Chew Microwave and Optical Fiber PDF

Lectures on
Theory of Microwave and Optical Waveguides
1
Weng Cho CHEW
Fall 2012
1
updated December 11, 2012
Contents
Preface vii
1 Preliminary Background 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 History of Electricity and Magnetism . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Maxwell-Heaviside Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.6 Reciprocity Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.6.1 Lorentz Reciprocity Theorem . . . . . . . . . . . . . . . . . . . . . . . 10
1.7 Energy Conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.7.1 Time Domain Poynting Theorem . . . . . . . . . . . . . . . . . . . . 10
1.7.2 Frequency Domain Poynting Theorem . . . . . . . . . . . . . . . . . . 11
1.7.3 Complex Power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.7.4 Lossless Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.8 Energy Density in Dispersive Medium . . . . . . . . . . . . . . . . . . . . . . 13
1.9 Symmetries in Electromagnetics . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.9.1 Time Reversal Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.9.2 Reflection Symmetry . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.9.3 Polar Vectors and Pseudovectors . . . . . . . . . . . . . . . . . . . . . 17
1.10 Green’s Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.11 Uniqueness Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.11.1 Scalar Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.11.2 Vector Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.12 Transformation Matrices for Microwave Circuits . . . . . . . . . . . . . . . . 25
1.12.1 Impedance and Admittance Matrices . . . . . . . . . . . . . . . . . . 25
1.12.2 Scattering Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
1.12.3 Chain Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2 Hollow Waveguides 35
2.1 General Uniform Cylindrical Waveguides . . . . . . . . . . . . . . . . . . . . 35
2.2 Transmission Line Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.2.1 TEM Mode of a Transmission Line . . . . . . . . . . . . . . . . . . . 38
i
ii Theory of Microwave and Optical Waveguides
2.2.2 Lossy Transmission Lines . . . . . . . . . . . . . . . . . . . . . . . . . 43

2.3 TE and TM Modes (H and E Modes) . . . . . . . . . . . . . . . . . . . . . . 45
2.3.1 Mode Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.4 Power Flow in a Waveguide . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.4.1 Power Flow and Group Velocity . . . . . . . . . . . . . . . . . . . . . 50
2.4.2 Pulse Propagation in a Waveguide . . . . . . . . . . . . . . . . . . . . 52
2.4.3 Attenuation in a Waveguide . . . . . . . . . . . . . . . . . . . . . . . 53
2.5 Wave Impedance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.6 Rectangular Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.6.1 TE Modes (H Modes) . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.6.2 TM Modes (E Modes) . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
2.7 Circular Waveguides . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.7.1 TE Modes (H Modes) . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.7.2 TM Modes (E Modes) . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.8 Excitation of Modes in a Waveguide . . . . . . . . . . . . . . . . . . . . . . . 63
2.8.1 Vector Wave Functions in a Waveguide . . . . . . . . . . . . . . . . . 64
2.8.2 Dyadic Green’s Function . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.8.3 Excitation of Modes by a Filamental Current . . . . . . . . . . . . . . 73
2.9 Modes of a Hollow Waveguide of Arbitrary Cross-Section . . . . . . . . . . . 75
2.9.1 Differential Equation Method . . . . . . . . . . . . . . . . . . . . . . . 75
2.9.2 Integral Equation Method . . . . . . . . . . . . . . . . . . . . . . . . 77
2.9.3 Ad Hoc Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3 Inhomogeneously Filled Waveguides 91

3.1 The Need for Hybrid modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.2 Derivation of Pertinent Equation . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.2.1 Ez -Hz Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
3.2.2 Transverse Field Formulation . . . . . . . . . . . . . . . . . . . . . . . 94
3.2.3 Physical Interpretation of the Depolarization Effect . . . . . . . . . . 94
3.2.4 Mode Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.2.5 Reflection Symmetry and Conservation of Parity . . . . . . . . . . . . 98
3.3 General Anisotropic Waveguide . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.4 Proof of Transpose of Operators . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.5 Dielectric-Slab-Loaded Rectangular Waveguides . . . . . . . . . . . . . . . . 102
3.6 Transverse Resonance Condition . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.7 Fabry-Perot Etalon . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
3.8 Rod-Loaded Circular Waveguide . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.8.1 Reflection off a Dielectric Rod . . . . . . . . . . . . . . . . . . . . . . 111
3.8.2 Reflection off a PEC Waveguide Wall . . . . . . . . . . . . . . . . . . 114
3.8.3 Reflection off an Outer Dielectric Wall . . . . . . . . . . . . . . . . . 115
3.8.4 The Guidance Condition . . . . . . . . . . . . . . . . . . . . . . . . . 115
3.9 Applications of Inhomogeneously Filled Waveguides . . . . . . . . . . . . . . 115
3.9.1 The Effect of Inhomogeneous Fillings on the Phase Velocity . . . . . 117
3.9.2 Quarter-Wave Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
3.9.3 Variable Phase Shifter . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
Contents iii
3.9.4 Variable Attenuator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

3.10 Spin Dynamics and Ferrite Materials . . . . . . . . . . . . . . . . . . . . . . 122
3.10.1 Natural Plane Wave Solutions in an Infinite Homogeneous Ferrite
Medium . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
3.10.2 Faraday Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
3.10.3 Applications of Faraday Rotation . . . . . . . . . . . . . . . . . . . . 129
4 Coupling of Waveguides and Cavities 135

4.1 Excitation of Waveguides by a Probe . . . . . . . . . . . . . . . . . . . . . . 135
4.1.1 Derivation of the Equivalent Problem and the Integral Equation . . . 135
4.1.2 Generalization to Other Structures . . . . . . . . . . . . . . . . . . . . 138
4.2 Input Impedance of the Probe . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.2.1 Variational Expressions for Input Admittance . . . . . . . . . . . . . 139
4.2.2 Rayleigh-Ritz Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.2.3 Mode Matching Method—A Tour de Force Calculation . . . . . . . . 145
4.3 Excitation of a Microstrip Patch Antenna . . . . . . . . . . . . . . . . . . . . 148
4.3.1 Magnetic Wall Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.3.2 The Q of the Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
4.3.3 Circular Polarization Excitation . . . . . . . . . . . . . . . . . . . . . 154
4.3.4 Perturbation Formula for Resonant Frequency Shift . . . . . . . . . . 155
4.3.5 Variational Impedance Formula for a Current Source . . . . . . . . . . 157
4.4 Aperture Coupling in Waveguide . . . . . . . . . . . . . . . . . . . . . . . . . 160
4.4.1 Bethe Coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
4.4.2 Equivalence Principles in Aperture Coupling . . . . . . . . . . . . . . 167
5 Discontinuities in Waveguides 179

5.1 Transmission Line Equivalence of Waveguide . . . . . . . . . . . . . . . . . . 179
5.2 Waveguide Junction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
5.2.1 Mode Matching–Eigenmode Expansion Method . . . . . . . . . . . . 182
5.2.2 Equivalence Principle and Integral Equation Formulation . . . . . . . 188
5.2.3 Relative Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
5.3 Numerical Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
5.4 Solution to the Multiple Waveguide Junction Problem . . . . . . . . . . . . . 195
5.4.1 A Two-Waveguide-Junction Problem . . . . . . . . . . . . . . . . . . 195
5.4.2 An N-Waveguide-Junction Problem . . . . . . . . . . . . . . . . . . . 197
5.4.3 Filter Design–A Resonance Tunneling Problem . . . . . . . . . . . . . 199
5.5 Hybrid Junctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
5.6 Periodic Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
5.6.1 Floquet Modes and Brillouin Zone . . . . . . . . . . . . . . . . . . . . 206
5.7 Stop Band and Coupled-Mode theory . . . . . . . . . . . . . . . . . . . . . . 207
5.7.1 Circuit Analysis of Periodic Structure . . . . . . . . . . . . . . . . . . 210
5.8 Metamaterial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
5.8.1 Evanescent Amplification by a Matched DNG Slab . . . . . . . . . . . 214
5.8.2 Composite Right-Left Handed Transmission Line . . . . . . . . . . . . 216
iv Theory of Microwave and Optical Waveguides
6 Optical Waveguides 229

6.1 Surface Waveguides–Dielectric Slab Waveguides . . . . . . . . . . . . . . . . 230
6.2 Circular Dielectric Waveguide . . . . . . . . . . . . . . . . . . . . . . . . . . 232
6.3 Weak-Contrast Optical Fiber . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
6.4 Perturbation Formula for Dielectric Waveguides . . . . . . . . . . . . . . . . 240
6.5 Mode Dispersion in an Optical Fiber . . . . . . . . . . . . . . . . . . . . . . 241
6.6 A Rectangular Dielectric Waveguide . . . . . . . . . . . . . . . . . . . . . . . 243
6.6.1 Harmonic Expansion Method . . . . . . . . . . . . . . . . . . . . . . . 243
6.6.2 Variational Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
6.7 Discontinuities in Dielectric Waveguides . . . . . . . . . . . . . . . . . . . . . 249
6.7.1 Reflection at a Laser Facet . . . . . . . . . . . . . . . . . . . . . . . . 249
6.7.2 Determination of the Modes . . . . . . . . . . . . . . . . . . . . . . . 252
6.8 Analyzing Weak Contrast Optical Fiber with WKB method . . . . . . . . . 265
6.8.1 The WKB Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 266
6.8.2 Solution in the Vicinity of a Turning Point . . . . . . . . . . . . . . . 268
6.8.3 Asymptotic Matching . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
6.9 Effective Index Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
6.9.1 Effective Index Concept . . . . . . . . . . . . . . . . . . . . . . . . . . 272
6.9.2 Quasi-TE polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
6.9.3 Quasi-TM Polarization . . . . . . . . . . . . . . . . . . . . . . . . . . 274
6.10 The Beam-Propagation Method . . . . . . . . . . . . . . . . . . . . . . . . . 275
6.11 Ray Tracing Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
6.11.1 Ray Tracing Equations in an Optical Fiber . . . . . . . . . . . . . . . 279
6.11.2 Determination of Initial Conditions . . . . . . . . . . . . . . . . . . . 281
7 Microwave Integrated Circuits 297

7.1 Quasi-TEM Approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
7.1.1 Microstrip Line—Line Capacitance . . . . . . . . . . . . . . . . . . . 299
7.1.2 Variational Expressions and Bounds for Capacitance . . . . . . . . . 306
7.2 Microstrip Line—A Frequency Dependent Theory . . . . . . . . . . . . . . . 308
7.2.1 Derivation of the Integral Equation . . . . . . . . . . . . . . . . . . . 309
7.2.2 Vector Fourier Transform (VFT) . . . . . . . . . . . . . . . . . . . . . 312
7.3 Microstrip Patch Revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
7.3.1 Integral Equation for the Resonance Case . . . . . . . . . . . . . . . . 317
7.3.2 Integral Equation for the Excitation Case . . . . . . . . . . . . . . . . 317
7.3.3 Far Field Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
7.4 Edge Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
7.5 Discontinuities in Microstrip Lines . . . . . . . . . . . . . . . . . . . . . . . . 321
7.5.1 An Open-End Discontinuity . . . . . . . . . . . . . . . . . . . . . . . 321
7.5.2 A Step Discontinuity . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
7.5.3 A Gap Discontinuity . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
7.5.4 A Slit Discontinuity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
7.5.5 A Microstrip Bend . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
7.5.6 A T Junction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
7.6 Directional Coupler Using Microstrip Line . . . . . . . . . . . . . . . . . . . . 324
Contents v
7.7 A Branch Line Directional Coupler . . . . . . . . . . . . . . . . . . . . . . . 327
8 Solitons 335
8.1 Optical Solitons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
8.2 The Korteweg de Vries Equation . . . . . . . . . . . . . . . . . . . . . . . . . 335
8.3 Derivation of the Nonlinear Schrödinger Equation . . . . . . . . . . . . . . . . 337
8.3.1 Dispersive effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
8.3.2 Solution of the Nonlinear Schrödinger Equation . . . . . . . . . . . . . 341
8.4 Solution of the KdV Equation via Inverse Scattering Transform . . . . . . . 343
8.4.1 Inverse Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
8.4.2 Solution of the KdV Equation . . . . . . . . . . . . . . . . . . . . . . 347
8.4.3 Inverse Scattering with Schrödinger Equation . . . . . . . . . . . . . . 349
8.4.4 Time-Domain Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . 350
vi Theory of Microwave and Optical Waveguides
Preface
This monograph is an outgrowth of many years of teaching at the University of Illinois at

Urbana-Champaign since Fall 1987.
Weng Cho CHEW

Fall 2012
vii
viii Theory of Microwave and Optical Waveguides
Chapter 1
Preliminary Background
1.1 Introduction
Waveguiding phenomena occur naturally or are man made. For instance, waveguides are
a fundamental component of radio wave, microwave and optical circuits [1–5]. They are
indispensable in modern technology in the radio frequency to the optical frequency range.
They are used in the telecommunication as well as in wireless communication, for example,
in the design of a cell phone. The purpose of a waveguide is to guide the energy of a wave
through a channel or a path with little attenuation. Waveguides are also used to prevent
interference between two electromagnetic signals.
The precursor to electromagnetic waveguides were acoustic waveguides as acoustic wave
theory, being scalar, was well established before electromagnetic theory [6]. Since acoustic
waves are longitudinal waves. They can be guided as a longitudinal mode in a hollow tube
for all frequencies. As a result, tubes of acoustic waveguides of different lengths have been
used as musical instruments since ancient times. The first analysis of electromagnetic guided
wave was probably done by Lord Rayleigh [7].
As we shall see later, a simple way to guide electromagnetic wave for all frequencies, is to
use two metallic conductors, usually an inner one and an outer one as in a coaxial cable. As
optical fiber guides a mode for all frequencies too, but as shall be shown, when the frequency
is very low, the mode’s energy is weakly trapped inside the fiber, making it impractical as
a waveguide for extremely low frequencies. Since most sources are finite in extent, e.g.,
antennas, they generate spherical waves in the far-field which decays algebraically. However,
waveguides, by confining the energy of the wave to a tube or a line, can cause a wave to
traverse great distances with little attenuation. An example is an optical fiber, which can
guide a signal with less than 0.3 dB/Km of attenuation [8].
Other emerging waveguiding technologies are plasmonic waveguides at optical frequencies
in nano-optics, or guiding waves using a chain of nano particles [28]. As nanoelectronic
devices are getting smaller, their dimensions are approaching the wavelengths of electron wave
functions. The propagation of electron waves in a channel can be viewed as a waveguiding
problem [29].
There are two main types of waveguides: the closed waveguide and the open waveguide. In
1
2 Theory of Microwave and Optical Waveguides
a closed waveguide, the electromagnetic energy is completely trapped within metallic walls.
The only way to gain access to the energy is to tap holes in the waveguide wall. Hence,
it transmits signals with very good shielding and very little interference from other signals.
Figure 1.1 shows some examples of closed waveguides. Notice that a closed waveguide can be
of one or more conductors. On the other hand, an open waveguide allows its field to permeate
all of space, even though most of the energy is still trapped and localized around the guidance
structure. As shown in Figure 1.2, an open waveguide is either a multi-conductor waveguide
or a dielectric waveguide. It is usually easier to fabricate an open waveguide. However, as a
result of their openness, such waveguides usually radiate at discontinuities and bends.
Because open waveguides radiate at discontinuities and bends, some of them are even used
as antennas [9]. There are also modes that are weakly guided by an open waveguide, i.e., it
radiates as it is being guided. Examples of such modes are the leaky modes. An antenna
built using such a mode is known as a leaky wave antenna.
The analysis of waveguides requires a basic understanding of electromagnetic theory. We
will review our basic electromagnetic theory in the following section.
1.2 History of Electricity and Magnetism

Humans are exposed to electromagnetic phenomena on a daily basis. Light wave is an elec-
tromagnetic phenomenon, so is lightning. Lodestone is probably the first human contact with
something magnetic. Ancient Chinese knew about the magnetic properties of lodestones, and
made compasses out of them. Static electricity was a phenomenon popularly demonstrated
in European courts to entertain the nobilities. But it was not until 1771-1773 that serious
experiments were done on static electricity by Henry Cavendish (1731-1810). To this day,
the Cavendish Laboratory stands in the University of Cambridge in England to the honor of
Cavendish [10].
Faraday’s law was formulated by Michael Faraday (1791-1867) to describe the fact that a
changing magnetic flux, linked to a metallic loop, will induce a voltage in the loop [12]. This
fact can be used to design generators that produce electricity for our homes. A multi-turn coil
can be immersed in the magnetic field of a permanent magnet, and rotated rapidly. A voltage
is then induced in the coil, which can be tapped to deliver electricity for a large number of
applications. Conversely, a DC current in a static magnetic field experiences a force due to
Lorentz force law. This idea can be used to design a motor. In fact, a DC motor was invented
by William Sturgeon in 1832 [27].
Ampere’s law was later formulated by André Marié Ampère (1775-1836) who stipulated
that a wire carrying a current produces a magnetic field [13]. Moreover, the magnetic field
is produced according to the right-hand rule. (Note: The stipulation that the magnetic field
goes from the north pole of a bar magnet to its south pole is entirely by convention. Hence,
the right-hand rule in Ampere’s law is also entirely by convention. Also, the concept of right-
handedness and left-handedness is hard to describe to an extra-terrestrial creature living in
another universe who has never seen a human before. Try that for yourself [11]!)
Gauss’ law by Carl Friedrich Gauss (1777-1855) describes that if a charge generating an
electric field is enclosed by a surface S, the sum of the total flux flowing through the surface
is equal to the total charge contained within the surface [15]. If the surface does not enclose
Preliminary Background 3
any charge, the sum of the total charge through the surface is equal to zero. Coulomb’s law
can be derived from Gauss’ law.
The above period represented the era during which the understanding of electromagnetism
was incomplete. Nevertheless, technology using electricity and magnetism was prevalent. As
soon as Alessandro Volta invented the battery, Ampere developed telegraphy in the early
1800s. In fact, submarine cables were laid during a large part of the nineteenth century by
the British empire around the world to enable telegraphic communication. So it was quite
well known that wave phenomena exist on telegraphic lines before the completion of Maxwell’s
theory as we shall discuss next.
Electromagnetic theory was completely formulated by the work of James Clerk Maxwell
(1831-1879) [16]. In 1864, he put forth the theory that there should be a term, called the
displacement current term, to be added to Ampere’s law. The work completed electromag-
netic theory and it was proven mathematically that electromagnetic wave was a possible
electromagnetic phenomenon. Consequently, it was realized that light waves were electro-
magnetic waves. Because of this important discovery, electromagnetic theory is also known
as Maxwell’s theory, and the set of equations is also known as Maxwell’s equations. However,
when Maxwell first wrote down the complete form of electromagnetic theory, it was in some
20 equations. It was Oliver Heaviside who recast those equations in their present succinct
form. Rightfully, these equations should be called the Maxwell-Heaviside equations [17].
In 1888, Heinrich Rudolf Hertz (1857-1894) performed an experiment to verify the exis-
tence of electromagnetic wave. Two spheres in close proximity to each other were used as
capacitors to store electric charges. The charges generate an electric field. A rapid discharge
of the electric charge causes the electric field to collapse, producing an electromagnetic wave.
The wave has both electric and magnetic field in it. Therefore, a wire loop, via Faraday’s law,
can be linked to the time varying magnetic flux, producing a voltage. This voltage creates
a spark in a gap left in the loop, even when the loop is at a distance from the spheres. To
his honor, the unit for frequency, which was cycles/second, is now named Hertz. The term
megahertz (MHz), or gigahertz (GHz) now adorns the spec sheets of most computers.
In 1901, Guglielmo Marchese Marconi (1874-1937) successfully transmitted an electromag-
netic signal across the Atlantic Ocean from Cornwall, England to Saint John, Newfoundland
in North America [18]. Many nay sayers predicted that he would be doomed to failure as
the earth surface is curved. Fortunately, the ionosphere in the outer atmosphere acted like a
mirror, and the electromagnetic waves bounced back to earth. It was in fact a serendipitous
experiment. It was after his experiments that wireless telegraphy was established. However,
Marconi never received a patent for his invention. The patent for telecommunication was
claimed by Nikola Tesla, who had the idea before Marconi.
Since then, electromagnetic theory has spurred the development of myriads of technolo-
gies, many of which are electrical engineering related. Some of the more prominent ones are
the development of the radar, various antennas for telecommunication, remote sensing sys-
tems, lasers and optics and more recently, wireless communication, computer chip design, and
electromagnetic compatibility and electromagnetic interference. The advent of quantum tech-
nologies as seen in quantum optics, quantum computers, quantum communications, Casimir
force in MEMS/NEMS, quantum transport in electronic devices, photonics, will also dwell
on classical electromagnetics in combination with modern physics concepts.
As of this date, electromagnetic theory continues to help in the conception, analysis,
and design of many new technologies. Hence, Maxwell-Heaviside equations are solved over
and again for many of these analysis tasks. As a consequence, much research has gone into
developing methods to impact many analyses in science and engineering.
1.3 Maxwell-Heaviside Equations

Soon after the advent of Maxwell’s theory, much analysis was performed with Maxwell-
Heaviside equations. By 1897, Lord Rayleigh had already studied the propagation of electro-
magnetic waves through tubes. There was then much knowledge on propagation and guidance
of acoustic waves. Hence, analogue between acoustic waves and electromagnetic waves were
drawn as much as possible, although acoustic waves are scalar while electromagnetic waves
are vector in nature.
In vector notation, and MKS units, Maxwell-Heaviside equations are given as
–
+ – +–
+
COAX Twinax Rectangular Finline

Waveguide Circular
Waveguide
Figure 1.1: Examples of closed waveguides.
+ + –
+ –
Twin Line
Coplanar
Microstrip Line Optical
Waveguide
Fiber
Optical Thin-Film Periodic Dipole Array

Waveguide Structure
Figure 1.2: Examples of open waveguides.
∂
∇ × E(r, t) = − B(r, t), (1.3.1)
∂t
∂
∇ × H(r, t) = D(r, t) + J(r, t), (1.3.2)
∂t
∇ · B(r, t) = 0, (1.3.3)
∇ · D(r, t) = ρ(r, t). (1.3.4)
where E is the electric field in volts/m, H is the magnetic field in amperes/m, D is the
electric flux in coulombs/m2 , B is the magnetic flux in webers/m2 , J(r, t) is the current
density in amperes/m2 , and ρ(r, t) is the charge density in coulombs/m3 . For time varying
electromagnetic fields, only two of the four Maxwell-Heaviside equations are independent.
Equations (1.3.3) and (1.3.4) can be derived from Equations (1.3.1) and (1.3.2) by using the
continuity equation:
∂ρ(r, t)
∇ · J(r, t) + = 0. (1.3.5)
∂t

If we assume
that A(r, t) = <e A(r)e−iωt , for all E(r, t), H(r, t), J(r, t), and ρ(r, t) =
<e ρ(r)e−iωt ; namely, the fields are time harmonic, the above equations become,
∇ × E(r) = iωB(r), (1.3.6)
∇ × H(r) = −iωD(r) + J(r), (1.3.7)

∇ · B(r) = 0, (1.3.8)
∇ · D(r) = ρ(r). (1.3.9)
The electric and magnetic fluxes are related to the electric and magnetic fields via the
constitutive relations, the most general of which are
D = · E + ξ · H, (1.3.10)
B = µ · H + ζ · E, (1.3.11)
where , ξ, µ and ζ are tensors. It is also the constitutive relations that characterize the
medium we are describing. A medium with the above constitutive relations is known as a
bianisotropic medium. A more commonly encountered medium is an anisotropic medium
with the constitutive relations
D = · E, (1.3.12)
B = µ · H. (1.3.13)
When , ξ, µ and ζ are functions of space, the medium is also known as an inhomogeneous
medium. When they are functions of frequency, the medium is frequency dispersive. When
they are functions of wavelength, it is spatially dispersive. For an isotropic medium, the
constitutive relations simply become
D = E, B = µH. (1.3.14)
In free-space, = 0 = 8.854 × 10−12 farad/m, µ = µ0 = 4π × 10−7 henry/m. The constant

c = √µ10 0 is related to the velocity of light, which has been very accurately measured. The
unit of meter is defined such that c is exactly equal to 299,792,458 m/s. The value of µ0 is
assigned to be 4π × 10−7 henry/m while the value of 0 is calculated from c.
1.4 Wave Equation

For an anisotropic, inhomogeneous medium, Maxwell-Heaviside equations for time-harmonic
fields could be written as
∇ × E(r) = iωµ · H(r), (1.4.1)
∇ × H(r) = −iω · E(r) + J(r), (1.4.2)
∇ · µ · H(r) = 0, (1.4.3)
∇ · · E(r) = ρ(r). (1.4.4)
If we take the curl of µ−1 · (1.4.1), we obtain, via the use of (1.4.2), that
∇ × µ−1 · ∇ × E(r) − ω 2 · E(r) = iωJ(r). (1.4.5)
Similarly, we can show that
∇ × −1 · ∇ × H(r) − ω 2 µ · H(r) = ∇ × −1 · J(r). (1.4.6)
Equations (1.4.5) and (1.4.6) are two vector wave equations governing the solutions of elec-
tromagnetic fields in an inhomogeneous, anisotropic medium. Here, µ and are functions of
positions; hence, they do not commute with the ∇ operator. Also, for time-varying fields,
E and H are derivable from each other; only one of the two equations (1.4.5) and (1.4.6) is
necessary to fully describe the electromagnetic fields.
For an isotropic medium, (1.4.5) and (1.4.6) reduce to
∇ × µ−1 ∇ × E(r) − ω 2 E(r) = iωJ(r), (1.4.7)
∇ × −1 ∇ × H(r) − ω 2 µH(r) = ∇ × −1 J(r). (1.4.8)

For electrodynamics, either one of the above equations is self-contained. We can derive the
phenomena of dynamic electromagnetic fields by just studying one of them.
1.5 Boundary Conditions

We cannot find a unique solution to a partial differential equation unless we specify the
boundary conditions as well. Equations (1.4.5) to (1.4.8) are vector wave equations whose
solutions we will seek over and again. One common method of solving the above equations is
to find the solutions in each of the homogeneous regions that constitute the inhomogeneity,
provided that the inhomogeneity is piecewise constant. The unique solution is then obtained
by matching the boundary conditions at the interface.
Since either Equation (1.4.5 ) or (1.4.6 ) is sufficient in describing electromagnetic fields,
the boundary conditions must be buried in them. Therefore, we can derive the boundary
conditions from them. To do this, we integrate (1.4.5) about a small area between the
interface of two media. Invoking Stokes’ theorem, we have
I Z Z
dl · (µ−1 · ∇ × E) − ω 2 dS · · E = iω dS · J. (1.5.1)
C A A
JS, MS
δ
^n
^
E1, H1
C
A
µ 1, ε 1
µ2 , ε 2 E2, H2
Figure 1.3: Boundary conditions at an interface.
Letting δ → 0, the surface integral on the left-hand side of the above equation vanishes.
Assuming that we have a current sheet Js , we can show that
n̂ × (µ−1 −1
1 · ∇ × E1 ) − n̂ × (µ2 · ∇ × E2 ) = iωJs . (1.5.2)
Since ∇ × E = iωµ · H, we have
n̂ × H1 − n̂ × H2 = Js . (1.5.3)
Performing the same analysis for Equation (1.4.6), we arrive at
n̂ × E1 − n̂ × E2 = 0. (1.5.4)
Equations (1.5.3) and (1.5.4) are the important boundary conditions we will use over and
again.
The boundary condition (1.5.3) can also be gleaned from (1.3.7). If J(r) is a current sheet
at on interface, represented by a delta function singularity, then this singularity must be from
the normal derivative of the tangential component of the magnetic field. From this fact we
can derive (1.5.3). By the some token, (1.5.4) can be derived from (1.3.8).
1.6 Reciprocity Theorem

If we have two sources J1 and J2 radiating in an anisotropic, inhomogeneous medium, and
J1 produces the field E1 , J2 produces the field E2 , the reciprocity theorem requires that for
a reciprocal medium,
hE1 , J2 i = hE2 , J1 i, (1.6.1)
R
where hA, Bi stands for drA · B. This theorem is derivable from Equation (1.4.5) with
constraints on and µ. When the source J1 is radiating, the field E1 satisfies the equation
∇ × µ−1 · ∇ × E1 − ω 2 · E1 = iωJ1 . (1.6.2)
When J2 is radiating, the field E2 satisfies the equation
∇ × µ−1 · ∇ × E2 − ω 2 · E2 = iωJ2 , (1.6.3)
where µ and in (1.6.2) and (1.6.3) represent the same medium. Dot-multiplying (1.6.2) by
E2 and integrating, and (1.6.3) by E1 and integrating, we have
hE2 , ∇ × µ−1 · ∇ × E1 i − ω 2 hE2 , · E1 i = iωhE2 , J1 i, (1.6.4)
hE1 , ∇ × µ−1 · ∇ × E2 i − ω 2 hE1 , · E2 i = iωhE1 , J2 i. (1.6.5)

Since Z
hE2 , ∇ × µ−1 · ∇ × E1 i = drE2 · ∇ × µ−1 · ∇ × E1 , (1.6.6)
V
we can use the identity
∇ · (A × B) = B · ∇ × A − A · ∇ × B (1.6.7)
and Gauss’ theorem to get

Z
−1
hE2 , ∇ × µ · ∇ × E1 i = dr(∇ × E2 ) · µ−1 · (∇ × E1 ) (1.6.8)
ZV
+ dS n̂ · (µ−1 · ∇ × E1 ) × E2
S
Z
dr(∇ × E2 ) · µ−1 · (∇ × E1 ) (1.6.9)
V
Z
+ iω dS n̂ · (H1 ) × E2 , (1.6.10)
S
where V and S are a volume and a surface tending to infinity. When S → ∞, µ becomes
isotropic and homogeneous. Furthermore, the solutions to the vector wave equation become
plane waves. Hence, ∇ → ik, and we have
(µ−1 · ∇ × E1 ) × E2 |rs = iµ−1 −1

0 (k × E1 ) × E2 = −iµ0 k(E1 · E2 ). (1.6.11)
In this manner, the surface integral in (1.6.10) is symmetric about E1 and E2 . If µ−1 is
symmetric, then the first integral on the right-hand side of (1.6.10) is also symmetric about
E1 and E2 . Hence, if µ−1 is symmetric, the first term of (1.6.4) and (1.6.5) are equal. If
is also symmetric, then the second term of (1.6.4) and (1.6.5) are also equal. Therefore, we
deduce that (1.6.1) is satisfied or that reciprocity holds when
µ = µt , = t . (1.6.12)
In other words, µ and are symmetric (if µ is symmetric, µ−1 is symmetric). The condi-
tion expressed in Equation (1.6.12) is necessary for an anisotropic medium to be areciprocal
medium. It also follows that all isotropic media are reciprocal.
The integral defined in Equation (1.6.1) is also known as a reaction. It could be thought of
as a generalized measurement. In words, the reciprocity theorem states that for a reciprocal
S
J2
J1 µ (r) , ε (r)
Figure 1.4: Proof of Reciprocity.
medium, the E-field due to J1 measured by J2 is the same as the E-field due to J2 measured
by J1 .
Examples of non-reciprocal media are plasma and ferrite media biased by a magnetic field.
A medium can be lossy and still be reciprocal.
In electromagnetics, it is customary to add a fictitions magnetic current M to Faraday’s
law such that
∇ × E = −iωD − M (1.6.13)
A reciprocity theorem that can be derived involving magnetic current is
hE1 , J2 i − hH1 , M2 i = hE2 , J1 i − hH2 , M1 i (1.6.14)
in replacement of (1.6.1).
Reciprocity theorem is deeply related to the symmetry of differential operators related to
Maxwell-Heaviside equations. For example, we can express (1.6.2) and (1.6.3) as
DE1 = J1 (1.6.15)
DE2 = J2 (1.6.16)
Where D is the pertinent differential operators. Then, hE2 , J1 i = hE1 , J2 i implies
hE2 , DE1 i = hE1 , DE2 i (1.6.17)
The above is the analogue of

at · A · b = bt · A · a (1.6.18)
t
which implies the A = A or A is symmetric. Hence, (1.6.17) implies that D is symmetric,
and this is possible only if (1.6.12) is satisfied.
The symmetry of D is the deeper underlying reason for the reciprocity theorem. For media
that are reciprocal, the symmetry of the electromagnetic equations will give rise to a number
of operators that are also symmetrical such as the impedance and admittance matrices, as
we shall learn later.
1.6.1 Lorentz Reciprocity Theorem

If the volume integrals in (1.6.4) and (1.6.5) are taken over a finite volume, then on subtracting
the two equations, making use of (1.6.10), and assuming the symmetry of the permeability
and permittivity tensors, we arrive at the general case of the reciprocity theorem,
I
hE2 , J1 i − hE1 , J2 i = dS n̂ · (E1 × H2 − E2 × H1 ) (1.6.19)
S
When the volume does not enclose the sources, we arrive at

I I
dS n̂ · (E1 × H2 ) = dS n̂ · (E2 × H1 ) (1.6.20)
S S
The above is generally known as the Lorentz reciprocity theorem. It is useful in waveguides
when sources are not involved.
1.7 Energy Conservation

Energy conservation in electromagectics is defined by the Poynting theorem. Poynting the-
orem holds for the time domain as well as the frequency domain. The theorem in the time
domain is actually quite different from that in the frequency domain. We shall present first
the time domain version.
1.7.1 Time Domain Poynting Theorem

The time domain Poynting theorem, sometimes known as the real Poynting theorem, governs
the conservation of instantaneous energy for electromagnetic field. To derive it, we start with
∇ · [E(r, t) × H(r, t)] = H · ∇ × E − E · ∇ × H

∂B ∂D (1.7.1)
= −H · −E· −E·J
∂t ∂t
Defining the Poynting vector
S (r, t) = E (r, t) × H (r, t) (1.7.2)
we have
∂B ∂B
∇ · S(r, t) = − H · +E· −E·J (1.7.3)
∂t ∂t
For free space where B = µ0 H, D = 0 E, we can show that
∂B ∂H 1 ∂
H· = H · µ0 = µ0 H · H (1.7.4)
∂t ∂t 2 ∂t
and similarly, for the electric flux term, we have
∂ 1
∇ · S(r, t) = − (µ0 H · H + 0 E · E) − E · J (1.7.5)
∂t 2
Then term
1
WT = (µ0 H · H + 0 E · E) (1.7.6)
2
corresponds to the total energy stored in the magnetic field and the electric field. When
∂
∂t WT is positive, it corresponds or contributing the negative term to ∇ · S implying a influx
of power at a point. The last term E · J corresponds to power absorbed or generated by the
current J. When E · J is positive, the current J is absorptive. This is true of a conductive
medium where J = σE.
Please note that the above derivation that leads to expression (1.7.6) is not valid for
material media. All material media have to be frequency dispersive, and hence, in the time
domain, the constitutive relations are denoted by time convolutions.
1.7.2 Frequency Domain Poynting Theorem

The frequency domain Poynting theorem governs energy conservation for complex power.
Hence, it is also known as the complex Poynting theorem. We start with
∇ · (E × H∗ ) = iωH∗ · B − iωE · D∗ − E · J∗
(1.7.7)
= iωH∗ · µ · H − iωE · ∗ · E∗ − E · J∗
Defining the complex Poynting vector
S̃(r) = E × H∗ (1.7.8)
the above becomes

∇ · S̃ = iω(H∗ · µ · H − E · ∗ · E∗ ) − E · J∗ (1.7.9)
For a source-free region, this becomes
∇ · S̃ = iω(H∗ · µ · H − E · · E∗ ) (1.7.10)
If
H∗ · µ · H = E · · E∗ (1.7.11)
in a region, then the right-hand side is zero, and there is no net power flux into or out of the
region. This occurs at resonance in a cavity.
1.7.3 Complex Power

The complex Poynting theorem is quite different from the real Poynting theorem. It can
be shown that half the real part of the complex Poynting vector is the time average of the
instantaneous Poynting vector, viz., [see problem 1.2]
1 h i
hS(r, t)i = <e S̃(r) (1.7.12)
2
The part that corresponds to the stored energy in the complex Poynting theorem is the
difference of the magnetic energy and electric energy stored, whereas that in the real Poynting
theorem is the sum of the two. This is because the imaginary part of the complex power is
reactive power [see problem 1.3]. Reactive power in a time harmonic system corresponds to
power that flows into a system, and later flows out of a system. Hence, its time average is
zero.
Notice that in a resonance system or circuit such as the LC tank circuit, the stored
magnetic energy and electric energy are equal to each other, and they exchange with each
other. Since the reactive power is the difference in the store magnetic and electric energy, it
is zero in this case. Therefore, when an LC tank circuit is at resonance, there is no need for
an external supply of reactive power.
1
Instantaneous Power V(t)I(t)
0.5
φ=0
φ=π/6
φ=π/3
φ=π/2
−0.5
0 2 4 6 8 10 12 14
t
Figure 1.5: Plots of P (t) = V (t)I(t) versus time t where V (t) = cos(t), and I(t) = cos(t + φ)
for various values of φ.
Even though no net power is delivered in the reactive power, a power utility company will
still charge its customer for the use of this power for two reasons: First, not all reactive power
is retrievable as it has to be sent over power lines that have conductive losses. Second, the
power company has to maintain a generator that can absorb the oscillation in the total power
caused by the presence of reactive power. Figure 1.5 shows that the instantaneous power can
be negative as well as positive when there is a phase shift between the voltage and the current
in a circuit.
1.7.4 Lossless Conditions

For an isotropic medium, the conditions for it to be lossless are that =m(µ) = 0, and =m() =
0 where “=m” implies “imaginary part.” However, the condition for an anisotropic medium
is quite different. We can derive the general lossless condition from energy conservation.
For a lossless medium, for energy conservation, and from the complex paynting’s theorem,
we require that Z I
<e dV ∇ · (E × H∗ ) = <e dS · (E × H∗ ) = 0, (1.7.13)
V S
∗
since <e[E × H ] corresponds to time average power flow. The above implies that
 
 Z 
<e iω dV (H∗ · µ · H − E · ∗ · E∗ ) = 0. (1.7.14)
 
V
A sufficient condition for arbitrary V is to require that H∗ · µ · H and E · ∗ · E∗ to be purely

real or their conjugates to be themselves, i.e.,
(H∗ · µ · H)∗ = H · µ∗ · H∗ = H∗ · µ† · H = H∗ · µ · H. (1.7.15)
Therefore, µ† = µ. Similarly the condition on E · ∗ · E∗ to be purely real is † = .

Consequently, the lossless conditions for an anisotropic medium is
= † , µ = µ† . (1.7.16)
In other words, the permittivity tensor and the permeability tensor have to be Hermitian.
1.8 Energy Density in Dispersive Medium

In the following derivation, we assume that E, H, and J have e−iωt time dependence, where
ω is a complex frequency [30]. Then
∇ · [E(t) × H∗ (t)] = H∗ (t) · ∇ × E(t) − E(t) · ∇ × H∗ (t)

= H∗ (t) · [iωµ · H(t)] − E(t) · [iω ∗ ∗ · E∗ (t) + J∗ (t)]
= iωH∗ (t) · µ · H(t) − iω ∗ E(t) · ∗ · E∗ (t) − E(t) · J∗ (t) (1.8.1)
Next, we let ω = ω 0 + iω 00 where ω 0 and ω 00 are real numbers. Then
∇ · [E(t) × H∗ (t)] = i(ω 0 + iω 00 )H∗ (t) · µ(ω 0 + iω 00 ) · H(t)

− i(ω 0 − iω 00 )E(t) · ∗ (ω 0 + iω 00 ) · E∗ (t) − E(t) · J∗ (t) (1.8.2)
Ordinarily, if ω is pure real, the time dependence would have canceled in the above, but
because ω is complex, the has time dependence of exp(2ω 00 t). Assuming that ω 00 ω 0 , we
can Taylor expand the right-hand side to get

∗ . 0 00 ∗ 0 00 ∂ 0
∇ · (E × H ) = i(ω + iω )H · µ(ω ) + iω µ(ω ) · H
∂ω 0

0 00 ∗ 0 00 ∂
− i(ω − iω )E · (ω ) − iω (ω ) · E∗ − E · J∗
∗ 0
(1.8.3)
∂ω 0
Collecting leading order and first order terms, we have

.
∇ · (E × H∗ ) = iω 0 [H∗ · µ(ω 0 ) · H − E · ∗ (ω 0 ) · E∗ ]

00 ∗ 0 0 ∂ 0
− ω H · µ(ω ) + ω µ(ω ) · H
∂ω 0

∗ 0 0 ∂
+ E · (ω ) + ω (ω ) · E − E · J∗
∗ 0
∂ω 0
= iω 0 [H∗ · µ(ω 0 ) · H − E · ∗ (ω 0 ) · E∗ ]

00 ∗ ∂ 0 0 ∂
−ω H · [ω µ(ω )] · H + E · [ω (ω )] · E − E · J∗
0 ∗ 0
(1.8.4)
∂ω 0 ∂ω 0
For lossless media, the first term is purely imaginary, while the second term is purely real.
The first term corresponds to reactive power. The second term comes about because for a
0 00 0 00
complex exponential of the form e−i(ω +iω )t = e−iω +ω t , the field strength is growing with
00
e2ω t time dependence. If we take the real part of (1.8.4), and focussing on the part of space
where J = 0, we have

1 ∗ 1 00 ∗ ∂ 0 0 ∂ 0 ∗ 0 ∗
∇ · <e(E × H ) = − ω H · ω µ(ω ) · H + E · ω (ω ) · E
2 2 ∂ω 0 ∂ω 0
∂
= − WT (1.8.5)
∂t
The above has the physical meaning that the divergence of the time-average real power flow
on the left-hand side is due to the time variation of the energy density on the right-hand side.
00
The energy density has a time dependence of e2ω t . Consequently, we identify the energy
density for dispersive media as

1 ∗ ∂ 0 0 ∂ 0 ∗ 0 ∗
WT = H · ω µ(ω ) · H + E · ω (ω ) · E (1.8.6)
4 ∂ω 0 ∂ω 0
When the medium is free space, we have

1
WT = {H∗ · µ0 H + E · 0 E∗ } (1.8.7)
4
which agrees with what we have derived from time-domain Poynting theorem.
1.9 Symmetries in Electromagnetics

Symmetries play an important role in the solutions of Maxwell-Heaviside equations. They
can be used for simplifying solutions to Maxwell-Heaviside equations, or they can be used to
derive new solutions. Alternatively, they can be used to predict how solutions behave once
symmetry is broken.
In solid state physics, symmetry is used to understand the propagation of electronic waves
in crystalline structures which have a high degree of symmetry. Due to the symmetries, group
theory can be used to analyze the physical characteristics of waves in crystalline structures
[20, 21]. Unlike electronic waves, electromagnetic waves, for most applications, propagate in
in nonsymmetric structures. Hence, the exploitation of symmetry in electromagnetics has
not reached the level in solid state physics. However, there are still a few symmetries we can
exploit, especially in waveguides and resonators which usually have some degrees of symmetry
associated with them.
Some of the obvious symmetries are translational symmetry and rotational symmetry.
Translational symmetry exploits the fact that Maxwell-Heaviside equations are invariant after
a translation in space. Rotational symmetry implies that a solution of Maxwell-Heaviside
equations remains a solution after rotation. Other symmetries are time-reversal symmetry
and reflection symmetry that we shall discuss next.1
1.9.1 Time Reversal Symmetry

In a lossless environment, solutions to Maxwell-Heaviside equations are time reversible in the
same medium. That is if a solution is found, and if we change t to −t, the solution remains a
valid solution to Maxwell-Heaviside equations within the same medium. This is like playing
a movie backward.
It is clear that the solution to the wave equation is time reversible. However, when we
have a lossy wave equation, the solution decays forward in time, but grows backward in
time. Therefore, the solution is not time reversible, namely, the reverse-time solution is a
solution to an active medium (amplifying medium like a laser cavity) but the forward-time
solution corresponds to a lossy medium. For the same reason, solutions to Maxwell-Heaviside
equations are not time reversible in a lossy medium.
To obtain a time-reversed field, we let t → −t. For example, a time-reversed B field is
B(r, −t). Then the time derivative of this time-reversed B field is
∂ ∂
B(r, −t) = − 0 B(r, t0 ) (1.9.1)
∂t ∂t
which is the negative of the original time derivative. Consequently, when time-reversed fields
are substituted back into Maxwell-Heaviside equations, they can be written as
∂
∇ × E(r, t0 ) = B(r, t0 ), (1.9.2)
∂t0
∂
∇ × H(r, t0 ) = − D(r, t0 ) + J(r, t0 ), (1.9.3)
∂t0
∇ · B(r, t0 ) = 0, (1.9.4)
∇ · D(r, t0 ) = ρ(r, t0 ). (1.9.5)
We will retrieve the original Maxwell-Heaviside equations if the signs of H, B, and J are
reversed. The need to reverse these quantities is also necessary for energy conservation in
Poynting theorem. The sign of E × H and has to change for time-reversed solution to reflect
1 Some of these symmetries have been used successfully in computational electromagnetics to expedite
numerical solutions of Maxwell-Heaviside equations [22]. See also discussions in [23, p. 268].
that the energy flow has to change direction. Note that the sources J and ρ are also time
reversed.
Alternatively, we can change the sign of E, D, and ρ. But the convention is to change
the signs of H, B, and J. A positive charge, when moving through space, remains a positive
charge when time-reversed. However, it produces a current of opposite polarity.
Since time always occurs as exp(−iωt) in the frequency domain, replacing t with −t is
the same as replacing i with −i. Hence, a time reversed field is obtained by conjugating the
frequency-domain field. The design of phase-conjugate mirror was in vogue in optics to create
a time-reversed optical field [31–33].
1.9.2 Reflection Symmetry

It was believed once that all laws of physics can be replicated in the mirror world, namely,
laws of physics remain the same under reflection. This is known as the conservation of parity.
However, it is now known that some laws of physics do not satisfy parity conservation [24].
However, the law of electromagnetics satisfies parity conservation. We just need to replace
a right-hand rule with a left-hand rule for the reflected solution, namely, the solution in the
mirror world.
A symmetry closely related to reflection symmetry is inversion symmetry [21]. In inversion,
we let r → −r, or in detail, x → −x, y → −y, and z → −z. A reflected function or object can
always be obtained from an inverted function or object by a rotation. For instance, if we have
a mirror in the xy plane, and we put an object in front of the mirror, the reflected object will
have z → −z with its xy coordinates unchange. However, this can also be obtained by first
inverting the object, followed by a 180 degree rotation about the z axis. Since a rotation of a
solution is still a solution to Maxwell-Heaviside equations, we will just discuss what inversion
does to a solution.
When we have a vector field such as E(r, t), we assume that the direction of the field also
change after inversion by replacing x̂ → −x̂, ŷ → −ŷ, and ẑ → −ẑ. Hence, a vector field
under inversion becomes −E(−r, t). If we take the curl of this inverted field, we have
∇ × [−E(−r, t)] = ∇0 × E(r0 , t) (1.9.6)
after we let r0 = −r, and under this change of variables, ∇ = −∇0 . We can substitute these
inverted fields into Maxwell-Heaviside equations to see if these equations retain their original
forms.
Consequently, Maxwell-Heaviside equations under substitution of inverted fields, and with
a change of variables r0 = −r, become
∂
∇0 × E(r0 , t) = B(r0 , t), (1.9.7)
∂t
∂
∇0 × H(r0 , t) = − D(r0 , t) − J(r0 , t), (1.9.8)
∂t
∇0 · B(r0 , t) = 0, (1.9.9)
∇0 · D(r0 , t) = ρ(r0 , t). (1.9.10)
However, the above is not the original Maxwell-Heaviside equations. The original Maxwell-
Heaviside equations can be retrieved if we can change the signs of B and H. This is under-
standable, since the right-hand rule becomes a left-hand rule in the mirror or reflected world
(since the reflected world is related to the inverted world by just a rotation). Hence, a change
of the signs of B and H will convert the left-hand rule back to the right-hand rule.
1.9.3 Polar Vectors and Pseudovectors

A word is in order about polar vectors versus pseudovectors (also known as axial vectors) [23].
A polar vector (or vector) changes sign under inversion, but a pseudovector does not. For
instance, if A and B are polar vectors, they will change sign under inversion. However, a
vector C = A × B will not change sign under inversion if the cross product is defined in the
original right-handed coordinate system. Hence, C is a pseudovector. In electromagnetics,
we can regard H and B as pseudovectors that do not change sign under inversion. In this
case, Maxwell-Heaviside equations are invariant under inversion or reflection.
By the same token, pseudoscalars exist. The scalar a · b × c, where a, b, and c are polar
vectors, is a pseudoscalar which changes sign under inversion.
If we define B and H to be pseudovectors instead, and they do not change sign under
inversion, then we do not have to change the sign of B and H instead in
1.10 Green’s Function

