector
function e × h is known as the Poynting vector.
The result is often stated as the integral form of Poynting’s theorem:
Z  
d 1 1
Z I Z
pext (t) = − e · jext dV = (e × h) · un dS + ǫe · e + µh · h dV + σe · e dV (F.11)
V S dt V 2 2 V

where V is a volume of space bounded by the closed surface S. The left side of this equation represents
the rate of energy deliverance (power flow) to the system by the sources within V , the first term on the
right side is the rate of energy flow outward through the surface S (whose outward unit normal vector
is un ), the second term on the right is the rate of increase of stored energy within V , and the third is
the power dissipated as heat in the conductivity σ.
In the time-harmonic case, we can represent the fields and sources in terms of phasors, as we did
with voltages and currents in Appendix E. Again, the time-dependent power and energy quantities are
expressed as the sum of the time-average (constant) part and an oscillatory part. The time-average
power flow through the surface S, for example, is given by

pav = Re(P̂ )

where the complex power P̂ is

1 1
I I
∗
P̂ = (E × H ) · un dS = Ŝ · un dS
2 S 2 S

while the oscillatory part of this power is given by

posc (t) = Re(Posc ej2ωt )

where the complex oscillatory power is

1 1
I I
Posc = (E × H) · un dS = Sosc · un dS
2 S 2 S

We have defined the complex Poynting vector Ŝ = E × H∗ and the complex oscillatory Poynting vector
Sosc = E × H in these equations. As was noted for the case of lumped-element circuits and transmission
lines in Appendix E, we cannot expect that |P̂ | = |P̂osc | in general.
The analogs of (F.10) for phasor quantities are:

∇ · Ŝ = −jω [µH · H∗ − ǫE · E∗ ] − σE · E∗ − E · J∗ext (F.12)

∇ · Sosc = −jω [µH · H + ǫE · E] − σE · E − E · Jext (F.13)

429
Integrating these equations over the volume V and using the divergence theorem yields

P̂ext = P̂ + 2jω(wM,av − wE,av ) + pL,av

Pext,osc = Posc + 2jω(WM,osc + WE,osc ) + PL,osc

where
1
Z
wM,av = µH · H∗ dV
4 V

1
Z
wE,av = ǫE · E∗ dV
4 V

are the parts of the time-average stored magnetic and electric energies in V not due to material dispersion,
1
Z
pL,av = σE · E∗ dS
2 V

is the time-average ohmic power loss in V , and

1
Z
P̂ext = − E · J∗ext dV
2 V

is the complex power delivered by the sources to the system. The quantities
1
Z
WM,osc = µH · H dV
4 V

1
Z
WE,osc = ǫE · E dV
4 V
1
Z
PL,osc = σE · E dS
2 V
1
Z
Pext,osc = − E · Jext dV
2 V
are the corresponding complex oscillatory powers or energies.
The real part of (F.12) relates the change in power flow through the surface S to the difference between
the time-average rate of energy generation and dissipation within V . The imaginary part of (F.12) has
no direct interpretation in terms of energy conservation (since it plays no part in the description of time
dependent power flow), but does relate the rate of change of the imaginary part of P̂ to the difference
between stored time-average magnetic energy and electric energy, and the imaginary part of the complex
externally supplied power. Equation (F.13) relates the oscillatory power flow to the oscillatory parts of
the stored and dissipated energies.

F.4 Elementary Dipoles

Elementary (or Hertzian) dipole sources are idealizations of very small electric or magnetic current
sources. The elementary electric dipole is described by a volume electric current density of

J = jωpδ(x − x0 )δ(y − y0 )δ(z − z0 ) (F.14)

while the elementary magnetic dipole is described by the magnetic current density

M = jωµmδ(x − x0 )δ(y − y0 )δ(z − z0 ) (F.15)

430
where the vector p is called the electric dipole moment and m the magnetic dipole moment. The
strengths of these dipole moments are denoted by the scalars p = |p| and m = |m| respectively.
In (F.14) and (F.15), δ represents the so-called Dirac delta-function, more properly called a generalized
function, which has the properties
δ(x) = 0

if x 6= 0, and
Z x2
δ(x) dx = 1
x1

if x1 < 0 < x2 . Thus, the sources in (F.14) and (F.15) are concentrated at the single point in space
(x0 , y0 , z0 ).
The elementary electric dipole is a limiting case of a short electric current filament of length ∆l
carrying a current I, as ∆l → 0 and I → ∞ in such a way that I∆l → jωp, with p pointing in the
direction of positive I. The elementary magnetic dipole is a similar limiting case of a small loop of
electric current I which spans an area ∆S → 0 such that I∆S → m and m is perpendicular to the
surface spanned by the loop, and related to the direction of current flow by the right-hand rule (see
problem p9-1).
The fields of such dipoles in a homogeneous medium of infinite extent (permittivity ǫ, permeability µ,
wave impedance ζ and wavenumber k) are found by standard methods of antenna theory, for example.
The fields of an electric dipole are given by:

1 1
H= ∇ × A; E= ∇×H
µ jωǫ

where the vector potential A is given by

jωµ −jkR
A=p e
4πR
p
and R = (x − x0 )2 + (y − y0 )2 + (z − z0 )2 is the distance between the observation point (x, y, z) and
the dipole location (x0 , y0 , z0 ). If p = puz , these fields are expressed in spherical coordinates as:
 
(jωp)k sin θ 1
Hφ = j 1+ e−jkR
4πR jkR
 
(jωp) cos θ 1
Er = ζ 1 + e−jkR
2πR2 jkR
 
(jωp)k sin θ 1 1
Eθ = jζ 1+ − e−jkR
4πR jkR (kR)2
The fields of a magnetic dipole oriented along the z-axis are:
 
2 m sin θ 1
Eφ = k ζ 1+ e−jkR
4πR jkR
 
m cos θ 1
Hr = jk 1 + e−jkR
2πR2 jkR
 
m sin θ 1 1
Hθ = −k 2 1+ − e−jkR
4πR jkR (kR)2

431
 J
M

h z

(z=0)

(a)
J
M

h z

(z=0)

i i
J M
(b)

Figure F.1: (a) Sources above a perfectly conducting plane; (b) Image equivalent for z > 0.