The Green’s function to a wave equation is the solution when the source is a point source
[25, 26]. When we know the solution to the wave equation due to a point source, the solution
due to a general source can be obtained by the principle of linear superposition. This is
because the wave equation is linear, and a general source could be thought of as a superposition
of point sources.
For example, if we need to find the solution to the following equation,
(∇2 + k 2 )ψ(r) = S(r), (1.10.1)
we can first find the Green’s function which is the solution to the following equation,
(∇2 + k 2 )g(r − r0 ) = −δ(r − r0 ). (1.10.2)
If we know g(r − r0 ), ψ(r) can be found formally. Multiplying (1.10.1) by g(r − r0 ) and
(1.10.2) by ψ(r), and integrating over volume, and subtracting, we obtain
Z Z
dr[g(r − r )∇ ψ(r) − ψ(r)∇ g(r − r )] = g(r − r0 )S(r)dr + ψ(r0 ).
0 2 2 0
(1.10.3)
V V
Noting that g∇2 ψ − ψ∇2 g = ∇ · (g∇ψ − ψ∇g), we can rewrite the left-hand side, using Gauss’
divergence theorem as
I Z
dS · (g∇ψ − ψ∇g) = g(r − r0 )S(r)dr + ψ(r0 ). (1.10.4)
S V
When S → ∞, all fields look like plane waves, and we can replace ∇ → ik. The left-hand
side of (1.10.4) then vanishes, and we have
Z
ψ(r) = − dr0 g(r − r0 )S(r0 ). (1.10.5)
V
Hence the solution of (1.10.1) can be written as an integral superposition of the solution of
(1.10.2).
To find the solution of Equation (1.10.2), we solve it in spherical coordinates with the
origin at r0 . Then, it becomes
(∇2 + k 2 )g(r) = −δ(r) = −δ(x)δ(y)δ(z). (1.10.6)
For r 6= 0, the homogeneous, spherically symmetric solution to (1.10.6) is
e+ikr e−ikr
g(r) = C +D . (1.10.7)
r r
V
S
S(r)
Figure 1.6: The radiation of a source S(r) in a volume V .
Physical grounds require that we have only outgoing solutions; hence,
eikr
g(r) = C . (1.10.8)
r
We can match the constant C to the singularity at the origin by substituting (1.10.8) into
(1.10.6), and integrating Equation (1.10.6) over a small volume about the origin.
Z Z
Ceikr Ceikr
dV ∇ · ∇ + dV k 2 = −1. (1.10.9)
r r
∆V ∆V
The second integral vanishes when ∆V → 0, because dV = 4πr2 dr. We can convert the first
integral in (1.10.9) into a surface integral using Gauss’ theorem, and obtain
∂ eikr
lim 4πr2 C = −1, (1.10.10)
r→0 ∂r r
or that C = 1/4π. Therefore, in general
0
eik|r−r |
g(r − r) = (1.10.11)
4π|r − r0 |
The solution to (1.10.1), from Equation (1.10.5) is then

Z 0
eik|r−r |
ψ(r) = − dr0 S(r0 ). (1.10.12)
4π|r − r0 |
V
Equation (1.10.12) is a convolutional integral, a consequence of the principle of linear super-

position.
For the vector wave equation in a homogeneous, isotropic medium, the equation is
∇ × ∇ × E(r) − k 2 E(r) = iωµJ(r). (1.10.13)
By using the fact that ∇ × ∇ × E = −∇2 E + ∇∇ · E, and that ∇ · E = 1 ρ = 1

iω ∇ · J, we
can rewrite (1.10.13) as

2 2 ∇∇
∇ E(r) + k E(r) = −iωµ I + 2 · J(r). (1.10.14)
k
There are three scalar wave equations embedded in the above equation. We can solve each of
them in the manner of Equation (1.10.5), and we have
Z
∇ 0 ∇0
E(r) = iωµ dr0 g(r0 − r) I + · J(r0 ). (1.10.15)
k2
V
It can be shown that

Z Z
dr0 g(r − r0 )∇0 f (r0 ) = ∇ dr0 g(r − r0 )f (r0 ), (1.10.16)
V V
Z Z
dr0 g(r − r0 )∇0 · F(r0 ) = ∇ · dr0 g(r − r0 )F(r0 ), (1.10.17)
V V
by using the vector identities ∇gf = f ∇g + g∇f, ∇ · gF = g∇ · F + (∇g) · F, and that

∇0 g(r − r0 ) = −∇g(r − r0 ). Hence, Equation (1.10.15) can be rewritten as
Z
∇∇
E(r) = iωµ I + 2 · dr0 g(r − r0 )J(r0 ). (1.10.18)
k
V
Sometimes, Equation (1.10.18) is written as

Z
E(r) = iωµ dr0 G(r, r0 ) · J(r0 ), (1.10.19)
V
where
∇∇
G(r, r ) = I + 2 g(r − r0 ),
0
(1.10.20)
k
is a dyad known as the dyadic Green’s function. Equation (1.10.19), with the ∇∇ operator
inside the integration, has to be clarified since it does not converge uniformly when r is also
in the source region occupied by J(r). Hence, it is only a convenient notation when the
observation point is outside the source region.
1.11 Uniqueness Theorem

The uniqueness theorem provides conditions under which the solution to the wave equation
is unique. This is especially important because the solutions to a problem should not be
indeterminate. These conditions under which a solution to a wave equation is unique are the
boundary conditions and the radiation condition. Uniqueness also allows one to construct
solutions by inspections; if a candidate solution satisfies the conditions of uniqueness, it is the
unique solution. Because of its simplicity, the scalar wave equation shall be examined first
for easier insight into this problem.
1.11.1 Scalar Wave Equation

Given a scalar wave equation with a source term on the right-hand side, we shall derive
the conditions under which a solution is unique. First, assume that there are two different
solutions to the scalar wave equation, namely,
[∇2 + k 2 (r)] φ1 (r) = s(r), (1.11.1)
[∇2 + k 2 (r)] φ2 (r) = s(r), (1.11.2)
where k 2 (r) includes inhomogeneities of finite extent. Then, on subtracting the two equations,
we have
[∇2 + k 2 (r)] δφ(r) = 0, (1.11.3)
where δφ(r) = φ1 (r) − φ2 (r). Note that the solution is unique if and only if δφ = 0 for all r.
Then, after multiplying (1.11.3) by δφ∗ , integrating over volume, and using the vector
identity ∇ · ψA = A · ∇ψ + ψ∇ · A, we have
Z Z Z
n̂ · (δφ∗ ∇δφ) dS − |∇δφ|2 dV + k 2 |δφ|2 dV = 0, (1.11.4)
S V V
where n̂ is a unit normal to the surface S. Then, the imaginary part of the above equation is
Z Z
=m n̂ · (δφ∗ ∇δφ) dS + =m(k 2 )|δφ|2 dV = 0. (1.11.5)
S V
2
Hence, if =m[k (r)] 6= 0 in V , and
(i) δφ = 0 or n̂ · ∇δφ = 0 on S,
(ii) δφ = 0 on part of S and n̂ · ∇δφ = 0 on the rest of S, or
2
(iii) δφ + αn̂ · ∇δφ = 0 on S, where α is real,
then the first integral above vanishes, and we have
Z
=m[k 2 (r)]|δφ|2 dV = 0. (1.11.6)
V
Since |δφ|2 is positive definite for δφ 6= 0, and =m(k 2 ) 6= 0 in V ,3 the above is only possible
if δφ = 0 everywhere inside V . Also, in the third case above, α can vary on the surface S. It
can also be chosen so that the first two cases are the special cases of the third case.
Therefore, in order to guarantee uniqueness, so that φ1 = φ2 in V , the above conditions
are equivalent to either
(i) φ1 = φ2 on S or n̂ · ∇φ1 = n̂ · ∇φ2 on S,
(ii) φ1 = φ2 on one part of S, and n̂ · ∇φ1 = n̂ · ∇φ2 on the rest of S, or
(iii) φ1 + αn̂ · ∇φ1 = φ2 + αn̂ · ∇φ2 , where α is real.
The specification of φ on S is also known as the Dirichlet boundary condition, while the spec-
ification of n̂ · ∇φ, namely, the normal derivative, is also known as the Neumann boundary
condition. The third is the reactive impedance boundary condition. In words, the uniqueness
theorem says that if two solutions satisfy the same Dirichlet or Neumann boundary condition
or a mixture thereof on S, or the reactive impedance boundary condition, the two solutions
must be identical.
Notice that the difference solution, δφ satisfies the boundary conditions above (1.11.6)
are all lossless (non-dissipative or non-gain) boundary conditions. When =m[k 2 (r)] 6= 0, and
when such boundary conditions are satisfied by the difference solution, (1.11.6) implies that
only trivial solution δφ = 0 exists. In other words, no time-harmonic difference solution can
exist in such media with loss or gain.
When =m(k 2 ) = 0, i.e., when k 2 is real, the condition δφ = 0 or n̂ · ∇δφ = 0 on S in
(1.11.4) does not necessarily lead to δφ = 0 in V , or uniqueness. The reason is that solutions
for δφ = φ1 − φ2 where
Z Z
|∇δφ|2 dV = k 2 |δφ|2 dV (1.11.7)
V V
2 The author is grateful to J. Mamou for pointing out this case.
3 More specifically, =m[k2 (r)] > 0, ∀ r ∈ V , or =m[k2 (r)] < 0, ∀ r ∈ V .
can exist. These are the resonance solutions in the volume V . These resonance solutions are
the homogeneous solutions4 to the wave Equation (1.11.1) at the real resonance frequencies of
the volume V . Because the medium is lossless, they are time harmonic solutions which satisfies
the boundary conditions, and hence, can be added to the particular solution of (1.11.1). In
fact, the particular solution usually becomes infinite at these resonance frequencies if S(r) 6= 0.
Equation (1.11.7) implies the balance of two energies. In the case of acoustic waves, for
example, it represents the balance of the kinetic energy and the potential energy in a volume
V.
When =m(k 2 ) 6= 0, however, the resonance
solutions
of the volume V are exponentially
2
decaying with time for a lossy medium
=m(k ) > 0 , and they are exponentially growing
with time for an active medium =m(k 2 ) < 0 . But if only time harmonic solutions φ1 and
φ2 are permitted in (1), these resonance solutions are automatically eliminated from the
class of permissible solutions. Hence, for a lossy medium [=m(k 2 ) > 0] or an active medium
[=m(k 2 ) < 0], the uniqueness of the solution is guaranteed if we consider only time harmonic
solutions, namely, two solutions will be identical if they have the same boundary conditions
for φ and n̂ · ∇φ on S.5
When S → ∞ or V → ∞, the number of resonance frequencies of V becomes denser. In
fact, when S → ∞, the resonance frequencies of V become a continuum implying that any
real frequency could be the resonant frequency of V . Hence, if the medium is lossless, the
uniqueness of the solution is not guaranteed at any frequency, even appropriate boundary
conditions on S at infinity, as a result of the presence of the continuum of resonance frequen-
cies. One remedy then is to introduce a small loss. With this small loss [=m(k) > 0], the
solution is either exponentially small when r → ∞ (if a solution corresponds to an outgoing
wave, eikr ), or exponentially large when r → ∞ (if a solution corresponds to an incoming
wave, e−ikr ). Now, if the solution is exponentially small, namely, keeping only the outgoing
wave solutions, it is clear that the surface integral term in (1.11.5) vanishes when S → ∞,
and the uniqueness of the solution is guaranteed. This manner of imposing the outgoing
wave condition at infinity is also known as the Sommerfeld radiation condition [34, p.
188]. This radiation condition can be used in the limit of a vanishing loss for an unbounded
medium to guarantee uniqueness.
The uniqueness of the solution to the Helmholtz wave equation is similar to the uniqueness
of the solution to the the matrix equation
A·x=b (1.11.8)
If a solution to the equation
A · xN = 0 (1.11.9)
exists, then the solution to the first equation is not unique. xN is the null-space solution to
the matrix.
4 “Homogeneous solutions” is a mathematical parlance for solutions to (1.11.1) without the source term.
5 The
nonuniqueness associated with the resonance solution for a lossless medium can be eliminated if we
consider time domain solutions. In the time domain, we can set up an initial value problem in time, e.g.,
by requiring all fields be zero for t < 0; thus, the nonuniqueness problem can be removed via the causality
requirement. The resonance solution, being time harmonic, is noncausal.
The right-hand side of (1.11.8) is the driving term. If the driving term to Helmholtz wave
equation is zero, and yet, a solution exists, it is usually called the resonance solution.6 The
resonance solution is equivalent to the null-space solution in matrix theory.
1.11.2 Vector Wave Equation

Similar to the uniqueness conditions for the scalar wave equation, analogous conditions for the
vector wave equation can also be derived. First, assume that there are two different solutions
to a vector wave Equation, i.e.,
∇ × µ −1 · ∇ × E1 (r) − ω 2 · E1 (r) = S(r), (1.11.10)
∇ × µ −1 · ∇ × E2 (r) − ω 2 · E2 (r) = S(r), (1.11.11)
where S(r) = iω J(r) − ∇ × µ −1 · M(r) corresponds to a source of finite extent. Similarly, µ

and correspond to an inhomogeneity of finite extent. Subtracting (1.11.10) from (1.11.11)
then yields
∇ × µ −1 · ∇ × δE − ω 2 · δE = 0, (1.11.12)
where δE = E1 − E2 . The solution is unique if and only if δE = 0. Next, on multiplying
the above by δE∗ , integrating over volume V , and using the vector identity A · ∇ × B =
−∇ · (A × B) + B · ∇ × A, we have
Z Z
− n̂ · (δE × µ · ∇ × δE) dS + ∇ × δE∗ · µ −1 · ∇ × δE dV
∗ −1
S V
Z
−ω 2 δE∗ · · δE dV = 0.
V
(1.11.13)
Since ∇ × δE = iω µ · δH, the above can be rewritten as
Z Z
∗ 2
iω n̂ · (δE × δH) dS + ω (δH∗ · µ† · δH − δE∗ · · δE) dV = 0. (1.11.14)
S V
Then, taking the imaginary part of (1.11.14) yields

 
 Z 
=m iω n̂ · (δE∗ × δH) dS
 
S
Z
iω 2
− [δH∗ · (µ† − µ) · δH + δE∗ · († − ) · δE] dV = 0. (1.11.15)
2
V
(1.11.16)
6 This is called the homogeneous solution in mathematical parlance. The solution that corresponds to the
driving term on the right-hand side is called the inhomogeneous solution.

But if the medium is not lossless (either lossy or active), then µ† 6= µ and † 6= , and the
second integral in (1.11.16) may not be zero. Moreover, if
(i) n̂ × δE = 0 or n̂ × δH = 0 on S,
(ii) n̂ × δE = 0 on one part of S and n̂ × δH = 0 on the rest of S, or
(iii) δH − iζ n̂ × δE = 0 on S, where ζ is a real number,
then the first integral in (1.11.16) vanishes. The above corresponds to lossless boundary
conditions for the difference field. The third case corresponds to a lossless reactive impedance
boundary condition. 7 Again, ζ can vary on S, and the first two cases can be made special
cases of the third case.
The above implies that,
Z
ω2
[δH∗ · i(µ† − µ) · δH + δE∗ · i(† − ) · δE] dV = 0. (1.11.17)
2
V
In the above, i(µ† − µ) and i(† − ) are Hermitian matrices. Moreover, the integrand will
be positive definite if both µ and are lossy, and the integrand will be negative definite if
both µ and are active. Hence, the only way for (1.11.17) to be satisfied is for δE = 0 and
δH = 0, or that E1 = E2 and H1 = H2 , implying uniqueness.
Consequently, in order for uniqueness to be guaranteed, either
(i) n̂ × E1 = n̂ × E2 on S or n̂ × H1 = n̂ × H2 on S,
(ii) n̂ × E1 = n̂ × E2 on a part of S while n̂ × H1 = n̂ × H2 on the rest of S, or
(iii) H1 − iζ n̂ × E1 = H2 − iζ n̂ × E2 on S.
In other words, if two solutions satisfy the same boundary conditions for tangential E or
tangential H, or a mixture thereof on S, or the same reactive boundary condition, the two
solutions must be identical.
Again, the requirement for a nonlossless condition is to eliminate the real resonance solu-
tions which could otherwise be time harmonic, homogeneous solutions to (1.11.10) satisfying
the boundary conditions. For example, if the appropriate boundary conditions for δE and
δH are imposed so that the first term of (1.11.14) is zero, then
Z
(δH∗ · µ† · δH − δE∗ · · δE) dV = 0. (1.11.18)
V
The above does not imply that δE or δH equals zero, because at resonances, a perfect balance
between the energy stored in the electric field and the energy stored in the magnetic field is
maintained. As a result, the left-hand side of the above could vanish without having δE and
δH be zero, which is necessary for uniqueness. But away from the resonances of the volume
V , the energy stored in the electric field is not equal to that stored in the magnetic field.
7A more complicated boundary condition for the third case may be designed.
Hence, in order for (1.11.18) to be satisfied, δE and δH have to be zero since each term in
(1.11.18) is positive definite for lossless media due to the Hermitian nature of µ and .
When V → ∞, as in the scalar wave equation case, some loss has to be imposed to guar-
antee uniqueness. This is the same as requiring the wave to be outgoing at infinity, namely,
the radiation condition. Again, the radiation condition can be imposed for an unbounded
medium with vanishing loss to guarantee uniqueness.
1.12 Transformation Matrices for Microwave Circuits

1.12.1 Impedance and Admittance Matrices
A general microwave circuit consists of many ports. A convenient way to characterize an
N -port network is to describe the network in terms of impedance matrices or admittance
matrices [5]. For example, if the N -port network can be characterized by a pair of voltage
and current at each port, then a column vector of voltages can be defined and also a column
vector of currents. We can write down a relationship between the voltages and the currents
as     
V1 Z11 Z12 · · · Z1N I1
 V2   Z21 Z22 · · · Z2N   I2 
    
V= . = . .. .. ..   ..  = Z · I (1.12.1)
 ..   .. . . .  . 
VN ZN 1 ZN 2 ··· ZN N IN
By the same token, we can express
I=Y·V (1.12.2)
where Y is the admittance matrix. For reciprocal circuits, it can be shown that Z and Y
are symmetric matrices. For lossless circuits, it can be shown that these matrices have pure
imaginary elements.
For a two port network which is reciprocal, there are three independent matrix elements.
Therefore, a two-port network can often be modeled by a T or a Π equivalent circuit.
Z 11 – Z 12 Z22 – Z12 –Y12
1 Z12 2 1 Y11 + Y12 Y22 + Y12 2
Figure 1.7: The T and Π equivalent circuits for a two-port circuit.
1.12.2 Scattering Matrices

For high frequencies, it is more pertinent to think about waves. Then at each port, we can
define an incident and a reflected wave. For instance, we can define an incident voltage wave
V + and a reflected voltage wave V − . A relationship can then be written between the reflected
waves at all the ports to the incident waves at all the ports.
 −    +
V1 S11 S12 · · · S1N V1
V2   S21 S22 · · · S2N  V2+ 
−
     +
V= . = . .. . ..   ..  = S · V (1.12.3)
.
 .   . . . . . .  . 
VN− SN 1 SN 2 · · · SN N VN+
It can be proved that S has to be symmetric for reciprocal circuits, and that it has to be
unitary if the circuit is lossless.
1.12.3 Chain Matrices

When one needs to cascade a series of two port networks, it is more convenient to work with
chain matrices or transmission matrices. A voltage and current transmission matrix relates
the voltage and current at one port to the voltage and current at the second port.
I1 I2
V1 V2
Figure 1.8: A diagram for defining the voltage and current transmission matrix.
Written explicitly, we have

V1 A1 B1 V2
= (1.12.4)
I1 C1 D1 I2
Notice that the current at Port 2 is flowing out of the port rather than into the port. In this
manner, if we have a second transmission matrix of a second network that relates V2 , I2 to
V3 , I3 , viz.,
V2 A2 B2 V3
= (1.12.5)
I2 C2 D2 I3
Hence, when these two networks are cascaded together, the resultant transmission matrix is
the product of the two matrices

V2 A1 B1 A2 B2 V3
= (1.12.6)
I2 C1 D1 C2 D2 I3
For a two port network, it can be shown that

2
A = Z11 /Z12 , B = (Z11 Z22 − Z12 )/Z12 , (1.12.7a)
C = 1/Z12 , D = Z22 /Z12 , (1.12.7b)

for a reciprocal network. It is also readily verified that
AD − BC = 1 (1.12.8)
for this case. Hence, the determinant of a chain matrix is always 1 for a reciprocal network.
Exercises for Chapter 1
Problem 1-1: The fundamental units in electromagnetics can be considered to be meter,

kilogram, second, and coulomb.
(a) Show that 1 volt, which is 1 watt/amp, has the dimension of (kilogram meter2 )/(coulomb
sec2 ).
(b) From Maxwell-Heaviside equations, show that µ0 has the dimension of (second volt)/(meter
amp), and hence, its dimension is (kilogram meter)/coulomb2 in the more fundamental
units.
(c) If we assign the value of µ0 to be 4π instead of 4π × 10−7 , what would be the unit of
coulomb in this new assignment compared to the old unit? What would be the present
value of 1 volt and 1 amp in this new assignment?
Problem 1-2: Show that for two time-harmonic functions,
1
hA(r, t)B(r, t)i = <e[A(r)B ∗ (r)], (1.12.9)
2
where A(r) and B(r) are the phasors of A(r, t) and B(r, t).The angular brackets above imply
time averaging.
Problem 1-3: Assume that a voltage is time harmonic, i.e., V (t) = V0 cos ωt, and that a
current I(t) = II cos ωt + IQ sin ωt, i.e., it consist of an in-phase and a quadrature component.
(a) Find the instantaneous power due to this voltage and current, viz.,V (t)I(t).
(b) Find the phaser representations of the voltage and current, and hence the complex
power due to this voltage and current.
(c) Establish a relationship between the real part and reactive part of the complex power
to the instantaneous power.
(d) Show that the reactive power is due to the quadrature component of the current, which
is related to a time-varying part of the instantaneous power with zero-time average.
Problem 1-4: For a scalar-wave equation, ∇ · −1 (r)∇φ(r) + k 2 φ(r) = S(r):
(a) Show that a reciprocal relationship hφ1 (r), S2 (r)i = hφ2 (r), S1 (r)i exists.
(b) What is the boundary condition satisfied by φ at an interface where (r) has a step
discontinuity?
Problem 1-5:
(a) Prove that for reciprocal circuits, the impedance matrix and the admittance matrix are
symmetric.
(b) Prove that for lossless circuits, the impedance matrix and the admittance matrix have
imaginary elements.
Problem 1-6:
(a) Prove that for reciprocal circuits, the scattering matrix is symmetric.
(b) Prove that for lossless circuits, the scattering matrix is unitary.
(c) Prove that for reciprocal circuits, the determinant of the chain matrix is always equal
to one.
Bibliography
[1] R.E. Collin, Field Theory of Guided Waves, IEEE Press, Piscataway, NJ, 1991.
[2] R. Mittra and S.W. Lee, Analytical Techniques in the Theory of Guided Waves, The
MacMillan Company, New York, 1971.
[3] L. Levin, Theory of Waveguides: Techniques for the Solution of Waveguide Problems,
Newnes-Butterworth, London, 1975.
[4] N. Marcuvitz, ed., Waveguide Handbook, MIT Radiation Laboratory Series, vol, 10,
McGraw-Hill, New York, 1951.
[5] R.E. Collin, Foundation for Microwave Engineering, IEEE Press, Piscataway, NJ, 2001.
[6] J. W. Strutt Rayleigh (Lord Rayleigh), Theory of Sound, New York: Dover Publ.,
1976. (Originally published 1877.)
[7] J. W. Strutt Rayleigh (Lord Rayleigh), “On the passage of electric waves through tubes,
or the vibra cylinder,” Phi. Mag., vol. 43, pp. 125–132, 1897.
[8] J. Hecht, City of Light: The Story of Fiber Optics, Oxford University Press, Oxford, U.K.,
1999.
[9] A.A. Oliner, “Leakage from higher modes on microstrip line with application to antennas,”
Radio Sci., 22(6), pp. 907-912, 1987.
[10] Encyclopaedia Britanica, Encyclopaedia Britanica Inc., 2004.
[11] R. Feynman, R.B. Leighton, and M.L. Sands, The Feynman Lectures on Physics, vol. I,
Chapter 52, Addison-Wesley Publishing Co., 1965.
[12] M. Faraday, “On static electrical inductive action,” Phil. Mag., 1843. M. Faraday, Exper-
imental Researches in Electricity and Magnetism. Vol. 1, Taylor & Francis, London, 1839.;
Vol. 2, Richard & John E. Taylor, London, 1844; Vol. 3, Taylor and Francis, London, 1855.
Reprinted by Dover in 1965. Also see M. Faraday, ”Remarks on Static Induction,” Proc.
Roy. Inst., Feb. 12, 1858.
[13] A. M. Ampère, “Mémoire sur la théorie des phénomènes électrodynamiques,” Mem.
Acad. R. Sci. Inst. Fr., 6, 228-232, 1823.
31
[14] C. S. Gillmore, Charles Augustin Coulomb: Physics and Engineering in Eighteenth Cen-
tury Frrance, Princeton, NJ, 1971.
[15] C. F. Gauss, “General theory of terrestrial magnetism,” Scientific Memoirs, vol. 2, ed.
R. Taylor (R & J.E. Taylor, London), pp. 184-251, 1841.
[16] J. C. Maxwell, A Treatise of Electricity and Magnetism, 2 vols, Clarendon Press, Oxford,
1873. Also, see P. M. Harman (ed.), The Scientific Letters and Papers of James Clerk
Maxwell, Vol. II, 1862-1873, Cambridge, U.K.: Cambridge University Press, 1995.
[17] O. Heaviside, “On electromagnetic waves, especially in relation to the vorticity of the
impressed forces, and the forced vibration of electromagnetic systems,” Phil. Mag., 25,
130-156, 1888. Also, see P. J. Nahin, “Oliver Heaviside,” Scientific American, pp. 122-129,
June 1990.
[18] Nobel Lectures, Physics 1901-1921, Elsevier Publishing Company, Amsterdam, 1967.
[19] J. Glenn, ed., The Complete Patents of Nikola Tesla, New York: Barnes and Noble
Books, 1994.
[20] W. K. Tung, Group Theory in Physics, Philadelphia, PA: World Scientific Publ., 1985.
[21] L. M. Falicov, Group Theory and its Physical Applications, Chicago: University of
Chicago Press, 1966.
[22] W. C. Chew, J. M. Jin, E. Michielssen, and J. M. Song, eds., Fast and Efficient Algo-
rithms in Computational Electromagnetics, Artech House, Boston, MA, 2001.
[23] J.D. Jackson, Classical Electrodynamics, Third Edition, John Wiley & Sons, Inc., NJ,
1999.
[24] T.D. Lee and C.N. Yang, “Question of parity conservation in weak interaction,” Phys.
Rev., 104(1), pp. 254-257, 1956.
[25] G. Green, An Essay on the Application of Mathematical Analysis to the Theories of

Electricity and Magnetism, T. Wheelhouse, Nottingham, 1828. Also, see L. Challis and
F. Sheard, “The green of Green functions,” Physics Today, pp. 41-46, December 2003.
[26] W.C. Chew, Waves and Fields in Inhomogeneous Media, Van Nostrand Reinhold, New
York, 1990, reprinted, IEEE Press, Piscataway, NJ, 1995.
[27] K.T. Chau, W.L. Li and C.H.T. Lee, “Challenges and opportunities of electric machines
for renewable energy,” (invited paper), Prog. in Electromag. Research B, vol. 42, 45-74,
2012.
[28] L. Novotny and B. Hecht, Principles of Nano-Optics, Cambridge University Press, 2006.
[29] S. Datta, Electronic Transport in Mesoscopic Systems, Cambridge University Press, 1995,
Paperback Edition 1997.
[30] H.A. Haus, Electromagnetic Noise and Quantum Optical Measurements, Springer-Verlag,
Berlin, 2000.
[31] A. Yariv, “Phase conjugate optics and real-time holography,” IEEE J. Quantum Elec-
tron., QE-14, p. 117, 1978.
[32] G.S. Agarwal, A.T. Friberg, and E. Wolf, “Elimination of distortions by phase conjuga-
tion without losses or gains,” Opt. Commun., 43, p. 446, 1982.
[33] W.C. Chew and T.M. Habashy, “Phase-conjugate mirror and time reversal,” J. Opt. Soc.
Am., vol. 2, no. 6, p. 808, 1985.
[34] A. Sommerfeld, Partial Differential Equation. New York: Academic Press, 1949.
Chapter 2
Hollow Waveguides
Much work has been written on hollow waveguides [1–7]. Theory on guided waves dates back
even to earlier dates as alluded to in Chapter 1. Hollow waveguides, when filled with air, allow
for high power microwave transmission. However, a hollow electromagnetic waveguide, unlike
an acoustic waveguide, has a lower cutoff frequency of operation. The way to counter this
is to use a two-conductor waveguide such as a coaxial waveguide. The support of the inner
conductor in such waveguides requires the filling of the waveguide with dielectric materials.
It will be shown that when a waveguide is homogeneously filled with materials, the theory is
essentially the same as that for one which is hollow.
2.1 General Uniform Cylindrical Waveguides

General uniform cylindrical waveguides include transmission lines, and hollow waveguides,
as well as multi-conductor waveguides. If the regions between the conductors are filled with
a homogeneous material, the waveguide can support purely transverse electric (TE), trans-
verse magnetic (TM) or transverse electromagnetic (TEM) modes. If the wave is TEM, the
waveguide is operating in the transmission line mode. It can be shown that a hollow waveg-
uide cannot support a TEM wave. Therefore, a transmission line needs to have at least two
conductors.
Because the fields of a general uniform cylindrical waveguide can be decomposed into TE
and TM types, we can characterize the TE wave with the ẑ-component of the magnetic field
or Hz , since Hz 6= 0 for this type of wave. Similarly, we can characterize the TM wave with
the ẑ-component of the magnetic field or Ez since Ez 6= 0 for this type of wave.
We shall derive the equations governing the Ez and Hz components of the fields in a
general, uniform cylindrical waveguide filled with a homogeneous material. We shall call such
waveguides hollow waveguides. We let [6]
E = Es + ẑEz , H = Hs + ẑHz , (2.1.1)
where the subscript s represents the transverse to z components. Substituting the above
35
Figure 2.1: Cross-section of a general uniform cylindrical waveguide.
equations into Maxwell’s equations, we have

∂
∇s + ẑ × (Es + ẑEz ) = iωµ(Hs + ẑHz ), (2.1.2)
∂z

∂
∇s + ẑ × (Hs + ẑHz ) = −iω(Es + ẑEz ). (2.1.3)
∂z
Equating the s components of (2.1.2) and (2.1.3), we have
∂
∇s × ẑEz + ẑ × Es = iωµHs , (2.1.4)
∂z
∂
∇s × ẑHz + ẑ × Hs = −iωEs . (2.1.5)
∂z
Substituting for Es from (2.1.5) into (2.1.4), we have

∂ ∂
ω 2 µHs = −iω∇s × ẑEz + ẑ × ∇s × ẑHz + ẑ × Hs . (2.1.6)
∂z ∂z
Using the vector identities ẑ × ∇s × ẑ = ∇s , ẑ × ẑ × Hs = −Hs , and assuming that the

∂2
field has e±ikz z dependence, so that ∂z 2
2 → −kz , we can rewrite (2.1.6) as

1 ∂
Hs = 2 ∇s Hz + iωẑ × ∇s Ez , (2.1.7)
k − kz2 ∂z
Hollow Waveguides 37
where k 2 = ω 2 µ. By the same token, we have

1 ∂
Es = ∇ E
s z − iωµẑ × ∇ s z .
H (2.1.8)
k 2 − kz2 ∂z
Equations (2.1.7) and (2.1.8) allow us to derive all the other components of the field in a
waveguide once we know the z components of the field.
If we equate the z components of (2.1.2) and (2.1.3), we have
∇s × Es = iωµHz , (2.1.9)
∇s × Hs = −iωEz . (2.1.10)
Substituting (2.1.7) and (2.1.8) into (2.1.9) and (2.1.10), we have
(∇2s + ks2 )Hz = 0, for TE waves, (2.1.11)
(∇2s + ks2 )Ez = 0, for TM waves, (2.1.12)
where ks2 = k 2 − kz2 . Therefore, Hz and Ez satisfy a two-dimensional scalar wave equation
also known as the reduced wave equation. Once Ez and Hz are solved for from (2.1.11) and
(2.1.12), we can derive all the other field components using (2.1.7) and (2.1.8).
In the above, (2.1.11) and (2.1.12) clearly show that Hz and Ez satisfy the wave equation
independently. The Ez component satisfies the homogeneous Dirichlet boundary condition,
Ez = 0 on the waveguide wall. The general boundary condition for the TE field is that
n̂ × Es = 0 on the waveguide wall. From (2.1.8), we can show that this is equivalent to
n̂ · ∇Hz = 0 on the waveguide wall, which is the homogeneous Neumann boundary condition.
The boundary condition does not couple the TE and TM waves implying that they can exist
independently of each other. This substantiates our assumption in the very beginning.
It is to be noted that an alternative way of solving the above problem is to define two
scalar potentials ψh and ψe , and let
ET E = ∇ × ẑψh (2.1.13)
TM
H = ∇ × ẑψe (2.1.14)
It can be easily shown that Hz from the TE field is proportional to ψh and Ez from the TM
field is proportional to ψe . Hence, this method of solution is completely equivalent to our
previous method of solution.
We can envision that a wave is guided in a waveguide because the wave is bouncing around
the wall of the waveguide. This is the bouncing wave picture of the wave. In this case, the k
vector is not pointing in the z direction completely, and kz < k. The mode is either TE or TM.
However, it is also possible to have a wave directly transmitted through a waveguide, such as
in the transmission line. In this case, the k vector is pointing entirely in the z direction, and
kz = k. We shall study this mode next.
2.2 Transmission Line Theory

The propagation of waves on the transmission line was first formulated in terms of telegra-
pher’s equations. Telegraphy was in use in the early 1800s even before the completion of
Maxwell’s equations in 1864. The telegrahers equations can be derived using circuit theory
(see Fig. 2.2), and they are valid even for meandering lines. The transmission line can be
thought of as consisting of a sequence of coupled L and C tank circuits. Each tank circuit
forms a simple harmonic oscillator. But on a transmission line, these harmonic oscillators are
coupled together. It is through the coupling of these harmonic oscillators that a wave can
propagate on a transmission line.
Figure 2.2: Two pieces of parallel wires form a transmission line, even when they are sinuous
in nature. A wire has inductance, and capacitance exists between two pieces of metal. Hence,
circuit model used to derive the telegrahers equations.
2.2.1 TEM Mode of a Transmission Line

In a transmission line, a TEM (transverse electromagnetic) mode can propagate. For TEM
waves, both Ez and Hz are zero. By looking at (2.1.7) and (2.1.8), Hs and Es will be non-
zero only if kz = k. Therefore, all TEM waves, or TEM modes in a waveguide have eikz
dependence. Furthermore, from (2.1.9) and (2.1.10), we conclude that for TEM waves
∇s × Es = 0, ∇s × Hs = 0. (2.2.1)
In other words, Es is electrostatic in the xy plane while Hs is magnetostatic in the xy plane.

Hence, we can let
Es = −∇s φs (x, y)eikz , Hs = −∇s ψs (x, y)eikz . (2.2.2)

Since ∇ · Es = 0 and ∇ · Hs = 0 inside the waveguide, φs and ψs satisfy Laplace equations
∇2s φs (x, y) = 0, ∇2s ψs (x, y) = 0. (2.2.3)
If we have perfect electric conductors, the boundary conditions are n̂ × Es = 0, and n̂ · Hs = 0

on the metallic surfaces. These boundary conditions translate to
φs = constant, (2.2.4)
∂
n̂ · ∇s ψs = ψs = 0. (2.2.5)
∂n
Equation (2.2.4) is known as the Dirichlet boundary condition while (2.2.5) is the Neumann
boundary condition. The constants in (2.2.4) are the potentials on the conductors which
may be different for different conductors. ψs is a multi-value function because magnetic field
always goes in a loop and ends on itself. In order to avoid dealing with multi-value functions,
it is a lot easier to solve for φs . Alternatively, one can solve for the magnetic field using a
vector potential.
Es
J1
φs = V1
Hs
⊗ J2
φs = V2
Figure 2.3: TEM mode in a transmission line.
At this point, it seems that φs and ψs are decoupled, and hence, the electric field and
the magnetic field are independent of each other. This could not be true, as the coupling is
expressed in Equations (2.1.4) and (2.1.5) (if we set Ez = Hz = 0 for discussing TEM modes).
Hence, the coupling of the fields is only through the ẑ-variation of the fields, i.e.,
∂
ẑ × Es = iωµHs , (2.2.6)
∂z
∂
ẑ × Hs = −iωEs . (2.2.7)
∂z
Furthermore, from the above, we deduce that Hs and Es are mutually orthogonal in the TEM
mode. Since the fields have eikz dependence, we conclude that
r
µ
ẑ × Es = Hs = ηHs , (2.2.8)

if we assume that the wave is only propagating in one direction. Then η is also known as the
intrinsic impedance of the medium, and all TEM waves satisfy (2.2.8).
In (2.2.3), the electrostatic and the magnetostatic problems are seemingly decoupled from
each other, but the fields are coupled via their z variation, as indicated by the above equations.
In (2.2.8), it says that once we know the electrostatic field, the magnetostatic field can be
easily derived. Hence, we need only to solve the electrostatic problem to fully characterize
the TEM solution.
Derivation of the Telegrapher Equations

We can integrate Equation (2.2.6) about a line contour around one of the conductors to obtain
I I
∂
ẑ × Es · dl = iωµ Hs · dl. (2.2.9)
∂z C C
H
By Ampere’s law, we have C Hs · dl = I, the total current on one of the conductors. For the
left-hand side, we have
I I
ẑ × Es · dl = dl × ẑ · Es
C C (2.2.10)
Q
= ,

Since dl × ẑ is an outward normal to C and Gauss’ theorem can be invoked. Therefore,
Equation (2.2.9) becomes
d
Q = iωµI. (2.2.11)
dz
Since the transverse field is purely static, we can define Q = CV where C is the capacitance
per unit length and V = V1 − V2 . Hence, (2.2.11) becomes
d µ
V = iω I. (2.2.12)
dz C
Since µ/C has the dimension of henry per meter, we can define L = µ/C, an inductance
per unit length, and (2.2.12) becomes
dV
= iωLI. (2.2.13)
dz
With similar manipulations to ẑ × (2.2.7), we obtain
dI
= iωCV. (2.2.14)
dz
C
S
I
C1
⊗
C2
Figure 2.4: Derivation of the telegraphers equations for a transmission line.
Equations (2.2.13) and (2.2.14) are the telegraphers equations (also known as telegraphists
equations) for a transmission line, which can also be derived from a circuits viewpoint. The
velocity of the wave on the line is given by
1 1
v=√ =√ . (2.2.15)
µ LC
Notice that in the above, we need only to solve the electrostatic problem to obtain C, and
L can be derived from C. There is no need to solve the magnetostatic problem.
Characteristic Impedance versus Intrinsic Impedance

√
Since the fields, and hence the voltage and current, have eikz dependence,where k = ω µ =
√
ω LC, we deduce either from (2.2.13) or (2.2.14) that
r
V L
= = Z0 , (2.2.16)
I C
if the wave is only propagating in the positive z direction. Z0 is also known as the charac-
teristic impedance of a transmission line. Since C and L are dependent on the geometry of
the transmission line, Z0 is a geometry dependent impedance. This is unlike η, the intrinsic
impedance. It can be easily shown from (2.2.13) and (2.2.14) that
d2 V
+ ω 2 LCV = 0, (2.2.17)
dz 2
d2 I
+ ω 2 LCI = 0, (2.2.18)
dz 2
which are one-dimensional scalar wave equations (or Helmholtz wave equations).
Energy Density and Power Flow

The time average energy stored per unit length in a transmission line for a single propagating
wave is given by Z Z
1 1
hWe i = Es · E∗s dS = (∇s φs )2 dS, (2.2.19)
4 S 4 S
Z Z
1 1
hWm i = µ Hs · H∗s dS = µ (∇s ψs )2 dS, (2.2.20)
4 S 4 S
where hWe i and hWm i are the time average energy stored in the electric field and the magnetic
field respectively. Using the fact that ∇ · (φ∇φ) = (∇φ)2 + φ∇2 φ, we can write
I
1 ∂ 1 1
hWe i = φ φdl = (V1 − V2 )Q = CV 2 , (2.2.21)
4 C1 +C2 ∂n 4 4
V1 C1
V2 C2
Figure 2.5: Power flow in a transmission line.
where V = V1 − V2 , Q =p CV . The above could also be derived from circuit theory. As

we have shown before, |Es | = µ Hs |, hence, the time average energy stored in the magnetic
field is
1
hWm i = hWe i = CV 2 . (2.2.22)
4
L 2
Since V 2 = Z02 I 2 = CI , and µ = LC, we can also write
1 2
hWm i = LI . (2.2.23)
4
Equation (2.2.23) can also be established by circuit theory. It also establishes our definition
of L as an inductance per unit length.
The time average power flow down a transmission line is given by

Z
1
hP i = <e dS ẑ · (Es × H∗s ). (2.2.24)
2 S
q
Since Hs = µ ẑ × Es from (2.2.18), we have
r Z
1
hP i = dS|Es |2 = 2vhWe i = vhWe + Wm i. (2.2.25)
2 µ S
Hence, the time average stored energy hWe + Wm i moving at velocity v, contributes to power
flow. Equations (2.2.13) and (2.2.14) can also be derived from a circuit model.
2.2.2 Lossy Transmission Lines

Since the telegraphists equations have strictly circuit theory interpretation, using circuit the-
ory concept, the extension to a lossy transmission line is straight forward: we replace the
series impedance per unit length −iωL by −iωL + R, and the shunt admittance per unit
length −iωC by −iωC + G where R is the series resistance per unit length in the conductor,
while G is the shunt conductance per unit length in the insulator. The telegraphists equations
then become
Figure 2.6: Circuits equivalence of a transmission line.
dV
= (iωL − R)I = −ZI, (2.2.26)
dz
dI
= (iωC − G)V = −Y V. (2.2.27)
dz
The characteristic impedance is now
r
Z
Z0 = , (2.2.28)
Y
while the propagation constant becomes
s
2 iR iG
k = −(iωL − R)(iωC − G), k=ω LC 1 + 1+ . (2.2.29)
ωL ωC
Hence k is complex and the wave eikz is attenuating.

Strictly speaking, when we have a lossy transmission line due to conductor loss, a pure
TEM wave cannot exist. This is because the axial current flow meets a resistance, and hence,
an axial component of the electric field is necessary now to drive a current in the conductor.
Therefore, the field is only quasi-TEM. However, the conductor loss can be thought of as a
small perturbation of the perfect conductor case, and the electromagnetic field in the lossy
line will not be vastly different from that of a lossless case.
The shunt conductance G in a lossy line can be found as follows. If the capacitance per
unit length between two conductors is given by the formula
C = K, (2.2.30)
where K is a geometry dependent factor, the shunt admittance would be given by Y =

−iωC = −iωK. If now the dielectric medium is lossy so that = 0 + iσ
ω , then the shunt
admittance is given by
Y = −iω0 K + σK. (2.2.31)
Hence, we identify G = σK. Note that the derivations in (2.2.6) to (2.2.16) hold true even if
is complex. For this reason, Y in (2.2.31) is exact.
The series resistance R can be found by calculating the resistance of the conductor in a
perturbative manner when it is lossy. The skin-effect will confine the current toqflow only
2
on the surface of the conductor. Since the skin depth in a conductor is δ = ωµσ , the
current is confined to flow in a thinner region at higher frequencies, hence, increasing this
series resistance.
Another way of calculating transmission line loss is via a perturbation argument and the
use of energy conservation. If a transmission line is lossy such that k = k 0 + ik 00 , and
0 00
V, I ∼ eik z−k z , (2.2.32)
Then, the power flow in a line, which is proportional to |V |2 or |I|2 is

00
P ∼ e−2k z . (2.2.33)
By energy conservation,
dP
= −Pd = −2k 00 P, (2.2.34)
dz
where Pd is the power dissipated per unit length on the line. Therefore, the attenuation
constant k 00 can be derived to be
Pd
k 00 = , (2.2.35)
2P
if we know Pd . We can assume P to be close to that of a lossless line in using (2.2.35) in a
perturbative concept.
Hs
Figure 2.7: Absence of TEM mode in a hollow, enclosed waveguide.
Absence of TEM Mode in a Hollow Waveguide

Before ending this section, we would like to prove by contradiction that a hollow waveguide
(i.e. without a center conductor) cannot support a TEM mode as follows. If we assume that
it does, then Hthe magnetic field has to end on itself due to the absence of magnetic charges. It
is clear that C Hs · dl 6= 0 about any closed contour following the magnetic field H lines. This
is clearly in violation of Equation (2.2.1) for a TEM mode which implies that Hs · dl = 0
if C does not enclose any conducting current. These two results are contradictory implying
the absence of a TEM mode in a hollow waveguide.
2.3 TE and TM Modes (H and E Modes)

2.3.1 Mode Orthogonality
Mode Orthogonality for Axial Fields
For TE and TM waves in a cylindrical waveguide, we characterize the waves by Hz and
Ez respectively. The equations governing these two scalar field components are given by
(2.1.11) and (2.1.12). In order to obtain a unique solution to (2.1.11) and (2.1.12), we have
to specify the boundary conditions for Hz and Ez . For a metallic waveguide, we need to
impose the boundary condition that n̂ × E = 0 on the metallic surface. This requires Ez = 0
on the metallic surface. From Equation (2.1.8), we see that if Ez = 0, and furthermore, if
∂
n̂ · ∇s Hz = ∂n Hz = 0, then n̂ × E = 0 on the metal surface. Hence, the equations governing
the TE and TM modes are
∂
(∇2s + ks2 )Hz = 0, Hz = 0 on S, TE, (2.3.1)
∂n
(∇2s + ks2 )Ez = 0, Ez = 0 on S, TM, (2.3.2)
where S is the metallic surface. The homogeneous boundary conditions in Equation (2.3.1)
is the Neumann boundary condition, while that for Equation (2.3.2) is the homogeneous
Dirichlet boundary condition. They are sufficient to uniquely determine the solutions to the
partial differential equations.
For a closed waveguide, solutions exist for (2.3.1) and (2.3.2) at only discrete values of
ks2 . Furthermore, ks2 is purely real because (2.3.1) and (2.3.2) are self-adjoint problems (see
Problem 2.6). This property is independent of the homogeneous material filling the waveguide,
and hence is true even for a lossy isotropic material. It can be shown easily that different
2
solutions of (2.3.1) or (2.3.2) corresponding to different kis are orthogonal, i.e.,
Z
Hiz Hjz dS = Cδij , TE, (2.3.3)
S
Z
Eiz Ejz dS = Cδij , TM, (2.3.4)
S
where the integration is over S, the cross-section of the waveguide. We can prove the above
assertion quite easily, e.g., by writing down
(∇2s + kis
2
)ψiz = 0, (2.3.5)
(∇2s + kjs
2
)ψjz = 0, (2.3.6)
where ψz in this case can be either Hz or Ez . Multiplying the first equation by ψjz and the
second equation by ψiz , subtracting the two equations, and integrating over S, we have
Z Z
2 2
(kis − kjs ) ψiz ψjz dS = (ψiz ∇2s ψjz − ψjz ∇2s ψiz )dS,
S S
I
Gauss’ theorem −→= n̂ · (ψiz ∇s ψjz − ψjz ∇s ψiz )dl. (2.3.7)
C
The above usage of Gauss’ theorem is also known as Green’s theorem. With either a homo-
geneous Neumann, Dirichlet or mixed boundary condition (Neumann on one part of C, and
Dirichlet on the other parts), the right-hand side of (2.3.7) vanishes, and we have
Z
2 2
(kis − kjs ) ψiz ψjz dS = 0. (2.3.8)
S
For i 6= j, we have Z
ψiz ψjz dS = 0. (2.3.9)
S
2 ∗
Since kjscan be shown to be pure
R real [see Problem 2-6], ψjz is also a solution of (2.3.6).
∗ ∗
Hence, we can further say that ψiz ψjz dS = 0, i 6= j, since ψjz is also a solution to (2.3.6).
S
The property described by (2.3.3) and (2.3.4), is known as the mode orthogonality of the
axial components of the field.
Mode Orthogonality for Transverse Fields

The transverse components of E or H fields are also orthogonal. This property can be proven
a number of ways. One way is to relate their orthogonality to the orthogonality of the scalar
wave functions mentioned before. However, to demonstrate that the orthogonality is also
related to the symmetry of the differential equation, which will prove this property as follows:
the E-field of either the TE or the TM mode of a waveguide satisfies
∇ × ∇ × Ei − k 2 Ei = 0. (2.3.10)
Assuming that the field has eikz z dependence, and extracting the transverse component of
the above equation, we have, for any two distinct modes,
2
∇s × ∇s × Eis − ∇s ∇s · Eis − kis Eis = 0, (2.3.11)
2
∇s × ∇s × Ejs − ∇s ∇s · Ejs − kjs Ejs = 0, (2.3.12)
2
where kis = k 2 − kiz
2 2
, kjs = k 2 − kjz
2
. Dot multiplying (2.3.11) by Ejs , (2.3.12) by Eis , and
subtracting, we have, after integrating,
Z
2 2
(kjs − kis ) Eis · Ejs dS
S
Z
= dS(Ejs · ∇s × ∇s × Eis − Eis · ∇s × ∇s × Ejs ) (2.3.13)
S
Z
− dS(Ejs · ∇s ∇s · Eis − Eis · ∇s ∇s · Ejs ).
S
By noting that ∇ · (A × B) = B · ∇ × A − A · ∇ × B, and hence,
−∇s · (Ejs × ∇s × Eis − Eis × ∇s × Ejs )

(2.3.14)
= Ejs · ∇s × ∇s × Eis − Eis · ∇s × ∇s × Ejs ,
and that ∇ · (Aφ) = φ∇ · A + A · ∇φ, and hence,
∇s · (Ejs ∇s · Eis − Eis ∇s · Ejs ) = Ejs · ∇s ∇s · Eis − Eis · ∇s ∇s · Ejs , (2.3.15)
we can convert the right-hand side of (2.3.13) into line integrals using Gauss’ theorem in two
dimensions, giving
Z
2 2
(kjs − kis ) Eis · Ejs dS
S
I
=− dl n̂ · [Ejs × (∇s × Eis ) − Eis × (∇s × Ejs )] (2.3.16)
C
I
− dl n̂ · [Ejs ∇s · Eis − Eis ∇s · Ejs ].
C
The right-hand side of (2.3.16) vanishes by virtue of the boundary condition. Since n̂ · [Ejs ×
(∇s × Eis )] = n̂ × Ejs · ∇s × Eis ,1 n̂ × Es = 0 implies the zero of the first integral on the right
hand size of (2.3.16). Furthermore, ∇s · Eis = −ikz Ez = 0 on the waveguide wall implies the
zero of the second term in the right-hand side of (2.3.16). Therefore,
Z
Eis · Ejs dS = 0, i 6= j, (2.3.17)
S
for any two distinct modes with different ks2 , irrespective of whether they are TE or TM
modes. A similar proof follows for Hs , i.e.,
Z
His · Hjs dS = 0, i 6= j. (2.3.18)
S
We can further show that Z

Fis · F∗js dS = 0, i 6= j, (2.3.19)
S
where F is either E or H. This is because E∗ or H∗ is also a solution of (2.3.11).