F.5 Image Theory

Consider the electric and magnetic currents located above the plane z = 0 which is taken to be a perfect
conductor, as shown in Fig. F.1(a). We seek an equivalent auxiliary problem (the image problem) in
which the ground plane is absent and additional sources are present in z < 0 such that the resulting
fields in z > 0 are identical with those of case (a). These additional sources must be such that, together
with the original sources in z > 0, they produce a zero value of uz × E at z = 0 in the image problem
(Fig. F.1).
For example, if M = 0 and J = uz Jz (x, y, z) in the original problem, and if we denote the fields
this source distribution would produce in infinite free space (in the absence of the ground plane) as
E0 (x, y, z) and H0 (x, y, z), then the image sources
Ji (x, y, z) = uz Jz (x, y, −z)
Mi (x, y, z) = 0 (F.16)
will produce the fields (also in the absence of the ground plane)
Ei (x, y, z) = uz E0z (x, y, −z) − E0t (x, y, −z)
i
H (x, y, z) = −uz H0z (x, y, −z) + H0t (x, y, −z) (F.17)
Here “t ” denotes components of a vector transverse (perpendicular) to z in the xy-plane. By superposing
these with the original fields to get the total field in case (b):
E = E0 + Ei
H = H0 + Hi (F.18)
we find that uz × E|z=0 is indeed zero, so that (F.18) does give the proper total field in the region z > 0.
In similar fashion, we can show that the image currents for horizontal electric currents Jt are:
Ji = −Jt (x, y, −z) (F.19)
while for vertical magnetic currents Mz we have
Mi = −uz Mz (x, y, −z) (F.20)

432
and for horizontal magnetic currents Mt ,

Mi = Mt (x, y, −z) (F.21)

These results are summarized schematically in Fig. F.1.

F.6 Notes and References

For the fields of elementary dipoles, see, e. g.,
Balanis (1982), chapters 3 and 4.

433
434
Appendix G

ELECTROMAGNETIC
MATERIAL PROPERTIES

The quantity
σ
ǫ̂ = ǫ − j = ǫ0 ǫr (1 − j tan δ)
ω

is called the complex permittivity; we will often simply write ǫ, with the understanding that it can
be made complex to account for dielectric or conduction losses if need be. The loss tangent tan δ is
a measure of the degree of losses in the material. Magnetic losses could similarly be accounted for by
letting µ be complex, but these are less often encountered in waveguide applications. We have restricted
the material parameters to be scalar quantities; this simplifies much of the analysis done in the book.
Nevertheless, anisotropic and gyrotropic materials, for which ǫ and/or µ are dyadics or tensors, are of
considerable interest in waveguide technology. Such materials include plasmas, ferrites, electro-optic
materials and more, and make possible the implementation of isolators, circulators and other important
devices. Detailed treatment of anisotropic materials can be found in the references given at the end of
this appendix.
Some materials are chiral or bianisotropic. This means that D can depend not only on E but also
on B, and likewise H depends not only on B but also on E. Typically, chiral materials are made up of
complex molecules that have a spiral shape or other structural feature that gives it a “right-handed” or
“left-handed” sense.
Many materials have electromagnetic properties that vary significantly with frequency. Such mate-
rials are said to be dispersive. This often occurs because of atomic or molecular resonances at one or
more frequencies in the vicinity of the operating frequency. Very close to such resonances, it is common
for a material to have increased losses, so that the loss tangent becomes larger. Losses may also be
increased by the absorption of water in a humid environment. Some materials should be avoided for
certain applications because of this effect.
If a material with some desired set of electromagnetic constitutive parameters cannot be found in
nature, or is unsuitable because of its other physical properties (toxicity, mechanical strength, etc.), it
may be possible to create engineered materials that have the desired behavior. In the early literature,
these were called artificial dielectrics, but more recently the idea has been extended to a wider class
of materials called metamaterials. These have made possible values of ǫ and µ that are simultaneously
negative in some frequency ranges, leading to a negative index of refraction. This has become a very
active area of research.

435
G.1 Permittivity and Conductivity
A brief collection of the permittivities and conductivities of common materials is given in Table G.1.
This table is a work in progress.

Material ǫr σ (S/m) tan δ

Aluminum Oxide (Sapphire; Al2 O3 ) 9.1 7 × 10−4

at f = 1 MHz

Copper 1 5.7 × 107

Fused Quartz at f = 3 GHz 3.8 10−4

Ice (−12◦ C) at f = 3 GHz 3.2 9 × 10−4

Paraffin Wax at f = 3 GHz 2.25 2 × 10−4

Plexiglas at f = 3 GHz 2.6 5.7 × 10−3

Polyethylene at f = 1 MHz 2.26 4 × 10−4

Ruby Mica at f = 3 GHz 5.4 3 × 10−4

Teflon at f = 100 MHz 2.1 5 × 10−4

Titanium Dioxide (TiO2 ) at f = 1 85 5 × 10−4

MHz

Water (distilled) at f = 1 MHz 78.2 2 × 10−4

Water (sea) at f = 1 MHz 77.7 2 × 104

Table G.1: Electromagnetic properties of common materials.

G.2 Notes and References

Good collections of data on the permittivity, permeability and conductivity of various materials are
given in

Moreno (1958), pp. 201-209;

ITT Staff (1975), chapter 4.

and can also be found on the internet.

Properties and applications of anisotropic and gyrotropic materials are discussed in:

Johnson (1965), chapters 11 and 12;

Kurokawa (1969), sections 5.6-5.8;
I. Lebedev (1973), sections 8.5 and 8.11;
Helszajn (1975);

436
 Atwater (1981), chapter 8;
Sazonov et al. (1982), chapter 9;
Baden Fuller (1987);
Linkhart (1989).
For information on chiral materials, see:
Lakhtakia et al. (1989).
Treatments of artificial dielectrics and metamaterials can be found in:
J. Brown, “Artificial dielectrics,” in Progress in Dielectrics, vol. 2, pp. 195- 225 (1960);
Eleftheriades and Balmain (2005);
Engheta and Ziolkowski (2006);

437
438
Appendix H

BESSEL LINES AND TURNING

POINTS

In this appendix, we will consider the exact solutions for waves on nonuniform transmission lines of a
special form—those whose line parameters vary as powers of the distance along the line. These solutions
will allow us to examine the behavior of waves that encounter turning points: locations where either x or
b becomes zero. We assume throughout this appendix that the line is lossless, i. e., z = jx and y = jb.

H.1 The Bessel Line

Let the line parameters vary with position as
x(z) = X0 z p (p > −1); b(z) = B0 z q (p > −1) (H.1)
for some real constants X0 , B0 , p and q. These parameters exhibit a turning-point behavior at z = 0 if
p or q is positive; we will consider z = 0 to be a generalized turning point as long as p > −1 and q > −1.
Equations (1.4) then yield the following equation for the voltage V (z):
d2 V p dV
− + X0 B0 z p+q V = 0 (H.2)
dz 2 z dz
Introducing the new variables
V = zrU ; w = zs (H.3)
where
p+1 p+q+2
r= ; s= (H.4)
2 2
transforms (H.2) to
d2 U ν2
 
1 dU 2
+ + K0 − 2 U = 0 (H.5)
dw2 w dw w
where we have defined √
p+1 r X 0 B0
ν= = ; K0 = (H.6)
p+q+2 s s
Note that 0 < ν < 1 under the conditions on p and q.
Now, (H.5) is none other than Bessel’s differential equation (B.1), so its solution can be written in
(1) (2) (1) (2)
the form U = AHν (K0 w) + BHν (K0 w), where A and B are constants, and Hν and Hν are Hankel
functions of the first and second kind, respectively. Thus,
h i
V = z r AHν(1) (K0 z s ) + BHν(2) (K0 z s ) (H.7)

439
The current can then be recovered from the second of (1.4), using recurrence relations of the type
(B.34)-(B.35):
K0 s−r h (1) (2)
i
I = js z AHν−1 (K0 z s ) + BHν−1 (K0 z s ) (H.8)
X0
Using the identities (B.22), we can express the current in terms of Hankel functions of positive order:

K0 s−r h −jνπ (1) (2)

i
I = −js z Ae H1−ν (K0 z s ) + Bejνπ H1−ν (K0 z s ) (H.9)
X0
The constants A and B can be expressed in terms of the voltage and current at the turning point by
using (B.7) and  ν
(1) (2) Γ(ν) 2
Hν (x) ∼ −Hν (x) ∼ −j (x → 0) (H.10)
π x
which are consequences of (B.8) (recall that Γ here denotes the gamma function). We obtain
 ν
jΓ(ν) 2
V (0) = (B − A) (H.11)
π K0

and  ν  ν
Γ(1 − ν) 2s K0 2
I(0) = (Bejνπ − Ae−jνπ ) (H.12)
π X0 2 K0
Solving (H.11)-(H.12), we have
  ν  ν 
π X0 2 I(0) jνπ K0 V (0)
A= + je (H.13)
2j sin νπ 2s K0 Γ(1 − ν) 2 Γ(ν)
  ν  ν 
π X0 2 I(0) −jνπ K0 V (0)
B= + je (H.14)
2j sin νπ 2s K0 Γ(1 − ν) 2 Γ(ν)
With these results in hand, we can substitute (H.13)-(H.14) into (H.7) and (H.9) to get V and I at
an arbitrary position z in terms of the voltage and current at the turning point. In other words, we get
a chain parameter description for the Bessel line of the type (1.56):

V (z) = A(z, 0)V (0) + B(z, 0)I(0)

I(z) = C(z, 0)V (0) + D(z, 0)I(0) (H.15)

We obtain after some algebra

 ν
Q2 (z)
A(z, 0) = Γ(1 − ν) J−ν [Q2 (z)]
2
 1−ν
Q2 (z)
B(z, 0) = −jΓ(ν)Zc (d) Jν [Q2 (z)]
2
 ν
1 Q2 (z)
C(z, 0) = −jΓ(1 − ν) J1−ν [Q2 (z)]
Zc (d) 2
 1−ν
Q2 (z)
D(z, 0) = Γ(ν) Jν−1 [Q2 (z)] (H.16)
2

where Q2 (z) is the phase integral (1.97) for this line:

Z z
Q2 (z) = β(z ′ ) dz ′ = K0 z s (H.17)
0

440
and r
X0 (p−q)/2
Zc (z) = z (H.18)
B0
We have made extensive use of Bessel function identities from Appendix B to simplify these expressions.
If the line is cutoff β → −jα, we should make the substitution Q2 → −jQ1 in these expressions, where
Q1 is given by (1.97).

H.2 Notes and References

This treatment of Bessel lines is adapted from
Ghose (1963), section 12.3.

441
442
Appendix I

MODES IN HOLLOW
WAVEGUIDES WITH ZERO
CUTOFF FREQUENCY

As mentioned in Chapter 5, the equations


∂Hz 
(∇2T + kc2 )Hz = 0; =0
∂n C0

describing TE modes in a hollow metallic waveguide filled with a homogeneous medium admit the special
solution:
Hz = K0 = constant; kc = 0 (I.1)
Because kc vanishes, however, it is not possible to use (5.38) and (5.39) to recover the other field
components, because they have the indeterminate form 0/0. We must use (5.15)-(5.18) to study modes
which have zero cutoff frequency.
Inserting Hz = K0 and Ez = 0 into (5.17)-(5.20), we arrive at

uz × ET = ζHT (I.2)

uz · ∇T × HT = 0 (I.3)
uz · ∇T × ET = −jωµK0 (I.4)
where ζ = (µ/ǫ)1/2 is the wave impedance of the medium filling the guide. We have used the fact that
√
γ = jω µǫ (since kc = 0), assuming the forward-propagating mode. The boundary conditions imply
that
Eℓ |C0 = 0 and Hn |C0 = 0 (I.5)
(cf. Fig. 5.5).
Now, equations (I.2) and (I.4) imply that
√
∇T · HT = jω µǫK0 (I.6)

If we apply the two-dimensional divergence theorem (cf. eqn. (D.16) in Appendix A),
I Z
− AT · un dℓ = ∇T · AT dS (I.7)
C S

443
 Figure I.1: Wall currents and magnetic fields for the “TE0 ” mode.

(where S is a surface in the xy-plane bounded by a curve C whose inward unit normal vector is un ) to
the vector
AT = Hz HT
and take C to be the conducting waveguide wall C0 , then
I
0 = − Hz Hn dℓ
Z C0
= ∇T · (Hz HT ) dS
S
Z 0
= [∇T Hz · HT + Hz ∇T · HT ] dS (I.8)
S0
√
= jω µǫK02 S0 (I.9)

where we have used (I.5) and (I.6), and used the symbol S0 also to denote the area of the waveguide
cross-section.
Equation (I.9) implies that ω = 0 or K0 = 0, and so (I.4) becomes

uz · ∇T × ET = 0 (I.10)

and the equations (I.2), (I.3) and (I.10) which describe the transverse fields are completely decoupled
from Hz , and are, in fact, the equations for TEM mode fields in the waveguide. We study such modes
in detail in Chapter 8, and show in particular that no TEM mode is possible in a hollow waveguide with
a simply-connected boundary wall.
The nontrivial “TE0 ” mode (one with Hz = constant 6= 0) which can exist therefore does so only at
D.C. (ω = 0), and consists of a constant, z-directed H-field maintained by transverse currents circulating
in the waveguide wall (Figure I.1). Since this DC effect is usually not important at the operating
frequency of a metallic waveguide, the contribution of this mode to the total field of the waveguide is
generally ignored.

444
Appendix J

FIELD LINE PLOTTING

Field-line plots of the transverse fields of waveguide modes are conventionally drawn with solid lines for
the electric field and dashed lines for the magnetic field as shown in Figure J.1. The lines are drawn so
as to be tangent to the appropriate field vector at each point, with the amplitude of the vector being
proportional to the density of the lines in the vicinity of each point. An E-field line in a two-dimensional

E field amplitude smaller

E field amplitude larger

Figure J.1: Typical field plot: E lines; – – – – H lines.

plot (x, y or ρ, φ) should thus be everywhere tangent to ET ; if the field line is described by the relation
y = y(x), we require that
dℓy dy Ey
= = (J.1)
dℓx dx Ex
be satisfied, whereas in cylindrical coordinates, a field line ρ = ρ(φ) should satisfy
dℓρ dρ Eρ
= = (J.2)
dℓφ ρdφ Eφ
Inserting the specific expressions for the components of ET into (J.1) or (J.2) allows us to solve for y(x)
or ρ(φ) and thus plot the field line (increasing the density of lines as the amplitude of ET warrants).