The orthogonality of modes in (2.3.17) follows from the fact that the differential operator
in (2.3.11) is symmetric with the defined boundary conditions, i.e.,
hEjs , (∇s × ∇s × − ∇s ∇s ·)Eis i = hEis , (∇s × ∇s × − ∇s ∇s ·)Ejs i. (2.3.20)
It is analogous to the fact that eigenvectors of a symmetric matrix with distinct eigenvalues
are orthogonal. When the medium is inhomogeneous, (2.3.11) is not valid for describing the
field, and the differential operators are not symmetric anymore. For an inhomogeneously
filled waveguide, (2.3.17) and (2.3.18) are not true in general.
Mode Orthogonality for Reaction

A more general orthogonality condition which we shall derive later, and is true even for
inhomogeneously filled waveguides is
Z
(Eis × Hjs ) · ẑdS = 0, i 6= j. (2.3.21)
S
The above is the reaction as is used in the Lorentz reciprocity theorem. To prove (2.3.21) for
hollow waveguides, we first write down the equations satisfied by Eis and Hjs , i.e.,
2
∇s × ∇s × Eis − ∇s ∇s · Eis − kis Eis = 0, (2.3.22)
2
∇s × ∇s × Hjs − ∇s ∇s · Hjs − kjs Hjs = 0. (2.3.23)
1 This follows from a · b × c = c · a × b = b · c × a.
We cross multiply (2.3.22) by Hjs and (2.3.23) by Eis . Upon subtraction and integration, we
have
Z
2 2
(kjs − kis ) (Eis × Hjs ) · ẑdS
S
Z
= [(∇s × ∇s × Eis ) × Hjs − Eis × (∇s × ∇s × Hjs )] · ẑdS (2.3.24)
S
Z
− [(∇s ∇s · Eis ) × Hjs − Eis × (∇s ∇s · Hjs )] · ẑdS.
S
Using the fact that
(∇s × ∇s × Es ) × Hs · ẑ = (Hs × ẑ) · (∇s × ∇s × Es )

= ∇s · [(∇s × Es ) × (Hs × ẑ)] (2.3.25)
+ (∇s × Es ) · ∇s × (Hs × ẑ)
and that
ẑ · Es × ∇s ∇s · Hs = (ẑ × Es ) · ∇s ∇s · Hs
(2.3.26)
= ∇s · (∇s · Hs ẑ × Es ) − ∇s · Hs ∇s · ẑ × Es ,
plus the fact that ∇s × (Hs × ẑ) = −ẑ∇s · Hs , and that ∇s · ẑ × Es = −ẑ · ∇s × Es , it can
be seen that the last terms in (2.3.25) and (2.3.26) are identical except for a sign difference.
Therefore, after using Gauss’ theorem,
Z I
dS[(∇s × ∇s × Es ) × Hs + Es × ∇s ∇s · Hs ] · ẑ = dln̂ · [(∇s × Es )
S C (2.3.27)
× (Hs × ẑ) + (∇s · Hs )ẑ × Es ].
Since ∇s × Es = iωµHz and ∇s · Hs = −ikz Hz , the right-hand side vanishes if we have either
an electric wall or a magnetic wall or a mixture thereof. Similarly, the other terms on the
right hand side of (2.3.24) vanish by the same argument. Therefore, in general,
Z
(Eis × Hjs ) · ẑdS = 0, i 6= j, (2.3.28)
S
for any two distinct modes with different propagation constants kz or ks .
Power Orthogonality
Since, H∗js is also a solution to (2.3.23), we have
Z
(Eis × H∗js ) · ẑdS = 0, i 6= j, power orthogonality. (2.3.29)
S
Since E × H∗ represents the complex Poynting vector, Equation (2.3.29) implies that the
power flow in a waveguide is independently carried by each mode. Cross interactions between
the E and H fields of two different modes do not result in power flow as testified by Equation
(2.3.29).
2
The above orthogonality principles assume that the modes have distinct eigenvalues kis or
2 2
distinct axial propagation constants kiz . When kis for two different modes are the same, the
modes are called degenerate. If there are N degenerate, independent modes, we can use the
Gram-Schmidt orthogonalization procedure to obtain N orthogonal modes if we so desire.
2.4 Power Flow in a Waveguide

Because of the power orthogonality of the distinct modes in a waveguide with perfectly im-
penetrable walls (perfect electric conductors or perfect magnetic conductors), we can study
the power flow due to each mode and the total power flow is the sum of the power flow from
each mode.
2.4.1 Power Flow and Group Velocity

The time average power flow in a waveguide is given by
Z
1
Pf = <e (Es × H∗s ) · ẑdS. (2.4.1)
2
S
For a TE mode with eikz z dependence, from (2.1.8),

iωµ
Es = ∇s × ẑHz , (2.4.2)
ks2
ikz
Hs = ∇s Hz . (2.4.3)
ks2
Substituting into (2.4.1), we can show that
 
Z
1  ωµkz∗ ∗

Pf = <e ẑ (∇ s Hz ) · (∇ s Hz )dS . (2.4.4)
2  ks4 
S
(ks2is always real in a homogeneously-filled waveguide). Using the fact that ∇s · (Hz ∇s Hz∗ ) =
(∇s Hz ) · (∇s Hz∗ ) + Hz ∇2s Hz∗ , the above becomes
  
Z Z
1  ωµkz∗  ∗ 2 ∗

Pf = <e ẑ dln̂ · H ∇ H
z s z − dSH ∇ H
z s z . (2.4.5)
2  ks4 
C S
The first integral vanishes by virtue of the boundary condition. While using ∇2s Hz∗ = −ks2 Hz∗ ,
we have  
 ∗ Z 
ẑ ωµkz 2
Pf = <e dS|H z | . (2.4.6)
2  ks2 
S
For a TM mode, we can similarly show that

 
 ∗ Z 
ẑ ω kz 2
Pf = <e dS|Ez | . (2.4.7)
2  ks2 
S
p
When the mode is cutoff, i.e., when k 2 < ks2 , so that kz = k 2 − ks2 is pure imaginary, and
the waveguide is lossless, there is no time average power flow down the waveguide. In such a
case, the mode is evanescent.
The time average energy stored per unit length in the electric field for the TE mode is
given by
Z
1
hWe i = Es · E∗s dS
4
S
Z (2.4.8)
ω |µ|2
2
∗
= (∇ H
s z ) · (∇ H
s z )dS.
4ks4
S
In the same manner as (2.4.4), we can show that

Z
ω 2 |µ|2
hWe i = |Hz |2 dS. (2.4.9)
4ks2
S
It can be shown that hWe i = hWm i for a lossless waveguide where hWm i is the time average
energy stored in the magnetic field. Therefore, the total time average energy stored per unit
length is Z
ω 2 µ2
hWT i = hWe i + hWm i = |Hz |2 dS. (2.4.10)
2ks2
S
For a lossless waveguide with a propagating mode, kz is pure real and (2.4.6) for the TE mode
becomes Z
ẑ ωµkz
Pf = dS|Hz |2 . (2.4.11)
2 ks2
S
Comparing (2.4.10) and (2.4.11), we note that for a lossless waveguide,
kz
Pf = hWT i, (2.4.12)
ωµ
where kz /ωµ has the dimension of velocity. In a waveguide,
kz2 = ω 2 µ − ks2 . (2.4.13)
The group velocity in a waveguide is derived to be

dω kz
vg = = . (2.4.14)
dkz ωµ
Therefore, (2.4.12) is just

Pf = vg hWT i. (2.4.15)
In other words, in a lossless waveguide, the time average energy stored per unit length, moving
at the group velocity vg contributes to the power flow. The group velocity in a waveguide is
the velocity of energy propagation, and it is also the signal velocity.
The phase velocity of a wave in a waveguide is defined to be
ω ω
vph = =p . (2.4.16)
kz k − ks2
2
It is the velocity of the phase of the wave. Since a signal does not travel at the phase velocity,
it could be larger than the speed of light. This happens near cutoff when k 2 → ks2 for a mode.
Group velocity or signal velocity cannot be larger than the speed of light, a limit dictated by
Einstein’s theory of special relativity. Note that vph vg = c2 where c is the velocity of light in
the medium.
2.4.2 Pulse Propagation in a Waveguide

Since the phase and the group velocities inside a hollow waveguide are frequency dispersive,
a pulse propagating inside a hollow waveguide will be distorted due to frequency dispersion.
Different frequency components will travel with different velocities. Hence, after a certain
distance of propagation, different Fourier components lose their phase coherence, causing
pulse distortion. Therefore, to minimize pulse distortion, the bandwidth of the pulse should
be narrow. It can be shown that for a narrow-band pulse, the envelope of the pulse propagates
with the group velocity while the carrier signal propagates with the phase velocity.
A narrow-band pulse can be written as:
Z ∞
1
p(r, t) = dω p̃(r, ω)e−iωt
2π −∞
Z ∞ (2.4.17)
1
= <e dω p̃(r, ω)e−iωt
π 0
where p(r, t) may represent a component (e.g., the z component) of the electric field or
magnetic field inside a waveguide. Since

2 1 ∂2
∇ − 2 2 p(r, t) = 0, (2.4.18)
c ∂t
by substituting (2.4.17) into (2.4.18), we require that

ω2
∇2 + 2 p̃(r, ω) = 0, (2.4.19)
c
If p(r, t) corresponds to a particular mode, for example, in the case of a single mode propa-
gation, then
∇2s + ks2 p̃(r, ω) = 0, (2.4.20)
and subtracting (2.4.19) from (2.4.20), we have

2
∂ 2
+ kz p̃(r, ω) = 0, (2.4.21)
∂z 2
ω2
where kz2 = c2 − ks2 . Hence
p̃(r, ω) = p̃o (rs , ω)eikz z . (2.4.22)
where rs = x̂x + ŷy. Consequently, we can rewrite (2.4.17) as
Z ∞
1
p(r, t) = <e dω põ (rs , ω)eikz z−iωt . (2.4.23)
π 0
If p(r, t) is a narrow-band pulse with a carrier frequency at ωo , the above can be approximated
by an integral
Z ω0 +∆
1
p(r, t) ' <e dω p̃o (rs , ω)eikz z−iωt . (2.4.24)
π ω0 −∆
In the above, we can approximate kz (ω) as
dkz (ω0 ) (ω − ω0 )
kz (ω) ' kz (ω0 ) + (ω − ω0 ) = kz (ω0 ) + (2.4.25)
dω vg
and obtain
( Z ω0 +∆ )
1 i(ω−ω0 )(z−vg t)
p(r, t) ' <e ei[kz (ω0 )z−ω0 t] dω p̃o (rs , ω)e vg
π ω0 −∆ (2.4.26)
n o
' 2<e ei[kz (ω0 )z−ω0 t] F (rs , z − vg t)
where Z ω0 +∆
1 i(ω−ω0 )z
F (rs , z) = dω p̃o (rs , ω)e vg
(2.4.27)
2π ω0 −∆
Since ω − ω0 is small, F (rs , z) is a slowly varying function of z. It represents an envelope

function, which in (2.4.26), propagates at the group velocity vg . The envelope function
modulates a rapidly varying function
ei(kz z−ω0 t) (2.4.28)
which travels at the phase velocity vph = ωkz0 . Hence for a narrow-band signal, a shape-
retaining envelope pulse can propagate in a dispersive waveguide.
2.4.3 Attenuation in a Waveguide

When we have a lossy dielectric medium inside a perfectly conducting waveguide, the atten-
uation due to the lossy dielectric can be ascertained from the formula
p
kz = ω 2 µ − ks2 = kz 0 + ikz 00 , (2.4.29)
where = 0 + i00 is complex. This is because the mathematical boundary value problem
has not changed when becomes complex. Hence, (2.3.1) and (2.3.2) hold true even for lossy
dielectric.
When the attenuation is due to wall losses because of the finite conductivity of the metallic
wall, the calculation is more involved. In this case, we can use the following formula derived
from energy conservation
Pd
kz 00 = , (2.4.30)
2Pf
where Pd is the time average power dissipated per unit length while Pf is the total time
average power flow in the waveguide. Since n̂ × E is not identically zero on the waveguide
wall now, n̂ · (E × H∗ ) is not zero on the waveguide wall. We can calculate the time average
power dissipated per unit length by integrating n̂ · (E × H∗ ) over the circumference of the
waveguide wall, i.e., I
1
Pd = <e dln̂ · (E × H∗ ), (2.4.31)
2 C
where C is a contour defining the cross-section of the waveguide. Since n̂ × E is not zero, it
can be approximated by r
µω 1
n̂ × E = H = (1 − i) H, (2.4.32)
iσ σδ
q
2
where δ = ωµσ is the skin depth in a metallic conductor. The above follows from that in
the metalpof the waveguide, the E and H are related by the intrinsic impedance of metal
which is µω/iσ [see Problem 2-2]. Therefore, (2.4.31) becomes
I
1
Pd = dl|H|2 , (2.4.33)
2σδ C
where H is purely tangential on the waveguide wall, if the wall is a perfect conductor. For
a perturbation calculation, we can use the H field of a perfectly conducting waveguide to
estimate (2.4.33). Note that σδ is also the surface conductance of the waveguide.
For a TE mode,
ikz
H = ẑHz + 2 ∇s Hz . (2.4.34)
ks
Therefore, from (2.4.33), Pd becomes
I
1 |kz |2
Pd = dl |Hz |2 + 4 |∇s Hz |2 . (2.4.35)
2σδ C ks
Using (2.4.11), (2.4.30), and (2.4.35), we obtain that
H n o
2 |kz |2 2
2
k δ C dl |H z | + ks4 |∇ H
s z |
kz 00 = s R
2
. (2.4.36)
4kz dS|Hz |
S
For a TM mode,
iω
H=− ∇s × ẑEz . (2.4.37)
ks2
Then, Pd becomes I
ω 2 2
Pd = dl|∇s Ez |2 , (2.4.38)
2σδks4 C
and H
00 k 2 δ C dl|∇s Ez |2
kz = R . (2.4.39)
4kz ks2 dS|Ez |2
S
The above method of computing the attenuation of a waveguide is also known as the power-
loss method. It is inadequate when the modes of the waveguides are degenerate.
For the TE case, kz → k when ω → ∞, while ks2 remains a constant independent of
frequencies. Therefore, from (2.4.36), we have
H
00 δk C dl|∇s Hz |2
kz ∼ 2 R , (2.4.40)
4ks dS|Hz |2
S
p
which increases as the frequency increases since δk = 2ω/σ.
For the TM case, when ω → ∞, from (2.4.39)
H
δk dl|∇s Ez |2
kz 00 ∼ 2 CR , (2.4.41)
4ks dS|Ez |2
S
which also increases as the frequency increases. Therefore, a metallic waveguide becomes
more inefficient at high frequencies.
From (2.4.36) and (2.4.39), we see that kz 00 diverges when kz → 0, i.e., when the wave
tends to cutoff. This is because there is no real power flow Pf at cutoff, and hence Pf → 0
in (2.4.30). Since Equation (2.4.30) embodies a perturbation concept, it is only valid when
kz 00 kz 0 . Therefore, Equations (2.4.36) and (2.4.39) are invalid near the cutoff of the wave.
But still, the trend is that kz00 becomes larger close to cutoff.
The Magic Modes

For some special modes of a waveguide, due to symmetry, the second term in (2.4.34) vanishes.
This can happen to some modes of a highly symmetrical waveguide such as a parallel plate
waveguide or a circular waveguide. In this case, (2.4.36) becomes
H
00 ks2 δ C dl |Hz |2
kz = R . (2.4.42)
4kz dS|Hz |2
S
In this case, kz00 becomes smaller as the frequency increases. For these modes, the electric
field is tangential to the waveguide wall, with magnetic field normal to the k vector and
the electric field. As the frequency increases, the k vector becomes almost parallel to the
z axis. The magnetic field becomes almost vertical to the waveguide wall with a small
tangential component. Hence, the induced surface current on the waveguide wall actually
becomes smaller. Consequently, the attenuation of the waveguide mode actually decreases
Metallic
wall
Figure 2.8: The field configuration of a magic mode on the wall of a waveguide. As the
frequency increases, the k vector becomes more parallel to the waveguide wall, and tangential
H becomes smaller, since H is orthogonal to both E and k. This reduces the surface current,
and hence, the wall loss.
with increasing frequency. Some TE modes of the circular waveguide are such a mode, and
they are known as “magic modes” (see Problems 2.9 and 2.12).
Figure 2.9 shows some typical losses of different modes in a rectangular and a circular
waveguide including a “magic mode”. Such low loss is desirable in radio astronomy where
the frequency is high and the signal low. Hence, circular waveguides with a corrugated wall
to “discourage” other modes, but promote the propagation of this magic mode, is actually
used in the design of the VLA (very large array) of NRAO (National Radio Astronomy
Observatory) in New Mexico.
2.5 Wave Impedance

Just as the intrinsic impedance in a homogeneous medium relates the electric field to a
magnetic field, we may relate the transverse components of E and H in a waveguide by a
wave impedance.
−4
x 10 Attenuation as a function of frequency
1
0.9
0.8
0.7
Attenuation (Np/m)
0.6
TE10 of rectangular W.G.
0.5 TE11 of circular W.G.
TE01 of circular W.G.
0.4
0.3
0.2
0.1
0
0 2 4 6 8 10 12 14 16 18 20
Normalized frequency
Figure 2.9: Loss as a function of frequency for different waveguide modes.
For a TE mode, the transverse components are

ikz −iωµ
Hs = ∇s Hz , Es = ẑ × ∇s Hz . (2.5.1)
ks2 ks2
Therefore, the wave impedance is
ẑ × Es ωµ
ZT E = = . (2.5.2)
Hs kz
For a TM mode, the transverse components are
iω ikz
Hs = ẑ × ∇s Ez , Es = ∇s Ez . (2.5.3)
ks2 ks2
Therefore, the wave impedance is
ẑ × Es kz
ZT M = = . (2.5.4)
Hs ω
The above is valid for a general cylindrical waveguide which is homogeneously filled. It is
useful in deriving equivalent transmission line models for a waveguide.
2.6 Rectangular Waveguides

The rectangular waveguide is the most commonly used hollow waveguide. By adjusting the
aspect ratio, a to b, of the waveguide, one can obtain a good bandwidth for single mode
propagation. Furthermore, the analysis of this waveguide is simple.
x
a
z
Figure 2.10: A rectangular waveguide.
2.6.1 TE Modes (H Modes)

A TE mode in a rectangular waveguide is characterized by Hz satisfying Equation (2.3.1) with
∂
the requisite Neumann boundary condition. An Hz that will satisfy (2.3.1) with ∂n Hz = 0
on the waveguide wall is
mπ nπ
Hz = H0 cos x cos y eikz z , (2.6.1)
a b
p
where ks2 = ( mπ 2 nπ 2
a ) + ( b ) , kz = k 2 − ks2 . The fact that ks2 has to satisfy the prescribed
values is also known as the guidance condition. The transverse fields can be found using
(2.1.7) and (2.1.8). The mode becomes evanescent or non-propagating when ks2 > k 2 , i.e.,
when kz becomes imaginary. Since k 2 = ω 2 µ, the cutoff frequency (the frequency below
which the mode is evanescent) is given by
1
1 mπ 2 nπ 2 2
ωmnc = √ + . (2.6.2)
µ a b
The corresponding mode is usually labeled as TEmn (or Hmn ) mode. The wavelength of a
wave at ωc in the medium denoted by µ, is the cutoff wavelength. It is
2
λmnc = 1 . (2.6.3)
[( m
a)
2 + ( nb )2 ] 2
In waveguide conventions, a is assumed larger than b. Then, the dominant mode (funda-
mental mode) with the lowest cutoff frequency is the m = 1, n = 0 mode, also known as the
TE10 mode (or H10 mode). The TE00 mode does not exist because, in this case, ks = 0, and
Es and Hs diverge from (2.1.7) and (2.1.8) when ks → 0.
2.6.2 TM Modes (E Modes)

A TM mode in a rectangular waveguide is characterized by Ez satisfying Equation (2.3.2)
with the requisite Dirichlet boundary condition. An Ez that will satisfy (2.3.2) with Ez = 0
on the waveguide wall is

mπ nπ
Ez = E0 sin x sin y eikz z , (2.6.4)
a b
p
a ) + ( b ) , kz = k 2 − ks2 . The TMmn mode has the same cutoff frequency
as the TEmn mode. However, when either m = 0, or n = 0, the mode does not exist since
Ez = 0 then. Therefore, the lowest TM mode is the TM11 mode with a cutoff frequency above
that of the TE10 mode. Given the z components of the fields, all other field components of a
waveguide can be derived. Figure 2.20 shows the field plots of some modes of a rectangular
waveguide [8].2
2.7 Circular Waveguides

Certain modes of a circular waveguide have less attenuation from wall loss compared to a
rectangular waveguide. Hence, it is sometimes preferred over a rectangular waveguide.
2.7.1 TE Modes (H Modes)
a φ
x
Figure 2.11: A circular waveguide.
The Ez component of a TE mode satisfies Equation (2.3.1) in cylindrical coordinates, i.e.,

1 ∂ ∂ 1 ∂2 2
ρ + + ks Hz = 0, (2.7.1)
ρ ∂ρ ∂ρ ρ2 ∂φ2
∂
with the boundary condition that ∂ρ Hz = 0, ρ = a, where a is the radius of the waveguide.
±inφ
If we assume that Hz has e , sin nφ or cos nφ dependence, where n is an integer, we can
∂2 2
replace ∂φ 2 by −n . Equation (2.7.1) then becomes

1 ∂ ∂ n2
ρ − 2 + ks2 Hz = 0. (2.7.2)
ρ ∂ρ ∂ρ ρ
2 The plots here are reproduced by A. Greenwood according to this reference.
Equation (2.7.2) is the Bessel equation, whose solutions are either Jn (ks ρ), Yn (ks ρ),
(1) (2)
Hn (ks ρ) or Hn (ks ρ). Of these four solutions, only two are independent because of the
relations
Hn(1) (ks ρ) = Jn (ks ρ) + iYn (ks ρ), (2.7.3a)
Hn(2) (ks ρ) = Jn (ks ρ) − iYn (ks ρ), (2.7.3b)
(1)
Jn (ks ρ) is regular about the origin when ρ → 0, but Yn (ks ρ) is singular (so are Hn (ks ρ)
(2)
and Hn (ks ρ)). Since the field cannot be infinite at the center of the waveguide due to the
absence of sources, the solution to (2.7.1) is of the form
Hz = H0 Jn (ks ρ)e±inφ+ikz z . (2.7.4)
∂
We require that ∂ρ Hz = 0 at ρ = a, implying that
Jn0 (ks a) = 0, (2.7.5)

p
with kz = k 2 − ks2 . If the m-th zero of Jn0 (x) is defined to be βnm such that Jn0 (βnm ) = 0,
the values of possible ks are
βnm
ks = . (2.7.6)
a
The above is also the guidance condition for the waveguide mode. The subscript n denotes
the orders of the Bessel function Jn (x) and the circular harmonic e±inφ . The subscript m
denotes the m-th zero of Jn0 (x) discounting the zero at the origin. The corresponding mode
√
is usually denoted as the TEnm mode. Cutoff occurs when k = ω µ < ks . From Figure
2.12, we see that the TE11 mode corresponding to the first zero of J1 (x) has the lowest cutoff
frequency. The cutoff frequency for the TEnm mode is given by
1 βnm
ωnmc = √ , (2.7.7)
µ a
and the corresponding cutoff wavelength is
2π
λnmc = a. (2.7.8)
βnm
2.7.2 TM Modes (E Modes)

Similar to a TE mode, the Ez component of a TM mode has e±inφ dependence, satisfying
the equation
1 ∂ ∂ n2 2
ρ − 2 + ks Ez = 0, (2.7.9)
ρ ∂ρ ∂ρ ρ
with the boundary condition Ez (ρ = a) = 0. Hence,
Ez = E0 Jn (ks ρ)e±inφ+ikz z , (2.7.10)
with Jn (ks a) = 0. If we denote the m-th zero of Jn (x) by αnm , then ks has possible values of
αnm
ks = , (2.7.11)
a
1 1
J0
J1 J’1
0.5 J2 0.5 J’2
0 0
−0.5 −0.5
J’0
−1 −1
0 5 10 15 20 0 5 10 15 20
(a) (b)
Figure 2.12: Plots of Bessel functions and their derivatives.
where the subscript n denotes the order of the Bessel function, Jn (x), and the subscript m
denotes the m-th zero of Jn (x), discounting the zero at the origin. It is also the guidance
condition, and the corresponding mode is known as the TMnm mode. The cutoff frequency
of the TMnm mode is given by
1 αnm
ωnmc = √ , (2.7.12)
µ a
and the corresponding cutoff wavelength is
2π
λnmc = a. (2.7.13)
αnm
Looking at Figure 2.12, we see that the lowest TM mode is the TM01 mode, but it has a
higher cutoff frequency compared to the TE11 mode.
Table 2.3.1. Roots of Jn0 (x) = 0.

n βn1 βn2 βn3 βn4
0 3.832 7.016 10.174 13.324
1 1.841 5.331 8.536 11.706
2 3.054 6.706 9.970 13.170
3 4.201 8.015 11.346 14.586
4 5.318 9.282 12.682 15.964
5 6.416 10.520 13.987 17.313
Table 2.3.2. Roots of Jn (x) = 0.

n αn1 αn2 αn3 αn4
0 2.405 5.520 8.654 11.792
1 3.832 7.016 10.174 13.324
2 5.135 8.417 11.620 14.796
3 6.380 9.761 13.015 16.223
4 7.588 11.065 14.373 17.616
5 8.771 12.339 15.700 18.980
It can be shown that the TE01 mode has the lowest loss at high frequencies. The TE01
mode is axially symmetric with E = φ̂Eφ . Therefore, it can be enhanced by various means.
One way is to use a mode filter as shown in Figure 2.13(a) [9]. The radial conducting wire
will short out modes with radial components of the electric field. However, the TE01 mode
is oblivious to the presence of the radial conducting strips, and is little affected. Another
way to enhance the TE01 mode in a circular waveguide is to use a ribbed waveguide. For
the TE01 mode, the current is purely circumferential and is oblivious to the presence of the
ribs. However, the TM modes, which have axial components of the current, will be affected
by the ribbed wall of the waveguide. In other words, the ribbed waveguide wall does not
support the axial current flow effectively. If the periodicity of the waveguide corrugation is
small compared to the wavelength, the TE01 mode will not be affected much.
Radial Conducting (a) Mode Filter

Wires
(b) Ribbed Waveguide
Figure 2.13: (a) A mode filter used to filter out most other modes, but it allows the TE01
modes to pass through. (b) A ribbed waveguide prevents axial current flow, hence attenuates
the TM modes, which has Ez 6= 0.
The transverse field components of a circular waveguide are easily obtained given the axial
components. Figure 2.21 shows the field plots of some modes in a circular waveguide [8].
2.8 Excitation of Modes in a Waveguide

The modes of a waveguide are excited by putting sources inside a waveguide. This is usually
in terms of waveguide probe carrying a current or a charge. So the probe can be dipole-like,
producing mainly an electric field, or it can be loop-like producing mainly a magnetic field.
The location of the probe is imperative if certain desirable modes are to be excited. We will
study the relation between the probe location, the source type, and the modes
Generalized Eigenfunction Expansion for Vector Wave Equation

Given an electric field that satisfies
∇ × ∇ × E(r) − k02 E(r) = iωµJ(r) (2.8.1)
we can expand the field in terms of the eigenfunctions of the following equation
2
∇ × ∇ × Fm (r) − km Fm (r) = 0 (2.8.2)
assuming that Fm satisfies the same boundary condition as the electric field E. First, we let
X
E(r) = am Fm (r) (2.8.3)
m
On substituting (2.8.3) into (2.8.1), we have

X
2
am (km − k02 )Fm (r) = iωµJ(r) (2.8.4)
m
Assuming an orthonormal relationship for the eigenfunctions such that3

Z
drF∗m0 (r)Fm (r) = δm0 m (2.8.5)
V
we can deduce that

hF∗m , Ji
am = iωµ 2 − k2
(2.8.6)
km 0
Consequently, we get
X hF∗m , Ji
E(r) = iωµ Fm (r) 2 − k2
(2.8.7)
m
km 0
3 Such an orthogonality relation with a conjugation is needed since the operator ∇ × ∇× is a Hermitian
(self-adjoint) operator.
The above is a general eigenfunction expansion formula if we orthonormalize the eigenfunc-

tions. It says physically that the field due to a source in a waveguide can be expanded in
terms of the eigenfunctions of the waveguide. The same expression also holds for cavity mode
expansion. The excitation coefficients of the eigenfunctions are given by (2.8.6). These coef-
ficients are proportional to hF∗m , Ji which is the inner product between the eigenfunction and
the source. Hence, it is important that we learn how to find these eigenfunctions.
Modes are often excited by current probes in the waveguide or cavity. The above expression
tells us if we want a certain mode Fm to be strongly excited, we need the inner product hF∗m , Ji
to be large. Hence, the current on the probe should be located at where the field of the mode
Fm is strong. If the probe is a short wire, it can be approximated by an electric dipole with
strong charge accumulation that produces a strong electric field. This electric dipole should
be placed close to the maxima of the mode in order to excite it.
On the other hand, if the current source J consists of a current loop with constant current,4
then the inner product
Z I Z
∗ ∗ ∗
hFm , Ji = drFm (r) · J(r) = dl · Fm (r) = dS n̂∇ × F∗m (r) (2.8.8)
V C A
where C is the contour of the loop, and A is the cross section area of the loop. If Fm
represents electric field, then ∇ × Fm represents the magnetic field. Hence, the current loop
has to be placed in location where the magnetic field of the mode is strong in order to excite
it. A small current loop behaves like a magnetic dipole and hence, it needs to be placed near
strong magnetic field in order to excite the mode strongly.
2
Also, due to the 1/(km − k02 ) dependence of the excitation coefficient, if the operating
frequency of the source is closed to the resonant frequency of the mode, that mode will be
strongly excited. This is the phenomenon of resonance coupling. Energy can be coupled to a
mode if we operate close to the resonant frequency of the mode.
2.8.1 Vector Wave Functions in a Waveguide

In
P∞ scalar function theory, we know that an arbitrary function can be expanded as f (x) =
nπx
n=0 an cos a by the completeness of the Fourier cosine basis for 0 < x < a. It turns out
that the waveguide modes that we have previously studied do not constitute a complete set.
First, they satisfy source-free Maxwell’s equations and hence, they are divergence free. Hence,
they cannot constitute a non-divergence free field in a waveguide. Second, they propagate in
either plus or minus z directions. If a source is present in a waveguide, it is not clear what
direction the field is propagating in the source region. To obtain a complete set, we need to
derive the vector wave functions.
An arbitrary vector function in a waveguide can be expanded in terms of vector wave
functions of the waveguide which is the analogue of Fourier basis. However, drawing such an
analogy has its pitfall. The expansion of modes in a cavity excited by a current source has
a colorful history. There are basically two kinds of vector wave functions in a waveguide or
a cavity: the divergence-free type (solenoidal) and the curl-free type (irrotational). It was
believed by some for a while that the usual divergence free functions are complete. But in
4 This is possible if the loop size is much smaller than the wavelength.
fact, both kinds of functions are needed for completeness [4, 10, 11]. More references on this
topic can be found in [13].
We shall discuss the derivation of such vector wave functions for a uniform hollow waveg-
uide. Both the electric field and the magnetic field in a waveguide satisfy the equation
∇ × ∇ × F − k 2 F = 0. (2.8.9)
It is to be noted that the above is an eigenvalue problem where k 2 is the eigenvalue, and F
is the eigenfunction. It can be shown that with the appropriate boundary conditions, the
∇ × ∇× operator is Hermitian, and that the eigenvalues k 2 are always real.
The above is not actually the equation satisfied by the fields of the waveguide since k 2
is arbitrary and not fixed to be ω 2 µ. However, solving the above equation for all possible
k 2 or eigenvalues generates enough eigenfunctions that form a complete set. Since k 2 is
equivalent to frequency, we can think of the above as solving for all the resonance solutions
(eigensolutions) of a very long cavity. As we shall see next, if the cavity is of finite length,
then k 2 will assume discrete values. But if the cavity is of infinite length, then k 2 assumes a
continuum of values.
Periodic Boundary Condition

We can think of the F functions as cavity modes of a waveguide with periodic boundary
condition in the z direction. If we have a traveling wave in the z direction indicated by
exp(ikz z) with the requirement that this function repeats itself after distance d, then
eikz z |z=0 = eikz z |z=d (2.8.10)
The above implies that
eikz d = 1 (2.8.11)
implying that
2pπ
kz = , ∀ integer p (2.8.12)
d
The traveling wave then becomes a Fourier mode
2pπz
eikz z = ei d (2.8.13)
When d → ∞, kz assumes a continuum of modes as in Fourier transform.
Divergence-Free Eigenfunctions
We can convert the (2.8.9) into a scalar wave equation via the transformation
F = ∇ × cψ. (2.8.14)
The above vector function is clearly divergence free. Substituting (2.8.14) into (2.8.9), and if
ψ satisfies
(∇2 + k 2 )ψ = 0, (2.8.15)
then F in (2.8.14) will satisfy (2.8.9). It is clear that F is transverse to c. Since in a waveguide,
as well as an infinitely long cavity, we can decompose the field into TE and TM types with
respect to z, it is natural to choose c = ẑ. For example, if ψh satisfies (2.8.15) with the
Neumann boundary condition, i.e.,
∂ψh
(∇2 + k 2 )ψh = 0, = 0, on C (2.8.16)
∂n
then the vector wave function
Me = ∇ × (ẑψh ), (2.8.17)
is analogous to the E-field of a TE mode. Again, it is to be emphasized that (2.8.16) defines
an eigenvalue problem.
Since ∇ × Me is also a solution to (2.8.9), we can define another vector wave function
1 1
Nh = ∇ × Me = ∇ × ∇ × ẑψh . (2.8.18)
k k
Nh is a vector wave function analogous to the magnetic field of a TE mode.
Similarly, we define a ψe satisfying
(∇2 + k 2 )ψe = 0, ψe = 0, on C. (2.8.19)
Then,
Mh = ∇ × ẑψe , (2.8.20)
is analogous to the H-field of the TM mode in a waveguide. Similarly, we have
1 1
Ne = ∇ × Mh = ∇ × ∇ × ẑψe , (2.8.21)
k k
which is analogous to the E field of a TM mode.
In the previous discussion, the subscript e denotes a quantity that is related to the E
field while the subscript h denotes a quantity that is related to the H-field in terms of their
boundary conditions. ψh and ψe are sometimes known as the magnetic Hertzian potential and
the electric Hertzian potential respectively. By looking back at Equations (2.1.7) and (2.1.8),
one notes that they are analogous to the Hz and Ez components of the field respectively.
As an example, for a rectangular waveguide,
mπx nπy
ψhmn (kz , r) = cos cos eikz z , (2.8.22a)
a b
mπx nπy
ψemn (kz , r) = sin sin eikz z . (2.8.22b)
a b
We can generate the vector wave function from the above by
mπx nπy
Memn (kz , r) = ∇ × ẑ cos cos eikz z , (2.8.23a)
a b
1 mπx nπy
Nhmn (kz , r) = ∇ × ∇ × ẑ cos cos eikz z , (2.8.23b)
k a b
and mπx nπy

Mhmn (kz , r) = ∇ × ẑ sin sin eikz z , (2.8.24a)
a b
1 mπx nπy
Nemn (kz , r) = ∇ × ∇ × ẑ sin sin eikz z . (2.8.24b)
k a b
We allow kz to be a free variable (arbitrary variable) in Equations (2.8.22a) and (2.8.22b).
This would render ψhmn and ψemn to be complete scalar functions in a hollow waveguide. At
least, we know that an arbitrary function of z can be Fourier expanded in terms of eikz z .
The above is analogous to finding the eigenvalues of a rectangular cavity of dimension
a × b × d with periodic boundary condition in the z direction. In this case, the eigenvalue is
2 2 2
given by k 2 = mπ a + nπ b + 2pπ d , where m, n, and p are all integers. We can identify
2pπ
kz = d . As we let d → ∞, kz which previously takes on discrete values, becomes a
continuum variable. Since kz can be any continuum real variable, the eigenvalue k 2 = ks2 + kz2
can also take on any continuum real variable. In the above ks is the transverse eigenvalue of
the waveguide problem similar to (2.3.1) and (2.3.1). We shall denote it as kis subsequently,
where the index i implies an ordered pair (m, n) in the case of a rectangular waveguide.
As of this point, these vector wave functions are not physical modes of a waveguide. In
order for them to be physical modes, kz2 has to satisfy the dispersion relation k02 = ks2 + kz2
where k0 is the wavenumber inside the waveguide. Since the dispersion relation describes an
equation of a sphere in the k space, this sphere is known as the Ewald sphere or the energy
shell in physics. The eigenfunctions derived so far do not satisfy the dispersion relation, and
they are said to be off the energy shell. When they are forced to satisfy the dispersion relation,
they are said to be on the energy shell.
Given the above information, we can easily show that
Z
dV ψhmn (kz , r)ψhm0 n0 (−kz 0 , r) =
(2.8.25)
πab
(1 + δ0m )(1 + δ0n ) δ(kz − kz 0 )δmn,m0 n0 ,
2
where δ(x) is the Dirac delta function and δij , the Kronecker delta function. In general, for
a waveguide of arbitrary cross-section, orthogonality relations exist for the ψh ’s and ψe ’s as
[analogous to (2.3.3) and (2.3.4)] 5
Z
dV ψhi (kz , r)ψhj (−kz 0 , r) = Ahi δ(kz − kz 0 )δij , (2.8.26a)
Z
dV ψei (kz , r)ψej (−kz 0 , r) = Aei δ(kz − kz 0 )δij , (2.8.26b)
where Ahi and Aei are the appropriate normalization constants.6

5 See[13] p. 388.
6 Notice ∗ (k ), but since k will become complex in Cauchy integration
that when kz is real, ψej (−kz ) = ψej z z
technique applied on the complex plane, a technique that we will use later, we will retain this notation in the
above.
For the vector wave functions, the orthogonality relationships are

Z
dV M e i (kz , r) · M e j (−kz 0 , r) = k 2 h A h i δ(kz − kz 0 )δij , (2.8.27a)
h h is e e
Z
dV N e i (kz , r) · N e j (−kz 0 , r) = kis
2 0
e A e i δ(kz − kz )δij , (2.8.27b)
h h h h
Z
dV M e i (kz , r) · N e j (−kz 0 , r) = 0, for all i, j, kz , kz 0 . (2.8.27c)
h h
Because the vector functions M and N are divergence free, and that their curls are not
zero, these functions are also termed the solenoidal vector wave functions. They can be used
to expand divergence-free fields.
Curl-Free Eigenfunctions
The expressions (2.8.20) and (2.8.22a) and (2.8.22b) are clearly divergence free. However, M
and N vector wave functions are not, in general, complete. They cannot be used to represent
fields whose divergence is non zero. To remedy this, we need the L functions, which are
defined as7
Lh = ∇ψh , Le = ∇ψe , (2.8.28)
where ψh and ψe are as defined in (2.8.16) and (2.8.19). Lh satisfies the magnetic field
boundary condition while Le satisfies the electric field boundary condition on the waveguide
wall. The vector wave functions have zero curl and non-zero divergence. Hence, they are
also known as the irrotational vector wave functions. They are solutions to Equation (2.8.9)
corresponding to when k 2 = 0. Hence, they are the null-space solution of the ∇ × ∇×
operator.
Their divergence is
∇ · Lh = −k 2 ψh , ∇ · Le = −k 2 ψe . (2.8.29)
Since ψh and ψe form complete sets, they can be used to expand field due to an arbitrary
source inside the waveguide.
Z
dV L e i (kz , r) · L e j (−kz 0 , r) = k 2 A e i δ(kz − kz0 )δij , (2.8.30a)
h h h
Z
dV M e i (kz , r) · L e j (−kz 0 , r) = 0, all i, j, kz , kz 0 , (2.8.30b)
h h
Z
dV N e i (kz , r) · L e j (−kz 0 , r) = 0, all i, j, kz , kz 0 , (2.8.30c)
h h
7 It is to be noted that Helmholtz theorem says that an arbitrary vector field can be decomposed into the
sum of divergence-free field and curl-free field. In other words, F = ∇ × A + ∇ψ. We expect to see such
decomposition here.
Eigenfunction Expansion of Arbitrary Fields

An arbitrary field in a waveguide can, in general, be expanded as
Z ∞ X
E(r) = dkz [aei (kz )Mei (kz , r) + bei (kz )Nei (kz ) + cei (kz , r)Lei (kz , r)], (2.8.31a)
−∞ i
Z ∞ X
H(r) = dkz [ahi (kz , r)Mhi (kz , r) + bhi (kz )Nhi (kz , r) + chi (kz )Lhi (kz , r)]. (2.8.31b)
−∞ i
The coefficients can be found from the orthogonality relationships.
2.8.2 Dyadic Green’s Function

The dyadic Green’s function in a waveguide is the solution to the equation
∇ × ∇ × G(r, r0 ) − k02 G(r, r0 ) = Iδ(r − r0 ), (2.8.32)
satisfying the boundary condition n̂ × G(r, r0 ) = 0, for r on the waveguide wall. Once this
Green’s function is known, the field due to an arbitrary distributed source in a waveguide can
be written as Z
E = iωµ G(r, r0 ) · J(r0 )dr0 . (2.8.33)
A General Dyadic Green’s Function

From the definition of the dyadic Green’s function as given by (2.8.33), and from the gener-
alized formula for mode expansion due to a current source, as given in (2.8.7), we can deduce
that the general dyadic Green’s function is of the form
X Fm (r)F∗ (r0 )
m
G(r, r0 ) = 2 − k2
(2.8.34)
m
k m 0
The above is a succinct way to express the dyadic Green’s function in terms of the modes of
the cavity, where the modes are assumed to be orthonormal.
Dyadic Green’s Function for Hollow Waveguide

The derivation of the dyadic Green’s function has a colorful history as the mode expansion in
a cavity [4, 14, 15, 17]. The controversy comes from the incompleteness of the divergence-free
modes in a waveguide or a cavity. More references can be found in [13].
To solve Equation (2.8.32), we expand G(r, r0 ) in terms of the vector wave functions Me ,
Ne , and Le . In other words,
Z ∞ X
G(r, r0 ) = dkz [Mei (kz , r)aei (kz , r0 ) + Nei (kz , r)bei (kz , r0 )
−∞ i (2.8.35)
+ Lei (kz , r)cei (kz , r0 )].
Substituting G(r, r0 ) into (2.8.32), and using Equation (2.8.49), we have

Z ∞ X
dkz {(k 2 −k02 )[Mei (kz , r)aei (kz , r0 ) + Nei (kz , r)bei (kz , r0 )]
−∞ i (2.8.36)
− k02 Lei (kz , r)cei (kz , r0 )} = Iδ(r − r ). 0
From the orthogonality relations for M, N, and L functions, we have

1
aei (kz , r0 ) = 2 A (k 2 2 ) Mei (−kz , r0 ), (2.8.37a)
kish hi z − kizh
1
bei (kz , r0 ) = 0
2 A (k 2 − k 2 ) Nei (−kz , r ), (2.8.37b)
kise ei z ize
−1
cei (kz , r0 ) = 2 Lei (−kz , r
0
). (2.8.37c)
k2 A ei k0
In the above, we have replaced k 2 −k02 with kz2 −kiz2 2 2 2 2 2 2

e , because k = kis e +kz , k0 = kis e +kiz e .
h h h h
Here, kiz denotes values on the energy shell, while kz denotes values off the energy shell. The
subscripts e and h on the transverse eigenvalue kis and longitudinal wavenumber kiz denote
the association of these values with either the Dirichlet or the Neumann problem respectively.
Therefore, the dyadic Green’s function is
Z ∞ X Mei (kz , r)Mei (−kz , r0 )
G(r, r0 ) = dkz 2 A (k 2 − k 2 )
−∞ i
kish hi z izh
(2.8.38)
Nei (kz , r)Nei (−kz , r0 ) Lei (kz , r)Lei (−kz , r0 )
+ 2 A (k 2 − k 2 ) − .
kise ei z ize k 2 Aei k02
Cauchy Integration Technique

The first two integrals are of the form
Z ∞ 0
eikz (z−z ) f (kz )
I= dkz 2 . (2.8.39)
−∞ kz2 − kiz
There are poles at kz = ±kiz . If we assume a small loss in the medium, then the poles are
off the real axis and the integral (2.8.39) is well-defined. If f (kz )/kz2 → 0 when kz → ∞, for
z > z 0 , we can deform the contour of integration from the real axis to the contour C. By
virtue of Jordan’s lemma, the integral over C vanishes and the integral (2.8.39) is then equal
to the residue of the pole at kz = kiz . When z < z 0 , we can deform the path of integration
downward, and equate the integral (2.8.39) to the residue of the pole at kz = −kiz . Therefore,
it follows that 0
e±ikiz (z−z ) z>0
I = πi f (±kiz ), (2.8.40)
kiz z < 0.
Note that the process of Cauchy integration technique forces kz to be on the energy shell or
on the Ewald sphere.
We can apply Cauchy integration technique to the first two integrals since Mei and Nei
tend to be constants when kz → ∞. However, because
1 1
Nei = ∇ × Mhi = ∇ × ∇ × ẑψei , (2.8.41)
k k
k′′z
C
kiz
⊗ kz′
×-kiz
Figure 2.14: Contour integration on the complex kz plane.
Nei (kz , r)Nei (−kz , r0 ) is proportional to 1/k 2 . But k 2 = kise

2
+ kz2 , implying that there
are additional poles at kz = ±ikise for the second term in (2.8.38). Similarly, the third term
in (2.8.38) also has a 1/k 2 dependence with poles at kz = ±ikise . It can be shown that the
pole contributions at kz = ±ikise for the second and the third terms cancel each other.
Since
Lei (kz , r) ∼ ẑikz ψei , kz → ∞, (2.8.42)
the third term tends to be a constant when kz → ∞. Therefore, contour integration cannot
be applied to the third term, since Jordan’s lemma is not satisfied. To remedy this, we write
(2.8.38) as
Z ∞ X Mei (kz , r)Mei (−kz , r0 )
G(r, r0 ) = dkz 2 A (k 2 − k 2 )
−∞ i
kish hi z izh
Nei (kz , r)Nei (−kz , r0 )

+ 2 A (k 2 − k 2 )
kise ei z ize
(2.8.43)
Lei (kz , r)Lei (−kz , r0 ) ẑ ẑψei (kz , r)ψei (−kz , r0 )
− −
k 2 Aei k02 k02 Aei
Z ∞ X ẑ ẑψei (kz , r)ψei (−kz , r )
0
− dkz .
−∞ i
k02 Aei
Contour integrations can now be performed on the third term, giving rise to a pole contribu-
tion that cancels a similar contribution from the second term. The second term contains a
pole at k = 0 becuse the N function is proportional to 1/k as evident from (2.8.18). Hence,
the second term is proportional to 1/k 2 just as the third term. Since k = 0 for this pole
contribution, for the sake of discussion, we will call this the static pole.
It is to be noted that if the L functions are not used (which is erroneous) in the expansion,
a modal contribution from the second term due to this static pole will exist. This mode will
satisfy the dispersion relation kx2 +ky2 +kz2 = 0, making it an nonphysical mode in the presence
of a time-harmonic excitation. Fortunately, it is cancelled by a similar contribution from the
third term, the curl-free term. On first sight, it may seem strange that a contribution from the
divergence-free term should cancel one from the curl-free term. However, a closer examination
shows that this static pole contributes both to a divergence-free and curl-free field. The role
of the curl-free modes outside the source region is also discussed in [16].
Consequently, we have
X Mei (±kizh , r)Mei (∓kizh , r0 )
G(r, r0 ) = πi 2 A
i
kizh kish hi
(2.8.44)
Nei (±kize , r)Nei (∓kize , r0 ) ẑ ẑ 0 z > z0
+ − 2 δ(r − r ),
2 A
kize kise ei k0 z < z0.
The upper sign is chosen when z > z 0 and the lower sign is chosen when z < z 0 . The identity
Z ∞ X ψei (kz , r)ψei (−kz , r0 )
δ(r − r0 ) = dkz , (2.8.45)
−∞ i
Aei
has been used to simplify the last integral. All the vector wave functions in (2.8.44) evaluated
on the Ewald sphere. Hence, they are now physical wave functions which are solutions to
Maxwell’s equations.
The Dirac delta function part of the Green’s function in (2.8.44) has a ẑ ẑ component. This
is because we have performed the dkz integration first, letting kz → ∞, before letting the index
i goes to infinity. When kz → ∞, it also implies that we are looking at length scales in the ẑ
direction with infinite resolution before the length scales in the transverse direction. Hence,
the singularity in (2.8.44) is exactly the one obtained if one were to use a disk-shaped pill box
in performing the principal volume integral. Therefore, in order to obtain a unique, correct
solution when applying (2.8.44), one has to use a disk-shaped principal volume integral.8
Once the dyadic Green’s function of a hollow waveguide is known, the excitation of modes
due to an arbitrary current source in a waveguide can be found using (2.8.33). By substituting
Equation (2.8.44) into (2.8.33), we have
E(r) =
X 1
Z
− ωµπ 2 A Mei (±kizh , r)Mei (∓kizh , r0 ) · J(r0 )dr0
i
kizh kish hi
Z (2.8.46)
1
+ 2 A Nei (±kize , r)Nei (∓kize , r0 ) · J(r0 )dr0
kize kise ei
i
− ẑJz (r).
ω
The above integral is over the support of the current J(r0 ). If z > zmax
0 0
or z < zmin , where
zmax and zmin define the range of the support of J(r ) in the z-direction, i.e., J(r0 ) = 0 for
0 0 0
8 See [13], Chapter 7.

0 0
z > zmax and z < zmin , then the above can be written as
E(r) =
X 1
Z
− ωµπ 2 A M ei (±k izh , r) Mei (∓kizh , r0 ) · J(r0 )dr0
i
k izh k ish hi (2.8.47)
Z 0
1 z > zmax
+ Nei (±kize , r) Nei (∓kize , r0 ) · J(r0 )dr0 , 0
kize kise Aei z < znmn .
Notice that Mei and Nei denote the TE and TM modes of a waveguide, respectively. The
integrals are proportional to the excitation coefficients of the waveguide modes. From the
above, we can see that the excitation coefficient of a waveguide mode Ej is proportional to
Z
Excitation Coefficient ∼ Ej (r0 ) · J(r0 )dr0 . (2.8.48)
In other words, to excite a certain mode Ej in a waveguide maximally, there should be as

much projection of J onto Ej . To avoid the excitation of a mode Ej , the current J should be
chosen to be orthogonal to the mode Ej .
2.8.3 Excitation of Modes by a Filamental Current

Consider a probe in a waveguide as shown in Figure 2.15 [6]. A current in the probe will
produce an electromagnetic field that couples to the modes of the waveguide. We shall
discuss how to calculate the amplitudes of the excited waveguide modes. Let us assume that
the current on the probe is described by a current sheet
Js = ŷI0 δ(x − d). (2.8.49)
This probe current does not produce an Ez component of the electric field. Hence, we do
not expect the TM modes to be excited. However, the TE modes will be excited because the
probe current will produce an Hz component of the magnetic field. If, on the other hand,
the current on the probe is not a constant, there will be charge build up on the probe from
∇ · J − iωρ = 0. This charge will induce an Ez component of the electric field, coupling to
the TM modes [see Problem 2-11].
For the current probe described by (2.8.49), only TE modes will be excited and we can
write the field as
X mπx nπy ikz z

 +
Hmn cos( ) cos( )e , z>0

 a b
m,n
Hz = X mπx nπy −ikz z (2.8.50)

 H −
cos( ) cos( )e , z < 0.

 mn
a b
m,n
q
mπ 2 nπ 2
where kz = k02 − a − b . The boundary condition across a current sheet is that
ẑ × [H(z+) − H(z−)] = Js , ẑ × [E(z+) − E(z−)] = 0, (2.8.51)

x
d a
Figure 2.15: Excitation of a rectangular waveguide by a current probe.
where ẑ × H gives the transverse field component of the magnetic field. Since Js has no y
±
variation, we expect ẑ × H not to have any y variation. Hence, we can assume that Hmn = 0,
n 6= 0. Assuming this, we can derive the transverse field components from (2.8.50) using
(2.1.7) and (2.1.8), giving
X mπx
+ ikz mπ
Hs = − Hm0 2
x̂ sin eikz z , z > 0, (2.8.52a)
m
ks a a
X ikz mπ mπx
−
Hs = Hm0 2
x̂ sin e−ikz z , z < 0, (2.8.52b)
m
ks a a
a ) + ( b ) . Applying the first boundary condition, we get
X ikz mπ mπx
+ −
− 2
ŷ sin (Hm0 + Hm0 ) = ŷI0 δ(x − d). (2.8.53)
m
ks a a
Therefore, from the above boundary condition, only the TEm0 modes will be excited. Equa-
+ −
tion (2.8.53) alone is not sufficient to determine the unknowns Hm0 and Hm0 . Therefore,
we need to apply the second boundary condition in (2.8.51). However, if one were to note
that Hz (z+) = Hz (z−) (a boundary condition which is the subset of the second boundary
condition in (2.8.51), we note that
+ −
Hm0 = Hm0 . (2.8.54)
Using (2.8.54) in (2.8.53) will uniquely determine the solution of (2.8.53). Consequently,
(2.8.53) becomes
X∞
2ikz + mπ mπx
− 2 Hm0 sin = I0 δ(x − d). (2.8.55)
m=1
ks a a
A simple Fourier series analysis shows that

+ −ks2 I0 mπd
Hm0 = sin . (2.8.56)
ikz mπ a
p
where ks2 = ( mπ 2
a ) , and kz = k02 − ks2 . This analysis shows that only TE modes with no ŷ-
variation of the field are excited. This is because the probe current in (2.8.49) generates a field
+
with no y-variation. Hm0 is sometimes known as the excitation coefficient of the mode. For
a particular TEm0 mode, we can change d to alter the amplitude of the excitation coefficient.
For example, letting d/a = 1/2 will maximize the excitation coefficient of the TE10 mode,
while the TEm0 mode with m even will not be excited.
2.9 Modes of a Hollow Waveguide of Arbitrary Cross-

Section
If a metallic waveguide has an arbitrary cross-section, whose shape does not fall on any of
the curvilinear coordinates, we will have to find the modes numerically. We have learned that
the TE and TM modes are characterized by solutions of the following equations:
∂Hz
TE (∇2s + ks2 )Hz (rs ) = 0, =0 on C, (2.9.1)
∂n
TM (∇2s + ks2 )Ez (rs ) = 0, Ez = 0 on C. (2.9.2)
2.9.1 Differential Equation Method

The above are eigenvalue problems. Eigenvalue problems can be converted into variational
problems by defining Rayleigh quotient
hφ, ∇2s φi
ks2 = − (2.9.3)
hφ, φi
The numerator can be simplified by integration by parts.

Z
2
hφ, ∇s φi = dSφ∇2s φ
S
Z I
= − dS∇s φ · ∇s φ + dln̂ · (φ∇s φ) (2.9.4)
S C
With the choice of appropriate boundary conditions, the last term can be made to vanish.
Hence, the Rayleigh quotient becomes
h∇s φ, ∇s φi
ks2 = (2.9.5)
hφ, φi
The above can be shown to be variational, meaning that a first order error in φ gives rise to
a second-order error in ks2 . By letting
φ = φe + δφ (2.9.6)
ks2 = kse
2
+ δks2 (2.9.7)
2
where φe and kse are exact values for the function and the eigenvalue, respectively. Then,
after cross-multiplying, and taking the first variation, we have
2
kse 2hφe , δφi + hφe , φe iδks2 = −2h∇s φe , ∇s δφi (2.9.8)
After doing integration by parts on the right-hand side, the term involving δφ cancel each
other, and hence δks2 = 0. In other words, the exact eigenfunctions and eigenvalues of the
problems (2.9.1) and (2.9.2) are at the stationary values or stationary points of the Rayleigh
quotient (2.9.5).
Rayleign-Ritz Method
In this method, we let
N
X
φ= an φn (2.9.9)
n=1
and pick an to make (2.9.5) stationary. By using (2.9.9) in (2.9.5), we arrive at

P
n,n0 an an Mnn at · M · a
0 0
2
ks = P = t (2.9.10)
n,n0 an an Bnn a · B̄ · a
0 0
where M and B are symmetric matrices, Mnn0 = h∇s φn , ∇s φn0 i, Bnn0 = hφn , φn0 i. The
equation above, (2.9.10), has stationary points. When we increase the number of unknowns
in (2.9.9), the stationary points of (2.9.10) will approach the exact stationary points and
hence, the exact answers. Meanwhile, (2.9.9) will approach the exact eigenfunction. We
assume that
a = a0 + δa, ks2 = ks0
2
+ δks2 (2.9.11)
where a0 is the value that will optimize (2.9.10), and ks0 2 is the optimal value of ks 2 . In
other words, we want a0 to be at the stationary point of (2.9.10). In this case, δks2 will be
zero. Subsequently, after cross-multiplying, and taking the first variation, (2.9.10) becomes
ks0 2 2δat · B · a0 + δks 2 a0 t · B · a0 = 2δat · M · a0 (2.9.12)
In order for the δks 2 to be zero, so that a0 represents the optimal solution, we require that
the δat terms cancel each other. Then it is necessary that
M · a0 = ks0 2 B · a0 (2.9.13)
The above is the matrix eigenvalue form which we can solve for a0 and ks0 2 . Once a0 is
found, the eigenfunction φ is found via (2.9.9).
Myriads of methods can be used to choose φn in (2.9.9). If the cross section of the
waveguide is arbitrary, it is more practical to triangulate the cross section and pick subdomain
basis functons such as pyrimidal functions. Such a method of solution is known as the finite
element method. Finite element method (FEM) is vastly popular in solving many differential
equation problems [18–21].
In the choice of basis functions in (2.9.9), we can pick the functions to satisfy the homo-
geneous Dirichlet or Neumann boundary condition to make the last term in (2.9.4) vanish.
Figure 2.16: Two-dimensional FEM mesh for the waveguide.
If we pick basis functions whose values float at the contour C, the solutions will satisfy the
homogeneous Neumann boundary condition. Hence, the homogeneous boundary condition is
also known as the natural boundary condition [13, p. 308].
2.9.2 Integral Equation Method

A more rigorous way of finding the modes of an arbitrarily shaped waveguide is to use the
integral equation approach [22–24]. For this, we need a Green’s function which is the solution
to the equation
(∇2s + ks2 )g(rs , r0s ) = −δ(rs − r0s ), rs ∈ S1 , r0s ∈ S. (2.9.14)
Multiplying (2.9.1) by g(rs , r0s ) and (2.9.14) by Hz (rs ), subtracting the two resultant equa-
tions, and integrating over S, we obtain
Z
[g(rs , rs )∇2s Hz (rs ) − Hz (rs )∇2s g(rs , r0s )]dS = Hz (r0s ), r0s ∈ S (2.9.15)
S
But since g∇2s Hz − Hz ∇2s g = ∇ · (g∇Hz − Hz ∇g), Gauss’ theorem can be used to convert
(2.9.15) into
I
dln̂ · [g(rs , r0s )∇Hz (rs ) − Hz (rs )∇g(rs , r0s )] = Hz (r0s ), r0s ∈ S. (2.9.16)
C
∂
Since n̂ · ∇s Hz (rs ) = ∂n Hz (rs ) = 0 for rs ∈ C from Equation (2.9.1) we have
I
− dln̂Hz (rs )n̂ · ∇g(rs , r0s ) = Hz (r0s ), r0s ∈ C. (2.9.17)
We have imposed r0s ∈ C in the above so that it is now an integral equation with the unknown
Hz (r0s ), r0s ∈ C. Since rs can be equal to r0s in the integral, the singularity of the Green’s
function n̂ · ∇g(rs , r0s ) makes a straight forward evaluation of the above integral divergent. A
principal value integral has to be taken to obtain a convergent integral [13, p. 455].
Figure 2.17: Choice of basis functions for Hz (l). Pi (l) is a pulse basis function.
In the above, Hz (rs ) can be parameterized as a function of l where l is a variable defining

the contour C of the waveguide. For simplicity, we can expand Hz (l) in terms of pulse basis
P
N
functions, i.e., Hz (l) = ai Pi (l). Then, (2.9.17) becomes
i=1
li + ∆
N
X Z 2 N
X
ai dln̂ · ∇g(l, l0 )pi (l) = − ai pi (l0 ). (2.9.18)
i=1 i=1
li − ∆
2
A matrix equation can be obtained by point matching and fixing the above equation at l0 = lj ,
j = 1, . . . , N, yielding
N
X
Aji aj = 0, j = 1, . . . , N, (2.9.19)
i=1
where
li + ∆
Z2
Aji = dln̂ · ∇g(l, lj )pi (l) + pi (lj ). (2.9.20)

li − ∆
2
Equation (2.9.19), hence, reduces to a matrix equation
A(ks ) · a = 0. (2.9.21)
A nontrivial solution exists for a only if
det (A(ks )) = 0. (2.9.22)
Here, A is a function of ks because the Green’s function in Equation (2.9.14) is a function of

ks . The above can be satisfied only at certain values of ks . At these values, (2.9.21) has a
non-trivial null space solution, and hence a is nonzero, and can be found.
The derivation so far requires g(rs , r0s ) to be a solution of (2.9.14) for rs ∈ S, and r0s ∈ S.
A simple Green’s function that satisfies this requirement is
i (1)
g(rs , r0s ) = g(rs − r0s ) = H (ks |rs − r0s |). (2.9.23)
4 0
A similar integral equation can be derived for TM polarization which is
I
dlg(rs , r0s )n̂ · ∇s Ez (rs ) = 0 r0s ∈ C. (2.9.24)
C
This can also be used to find the waveguide modes of an arbitrarily shaped waveguide. In
the above, there is no derivative on the Green’s function. Hence, this integral equation is less
singular compared to the one for TE polarization.
2.9.3 Ad Hoc Method

We describe an ad hoc method for solving for the waveguide modes, but this method is not
rigorous. It can only yield the modes satisfactorily if the shape of the waveguide is not too
oblong, or if the wall of the waveguide does not have sharp corners.
One way to solve (2.9.1) and (2.9.2) is to expand the scalar field Hz or Ez in terms of
functions which are known to be solutions of (2.9.1) or (2.9.2), but do meet the specified
boundary conditions. For example, we can let
X
Hz (rs ) = an ψn (rs ), (2.9.25)
n
9
where ψn (rs ) could be
ψn (rs ) = Jn (ks ρ)einφ . (2.9.26)
ψn (rs ) is clearly a solution to (2.9.1), but the boundary condition is not met. In order to
satisfy the boundary condition, we require that the normal derivative of Hz = 0, or
∂Hz X ∂ψn (rs )
= an = 0 on C. (2.9.27)
∂n n
∂n
9 The basis given by (2.9.26) is actually not complete in an arbitrarily shaped waveguide. This method is
hence akin to the method of Rayleigh’s hypothesis [13].

Figure 2.18: A waveguide of arbitrary cross-section.
The general method to solve (2.9.27) is to weight it with a test function and integrate
over C, e.g.,
X ∂ψn

an φm , = 0, (2.9.28)
n
∂n
where Z
∂ψn ∂ψn
φm , = dlφm (rs ) = Amn (ks ) (2.9.29)
∂n ∂n
C
is only a number dependent on ks and the indices m and n. Equation (2.9.28), hence, reduces
to a matrix equation
A(ks ) · a = 0. (2.9.30)
A nontrivial solution exists for a only if

det A(ks ) = 0. (2.9.31)
We can search Equation (2.9.31) numerically for the values of ks that satisfy the equation.
These are the eigenvalues of the problem for the TE modes. The eigenvector a can be found
from (2.9.30) at these values of ks . With the knowledge of a, we can construct Hz , the
eigenfunction.
For the TM problem, similar analysis gives rise to an A matrix given by
Z
Amn (ks ) = dlφm (rs )ψn (rs ). (2.9.32)
C
Another way of solving (2.9.27) is to first assign N points on the perimeter of the wave
guide C and evaluate the Equation (2.9.28) at these location. Equation (2.9.27) then becomes
X ∂ψn (rm )
an = 0, m = 1, . . . , N. (2.9.33)
n
∂n
rN-1 rN r1 r2 r3
µ, ε
Figure 2.19: The solution of a waveguide problem via the method of point-matching.
The above is a matrix equation similar to (2.9.30) where
∂ψn (rm )
Amn (ks ) = . (2.9.34)
∂n
Similar analysis for the TE modes gives
Amn (ks ) = ψn (rm ). (2.9.35)
This particularly simple way of solving Equation (2.9.27) is known as the point matching
technique.
Problem 2-1: Many transmission line systems have more than two conductors, like your
home telephone line, computer bus etc. If you have an N conductors (a general cylindrical
waveguide with translational invariance in z direction) embedded in a homogeneous isotropic
medium, how many independent TEM (Transverse Electromagnetic) modes can propagate
down this line.
Problem 2-2: The reflection of a plane wave by a half space has a closed form solution.
Write down the solutions of the reflection of plane TE and TM waves from a half space. Now
make the lower half space a metallic conductor whose conductivity is high but not infinite.
Deduce a relationship between the tangential component of the electric field and the magnetic
field at the metallic surface, i.e.,
Et = Zm Ht .
Zm is also known as the surface impedance of the metallic conductor. It is useful for calcu-
lating the power absorbed by the metallic conductor.
Problem 2-3: For a coaxial transmission line with an inner conductor of radius a and an
outer conductor of radius b, find the electric potential φs and the magnetic potential ψs . Is
ψs a single value or a multi-value function for each point in space? Comment on this.
There are two ways to calculate the attenuation in a transmission line when the conductor
is not perfect. One way is to first calculate the series resistance per unit length, R, and then
find the attenuation constant from the formula (k 0 + ik 00 )2 = −Y Z where Y is the shunt
admittance per unit length, and Z is the series impedance per unit length, and k 00 is the
attenuation constant. Another way is to find Pd , the power dissipated per unit length by
integrating the time average Poynting vector hE × Hi over the surface of the conductors.
Since the tangential electric field is not zero anymore on a non-perfectly conducting metallic
surface, the time average Poynting vector is not zero on the metallic surfaces. After finding
the Pd , one can calculate the attenuation constant from
Pd
k 00 =
2P0
where P0 is the time average total power flow down the transmission line. You can use the
surface impedance found from Problem 2 to calculate the power absorbed by the metallic
conductor.
(a) Find the attenuation constant for a coaxial line using these two methods, and show that
they are the same.
(b) For a 50 ohm coaxial transmission line with an outer radius of .5 cm, filled with teflon
with = 20 , calculate the loss due to attenuation in terms of dB/Km at 20 KHz and 1
GHz, if we assume that the conductor is made out of copper with σ = 5.8 × 107 mho/m.
Problem 2-4: Can a single piece of conductor embedded in a homogeneous medium support
a TEM mode? Discuss why and why not.
Problem 2-5:
(a) Explain why the line capacitance (farad/m) of the TEM mode in a transmission line
can be found by solving an electrostatic problem.
(b) Given the knowledge of the line capacitance, show that it has to be of the form C = K
where K is a dimensionless, geometry dependent factor.
(c) Show that now if there is dielectric loss such that = 0 + iσ/ω, then the admittance per
unit length to be used in the telegraphists equation (for lossy line) is Y = −iω0 K +σK.
Hence, G = σK.
2
Problem 2-6: The eigenvalue kis of the partial differential equation (∇2s + kis
2
)ψiz = 0 is
always real, and can be proven as follows:
2 ∗ ∗2
(a) Assume that kis is complex so that ψiz is a solution of (∇2s + kis ∗
)ψiz = 0. Show that
∗ ∗2
ψiz ∇2s ψiz − ψiz ∇2s ψiz
∗
= (kis 2
− kis )|ψiz |2 .
(b) Assume that ψiz is either Eiz or Hiz , integrate the equation in part (a) over the cross-
section of a metallic waveguide and show that
Z
∗2
(kis − kis ) |ψiz |2 dS = 0.
2
S
2
From the above, explain why kis has to be real.
Problem 2-7: Prove that Z
dSHis · Hjs = 0, i 6= j
S
for any two distinct modes for a homogeneously filled closed waveguide. S in the above refers
to the cross-sectional area of the waveguide that is permeated by the field.
Problem 2-8:
(a) Prove that Eis and Ejs are always orthogonal if one field is coming from a TM (TE)
mode while the other field is coming from a TE (TM) mode. Hint: Express these fields
in terms of the longitudinal components of the fields.
(b) Prove that Z
(Eis × H∗js ) · ẑdS = 0
S
for i 6= j.
(c) From the above, explain how the concept of energy orthogonality and power orthogo-
nality follow.
Problem 2-9: For the TE10 mode of a rectangular waveguide, find its attenuation constant
due to wall loss. Do the same for the TE11 and the TE01 modes of a circular waveguide.
Sketch the attenuation constants as functions of frequency. Which mode has the lowest loss
at high frequencies?
Problem 2-10: Derive the complete expression for the fields of the TE11 mode of a rectan-
gular waveguide and sketch the field patterns for both the E and H fields in the cross-section
of the waveguide. Do the same for the TM11 mode of the rectangular waveguide, and the
TE12 mode of the circular waveguide. (A good feel for the field pattern of the waveguide
mode is essential in determining how to excite it with a probe, see Equation (2.8.48).)
Problem 2-11:
A probe with a current distribution represented by a current sheet given by
πy
Js (r) = ŷδ(x − a/2) sin( )
b
is placed in a waveguide. Calculate the excitation coefficients of all the modes excited by this
probe.
Problem 2-12: Certain modes of metallic waveguides have decreasing loss for increasing
frequency.
(a) For the TE10 mode of a rectangular waveguide, find its attenuation constant due to
wall loss. Do the same for the TE11 and the TE01 modes of a circular waveguide. Hint:
Integral of product of Bessel functions is needed in the solution. Equation (11.4.5) of
Abramowitz and Stegun comes in useful here.
(b) Sketch the attenuation constants as functions of frequency. Which mode has the lowest
loss at high frequencies?
(c) For the TE1 mode of a parallel plate waveguide, does its attenuation constant increase
or decrease with frequency?
(d) From the above, explain the feature of the “magic” mode that has decreasing loss for
increasing frequency.
Problem 2-13:
(a) Give an intuitive explanation as to why the TE11 mode of a circular waveguide has a
lower cutoff frequency than the TE01 mode.
(b) Explain why the TE00 mode of a rectangular waveguide cannot exist.
Problem 2-14: Prove the mutual orthogonality of the vector wave functions in a hollow
waveguide, that is Z
dV M he i (kz , r) · N he j (−kz0 , r) = 0,
Z
dV L he i (kz , r) · N he j (−kz0 , r) = 0,
Z
dV L he i (kz , r) · M he j (−kz0 , r) = 0,
for all i, j, kz and kz0 .
Problem 2-15:
For the coaxial waveguide shown:
(a) Write down the guidance conditions for all the modes in the waveguide using Bessel and
Neumann functions. Is the TEM mode a special case of these guidance conditions?
(b) Derive the M, N, and L vector wave functions for the electric and magnetic fields of
this waveguide. (Hint: Use sine and cosine functions for the φ dependence.)
(c) Write down the orthogonality relationships for the vector wave functions with the Aei
and Ahi (defined in the text) derived explicitly.
(d) For a source in the circular waveguide described by

a+b
J(r) = ẑI`δ ρ − δ(φ)δ(z)/ρ,
2
find the E field in terms of the vector wave function.
Problem 2-16: Go through the exercise of deriving the dyadic Green’s function of a waveg-
uide. Most of the derivation is already outlined in the text. Fill in the details.
Problem 2-17: For a rectangular waveguide:
(a) Write down the guidance conditions for all the modes in the waveguide.
(b) Derive the M, N and L vector wave functions for the electric and magnetic fields of
this waveguide.
(c) You are given three Hertzian electric dipoles. Describe how you would place these
dipoles in a rectangular waveguide and weight their amplitudes so that the TE10 is
excited but the TE20 and TE30 modes would not be excited (note: you may excite all
the other modes).
Problem 2-18:
(a) Derive the integral equation for finding the modes of the TM polarization of a hollow
waveguide: I
dlg(rs , r0s )n̂ · ∇s Ez (rs ) = 0, r0s ∈ C.
C
(b) Describe how you would solve the above equation numerically.
a/2
a
TE10
1 2
TM01 TE11 TE21

TE20 TM11 TM21
2 3
TE31
TE30 TM31
3 4
TE41
TE22
TE40 TE12 TM41
TE02 TM12 TM22
4 5
TE50
TE32 TE51 TE42
TM32 TM51 TM42
5 6
TE61
TE23
TE60TE13 TM61 TE52 TE33
TE03TM13 TM23 TM52 TM33 TE70
6 7
Normalized modal cutoff frequencies for a 2:1
rectangular waveguide
(1) TE10 (2) TE01 (3) TE20 (4) TE11 (5) TM11 (6) TE 21
(7) TM21 (8) TE30 (9) TE31 (10) TM31 (11) TE40 (12) TE02
(13) TE12 (14) TM12 (15) TE41 (16) TE22 (17) TM41 (18) TM22
(19) TE50 (20) TE32 (21) TM32 (22) TE51 (23) TM51 (24) TE42
(25) TM42 (26) TE60 (27) TE03 (28) TE13 (29) TM13 (30) TE61
(31) TE23 (32) TM61 (33) TM23 (34) TE52 (35) TM33 (36) TE33
Figure 2.20: Cutoff frequencies (normalized with ω10c = 1) and field plots of the first 36
modes of a rectangular waveguide (courtesy of A. Greenwood).
TE11 TM01 TE21
1 2
TE41
TM11 TE12
TE01 TE31 TM21 TM02
2 3
TM31 TE02
TE51 TE22 TM12
3 4
TE13
TE61 TE71 TM51
TM41 TE32 TM22 TM03
4 5
TM61
TE81 TE23 TM13
TE42 TM32 TE03 TE52 TE91
5 6
Normalized model cutoff frequencies for a circular waveguide.
(1) TE11 (2) TM01 (3) TE21 (4) TM11 (5) TE 01 (6) TE 31
(7) TM21 (8) TE41 (9) TE12 (10) TM02 (11) TM31 (12) TE51
(13) TE22 (14) TE02 (15) TM12 (16) TE61 (17) TM41 (18) TE32
(19) TM22 (20) TE13 (21) TE71 (22) TM03 (23) TM51 (24) TE42
(25) TE81 (26) TM32 (27) TM61 (28) TE23 (29) TM13 (30) TE03
Figure 2.21: Cutoff frequencies (normalized with ω11c = 1) and field plots of the first 30
modes of a circular waveguide (courtesy of A. Greenwood).
Figure 2.22: Problem 2-1

Bibliography
[1] H.R.L. Lamont, Wave Guides, third ed., Methuen, London, 1950.
[3] K.G. Budden, The Wave-Guide Mode Theory of Wave Propagation, Englewood Cliffs,
NJ, Prentice Hall, 1961.
[6] J.A. Kong, Electromagnetic Wave Theory, EMW Publishing, Cambridge, MA, 2000.
[7] D.M. Pozar, Microwave Engineering,, John Wiley & Sons, 2000.
[8] C.S. Lee, S.W. Lee, and S.L. Chuang, “Plot of modal field distribution in rectangular and
circular waveguides,” IEEE Trans. Micro. Theory Tech., vol. MTT-33, no. 3, pp. 271-274,
March 1985.
[9] J.W. Archer, “TE0n-mode filter for VLA circular waveguide system,” Electronic Letters,
vol. 15, pp. 343-345, June 7, 1979.
[10] J.A. Stratton, Electromagnetic Theory, McGraw-Hill, NY, 1941.
[11] W.W. Hansen, “A new type of expansion in radiation problems,” Phys. Rev., 47, 139143,
1935).
[12] R.E. Collin, “On the incompleteness of E and H modes in waveguides,” Can. J. Phys.,
51, 1135-1140, 1973.
[14] C.T. Tai, Dyadic Green’s Functions in Electromagnetic Theory,, Intext Pub., NY, 1971.
[15] C.T. Tai, “On the eigenfunction expansion of dyadic Green’s functions,” Proc. IEEE, 61,
480-481, 1973.
89
[16] W.A. Johnson, A.Q. Howard, and D.G. Dudley, “On the irrotational component of the
electric Green’s dyadic,” Radio Sci., 14, 961-967, 1979.
[17] A.D. Yaghjian, “Electric dyadic Green’s functions in the source region,” Proc. IEEE, 68,
248-263, 1980.
[18] O.C. Zienkiewicz, The finite element method in engineering science, McGraw-Hill, 1971.
[19] P.P. Silvester and R.L. Ferrari, Finite Elements for Electrical Engineers, first edition,
Cambridge University Press, 1983.
[20] J.M. Jin, The Finite Element Method in Electromagnetics, Wiley, 1993.
[21] J.L. Volakis, A. Chatterjee, L.C. Kempel, Finite Element Method Electromagnetics: An-
tennas, Microwave Circuits, and Scattering Applications, Wiley, 1998.
[22] R.F. Harrington, Field computation by moment methods, Macmillan, 1968.
[23] W.C. Chew, J.M. Jin, E. Michielssen, and J.M. Song, (editors), Fast and Efficient Algo-
rithms in Computational Electromagnetics, Artech House, Boston, MA, 2001.
[24] W.C. Chew, M.S. Tong, and B. Hu, Integral Equations Methods for Electromagnetic and
Elastic Waves, Morgan & Claypool, 2008.
Chapter 3
Inhomogeneously Filled
Waveguides
The theory of inhomogeneously filled waveguides covers a large class of waveguides [1, 3–6].
Waveguides are filled with inhomogeneous material to give the waveguide a certain property.
For instance, phase-shifters, polarizers, and attenuators are made out of inhomogeneously
filled waveguides. Often, waveguides are filled inhomogenously with ferrite material to make
non-reciprocal waveguides. To make the fabrication of waveguides simpler, many waveguides
are also filled with inhomogeneous materials. Optical waveguides almost exclusively comprise
inhomogeneously filled material.
3.1 The Need for Hybrid modes

When a hollow waveguide is filled with inhomogeneity, the most general case is the existence
of the hybrid modes. In these modes, the TEz and TMz modes are coupled. Only for cases
with special symmetry are they decoupled. The reason is that both TEz and TMz fields are
needed to match the boundary condition at the dielectric interface.
Assume that we only have TEz field inside the waveguide. Then,

1 ∂ 1
Hs = 2 ∇s Hz = 2 ikz ∇s Hz (3.1.1)
ks ∂z ks
The boundary condition for Hz is that it is continuous, or H1z = H2z at the interface between
two dielectric regions. Furthermore, we require that,
n̂ × H1s = n̂ × H2s (3.1.2)
implying that at the dielectric interface,

1 1
2 ikz n̂ × ∇s H1z = k 2 ikz n̂ × ∇s H2z
k1s
(3.1.3)
2s
91
Due to phase matching, kz is the same in all regions. If H1z is continuous at a dielectric
interface, then n̂ × ∇s H1z = n̂ × ∇s H2z also, since these are tangential derivatives. Therefore,
(3.1.3) cannot be satisfied since k1s 6= k2s in general. In order to satisfy the boundary
condition, (3.1.1) has to be augmented with the contribution from TMz field.
However, under special circumstances, Equation (3.1.3) can be satisfied if:
(1) kz = 0, implying that, Hs = 0 and H = ẑHz only;
(2) n̂ × ∇s Hiz = 0 at the interface, implying symmetry that exists for certain modes, for
instance, in a dielectric slab or in an axi-symmetric geometry such as a circular optical
fiber;
(3) The surface is a PMC surface so that Hz = 0, n̂ × Hs = 0;
(4) The surface is a PEC surface so that the boundary condition is for n̂ × E = 0, giving
rise to n̂ · ∇Hz = 0 on the surface.
Similarly, arguments above apply to the TMz field. Under these special circumstances, the
field is not depolarized at the interface. Namely, if the field is TEz or TMz before impinging on
the interface, the scattered field off the interface remains the same as the original polarization.
3.2 Derivation of Pertinent Equation

When a uniform waveguide is filled with inhomogeneous materials, the guided modes of
the structure cannot be decomposed into TE and TM modes, except for some very special
cases. In other words, the Ez and Hz components of the fields are always coupled together.
Such modes are also called the hybrid modes. This coupling can be shown from Maxwell’s
equations, which imply
∇ × µ−1 ∇ × E − ω 2 E = 0. (3.2.1)
The above governs the E-field propagating in an inhomogeneously filled waveguide. Assuming
eikz z dependence, we can decompose
E = Es + Ez , ∇ = ∇s + ẑikz , (3.2.2)
for a guided mode in the waveguide. With the use of the above,
∇ × µ−1 ∇ × E
= ∇s × µ−1 ∇s × Es + ikz ẑ × µ−1 ∇s × Ez − kz2 µ−1 ẑ × ẑ × Es (3.2.3)
−1 −1
+ ∇s × µ ∇s × Ez + ikz ∇s × µ ẑ × Es .
Because µ is a function of rs , it does not commute with ∇s . The last two terms in (3.2.3)
are directed in the z direction. Therefore, by equating the z components in (3.2.1), we have
∇s × µ−1 ∇s × Ez + ikz ∇s × µ−1 ẑ × Es − ω 2 Ez = 0. (3.2.4)

Inhomogeneously Filled Waveguides 93
µµ, ε
µ 1, ε 1
Figure 3.1: An inhomogeneously filled waveguide.
3.2.1 Ez -Hz Formulation

∇s × µ−1 ∇s × Ez = −∇s · µ−1 ∇s Ez , (3.2.5a)
∇s × µ−1 ẑ × Es = ẑ∇s · µ−1 Es . (3.2.5b)
1
Using
i
Es = [kz ∇s Ez + ωµ∇s × Hz ], (3.2.6)
ks2
p
where now ks2 = ω 2 µ − kz2 is a function of r, we can rewrite (3.2.4) as

ikz iω
∇s · µ−1 ∇s Ez − ikz ∇s · ∇ E
s z + ∇ s × H 2
z + ω Ez = 0. (3.2.7)
µks2 ks2
p
In the above, ks = k 2 − kz2 . Since k is a function of rs , so is ks . Therefore, in general, the
equation governing Ez is coupled to Hz as well. By duality principle, the equation governing
Hz , which is also coupled to Ez , is

−1 ikz iω
∇s · ∇s Hz − ikz ∇s · ∇s Hz − 2 ∇s × Ez + ω 2 µHz = 0. (3.2.8)
ks2 ks
However, if µ and are constants, then ks2 is a constant, and ∇s · ∇s × Ez = ∇s · ∇s × Hz =

0, the Ez and Hz equations are decoupled again. Therefore, for general µ and which
are inhomogeneous, the TE and the TM fields in a waveguide are coupled. Exceptions are
sometimes found, e.g., in the axial symmetric modes of a circular optical fiber.
The above equations show that the TE and TM fields have to co-exist in the waveguide,
hence giving rise to hybrid modes. However, they are not cast in terms of eigenvalue problems.
1 This equation is valid in an inhomogeneously filled waveguide, because the only assumption made in
deriving it is that the field varies as exp(ikz z). However, ks is not constant anymore.
We will derive equations from which the eigenvalues and eigenfunctions of the waveguide can
be derived.
3.2.2 Transverse Field Formulation

The transverse components of the field in Equation (3.2.1) can be isolated to obtain
∇s × µ−1 ∇s × Es + ikz µ−1 ẑ × ∇s × Ez + kz2 µ−1 Es − ω 2 Es = 0. (3.2.9)
Using ∇ · E = 0, we have
ikz Ez = −−1 ∇s · Es . (3.2.10)
Therefore
ikz ẑ × ∇s × Ez = ikz ∇s Ez = −∇s −1 ∇s · Es . (3.2.11)
Consequently, an equation that governs the transverse electric field for the i-th mode is
µ∇s × µ−1 ∇s × Eis − ∇s −1 ∇s · Eis − k 2 Eis + kiz

2
Eis = 0, (3.2.12)
where k 2 = ω 2 µ. For reason to be explained later, we can multiply the above by z× to get
z × µ∇s × µ−1 ∇s × Eis − z × ∇s −1 ∇s · Eis − k 2 z × Eis + kiz

2
z × Eis = 0, (3.2.13)
In addition, an equation for the transverse magnetic field for the j-th mode is easily obtained
by invoking duality, namely,
z × ∇s × −1 ∇s × Hjs − z × ∇s µ−1 ∇s · µHjs − k 2 z × Hjs + kjz

2
z × Hjs = 0. (3.2.14)
In the above, kz2 are the eigenvalues since it is constant throughout the uniform waveguide.
Due to the phase matching condition, kz is a constant everywhere in an inhomogeneous
waveguide. We need to solve either Equation (3.2.13) or (3.2.14) since they can be shown to
be transpose of each other.
Notice that whether if we describe the modes in a waveguide with Equation (3.2.7) and
(3.2.8), or Equations (3.2.13) and (3.2.14), only two components of the electric field or the
magnetic field are required to describe the modes in an inhomogeneously filled waveguide.
This is because ∇ · B = 0 and ∇ · D = 0 and hence, not all three components of the B field
or the D field are independent of each other.
As a last note, as shall be shown later, (3.2.13) and (3.2.14) are transpose of each other.
Hence, eigenfunctions from these equations are mutually orthogonal.
3.2.3 Physical Interpretation of the Depolarization Effect

The fact that the Ez and Hz waves are in general coupled at a dielectric interface is also
known as depolarization effect. An Hz (TE to z) wave incident at a dielectric rod, in general,
generates both Hz and Ez (TM to z) waves, and hence, causes the depolarization of the wave.
This happens conversely for Ez wave incident on a dielectric rod. This depolarization effect
occurs only for Ez or Hz waves that vary as a function of z (kz 6= 0). The depolarization effect
E
H
Magnetic Field k
Generated by the
Eddy Currents
Polarization
Current
Figure 3.2: Polarization current and magnetic field generated by the induced current on a
dielectric rod illuminated by an Ez wave.
disappears when the waves do not vary as a function of z (kz = 0), or when the scatterer is
a cylindrical perfect electric conductor (PEC) or a perfect magnetic conductor.
A guided mode can be thought of as a wave bouncing around in a waveguide such that
phase coherence or constructive interference occurs. The condition for phase coherence or
constructive interference is precisely the guidance condition of the waveguide modes. In the
case of a hollow waveguide, the waves are bouncing off a PEC cylindrical surface and hence,
the polarization purity can be maintained. Therefore, the mode can be either purely Ez or Hz
type. When a waveguide is inhomogeneously filled, the waves have to bounce off a dielectric
rod, and in general, Ez or Hz mode purity cannot be maintained. These modes are termed
the hybrid modes.
We can argue by contradiction, except for special symmetric cases, that only a hybrid
mode consisting of TE and TM waves is possible. When an Ez polarized (TM to z) wave is
obliquely incident on a dielectric slab as shown in Figure 3.2, polarization purity can be kept
if the slab is infinitely wide coming out of the paper. The magnetic field will be horizontal in
the slab with the electric polarization current flowing around it. The magnetic field alternates
in its polarity as one moves in the z direction. The k vector lies in the plane of the paper
both in the air and in the slab. Now assume that the slab is truncated so that it extends in
finitely into the paper as well as out of the paper. Assume that polarization purity is still
preserved. Then at the interface at the truncated surfaces, the k vectors are both parallel to
the surface inside the dielectric as well as in the air. This is an impossibility since the phase
velocity of the waves on two sides of the interface are different and the boundary condition
can never be met since phase matching is violated. Hence, the k vector has to “tilt” in order
to satisfy the boundary condition, introducing the z component of the magnetic field.
Alternatively, we can consider an Hz incident wave (TE to z). Let us assume that the
field remains TE inside the dielectric rod and see that it will lead to a contradiction. If this
is the case, the transverse electric field will at least induce polarization currents flowing in
the xy directions. When these currents meet a dielectric interface, polarization charges are
induced at the interface. Because of the z-variation of the incident field, these charges must
be sign-changing in the z-direction. The electric field has to turn around due to the different
phase velocity it has in the air compared to the dielectric. Therefore, an Ez field must exist
due to these charges.
The above explains the general depolarization of the wave. However, certain symmetrical
Polarization Polarization
Current Charges H
E
+ _
k
_ +
Electric Field
Generated by
the Charges
+ _
Figure 3.3: Induced polarization currents and charges in a dielectric rod if the field remains
TE to z when the rod is illuminated by an Hz wave.
rods may not depolarize certain incident field. For instance, an infinite dielectric slab will not
depolarize a TE (or TM) wave where the E (or H) field is aligned parallel to the dielectric
interface. A circular dielectric rod will not depolarize an axially symmetric Eφ or Hφ polarized
TE or TM wave incident on it. In the above discussion, we are referring to TE to z or TM
to z waves.
3.2.4 Mode Orthogonality

In an inhomogeneous waveguide, the differential operators in (3.2.13) and (3.2.14) are not
self-transpose (or symmetric) which is unlike a homogeneously filled cylindrical waveguide.
Therefore, general orthogonality between Eis and Ejs , or His and Hjs does not exist. How-
ever, mode orthogonality exists between Eis and Hjs . We can prove the orthogonality of Eis
and Hjs for a general, inhomogeneous, anisotropic waveguide using the Lorentz reciprocity
theorem.
Consider two waveguides with identical walls but different anisotropic, inhomogeneous
media: one waveguide is filled with µ, , while the other one is filled with µt (rs ), t (rs )
where the superscript t stands for transpose. It can be shown that
Z
(Eis × Hjs ) · ẑdS = 0, i 6= j, (3.2.15)
S
where Ei is the E-field inside the first waveguide, while Hj is the H-field inside the second
waveguide. In the first waveguide,
∇ × Ei = iωµ · Hi , ∇ × Hi = −iω · Ei , (3.2.16)
while in the second waveguide,
∇ × Ej = iωµt · Hj , ∇ × Hj = −iωt · Ej . (3.2.17)
Ei, Hi
Ej, Hj
µµ(rs),
µ t(rs)
(i) εε(rs) (ii) ε t(rs)
Figure 3.4: Proof of mode orthogonality.
We can show that
∇ · (Ei × Hj − Ej × Hi ) = iω(Hj · µ · Hi − Hi · µt · Hj )
(3.2.18)
+ iω(Ei · t · Ej − Ej · · Ei ).
t
Since A · B · C = C · B · A,
∇ · (Ei × Hj − Ej × Hi ) = 0. (3.2.19)
The above is the generalized Lorentz reciprocity theorem. If Ei , Hi ∼ eikiz z while Ej ,

Hj ∼ eikjz z in their z-dependence, the above becomes
∇s · (Ei × Hj − Ej × Hi ) = −i(kiz + kjz )ẑ · (Eis × Hjs − Ejs × His ) (3.2.20)
Integrating (3.2.20) over the cross-section of the waveguide, we have
I Z
dln̂ · (Ei × Hj − Ej × Hi ) = −i(kiz + kjz ) dS ẑ · (Eis × Hjs − Ejs × His ). (3.2.21)
C S
The line integral vanishes by virtue of the boundary conditions. Hence,

Z
(kiz + kjz ) dS ẑ · (Eis × Hjs − Ejs × His ) = 0. (3.2.22)
If the waveguide has reflection symmetry, a mode propagating in the +z direction is almost
the same as a mode propagating in the −z direction, except for a change of the sign of the
field. If Ejs , Hjs corresponding to a mode with kjz propagating in +z direction, then by
Poynting theorem, Ejs , −Hjs are the transverse fields of a mode with −kjz propagating in
the −z direction. Then, Equation (3.2.22) can be rewritten as
Z
(kiz − kjz ) dS ẑ · (Eis × Hjs + Ejs × His ) = 0. (3.2.23)
If kiz and kjz are both non-zero, then the integral in (3.2.22) must be zero. If kiz 6= kjz , then
the integral in (3.2.23) must be zero. The combination of (3.2.22) and (3.2.23) implies that
Z
dS ẑ · (Eis × Hjs ) = 0, i 6= j. (3.2.24)
S
If the medium is reciprocal, then µ = µt , = t and waveguides (i) and (ii) in Figure 3.1
are the same waveguide, and Eis and Hjs are the fields from the same waveguide.
Furthermore, if µ and are Hermitian corresponding to a lossless medium with reflection
symmetry, similar proof shows that
Z
dS ẑ · (Eis × H∗js ) = 0, i 6= j. (3.2.25)
S
The above is the power orthogonality condition for two different modes in an inhomogeneous,
anisotropic, lossless, waveguide with reflection symmetry. Otherwise, the mode of the original
waveguide is orthogonal to another mode of another waveguide where the medium is filled
will a conjugate transpose medium.
3.2.5 Reflection Symmetry and Conservation of Parity

A commonly accepted law of physics is that the classical laws of physics hold true in the
mirrored world (the reflected world) [8]. For electromagnetics, this means that a right-hand
rule becomes a left-hand rule. This is also known as the conservation of parity, and is found
to be violated in modern physics by some weak interactions.
If a waveguide has reflection symmetry, we say that it appears to be the same waveguide
in the mirrored world as it is in the real world. This is certainly true of all uniform hollow
waveguide. If a mode propagates in the +z direction in the real world, the corresponding
mode propagates in the mirrored +z direction, plus that all the field components are mirrored.
We can rotate the mirrored waveguide by 180o about an axis perpendicular to the z axis, and
we recover the original waveguide. The field of the mirrored mode satisfies the left-hand rule
rather than the right-hand rule. However, we can convert the field of the mirrored mode into
a real-world mode by switching H to −H, and now the field will satisfy the right-hand rule
as before.
When the waveguide is filled with an anisotropic material, the problem is more tricky,
because in the mirrored world, the anisotropic material may not reflect to be the same material
in the real world. However, if the permeability and permittivity tensors are of the form

µs 0 s 0
µ= , = , (3.2.26)
0 µzz 0 zz
then the waveguide has reflection symmetry, i.e., the waveguide in the mirrored world is the
same as the original waveguide. Again, if we switch H to −H, we can obtain a real world
mode.
We can convince ourselves more by looking at Maxwell’s equations. By separating it into
axial and transverse components, Maxwell’s equations for a guided mode for the above tensors
become
∇s × Hs = −iωzz Ez , (3.2.27a)
180°
Mirror H –H Rotation
Real Mode Mirrored
Mode Real Mode Real Mode
Out Out Out In
E E E E
H H H H
(a) (b) (c) (d)
Figure 3.5: Reflection and the conservation of parity of a waveguide.
ikz ẑ × Hs + ∇s × Hz = −iωs · Es , (3.2.27b)
∇s × Es = iωµzz Hz , (3.2.27c)
ikz ẑ × Es + ∇s × Ez = iωµs · Hs . (3.2.27d)

If the direction of propagation changes, which is what a mirror reflection does, kz changes to
−kz . Notice that now if we let
Hs → −Hs , Hz → Hz , Es → Es , Ez → −Ez , (3.2.28)
the above equations remain invariant. Therefore if the waveguide mode changes direction, the
above transformation in the field is needed to obtain another solution to Maxwell’s equations.
To see how this is related to the conservation of parity, we imagine a waveguide in (a)
above with the E and H fields as shown. They could have a ẑ component perpendicular to
the paper which is not shown. We assume that the mode is propagating out of the paper.
In the mirrored world, as shown in (b), we have a mirrored mode and mirrored waveguide.
The electromagnetic field does not satisfy Maxwell’s equations according to the right-hand
rule. To make the mode in (b) into a real mode, we let H → −H as shown in (c). However,
compared to (a), the waveguide is not the original waveguide. If the constitutive parameters
are those in (3.2.26), then (c) is just a 180o rotation of the waveguide in (a) about an axis
perpendicular to the z axis. By rotating the waveguide by 180o , we obtain (d). However, Hs
has changed sign while Hz remains unchanged compared to (a). This is precisely the field
transformation prescribed by (3.2.28). A closer examination also indicates that Es does not
change sign, while Ez changes sign.
3.3 General Anisotropic Waveguide

The vector wave equation governing the field inside an anisotropic waveguide is
∇ × µ−1 · ∇ × E − ω 2 · E = 0. (3.3.1)
In general, the modes inside such a waveguide cannot be decomposed into TE and TM modes.
Hence, the problem is again vector, requiring at least two components of the field. Assuming
the waveguide to have reflection symmetry in the z direction, then

s 0 µs 0
= , µ= , (3.3.2)
0 zz 0 µzz
where s and µs are 2×2 tensors with components in the transverse direction. The transverse
component of (3.3.1) can be extracted to obtain
∇s × µ−1 −1 2 −1 2
zz ∇s × Es − ikz ẑ × µs · ẑ × ∇s Ez − kz ẑ × µs · ẑ × Es − ω s · Es = 0. (3.3.3)
With the use of the divergence condition,
ikz Ez = −−1
zz ∇s · s · Es . (3.3.4)
Hence, the z component of the field in (3.3.3) can be replaced to yield
∇s × µ−1 −1 −1
zz ∇s × Es + ẑ × µs · ẑ × ∇s zz ∇s · s Es
(3.3.5)
− kz2 ẑ × µ−1 2
s · ẑ × Es − ω s · Es = 0.
Consequently, the field in an anisotropic waveguide can be characterized by Es alone. We

can multiply the above by µs · ẑ× to get
µs · ẑ × ∇s × µ−1 −1
zz ∇s × Es − ẑ × ∇s zz ∇s · s · Es
(3.3.6)
− ω 2 µs · ẑ × s · Es + kz2 ẑ × Es = 0.
Note that the above equation is a function of kz2 , implying that if Es with eikz z dependence is
a solution to (3.3.6), an Es with e−ikz z dependence is also a solution. This is a consequence
of reflection symmetry, or the assumptions about and µ in (3.3.2). The corresponding
equation for the transverse magnetic field is obtained by duality, yielding
s · ẑ × ∇s × −1 −1
zz ∇s × Hs − ẑ × ∇s µzz ∇s · µs · Hs
(3.3.7)
− ω 2 s · ẑ × µs · Hs + kz2 ẑ × Hs = 0.
Since the solution to (3.3.6) is orthogonal to the solution to (3.3.7), for two different modes,
Equation (3.3.7) is also the transpose equation of (3.3.6) [9, 15].
3.4 Proof of Transpose of Operators