445
Clearly the ratio of the components of ET must be real for (J.1) and (J.2) to make sense. Equations for
the field lines of HT can be treated in a similar manner.
In cases where ET has been obtained by separation of variables, (J.1) and (J.2) can be solved more
explicitly. For example, if Eρ and Eφ are the fields of a TElm mode of a hollow metallic circular waveguide
[equations (5.82) and (5.83) of chapter 5], then we put dℓρ = dρ and dℓφ = ρ dφ, and (J.2) gives
 ′ 
j ρ
Jl′ lma j′
 ′  lm dρ = l tan(lφ)dφ (J.3)
j ρ
Jl lm a
a

or, integrating,
′
 
jlm ρ
Jl cos lφ = K = constant (J.4)
a
By choosing different values of the constant K, we can generate a whole family of curves corresponding
to the mode fields. The plots of Figure 5.9 are special cases of the result (J.4).
A more general example is furnished by the two-dimensional magnetostatic field given in terms of
the z-component of a vector potential:
∂Az ∂Az
BT = ∇T × (uz Az ) = −uz × ∇T × Az = ux − uy (J.5)
∂y ∂x
Then the field line equations for HT are given in rectangular coordinates as

dy By ∂Az /∂x
= =− (J.6)
dx Bx ∂Az /∂y
or
∂Az ∂Az
dx + dy = ∇Az · dℓ = 0 (J.7)
∂x ∂y
where dℓ = ux dx + uy dy. Hence, the HT field lines must follow a path along which ∇Az = 0, or
Az = constant. The magnetostatic field of a line current shown in Fig. J.2 illustrates this result.

Figure J.2: Magnetostatic field of a line current.

446
Appendix K

INTEGRAL IDENTITIES FOR

FIELDS OF GUIDED MODES

It is possible, using the identities (9.5) and (9.6), or the Lorentz reciprocity theorem which follows from
them, to obtain a number of useful relationships satisfied by the fields and propagation coefficient of
a guided mode on an arbitrary waveguide structure. These can be used to gain useful physical insight
into the properties of the mode, and even give approximate expressions for the propagation coefficient
if suitable approximate knowledge of the fields is available. We assume throughout that ω 6= 0, but no
assumption has been made about the normalization of the mode fields, so that (5.27) is not assumed to
hold.

K.1 Identities for General, Lossy Waveguides

In (9.5), let us take

Ea = Em e−γm z
Ha = Hm e−γm z (K.1)

and

Eb = E−m eγm z
Hb = H−m eγm z (K.2)

where (Em , Hm ) and γm = αm + jβm are the fields and propagation coefficient of a waveguide mode,
while (E−m , H−m ) and −γm are those of the mode corresponding to it which propagates in the opposite
direction (see (5.25)). Let us integrate this expression over the volume V of the slice of waveguide shown
in Figure 9.4. The result is
Z
0 = −jω (µHm · H−m + ǫEm · E−m ) dV (K.3)
V

But since Z Z z1 Z
(−) dV = dz dS(−)
V 0 S0

and the integrand of (K.3) is independent of z, we conclude from (K.3) that

Z Z
µHm · H−m dS = − ǫEm · E−m dS (K.4)
S0 S0

447
since z1 is arbitrary.
If, instead of (K.2), we had chosen the “b” fields to be the same as the “a” fields, we would have,
after integrating (9.5),
2
+ µH2m ) dS
R
jω S0 (ǫEm jωUm,osc
γm = R = (K.5)
2 S0 (Em × Hm ) · uz dS Pm,osc
2
where by Em , we mean Em · Em ,
1
Z
Pm,osc = (Em × Hm ) · uz dS (K.6)
2 S0

is the complex oscillatory power carried by the mode, and

1
Z
Um,osc = (ǫE 2 + µH2m ) dS (K.7)
4 S0 m

is the complex oscillatory energy per unit length stored by the mode. The latter, by virtue of (K.4), can
also be expressed in the two alternative forms
1 1
Z Z
2
Um,osc = (ǫEmT · EmT + µHmz ) dS = (ǫE 2 + µHmT · HmT ) dS (K.8)
2 S0 2 S0 mz

Consider next the quantity Da × Bb . We can write

1 1
uz · Da × Bb = − uz · [Da × (∇ × Eb )] − uz · (Da × Mb )
jω jω
1 a 1
= D · [uz × (∇ × Eb )] − uz · (Da × Mb ) (K.9)
jω jω
1 ∂Eb 1
= [Da · ∇Ezb − Da · ]− uz · (Da × Mb )
jω ∂z jω
where the last equality follows from the vector identity (D.19) in Appendix A. Then we have

∂Eb
 
1
uz · Da × Bb = Da · ∇Ezb + Ezb ∇ · Da − Da ·
jω ∂z
a
1 ∇ · J
− uz · (Da × Mb ) − (K.10)
jω ω2
b
 
1 a b a ∂E
= ∇ · (D Ez ) − D ·
jω ∂z
1 ∇ · Ja b
− uz · (Da × Mb ) − Ez
jω ω2
since
−jω∇ · Da = ∇ · Ja
and where the identity (D.6) has been used in the last step. Putting the “a” and “b” fields equal to
those in (K.1) and (K.2), respectively, and taking into account that Ja and Mb are zero, we have
1
µǫuz · (Em × H−m ) = {∇ · [ǫEm E−mz ] − γm ǫEm · E−m } (K.11)
jω
Integrating over the same volume V and carrying out the z integrals as before leaves us with
γm
Z Z
µǫ (Em × Hm ) · uz dS = ǫEm · E−m dS (K.12)
S0 jω S0

448
Because of (K.4), there are several alternative ways to express (K.12):
1
R
1 jω S0 ǫEm · E−m dS
= R
γm S0 µǫ(Em × Hm ) · uz dS
1
R
− jω S0 µHm · H−m dS
= R (K.13)
S0
µǫ(Em × Hm ) · uz dS
1
R
jω S0 (ǫEm · E−m − µHm · H−m ) dS
= R
2 S0 µǫ(Em × Hm ) · uz dS

Another class of relationships can be obtained by taking derivatives with respect to ω. Let E, H be
any solution of source-free Maxwell’s equations. Then their derivatives with respect to ω satisfy

∂E ∂H ∂(ωµ)
∇× = −jωµ −j H
∂ω ∂ω ∂ω
∂H ∂E ∂(ωǫ)
∇× = jωǫ +j E (K.14)
∂ω ∂ω ∂ω
where we have allowed for the possibility that µ and ǫ may be dispersive (i.e., depend on frequency).
Therefore, we may consider the fields

Eb = ∂E/∂ω and Hb = ∂H/∂ω (K.15)

to be legitimate field solutions provided we regard the quantities

∂(ωǫ) ∂(ωµ)
Jb = j E and Mb = j H (K.16)
∂ω ∂ω
as the “sources” of Eb and Hb .
Letting

E = E−m eγm z
H = H−m eγm z (K.17)

we have from (K.15) and (K.16) that

 
b ∂E−m ∂γm
E = +z E−m eγm z
∂ω ∂ω
 
∂H−m ∂γm
Hb = +z H−m eγm z (K.18)
∂ω ∂ω

and
∂(ωǫ)
Jb = j E−m eγm z
∂ω
∂(ωµ)
Mb = j H−m eγm z (K.19)
∂ω
Using the reciprocity theorem (9.7) and integrating over V as before, we get:
nR o
j S0 ∂(ωǫ)
R ∂(ωµ)
∂γm ∂ω Em · E−m dS − S0 ∂µ Hm · H−m dS
= R (K.20)
∂ω 2 S0 Em × Hm · uz dS

449
Finally, the use of

E = Em e−γm z
H = Hm e−γm z (K.21)

in (K.15) and (K.16), along with (K.1), followed by an application of Lorentz’ theorem (9.7) and an
integration over V yields
R h ∂(ωǫ) 2 ∂(ωµ) 2
i
S0 ∂ω Em − ∂ω Hm dS
γm = j R  (K.22)
2 S0 Em × ∂H ∂Em

∂ω − ∂ω × Hm · uz dS
m

which is interesting to compare with (K.5).

K.2 Identities for Modes in Lossless Waveguides

Of special interest is the case where ǫ and µ are real, making the waveguide a lossless one. Properties
of the modes of lossless waveguides analogous to those of the foregoing section can be derived in a very
similar fashion, and often have direct physical interpretations, which we will give below.
The derivation of these identities proceeds as before using (K.1) as a-state fields, but with a different
choice for the b-state fields. An examination of the time-harmonic source-free Maxwell equations in a
lossless medium (ǫ and µ are real) shows that, if E, H is a solution, so is E∗ , −H∗ , where ∗ denotes a
complex conjugate. Thus, one possible choice for the b-state field would be
∗
Eb ∗ −γm z
= Em e
∗
Hb = −H∗m e−γm z (K.23)

Following the derivation that led to (K.4) leads to the result

αm P̂ = jω(UM0 − UE0 ) (K.24)

where
1
Z
∗
UE0 = ǫEm · Em dS (K.25)
4 S0
1
Z
UM0 = µHm · H∗m dS (K.26)
4 S0
are respectively the nondispersive parts of the time-average electric and magnetic stored energy per unit
length in the waveguide, and
1
Z
P̂ = Em × H∗m · uz dS (K.27)
2 S0
is the complex power carried through the cross section S0 of the waveguide (assuming it valid to make
this interpretation even though S0 is not a closed surface).
On the other hand, if we take as the b-state field
∗
Eb ∗
= E−m e γm z
∗
Hb = −H∗−m eγm z (K.28)

then we are led to the result

βm P̂ = ωT (K.29)
where
1
Z h    i
∗ 2 2
T = ǫ EmT · EmT − |Emz | + µ HmT · H∗mT − |Hmz | dS (K.30)
4 S0

450
is a quantity similar to but not quite identical with the stored energy. Since T , UE0 and UM0 are all
real, (K.24) and (K.29) together imply that if P̂ is not equal to zero, then either αm = 0 (the mode
propagates, in which case P̂ is real) or βm = 0 (the mode is cutoff, in which case P̂ is imaginary). Only
if P̂ = 0 can both αm and βm be different from zero for a lossless waveguide, in which case the mode is
called a complex mode.
For many purposes, the most important modes of lossless waveguides are the propagating ones. For
these we set P̂ ≡ P , which is real, and by (K.24) we will have
1
UE0 = UM0 = U0 (K.31)
2
where U0 = UE0 + UM0 is the part of the total stored energy per unit length of the waveguide not
associated with material dispersion (i. e., the dependence of µ and ǫ on frequency). Other identities
for the propagating modes of lossless waveguides can be found using (K.28) as the b-state fields. For
example, a derivation parallel to that leading to (K.13) gives

ω U0
vp = = (K.32)
β G0
where vp is the phase velocity of the mode, while

1
Z
∗
G0 = Dm × Bm · uz dS (K.33)
2 S0

is the nondispersive part of the time average pseudomomentum of the fields of the mode per unit length
of waveguide: Dm = ǫEm and Bm = µHm .1
Analogous to (K.20), we find
1 P
vg = = (K.34)
dβ/dω U
where vg is the group velocity of the mode, and U is the total stored energy per unit length of the
waveguide:
Z  
1 ∂(ωǫ) ∗ ∂(ωµ) ∗
U = Em · Em + Hm · Hm dS
4 S0 ∂ω ∂ω
 
ω ∂ǫ ∂µ
Z
∗ ∗
= U0 + Em · Em + Hm · Hm dS (K.35)
4 S0 ∂ω ∂ω

From (K.29), we have

P
vp = (K.36)
T
We note from (K.32), (K.34) and (K.36) that T satisfies the relation

U0 P vg U
= = (K.37)
G0 T T
or,
T U0 T
vg = = (K.38)
G0 U G

1 The momentum density associated with a field in free space is d × b; the corresponding time-average momentum

density is 21 Re(D × B∗ ). In material media, there is some cloudiness in the use of this expression, because some of it is
associated with mechanical momentum of the atoms and molecules of the material itself, and only a portion can truly be
claimed to be “electromagnetic” momentum. We thus use the term pseudomomentum for the integral of 21 D × B∗ in an
arbitrary medium.

451
where G is the total time average pseudomomentum carried by the mode:
 
U
G = G0
U0
Z  
G0 ω ∂ǫ ∗ ∂µ
= G0 + Em · Em + Hm · H∗m dS (K.39)
U0 4 S0 ∂ω ∂ω
Z  
β ∂ǫ ∗ ∂µ
= G0 + Em · Em + Hm · H∗m dS
4 S0 ∂ω ∂ω
as follows from (K.32) and (K.35). Hence we interpret T as the time average of pseudomomentum flow
in the waveguide mode passing the cross section S0 , again assuming this interpretation to be valid for
an unclosed surface like S0 .

K.3 The Partial Power Law

When ǫ and µ are nondispersive (∂ǫ/∂ω = 0 and ∂µ/∂ω = 0), equations (K.32) and (K.34) can be
combined to give the so-called partial power law of N. A. Semenov:
µǫEm × H∗m · uz dS
R
1 G0
= = SR0 ∗
(K.40)
vp vg P S0 Em × Hm · uz dS

This states that the inverse product of vp and vg is equal to an average of µǫ over the guide cross-
section, weighted by the z-component of the time-average Poynting vector. For a homogeneously-filled
guide where ǫ and µ are independent of position, (K.40) simplifies to
1
vp vg = (K.41)
µǫ
which is easily verified for the hollow guides of Chapter 5 and the TEM transmission lines of Chapter 8.
For more general modes on lossy waveguides, the relation corresponding to (K.40) follows from (K.20)
and (K.13): R
S0 µǫEm × Hm · uz dS
2
γm dγm d(γm )
= = − R (K.42)
ω dω d(ω 2 ) S0 Em × Hm · uz dS