For an anisotropic waveguide with reflection symmetry,
−1
µs · ẑ × ∇s × µzz ∇s × Es − ẑ × ∇s −1
zz ∇s · s · Es
−ω 2 µs · ẑ × s · Es + kz2 ẑ × Es = 0 (3.4.1)
Way to discretize these equations has been

done by K. Radhakrishnan using lattice
electromagnetics theory in his PhD thesis at
UIUC. Recently, Q. Dai has an FEM method
to solve them correctly in his PhD thesis at
HKU.
We shall show that the above equation is transpose to the equation
ts · ẑ × ∇s × −1 −1 t
zz ∇s × Hs − ẑ × ∇s µzz ∇s · µs · Hs
−ω 2 ts · ẑ × µts · Hs + kz2 ẑ × Hs = 0 (3.4.2)
The definition of the transpose operator is2
hu, Lvi = hv, Lt ui (3.4.3)
where inner product betweenR two vector fields in the infinite dimensional space also called the
Hilbert space) is hf , gi = dSf ·g. The integration in this case is taken over the cross-sectional
area of the waveguide.
To prove this, we start with the expression
I = hHs , µs · ẑ × ∇s × µ−1
zz ∇s × Es i (3.4.4)
It can be easily shown that
I = hẑ × µts · Hs , ∇s × µ−1

zz ∇s × Es i (3.4.5)
Using Gauss theorem, or integration by parts, and that ∇s ·(A×B) = B·∇s ×A−A·∇s ×B
we have
Z Z
I = dS∇s · [(ẑ × µts · Hs ) × (µ̄−1
zz ∇ s × Es )] − dS(∇s × ẑ × µts · Hs ) · µ−1
zz ∇s × Es ) (3.4.6)
It can be shown that ∇s ×(ẑ ×µts ·H1 ) = ẑ∇s ·(µts ·Hs ), and that ẑ ·∇s ×Es = −∇s ·(ẑ ×Es ).
The first term above can be converted to a contour integral over the waveguide wall using
Gauss’ divergence theorem. It vanishes by virtue of the boundary condition on the waveguide
wall. Hence, only the second term remains. Using integration by parts, and then a vector
identity,
Z Z
t
−1
I = dS∇s · µ̄s · Hs µzz ∇s · (ẑ × Es ) = dS ẑ × ∇s µ−1 t
zz ∇s · (µs · Hs ) · Es (3.4.7)
It can be shown that (ẑ×)t = −ẑ×, ( · ẑ × µs )t = −µts · ẑ × ts . Hence if,
L = µs · ẑ × ∇s × µ−1 −1 2 2
zz ∇s × −ẑ × ∇s zz ∇s · s − ω µs · ẑ × s + kz ẑ× (3.4.8)
Then
Lt = ts · ẑ × ∇s × −1 −1 t 2 t t 2
zz ∇s × −ẑ × ∇s µzz ∇s · µs − ω s · ẑ × µs + kz ẑ× (3.4.9)
In general, (3.4) is the equation for a waveguide filled with transpose medium compared to the
original waveguide equation (3.4). However, if the waveguide is filled with reciprocal medium,
then (3.4) and (3.4) are equations in the same waveguide.
(a) (b)
0 x1 x2 a
Figure 3.6: Dielectric-slab-loaded waveguide.
3.5 Dielectric-Slab-Loaded Rectangular Waveguides

For the analysis of a general inhomogeneously filled waveguide, a numerical method has to
be sought [9–11]. However, when the waveguide has certain symmetry such as a slab loaded
rectangular waveguide, analytic method for its analysis is possible.
Dielectric-slab-loaded waveguides find applications in a number of microwave components,
because the phase velocity of a mode can be altered with dielectric loading [3,12]. A waveguide
can also be loaded with a ferrite slab [13, 14]. When a ferrite slab is biased with a magnetic
field, it becomes an anisotropic, gyrotropic medium. Such a medium is non-reciprocal. Hence,
ferrite slabs can be used to design non-reciprocal devices such as isolators.
The analysis of a ferrite-slab-loaded waveguide is complicated. We will focuss our analysis
on a dielectric-slab-loaded waveguide. In such a waveguide, with the exception of special cases,
the modes cannot be decomposed into TE and TM modes with respect to the z direction.
Hence, a general mode is hybrid. However, for the slab-loaded waveguide as shown in Figure
3.6(a), the modes in the waveguide can be decomposed into modes with E field transverse to
x or modes with H field transverse to x. Similar decomposition exists for Figure 3.6(b) since
it is just a 90o rotation of Figure 3.6(a). The modes with E field transverse to x are known
as the LSE (Longitudinal Section Electric) modes, while the modes with H field transverse
to x are known as the LSM (Longitudinal Section Magnetic) modes. There exist closed form
expressions for the guidance condition of these modes, but the exact wave number kz has to
be found numerically.
Since the LSE mode is transverse to x, it can be characterized with Hx . Similarly, the LSM
mode can be characterized with Ex . The field has to have eikz z dependence everywhere inside
the waveguide due to the phase matching condition. Consequently, the equations satisfied by
Hx and Ex are
(∇2s + ki2 − kz2 )Hix = 0, LSE modes, (3.5.1)
(∇2s + ki2 − kz2 )Eix = 0, LSM modes. (3.5.2)
where subscript i denotes the region i and subscript s denotes transverse to x. The solutions
to (3.5.1) and (3.5.2) are of the form
Hix = H0 {e±ikix x }{e±iky y }eikz z , (3.5.3)

2 When applied to a matrix operator, this becomes ut · L · v = vt · Lt · u . Similar formula can be derived
for defining conjugate transpose or adjoint of an operator [15].

t3 t2
b
k3 k2 k1
3 2 1
µ3, ε3 µ2, ε2 µ1, ε1

x
0 x2 x1 a
n2, e2
Figure 3.7: A special case of a slab loaded waveguide.
Eix = E0 {e±ikix x }{e±iky y }eikz z , (3.5.4)

where the braces imply linear superpositions. Here, ky in each region must be the same due
to the phase matching condition. The fields transverse to x can be found from Hix and Eix
in each region, i.e.,

1 ∂
Eis = 2 ∇ E
s ix − iωµ i x̂ × ∇ H
s ix , (3.5.5a)
ky + kz2 ∂x

1 ∂
His = ∇ H
s ix + iωi x̂ × ∇ E
s ix , (3.5.5b)
ky2 + kz2 ∂x
where the subscript s implies transverse to x. It can be seen from the above that the boundary
conditions for Hix and Eix are that
∂Hix
= 0, at y = 0 and y = b, (3.5.6a)
∂y
Eix = 0, at y = 0 and y = b. (3.5.6b)

Therefore, forms for Hix and Eix that satisfy the above boundary conditions are
nπy
Hix = H0 {e±ikix x } cos eikz z , (3.5.7a)
b
nπy
Eix = E0 {e±ikix x } sin eikz z , (3.5.7b)
b
q
where we have let ky = nπb , and n is an integer. In the above, kix = ki2 − kz2 − ( nπ 2
b ) . It is
now clear that Hix represents bouncing waves that are TE to x while Eix represents bouncing
waves that are TM to x. Hence, in region 1, we can write the solution as
nπy
H1x = H0 [e−ik1x (x−x1 ) + R̃12
T E ik1x (x−x1 )
e ] cos eikz z , (3.5.8a)
b
nπy
E1x = E0 [e−ik1x (x−x1 ) + R̃12
T M ik1x (x−x1 )
e ] sin eikz z , (3.5.8b)
b
T E,T M
where R̃12 is a generalized Fresnel reflection coefficient for a TE or a TM wave incident
from the left at x = x1 . It includes subsurface reflections. However, we need to impose the
boundary conditions that
∂Eix
Hix (x = a) = 0, (x = a) = 0. (3.5.9)
∂x
The above boundary conditions can be derived from (3.5.5a) and (3.5.5b) as the most general
case. They can also be derived from n̂ · H = 0 and ∇ · E = 0, respectively. They could only
be satisfied if
T E 2ik1x (a−x1 )
1 + R̃12 e = 0, LSE modes, (3.5.10a)
T M 2ik1x (a−x1 )
1 − R̃12 e = 0, LSM modes. (3.5.10b)
The above are the guidance conditions for the LSE modes and the LSM modes in a slab
loaded rectangular waveguide.
To find R̃12 , we note that a wave impinging on a slab with one subsurface interface will
consist of a surface reflection from the top interface plus a sequence of subsurface reflections.
The single interface reflections are governed by the Fresnel reflection coefficients. Hence [15]
R̃12 = R12 + T12 R23 T21 e2ik2x t2 + T12 R23

2
R21 T21 e4ik2x t2
3 2
(3.5.11)
+ T12 R23 R21 T21 e6ik2x t2 + . . . ,
where Tij = 1 + Rij is the Fresnel transmission coefficient at the ij interface. The above
could be summed to yield
T12 R23 T21 e2ik2x t2 R12 + R23 e2ik2x t2

R̃12 = R12 + = . (3.5.12)
1 − R23 R21 e2ik2x t2 1 − R23 R21 e2ik2x t2
In the above,
TE µj kix − µi kjx TM j kix − i kjx
Rij = , Rij = , (3.5.13)
µj kix + µi kjx j kix + i kjx
depending on whether we are calculating R̃12 for a TE wave or a TM wave.
If there are subsurface layers below region 3, R23 in (3.5.12) can be replaced with R̃23 , or
R12 + R̃23 e2ik2x t2

R̃12 = . (3.5.14)
1 − R21 R̃23 e2ik2x t2
t2
3 2 1
T12
T12eik2xt
R12
T12R23eik2xt
T12R23T21e2ik2xt
T12R23e2ik2xt
2
T12R23R21T21e2ik2xt
3 2
T12R23R21T21e4ik2xt
Figure 3.8: Multiple reflections and transmissions through a dielectric slab.
The above is a recursive relation from which one can calculate R̃ij for any number of layers.
TE TM
For example, if there is a metallic wall at x = 0, as in Figure 3.7, R34 = −1, R34 = 1, then
TE
TE R23 − e2ik3x t3 TM
TM
R23 + e2ik3x t3
R̃23 = T E e2ik3x t3
, R̃23 = T M e2ik3x t3
. (3.5.15)
1 + R32 1 − R32
With R̃12 defined by (3.5.14) and (3.5.15), Equations (3.5.10a) and (3.5.10b), in general,
have to be solved numerically with a root solver like the Newton-Raphson method, or the
Muller’s method. All the Rij ’s are defined in terms of kix in Equation (3.5.13), where
r nπ 2
kix = ki2 − kz2 − . (3.5.16)
b
Hence, Equations (3.5.10a) and (3.5.10b) can be solved for values of kz . For simple geometry,
graphical solutions to Equations (3.5.10a) and (3.5.10b) may be found.
The guidance condition given by Equations (3.5.10a) and (3.5.10b) are also known as the
transverse resonance condition which will be discussed in greater detail in the next section.
The guidance conditions in (3.5.10a) and (3.5.10b) are obtained by considering waves bouncing
in the air region. If a guided mode is trapped in the dielectric slab region, the wave becomes
evanescent in the air region and Equations (3.5.10a) and (3.5.10b) can become ill-posed. To
remedy this, it is better to write down the guidance condition in the slab region using the
transverse resonance condition described in the next section.
Figure 3.9 shows the dispersion curves of the TE10 and TE01 (with respect to z) modes
of a slab loaded dielectric waveguide. Case 1 is when the whole waveguide is filled with
TE10 Dispersion Curve
140
εr = 2.0
a = 0.12 m.
120 a/4 a/2
1
2
100 εo εr εo 3
Wave Number kz
a/2
80
a
60
40
1: Pure Dielectric Mode

20 2: LSE Mode
3: Free Space Mode
0
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Frequency in GHz
(a)
TE 01 Dispersion Curve
160 εr = 2.0 1
a = 0.12 m. 2
140 a/4 a/2
3
120 εo εr εo
a/2
Wave Number kz
100
a
80
60
40
1: Pure Dielectric Mode
20 2: LSM Mode
3: Free Space Mode
0
1 2 3 4 5 6 7
Frequency in GHz
(b)
Figure 3.9: Dispersion curves for various dielectric loading for (a) TE10 mode, and (b) TE01
mode.
Square Waveguide Modified Dispersion Curves

1.6
...............................................................
...............................................
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
...............................................................
...............................................................
1.4 ...............................................................................................................
Relative Wave number (kz / k o) 2

1.2 ...............................................................................................................
4
1 ...............................................................................................................
0.8 ...............................................................................................................
ε r = 2.0 a/4
0.6 ........................ .....................................................
ε0 εr ε0
0.4 ........................
a ....
1: Purely Dielectric Modes

..
2: LSE Mode
0.2 ........................ ....
3: LSM Mode ..
a 4: Free Space Modes

. .
..
. .
..
..
0
..
0 0.5 1 1.5 2 2.5 3 3.5 4
Normalized Frequency (a / λ)
Figure 3.10: The dispersion curves for a slab-loaded square waveguide showing anisotropy.
dielectric material, and case 3 is when the waveguide is empty. Notice that for the perturbed
modes (case 2), the dispersion curve is parallel to case 1, The reason is that the mode is
entrapped in the dielectric slab for high frequencies, and the group velocity of the mode
(dω/dkz ) approaches that of the dielectric slab.
Also, notice that at lower frequencies, the TE10 mode is perturbed more by the slab than
the TE01 mode, because for the TE01 mode, the electric field is normal to the slab while
for the TE10 mode, the electric field is parallel to the slab. This fact can be used to create
anisotropy in a symmetric waveguide like a circular waveguide or a square waveguide. Figure
3.10 shows the dispersion curves for the LSE mode, which is the perturbed TE10 mode, and
the LSM mode, which is the perturbed TE01 mode. Notice that the LSE mode is affected
more by the dielectric slab than the LSM mode. This anisotropic effect can be used to make a
quarter-wave plate out of a dielectric slab loaded waveguide. This will be discussed in greater
detail later.
3.6 Transverse Resonance Condition

The guidance conditions given by (3.5.10a) and (3.5.10b) can also be derived by the transverse
resonance condition. The transverse resonance condition is a powerful condition that can be
used to derive the guidance condition of a mode in a layered medium.
To derive this condition, we first have to realize that a guided mode in a waveguide is due
to the coherent or constructive interference of the waves. This implies that if a plane wave
starts at position 1 and is multiply reflected as shown, it will regain its original phase in the
x = x1 x = x2
Region 0
y
5 x
~
R– 3
~
4 R+
1 2
t
Figure 3.11: The transverse resonance condition for a layered medium. The phase of the wave
at position 5 should be equal to the phase at position 1.
x direction at position 5. Since this mode progresses in the y direction. it will gain a phase
in the y direction. But, for it to coherently interfere in the x direction, the transverse phase
at 5 must be the same as 1.
Assuming that the wave starts with amplitude 1 at position 1, it will gain a transverse
phase of eik0x t when it reaches position 2. Upon reflection at x = x2 , at position 3, the
wave becomes R̃+ eik0x t . Finally, at position 5, it becomes R̃− R̃+ e2ik0x t . For constructive
interference to occur or for the mode to exist, we require that
R̃− R̃+ e2ik0x t = 1. (3.6.1)
The above is the transverse resonance condition. It is also the guidance condition for a mode
travelling in a layered medium.
In (3.5.10a), a metallic wall has a reflection coefficient of 1 for a TM wave, hence if R̃+ is
1, Equation (3.6.1) becomes
1 − R̃− e2ik0x t = 0. (3.6.2)
In (3.5.10b), a metallic wall has a reflection coefficient of −1, and Equation (3.6.1) becomes
1 + R̃− e3ik0x t = 0. (3.6.3)
3.7 Fabry-Perot Etalon

Since we have the machinery in place, it is convenient to study the Fabry-Perot etalon which
is often used as an optical filter [5]. By tracing the plane wave or ray through the slabs, we
can show that the generalized transmission coefficient from region 1 to region 3 is
T12 T23 eik2x d

T̃13 = (3.7.1)
1 − R23 R21 e2ik2x d
where d is the thickness of the slab. In etalon application, regions 1 and 3 are free space while
region 2 is dielectric. Specializing to the case when regions 1 and 3 have the same parameters,
we have
T12 T21 eik2x d
T̃ = 2 e2ik2x d (3.7.2)
1 − R21
Since T12 = 1 + R12 , T21 = 1 + R21 = 1 − R12 , we have

2
(1 − R21 )eik2x d
T̃ = 2 e2ik2x d (3.7.3)
1 − R21
For normally incident wave, k2x = k2 , and the above becomes

2
(1 − R21 )eik2 d
T̃ = 2 (3.7.4)
1 − R21 e2ik2 d
where √ √
k2 − k1 2 − 1
R21 = =√ √ (3.7.5)
k2 + k1 2 + 1
assuming that µ1 = µ2 = µ . The transmissivity is
2 2
|1 − R21 |
t = |T̃ |2 = 2 (3.7.6)
|1 − R21 e 2 t |2
2ik
The above is maximum with t = 1 when e2ik2 d = 1 or 2k2 d = 2mπ with integer m. In other
words, at such a frequency, the etalon is transparent with no reflected wave. Specializing
(3.5.14) to this case, and using the fact that R12 = −R21 ,

R12 1 − e2ik2 d
R̃ = 2 e2ik2 d . (3.7.7)
1 − R21
It is seen that the above can be zero if 2k2 t = 2mπ. The zero comes about because of the
destructive interference of the reflected waves. The transmissivity of the etalon as a function
of frequency is shown in Figure 3.12.
Next, we can perform the pole analysis of the etalon. In the vicinity of the maxima,
2id 2id
e2ik2 d = 1 + (ω − ωm ) = 1 + ∆ω (3.7.8)
c2 c2
where ∆ω = ω − ωm ,
mπc2
ωm = (3.7.9)
d
and
1
c2 = √ (3.7.10)
µ2
Figure 3.12: The transmissivity of the etalon as a function of frequency.
Hence,
2
1 − R21
T̃ ≈ 2 2 2id ∆ω
(3.7.11)
1− R21 − R21 c2
near the maxima. The pole locations can be found to be when
2

∼ 1 − R21 −ic2
∆ω = 2 (3.7.12)
R21 2d
Hence, the pole location for the m-th resonance mode in the complex plane is
2 2
ic2 1 − R21 mπc2 ic2 1 − R21
ωpm ' ωm − 2 = − 2 (3.7.13)
2d R21 d 2d R21
2
The above is accurate when 1 − R21 is small. It is slightly below the real ω axis which is in
agreement with a lossy resonant mode with e−iωt time convention. Notice that the imaginary
part of the pole location is independent of frequency (See Figure 3.13).
The mechanism for transmission here is resonance tunneling. The Fabry-Perot etalon is
used as an optical filter. To obtain a narrow band filter, the Q’s of these modes have to be
high. The Q of a resonator is
ωr W T
Q= (3.7.14)
Pd
where
WT
=T
Pd
is the time constant for which the energy store WT will diminish to e−1 of its original value.
So Q = ωr T is also the number of cycles in radian for this decay to happen. Q is an
Im(ω )
m=0 m=1 m=2 m=3 m=4

Re(ω )
x x x x x
Figure 3.13: The locations of the poles in the complex ω plane for the Fabry-Perot modes.
asymptotic concept that has meaning only when Q is large. When the pole location is known,
0 00
ωp = ωp + iωp , then
0
ωp
Q = − 00 (3.7.15)
2ωp
Applying the above to the m-th mode in (11), we have
2
mπR21
Qm = 2 ) (3.7.16)
(1 − R21
3.8 Rod-Loaded Circular Waveguide

When a circular waveguide is loaded with a concentric circular dielectric rod, the equation
for the guidance condition can be found in closed form. Ferrite rods are usually used to load
a circular waveguide to engender Faraday rotation. Since ferrites are anisotropic, the analysis
of a ferrite loaded circular waveguide is in general very complex. Hence, we will analyze
the case of a circular waveguide loaded with a circular dielectric rod. The general case of a
waveguide periodically loaded with dielectric rods can be used to make microwave filters [16].
3.8.1 Reflection off a Dielectric Rod

Since Ez and Hz waves are in general coupled in uniform dielectric rod, we have to consider
both polarizations together in this reflection problem. We will follow an analysis presented
in [15]. The z-components of the fields satisfy

2 2
Ez
∇ +k = 0. (3.8.1)
Hz
Assuming eikz z+inφ dependence in the wave, the equation becomes

1 ∂ ∂ n2 2 2 Ez
ρ + 2 + k − kz = 0. (3.8.2)
ρ ∂ρ ∂ρ ρ H z
Region 0 Region 1 µ1, ε1 µ0, ε0
ρ=a
ρ
O
Figure 3.14: Geometry for defining the reflection of a wave off a circular dielectric rod.
The general solution to the above is of the form

Ez
= [an Jn (kρ ρ) + bn Hn(1) (kρ ρ)]eikz z+inφ , (3.8.3)
Hz
p (1)
where kρ = k 2 − kz2 , and Hn (x) is the Hankel function of the first kind.
We can assume an incident field on a dielectric rod in region 0 as
i
i E0z
f0z = i = a0n Jn (k0ρ ρ) (3.8.4)
H0z
p
where k0ρ = k02 − kz2 , and the eikz z+inφ dependence is implied. The above is the incident
wave in the absence of the dielectric rod. Hence, it cannot have a Hankel wave for it would
be singular at the origin. When a circular dielectric rod is put at the origin, it will reflect
the incident wave generating an outgoing wave satisfying the radiation condition at infinity.
Therefore, The total solution in region 0 must be of the form

Eoz
f0z = = a0n Jn (k0ρ ρ) + b0n Hn(1) (k0ρ ρ). (3.8.5)
H0z
Since bn must be linearly dependent on an , we express
b0n = R01 · a0n (3.8.6)
In region 1, the solution can only admit Bessel waves since a Hankel wave is singular at the
origin. Therefore, the general solution is of the form

E1z
f1z = = a1n Jn (k1ρ ρ). (3.8.7)
H1z
Since an is linearly related to a0n , we let
a1n = T01 · a0n . (3.8.8)
The transverse to z components of the field can be obtained from the following equations in
the j-th region
1
Ejs = 2 [ikz ∇s Ejz − iωµẑ × ∇s Hjz ], (3.8.9a)
kjρ
1
Hjs = 2 [ikz ∇s Hjz + iωẑ × ∇s Ejz ], (3.8.9b)
kjρ
∂
where ∇s = ρ̂ ∂ρ + φ̂ ρ1 ∂φ
∂ ∂
= ρ̂ ∂ρ + φ̂ in 2 2 2
ρ , and kjρ = kj −kz . From the phase-matching condition,
kz is the same in all regions.
In order to find R01 and T01 in (3.8.6) and (3.8.8), we need to match the continuity of
the tangential components of E and H across the interface. These tangential components are
the z and the φ components. The z components are already given in (3.8.5) and (3.8.7). The
corresponding φ components can be derived using (3.8.9a) and (3.8.9b). From (3.8.5) and
(3.8.9a) and (3.8.9b), we deduce that

H0φ (1)
f0φ = = Jn (k0ρ ρ) · a0n + Hn (k0ρ ρ) · R01 · a0n (3.8.10)
E0φ

H1φ
f1φ = = Jn (k0ρ ρ) · T01 · a0n , (3.8.11)
E1φ
where
1 iωj kjρ ρBn0 (kjρ ρ) −nkz Bn (kjρ ρ)
Bn (kjρ ρ) = . (3.8.12)
2 ρ
kjρ −nkz Bn (kjρ ρ) −iωµj kjρ ρBn0 (kjρ ρ)
(1) (1)
In the above, Bn is either Hn or Jn depending on if we are defining Hn or Jn matrix. Note
that Bn is diagonal when n = 0. This also implies the decoupling of Ez and Hz waves when
n=0.
Matching the boundary condition at the boundary where ρ = a, we have
[Jn (k0ρ a) + Hn(1) (k0ρ a)R01 ] · a0n = Jn (k1ρ a)T01 · a0n , (3.8.13a)
(1)
[Jn (k0ρ a) + Hn (k0ρ a) · R01 ] · a0n = Jn (k1ρ a) · T01 · a0n , (3.8.13b)
The above can be solved to yield
−1
R01 = D · [Jn (k1ρ a)Jn (k0ρ a) − Jn (k0ρ a)Jn (k1ρ a)], (3.8.14a)

2ω −1 0 0
T01 = 2 a D · , (3.8.14b)
πk0ρ 0 −µ0
where h i
(1)
D = Jn (k1ρ a)Hn(1) (k0ρ a) − Hn (k0ρ a)Jn (k1ρ a) . (3.8.14c)
The Wronskian for Hankel function, which is
0 2i
Hn(1) (x)Jn0 (x) − Jn (x)Hn(1) (x) = − (3.8.15)
πx
has been used to simplify the above.
In general R01 and T01 are non-diagonal implying the coupling of the Ez and Hz waves
by the dielectric rod.
Region 0
Region 2 Region
σ 1
µ1, ε1 µ0, ε0
Figure 3.15: Geometry of a dielectric-rod-loaded circular waveguide.
3.8.2 Reflection off a PEC Waveguide Wall

We have asserted that a PEC or a PMC cylindrical surface does not depolarize an Ez or Hz
wave, We can further confirm this assertion by looking at Equation (3.8.9a). For a PEC, we
require that Ez = 0 and that n̂ × Es = 0 on the cylindrical surface. It is seen that if we
set Ez = 0 on a surface, then n̂ × ∇Ez = 0 or the first term in (3.8.9a) is zero without any
help from the second term in (3.8.9a). Therefore, the Ez wave alone can satisfy the boundary
condition on a PEC independently of the Hz wave. In order for n̂ × Es = 0 for the Hz wave,
we require that
n̂ × ẑ × ∇s Hz = 0, (3.8.16a)
or that
n̂ · ∇s Hz = 0. (3.8.16b)
In other words, if we impose the homogeneous Neumann boundary condition (3.8.16b) on a
PEC cylindrical surface, the tangential electric field that arises from the Hz wave will satisfy
the requisite boundary condition independently of the Ez wave.
As a result of the above discussion, when an outgoing Hankel Wave impinges on a PEC
waveguide wall which is circular, it reflects back into a Bessel wave. Hence, the wave in region
0 can be written as

E0z
f0z = = Hn(1) (k0ρ ρ)b0n + Jn (k0ρ ρ)R02 · b0n (3.8.17)
H0z
where R02 is a diagonal 2×2 matrix due to the decoupling of the Ez and Hz waves. Matching
the requisite boundary condition on the waveguide wall at ρ = b, we obtain that
" #
(1)
−Hn (k0ρ b)/Jn (k0ρ b) 0
R02 = (1)0 . (3.8.18)
0 −Hn (k0ρ b)/Jn0 (k0ρ b)
Note that if region 2 is a dielectric region, the above will be a non-diagonal matrix as shown
in the next subsection.
3.8.3 Reflection off an Outer Dielectric Wall

If region 2 is a dielectric region rather than a perfectly conducting region, a reflection matrix
can be similarly derived as in Subsection 3.5.1. In this case, we can show that
−1 (1) (1)
R02 = D · [Hn(1) (k0ρ a)Hn (k2ρ a) − Hn(1) (k2ρ a)Hn (k0ρ a)] (3.8.19a)

2ω −1 0 0
T02 = 2 a D · . (3.8.19b)
πk0ρ 0 −µ0
where
(1)
D = [Jn (k0ρ a)Hn(1) (k2ρ a) − Hn (k2ρ a)Jn (k0ρ a)]. (3.8.19c)
3.8.4 The Guidance Condition

The guidance condition in a dielectric-rod-loaded circular waveguide can be obtained by
considering the solution in region 0 and matching boundary condition on the waveguide wall
at ρ = b. The solution in region 0 is given by (3.8.5) is rewritten here as
f0z = [Jn (k0ρ ρ)I + Hn(1) (k0p ρ)R01 ] · a0n . (3.8.20)
However, according to (3.8.17), the field in region 0 can also be written as
f0z = [Jn (k0ρ ρ)R02 + Hn(1) (k0ρ ρ)I] · b0n . (3.8.21)
Hence, we conclude that

R02 · b0n = a0n , (3.8.22a)
R01 · a0n = b0n , (3.8.22b)

or that
(R02 · R01 − I) · a0n = 0. (3.8.23)
In order for a0n 6= 0, we require that
det(R02 · R01 − I) = 0. (3.8.24)
The above is the guidance condition or the transverse resonance condition for a cylindrically
layered circular waveguide.
3.9 Applications of Inhomogeneously Filled Waveguides

Inhomogeneous filled waveguides can be used to make variable phase shifters and attenuators
[3, 4]. When the inhomogeneity filling the waveguide is nonreciprocal like ferrite, isolators,
gyrators, and attenuators can be made.
(a)
εr (b)
εr (c)
Figure 3.16: Different dielectric-slab loading of a parallel plate waveguide.

3.9.1 The Effect of Inhomogeneous Fillings on the Phase Velocity

An intuitive understanding of how inhomogeneous fillings affect the phase velocity of a guided
mode can be acquired by studying a parallel plate waveguide. Considering a square region
of a parallel plate waveguide away from the edges so that fringing field effect can be ignored.
Hence, we assume that the field lines for the TEM mode is perfectly vertical.
The capacitance of (a) can be thought of as three capacitances in parallel or as three
capacitances in series. When the waveguide is filled as in (b), it affects one of the three
capacitances in parallel, yielding a resultant capacitance given by
2 r 2 + r
Cb = Ca + Ca, = Ca (3.9.1)
3 3 3
where Ca is the capacitance of (a). When the waveguide is filled as in (c), it affects one of
the three capacitances in series, yielding a resultant capacitance given by
Cc−1 = 2(3Ca )−1 + (3r Ca )−1 (3.9.2)
or
3r Ca
Cc = . (3.9.3)
2r + 1
It is seen that
3r 2 + r
< , r > 1. (3.9.4)
2r + 1 3
Therefore Cb > Cc always for r > 1. The reason is that enhancing a capacitor in parallel
has more effect on the total capacitance than enhancing a capacitor in parallel. When r →
∞, Cc → 23 Ca or saturates while Cb → ∞. Since the phase velocity of a TEM mode in a
parallel plate waveguide is given by
1
υ=√ (3.9.5)
LC
where L and C are line inductance and line capacitance respectively, a dielectric loading in
case (b) slows down the wave more than case (c).
When the waveguide is a square waveguide, the TE10 mode is affected even more by a
symmetrically located dielectric slab because the electric field has a maximum at the center
of the waveguide. Hence, by dielectric-slab loading as in case (b), the TE10 mode propagates
with a slower phase velocity than the TE01 mode. This gives rise to anisotropy in a waveguide.
3.9.2 Quarter-Wave Plate

By slab loading a circular waveguide, one can make a quarter wave plate. If a dielectric slab
is oriented at 45o with respect to the TE11 mode of a circular waveguide, the mode can be
decomposed into two orthogonal modes, one with E field perpendicular to the dielectric slab,
and another parallel to the slab. The one with electric field parallel to the slab is going to
be slowed down more than the one with electric field perpendicular to the slab. Hence after
a certain distance, the phases of these two modes are going to be out of phase. If the length
of the dielectric slab is chosen judiciously such that these two modes are 90o out of phase,
TE10
E Field εr εr
(a) (b) (c)
Figure 3.17: Dielectric-slab loading of a square waveguide. The TE10 mode is affected more
AAAA AAAAA AAAA

by case (b) than case (c).
AAAA AAAAA AAAA

AAAA
AAAA
= AAAAA
AAAAA
+ AAAA
AAAA
Figure 3.18: Decomposition of a TE11 mode of a circular waveguide into two orthogonal
modes. One perpendicular to the slab and another parallel to the slab.
then one obtains a circularly polarized mode at the other end of the waveguide. Such an
application of a dielectric slab loading is similar to a quarter-wave plate in optics and hence
its name. A half-wave plate will shift one component of the wave by 180o compared to the
other orthogonal component.
3.9.3 Variable Phase Shifter

Dielectric loading can be used to make variable phase shifters. One way of achieving this is
to vary the position of the slab position in a rectangular waveguide. Since the TE10 mode of
a rectangular waveguide has a weaker field near the side walls of the waveguide, the phase
velocity will be slowed down less when the slab is near to the waveguide wall compared to
near the center of the waveguide (See Figure 3.20).
A linear phase changer can also be constructed by dielectric slab loading as shown below.
When the center slab is moved by a distance ∆z, line 1 and line 3 are increased by a length
∆z while line 2 and line 4 are decreased by a length ∆z. Therefore, the total phase shift is
∆φ = (kz1 + kz3 − kz2 − kz4 )∆z. (3.9.6)
Because of the sinusoidal distribution of the electric field of a TE10 mode of a rectangular
waveguide, it is clear that

kz4 < kz1 < kz3 < kz2 . (3.9.7)
However, if the center section is chosen to be about 0.3a, we can have
kz1 + kz3 > kz2 + kz4 , (3.9.8)
and a net positive phase shift linearly proportional to ∆z becomes possible.

A rotary phase shifter can be made by sandwiching a section of half-wave plate between
two sections of quarter-wave plates as shown in Figure 3.21. The quarter-wave plate and half-
wave plate sections are made by dielectric slab loading as described previously. By rotating
the middle section, one can vary the phase of the wave going from port A to port B.
To explain the operating principle of this device, we will use linearly polarized waves in
free space. The first quarter-wave plate converts a linearly polarized wave into a circularly
polarized wave where the x̂ and ŷ components are 90o out of phase, viz.,
E = (ix̂ + ŷ)E0 , (3.9.9)
where x̂ direction is parallel to the dielectric slab in the quarter-wave plate section, and E0
is a complex number. The above represents a left-hand circularly polarized wave. When
this wave impinges on the half-wave plate section, the dielectric slab can be oriented at any
angles. In Figure 3.22, we assume it to be at an angle θ with respect to the coordinates of
the dielectric slab in the first section. In this case, we have
x̂ = x̂0 cos θ − ŷ 0 sin θ, ŷ = x̂0 sin θ + ŷ 0 cos θ. (3.9.10)

Using (3.9.10) in (3.9.9), we have
E = (ix̂0 + ŷ 0 )e−iθ E0 (3.9.11)
The above is still a left-hand circularly polarized wave, and e−iθ comes about because we are
rotating the coordinates counterclockwise while the polarization is rotating clockwise. After
this wave has gone through the half-wave section, the ŷ 0 component will be 180o out of phase
with respect to the x̂0 component and we have
E = (−ix̂0 + ŷ 0 )e−iθ E00 (3.9.12)
which is a right-hand circularly polarized wave, and E00 is a new complex number accounting
for the additional phase delay the wave has acquired in propagating through the middle
section. Projecting this back to the original coordinates by using
x̂0 = x̂ cos θ + ŷ sin θ, ŷ 0 = −x̂ sin θ + ŷ cos θ, (3.9.13)
we have
E = (−ix̂ + ŷ)e−2iθ E00 . (3.9.14)
If the wave had remained left-hand circularly polarized, the projection back would have an-
nulled the phase e−iθ . But because it becomes a right-hand circularly polarized, it introduces
an additive phase instead.
120
Relative Wave Number (kz / k o)
0.8
0.85
0.9
0.95
1
1.05
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
0
............
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
0.1
εr
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
0.2
a / λ = 0.75
Perturbed TE 10
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
(a)
(b)
0.3
..
............................................
a/2
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ε r = 2.0
0.4
x
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ε0
0.5
a
εr
Relative Position of Slab (x/a)

a/4
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
ε0
0.6
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
0.7
Wave Number as a Function of Dielectric Slab Position
...............
....................................................................................................
...............
....................................................................................................
Variable dielectric slab locations affects the phase velocity of the TE10 mode.
Figure 3.19: (a) Variable dielectric slab locations can be used as a variable phase shifter. (b)
Line 1 Line 2 Line 3 Line 4

kz1 kz2 kz3 kz4
0.30a a
Top View End View

Dielectric
Slabs
Figure 3.20: A linear phase shifter using four different sections of dielectric loading.
Rotating Section
A B
λ/4 Plate λ/2 Plate λ/4 Plate
Figure 3.21: A rotary phase shifter.
y
y′
x′
θ
θ x
Dielectric Slab
Figure 3.22: Orientation of the dielectric slab in the half-wave plate section compared to the
original coordinates.
AAAA
AAAA
AAAAAAAAA
AAAAA
Rotatable Resistive Sheet
AAAA
AAAAAAAAAAAAA
AAAAA
A B
Rotating Section
Horizontal Resistive Sheets
Figure 3.23: A rotary variable attenuator.
When this wave in (3.9.14) passes through the last section of the quarter-wave plate, the
x̂ component gains another 90o of phase with respect to the ŷ component, and it becomes
E = (x̂ + ŷ)e−2iθ E000 (3.9.15)
where E000 is a new complex number. This wave is again linearly polarized with respect to the
rectangular waveguide in the output port. Note that there is a phase shift of −2θ which is
dependent on the orientation of the center rotating section .
3.9.4 Variable Attenuator

For the same reason that a dielectric slab parallel to the electric field affect a mode more
than a slab perpendicular to the electric field, a resistive (lossy) dielectric sheet parallel to
the electric field will incur more loss on a mode than a resistive sheet perpendicular to the
electric field. Hence, a variable attenuator can be made similar to a variable phase shifter,
except that in the first and the last section, the resistive sheets are loaded horizontally, and
the middle section the resistive sheet can be rotated.
The horizontal resistive sheet in the first section ensures that the mode is predominantly
vertically polarization. The resistive sheet in the middle section will attenuate the mode
proportional to the component of the electric field parallel to the resistive sheet. Hence, the
attenuation of the mode by the middle section can be varied by rotating it. The last section
ensures that any horizontally polarized modes due to mode conversion be removed, and only
the vertically polarized component will exit from port B.
3.10 Spin Dynamics and Ferrite Materials

The understanding of the interaction of particle spins with electromagnetic field is commonly
encountered in the study of nuclear magnetic resonance, and ferrite materials. The first
is extremely useful in magnetic resonance imaging (MRI) which is an important medical
imaging modality. A particle spin has angular momentum as well as a magnetic moment. The
magnetic moment will interact with an ambient magnetic field giving rise to the precession
of the spins. The precession of the spins, in the case of MRI, yields spin echoes that can
be measured for imaging and spectrocopic purposes. In the case of ferrites, it gives rise to
anisotropic, gyrotropic materials that exhibit Faraday rotation. Such effect can be used to
design non-reciprocal microwave devices such as isolators.
As mentioned, ferrite is commonly used as an anisotropic material in a waveguide to
make nonreciprocal waveguides [3, 4, 18–20]. Ferrite obtains its anisotropy by having its
electron spins interact with a static magnetic field. An electron spin has a magnetic dipole
moment as well as angular momentum. In the presence of a static magnetic field, this dipole
moment aligns itself with the magnetic field. If a transverse force, via an electromagnetic
field, is applied to tilt the direction of the dipole moment, and hence, the direction of angular
momentum, the spin precesses about the static magnetic field just as a spinning top precesses
about a gravitational field.
The correct description of the motion of an electron in the presence of a magnetic field
requires quantum mechanics. However, when a large number of electrons are considered, their
average motion can be described by a classical equation of motion similar to the equation
governing the motion of a spinning top. From this equation, we can understand the anisotropic
nature of ferrites being biased by a magnetic field.
For an electron with a magnetic dipole moment m in the presence of a magnetic field B0 ,
the torque exerted on the electron is given by
T = m × B0 . (3.10.1)
This torque is exerted on the angular momentum P of the electron, causing it to change.
Therefore, we have
dP
= T = m × B0 . (3.10.2)
dt
But the magnetic dipole moment of an electron is antiparallel to its angular momentum, i.e.,
m = −γP (3.10.3)
where γ is known as the gyromagnetic ratio. Consequently, the equation of motion for a
spinning electron in a magnetic field is
dP
= −γP × B0 = +ω 0 × P (3.10.4)
dt
where ω 0 = γB0 , and ω0 = |ω 0 | = γ|B0 | is also known as the Larmor frequency. For electron
spins in a reasonably strong magnetic field, the Larmor frequency can be in the microwave
regime.
Even though we say that (3.10.4) is the equation of motion for an electron, it actually
governs the average motion over an ensemble of a large number of electrons. Hence, a classical
picture applies here. Other subatomic particles like protons also possess a spin, but because
of their larger mass, the Larmor frequency is much lower. Proton spins, prevalent in MRI, has
a Larmour frequency of 42.6 MHz per Tesla,3 while electron spins have Larmor frequencies
in the GHz regime.
31 T=104 gauss=1 weber/m2 =1 Volt Second/m2 .
The Larmor frequency is the free precession frequency of a spin when it is tipped from
the vertical position in the presence of the static biasing magnetic field. Equation (3.10.4) is
sometimes known as the Bloch equation.
Assuming that the field B0 is static, then it is easy to show that the natural solution to
(3.10.4) consists of a P with a ẑ component and a circulating x̂ and ŷ component. Let us
assume that
1
P = √ (x̂ + iŷ)Ps e−iωt + ẑPz . (3.10.5)
2
The above is not a pure time-harmonic p signal as the second term can have a different frequency
from the first term. Note that |P| = Ps2 + Pz2 = P0 =constant even though it is time varying
as angular momentum has to be conserved. Also, this angular momentum comes from the
intrinsic spin of the particle, which is a constant. Also, P(t) can be obtained in the real
world by taking the real part of Equation (3.10.5), or by adding a complex conjugate term to
the above. For convenience, we will leave (3.10.5) with its phasors. Then, using (3.10.5) in
(3.10.4) yields
e−iωt dPz e−iωt
−iω(x̂ + iŷ)Ps √ + ẑ = (ω0 ŷ − iω0 x̂)Ps √ . (3.10.6)
2 dt 2
It is seen that the above is satisfied when we have
dPz
ω = ω0 , = 0. (3.10.7)
dt
In other words, the spin precesses at the Larmor frequency ω0 while the ẑ-component of its
angular momentum remains unchanged since the system is non-dissipative. In a dissipative
system, the kinetic energy in the angular momentum will be lost to the environment. The
precession of the spin will slow down, and eventually, the spin will be completely aligned with
the background static magnetic field.
Now if we include an additional RF field B1 , in the transverse direction which is circularly
polarized such that
1
B1+ = √ (x̂ + iŷ)B1+ e−iωt , (3.10.8)
2
then the equation of motion becomes
dP
= ω 0 × P + γB1+ × P. (3.10.9)
dt
The transverse RF field, which is the driving field, will tilt the spin and force it to precess at
the same frequency. Therefore, we let
1
P = √ (x̂ + iŷ)Ps e−iωt + ẑPz . (3.10.10)
2
Using (3.10.10) in (3.10.9), we have
e−iωt dPz e−iωt
−iω(x̂ + iŷ)Ps √ + ẑ = −iω0 (x̂ + iŷ)Ps √
2 dt 2
(3.10.11)
e−iωt
+ iγ(x̂ + iŷ)B1+ Pz √
2
B0
m
Electron
x
Figure 3.24: The magnetic dipole moment of an electron precesses just like a spinning top
when an RF transverse to z magnetic field is applied.
Again we have dPz /dt = 0 and

γB1+
Ps = Pz . (3.10.12)
ω0 − ω
The precession now is at the driving frequency ω. This equation so far, has been derived
with no approximation. Notepthat if the frequency ω = ω0 , a resonance occur, but Ps does
not go to infinity as |P| = Ps2 + Pz2 = P0 =constant and Pz → 0 when Ps → P0 . In
other words, as one drives the spin system closer to the resonance frequency, Ps Pz , but
Ps2 + Pz2 = constant. Therefore, the vector P has to tilt, and at resonance, the vector P is
precessing horizontally.
Next, we assume that the applied RF field has a magnetic field component given by a
circularly polarized field of opposite polarity such that
1
B1− = √ (x̂ − iŷ)B1− e−iωt (3.10.13)
2
and assuming then the spins are tipped from the vertical and precess at the same frequency
as the driving RF field, we let
1
P = √ (x̂ − iŷ)Ps e−iωt + ẑPz . (3.10.14)
2
Using (3.10.14) in (3.10.9) with B1+ replaced by B1− , we have
e−iωt dPz e−iωt e−iωt

−iω(x̂ − iŷ)Ps √ + ẑ = iω0 (x̂ − iŷ)Ps √ − iγ(x̂ − iŷ)B1− Pz √ . (3.10.15)
2 dt 2 2
The above implies that dPz /dt = 0, and that
γB1−
Ps = Pz . (3.10.16)
ω0 + ω
Assuming that the RF field is much smaller than the static field so that B1 B0 , then
we can assume that Ps is small or that Pz ' P0 . In this case, (3.10.12) and (3.10.16) together
become
γB1±
Ps± ≈ P0 , (3.10.17)
ω0 ∓ ω
and the transverse part of the spin angular momentum becomes
1 γB1±
Ps± ≈ √ (x̂ ± iŷ) P0 e−iωt . (3.10.18)
2 ω0 ∓ ω
By the above approximation, we have simplified the relationship between Ps± and the driving
field B1± .
The magnetic moment of a spin is given by m = −γP. Hence, the transverse part of the
magnetic moment is
1 γB1± m0 −iωt
ms± = √ (x̂ ± iŷ) e . (3.10.19)
2 ω0 ∓ ω
where m0 = −γP0 is the component of the magnetic moment in the ẑ-direction, where in the
approximation here it is the magnitude of the magnetic moment.
The magnetic dipole moment density or magnetization density is given by M = N m and
we have
1 γB1± M0 −iωt
Ms± = √ (x̂ ± iŷ) e . (3.10.20)
2 ω0 ∓ ω
where Ms = N ms and M0 = N m0 are the dipole moment density in the transverse and axial
directions respectively.
The above is the RF response of the transverse magnetization density in response to an
applied RF magnetic field given by
1
B1± = √ (x̂ ± iŷ)B1± e−iωt . (3.10.21)
2
In other words, we have
γM0
Ms± = B1± . (3.10.22)
ω0 ∓ ω
One can always make a linear polarization out of a linear superposition of two circular polar-
ization via
1
B = x̂Bx e−iωt = [(x̂ + iŷ)Bx + (x̂ − iŷ)Bx ] e−iωt . (3.10.23)
2
When this RF field is used to excite the spins, the RF magnetization response is given by

γM0 1 γM0 1
Ms = (x̂ + iŷ)Bx e−iωt + (x̂ − iŷ)Bx e−iωt
ω0 − ω 2 ω0 + ω 2

ω0 −iωt ω (3.10.24)
= x̂γM0 Bx e + iŷγM0 Bx e−iωt .
ω02 − ω 2 ω02 − ω 2
Similarly, when we make the RF field ŷ polarized such that

1
B = ŷBy e−iωt = [(x̂ + iŷ)By − (x̂ − iŷ)By ]e−iωt , (3.10.25)
2i
the RF magnetization response is

γM0 1 γM0 1
Ms = (x̂ + iŷ)By e−iωt − (x̂ − iŷ)By e−iωt
ω0 − ω 2i ω0 + ω 2i

ω −iωt ω0 (3.10.26)
= −ix̂γM0 By e + ŷγM B
0 y e−iωt .
ω02 − ω 2 ω02 − ω 2
Consequently, Bx = µ0 Hx , and By = µ0 Hy , we have in matrix form

Mx χxx χxy Hx
= , or M = χ · H (3.10.27)
My χyx χyy Hy
where
µ0 γM0 ω0
χxx = χyy = , (3.10.28a)
ω02 − ω 2
−iµ0 γM0 ω
χxy = −χyx = . (3.10.28b)
ω02 − ω 2
The above is the RF magnetization response to an RF magnetic field excitation in a ferrite
medium when the RF signal is assumed small, and we assume that ẑ component of the
magnetization response is small and is of higher order. The reason being that the z component
of the RF field has little effect on the spin momentum when it is pointed primarily in the z
direction. In general,
B = µ0 (H + M) = µ0 (I + χ) · H = µ · H (3.10.29)
where  
1 + χxx χxy 0
µ = µ0  χyx 1 + χyy 0 . (3.10.30)
0 0 1
This is an example of an anisotropic magnetic medium, or a gyrotropic magnetic medium.
The above permeability tensor is also Hermitian implying that it represents a lossless medium.
3.10.1 Natural Plane Wave Solutions in an Infinite Homogeneous

Ferrite Medium
A ferrite medium,which is a gyrotropic material, admits circularly polarized plane waves as
the natural plane waves propagating in a homogeneous anisotropic ferrites [21]. We can see
this by looking at the solution of the vector wave equation. When is homogeneous and
isotropic, and µ is anisotropic, the magnetic field satisfies the following vector wave equation:
∇ × ∇ × H − ω 2 µ · H = 0. (3.10.31)
Assuming that a plane wave solution propagating in the z direction exists such that
H = H0 eikz , (3.10.32)
then
−k 2 ẑ(ẑ · H) + k 2 H − ω 2 µ · H = 0. (3.10.33)
If µ is of the form
µs 0
µ= (3.10.34)
0 µ
where µs is a 2 × 2 tensor, then Hz = 0 by equating the z component of (3.10.33). Therefore
H has only transverse to z components, and Equation (3.10.33) becomes
k 2 Hs − ω 2 µs · Hs = 0 (3.10.35)
where Hs is a vector in the xy plane. From Equation (3.10.30), it is clear that
Bs± = µs · Hs± = µ± Hs± (3.10.36)
for a ferrite medium where

Hs± = H0± (x̂ ± iŷ). (3.10.37)
Consequently, (3.10.35) becomes
k 2 Hs± − ω 2 µ± Hs± = 0 (3.10.38)
where k can have two possible values given by

√
k± = ω µ± . (3.10.39)
Therefore, the natural solution in a ferrite medium is of the form
H± = H0± (x̂ ± iŷ)eik± z . (3.10.40)
In essence, in a ferrite medium, a right-hand circularly polarized wave “feels” a different per-
meability compared to a left-hand circularly polarized wave. Therefore, the two polarizations
propagate with different velocities. When the wave is not propagating in the z direction, the
propagation of the wave is more complicated, and will not be discussed here.
3.10.2 Faraday Rotation