K.4 Expressions for Filling Factors

We can obtain some interesting insights into the dielectric filling factors considered in section 9.4 based
upon a generalization of the identity (K.20). Suppose now that the frequency ω(s), the permittivity
ǫ(x, y; s) and the permeability µ(x, y; s) are all functions of some parameter s. For example, s might be
the relative permittivity ǫr of the substrate of a planar transmission line, whose variation causes ǫ(x, y; s)
to vary only in a portion of space:

ǫ(x, y; ǫr ) = ǫ0 in S2 (y > h)
= ǫ0 ǫr in S1 (0 < y < h) (K.43)

We can repeat the derivation leading from (K.14) to (K.20), only now we consider the fields of the “b”
state to be
Eb = ∂E/∂s and Hb = ∂H/∂s (K.44)
Our result is h i
R ∂(ωǫ) ∂(ωµ)
∂γm S0 ∂s Em · E−m − ∂s Hm · H−m dS
=j R (K.45)
∂s 2 S0
Em × Hm · uz dS

452
As in section K.2, we can also obtain a version of (K.45) suitable for propagating modes on lossless
waveguides:
R h ∂(ωǫ) ∗ ∂(ωµ) ∗
i
∂βm S0 ∂s Em · Em + ∂s Hm · Hm dS
= R (K.46)
∂s 2 S0 Em × H∗m · uz dS
Now consider the filling factor of a microstrip or other quasi-TEM line as defined by (9.44). If we
assume ǫr to be independent of ω, but to affect ǫ(x, y) as in (K.43), then

∂(ωǫ)
= ωǫ0 in S1
∂ǫr
= 0 in S2 (K.47)

while ∂(ωµ)/∂ǫr is identically zero. Hence (K.46) becomes

∗
R
∂β ωǫ0 S1 Em · Em dS
= R ∗
(K.48)
∂ǫr 2 S0 Em × Hm · uz dS

But
∂ǫeff ∂(β 2 /k02 ) 2β ∂β
q= = = 2 (K.49)
∂ǫr ∂ǫr k0 ∂ǫr
This in conjunction with (K.34) and (K.31) yields
∗
R
c2 /ǫr S1 ǫEm · Em dS
q= R
∗ dS
(K.50)
vp vg S0 ǫEm · Em

We see that q is proportional to the fraction of stored energy carried in the substrate region of the
microstrip, since S1 represents the cross-section of that substrate. If further, the mode can be described
quasistatically, where p
β ≃ ω ls cs
and ls , cs are independent of ω, then
c c √
≃ ≃ ǫeff (K.51)
vp vg
and hence
∗
R
∂ǫeff ǫeff S1 ǫEm · Em dS
q= ≃ R
∗ dS
(K.52)
∂ǫr ǫr S0 ǫEm · Em
Equation (K.50) can clearly also be adapted to apply to the filling factors of a step-index dielectric
waveguide as defined in section 9.4. Let S1 and S2 represent the cross-sections of the core and cladding
as shown in Figure E.1. The refractive indices of these regions are n1 and n2 , respectively, and the
boundary contour between them is denoted C. From (K.48) and (K.49) generalized to the case of the
filling factors q1 and q2 defined by (9.40), we have
∗
R
cneff S1 ǫEm · Em dS
q1 = 2 R ∗
(K.53)
n1 S0 E × Hm · uz dS
∗
R
cneff S2 ǫEm · Em dS
q2 = 2
R (K.54)
n2 S0 Em × H∗m · uz dS
From (K.53) and (K.54), together with (K.31), (K.34) and the definition (9.55) for the group index ng ,
we can show quite generally that
n21 q1 + n22 q2 = neff ng (K.55)

453
 xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
C
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx
xxxxxxxxxxxxxxx

an

S1 (n1)

S2 (n2)

Figure K.1: Core and cladding geometry of a step-index dielectric waveguide of arbitrary geometry.

under no approximation.
An alternate representation for q1 and q2 is also possible, proceeding from the identity (K.11). If
(K.28) is used in (K.11), we have for propagating mode fields on lossless guides:

βm j
µǫuz · (Em × H∗m ) = ∗
Em · Em ∗
+ ∇ · [Dm Emz ] (K.56)
ωǫ ω

If this is integrated, first over S1 and then over S2 , and we apply (K.53), (K.54) and the divergence
theorem, then we obtain that

Em × H∗m · uz dS
R
E ∗ D · u dℓ
H
jc2
q1 = RS1
− 2
R C mz m n (K.57)
∗ ωn1 S0 Em × H∗m · uz dS
S0 Em × Hm · uz dS

Em × H∗m · uz dS
R
∗
H
S2 jc2 C Emz Dm · un dℓ
q2 = R + (K.58)
ωn22 S0 Em × H∗m · uz dS
R
E × H∗m · uz dS
S0 m

When the mode exists under weak-guidance conditions ((n21 − n22 )/n22 ≪ 1), Emz∗
can be set equal to
zero, and we have (a) an independent proof of (9.54), and (b) that q1 and q2 in this case represent the
percentages of power in the mode being carried in the core and cladding respectively.

K.5 Bounds on Phase and Group Velocities

We have seen examples (as for instance in chapter 5) where β/ω → 0 and dβ/dω → ∞ for a propagating
mode on a lossless guide; the situation occurs at cutoff and implies that vp → ∞ and vg → 0. There is
thus no general upper bound on vp nor lower bound on vg . However, some partial bounds are obtainable
from the results of this Appendix, as we will show below.
Suppose that µ and ǫ are positive, consider a forward mode with γ = jβ. By the definition of a

454
forward mode, P > 0.2 Now for any fields E and H, the inequality
 
1 1 1
|E × H∗ · uz | ≤ √ ǫE · E ∗ + µH · H∗ (K.59)
µǫ 2 2
is true. We prove this by first noting that

|E × H∗ · uz | ≤ |E||H| (K.60)

from the definition of the cross product. But also

"  #2
1/4
ǫ  µ 1/4
|E| − |H| ≥ 0 (K.61)
µ ǫ
or,  
1 1 1
|E||H| ≤ √ ǫE · E ∗ + µH · H∗ (K.62)
µǫ 2 2
Combining (K.60) with (K.62) yields (K.59). Now (K.59) can be used to show that
 
∗ √ 1 ∗ 1 ∗
|µǫE × H · uz | ≤ µǫ ǫE · E + µH · H
2 2
 
p 1 ∗ 1 ∗
≤ (µǫ)max ǫE · E + µH · H (K.63)
2 2
where (µǫ)max is the maximum value of the product µǫ that occurs over the cross-section of a waveguide.
Also,
1 1 √
ǫE · E ∗ + µH · H∗ ≥ µǫ|E × H∗ · uz |
2 2 p
≥ (µǫ)min |E × H∗ · uz | (K.64)

From (K.63) and (K.32), we have then

1
|vp | ≥ p ≡ vp,min (K.65)
(µǫ)max
or p
c (µǫ)max
|neff | = ≤ √ ≡ nmax (K.66)
|vp | µ0 ǫ 0
From (K.64) and (K.34), if ǫ and µ can be reckoned independent of ω,
1
vg ≤ p ≡ vg,max ≥ vp,min (K.67)
(µǫ)min
or p
c (µǫ)min
ng = ≥ √ ≡ nmin (K.68)
vg µ0 ǫ 0
2 It is possible, however, for β to be negative, implying that v < 0 as well. From (K.32), we see that G is required
p 0
to be negative for this to occur, and likewise from (K.36), we see that T < 0. In such a case, (K.34) implies that vg > 0
since U > 0. Only if µǫ is not a constant throughout the cross-section can G0 < 0 at the same time that P > 0. A mode
such as this, with vp vg < 0 is called a backward-wave mode, in contrast to the more usual direct or normal-wave mode for
which vg vp > 0. Examples of backward-wave modes are well-known in the literature, often appearing in conjunction with
complex modes. Properties analogous to those we are about to derive can also be found for backward-wave modes.
Note the distinction in terminology with forward and reverse modes, which are used merely to indicate in which direction
(±z) energy flow occurs for a given mode of a lossless waveguide.