Faraday rotation occurs in a ferrite. It can also occur in the earth ionosphere where the
electron spins are biased by the earth magnetic field. To understand Faraday rotation, we
decompose a linearly polarized wave into two circularly polarized waves, viz.,
1 1
E = x̂E0 = (x̂ + iŷ)E0 + (x̂ − iŷ)E0 , (3.10.41)
2 2
where the first term is left-hand circularly polarized for a wave propagating in the ẑ direction,
and likewise, the second term is right-hand circularly polarized. In a ferrite medium, these
AAAA AAAAAAAA
AAAA
AAAA AAAAAAAA
AAAA
45° Twist B0
1
AAAA AAAAAAAA
AAAA
Ferrite Rod
45° Faraday
2
Resistive Resistive
Card Rotation Card
Figure 3.25: A Faraday-rotation isolator.
two polarizations will propagate with different phase velocities. After a certain distance, there
could be a phase difference between them, and we have
1 1
E= (x̂ + iŷ)E00 eiθ + (x̂ − iŷ)E00 . (3.10.42)
2 2
Combining the terms in the above, we have

i θ2 θ 0 i θ2 θ
E = x̂e cos E0 − ŷe sin E0
2 2

θ θ (3.10.43)
= x̂ cos − ŷ sin E00 eiθ/2 .
2 2
Hence, the wave vector is now tilted by an angle − θ2 or has rotated clockwise.
For a wave propagating in the negative z direction, the polarizations in the first and second
term reverse roles. Now, the second term in (3.10.42) will have a phase gain over the first
term, and the wave vector is rotated counterclockwise after propagating through a certain
distance. Hence, the phenomenon is nonreciprocal meaning that the waves propagating in
the ±z directions are quite different in behavior.
3.10.3 Applications of Faraday Rotation

Faraday rotation can be used to make a number of nonreciprocal microwave devices. One
simple example is a gyrator, which is defined as a device whose transmission from port 1
to port 2 has a 180o phase shift compared to its transmission from port 2 to port 1. The
polarization rotation and the nonreciprocal natures of ferrites can be used to make such a
device.
Another device that can be made from ferrite loading is an isolator. An isolator consists
of first a 45o mechanical twist which is a reciprocal section. Then it is followed by another
section of 45o Faraday rotation which is the nonreciprocal section. Resistive cards are added
to filter undesired modes other than the TE10 mode.
For a mode transmitting from port 1 to port 2, the mechanical twist section rotates
the polarization counterclockwise by 45o while the ferrite section undoes it by rotating the
polarization clockwise by 45o . Hence the transmission from port 1 to port 2 is little affected.
AAA
AAA AAAAAA
AAAAAA
Static
AAA Static
AAAAAA
AAA AAAAAA
Magnetic Ferrites Magnetic
AAA
Field Field
Ferrites
Figure 3.26: Ferrite loading for resonance isolators.
For transmission from port 2 to port 1, the rotation in the ferrite section is counterclock-
wise, and hence is additive with respect to the rotation in the twisted section. Hence, when
the wave arrives at port 1, it has been rotated by 90o , and will not be able to transmit as the
TE10 mode.
Another way to use ferrite as an isolator is to notice the resonance behavior of ferrites.
The precessing ferrite spins acquire a larger component transverse to the static magnetic fields
when it is excited by one circular polarization compared to the other. When there is a loss
mechanism to dissipate the energy of the spins, the polarization that excites a larger spin
amplitude will dissipate more energy into the ferrite than the other polarization.
If we take the top view of the magnetic field distribution of the TE10 mode of a rectangular
waveguide, the magnetic field is actually circularly polarized away from the midsection of the
waveguide. The polarization changes sign for a +z or a −z propagating wave. Therefore, if
a waveguide is loaded with ferrites away from the middle of the waveguide as shown, it will
attenuate a TE10 mode by a different amount depending on the direction of propagation of
the mode. This can be used as an isolator, and it is called the resonance isolator.
Problem 3-1: Prove the identities in Equations (3.2.5a) (3.2.5b) of Section 3.1.
Problem 3-2: Consider the scattering problem involving a circular dielectric rod. The
incident wave is described by a TM wave given as
Ez = E0 J0 (kρ ρ)eikz z .
Match boundary condition at the surface of the dielectric rod, and find the scattered field.
Does the scattered field involve both TE and TM wave? Explain why.
Problem 3-3: If this world is dominated by left-handed people, we may have used left-
hand rule in cross products rather than right-hand rule. How should Maxwell’s equations be
rewritten if left-hand rule is used instead? Would the law of electromagnetics be affected if
left-hand rule is used instead?
Problem 3-4: Explain why if a general uniform waveguide is filled with an anisotropic
material such that and µ are 3 × 3 tensors, and that xz , yz , zx , zy , µxz , µyz , µzx , and
µzy are all not zero, then the waveguide does not have reflection symmetry. That is the
waveguide in the mirrored world is not the original waveguide anymore even after a 180o
rotation.
Problem 3-5:
(a) Prove that if two matrices are the transpose of each other, they share the same eigen-
values. Do they share the same eigenvectors also? Proof that their eigenvectors are
orthogonal to each other.
(b) Prove that Equation (3.3.7) is negative-transpose to Equation (3.3.6) in Section 3.2.
Problem 3-6: For the dielectric-slab-loaded rectangular waveguide shown:
(a) Assume that the dielectric slab is absent, plot the dispersion curve for the TE10 and
TE01 modes above cut-off. That is plot kz as a function of frequency.
(b) Now with the dielectric slab in place, the guidance properties of the aforementioned two
modes will be perturbed. Write down the guidance conditions from which you can find
the dispersion curves of the perturbed TE10 and TE01 modes. Define all the variables
in the guidance conditions so that if you need to calculate these guidance conditions,
you know how to.
(c) Write a computer program to solve for the roots of the guidance conditions and plot
the dispersion curves for the perturbed modes. (Note: You can use a Muller root solver
which is available in IMSL.)
Problem 3-7: Simplify Equation (3.8.24) of Section 3.5 for the n = 0 mode, or the axially
symmetric mode.
Problem 3-8: For an inhomogeneously filled waveguide, the equations governing the Ez and
Hz components of the fields are

ikz iω
µ∇s · µ−1 ∇s Ez − ikz µ∇s · ∇ E
s z + ∇ s × H 2
z + ω µEz = 0,
µks2 ks2

ikz iω
∇s · −1 ∇s Hz − ikz ∇s · ∇ H
s z − ∇ s × E 2
z + ω µHz = 0.
ks2 ks2
(a) For inhomogeneities which are piecewise constant, show that these two equations are
coupled only at the discontinuities of the piecewise constant inhomogeneity.
(b) For a homogeneously filled waveguide, show that the two equations are decoupled from
each other. What are the equations in this case?
Problem 3-9:
Find the guidance conditions for the symmetrically placed slab in the rectangular waveguide
for the LSM and LSE modes. Simplify the expressions for the guidance conditions as much
as possible.
Problem 3-10: If the equation for spin precession now has small loss terms, so that Pz → Pz0 ,
and Ps → 0, when t → ∞, namely,
dP 1 1
= ω 0 × P − Ps − ẑ (Pz − Pz0 ) ,
dt T2 T1
where T1 is the relaxation time for Pz , and T2 is the relaxation for Ps . Find the solution to
the above equation. What is the steady solution if a time-harmonic RF B1 field as in (3.10.8)
is applied to excite the system?
Problem 3-11: Analyze the problem in Subsection 3.10.1 when the wave is propagating in
the negative z direction for Faraday rotation.
Bibliography
[2] L.G. Chambers, “Compilation of the propagation constants of an inhomogeneously-filled
waveguide,” Br. J. Appl. Phys., 3, 19-21, 1952.
[5] A. Yariv, Optical Electronics, Holt, Rinehart, and Winston, New York, 1985.
[6] T. Okoshi, Opitcal Fibers, Academic Press, New York, 1992.
[7] R.H. Sheikh and M.W. Gunn, “Wave Propagation in a Rectangular Waveguide Inhomoge-
neously Filled with Semiconductors (Correspondence),” IEEE Trans. Micro. Theory and
Techniques, vol. MTT-19, no. 2, pp. 117-121, 1968.
[8] R. Feynman, R.B. Leighton, and M.L. Sands, The Feynman Lectures on Physics, vol. I,
Chapter 52, Addison-Wesley Publishing Co., 1965.
[9] W. C. Chew and M. Nasir, “A variational analysis of anisotropic, inhomogeneous dielectric
waveguides,” IEEE Trans. Microwave Theory Techniques, vol. 37, no. 4, pp. 661-668, Apr.
1989.
[10] J.F. Lee, D.K. Sun, and Z.J. Cendes, “Full wave analysis of dielectric waveguides using
tangential vector finite elements,” IEEE Trans. Microwave Theory and Techniques, vol.
39, no. 8, pp.1262-1271, August 1991.
[11] J. Jin, The finite element method in electromagnetics, John Wiley & Sons, Inc., New
York, 1993.
[12] P.H. Vartanian, W.P. Ayres, and A.L. Helgesson, “Propagation in Dielectric Slab Loaded
Rectangular Waveguide,” IRE Trans. Micro. Theory Tech., vol. 6, no. 4, pp. 215-222, April
1958.
[13] B. Lax, K.J.B Utkm, and L.M. Roth, “Ferrite phase shifters in rectangular waveguide,”
J. Appl. Phys., vol. 25, pp. 1413-1421, November, 1954.
133
[14] A.D. Bresler, “On the T En0 modes of a ferrite slab loaded rectangular waveguide and
the associated thermodynamic paradox,” IEEE Trans. Microwave Theory and Tech. vol.
8, no. 1, pp. 81-95, Jan 1960.
[16] S. Amari, R. Vahldieck, J. Bornemann and P. Leuchtmann, “Propagation in a circular

waveguide periodically loaded with thick dielectric disks,” IEEE MTT-S Int. Microwave
Symp. Dig., pp. 1535-1538, Baltimore, USA, June 1998.
[17] W. Li, M.L. Gong, Y.Y. Wei, H.Q. Xie, “The dispersive properties of a dielectric- rod
loaded waveguide immersed in a magnetized annular plasma,” Chinese Phys., 13, 54-59,
2004.
[18] B. Lax, “The Status of Microwave Applications of Ferrites and Semiconductors,” IRE
Trans. Micro. Theory Tech., vol. 6, no. 1, pp. 5-18, Jan 1958.
[19] L. Zhou and L. E. Davis, “Finite element method with edge elements for waveguides
loaded with ferrite magnetized in arbitrary direction,” IEEE Trans. Microwave Theory
and Techniques, vol. 44, no. 6, pp. 809-815, June 1996.
[20] B. Lax and K.J. Button, Microwave Ferrites and Ferrimagnetics, McGraw-Hill, NY,
1962.
[21] K.G. Budden, Radio Waves in the Ionosphere, Cambridge University Press, Cambridge,
UK, 1961.
Chapter 4
Coupling of Waveguides and

Cavities
Once we have a waveguide or a cavity, it is important to know how to couple energy

into it. Energy can be coupled into a waveguide by use of a probe, or an aperture, or
simply just by connecting one waveguide to another. We will first study the excitation of
modes in a waveguide by the use of a probe. We will develop the integral equation from
which such a problem can be solved exactly, as well as calculating the input impedance by a
variational formula. Such method can also be applied as well to cavity coupling, and coupling
of electromagnetic energy into free space as in antennas.
To study aperture coupling, we will also discuss the pertinent equivalent principle needed.
Such problems have also been addressed in [1, 2].
4.1 Excitation of Waveguides by a Probe

We will study the coupling of modes from a coaxial line to a waveguide as shown in Figure
4.1. Coaxial line is quite prevalent, and its characteristics can be easily understood via
transmission line theory. This coupling problem is important in understanding the transfer
of power from a coaxial cable to a waveguide system. By the proper adjustment of d and l,
we can cause almost all the power from the coaxial cable to be transferred to the waveguide.
From a transmission line theory viewpoint, the coupling to the waveguide is reflected in the
transmission line being terminated with a load. The load can be changed by the proper
adjustment of the dimension and location of the probe until a matched load is arrived at.
4.1.1 Derivation of the Equivalent Problem and the Integral Equa-

tion
In the waveguide volume enclosed by the surface S, the electric field satisfies the following
vector wave equation
∇ × ∇ × E(r) − k 2 E(r) = 0. (4.1.1)
135
y
l
^
n
Sp
V So
d Sa
S
Figure 4.1: Excitation of a cylindrical, hollow waveguide by a probe.
A dyadic Green’s function is defined to be a solution to the following equation, i.e.,
∇ × ∇ × G(r, r0 ) − k 2 G(r, r0 ) = Iδ(r − r0 ). (4.1.2)
Dot-multiplying (4.1.1) by G(r, r0 ) and (4.1.2) by E(r), upon subtraction and integration
over V , we have
Z
dr[E(r) · ∇ × ∇ × G(r, r0 ) − ∇ × ∇ × E(r) · G(r, r0 )] = E(r0 ). (4.1.3)
V
Using
∇ · [−E(r) × ∇ × G(r, r0 ) − (∇ × E(r)) × G(r, r0 )]

(4.1.4)
= E(r) · ∇ × ∇ × G(r.r0 ) − ∇ × ∇ × E(r) · G(r, r0 ),
and the divergence theorem, then

Z
E(r0 ) = − dS n̂ · [E(r) × ∇ × G(r, r0 ) + (∇ × E(r)) × G(r, r0 )], r0 ∈ V. (4.1.5)
S
The above can also be written as

Z
E(r) = − dS 0 [n̂0 × E(r0 ) · ∇0 × G(r0 , r) + iωµn̂0 × H(r0 ) · G(r0 , r)], r ∈ V. (4.1.6)
S
The above is actually a statement of Huygens’ principle for vector electromagnetic field: Given
the knowledge of tangential E and H fields on the closed surface of a volume V , the field is
known everywhere inside V .1
1 In (4.1.3), if r0 is outside V , the right-hand side will evaluate to zero, leading to the left-hand side of
(4.1.6) to be zero. This identity is known as the extinction theorem.

Coupling of Waveguides and Cavities 137
At this point, we have not specified the dyadic Green’s function to be used in Equation
(4.1.6) except that is is a solution of (4.1.2). A convenient choice is a dyadic Green’s function
that satisfies the boundary condition that
n̂ × G(r, r0 ) · a = 0, r ∈ wall, (4.1.7)
where a is an arbitrary vector. In other words,
n̂ × G(r, r0 ) = 0, r ∈ wall, (4.1.8)
where the wall is shown in Figure 4.2. In this case, the dyadic Green’s function is for an
electric field generated by an arbitrary point source in the geometry of Figure 4.2.
Figure 4.2: Equivalent sources for the probe excitation of a waveguide.
If this is the case, the last term in (4.1.6) can be written as
n̂0 × H(r0 ) · Ge (r0 , r) = −H(r0 ) · n̂0 × Ge (r0 , r) = 0, r0 ∈ wall. (4.1.9)
where the subscript e indicates an electric dyadic Green’s function. Hence, the second integral
in (4.1.6) is nonzero only on Sp . Since n̂ × E(r0 ) is nonzero only on Sa , the first integral is
only nonzero on Sa . Consequently, (4.1.6) becomes
Z
E(r) = − dS 0 n̂0 × E(r0 ) · ∇0 × Ge (r0 , r)
Sa
Z (4.1.10)
− iωµ dS 0 n̂0 × H(r0 ) · Ge (r0 , r).
Sp
The integral over S0 in Figure 4.1 can be made to vanish by taking S0 to infinity and
introducing an infinitesimal amount of loss. −n̂0 ×H(r0 ) can be identified as Jp (r0 ), the surface
current on the probe, and n̂0 ×E(r0 ) can be identified as Ma (r0 ), as equivalent magnetic current
impressed on Sa . Rewriting (4.1.10), we have
Z Z
0 0 0 0
E(r) = − dS Ma (r ) · ∇ × Ge (r , r) + iωµ dS 0 Jp (r0 ) · Ge (r0 , r). (4.1.11)
Sa Sp
Using the reciprocity relations, it can be shown that [3, p. 32] [see also Problem 4-1]
[Ge (r0 , r)]t = Ge (r, r0 ), (4.1.12a)
[∇0 × Ge (r0 , r)]t = ∇ × Gm (r, r0 ) (4.1.12b)

and (4.1.11) becomes
Z Z
0 0 0
E(r) = − dS ∇ × Gm (r, r ) · Ma (r ) + iωµ dS 0 Ge (r, r0 ) · Jp (r0 ). (4.1.13)
Sa Sp
where the subscript m indicates a magnetic dyadic Green’s function. If Ma (r0 ) is assumed
known, then using n̂ × E(r) = 0 on Sp , an integral equation can be set up
Z Z
iωµn̂ × dS 0 Ge (r, r0 ) · Jp (r0 ) = n̂ × dS 0 ∇ × Gm (r, r0 ) · Ma (r0 ), (4.1.14)
Sp Sa
from which Jp (r0 ) can be solved for. In Equation (4.1.13), the problem in Figure 4.1 is
replaced with an equivalent problem consisting of impressed current Jp on Sp and Ma on Sa .
These are impressed current because they are sources impressed in the space V radiating via
the waveguide dyadic Green’s function.
4.1.2 Generalization to Other Structures

The above theory is quite general. It obviously applies to arbitrarily shaped waveguides or
cavities. When applied to an arbitrarily shaped structure as shown in Figure 4.3, the onus is
on finding the dyadic Green’s function. However, one can also use free-space dyadic Green’s
function, but at the expense of adding more unknowns to the integral equation.
The above formulation can also be applied to a monopole antenna mounted on a ground
plane driven by a coaxial cable from below. In this case, we can use a dyadic Green’s function
that satisfies the boundary condition of a metallic half space. This Green’s function can be
found easily using image theorem.
It can also be applied to a metallic antenna driven by a magnetic current source in free
space. We can apply the free-space dyadic Green’s function in this case. One can assume
that Sa and Sp form a closed surface. In this case, there are impressed electric current on
the surface Sp . But on the surface Sa , there would be both impressed magnetic and electric
current.
In the above cases, there is a bounding surface at infinity, Sinf that has to be included.
By use of the radiation condition, the contribution from this bounding surface can be shown
to vanish.
Swall Sinf
Infinity
Sp
Sa
Sa
Sp Sp
Jp
Sa Sa Ma  J p
Figure 4.3: Excitation of a different structure with a magnetic current source. Top: A coaxial
driven probe exciting an arbitrary waveguide. Bottom left: A monopole antenna driven by
a coax via a ground plane. Bottom right: A general antenna driven by a voltage source
represented by an equivalent magnetic current source.
4.2 Input Impedance of the Probe

The input impedance of a probe exciting a waveguide can be calculated. This yields infor-
mation on how the position and length of the probe can be adjusted to arrive at the input
impedance we desire. For maximum transfer of the power, the input impedance should be
matched to the characteristic impedance of the transmission line to minimize reflections.
4.2.1 Variational Expressions for Input Admittance

Variational expressions for the input impedance of an antenna has been discussed in [4–6].
Such expressions can also be used to find the input impedance of a probe inside the waveguide,
as the two problems are very similar. The difference appears mainly in the Green’s function.
In an antenna, the source is radiating in free space, and hence, free-space Green’s function
usually suffices. However, in a waveguide, the free-space Green’s function has to be replaced
by the Green’s function of the waveguide structure.
A variational expression for the input impedance or admittance of a source driven complex
structure can be derived. The source can be either a voltage source or a current source.
When it is a voltage source, it is a magnetic current ring, and when it is a current source,
it is an electric dipole such as a Hertzian dipole or its equivalence. The case of a current
source has been adequately discussed in [22]. The voltage source is usually modeled by a
magnetic ring current, a ribbon current, or a magnetic frill. The case of a magnetic current
exciting an antenna structure has been discussed in [6]. However, our discussion here is in
accordance to [7], which is different from the previous treatment on the subject. However, the
variational formula for a magnetic current source driven antenna seems to have been missed
until presented in [7].
In any case, the input impedance of the structure is predominantly determined by the
induced current on the structure. A variational expression has the advantage of yielding a
second order error for the input impedance when the error of the current on the structure
is first order. The current on the complex structure can be roughly estimated, or solved for
from an integral equation.
If a probe current produces a magnetic field Hp , and the aperture magnetic current
produces an electric field Ea and a magnetic field Ha , then
Z Z
−hMa , HT i = − dS Ma (r ) · HT (r ) = − dS 0 n̂ · (Ea × HT )
0 0 0
(4.2.1)
Sa Sa
where HT = Hp + Ha . On the aperture Sa , we can assume that only the TEM mode of the
coax is important. Therefore,
Ea = ρ̂E0 , HT = φ̂H0 , (4.2.2)
Substituting (4.2.2) into (4.2.1), we have
Z Zb Z2π
0
− dS n̂ · (Ea × HT ) = dρE0 ρdφH0 = V I. (4.2.3)
Sa a 0
In the above, the fields are represented in a local coordinate system, and n̂ = −ẑ. Also, even
though HT may depart from the TEM mode field of a waveguide, by the mode orthogonality
theorem, only the TEM-mode component of HT will contribute to the integral in Equation
(4.2.3).
Does this mean that the higher-order modes in HT do not contribute to the current I?
No, it does not. The higher-order modes do contribute to the current I near the aperture,
but their effect diminishes rapidly away from the aperture. Therefore, the current I in (4.2.3)
is only the TEM component of the current, which can be easily related to the current in the
rest of the coaxial cable by transmission line theory.
Consequently,
V I = −hMa , HT i. (4.2.4)
By letting I = Yin V , we deduce that
hMa , HT i
Yin = − . (4.2.5)
V2
The above is an expression for the input admittance of the probe assuming only the TEM
mode in the coax, but it is not variational. To derive a variational expression, we write [7]
−hMa , HT i + hJp , ET i
Yin = (4.2.6)
V2
where ET = Ea + Ep , the total electric field produced both by Ma and Jp . The above also
falls under the category of the reaction formula for the input impedance of antennas. Notice
that if ET is exact, then the tangential component of ET is zero on the probe surface and
hJp , ET i would be zero. However, the second term in the numerator of (4.2.6) is required to
make it a variational expression. In other words, first order error Jp will result in a second
order error in Yin . To prove that (4.2.6) is variational, we let Ma and V be known and hence
fixed, and let
Jp = Jpe + δJ, Yin = Yine + δY (4.2.7)
where the subscript e stands for “exact”. Cross-multiplying (4.2.6), and taking the first
variation, we have
δY V 2 = −hMa , δHi + hδJ, ET e i + hJpe , δEi. (4.2.8)
From reciprocity,
hδJ, ET e i = −hMa , δHi + hJpe , δEi. (4.2.9)
Then
δY V 2 = 2hδJ, ET e i. (4.2.10)
Since
hδJ, ET e i = 0, (4.2.11)
because ET E has no tangential components on the probe surface and δJ is purely tangential
on the probe surface, (4.2.10) implies that
δY = 0. (4.2.12)
As a result, the first variation in the admittance about the exact admittance Yine is zero.
Equation (4.2.6) is a variational expression for the input admittance. Given Ma , Jp and V
with first order errors, the errors incurred in Yin is of second order.
The variational nature of Equation (4.2.6) can be better appreciated if its quadratic nature
is written more explicitly. To this end, it can be written as
hMa , Ha i hMa , Hp i hJp , Ep i hJp , Ea i
Yin = − − + + (4.2.13)
V2 V2 V2 V2
where we assume Ma , and hence, V , Ha , and Ea are fixed. When Jp is varied, only the last
three terms would vary. By reciprocity, −hMa , Hp i = hJp , Ea i, and the above becomes
hMa , Ha i hJp , Ea i hJp , Ep i
Yin = − 2
+2 + . (4.2.14)
V V2 V2
Furthermore,
hJp , Ep i = iωµhJp , Ge , Jp i
Z Z
= iωµ dSJp (r) · dS 0 Ge (r, r0 ) · Jp (r0 ). (4.2.15)
Sp Sp
The double commas in the above implies that there is a double integration in the inner
product. The above is analogous to at · A · a in linear algebra. When it is used in (4.2.14), it
becomes
hMa , Ha i hJp , Ea i hJp , Ge , Jp i
Yin = − +2 + iωµ . (4.2.16)
V2 V2 V2
The above is clearly quadratic and has a stationary point about the exact solution.
For an exact ET , hJp , ET e i = 0 and (4.2.6) reduces to (4.2.5) again. Since HT = Ha +Hp ,
we have from (4.2.5)
hMa , Ha i hMa , Hp i
Yin = − − (4.2.17)
V2 V2
for exact solutions.
Furthermore, −hMa , Hp i = hJp , Ea i = −hJp , Ep i from reciprocity and that hJp , ET i = 0
for exact solutions. Hence, (4.2.17) becomes
hMa , Ha i hJp , Ep i
Yin = − − . (4.2.18)
V2 V2
If Ma is assumed real, the first term is the complex conjugate of the complex power hMa , H∗a i
due to Ma alone and can be related to the gap capacitance at the base of the probe. The
second term in (4.2.18) is the complex conjugate of the complex power hEp , J∗p i due to Jp
alone. Hence, it is due to the probe admittance. Since the terms in (4.2.18) are additive, the
gap capacitor is in parallel connection with the probe admittance.
It is to be reminded that in the use of (4.2.6), HT and ET are to be calculated from Ma
and Jp . In particular,
Z
Hp (r) = dS 0 ∇ × Ge (r, r0 ) · Jp (r0 ), (4.2.19)
Sp
Z
Ea (r) = − dS 0 ∇ × Gm (r, r0 ) · Ma (r0 ). (4.2.20)
Sa
4.2.2 Rayleigh-Ritz Method

The Rayleigh-Ritz method is named after Lord Rayleigh [17], a prodigious English scientist,
and Walter Ritz [18], a Swiss mathematician. It is extremely useful in solving complex
physical problems. It seems that many physical phenomena are always described by an
equation which corresponds to the minimization of a certain quantity. So instead of solving
the equation direction, one can attempt to minimize the corresponding quantity instead.
This method is extremely useful if we have an expression for Yin which will have Yine
(exact Yin ) as the lower bound or the upper bound to all the approximate Yin . (However, Yin
is actually a complex number, but for the sake of the ease for discussion, we will assume that
Yin is real.) Without loss of generality, let us discuss the lower bound case. For example, if
Yin = f (Jp ) ≥ Yine , (4.2.21)

and the equality is satisfied only if Jp = Jpe , an optimal value of Yin can be obtained even
with an approximate Jp . We can let
N
X
Jp = an Jn , (4.2.22)
n=1
where Jn is a set of basis functions with which an arbitrary Jp can be approximated fairly
well. The coefficients an ’s are yet to be determined to give the best approximation to Jp .
The Rayleigh-Ritz procedure provides a systematic way to determine the optimal values of
an ’s so as to best determine Yin from (4.2.21). If we substitute (4.2.22) into (4.2.21), then
N
!
X
Yina = f an Jn > Yine . (4.2.23)
n=1
Figure 4.4: Gap capacitances can be important in the input impedance of a probe.
The best choice of an ’s will be those that minimizes the number Yina . In other words,
from Figure 4.6. the optimal values of an ’s are those that would make Yina stationary. Hence,
there are N equations N
P
∂f an Jn
n=1
= 0, i = 1, . . . , N, (4.2.24)
∂ai
from which we can solve for the optimal ai ’s, the aio ’s.
This concept may not work as well when Yin is a complex function, or when the stationary
point is not a global minimum or maximum, but a saddle point instead. However, for these
cases, the Rayleigh-Ritz procedure converges despite as we shall explain later.
Equation (4.2.14) is a variational expression when the solution is varied about Jpe . Hence,
the Rayleigh-Ritz procedure can be used to find the optimal Yin . To do so, we assume
N
X
Jp = an Jn . (4.2.25)
n=1
Yin
Yine
J pe Jp
Figure 4.5: A pictorial representation of a stationary point in a multidimensional space.
Yina
Yino
Yine
ao a
Figure 4.6: The minimum Yino achieved by the Rayleigh-Ritz procedure compared to the
exact minimum, Yine .
Then (4.2.16) becomes

P PP
an hJn , Ea i an an0 iωµhJn , Ge , Jn0 i
hMa , Ha i n n n0
Yin =− +2 + . (4.2.26)
V2 V2 V2
The above is of the form
hMa , Ha i at · e at · A · a
Yin = − + 2 + (4.2.27)
V2 V2 V2

where [e]n = hJn , Ea i, A nn0 = iωµhJn , Ge , Jn0 i, and [a]n = an . The matrix A is the
matrix representation of the dyadic Green’s function iωµGe , while the vector e is the vector
representation of E.
Taking the first variation of the above about ao , the optimal a, we have
δat · e δat · A · ao
δYin = 2 + 2 . (4.2.28)
V2 V2
The first variation δYin will vanish if
A · ao = −e. (4.2.29)
The above could be solved to yield the optimal ao which can in turn be used to find Jp
in (4.2.25). Yin0 can also be found by the direct substitution of (4.2.29) into (4.2.27) yielding
hMa , Ha i ato · e
Yin0 = − + . (4.2.30)
V2 V2
Equation (4.2.29) is also more directly obtained by solving the following integral equation
which follows from −n̂ × Ea = n̂ × Ep on the probe surface
Z
−n̂ × Ea = iωµn̂ × dS 0 Ge (r, r0 ) · Jp (r0 ). (4.2.31)
Sp
The use of Galerkin’s method in solving (4.2.31) yields (4.2.29).

The reason why the Rayleigh-Ritz procedure converges to the exact solution is because
most of the variational expressions we work with are quadratic in nature. This quadratic
nature is the generalization of the quadratic expression in one dimension such as ax2 + bx + c
to higher dimensions. For example, (4.2.27) is a quadratic expression in the variable a while
(4.2.13) is quadratic in the variable Jp . The space for the inner product in (4.2.27) is an
approximate finite dimensional space, while the space for the inner product in (4.2.13) is
infinite dimensional. All quadratic expressions have only one stationary point. Consequently,
as one increase the dimension of the approximate finite dimensional space, the stationary
point will approach that of the infinite dimensional space.
4.2.3 Mode Matching Method—A Tour de Force Calculation

In the previous section, the equation for the input admittance, Equation (4.2.5) is valid only
when the field at Sa can be approximated by only TEM modes. This is not true in general.
A more accurate analysis of the input impedance of the probe requires the use of the mode-
matching method [7]. This will allow us to use higher order modes in the coaxial waveguide.
It also allows us to control the accuracy of the calculation as much as we want to. When very
high accuracy solution is needed, we just need to add more modes in the solution procedure.
Inside the coaxial waveguide, the field is assumed to be
L−1
X
E(r) = E0 (rs )eikz + Γl El (rs )e−iklz z , (4.2.32)
l=0
where E0 (rs ) is the field distribution of the TEM mode, and Em (rs ), m > 0 are the fields of
the higher order modes which are evanescent. The corresponding magnetic field is
L−1
X
H(r) = H0 (rs )eikz − Γl Hl (rs )e−iklz z . (4.2.33)
l=0
Hence, the total electric field at the aperture in accordance with Equation (4.2.32) is
L−1
X
Ea (rs ) = E0 (rs ) + Γl El (rs ). (4.2.34)
l=0
The corresponding magnetic current is

L−1
X
Ma (rs ) = Ea × n̂ = M0 + Γl Ml (4.2.35)
l=0
where Ml = El × n̂.
The electric field in the waveguide region in the volume V due to the magnetic current
Ma is given by
Z
Ea (r) = dS 0 ∇ × Gm (r, r0 ) · Ma (r0 )
Sa
Z
= dS 0 ∇ × Gm (r, r0 ) · M0 (r0 )
(4.2.36)
Sa
L−1
X Z
+ Γl dS 0 ∇ × Gm (r, r0 ) · Ml (r0 ), r ∈ V,
l=0 Sa
where Gm (r, r0 ) is the magnetic-type dyadic Green’s function.

Ea from (4.2.36) can be substituted into (4.2.31) to yield
Z
−n̂ × dS 0 ∇ × Gm (r, r0 ) · M0 (r0 )
Sa
L−1
X Z
− Γl n̂ × dS 0 ∇ × Gm (r, r0 ) · Ml (r0 ) (4.2.37)
l=0 Sa
Z
= iωµn̂ × dS 0 Ge (r, r0 ) · Jp (r0 ).
Sp
Expanding Jp as in (4.2.25), we obtain

Z
−n̂ × dS 0 ∇ × Gm (r, r0 ) · M0 (r0 )
Sa
L−1
X Z
− Γl n̂ × dS 0 ∇ × Gm (r, r0 ) · Ml (r0 ) (4.2.38)
l=0 Sa
N
X Z
= iωµ an n̂ × dS 0 Ge (r, r0 ) · Jn (r0 ).
n=1 Sp
Testing the above equation with n̂ × Jm (r), m = 1, . . . , N, we have

−hJm (r), ∇ × Gm (r, r0 ), M0 (r0 )i
L−1
X
− Γl hJm (r), ∇ × Gm (r, r0 ), Ml (r0 )i
l=0 (4.2.39)
N
X
= iωµ an hJm (r), Ge (r, r0 ), Jn (r0 )i, m = 1, . . . , N.
n=1
In the above, the notation

Z Z
hf (r), G(r, r0 ), g(r0 )i = dSf (r) · dS 0 G(r, r0 ) · g(r0 ). (4.2.40)
Sp
Notice that the total field inside the waveguide is given by

ET = Ea + Ep (4.2.41)
where Ep is the field produced by the probe current, i.e.,
Z
Ep = iωµ dS 0 Ge (r, r0 ) · Jp (r0 ). (4.2.42)
Sp
Because of the electric dyadic Green’s function used here, n̂ × Ep = 0 on Sa . Therefore,

n̂ × ET = n̂ × Ea on Sa . (4.2.43)
Consequently, the total field given by (4.2.41) calculated via the use of (4.2.36), and (4.2.43)
satisfies the boundary condition that n̂ × ET is continuous at Sa from the coaxial waveguide
to the main waveguide.
Next, we need to impose the boundary condition that the tangential component of the
magnetic field is continuous across Sa . To this end, we find HT = Ha + Hp where
Z
Ha (r) = iω dS 0 Gm (r, r0 ) · Ma (r0 ), (4.2.44a)
Sa
Z
Hp (r) = dS 0 ∇ × Ge (r, r0 ) · Jp (r0 ). (4.2.44b)
Sp
On expanding Jp as in (4.2.25), and Ma as in (4.2.35), we have

Z
HT (r) = iω dS 0 Gm (r, r0 ) · M0 (r0 )
Sa
L−1
X Z
+ iω Γl dS 0 Gm (r, r0 ) · Ml (r0 ) (4.2.45)
l=0 Sa
N
X Z
+ an dS 0 ∇ × Ge (r, r0 ) · Jn (r0 ).
n=1 Sp
The boundary condition requires that the tangential components of H in (4.2.33) and (4.2.45)
be continuous. Equating (4.2.33) and (4.2.45), and testing the result with Mp (r), p =
0, . . . , L − 1, we have
δop λ0 − Γp λp = iωhMp (r), Gm (r, r0 ), M0 (r0 )i

L−1
X
+ iω Γl hMp (r), Gm (r, r0 ), Ml (r0 )i
l=0 (4.2.46)
N
X
+ an hMp (r), ∇ × Ge (r, r0 ), Jn (r0 )i, p = 0, . . . , L − 1.
n=1
In the above, we have made use mode orthogonality to arrive at

Z Z
hMp , Hl i = dSMp · Hl = dSEp × n̂ · Hl
Sa Sa
Z (4.2.47)
=− dS n̂ · Ep × Hl = −δpl λp
Sa
where Z
λp = dS n̂ · Ep × Hp (4.2.48)
Sa
Equations (4.2.39) and (4.2.46) constitute N + L equations for N + L unknowns, Γl , l =

0, . . . , L − 1, and an , n = 1, . . . , N . They can be solved by using matrix inversion. Once Γ0 ,
the reflection coefficient of the TEM mode is found, and assuming that only the TEM mode
propagates in the coaxial waveguide region, the input impedance of the probe is given by
1 − Γ0
Yin = Y0 . (4.2.49)
1 + Γ0
The advantage of this approach is that the input admittance can be found to any desired
numerical accuracy by increasing the number of terms in (4.2.25) and (4.2.35). When only
one mode is used in the coaxial region, it can be shown that this method, with the input
impedance given by (4.2.49), yields the same answer as the method using the variational
expression.
4.3 Excitation of a Microstrip Patch Antenna

A microstrip patch antenna is made by etching a patch on top of a dielectric substrate backed
by a ground plane. It was first proposed by Deschamps in 1953 [8], and put into practice
by Munson in 1972 [9]. Microstrip patch antenna is a very popular antenna because of its
ease of fabrication, light weight, and conformal nature. Because of the proximity of the
radiation source to a ground plane, and cancelation of the radiation field due to a negative
image current on the ground plane, the current on a microstrip patch is a poor radiator
ordinarily. However, it can be made to radiate well if resonant modes exist on the patch. At
the resonant frequency of the patch, the current amplitude can be greatly enlarged, enhancing
the radiation field despite negative image current cancelation. Hence, it radiates by resonance
coupling [10–13].
A microstrip patch antenna can be thought of as a cavity-backed slot antenna. The
radiation is actually from the side walls or slots of the antenna, and will not radiate well
unless it is backed by a resonant structure. Because of the cavity nature of the antenna, it is
generally narrow band, but much ingenious design has made these antennas operate with a
broader bandwidth.
4.3.1 Magnetic Wall Model

A microstrip patch can be approximated by a magnetic wall model where the side walls of
the patch is replaced with magnetic walls and the top and bottom patches remain metallic.
The magnetic walls have n̂ × H = 0 boundary condition while that on the metallic walls is
n̂ × E = 0. In short, they are perfectly conducting walls. We can assume that the substrate is
∂ ∂
thin so that ∂z = 0 for the field. Only very high order modes will have ∂z 6= 0. These modes
will be far away from the operating frequency of the patch, so that they are weakly excited.
We assume that only TMz modes are important since TEz modes will be shorted out. So
the field inside the cavity can be written as

mπx nπy
Emn = ẑEmn cos cos (4.3.1)
a b
Notice that we have chosen the solution to satisfy the Neumann boundary condition on the
magnetic wall so that tangential magnetic field is zero there.2
patch
ground
plane
substrate, r  1
Figure 4.7: Microstrip patch antenna: (a) Side view. (b) Top view.
2 We assume that the probe used is such that there is no charge accumulation on the probe and hence, only
divergence-free modes need to be considered in the mode expansion. Such is the case if the current on the
probe has constant current such that ∇ · J = 0.
The resonant frequency of the TMmn is given by

2 2
2 mπ nπ
kmn = + (4.3.2)
a b
The normalization constant is
12
4
Emn = (4.3.3)
ab(1 + δ0m )(1 + δ0n )
so that the modes are orthonormal. Using the fact that
∇ × ∇ × E − k 2 E = iωµJ (4.3.4)
and letting X
E= amn Emn (x, y) (4.3.5)
m,n
and that
2
∇ × ∇ × Emn − kmn Emn = 0 (4.3.6)
we obtain that
hEmn , Ji
amn = iωµ 2 − k2
(4.3.7)
kmn
If we assume that
J(x, y) = ẑI0 δ(y − y 0 )B(x − x0 ) (4.3.8)
where (
1 1, |x| ≤ w/2
B(x) = (4.3.9)
w 0, |x| > w/2
is a box function. Then
x0 + w
0 Z 2
nπy 1 mπx
hEmn , Ji = Emn cos cos dx (4.3.10)
b w a
x0 − w
2
By letting
x0 + w −1 x0 + w2
Z 2
1 mπx 1 mπ mπx
I= dx cos = sin 0 w
w a w a a x− 2
x0 − w
2
−1 w w
1 mπ mπ(x0 + 2) mπ(x0 − 2)
= sin − sin
w a a a
(4.3.11)
and by using
sin(A + B) − sin(A − B) = 2 cos A sin B (4.3.12)

then

2a mπx0 mπw mπx0 mπw
I= cos sin = cos sinc (4.3.13)
mπw a 2a a 2a
where sinc(x) = (sin x)/x.

Hence

mπx0 nπy 0 mπw
hEmn , Ji = Emn cos cos sinc (4.3.14)
a b 2a

φmn (x0 , y 0 )sinc mπw2a
amn = iωµ 2 − k2
(4.3.15)
kmn
X
1 mπw
E = iωµẑ φ (x, y)φmn (x0 , y 0 )sinc
2 − k 2 mn
(4.3.16)
k
m,n mn
2a
where
mπx mπy
φmn (x, y) = Emn cos cos (4.3.17)
a b
The above is the magnetic wall cavity model for the patch antenna. However, it has
no loss and the resonant frequencies of the modes of the cavity are purely real. When the
operating frequency coincides with the resonant frequency, from (4.3.7), it is seen that the ex-
citation coefficient of the mode becomes infinite. This is unphysical, as the resonant resonant
frequency of the cavity is never real in practice: the resonant modes of the patch antenna
are radiationally damped. Hence, its resonant frequencies are complex rather than real. In
addition, there are material loss and copper loss of the antenna that causes the modes to have
complex resonant frequencies, giving rise to damped resonances. The dielectric loss can be
easily incorporated by using a complex dielectric. The radiation damping can be modeled by
a lossy magnetic wall while the copper loss can be modeled by a lossy electric wall.3
Figure 4.8: Excitation of the patch antenna with a probe (cross-section view).
For a real resonant frequency, the fields of the mode will have a e−iωmn t time dependence.
If the resonant frequency is complex with ω̃mn = ωmn − iαmn , then the time dependence of of
the fields is e−iωmn t e−αmn t . The energy density of the mode is proportional to |E|2 and |H|2 ,
3 The excitation of the cavity modes by a source is elaborately dealt with in [2, 20]. An application to
microstrip antenna is given in [12].

and hence the stored energy Wmn,T ∼ e−2αmn t . By energy conservation, the power radiated
by the mode is proportional to the negative time rate of change of stored energy, or
∂
Pmn,rad = − Wmn,T = 2αmn Wmn,T
∂t
Hence, if the stored energy and the power radiated is known, the imaginary part of the
resonant frequency αmn can be found. This is equivalent to finding the Q of the resonant
modes of the antenna.
Actually, the reactive power leakage from the antenna also gives rise to a real resonant
frequency shift. The estimate of this shift is more difficult and method of estimating this shift
is given in [14–16] using perturbation approach and asymptotic approach. In retrospect, the
fringing field at the open edge of the patch makes the patch effectively larger, lowering the
real resonant frequency compared to that predicated by the magnetic wall model.
4.3.2 The Q of the Modes

Due to radiation damping, and other losses in the cavity, the resonant frequency of each mode
is not purely real. We shall discuss how to estimate the damping rate due to radiation. To
estimate the power radiated by the patch, we need to find the current on the patch. To this
end, we derive the magnetic field in the cavity:
∇ × Emn = iωµHmn (4.3.18)
or that
1 h nπ mπx nπy mπ mπx nπy i
Hmn = − x̂ cos sin − ŷ sin cos
Emn ωµ b a b a a b
(4.3.19)
The corresponding current on the top patch is
1 h mπ mπx nπy nπ mπx nπy i
Jmn = −ẑ × Hmn = x̂ sin cos + ŷ cos sin
iωµ a a b b a b
(4.3.20)
If this current is radiating in free space, its field is given by
Z Z 0
∇∇ eik|r−r |
Emn,R = iωµ dr0 Ḡ (r, r0 ) · Jmn (r0 ) = iωµ Ī + 2 · dr0 Jmn (r0 ) (4.3.21)
k 4π|r − r0 |
By letting
0
eik|r−r | eikr −ikr̂·r0
0
∼ e (4.3.22)
4π|r − r | 4πr
Hence, ikr Z
∇∇ e 0
Emn,R ≈ iωµ Ī + 2 · dr0 e−ikr̂·r Jmn (r0 ) (4.3.23)
k 4πr
Furthermore, we have
eikr eikr
∇∇ eikr
Ī + 2 ≈ Ī − k̂ k̂ = θ̂θ̂ + φ̂φ̂ (4.3.24)
k r r r
and
eikr Z 0
Emn,R ≈ iωµ θ̂θ̂ + φ̂φ̂ · dr0 e−ikr̂·r J(r0 )
4πr
eikr
= iωµ θ̂θ̂ + φ̂φ̂ · J̃(kr̂) (4.3.25)
4πr
The last integral corresponds to a Fourier integral with the Fourier spectral variable evaluated
on the energy shell. Since J(r) consists of sinusoidal functions, their Fourier transforms can
be evaluated in closed form. Hence, J̃(kr̂) can be found.
The above is the electric field radiated via the free-space dyadic Green’s function. In order
to account for the fact that this current is radiating on a dielectric substance backed by a
ground plane, we need only to add the reflected wave term. The reflected wave can be added
using ray physics since the observation point is in the far field where ray physics applies [3].
Consequently,
eikr n ˜ h i h io
Emn,R ≈ iωµ θ̂Jθ (kr̂) 1 − R̃T M (r̂) + φ̂J˜φ (kr̂) 1 + R̃T E (r̂) (4.3.26)
4πr
where R̃T M and R̃T E are the generalized reflection coefficient for the layered medium rep-
resenting the substrate with a ground plane. The minus sign in front of R̃T M is because Jθ
produces TM fields of opposite polarities above and below the source which is assumed to be
an infinitely thin sheet.
The power density radiated by this mode is then
1 2
Smn (r, θ, φ) = |Emn,R | (4.3.27)
2η
h i2 h i2
ω 2 µ2 1 ˜ TM ˜ TE
= Jθ (kr̂) 1 − R̃ (r̂) + Jφ (kr̂) 1 + R̃ (r̂) (4.3.28)
2η |4πr|2
where r̂ is a function of (θ, φ).

The total power radiated can be found by performing the following integral over a hemi-
sphere, namely,4
Z 2π Z π2
Pmn,rad = Smn (r, θ, φ)r2 sin θdθdφ (4.3.29)
0 0
The stored energy of the mode is
Z Z
1 2 2
1
Wmn,T = |Emn | + µ|Hmn | dV = |Emn |2 dV (4.3.30)
4 V 2 V
in the cavity between the patches. Therefore,
Z
1 1
Wmn,T = |Emn |2 dV = (4.3.31)
2 V 2
4 In addition, the patch current can excite a surface wave mode in the dielectric substrate layer causing
further loss, and damping of the mode. This is not included in this integral.
Im[ ]
 Re[ ]
10 01
Figure 4.9: Splitting of the degenerate TM10 and TM01 modes.
Consequently, the Q of the mn mode is
ωmn Wmn,T ωmn

Qmn = = = ωmn τmn (4.3.32)
Pmn,rad 2αmn
where τmn is the decay time constant of the mn modes, and the complex resonant frequency
of the mode is
ω̃mn = ωmn − iαmn (4.3.33)
From (4.3.32),
2Qmn
αmn = (4.3.34)
ωmn
The Q can be estimated using the integral (4.3.29) and expression (4.3.32).
4.3.3 Circular Polarization Excitation

In a square microstrip patch, the TM10 and TM01 modes are degenerate. If the operating
frequency is chosen close to that of these modes, they will be dominant. Hence, (4.3.5) can
be approximated by only two modes, namely,
E∼
= a10 E10 (x, y) + a01 E01 (x, y) (4.3.35)
where
hE10 , Ji hE01 , Ji
a10 = iωµ 2 − k2
, a01 = iωµ 2 − k2
(4.3.36)
k̃10 k̃01
where k̃10 and k̃01 are the complex resonant frequencies of the modes. If the probe is located
such that x0 = y 0 or along diagonal of the square patch, and sinc( mπw
2a ) ≈ 1 for both TM01 and
TM10 modes, then a10 ≈ a01 , as can be seen from (4.3.16). The mode currents of these modes
are or orthogonal to each other in space, but the field produced is not circularly polarization.
However, circular polarization can be obtained by making these modes non-degenerate by
destroying the symmetry. We can let a = b + 4, so that the TM10 mode has a slightly lower
resonant frequency compared to the TM01 mode. Consequently, we have
hE10 , Ji hE01 , Ji
a10 ' iωµ , a01 ' iωµ (4.3.37)
(k̃10 − k)2k̃10 (k̃01 − k)2k̃01
If we split the modes appropriately, and get k̃10 − k to be 90o out of phase with k̃01 − k. Then
a10 and a01 will be 90o out of phase, and E in (4.3.35) will become circularly polarized.
Ei = 0, Hi = 0 Ef ≠ 0, Hf ≠ 0
Sa Sa
Ei, Hi V Ef, Hf V
ωi ωf
Sw Sw
(a) (b)
Figure 4.10: Derivation of the perturbation formula: (a) Geometry before perturbation. (b)
Geometry after perturbation.
4.3.4 Perturbation Formula for Resonant Frequency Shift

The above derivation for resonant frequency shift due to radiation damping is intuitive. But
a more formal procedure for ascertaining the resonant frequency shift can be obtained by
using perturbation concepts [4]. These concepts have been applied to derive the resonant
frequency shift for microstrip antennas [14, 15]. To start, we take the divergence of the
following quantity:
∇ · (E∗i × Hf + Ef × H∗i )
= Hf · ∇ × E∗i − E∗i · ∇ × Hf + H∗i · ∇ × Ef − Ef · ∇ × H∗i
= −iωi Hf · µ∗ · H∗i + iωf E∗i · · Ef + iωf H∗i · µ · Hf − iωi Ef · ∗ · E∗i
(4.3.38)
Integrating the above over the original volume of the cavity, we arrive at
Z Z
∗ ∗
dS n̂ · (Ei × Hf + Ef × Hi ) = i(ωf − ωi ) dV (H∗i · µ · Hf + E∗i · · Ef ) (4.3.39)
S V
where we have assumed that the medium is lossless, and hence † = and µ† = µ. Since
this is a perturbation, the final fields and the initial fields are similar to each other. Hence,
we can approximate the right-hand side. Consequently,
Z Z
dS n̂ · (E∗i × Hf + Ef × H∗i ) ∼
= i(ωf − ωi ) dV [H∗i · µ · Hi + E∗i · · Ei ] (4.3.40)
S V
The integral on the right-hand side is purely real now, and it can be written in terms of the
initial time-average stored energy hWTi i.
Z Z
∗
dS n̂ · (Ei × Hf ) + dS n̂ · (Ef × H∗i ) = i(ωf − ωi )4hWTi i (4.3.41)
SA SW
In the above, the first integral on the left can be approximated as the complex power radiated
by the hole.
Z
∗ ∼
Prad = dS n̂ · (E∗f × Hf ) = 2hPrad i − iRrad (4.3.42)
SA
where Rrad is the reactive power leaked by the cavity to the outside. The second integral on
the left of (4.3.41) can be related approximately to the wall loss on the interior of the cavity.
Hence, it can be approximated by
Z Z
∼
Pwall = ∗
dS n̂ · (Ef × Hi ) = dS(n̂ × Ef ) · H∗i
SW SW
Z
1
= dS|H∗i |2 (4.3.43)
η SW
where
r r r
µ ωµ ωµ
η= = = (1 − i)
iσ 2σ
Consequently, the wall loss becomes

r Z
ωµ
Pwall = (1 − i) dS|Hi |2 = 2hPwall i − iRwall (4.3.44)
2σ SW
Hence, the resonant frequency shift is given by
2hPrad i + 2hPwall i − iRrad − iRwall

ωf − ωi ≈ −i (4.3.45)
4hWT i i
Therefore,
hPrad + Pwall i
=m(ωf − ωi ) ≈ −i (4.3.46)
2hWT i i
Rrad + Rwall
<e(ωf − ωi ) ≈ − (4.3.47)
4hWT i i
The dissipative loss causes the cavity to have a negative imaginary part of the resonant
frequency giving rise to damping. The reactive power leakage and absorption by the wall
make the cavity appear larger and lower the resonant frequency.
The fringing field effect at the edge of a microstrip patch can also be solved in closed form
using Wiener-Hopf technique [23, 24]. It has been used to ascertain resonant frequency shift
of microstrip antennas in [16].
4.3.5 Variational Impedance Formula for a Current Source
Figure 4.11: Excitation of a microstrip antenna by a current source.
Figure 4.12: Excitation of a general antenna by a current source.
In the previous input impedance formula, the magnetic current Ma is assumed known
and immutable by its environment. The magnetic current is related to the tangential electric
field; hence, it is equivalent to a voltage source in circuit theory. Another formula that is
of importance is when the structure is driven by an immutable applied (impressed) electric
current Ja . Hence, it is equivalent to a current source in circuit theory. The microstrip
antenna driven by such a current source is shown in Figure 4.11 and a general structure case
is shown in Figure 4.12. In this case, a variational formula for the input impedance based on
reaction is given by [4, 19]5
hET , JT i
Zin = − (4.3.48)
I2
where ET = Ea + Ep , and Ea and Ep are the fields produced by applied (or impressed)
current Ja and the induced current Jp , respectively, and JT = Ja + Jp . As shall be shown
later, the above formula can admit approximate solution for Jp with second order error in
the input impedance Zin .
In the exact limit, hET , Jp i = 0 and the above reduces to
hET , Ja + Jp i hET , Ja i
Zin = − =− (4.3.49)
I2 I2
To prove the above impedance formula, we assume that the exciting source is small and hence,
Ja is constant over space. Then
Z Z b
−hET , Ja i = − dV ET · Ja = −I dl · Ea = Vab I = V I = Zin I 2 (4.3.50)
V a
asserting the correctness of the impedance formula in the exact limit.