455
 Finally, note that from (K.30), (K.31), (K.25) and (K.26),

|T | ≤ U0 (K.69)

Then by (K.36) and (K.34), again for ǫ and µ independent of ω, we have

|vp | ≥ vg (K.70)

This also implies that ng ≥ neff , and that

ω dβ
≥1
β dω
and thus that  
dvp 1 ω dβ
= 1− ≤0 (K.71)
dω β β dω
i.e., the phase velocity is a nonincreasing function of ω at any frequency, while the effective index
neff = β/k0 is a nondecreasing function of ω. We have seen examples of this property many times
already in chapters 5-8.

K.6 Notes and References

The reader unfamiliar with the concept of momentum associated with an electromagnetic field should
consult
Javid and Brown (1963), pp. 129-133.
A complete discussion of the energy-power-momentum relations for waveguides is given in
H. A. Haus and H. Kogelnik, “Electromagnetic momentum and momentum flow in dielectric
waveguides,” J. Opt. Soc. Amer., vol. 66, pp. 320-327 (1976).
The various results are of differing ages, and can be traced in the following sources:
P. Chorney, “Power and energy relations in bidirectional waveguides,” in Proc. Symp. Elec-
tromagnetics and Fluid Dynamics of Gaseous Plasma. New York: Polytechnic Press,
1962, pp. 195-210.
N. A. Semenov, “The partial-power law,” Radio Eng. Electron. Phys., vol. 8, pp. 1419-1420
(1963).
N. A. Semenov, “Proof of the partial-power law,” ibid., vol. 10, pp. 1320-1321 (1965).
J. Brown, “Electromagnetic momentum associated with waveguide modes,” Proc. IEE (Lon-
don), vol. 113, pp. 27-34 (1966).
P. Chorney and P. Penfield, “Waveguide power-mode theorems for nonconservative systems,”
IEEE Trans. Micr. Theory Tech., vol. 19, pp. 767-772 (1971).
S. Kawakami, “Relation between dispersion and power-flow distribution in a dielectric wave-
guide,” J. Opt. Soc. Amer., vol. 65, pp. 41-45 (1975).
S. Nemoto and T. Makimoto, “A relationship between phase and group indices of guided
modes in dielectric waveguides,” Int. J. Electron., vol. 40, pp. 187-190 (1976).
E. F. Kuester, “Generalization of the partial-power law (Brown’s identity) to waveguides
with lossy media,” Electron. Lett., vol. 20, pp. 456-457 (1984).
Results similar to those of section K.4 relating loss (and filling factor) to group velocity and stored
energy fractions can be found in:
E. O. Schulz-DuBois, “Energy transport velocity of electromagnetic propagation in dispersive
media,” Proc. IEEE, vol. 57, pp. 1748-1757 (1969).

456
Some of the results of sections K.4 and K.5 can also be found in:
R. B. Adler, “Properties of Guided Waves on Inhomogeneous Cylindrical Structures,” Tech.
Rept. No. 102, Research Laboratory of Electronics, Massachusetts Institute of Technol-
ogy, 1949.
R. B. Adler, “Waves on inhomogeneous cylindrical structures,” Proc. IRE, vol. 40, pp.
339-348 (1952).
More results of this type (though in somewhat abstract form applicable to very general—not just
electromagnetic—waveguides) can be found in
Silbergleit and Kopilevich (1996), sects. 20 and 28.
Equations (K.45) and (K.46) are electromagnetic vector analogs of the Hellmann-Feynman theorem; see,
e.g.,
Arnaud (1976), pp. 216-218.
Equations (K.65)-(K.68) are given in
Mrozowski (1997), section 11.3.4.
For more about complex modes, see:
A. M. Belyantsev and A. V. Gaponov, “Waves with complex propagation coefficients in
coupled transmission lines without energy dissipation,” Radio Eng. Electron. Phys., vol.
9, pp. 980-988 (1964).
A. S. Omar and K. F. Schünemann, “Complex and backward-wave modes in inhomogeneously
and anisotropically filled waveguides,” IEEE Trans. Micr. Theory Tech., vol. 35, pp.
268-275 (1987).
A. Magos, “Singular frequencies in wave guides with inhomogeneous dielectric,” Int. J. Appl.
Electromagnetics Mater., vol. 4, pp. 189-196 (1994).
T. F. Jabloński, “Complex modes in open lossless dielectric waveguides,” J. Opt. Soc. Amer.
A, vol. 11, pp. 1272-1282 (1994)
and

Mrozowski (1997).
The terminology we have used (backward-wave versus direct or normal-wave modes) to indicate the
relative directions of phase and group velocity for a mode is not universal, as can be seen from a review
of the literature, but would seem to be most natural for our purposes. The reader should be aware of
this when consulting other published work.

457
458
Appendix L

DERIVATION OF SURFACE
IMPEDANCE FOR A GOOD
CONDUCTOR

In this Appendix, we derive the expression for the surface impedance ZS of a good conductor. Referring
to Figure 9.7, we set up a coordinate system (x1 , x2 , n) corresponding to the unit vectors u1 , u2 tangential
to the interface and un normal to the interface. From what we know about the behavior of plane waves
refracting into a dense or highly-conducting medium, we assume that, in the conducting medium

E = Ec (x1 , x2 , n)eψ(kc n) (L.1)

H = Hc (x1 , x2 , n)eψ(kc n) (L.2)

where
r
σc
kc = ω µc (ǫc − j )
ω
1−j
≃ (L.3)
δc
is the wavenumber in the conducting medium, and δc is the skin depth (9.58), under the condition
σc /ωǫc ≫ 1. The function ψ must be determined, but it represents the expectation that E and H vary
much more rapidly normal to the interface than in other directions, so that Ec and Hc are assumed to
be slowly varying within distances on the order of |kc |−1 .
Plugging (L.1) and (L.2) into Maxwell’s equations, we get

∇ × Ec + kc ψ ′ (kc n)un × Ec = −jωµ0 Hc

σc
∇ × Hc + kc ψ ′ (kc n)un × Hc = jω(ǫc − j )Ec (L.4)
ω
Because Ec and Hc are slowly-varying on the scale of |kc |−1 , we neglect the curl terms in (L.4) compared
to the second terms on the left sides. From the resulting equations, we have that