To prove the variational form of the above formula, we express the above in a quadratic
form. To this end, we have
hEa + Ep , Ja + Jp i hEa , Ja i + hEp , Ja i + hEa , Jp i + hEp , Jp i

Zin = − 2
=−
I I2
hEa , Ja i + 2hEa , Jp i + hEp , Jp i
=−
I2
hEa , Ja i + 2hEa , Jp i + iωµhJp , Ge , Jp i
=− (4.3.51)
I2
It is straightforward to prove that the stationary point of the above expression is at the exact
solution. In the above, the choice of Ge (r, r0 ) is important. It can be a free-space Green’s
function, or Green’s function that satisfies specific boundary conditions. For instance, on
a PEC surface where n̂ × Ge (r, r0 ) = 0, reciprocity implies immediately that an impressed
electric current on such a surface does not radiate, and its contribution can be ignored in the
above calculation.
Another popular formula for input impedance is the power formula where the input
impedance is given by
hET , J∗T i
Zin = − (4.3.52)
|I|2
5 This formula was used for input impedance calculation before its variational nature was known.
The above is based on power conservation, but its variational nature cannot be proved.
However, in the limit when ET is exact, the above reduces only to integration over the
current Ja , and it becomes
hET , J∗a i
Zin = − (4.3.53)
|I|2
If the current source Ja is small and constant phase, the above formula reduces to the varia-
tional formula based on reaction.
The names for these formulas have been rather confusing in the literature. The above
power formula has been called the induced EMF formula in [4], while the reaction formula
has been called the induced EMF formula in [6]. Much of the controversy between the induced
EMF formula and power formula has also been discussed in [6].
...
(O, O) (O, 1) (m, n) all

other
modes
Figure 4.13: Circuit model for the input impedance of a microstrip patch antenna.
Application to Patch Antenna

We will next show how to apply the above result to a microstrip patch antenna. In the exact
limit, as mentioned before,
hET , Ja i
Zin = − (4.3.54)
I2
In the above, ET is the total field in the structure produced by the applied current Ja and
the induced current Jp . For the microstrip patch antenna, we have learned how to find the
total field in the cavity using a magnetic wall model when a current source is inserted into it.
Hence, the total field ET is approximately given by6
X hE∗ , Ja i
i
ET = iωµ Ei (r) (4.3.55)
i
k̃i2 − k 2
6 More elaborate model is given in [2, 12, 20].
In the above, we have assumed that the dyadic Green’s function of the microstrip cavity can
be approximated by
X Ei (r)E∗ (r0 )
i
G(r, r0 ) = 2 − k2
(4.3.56)
i
k̃ i
Hence, this Green’s function automatically accounts for the field generated by the induced
current on the wall of the cavity. This is because we have chosen the eigenmodes to satisfy
the requisite boundary conditions on the wall of the cavity.
Therefore,
X hE∗ , Ja i
i
hET , Ja i = iωµ 2 − k2
hEi , Ja i (4.3.57)
i
k̃ i
Hence, the input impedance, derivable from (4.3.54), is

M
X
Zin = Za + Zi (4.3.58)
i
where Za is higher-order mode contributions, and the second summation comes from the
dominant mode contributions.
iωµ hE∗i , Ja ihEi , Ja i
Zi = − (4.3.59)
I2 k̃i2 − k 2
is the contribution to the input impedance from individual modes of the cavity. Notice that
the above frequency dependence can be fitted with a simple GLC model of a lossy tank
circuit resonator. Hence, for the i-th mode, we can pick Gi , Li , and Ci appropriately to fit
the mathematical formula. A circuit approximation of the microstrip patch antenna hence
can be expressed as in Figure 4.13 [11]. For the microstrip patch, there is a static mode TM00
mode with zero resonant frequency. This is denoted by the lossy capacitor model. Also, the
probe produces a singular field, which can only be constituted by a linear superposition of
many high order modes. Hence, the probe inductance comes from the higher order modes in
the cavity. Notice that in this model, unlike the magnetic frill model, the gap capacitance at
the base of the probe is ignored.
4.4 Aperture Coupling in Waveguide

In addition to using probes to couple energy into a waveguide, a simple way is to drill holes on
the walls of the waveguide, and let energy flow naturally from one waveguide to another. The
simple solution of aperture coupling was first derived by Hans Bethe [2,21,22], who eventually
received a Nobel prize, not for one given contribution, but for his numerous contributions in
physics.
4.4.1 Bethe Coupling

An E field in the vicinity of a waveguide wall, is predominantly normal to the waveguide
wall. If now, an aperture is opened at the waveguide wall, the electric field in the vicinity of
^
n B B
A A
E E
(a) (b)
B B
A
H A H
(c) (d)
Figure 4.14: Fields in the vicinity of an aperture in a waveguide. Case (b) resembles an
electric dipole, while case (d) resembles a magnetic dipole.
the waveguide wall will be as shown in Figure 4.14(b). The field looks like that of a vertical
electric dipole in region B. It has been shown by Bethe [2, 21] that the dipole moment of the
vertical electric dipole is proportional to the normal component of the electric field. For a
circular aperture of radius a0 , it is
2 3
p= a n̂(n̂ · 0 E) = αe n̂(n̂ · 0 E). (4.4.1)
3 0
where αe = 32 a30 , and a0 is the radius of the circular aperture.
x H E Aperture
⊗ ⊗
⊗ ⊗ ⊗ ⊗ ⊗ ⊗
TE10 ⊗⊗⊗ ⊗ ⊗⊗
Mode ⊗ ⊗ ⊗ ⊗ ⊗ ⊗
⊗ ⊗
y z
Figure 4.15: Top view of the field in the neighborhood of an aperture in a rectangular waveg-
uide.
A magnetic field in the vicinity of the waveguide wall, is predominantly tangential. Now,
if an aperture is present, the magnetic field will leak into region B as shown in Figure 4.14(d).
It looks like the field due to a horizontal magnetic dipole in region B. Similarly, the dipole
moment of the horizontal magnetic dipole is [2, 21]
4
m = − a30 Ht == −αm Ht . (4.4.2)
3
where αm = 34 a30 .
H H E E
E or H
Magnetic Dipole Electric Dipole
Figure 4.16: Equivalent sources at the aperture of a waveguide.
Incident θ
Wave
z
Coupled
Wave
Figure 4.17: A directional coupler using two rectangular waveguides, one on top of another.
If a TE10 mode is propagating in a rectangular waveguide, and we have a small aperture

on top of the waveguide, then a time harmonic electric and magnetic dipoles will be generated
at the aperture. In this case, the electric dipole and magnetic dipole are in phase with respect
to each other. For the case of Figure 4.15, the magnetic dipole is pointing in the x direction,
and the electric dipole is pointing in the y direction. The superposition of the vertical electric
dipole and horizontal magnetic dipole gives rise to the cancelation of fields in the +z direction
(see Figure 4.16). Hence, together, they radiate predominantly in the −z direction.7
7 Antennas made by a superposition of an electric dipole and a loop to increase their directivity are known
as Huygens antenna.
Now, if we lay another waveguide on top of the first waveguide, the radiating electric and
magnetic dipoles couple most efficiently into the TE10 mode of the top waveguide if the top
guide is oriented at an angle θ with respect to the bottom guide as shown in Figure 4.17.
This is because a TE10 mode is actually a bouncing plane wave in a rectangular waveguide.
4
3 d
2
1
Figure 4.18: An asymmetrically located aperture can also be used to make a directional
coupler where d 6= a/2.
When the aperture is located at the middle of the waveguide, the magnetic dipole is
pointing in the x direction, requiring the top waveguide to be tilted. We can generate a
dipole pointing away from the x direction by placing the aperture away from the center of the
waveguide, or by using an elliptically shaped aperture. In this case, we can couple efficiently
into the TE10 mode without having to tilt the top waveguide. Hence, a directional coupler can
also be made with θ = 0 if the aperture is not symmetrically located as shown in Figure 4.18
When an aperture is asymmetrically located, the magnetic field that excites it is elliptically
polarized. It can be shown that the exciting field of this aperture due to an incident TE10 in
port 1 of the bottom waveguide is given by

πd
E = ŷEy = ŷE0 sin , (4.4.3)
a

kz πd π πd
H = −E0 x̂ sin + iẑ cos . (4.4.4)
ωµ0 a kz a a
The equivalent electric dipole moment for radiation into the upper guide is

πd
p = ŷ0 αe E0 sin , (4.4.5)
a
The equivalent magnetic dipole moment is

kz πd π πd
m = α m E0 x̂ sin + iẑ cos , (4.4.6)
ωµ0 a kz a a
Excitation of Modes by Electric and Magnetic Current Sources

Since there will be an electric dipole source and a magnetic dipole source induced in a waveg-
uide due to coupling via a hole, it is prudent to study how they would couple to other modes
in a waveguide. We have learned previously that an electric current source excites modes in
a waveguide or cavity as
X hEi , Je i
E(r) = iωµ Ei (r) (4.4.7)
i
ki2 − k 2
We define
1
Hi = ∇ × Ei (4.4.8)
ki
If Ei is normalized, Hi is also normalized. In fact one can easily show that
1
hH∗i , Hi i = h∇ × E∗i , ∇ × Ei i
ki2
1
= 2 hE∗i , ∇ × ∇ × Ei i
ki
= hE∗i , Ei i = 1 (4.4.9)
Also, it can be easily shown that

1
∇ × Hi = ∇ × ∇ × Ei = ki Ei (4.4.10)
ki2
Hence, if we have a vector wave equation given by8
∇ × ∇ × H − k 2 H = iωJm (4.4.11)
the orthonormal eigenmode expansion gives

X hH∗ , Jm i
i
H = iω 2 − k 2 ) Hi (r) (4.4.12)
i
(k i
where Hi is normalized. The corresponding E field, via the use of (4.4.10) and Maxwell’s
equations, is
X hH∗ , Jm i
i
E= ki Ei (4.4.13)
(ki2 − k 2 )
The polarization density can be expressed as

πd
P = ŷαe E0 sin δ(r − x̂d) (4.4.14)
a
to imply that the hole is located at (x, y, z) = (d, 0, 0). The corresponding electric current
density arising from time-varying polarization density is
Jp = −iωP (4.4.15)
8 We will use Jm to denote magnetic current and reserve M to denote magnetization density in this section.
Similarly, the magnetization density is

kz E0 πd π
M = αm x̂ sin + iẑ cos (πda) δ(r − x̂d) (4.4.16)
ωµ a kz a
The corresponding magnetic current density arising from a time-varying magnetization den-
sity is given as
Jm = iωµM (4.4.17)
The corresponding TE10 mode in the upper waveguide is

πx
E10 = ŷE10 sin e±ikz z (4.4.18)
a
where E10 is for normalization. The corresponding H10 field is
E10 h πx π πx i
H10 = ±x̂ikz sin + ẑ cos e±ikz z (4.4.19)
k10 a a a
which is normalized. The ± sign implies ±z propagating waves.
To see if the mode excited by the electric dipole will cancel the one excited by the magnetic
dipole, we need to compare (4.4.7) and (4.4.13). Hence, we need to sum the coefficients
k10 hH?10 , Jm i
and
?
iωµhE10 , Jm i
In details,

πd π2 πd
k10 hH?10 , Jm i = αm ikz E0 E10 ±ikz sin2 +i 2
cos 2
(4.4.20)
a kz a a
Furthermore,
? πd
iωµhE10 , Jm i = ω 2 µαe E0 E10 sin2 ( ) (4.4.21)
a
Hence, the TE10 mode that is excited is proportional to
B± ∝ k10 hH?10 , Jm i + iωµhE?10 , Jp i

πd π 2 πd πd
= E0 E10 ±kz2 αm sin2 + αm cos2 + k 2 αe sin2
a a a a
= 0, z < 0. (4.4.22)
The solution of the above is

d 1 1 λ
= sin−1 √ 0 (4.4.23)
a π 6 a
where λ0 is the free-space wavelength. The above coupler using one hole to achieve directional
coupling is known as the Bethe hole coupler.
When a hole is dug in a waveguide wall, the induced dipoles also give rise to back action
in the original waveguide, altering the field. This in turn, gives rise to an alteration of the
coupled field from the first waveguide to the second waveguide. This effect is taken into
account by Collin [2]. The resulting hierarchy of equations are rather complicated but the
above condition still holds true. This can be thought of as a multiple scattering or coupling
effect, as in the Fabry-Perot etalon. It is the cancelation of the leading order term that is
AAA
AA AAA
AA
important, for the cancelation of the higher-order terms will follow suit.
4
A
AAA
AA AAA
AA B 3
AA
AAA
AAA
AAA
1 2
AAA λg/4
Figure 4.19: A two-hole directional coupler.
Other kinds of directional couplers are also possible. For instance, one can make a two-
hole coupler with the holes spaced λg /4 apart as shown in Figure 4.19. The apertures need
not have a preferred directional coupling. If the aperture coupling is weak, then the wave
reaching the second hole is essentially the same as the wave that reaches the first hole. Hence,
the two different waves that reach B via coupling through the two different holes are in phase
and will interfere constructively. Because of the λg /4 separation of the two holes, the waves
from the two different holes that reach A will be λg /2 or 180o out of phase. Therefore, at A,
the waves interfere destructively, and there is little energy coupled to port 4.
Figure 4.20: Equivalent problem of a hollow waveguide–equivalence principle I.
Since this coupler uses constructive and destructive interferences to enhance its directivity,
the directivity is frequency sensitive. One remedy is to use more holes so as to broaden its
bandwidth, or to use the Schwinger reversed-phase coupler. [2, 25]
Impressed
J Current
J
Jp
Figure 4.21: Equivalent problem of a waveguide with an aperture—nonradiating case.
4.4.2 Equivalence Principles in Aperture Coupling
Figure 4.22: Equivalent problem of a waveguide with an aperture—radiating case–equivalence

principle II.
Equivalence principles have been discussed in [2,4]. In the actual calculation of the electric
and magnetic dipole moments, certain equivalence principles have to be invoked. We shall
discuss them as follows.
There will be two kinds of currents in the following discussion: induced currents and
impressed currents. Induced currents are currents flowing in a conductor due to the presence
of an incident or exciting field next to the conductor. A perfect conductor, for instance,
cannot have a non-zero field in it, and hence, current flows on its surface to prevent the
fields from penetrating it. On the other hand, impressed currents are currents we assume as
sources in Maxwell’s equations. They are currents assumed to exist in free space. They are
the driving source terms in Maxwell’s equations, that are immutable as we seek the solutions.
On the other hand, induced currents follow from the solutions of Maxwell’s equations. They
are due to currents flowing in conductors as we seek solutions to Maxwell’s equations.
Equivalence Principle I
M M
E = 0, H = 0
J
(a) J (b)
Jp
Jp
Figure 4.23: Equivalent problem of a waveguide which is partly covered with a magnetic
wall-nonradiating case.
If we have a source in a metallic waveguide, the source will induce a current J on the inner
surface of the waveguide. The current J is known as the induced current. On the surface of
the waveguide, we have
n̂ × E = 0, n̂ × H = J. (4.4.24)
Now if we remove the metallic wall that supports the induced current and replace the induced
current with an impressed current in vacuum, the field inside the surface S is identical to
before. Furthermore, the field outside S is identically zero. This is because the impressed
current J supports a discontinuity in the magnetic field. Hence, n̂ × H = 0 and n̂ × E = 0 just
outside the impressed current J. By Huygens’ principle, the field must be zero everywhere
outside S. We can check if the solution satisfies all the requisite boundary conditions. If it
does, it is the only unique solution.
The equivalence principle can be proved by three means:
• By performing a Gedanken experiment whereby the conductive material with zero field
inside is been chiseled away until the induced current is replaced by impressed current
in free space. The induced current held the internal field of the cavity in place, and the
impressed current will still hold the internal fields in place.
• By using uniqueness principle argument, when the induced current is replaced by im-
pressed current, the boundary conditions for the fields remain the same. By uniqueness
principle, they must be the same;
• The equivalence principle can also be proved mathematically by the use of Huygens
principle.
Figure 4.24: Equivalent problem of a waveguide with an aperture—radiating case–equivalence

principle III.
Equivalence Principle II
From equivalence principle I, if we have a small aperture in the cavity now, and at the
aperture, the original induced current from Figure 4.20a is impressed, then the field will still
be identically zero outside the waveguide. Because of this, the cases in Figure 4.22a and
Figure 4.22b generate equivalent field outside the waveguide. Note that the current in Figure
4.22b is exactly opposite to that in Figure 4.21. Because of this, the two currents in Figure
4.21 generate zero field outside the waveguide.
Equivalence Principle III
Figure 4.25: Equivalent problem of a magnetic current radiating in the vicinity of an aperture.
By a similar argument, if we have a closed cavity with a magnetic wall over part of the
cavity as in Figure 4.23a, it is completely equivalent to the case of Figure 4.23b, with zero
field outside the waveguide. Consequently, a waveguide with a small aperture is Figure 4.24a
generates equivalent field outside the waveguide as in the case of Figure 4.24b.
Therefore, for radiation due to an aperture in a waveguide, there are two equivalent
problems denoted by Figure 4.22 and Figure 4.24. For the case of Figure 4.22, to find J,
we have to first solve the closed waveguide problem denoted by Figure 4.20a. For the case
(a) (b)
Figure 4.26: Equivalent problem of the E field around a waveguide wall partially covered by
a small magnetic wall.
of Figure 4.24, we have to first solve the problem denoted by Figure 4.23a to find M. It
turns out that equivalence principle III is preferred over equivalence principle II because the
radiation of a magnetic current in a small aperture is easier to calculate than the radiation
of an electric current in a small aperture. For example, if the surface of the waveguide is flat
enough, or that the aperture is small enough, we can replace the problem in Figure 4.25a
with that in Figure 4.25b.
When the PEC surface is flat, a horizontal magnetic dipole radiating in the aperture is the
same as the dipole radiating in free space, as the dipole produces only horizontal magnetic
field that satisfies the boundary condition on the flat PEC surface. So the source is oblivious
of the presence of the flat PEC surface. By image theorem, a horizontal magnetic dipole
radiating in free space is equivalent to one with half its original strength radiating on top of
a PEC ground plane.
(a) (b)
Figure 4.27: Equivalent problem of the H field around a waveguide wall partially covered by
a small magnetic wall.
As mentioned before, in order to find M, we need to solve the closed problem with a
magnetic wall patch as shown in Figure 4.23. If the magnetic wall patch is small enough,
the field around the magnetic wall may be approximated with a static field solution. For
example, if the field is predominantly electric, the field in the vicinity of the magnetic wall
patch (assuming that the wall is reasonably flat) resembles that of Figure 4.26a. Due to
the symmetry of the problem, it is equivalent to that of Figure 4.26b. What happens is
that the electric field induces circulating magnetic current on the magnetic wall patch that
expels the electric field. The circulating magnetic current generates a vertical electric dipole
moment that expels the electric field from the magnetic disk. If the field is predominantly
magnetic, the magnetic field around the magnetic wall patch looks like that in Figure 4.27a,
which is equivalent to that in Figure 4.27b. The horizontal magnetic field induces a horizontal
magnetic dipole moment on the magnetic wall patch.
To obtain the solution for Figure 4.26, we need to solve for the solution of a static electric
field in the vicinity of an ellipsoid. Consider the problem shown in Figure 4.28. When the
relative permittivity, r of the ellipsoid is less than one, the electric field avoids the ellipsoid
by skirting around it. The solution of the static electric field around the ellipsoid can be
obtained in ellipsoidal coordinates. By letting one of the axes of the ellipsoid shrink to zero,
the ellipsoid becomes a disk. Moreover, if we let r = 0, no field can penetrate the disk and
the requisite solution for Figure 4.26 is obtained.
Er < 1
Figure 4.28: Equivalent problem of a static electric field around an ellipsoid. When the
relative permittivity becomes zero, and the ellipsoid becomes a disk, the electric field around
the disk is the same as the electric around a PMC disk.
By the same token, the solution needed for Figure 4.27 can be obtained by studying the
static magnetic field around an ellipsoid with relative permeability µr larger than one. In this
case, the magnetic field in the vicinity of the ellipsoid is attracted t the ellipsoid as shown
in Figure 4.29. When we let the ellipsoid become a disk, and let µr → ∞, we obtain the
requisite solution for Figure 4.29. These closed-form solutions can be used to obtain the
coupling coefficients for small aperture coupling, as in Bethe coupling, in a waveguide.
µr > 1
Figure 4.29: Equivalent problem of a static magnetic field around an ellipsoid. When the
relative permeability becomes infinite, and the ellipsoid becomes a disk, the magnetic field
around the disk is the same as the magnetic around a PMC disk.
Problem 4-1: By making use of reciprocity, prove the validity of Equation (4.1.12a) and
(4.1.12b) of the text.
Problem 4-2: Given the Rayleigh quotient
at · A · a
,
at · B · a
find the solution that will minimize it. What equation does the variational solution solve?
Problem 4-3:
0
z

The dyadic Green’s function of a infinitely long waveguide is given by

X Mei (±khiz , r)Mei (∓khiz , r0 )
0
G(r, r ) = πi 2 A
i
khiz khis hi

0
Nei (±keiz , r)Nei (∓keiz , r ) ẑ ẑ 0 z > z0
+ − δ(r − r ),
2 A
keiz keis ei ko2 z < z0.
Using image theorem, find the dyadic Green’s function of a waveguide with a shorting plane
at z = 0.
Problem 4-4:
(a) For the geometry shown where a waveguide is fed by a coaxial probe flushed with the
waveguide wall, write down the expression for the input admittance using the variational
formula derived in the text assuming that the field at the aperture is that of the coaxial
mode.
(b) Now, using mode-matching method, and assuming only a TEM mode in the coaxial
waveguide, derive an expression for the input admittance using mode-matching method.
Show that this result is the same as that in part (a). (Hint: Do not write out the dyadic
Green’s function explicitly. Leave it as a symbolic operator.)
Problem 4-5:
(a) For a probe in a waveguide excited by a dipole as shown, show that a variational
expression for the input impedance at the base of the probe is [see e.g., Harrington]
iωµ
Zin = − hJp , G, Jp i.
I2
(b) If only the TE10 mode is propagating in a rectangular waveguide, we can approximate
the dyadic Green’s function only with the term associated with the TE10 mode. Assume
that the probe current is of the form Jp = ŷ sin[k(y − d)] so that the probe current is
zero at the tip of the probe. Use the variational formula above to find an approximation
to the probe input impedance with the simplifying approximation on the dyadic Green’s
function. (Note that this approximation is only good for calculating the real part of the
probe impedance. This is because the real part of the probe impedance is related to the
real power radiated to infinity which is carried by the TE10 mode. The inductance of
the probe will not be well approximated by this method because the probe inductance
is associated with the singular field near the probe which can be well approximated only
if we include the higher order evanescent modes in the Green’s function.)
Problem 4-6:
d
2a
z 0
For a TE10 mode propagating in a rectangular waveguide with a small circular aperture of
radius a on top of the waveguide, find the polarization, amplitude and phase of the induced
electric and magnetic dipole moments at the aperture due to the field in the waveguide.
Problem 4-7: Derive the expressions for the equations for Bethe coupling in (4.4.22), and
show that (4.4.23) does solve the equation above it.
Problem 4-8: Establish the equivalence principles I, II, and III using Green’s theorem or
Huygens’ principle for vector electromagnetic fields. In some of the cases, it will be useful to
assume that the dyadic Green’s function satisfies certain boundary conditions to reduce the
size of the surface integral that needs to be performed.
Problem 4-9: Write down the electrostatic solution of the potential due to an ellipsoid in
the presence of a constant electric field. Let one of its axes of the ellipsoid shrink to zero.
What should the permittivity of the ellipsoid be in order to get the solution of Figures 4.26
and 4.27
Bibliography
[4] R.G. Harrington, Time Harmonic Electromagnetic Field, New York: McGraw-Hill, 1961.
[6] E.C. Jordan and K.G. Balmain, Electromagnetic Waves and Radiating Systems, Second
Edition. Prentice-Hall, 1968.
[7] W.C. Chew, Z.P. Nie, Q.H. Liu and Y.T. Lo, “Analysis of probe-fed microstrip disk
antenna,” IEE Proceedings-H, vol. 138, no. 2, pp. 185-191, April 1991.
[8] G.A. Deschamps, “Microstrip Microwave Antennas,” presented at the 3rd USAF Sympo-
sium on Antennas, 1953.
[9] R.E. Munson, “Microstrip Phased Array Antennas, Proc. of Twenty-Second Symp. on
USAF Antenna Research and Development Program, October 1972.
[10] Y.T. Lo, W.F. Richards, “Theory and experiment on microstrip antennas,” IEEE Trans.
Ant. and Propagat., vol. AP-27, PP 137-145, March 1979.
[11] W. F. Richards, Y. T. Lo, and D. D. Harrison, “An improved theory for microstrip
antennas and applications,” IEEE Trans. Antennas Propag., AP-29, pp. 38-46, Jan. 1981.
[12] W. C. Chew, J. A. Kong, and L. C. Shen, “Radiation characteristics of a circular mi-

crostrip antenna,” J. Appl. Phys., 51, 3907, 1980.
[13] D. M. Pozar, “Considerations for millimeter wave printed antennas, IEEE Trans. An-
tennas Propagat., vol. AP-31, pp. 740-747, 1983.
[14] W.C. Chew and J.A. Kong, “Resonance of the axial-symmetric modes in microstrip disk
resonators,” J. Math. Phys., vol. 21, no. 3, pp. 582-591, Mar. 1980.
177
[15] W.C. Chew and J.A. Kong, “Resonance of nonaxial symmetric modes in circular mi-
crostrip disk antenna,” J. Math. Phys., vol. 21, no. 10, pp. 2590-2598, Oct. 1980.
[16] W.C. Chew and J.A. Kong, “Asymptotic formula for the resonant frequencies of a circular
microstrip antenna,” J. Appl. Phys., vol. 52, no. 8, pp. 5365-5369, Aug. 1981.
[17] J.W. Rayleigh, “In finding the correction for the open end of an organ-pipe,” Phil. Trans.,
161, 77, 1870.
[18] W. Ritz, “ber eine neue Methode zur Lsung gewisser Variationsprobleme der mathema-
tischen Physik,” J. Reine Angew. Math., 135, 1-61, 1908.
[19] P.S. Carter, “Circuit relations in radiating systems and applications to antenna prob-
lems,” Proc. IRE, vol. 20, no. 6, pp. 1004-1041, 1932.
[20] K. Kurokawa, “The expansions of electromagnetic. fields in cavities,” IRE Trans. Mi-
crowave Theory Tech., vol. 6 , no. 2, pp. 178-187, 1958.
[21] H.A. Bethe, “Theory of diffraction by small holes,” Phys. Rev., vol. 66, pp. 163-182,
1944.
[22] S. B. Cohn, “Electric polarizability of apertures of arbitrary shape,” Proc. IRE, vol. 40,
pp. 1069-1071, 1952.
[23] N. Wiener, E. hopf, “Ueber eine Klasses singulärer Integralgleichungen,” S.-B. Deutsch.
Akad. Wiss. Berlin KI. Math. Phys. Tech., pp. 696-706 1931.
[24] W. C. Chew and J. A. Kong, “Asymptotic formula for the capacitance of two oppositely
charged discs, Math. Proc. Camb. Philos. Soc., 89, pp. 373-384, 1981.
[25] T.N. Anderson, “Directional coupler design nomograms,” Microwave J., vol. 2, pp. 34-38,
May 1959.
Chapter 5
Discontinuities in Waveguides
Discontinuities in waveguides cannot be avoided. When two waveguides of different sizes

are connected together, the junction does not form a smooth transition introducing a dis-
continuity. Such discontinuities will reflect the waveguide mode. Moreover, infinitely many
modes are needed at the discontinuity in order to match the boundary condition. But most
of the modes “excited” by the presence of discontinuities are evanescent modes. They do not
convect energy away from a junction. Hence, the higher-order modes serve to store energy.
When most of the energy stored is in the magnetic field, the junction discontinuity effect is
inductive, while if most of the stored energy is in the electric field, the effect is capacitive.
Hence, simple equivalent models for the junction can be either a capacitor or an inductor.
However, the calculation of these inductive and capacitive effect requires the use of some tour
de force calculations, a subject that we will discuss in this Chapter.
However, waveguide junctions appear in more complex forms when applied to circulators
and T junctions [1]. The analysis of these waveguide junctions is not amenable to analytic
methods, and hence, much numerical methods have been invoked in analyzing them as is seen
from the reference list in this chapter.
5.1 Transmission Line Equivalence of Waveguide

A wealth of engineering knowledge is built on circuit theory and transmission line theory.
Due to our familiarity with circuit theory and transmission line theory, it is useful to relate
the propagation of modes in a waveguide to transmission line theory [2–4].
For a hollow waveguide of arbitrary shape, it is easy to show that the ratio of the transverse
components of the electric and magnetic fields are
ẑ × Es ωµ0 ẑ × Es kz
ZT E = = , ZT M = = . (5.1.1)
Hs kz Hs ω0
These are called the wave impedances of a waveguide. Notice that they are mode dependent.
If they are to be likened to the characteristic impedance of a transmission line, that it is
natural to define a voltage which is proportional to Es and a current which is proportional
179
to Hs for a transmission line equivalence. However, this proportionality constant should be

chosen so that the time average power given by <e[V I ∗ /2], is the same as the power flowing
down the waveguide.
As an example, a mode propagating down a waveguide in the positive z direction may be
expressed as 1
Es = C+ es e−jkz z (5.1.2a)
Hs = C+ hs e−jkz z (5.1.2b)
where ẑ × es = Zw hs where Zw is wave impedance of the particular mode under discussion.
For a mode propagating in the negative z direction, the fields are given by
Es = C− es e−jkz z , (5.1.3a)
−jkz z
Hs = −C− hs e . (5.1.3b)
The expression for the field can be replaced by equivalent voltage and current waves given by
V = V+ e−jkz z + V− ejkz z , (5.1.4a)
I = I+ e−jkz z − I− ejkz z , (5.1.4b)
where V+ = K1 C+ , V− = K1 C− , I+ = K2 C+ , and I− = K2 C− .
To ensure that the equivalent circuit carries the same power, we require that
Z
1 ∗ |C+ |2
V+ I+ = (es × h∗s ) · ẑdS (5.1.5)
2 2 s
or that Z
K1 K2∗ = (es × h∗s ) · ẑdS. (5.1.6)
s
The characteristic impedance of this equivalent transmission line is
Zc = V+ /I+ = K1 /K2 . (5.1.7)
Notice that there exists no unique way of choosing K1 and K2 . Therefore, es and hs in
(5.1.6) can be normalized such that the right hand side of (5.1.6) is 1. In this case, we require
K1 K2∗ = 1. Also, the characteristic impedance of the transmission line equivalence in (5.1.7)
can be made equal to 1 since all Smith charts are given for normalized impedance values.
Alternatively, one can choose K1 /K2 = Zw .
Armed with the transmission line model of a waveguide, much of the tools that are found
in transmission line theory like Smith chart, and impedance matching techniques can be used.
However, connecting two waveguides of different wave impedances and hence sizes gives rise
to waveguide discontinuities. These discontinuities scatters a propagating mode into higher
modes which are evanescent. Hence, they give rise to localized stored energy in their vicinity.
Depending on if the stored energy is of electric or magnetic type, these discontinuities are
often modelled by shunt capacitances and inductances which can be calculated. At other
times, one may deliberately introduce discontinuities in a waveguide call diaphragms to give
rise to a shunt capacitance or inductance for the purpose of matching to the load or designing
filters.
1 We shall use ejωt time convention for agreement with the time convention of circuit theory.
Discontinuities in Waveguides 181
(a)
Waveguide
Junction
(b) Inductive
J
Diaphragm
(c) +++++++
––––––– Q Capacitive
Diaphragm
Figure 5.1: Different kinds of waveguide junctions.
5.2 Waveguide Junction
Waveguide discontinuities have been studied by a number of workers [5,6]. Before the advent
of digital computers, analytic and variational methods were used to study these discontinu-
ities. With the advent of high speed digital computers, these discontinuities are routinely
studied with numerical method, requiring the solution of a large system of linear algebraic
equation (see reference list).
Here, we propose to characterize the scattering by a waveguide discontinuity by reflection

and transmission operators using mode matching. Then, the generalization to an arbitrary
number of discontinuities (see Figure 5.3) becomes routine. The waveguide discontinuity prob-
lem has also been studied previously, but a more generalized formulation, valid for arbitrarily-
shaped waveguides, is presented here. The formulation is then related to the integral equation
formulation, and a new criterion for convergence of the method is given and corroborated by
numerical simulation.
Waveguide discontinuities arise in many occasions, for example, at a waveguide junction

(Figure 5.1(a)), or when diaphragms are added to change the phase and amplitude of a wave
propagating through a waveguide. A diaphragm can be of the inductive type as shown in
Figure 1(b), or of the capacitive type as shown in Figure 1(c). An inductive diaphragm
induces stored magnetic energy while a capacitive diaphragm induces stored electric energy.
Moreover, a capacitive diaphragm changes the phase of a mode differently from an inductive
diaphragm. Hence, they can be used as tuning elements for matching purposes in waveguides.
Sa 2
1
z
z=0
Figure 5.2: Mode matching at a general waveguide junction.
4
1 2 3 6 8 9
5 7
Figure 5.3: A waveguide with many junction discontinuities.
5.2.1 Mode Matching–Eigenmode Expansion Method

Waveguide discontinuity problems are generally solved by the method of mode matching.2
In this method, the waves in waveguide 1 and waveguide 2 (see Figure 5.2) are expanded in
terms of the modes of the waveguide. The amplitudes of the modes are found by matching
the boundary conditions at the discontinuity. The field structure is quite complicated at the
waveguide junction as the field has to bend to adjust to the boundary conditions. A multitude
of modes is needed for matching the boundary condition. However, most of these modes are
evanescent giving rise to localized field at the junction discontinuity. The local field can store
electric field energy giving rise to a capacitive effect. If it stores magnetic field energy, it gives
rise to an inductive effect.
Another important physics that happens at a waveguide junction is that if there is a pure
waveguide mode impinging on the junction, due to the need to match boundary conditions, all
higher order modes are excited at the junction giving rise to infinitely many reflected modes
and transmitted modes. This is the physics of mode conversion.
In order to facilitate the ease for mode matching, we first develop a succinct and compact
notation that can encompass all modes in a waveguide. If we have a superposition of modes
in an arbitrarily-shaped, hollow waveguide, the z components of the fields can be written as
X
Hz = Hi ψhi (rs )eikhiz z , (TE modes) (5.2.1)
i
2 This section follows the development in [20].
X
Ez = Ei ψei (rs )eikeiz z . (TM modes) (5.2.2)
i
Here, ψhi (rs ) and ψei (rs ) are solutions to the equation (∇2s + kis
2
)ψi (rs ) = 0 with Neumann
∂2 ∂2
and Dirichlet boundary conditions, respectively, on the waveguide wall. ∇2s = ∂x 2 + ∂y 2 , and
2
kiz is the eigenvalue corresponding to the eigenmode ψi (rs ). The subscripts h and e are used
to denote the Neumann problem (TE) and the Dirichlet problem (TM), respectively. For
each mode, the transverse components of the fields can be found via the equations derivable
from Maxwell’s equations similar to those in Chapter 3:
1 ∂ iωµ
eis = 2 ∇s eiz − 2 ẑ × ∇s hiz , (5.2.3)
keis ∂z khis
1 ∂ iω
his = 2 ∇s hiz + 2 ẑ × ∇s eiz , (5.2.4)
khis ∂z keis
where eiz and hiz are the z components of the electric field and magnetic field respectively for
each individual mode. Consequently, eis and his are the transverse components of the electric
2
field and magnetic field respectively for each individual mode. Furthermore, keis = k 2 − keiz
2
2 2 2
and khis = k − khiz .
Using the above, we deduce that
X
Es = eis
i
X Ei iωµHi
(5.2.5)
ikeiz z ikhiz z
= 2 ikeiz ∇ s ψ ei (rs )e − 2 ẑ × ∇ s ψhi (rs )e ,
i
keis khis
X
Hs = his
i
X iωEi Hi
(5.2.6)
ikeiz z ikhiz z
= 2 ẑ × ∇s ψei (rs )e + 2 ikhiz ∇s ψhi (rs )e .
i
keis khis
We see that ẑ × Hs is closely related to Es , i.e.,

X iωEi Hi

ẑ × Hs = − 2 ∇s ψei (rs )eikeiz z + 2 ikhiz ẑ × ∇s ψhi (rs )eikhiz z . (5.2.7)
i
keis khis
Using vector notation, we can write Es more compactly as

" ∇s ψei (rs )
#t
X 2
keis eikeiz z 0 ikeiz Ei
Es = ẑ×∇s ψhi (rs )
2
0 eikhiz z −iωµHi
i khis
X (5.2.8a)
t
= ψ i (rs ) · e|ik{ziz z} · ei ,
| {z } |{z}
i 2×2 2×1
2×2
where ik z
2
∇s ψei (rs )/keis ikiz z e eiz 0
ψ i (rs ) = 2 , e = , (5.2.8b)
ẑ × ∇s ψhi (rs )/khis 0 eikhiz z
and
ikeiz Ei
ei = . (5.2.8c)
−iωµHi
Similarly, we can write
X t
ẑ × Hs = − ψ i (rs ) · gi · e|ik{ziz z} · ei , (5.2.9)
|{z} |{z} |{z}
i 2×2 2×1
2×2 2×2
where
ω/keiz 0
gi = .
0 khiz/ωµ
Using the fact that one can write
X
an λn bn = at · λ · b, (5.2.10a)
n
where
at = [a1 , a2 , a3 , . . .], bt = [b1 , b2 , b3 , . . .], (5.2.10b)
 
λ1
 λ2 
 
λ= λ3 , (5.2.10c)
 
..
.
we can rewrite (5.2.8a) and (5.2.9) compactly as
t
eiKz · |{z}
Es = Ψ (rs ) · |{z} e , (5.2.11a)
| {z }
2×∞ ∞×∞ ∞×1
t iKz
ẑ × Hs = − Ψ (rs ) · |{z}
G · e|{z} · |{z}
e , (5.2.11b)
| {z }
2×∞ ∞×∞ ∞×∞ ∞×1
where h t i
t t t
Ψ (rs ) = ψ 1 (rs ), ψ 2 (rs ), ψ 3 (rs ), . . . , (5.2.12a)
 
eik1z z
 
 eik2z z 
e iKz
= , (5.2.12b)
e ik3z z 
 
..
.
 
g1
 g2 
 
G= g3 , (5.2.12c)
 
..
.
et = [e1 , e2 , e3 , . . .]. (5.2.12d)

The above order of the elements of the matrices may be rearranged for bookkeeping
purposes. For instance, one may prefer to group the TE modes and the TM modes separately
rather than as a couplet in the above. In this case, for example,
 TM 
g1
 g2T M 
 
 . .. 
 
G=  . (5.2.13a)
g TE 
 1 
 T
g2 E 
 
..
.
where,
giT M = ω/keiz , giT E = khiz /ωµ. (5.2.13b)
Similar rearrangement need be done for (5.2.12a), (5.2.12b) and (5.2.12d) if this is in fact
preferred.
With the compact way to write the transverse components of the fields as in (5.2.11a),
we can write down with physical intuition the general solution for waveguides 1 and 2. In
waveguide 1, there will be an incident as well as a reflected wave. Hence, we have
t

E1s = Ψ1 (rs ) · eiK1 z · e + e−iK1 z · eR , (5.2.14a)
t

−ẑ × H1s = Ψ1 (rs ) · G1 · eiK1 z · e − e−iK1 z · eR . (5.2.14b)
In waveguide 2, we can write the fields as

t
E2s = Ψ2 (rs ) · eiK2 z · eT , (5.2.14c)
t
−ẑ × H2s = Ψ2 (rs ) · G2 · eiK2 z · eT . (5.2.14d)
We can define R and T which are reflection and transmission operators such that
eR = R · e, eT = T · e. (5.2.15)
These operators entail the physics of mode conversion at a waveguide junction. They are
infinite dimensional matrices, or operators. Then
t

E1s = Ψ1 (rs ) · eiK1 z + e−iK1 z · R · e, (5.2.16a)
t

−ẑ × H1s = Ψ1 (rs ) · G1 · eiK1 z − e−iK1 z · R · e, (5.2.16b)
and
t
E2s = Ψ2 (rs ) · eiK2 z · T · e, (5.2.17a)
t
−ẑ × H2s = Ψ2 (rs ) · G2 · eiK2 z · T · e. (5.2.17b)
From the continuity of the tangential electric field, we have

t t
Ψ1 (rs ) · I + R · e = Ψ2 (rs ) · T · e, rs ∈ Sa , (5.2.18)
Furthermore, we require that tangential electric fields equal zero for rs ∈ / Sa at the plane of
the discontinuity.
The continuity of the magnetic field implies
t t
Ψ1 (rs ) · G1 · I − R · e = Ψ2 (rs ) · G2 · T · e, rs ∈ Sa . (5.2.19)
The unknowns to be sought here are R and T.

t
The aperture Sa can be thought of as the cross-section of a waveguide with modes Ψa (rs ).
We can then equate (5.2.18) to
t t t
Ψ (rs ) · I + R · e = Ψ (rs ) · T · e = Ψ (rs ) · a , rs ∈ Sa , (5.2.20)
| 1{z } | {z } |{z} | 2{z } |{z} |{z} | a{z } |{z}
2×∞ ∞×1 2×∞ ∞×∞ ∞×1 ∞×N N ×1
∞×∞
t
where Ψa (rs ) = 0, rs ∈/ Sa . In this way, (5.2.18) plus the auxiliary condition after (5.2.18) are
satisfied. For practical purposes, we choose a to be a vector of length N even though in theory,
the summation should be infinite. This is obviated by the integral equation formulation in the
next section. Hence, the dot product for the last term in (5.2.20) implies N -term summation
while the rest of the dot products imply infinite summation. From this point onward, we
shall denote inner products with infinite summations with double-dot products, and leave
the single-dot product for inner product with a finite summation. Note that in general, the
length of the vector e is determined by the number of incident modes. We shall assume that
e is of infinite length, but the following formulation is also valid for e of finite length. It is
to be noted that in (5.2.20), any complete set of basis functions instead of waveguide modes
that can be used to expand the aperture field for rs ∈ Sa will also suffice.
Multiplying equation (5.2.20) by Ψ1 (rs ) and integrate over rs , we get

D1 : I + R : e = L1a · |{z} a , (5.2.21a)
|{z}
∞×N N ×1
where we define D E D E
t t
D i = Ψi , Ψi , Lia = Ψi , Ψa . (5.2.21b)
Di is diagonal due to mode orthogonality while Lia is in general non-square, infinite dimen-
sional by N dimensional matrix. Di can be made into an identity matrix if the modes are
made orthonormal. Similarly, multiplying (5.2.20) by Ψ2 (rs ) and integrating over rs , we have
D2 : T : e = L2a · a. (5.2.22)
From (5.2.21a) and (5.2.22), we deduce that

−1
R : e = D1 : L1a · a − e, (5.2.23a)
−1
T : e = D2 : L2a · a. (5.2.23b)
−1
As it is diagonal, Di is easily found even though it is infinite dimensional. Substituting
(5.2.23a) into (5.2.19), and rearranging terms, we have
t t −1
2Ψ1 (rs ) : G1 : e =Ψ1 (rs ) : G1 : D1 : L1a · a
t −1
(5.2.24)
+ Ψ2 (rs ) : G2 : D2 : L2a · a, rs ∈Sa .
Remember that the above is derived from the continuity of the magnetic field which needs to
be imposed only on Sa . Hence, we should weight the above equation with functions whose
support is over Sa . Weighting the above equation by Ψa (rs ) where Ψa (rs ) is a vector of
length N , we have, after using the definition for Lia given in (5.2.21b),
t
t −1 t −1

2L1a : G1 : e = L1a : G1 : D1 : L1a + L2a : G2 : D2 : L2a · a. (5.2.25)
t −1
Note that in general, Lia is nonsquare, but Lia : Gi : Di : Lia is an N × N square matrix.
Now, we can solve (5.2.25) (which is unlike (5.2.24)) for a giving
t −1 t −1
−1 t
a = L1a : G1 : D1 : L1a + L2a : G2 : D2 : L2a · 2 L1a : G1 : e. (5.2.26)
Substituting (5.2.26) in (5.2.23a), we can solve for R and T giving

−1
t −1
R = D1 : L1a · L1a : G1 : D1 : L1a
−1 (5.2.27a)
t −1 t
+L2a : G2 : D2 : L2a · 2L1a : G1 − I,
−1
t −1
T = D2 : L2a · L1a : G1 : D1 : L1a
−1 (5.2.27b)
t −1 t
+L2a : G2 : D2 : L2a · 2L1a : G1 .
The above expressions simplify further with the following assumptions. If Sa is as large
as the smaller waveguide, then ψ a (rs ) = ψ 1 (rs ), Lia = Li1 and L11 = D1 . Consequently, the
above simplifies to
t −1
−1
R = D1 · G1 + L21 : G2 : D2 : L21 · 2D1 · G1 − I
t −1
−1
= D1 · G1 + L21 : G2 : D2 : L21 (5.2.28a)
t −1

· D1 · G1 − L21 : G2 : D2 : L21 ,
−1
t −1
−1
T = D2 : L21 · D1 · G1 + L21 : G2 : D2 : L21 · 2D1 · G1 . (5.2.28b)
This is the case for the absence of the diaphragm. Note that R is now N ×N and T is ∞×N .
This follows from that the basis functions used at the aperture is orthogonal to the higher
order modes in waveguide 1, and hence, would not excite them.
Einc 2
Ma –Ma
1
Figure 5.4: Equivalent problem for a waveguide discontinuity.
If waveguides 1 and 2 are identical, but a diaphragm is present, (5.2.27a) simplifies to

−1
t −1
−1 t
R = D1 : L1a · L1a : G1 : D1 : L1a · L1a : G1 − I, (5.2.29a)
−1
t −1
−1 t
T = D1 : L1a · L1a : G1 : D1 : L1a · L1a : G1 . (5.2.29b)
In this case, I + R = T. Here, R and T are again ∞ × ∞ matrices.