Ecn ≃ 0; Hcn ≃ 0 (L.5)

while the components tangential to the interface can be solved for to give

ψ ′ = ±j (L.6)

459
or
ψ(kc n) = ±jkc n + const (L.7)
We choose the + sign in (L.7) in order that the solutions (L.1) and (L.2) decay away from the interface
(as n becomes negative), so that we can obtain the relation

Ec(tan) ≃ ζ̂c un × Hc(tan) (L.8)

where ζ̂c is the wave impedance

µc
r
ζ̂c = (L.9)
ǫc − j σωc
of the conducting medium. Since (L.8) holds in particular at the interface (n = 0), and tangential field
components are continuous across the interface, (L.8) implies the boundary condition (9.59) with

ZS = ζ̂c

References
The analysis here is somewhat simplified from
S.M. Rytov, “Calcul du skin-effect par la mèthode des perturbations”, J. Phys. (USSR), vol.
2, pp. 233-242 (1940).
See also
T.B.A. Senior, “Impedance boundary conditions for imperfectly conducting surfaces”, Appl.
Sci. Res., vol. B8, pp. 418-436 (1960).
In both papers, more details of how accurate the impedance boundary conditions are, are provided. See
also
O.I, Panych, “Approximate boundary conditions in diffraction problems [Russian],” Dokl.
Akad. Nauk SSSR, vol. 70, pp. 589-592 (1950).
O.I. Panych, “Asymptotic expansion of the solution of a boundary problem [Russian],”
Matem. Sbornik, vol. 32(74), pp. 385-406 (1953).
K.M. Mitzner, “An integral equation approach to scattering from a body of finite conductiv-
ity,” Radio Science, vol. 2, pp. 1459-1470 (1967).

460
Appendix M

CHANGE IN INDUCTANCE DUE

TO DEFORMATION OF
BOUNDARY

In this Appendix, we will evaluate the change in the inductance l per unit length of a transmission line
operating under quasistatic conditions when its boundary undergoes a small deformation. Specifically,
consider the transmission line of cross-section S0 bounded by two perfect conductors C1 and Cg as shown
in Figure M.1. Let the new boundaries C1′ and Cg′ enclosing a new cross section S0′ result from displacing
C1 inward and Cg outward (that is, into the conductor in each case) by an amount δn normal to each
curve. For the original line, there will be magnetic fields H and B = ∇ × A = ∇ × (uz Az ) satisfying
the static field equation
∇×H=0 (M.1)

inside S, and giving rise to an inductance l per unit length which can be defined by (see (8.40)):

1 2
lI = UM (M.2)
2 1

where I1 is the total current on C1 ,

I
I1 = H · dℓ (M.3)
C1

and UM is the static magnetic energy per unit length of the line,

1 1
Z Z
UM = H · B dS = H · ∇ × A dS
2 S0 2 S0
1
Z
= [∇ · (A × H) + A · ∇ × H] dS (M.4)
2 S0
1 1
I I
= − (A × H) · un dℓ − (A × H) · un dℓ
2 C1 2 Cg

Since un · B = 0 on C1 and Cg , we have that


∂Az 
0 = un · ∇ × A|C1,g = un · ∇ × (uz Az )|C1,g = (M.5)
∂ℓ C1,g

461
 Cg
Cg'

S0 C1' C1
ul
δn un

ul un
y
δn
z x

Figure M.1: Cross-section of TEM or quasi-TEM transmission line; original boundaries are C1 and Cg ;
deformed boundaries are C1′ and Cg′ .

or Az is a constant (Az1 , say) on C1 and also constant (Azg , say) on Cg . Hence,

1 1
I I
UM = Az1 H · dℓ + Azg H · dℓ
2 C1 2 Cg
1
= (Az1 − Azg )I1 (M.6)
2
The deformed line will have slightly different fields H′ and B′ = ∇ × (uz A′z ) where

∇ × H′ = 0 (M.7)

in S0′ , and will be characterized by a slightly different inductance l′ such that

1 ′ ′ 2 ′
l (I1 ) = UM (M.8)
2
where I
I1′ = H′ · dℓ (M.9)
C1′

and
′1 ′
UM (A − A′zg )I1′
= (M.10)
2 z1
To obtain a formula for the change in l, let us consider the quantity

∇ · (A′ × H) = B′ · H (M.11)

Integrating this identity over the cross section S0 and using the two-dimensional divergence theorem, we
get
Z I I
′ ′
(B · H) dS = − (A × H) · un dℓ − (A′ × H) · un dℓ
S0 C1 Cg
I I
= (A′z1 − A′zg )I1 + (A′z − A′z1 )H · dℓ + (A′z − A′zg )H · dℓ (M.12)
C1 Cg

462
Now on C1 ,
∂Az
A′z − A′z1 ≃ δn = −δnBℓ
∂n
and similarly on Cg . Since the first term on the right side of (M.12) is equal to l′ I1 I1′ , we can write
Z I I
(B′ · H) dS ≃ l′ I1 I1′ − δnB · Hdℓ − δnB · Hdℓ (M.13)
S0 C1 Cg

On the other hand, let us integrate the quantity

∇ · (A × H′ ) = B · H′ (M.14)

over S0 and proceed as above. We have

Z I I
(B · H′ ) dS = − (A × H′ ) · un dℓ + (A × H′ ) · un dℓ
S0 C1 Cg
= (Az1 − Azg )I1′ (M.15)

From (M.15), (M.6) and (M.2), we have

Z
(B · H′ ) dS = lI1 I1′ (M.16)
S0

Now for an isotropic medium, B′ · H = B · H′ , so subtracting (M.16) from (M.13) and recognizing
that we can replace I1 I1′ by I12 to leading order, we get
H H
′ C1 δn(B · H) dℓ + Cg δn(B · H) dℓ
l −l≃ (M.17)
I12

or, using (M.2) and (M.5):

H H
δl l′ − l C1 δn(B · H) dℓ + Cg δn(B · H) dℓ
≡ ≃ R
S0 (B · H) dS
l l
δUM
= (M.18)
UM
That is, the relative change in l is proportional to the relative change in UM .

463
464
Appendix N

CORRECTION OF SMALL HOLE

THEORY FOR ENERGY
CONSERVATION

We’ll get to it.

References
The original treatments of corrections to small-hole aperture theory to enforce energy conservation are
found in
R. E. Collin, “Small aperture coupling between dissimilar regions,” Electromagnetics, vol. 2,
pp. 1-24 (1982).
D. K. Cheng and C.-H. Liang, “Generalised hybrid-network parameters for electromagnetic
coupling between dissimilar regions through a small aperture,” IEE Proc. pt. H, vol.
129, pp. 325-332 (1982).
J. R. Mautz and R. F. Harrington, “An admittance solution for electromagnetic coupling
through a small aperture,” Appl. Sci. Res., vol. 40, pp. 39-69 (1983).

465