In general, R and T are nondiagonal matrices. Physically, this means that a mode coming
into the discontinuity will excite many modes, giving rise to numerous reflected and trans-
mitted modes. As mentioned before, this entails the physics of mode conversion. However,
if in both waveguides, only the dominant modes are propagating, e.g., the TE10 mode in a
rectangular waveguide, the excited higher order modes will be evanescent, and hence, local-
ized around the discontinuity. These evanescent fields store electromagnetic energy, making
the discontinuity either capacitive or inductive depending on if the energy is stored in the
electric field or magnetic field.
Even though the double-dot products in the above should involve infinite summations, a
practical implementation necessitates the truncation of the infinite summations. If the infinite
summation is truncated at P terms, as shall be shown in the next section, P N to ensure
the accuracy of the double-dot products (see also Problem 5-2).
If only a few elements of R and T are needed, to save on computer memory, the matrix
Lia , which is P × N , need not be calculated and stored. Explicit expressions can usually be
t −1
derived for the elements of Lia : Gi : Di : Lia , which is N × N , so that only these matrices
need be stored. Then, the corresponding elements of Lia are calculated to form the desired
elements of R and T.
Note that the above formulation is general enough that it encompasses the case of even if
the waveguides are of different shapes.
5.2.2 Equivalence Principle and Integral Equation Formulation

The above formulation shows a derivation of the reflection and transmission operators of a
waveguide discontinuity using a mode matching procedure. Their calculation involves ∞ × N
dimensional matrices. From the above equations, we see that the reflection and transmission
operators could be calculated even if we do not truncate the infinite dimension of the matrices.
The matrix that needs to be inverted is always N × N . However, to save computation

resources, it is necessary to truncate the infinite summations at some finite values. Many
authors have suggested using P1 modes only in waveguide 1 and P2 modes only in waveguide
2 in the mode matching procedure [5]. Consequently, the resultant matrices in the above
formulation, which are ∞ × N , become P1 × N or P2 × N . A rule is often given for the ratio
between P1 , P2 and N for a convergent result. This issue is generally regarded as relative
convergence. We shall give a new rule for the choice of P1 and P2 for this problem.
Alternatively, the problem on the choice of P1 , P2 , and N can be further enlightened by
looking at an equivalent formulation of the above problem. This equivalent formulation can
be achieved with the use of equivalence principle first, and later, deriving an integral equation
for the problem.
In the equivalent problem, the waveguide discontinuity problem is divided into two waveg-
uides with shorting planes at the location of the discontinuity (see Figure 5.4). Moreover,
magnetic current Ma = E × n̂ are impressed at the location of the aperture. By the equiva-
lence principle, if E × n̂ is known at the aperture, the reflected and the transmitted fields are
the same in the original problem and the equivalent problem. This equivalence problem can
be easily derived using vector Green’s theorem and vector Huygens principle.
If the magnetic dyadic Green’s functions are known for the shorted waveguide 1 and
waveguide 2, then the magnetic fields in waveguide 1 and waveguide 2 are
Z
H1 (r) = Hinc (r) + iω1 dS 0 G1m (r, r0 ) · Ma (r0 ), (5.2.30a)
Sa
Z
H2 (r) = −iω2 dS 0 G2m (r, r0 ) · Ma (r0 ). (5.2.30b)
Sa
Such a dyadic Green’s function can be derived using the method outlined in the previous
chapter, which expands the Green’s function in terms of an infinite number of waveguide
modes. Since tangential magnetic field is continuous across the aperture, we have
Z
n̂ × Hinc (r) = iωn̂ × dS 0 [2 G2m (r, r0 ) − 1 G1m (r, r0 )] · Ma (r0 ), r ∈ Sa . (5.2.31)
Sa
The above integral equation can be solved with the method of moments or Galerkin’s method,
and converted into an N × N matrix equation, by using N expansion functions and N testing
functions. In this procedure, we let
N
X
Ma (r0 ) = an fn (r0 ) (5.2.32)
n=1
By testing the above with −n̂ × fm (r), we obtain

∞
X
hfm , Hinc i = iω hfm , 2 G2m − 1 G1m , fn i, m = 1, . . . , N (5.2.33)
n=1
The above is a matrix equation of the form
b=A·a (5.2.34)
where,
[b]m = hfm , Hinc i, [A]mn = iωhfm , 2 G2m − 1 G1m , fn i, [a]n = an (5.2.35)
and
Z Z
hfm , G, fn i = dS fm (r) · dS 0 G(r, r0 ) · fn (r0 ) (5.2.36)
Sa Sa
Z
hfm , Hinc i = dS fm (r) · Hinc ) (5.2.37)
Sa
As has been seen, the method of moments yields the optimal solution when N basis
functions are used to solve (5.2.31). Equation (5.2.31) is an exact integral equation, and it is
clear that the numbers of waveguide modes needed to approximate G1m and G2m should be
infinite to render them as accurate as possible. This is equivalent to choosing P1 and P2 to
be infinite in the previous formulation using mode matching.
5.2.3 Relative Convergence

If Sa is made as large as waveguide 1, as is the case discussed in Equation (5.2.28a) in the
previous section, and the electric field of N modes from waveguide 1 is used to approximate
Ma via E1 × n̂, then the integral
Z
iω1 dS 0 G1m (r, r0 ) · Ma (r0 ) (5.2.38)
Sa
would only yield N waveguide modes for the reflected field in waveguide 1 due to mode
orthogonality. Hence, P1 = N in this case. What then is the number of modes required in
the expansion of dyadic Green’s function in the following integral?
Z
iω2 dS 0 G2m (r, r0 ) · Ma (r0 ). (5.2.39)
Sa
This integral yields the transmitted field in waveguide 2. Since mode orthogonality does not
apply here, in theory, an infinite number of modes is needed in G2m (r, r0 ). But in practice,
we need only a finite number of modes P2 . How large should P2 be to accurately represent
the field generated by (5.2.39)?
Further insight can be obtained by studying a simpler problem—the parallel plate waveg-
uide problem. The problem is scalar in this case and a typical integral for (5.2.39) looks
like Z d2 " ∞ #
X eikmz |z| mπx mπx0
0
I= dx cos cos Ma (x0 ), (5.2.40a)
0 m=0
kmz d2 d 2
where we have expanded the scalar Green’s function as
X∞ 0
eikmz |z−z | mπx mπx0
G2m (r, r0 ) ∝ cos cos . (5.2.40b)
m=0
kmz d2 d2
r 2
mπ
Here, kmz = k22 − d2 and d2 is the separation of the parallel plate waveguide 2. Ma (x0 ),
the aperture field, following the above method, is expanded in terms of the modes of waveguide
1, i.e.,
XN
0 nπx0
Ma (x ) = an cos , (5.2.41)
n=0
d1
where d1 is the separation of parallel plate waveguide 1. Hence, (5.2.40a), after using (5.2.41),
letting z = z 0 , and exchanging the order of summation and integration, becomes,
∞
X X
N Z d2
1 mπx mπx0 nπx0
I= cos an dx0 cos cos . (5.2.42)
m=0
kmx d2 n=0 0 d2 d1
The above integral could be integrated readily to yield

N
X ∞
X
1 mπx
I= an cos fnm , (5.2.43a)
n=0 m=0
kmz d2
where  h i h i
 mπ nπ mπ nπ
1 sin d2 + d1 d2 sin d2 − d1 d2 
fnm = mπ nπ + mπ nπ . (5.2.43b)
2 d2 + d1 d2 − d1

In the above, the integral I is the result of the action of the Green’s function on the expansion
or basis functions at the aperture. The integral in (5.2.42) is the cosine transform of the basis
functions in
waveguide
1 using the cosine functions of waveguide
2. 0In
other words, the basis
0
function cos nπxd1 is expanded in terms of the functions cos mπx
d2 . The coefficient of the
mπ nπ
expansion peaks when d2 = d1 , or when the wavelengths of the two Fourier harmonics match
or about equal.
When the integral (5.2.39) is used in (5.2.31) to match boundary condition, r ∈ Sa , or
z = 0 in
(5.2.40a)
and (5.2.43a) when used in (5.2.31). If we further test (or weight) (5.2.43a)
n0 πx
by cos d1 and integrate over x, as is required when the method of Galerkin is used to
solve (5.2.31), then (5.2.43a) can be transformed to (with z = 0)
N
X ∞
X 1
A= an fnm fn0 m
n=0 m=0
kmz
(5.2.44)
N
X P2
X
∼ 1 0
= an fnm fn0 m , n = 0, 1, · · · , N.
n=0 m=0
kmz
³ ´
c o s 2πx
d1
1
(a) 0
−1
d1 d2 x
³ ´
c o s 4πx
d2
1
(b) 0
−1
d2 x
P
X µ ¶
mπx
am cos P = 150
m=0
d2
1
(c) 0
−1
d2 x
P
X µ ¶
mπx
am cos P = 1000
m=0
d2
1
(d) 0
−1
d2 x
Figure 5.5: Convergence of the cosine Fourier series expansion of the mode of one waveguide
with dimesion d1 in terms of the modes of the second waveguide with dimension d2 .
0.9
0.8
Normalized Root Mean Squared Error
N=2 N=8 N=16 N=24 N=32 N=40 N=50 N=64

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 20 40 60 80 100 120 140 160
Number Of Modes (P)

Figure 5.6: Error convergence of the cosine Fourier series expansion cos Ndπx1
=
P
P
am cos mπx
d2 as a function of P with d2 = 2d1 . Notice that as P > N d2 /d1 , the
m=0
error drops rapidly.
The above is the matrix representation of the Green’s function using the eigenmodes from
waveguide 1. Note that in the above, fnm peaks at m = dd21 n, and fn0 m peaks at m = dd12 n0 .
Hence, in order to evaluate A accurately in (5.2.44), the summation over m in (5.2.44) must
at least be large enough so that the contribution from the peaks of fnm fn0 m are included.
Since the largest n and n0 are N , for the approximate summation in (5.2.44) to be accurate
when the infinite summation in (5.2.44) is replaced by a summation over m from 0 to P2 , we
require that
P2 N
. (5.2.45)
d2 d1
From the above analysis, it is clear that
(a) if the dimension of the diaphragm region is da such that da < d1 , da < d2 , and
(b) if N basis functions are used to approximate the aperture field, and moreover,
(c) if truncated numbers of modes, P1 and P2 are used to represent the fields in waveguides
1 and 2, respectively,
then in order for this truncation to be accurate,
P1 N P2 N
, . (5.2.46)
d1 da d2 da
This point has also been noted by Orta et al3 in the study of frequency selective surfaces. Note
that P1 and P2 need not be related by a specific ratio, but if the same degree of accuracies is
required of the fields in both waveguides, then
P1 P2
' . (5.2.47)
d1 d2
The inequalities in (5.2.46) can also be interpreted as that P1 and P2 should be chosen
large enough so that the spectral components used in G1m and G2m are large enough to
capture (or accurately represent) the dominant spectral components in Ma .
The above analysis can be extended to the rectangular waveguide case. Since the x and
y coordinates of a rectangular waveguides are separable, the x-spectral components can be
considered separately from the y-spectral components. Similar inequalities as in (40) will hold
separately for the x and y spectral components of the waveguide. For the case of a circular
waveguide which is axially symmetric, a similar analysis will yield the inequality as in (5.2.46)
but d1 , d2 , and da represent the diameter of the waveguides and aperture. This is because
Bessel functions behave like sinusoidal functions when their arguments are large.
It is to be noted that other basis functions can be used to expand the aperture field in
(5.2.41) other than the waveguide modes. In this case, fnm will not be of the form given by
(5.2.43a), but usually, a more complex form ensues. In this case P1 and P2 should be chosen
large enough to capture the dominant spectral components in Ma . The rule for choosing P1
and P2 will not be as simple as that given by (5.2.46). But for a fixed N , one should increase
P1 and P2 so that they are large enough until the calculated amplitudes of the reflected and
transmitted modes stabilize.
3 R. Orta, R. Tascone, and R. Zich, “Multiple dielectric loaded perforated screens as frequency selective
surfaces,” IEE Proc., vol. 135, pt. H, no. 2, pp. 75-82, 1988.
(a) (b)
Figure 5.7: Plots showing the convergence of the reflection coefficient amplitude of TE10
mode of a rectangular waveguide when (5.2.46) is satisfied. The dimension of the waveguide
is 0.8 × 0.6 wavelength. (a) The inductive diaphragm case. The dimension of the diaphragm
is 0.6 × 0.6 wavelength. (b) The capacitive diaphragm case. The dimension of the diaphragm
is 0.8 × 0.4 wavelength. In these cases, P1 = P2 = P (re-plotted by F. Ling).
5.3 Numerical Examples

A program has been developed using the formulation of Section 5.2. This program yields
the reflection and transmission operators due to a junction discontinuity with a diaphragm
of two rectangular waveguides. This program is used to study the convergence of the TE10
mode reflection coefficient when the two waveguides are identical but separated by either an
inductive diaphragm or a capacitive diaphragm. The dimension of the waveguide is 0.8 × 0.6
wavelength. The apertures are symmetrically located with dimension 0.6 × 0.6 wavelength
for the inductive diaphragm, and 0.8 × 0.4 wavelength for the capacitive diaphragm.
A TE10 mode is assumed incident onto this junction discontinuity. For the inductive
diaphragm, due to symmetry, only modes with x-variations will be excited. Figure 5.7(a)
shows the amplitude of the reflection coefficient as a function of the number of waveguide
modes when the number of basis function N in the aperture is kept fixed. It is seen that for
a fixed N , the solution converges when (5.2.46) is satisfied. It is also seen that N has to be
sufficiently
large before the reflection coefficient amplitude is accurate. Note that in this
case,
P dda need not be very much larger than N for the calculation to stabilize. P dda has to
be only a little larger than N to capture the dominant components of Ma .
Figure 5.7(b) shows a similar convergence plot for the case of a capacitive diaphragm.
Due to symmetry, only higher order modes with y-variation will be excited. Qualitatively,
the result is similar to that of Figure 5.7(a).
Note from Figures 5.7(a) and 5.7(b) that if the conventional rule of choosing Pd11 = dNa = Pd22
z=0 z=h
1 2 3
Figure 5.8: A two-waveguide junction problem.
is used, fairly good result could be obtained when N is large enough. This is an entirely an
artifact of choosing waveguide modes to expand the field at the aperture. The conventional
wisdom of choosing P1 and P2 for this case will capture most of the spectral components
(the aperture basis functions are expanded in terms of a Fourier series using the waveguide
modes, and we call the amplitude of a term of this series a spectral component) this series of
the basis functions at the aperture. If basis functions with edge conditions, for instance, are
used, high spectral components will ensue and the conventional wisdom will not apply.
As a final note, it can be shown that the mode matching method always conserves energy
irrespective of how many basis functions are used at the aperture. Thus the conservation of
energy can be used to check the correctness of the implementation of the method, but not the
accuracy. As seen in the previous simulation, an accurate solution can only be obtained by
slowly increasing N while ensuring that (5.2.46) is satisfied. An accurate solution is obtained
when the reflection and transmission coefficients cease to change with increasing N .
5.4 Solution to the Multiple Waveguide Junction Prob-

lem
Once the solution to the one waveguide junction problem is known, a two waveguide junction
problem can be solved. Then, the N -waveguide junction problem can be sought recursively
in the manner of [45]. The N -waveguide junction solution can be used to model complex
waveguide junctions. It can also be used to model tapered waveguide junctions.
5.4.1 A Two-Waveguide-Junction Problem

Given a two-waveguide-junction geometry as shown, we can write down the solution in waveg-
uide 1 for the electric field as consisting of

t e
E1s = Ψ1 (rs ) · eiK1 z + e−iK1 z · R 12 · e (5.4.1)
where e is the amplitude of the incident modes, and R e is the reflection operator including
multiple reflections between junctions 1 and 2. The first term consists of modes travelling to
the right while the second term consists of modes travelling to the left.
In waveguide 2, we can write the solution for the electric field as

t

E2s = Ψ2 (rs ) · eiK2 z · A2 + e−iK2 z · B2 . (5.4.2)
In waveguide 3, the corresponding electric field is

t
E3s = Ψ3 (rs ) · eiK3 z · A3 . (5.4.3)
A relationship can be easily established between the amplitudes A2 and B2 in waveguide 2,
because the left-going wave in waveguide 2 is a consequence of the right-going wave in the
same waveguide. Therefore, at z = h, we must have
e−iK2 h · B2 = R23 · eiK2 h · A2 , (5.4.4)

where R23 is the reflection operator for the single waveguide junction. Consequently, we can
rewrite (5.4.2) as
t

E2s = Ψ2 (rs ) · eiK2 z + e−iK2 (z−h) · R23 · eiK2 h · A2 . (5.4.5)
The right-going wave in waveguide 2 is a result of the transmission of right-going wave in

waveguide 1 plus a reflection of the left-going wave in waveguide 2. Therefore, at z = 0, we
must have
A2 = T12 · e + R21 · B2 . (5.4.6)
Using (5.4.4) in (5.4.6), we can solve for A2 to yield
−1
A2 = I − R21 · eiK2 h · R23 · eiK2 h · T12 · e (5.4.7)
Using (5.4.7) in (5.4.4), we can express B2 in terms of e. Therefore, A2 and B2 are found in
terms of e.
The right-going wave in waveguide 3 is a consequence of a transmission of the right-going
wave in waveguide 2. Therefore, at z = h, we must have
eiK3 h · A3 = T23 · eik2 h · A2 . (5.4.8)

Consequently, A3 can be found in terms of e via the use of (5.4.7) and (5.4.8).
The left-going wave in waveguide 1 is a consequence of the reflection of the right-going
wave in waveguide 1 plus a transmission of the left-going wave in waveguide 2. Therefore, we
can write
Re ·e=R ·e+T ·B . (5.4.9)
12 12 21 2
Using B2 from (5.4.4) and (5.4.7), we have

e = R + T · eiK2 h · R · eiK2 h
R 12 12 21 23
−1 (5.4.10)
· I − R21 · eiK2 h · R23 · eiK2 h · T12 .
Therefore, with e, the amplitude of the incident modes, known, one can find the solution of
the two waveguide junction problem in all the three waveguides. We call R e the generalized
12
reflection operator at the (1,2) junction that accounts for multiple reflections.
∼
∼
∼
∼
1 2 3 4 j–1 j j+1 N–1 N
∼
∼
∼
∼
z=h1 z=h2 z=h3 z=hj–1 z=hj z=hN–1
Figure 5.9: A multiple-junction waveguide.
5.4.2 An N-Waveguide-Junction Problem

Given an N -waveguide-junction problem where the discontinuities of the waveguide are at
z = hj , j = 1, · · · , N , one can write down the solution in region j as

t e
Ejs = Ψj (rs ) · eiKj z + e−iKj (z−hj ) · R iKj hj
j,j+1 · e · Aj . (5.4.11)
The amplitude of the left-going wave in the above is written in such a way so that at z = hj ,
the left-going wave amplitude is just related to the right-going wave amplitude by Re
j,j+1 , the
generalized reflection operator at the (j, j + 1) junction. Motivated by (5.4.10), we can write
down the expression for the generalized reflection operator as
e iKj (hj −hj−1 ) e

· Rj+1,j+2 · eiKj (hj −hj−1 )
R j,j+1 = Rj,j+1 + Tj+1,j · e
−1 (5.4.12)
e
· I − Rj,j+1 · eiKj (hj −hj−1 ) · R iKj (hj −hj−1 )
j+1,j+2 · e · Tj,j+1 .
Notice that we have replaced R23 in (5.4.10) with a generalized reflection operator R e
j+1,j+2
at the (j + 1, j + 2) junction because for a multiple-junction waveguide, multiple reflections
to the right of the (j, j + 1) junction has to be accounted for.
Using (5.4.12), starting at the right-most junction, the generalized reflection operators in
all the waveguides can be found. Next, Using an equation similar to (5.4.7), the amplitude
Aj in waveguide j can be related to Aj−1 in waveguide j − 1 as
−1
e
eiKj hj−1 · Aj = I − Rj,j−1 · eiKj (hj −hj−1 ) · R iKj (hj −hj−1 )
j,j+1 · e
(5.4.13)
· Tj−1,j · eiKj−1 hj−1 · Aj−1 .
Starting with the left-most waveguide, the field solution for all the waveguides can be obtained.
Figure 5.10: A multiple-junction waveguide used to design a filter.
Figure 5.11: The pass-band response of the filter in Figure 5.10 with six resonators. The
waveguide dimensions are 13.7 in by 6.22 in. The symmetric irises are 0.02 in thick. The
widths of apertures are 1.026, 0.958, 0.898, and 0.870 in respectively from one end to the
center. The cavity lengths are 0.626, 0.653, and 0.674 in respectively from one end to the
center (courtesy of F. Ling).
5.4.3 Filter Design–A Resonance Tunneling Problem

Multiple junctions in a waveguide can be used to design filters. Junctions set up interference
in a waveguide, giving rise to resonant modes. The resonant mode of a multiple junction
waveguide is defined as one that a reflected mode can exist without the presence of the
incident mode. This is the same as requiring that

det Re
j,j+1 = ∞ (5.4.14)
The above is equivalent to

e
det I − Rj,j+1 · eiKj (hj −hj−1 ) · R iKj (hj −hj−1 )
j+1,j+2 · e =0 (5.4.15)
The above is the generalized transverse resonance condition. One can vary the frequency ω
until the above equation is satisfied. It yields the poles of the system as is the case of the
Fabry-Perot etalon. The locations of the poles can be used to guide the design of filters as
shown in Figures 5.10 and 5.11.
5.5 Hybrid Junctions

A hybrid-T junction is a nifty device that when a TE10 mode is incident from port 1, the
wave will couple to a mode in ports 2 and 3 but not 4. Similarly, a mode incident from port
4 will couple to only ports 2 and 3 but not 1. A mode incident from port 2 couples to ports
1 and 4 but not 3, while a mode incident from port 3 couples to ports 1 and 4 but not 2.
The understanding of the working of the hybrid-T, sometimes known as a magic-T, can
be derived from symmetry arguments. For a TE10 mode incident at port 1, the E field
is symmetrical about the plane that bisects the center of the waveguide of port 1. From
the figure, the vertical component of the electric field is even symmetric about the plane of
symmetry, while the horizontal component is odd symmetric. Hence, a TE10 mode cannot
be excited in port 4.
For a TE10 mode incident from port 4, the horizontal component of the electric field is
even symmetric while the vertical component is odd symmetric about the plane of symmetry.
Hence, the modes excited in ports 2 and 3 are of opposite polarity. Furthermore, no TE10
mode can be coupled to port 1 because of the symmetry condition.
When ports 2 and 3 are terminated in matched loads so that only outgoing waves exist
in them, ports 1 and 4 can be matched by adding matching elements so that S11 and S44
are zero. In this case, only incoming wave exists in port 1 or port 4 when ports 2 and 3
are matched. The remaining property of the magic-T can then be explained by time-reversal
symmetry.
The magic-T is a linear and lossless device. By linearly superposing a mode incident at
port 1 and another mode incident at port 4, one can obtain a mode exiting at port 3 but
not a mode exiting at port 2. Hence, this linear superposition is a solution to Maxwell’s
equations. However, the time-reversed solution is also a solution to Maxwell’s equations. The
time-reversed solution corresponds to having ports 1 and 4 matched since there will only
be outgoing waves at these ports. Furthermore, port 3 is matched since only incoming wave
2 3
1
(a)
4 4
Plane of Plane of
Symmetry Symmetry
2 3 2 3
(b) (c)
Figure 5.12: (a) A hybrid-T junction. (b) Electric field pattern in port 2 and port 3 with
a TE10 mode incident from port 1. (c) Electric field pattern in ports 2 and 3 with a TE10
mode incident from port 4.
AA
AA AA
AA
AA
AA
AAAA
4
AA
AA
AAAA
4
AAAA AAAA
2
AAAA Magic-T 3 2
AAAA
Magic-T 3
AAAA AAAA
Junction Junction
(a) (b)
1 1
Figure 5.13: (a) Normal-time operation of a magic-T. (b) Reversed-time operation of a magic-
T.
exists at this port, and magically, there is no coupling from port 3 to port 2! Similar argument
leads to a matched port 2 with matched loads at ports 1 and 4, and no coupling from port 2
to port 3.
Another useful hybrid junction is the rat-race ring circuit. When a mode is incident in
port 1, it will split evenly in two directions into two waves. Due to the choice in the size
of the two ring, the waves will arrive out-of-phase at port 3. The clockwise travelling wave
will suffer a perturbation at port 2 while the counterclockwise travelling wave will suffer a
perturbation at port 4. Therefore, their amplitudes at port 3 are equal and out of phase,
and there is no coupling to port 3. The phases of these two waves are equal at ports 2 and
4 even though their amplitudes may not be equal because the clockwise travelling wave and
counterclockwise travelling wave are perturbed by different amount. Therefore, there will be
coupling to ports 2 and 4.
By the same argument, port 2 will not couple to port 4. Port 3 will not couple to port 1
and port 4 will not couple to port 2 by reciprocity.
Hybrid junctions is used to make microwave impedance bridges, frequency discriminator
circuits as in balanced mixers, circulators as well as other applications.
5.6 Periodic Structures

A periodic structure is useful as a filter or a slow-wave structure. The constructive and de-
structive interference phenomenon of a periodic structure can generate passbands and stop-
bands: A passband is a frequency band that allows the propagation of a wave with no
attenuation while a stopband forbids the propagation of a wave. A periodic structure can be
analyzed by the Bloch-Floquet theorem [2, 46, 47].
Due to the periodicity of this structure, a solution E(r) when translated by a distance d in
AAA
AAA AAA
AAA
AAA AAA
2 3
AAAAAAAA
AAA
AA
AAA
AAA
AAAA a = λg/4
AAAA
1
AAAA
AA
AAA
AAAA
AAAAAAA
AAAA
AAAAAA
AAAA
a
a
a
4
AAAA
AAA 3a
Figure 5.14: A rat-race ring-type hybrid junction.
Figure 5.15: A periodic structure in a parallel plate waveguide.

the z direction, must be itself with an added phase, viz.
E(r + ẑnd) = E(r)eikz nd , H(r + ẑnd) = H(r)eikz nd . (5.6.1)
This implies that

E(r) = eikz z Ep (r), (5.6.2)
H(r) = eikz z Hp (r), (5.6.3)
where Ep (r) and Hp (r) are periodic functions in z having the property
Ep (r + ẑnd) = Ep (r), (5.6.4)
Hp (r + ẑnd) = Hp (r), (5.6.5)

where n is any positive or negative integer. The expression of a field in (5.6.2) and (5.6.3) is
known as the Floquet theorem or the Bloch theorem.
Any periodic function can be expanded in terms of a Fourier series. Therefore, we can
write
X∞
2nπ
Ep (r) = Epn (rs )ei d z , (5.6.6)
n=−∞
∞
X 2nπ
Hp (r) = Hpn (rs )ei d z , (5.6.7)
n=−∞
where rs = x̂x + ŷy. Consequently, we can rewrite (5.6.2) and (5.6.3) as

∞
X ∞
X
2nπ
E(r) = Epn (rs )ei(kz + d )z = Epn (rs )eiknz z , (5.6.8)
n=−∞ n=−∞
∞
X ∞
X
2nπ
H(r) = Hpn (rs )ei(kz + d )z = Hpn (rs )eiknz z , (5.6.9)
n=−∞ n=−∞
where knz = kz + 2nπ

d . Each of the terms above is termed a Floquet mode, with a wavenumber
knz . However, each Floquet mode cannot exist by itself–they have to exist together as a
cluster. The phase velocity of the n-th Floquet mode is
ω ω
vpn = = , (5.6.10)
knz kz + 2nπ/d
while the group velocity is given to be

−1
dω dknz
vgn = = = vg , (5.6.11)
dknz dω
The group velocity is independent of the harmonics. Since vpn can be negative for some
negative n, vpn can be opposite in sign to vg .
Let us assume that a TM mode is propagating in the parallel plate waveguide such that
H = x̂Hx . There will be no depolarization since ∂/∂x = 0, and this field component is
parallel to the surfaces involved. Consequently, we have
∞
X
Hx = hn (y)eiknz z , (5.6.12)
n=−∞
Since (∇2 + k0 2 )Hx = 0, we have
d2 hn (y) 2
+ kny hn (y) = 0, (5.6.13)
dy 2
2
where kny = k02 − knz
2
.
For y > h, the coefficient hn (y) can be expanded as
hn (y) = an cos [kny (y − b)], (5.6.14)
Since the Neumann boundary condition n̂ · ∇Hx = 0 on a metallic surface, we have chosen a
cosine in (5.6.14) so that this condition is satisfied.
Inside the corrugation, we can expand Hx in terms of the modes of a shorted parallel
waveguide, or
X∞ mπz
Hx = gm (y) cos , (5.6.15)
m=0
w
By the same token, gm (y) satisfies
d2 gm (y) 2
+ γmy gm (y) = 0, (5.6.16)
dy 2
2

mπ 2
where γmy = k02 − w . Therefore, we derive that
gm (y) = bm cos(γmy y), (5.6.17)
Consequently, we can write

∞
X
Hx = an cos[kny (y − b)]eiknz z , h < y < b, (5.6.18)
n=−∞
∞
X mπz
Hx = bm cos(γmy y) cos , 0 < y < h. (5.6.19)
m=0
w
The bottom equation is valid for 0 < z < w, and it replicates itself with the correct phase
shift for different slots. To find the guidance condition for nontrivial solutions to an and bn ,
we match boundary condition across the y = h interface. At this interface, the tangential
component of the magnetic field is continuous, so is the tangential component of the electric
∂Hx
field. The latter is the same as ∂y and is continuous across this interface. Therefore, we
arrive at
∞
X ∞
X mπz
an cos[kny (h − b)]eiknz z = bm cos γmy h cos , 0<z<w (5.6.20)
n=−∞ m=0
w
∞
X
an kny sin[kny (h − b)]eiknz z
n=−∞
 ∞ mπz
 X (5.6.21)
 bm γmy sin (γmy h) cos , 0 < z < w,
= m=0
w


0 w<z<d
Equation (5.6.20) is a cosine series expansion for 0 < z < w. We can solve for bm by
multiplying (5.6.20) by cos( mπz
w ) and integrate from 0 to w to obtain
w X∞ D mπz E
bm cos(γmy h) (1 + δom ) = an cos[kny (h − b)] eiknz z , cos , (5.6.22)
2 n=−∞
w
where D mπz E Z w mπz

eiknz z , cos = eiknz z cos dz, m = 0, ..., (5.6.23)
w 0 w
and can be evaluated in closed form if needed.
In (5.6.21), by writing
2nπ
eiknz z = eikz z+i d z , (5.6.24)
it is apparently a Fourier series expansion. We can find the coefficients an as
∞
X D mπz E
an kny sin[kny (h − b)]d = bm γmy sin(γmy h) cos , e−iknz z , (5.6.25)
m=0
w
where D mπz E Z w mπz

cos , e−iknz z = cos e−iknz z dz, (5.6.26)
w 0 w
and n varies from −∞ to +∞. Equations (5.6.22) and (5.6.25) entail two infinite system
of equations. Since γmy and kny become large imaginary numbers when m and n are large,
cos(γmy h), sin(γmy h), cos[kny (h − b)], and sin[kny (h − b)] become exponentially large. In this
case, we can define
b̂m = bm cos(γmy h), (5.6.27)
ân = an cos[kny (h − b)], (5.6.28)
to rewrite (5.6.22) and (5.6.25) as
w X∞ D mπz E
b̂m (1 + δom ) = ân eiknz z , cos , (5.6.29)
2 n=−∞
w
∞
X D mπz E
ân kny tan[kny (h − b)]d = b̂m γmy tan(γmy h) cos , e−iknz z , (5.6.30)
m=0
w
To solve the above, we need to truncate the infinite system. We can rewrite (5.6.29) and
(5.6.30) as
XN
b̂m λm = ân Amn , m = 0, ..., M, (5.6.31)
n=−N
M
X
ân βn = b̂m Bnm , n = −N, ..., N. (5.6.32)
m=0
The above can be written as matrix equations
λ · b̂ = A · â, β · â = B · b̂, (5.6.33)
where λ and β are diagonal matrices, A is a (M +1)×(2N +1) matrix, B is a (2N +1)×(M +1)
matrix, a is a length (2N + 1) vector while b is a length (M + 1) vector. Equation (5.6.33)
can be rewritten as
−1 −1
b̂ − λ · A · β · B · b̂ = 0. (5.6.34)
Nontrivial solution for b will exist if
−1 −1
det(I − λ ·A·β · B) = 0. (5.6.35)
The above is the guidance condition from which one can solve for k given kz . It will be found
that k does not always exist for all values for a given kz . Also, it could be that for a certain
window of k, no value exists for all kz . Those are the stop band of the periodic structure.
5.6.1 Floquet Modes and Brillouin Zone

From the periodic structure theory, we notice that a wave propagating in a periodic waveguide
is of the form
φ(r) = eikz z φp (r), (5.6.36)
where φp (r) is a periodic function in z with φp (r + ẑnd) = φp (r), and
∞
X 2nπ
φp (r) = φpn (rs )ei d z . (5.6.37)
n=−∞
The physical picture is that this wave is a cluster of Floquet modes propagating in unison
through the waveguide in order to satisfy the boundary condition on the waveguide wall. In
other words, all the Floquet modes of the form exp(i(kz + 2nπ/d)) move in lock step with
respect to each other. We can express (5.6.36) more explicitly in terms of the Floquet modes,
namely,
X∞
φpn (rs , kz )ei(kz + d )z .
2nπ
φ(r) = (5.6.38)
n=−∞
66
55
44
kd
kd
33
22
11
00
-2−2π −1.5 −π
−1 −0.5 0
0
k d
0.5
π
1 1.5
2 π2
z
kz d
Figure 5.16: Dispersion (k − kz ) diagram of a corrugated parallel plate waveguide showing

bandpass and bandstop sections (w = d/2, b = d, h = 0.25b) (courtesy of L.J. Jiang).
2mπ
Now, if we let kz → kz + d , where m is an integer, the above becomes
∞
X
2mπ 2(n+m)π
φ(r) = φpn rs , kz + ei[kz + d ]z
n=−∞
d
∞ (5.6.39)
X 2mπ 0

i k + 2n π z
= φp,n0 −m rs , kz + e z d .
0
d
n =−∞
It is clear that if (5.6.38) is an eigensolution that can satisfy the boundary condition on
the waveguide wall, (5.6.39) can be easily made to satisfy the boundary condition on the
waveguide wall if
2mπ
φp,n −m rs , kz +
0 = cφpn0 (rs , kz )
d
where c is a multiplicative constant. Therefore, the cluster of Floquet modes is invariant
with respect to the transform kz → kz + 2mπ d . Consequently, the dispersion diagram for a
guided mode in a periodic structure is 2π d periodic in kz as shown in Figure 5.16. Each of
this periodic zone is called the Brillouin zone.
5.7 Stop Band and Coupled-Mode theory

The existence of pass bands and stop bands in a periodic waveguide is due to the coupling,
constructive and destructive interference of the Floquet modes. To see how the interfering
kd
kz d
Figure 5.17: Dispersion (k − kz ) diagram of a parallel plate waveguide without periodic

perturbation.
kd
1 2
– 4π –2π 2π 4π kz d
Figure 5.18: Dispersion (k − kz ) diagram of a parallel plate waveguide with weak or no

periodic perturbation. The Floquet modes are generated, but they remain weakly coupled to
each other.
Floquet modes can yield a stop band when they are coupled, we use the coupled modes
theory [48].
To do this analysis, we think of the corrugation in a waveguide as a perturbation. Before
the corrugation perturbations are introduced, assume that we have only the forward and
backward propagating TEM modes within the parallel-plate waveguide. The dispersion dia-
gram is as shown in Figure 5.17. The moment a small periodic perturbation is introduced,
higher order Floquet modes emerge in the waveguide. If the perturbation is weak enough,
little coupling occurs between the forward Floquet modes and the backward Floquet modes.
The forward and backward Floquet modes carry energy in the forward and backward direc-
tions respectively due to the sign of their group velocities. The dispersion curve in the case
of vanishingly small periodic perturbation is shown in Figure 5.18.
We can focus our attention on modes 1 and 2 and study the coupling behavior between
them. The couple-mode equation for describing these two contra-propagating modes can be
written as
dφ1
= ik1 φ1 + iα12 φ2 , (5.7.1)
dz
dφ2
= ik2 φ2 + iα21 φ1 , (5.7.2)
dz
αij is the coupling coefficient describing the coupling of energy between the two modes. If
αij = 0, coupling ceases to exist, and we have kz = k1 for mode 1, and kz = k2 for mode 2
as expected. More specifically, k1 = k, and k2 = −k + 2π d . We can rewrite (5.7.1) and (5.7.2)
using matrix notation as
d
φ = i K · φ, (5.7.3)
dz
where
φ1 k1 α12
φ= , K= . (5.7.4)
φ2 α21 k2
But letting the solution
φ = eikz z a0 , (5.7.5)
we have from (5.7.3) that
kz a 0 = K · a 0 , (5.7.6)
or that a0 is the eigenvector of K, and then kz is its eigenvalue given by

det kz I − K = 0. (5.7.7)
Equation (5.7.7) yields

(k1 − kz )(k2 − kz ) − α12 α21 = 0. (5.7.8)
Solving the above for kz , we have
(k1 + k2 ) 1 p
kz = ± (k1 − k2 )2 + α12 α21 . (5.7.9)
2 2
From energy conservation, we rewrite that
d
|φ1 |2 − |φ2 |2 = 0, (5.7.10)
dz
since two modes propagate in the opposite directions, and hence, their energy flow cancel
each other. Equation (5.7.10) is the same as
d †
φ · S · φ = 0, (5.7.11)
dz

1 0
where S = . From (5.7.11), and using (5.7.3), we have
0 −1

d † d † dφ
φ ·S·φ= φ · S · φ + φ† · S ·
dz dz dz
†
= −iφ† · K · S · φ + iφ† · S · K · φ (5.7.12)
†

= iφ† · S · K − K · S · φ = 0.
∗
In order for the above to be zero, it is required that α12 = −α21 . (If the two modes are
co-propagating, carrying energy the same direction, the condition for energy conservation is
∗
α12 = α21 .)
I n–1 In I n+1
jX jX jX
Vn–1 Vn Vn+1
Figure 5.19: Circuit equivalence of a periodic structure.
I1 d I2
V1 V2
Figure 5.20: A section of a transmission line.
For mode 1, k1 = k, and for mode 2, k2 = −k + 2πd . From (5.7.9), we have that
r
π π 2
kz = ± k− − |α12 |2 . (5.7.13)
d d
But in the dispersion diagram, we usually fix kz and solve for k. Alternatively, we can invert
Equation (5.7.13) to obtain
r
π π 2
k= ± kz − − |α12 |2 (5.7.14)
d d
In the vicinity of kz = πd , k = πd ± |α12 |. In other words, k splits into two distinct values
about πd due to forward and backward wave coupling.
5.7.1 Circuit Analysis of Periodic Structure

The above numerical analysis solves the problem exactly within numerical approximation.
However, it offers little insight to the problem. As an approximation to the problem, we can
think of the corrugated parallel plate waveguide problem as being a transmission line problem
with shorted stubs connected in series.
The structure in Figure 5.19 is best analyzed by the chain matrix or the ABCD trans-
mission matrix. It can be shown that the chain matrix connecting the voltages of a section
of transmission line of length d is

V1 cos(kd) jZc sin(kd) V2
= . (5.7.15)
I1 jYc sin(kd) cos(kd) I2
For a series reactance, the corresponding chain matrix is

V1 1 jX V2
= . (5.7.16)
I1 0 1 I2
Therefore, the chain matrix connecting Vn , In to Vn+1 , In+1 is

Vn cos(kd) jZc sin(kd) 1 j Vn+1
= . (5.7.17)
In jYc sin(kd) cos(kd) 0 1 In+1
But if a wave propagates on a periodic structure, then
Vn = Vn+1 ejkz d , In = In+1 ejkz d . (5.7.18)

jk d
cos(kd) jX cos(kd) + jZc sin(kd) e z 0 Vn+1
− = 0. (5.7.19)
jYc sin(kd) −XYc sin(kd) + cos(kd) 0 ejkz d In+1
The above will have a nontrivial solution only if

A − ejkz d B
det = 0, (5.7.20)
C D − ejkz d
The above is equivalent to
AD − BC − (A + D)ejkz d + e2jkz d = 0, (5.7.21)
Since AD − BC = 1 for a reciprocal network, we have

A+D
cos(kz d) = , (5.7.22)
2
The above gives the guidance condition for the wave number of a wave propagating on a
periodic structure. For the circuit of a series reactance loaded periodic structure, this becomes
XYc
cos(kz d) = cos(kd) − sin(kd), (5.7.23)
2
If X is due to a shorted stub, then X = Zs tan(kh), and we have
Zs Yc
cos(kz d) = cos(kd) − tan(kh) sin(kd), (5.7.24)
2
Figure 5.21 plots the right-hand side of (5.7.24) which is f (kd) as a function of kd, and
the left-hand side of (5.7.24) which is just cos(kz d). For every kz d, we can read off several
values of kd such that cos(kz d) = f (kd). Hence we arrive at the following diagram for kd
versus kz d.
As can be seen from Figure 5.22, there are frequency bands at which real values of kz d
could exist. These are the passbands. In Figure 5.22, we have only shown the first two bands.
The band of frequencies at which no real values of kz d can exist is the stopband. Because of
this property, a periodic structure can be used as a filter. Also, notice that the phase velocity
is smaller than that of just a TEM mode propagating in a parallel plate waveguide, hence,
the name slow-wave structure for a periodic waveguide. It is also possible to have the group
velocity opposite in sign to the phase velocity as can be seen from the above diagram. This
is used in a backward wave oscillator in microwave circuits.
f(kd) cos(kzd)
(kd)2 (kd)1 k zd
kd π/2 π 3π/2 kzd
-1
Figure 5.21: Plot of cos(kz d) versus kz d and f (kd) versus kd.
kd
-2π -π π 2π kzd
Figure 5.22: The kd − kz d dispersion diagram of a periodically loaded transmission line.

5.8 Metamaterial
The area of metamaterial has been inspired by the suggestion of Vesalago [49] that if we have
both negative and µ negative material (also called a double-negative or DNG material), then
the wave will be a backward wave with the phase velocity traveling in the opposite direction
to the group velocity. It was further suggested by Pendry [50] that a DNG material time
reverses a field, and hence, can be used to make a perfect lens (or superlens). As a result
there have been a flurry of activities in this field.
Unfortunately, many of the superlensing effect of DNG material disappears with the slight-
est amount of loss or imperfection. However, the excitement in search of the holy grail has
inspired many new ideas and structures that could be of interest. We will briefly review the
physics of DNG materials.
We can start with Maxwell’s equations for regular materials:
∂H
∇×E = −µ
∂t
∂E
∇×H = +J
∂t
∇ · E = ρ
∇ · µH = 0 (5.8.1)
If we were to change the sign of and µ in Equations (5.8.1), we arrive at:
∂H
∇×E = µ
∂t
∂E
∇ × H = − +J
∂t
∇ · E = −ρ
∇ · µH = 0 (5.8.2)
Notice that once the solution to (5.8.1) is obtained, we can obtain the solution to (5.8.2) by
changing the signs of H, J, and ρ. Hence power flow, which is defined to be E × H, points
in the opposite direction for solutions of Equations (5.8.2) compared to that of solutions of
Equations (5.8.1). Also, the E, H and k vectors form the left-hand rule (see Figure 5.23).
Hence, a DNG medium is also called a left-handed medium (LHM) as opposed to the regular
right-handed medium (RHM).
We can easily prove that LHM has to be frequency dispersive by reductio ad absurdum [53].
Assuming that the medium is frequency independent or non-dispersive. Then a conservation
law for power flow can be easily derived for a source-free region to be:
1 ∂
∇ · (E × H) = − µ|H|2 + |E|2
2 ∂t
or in integral form via the use of Gauss’ divergence theorem:
I Z
1 ∂
dS · (E × H) = − dV µ|H|2 + |E|2
S 2 ∂t V
Figure 5.23: An RHM plane wave in a regular medium versus an LHM plane wave that
propagates in a double negative medium.
In the above, when LHM and a non-dispersive RHM, such as vacuum, co-exist and mix with
each other, and µ can have different signs for different regions. Hence, with a proper choice
of surface and volume, the right-hand side of the above can be zero, while the left-hand side
is non-zero violating energy conservation! However, if the medium is dispersive, as we have
learned from Chapter 1, the above is not the correct expression for energy storage. Therefore,
double negative materials can only be made over a narrow bandwidth when it co-exists with
regular materials. Moreover, we still have to use E × H as the direction of power flow if
the DNG has been made from regular materials. Consequently, the k vector of a plane wave
points in opposite direction to that of power flow.
As a result, the group velocity of a DNG material has to be opposite to that of the phase
velocity. At a simple interface, due to phase matching and energy conservation requirements,
negative refraction occurs as shown in Figure 5.24. Phase matching requires the k vector to
be aligned as shown, but energy conservation requires that the group velocity be opposite to
that of the phase velocity. Because of negative refraction, it has proposed the use of negative
refraction for focusing [49].
5.8.1 Evanescent Amplification by a Matched DNG Slab

A matched DNG slab has the capability of amplifying evanescent wave, as shall be shown.
This has been suggested to use a DNG slab as a super lens [50].
The transmission coefficient, as we have seen from the section on Fabry-Perot etalon is
Figure 5.24: Negative refraction happens at an interface because of the need for phase match-
ing and energy conservation.
given by
T12 T21 eik2z d 2
(1 − R21 )eik2z d
T̃ = 2 2ik d
= 2 (5.8.3)
1 − R21 e 2z 1 − R21 e2ik2z d
with
TE µj kiz − µi kjz TM j kiz − i kjz
Rij = , Rij = , (5.8.4)
µj kiz + µi kjz j kiz + i kjz
depending on whether we are calculating T̃ for a TE wave or a TM wave. Also, Tij = 1 + Rij
in the above, and we have assumed that the z axis is normal to the slab.
For a matched medium, we pick µ2 = −µ1 and 2 = −1 . For propagating wave, we pick
the phase velocity in region 2 to be opposite to that of regions 1 and 3. In other words,
k2z = −k1z . Then it is quite clear that R12 = 0, and
T̃ = eik1z h (5.8.5)
In fact, it can be shown that the above result is independent of what sign we pick for k2z .
For the evanescent spectrum it can be shown that, irrespective of the branch of square
root for choose for k2z , the transmission coefficient is always
T̃ = eα1z h (5.8.6)
p
where α1z = k12 − kx2 where kx is the wavenumber parallel to the slab. Hence, the evanescent
wave has been amplified after it has passed through the DNG slab. This can be used as a
super lens for super resolution phenomena [50].
To understand this, we first look at the Weyl [15] identity
ZZ
∞
eik0 r i eikx x+iky y+ikz |z|
= dkx dky ,
r 2π kz
−∞
Figure 5.25: Super lens or super resolution effect of a matched DNG slab.
where kx2 + ky2 + kz2 = k02 , or kz = (k02 − kx2 − ky2 )1/2 . The above says that the field produced
by a point source consists of both evanescent spectrum as well as the propagating spectrum.
However, it is the evanescent spectrum that contains the high-resolution information of the
point source. As we move away from the point source, the evanescent spectrum becomes
smaller, and therefore, the high-resolution information is lost. However, if this evanescent
spectrum can be reconstituted by using the DNG slab, then the high-resolution information
can be regained. This is illustrated in Figure 5.25. However, this reconstitution of the
evanescent spectrum is a highly unstable undertaking and is easily upset by loss or geometry
imperfection [52, 53].
5.8.2 Composite Right-Left Handed Transmission Line

One interesting technology that has emerged in this area is the composite right-left handed
(CRLH) transmission line [54, 55]. We will analyze this transmission line next using circuit
theory. For the case shown in Figure 5.26, we can write down Kirchoff voltage law to get
V (z + ∆z) − V (z) = −I(z)Z∆z (5.8.7)
I(z + ∆z) − I(z) = −V (z + ∆z)Y ∆z (5.8.8)

where Z and Y are per unit length impedance and admittance. The above becomes the
Telegraphers equations when ∆z → 0
∂V (z)
= −ZI(z) (5.8.9)
∂z
∂I(z)
= −Y V (z) (5.8.10)
∂z
or
∂ 2 V (z)
= ZY V (z) (5.8.11)
∂z 2
I ( z ), V ( z ) I ( z  z ), V ( z  z )
Z
Figure 5.26: Lumped element model for deriving the CRLH transmission line equations.
∂ 2 I(z)
= ZY V (z) (5.8.12)
∂z 2
or
β 2 = −Y Z (5.8.13)
√
β = ±j Y Z (5.8.14)
If
1
Z= (5.8.15)
jωCL
1
Y = (5.8.16)
jωLL
then above imitates a double negative material. We can pick one branch of the square root
to get
1
β=− √ <0 (5.8.17)
ω LL CL
r r
Z LL
Z0 = = >0 (5.8.18)
Y CL
ω p
vph = = −ω 2 LL CL < 0 (5.8.19)
β
dω ω p
vg = = − = ω 2 LL CL > 0 (5.8.20)
dβ k
ZR ZL
LR
CL
YR YL
CR LL
Figure 5.27: The lumped element model for a CRLH transmission line.
The above indicates that the group velocity is opposite to the phase velocity: it supports a
backward wave.
For a CRLH transmission line, we have

1
Z = j ωLR − (5.8.21)
ωCL

1
Y = j ωCR − (5.8.22)
ωLL
Then if ω > ωse = √ 1 , ω > ωsh = √ 1
LR CL LL CR
s
1 1
β= ωLR − ωCR − >0 (5.8.23)
ωCL ωLL
If ω < ωse , ω < ωsh , then

s
1 1
β=− − ωLR − ωCR (5.8.24)
ωCL ωLL
and β < 0.
A band gap exists for min (ωse , ωsh ) < ω < max (ωse , ωsh ). If we make ωse = ωsh , the
band gap disappears.
At ω = ω0 > 0, there exists a wave with β = 0, or a constant phase wave. At this frequency,
the series resonance becomes a short and the shunt resonance becomes an open. This idea
  RH
 PRH
 PLH
max  se ,  sh 
 LH min  se ,  sh 

O
Figure 5.28: The ω-β diagrams of the pure RH line, the pure LH line, and that of the CRLH
line. For CRLH, and band gap exists where the frequency ω has no solution for β or a
propagating mode.
0

Figure 5.29: The choice of LR , LL , CR , and CL where ωse = ωsh , and the band gap disappears.
can be used to design equi-phase loop antenna for RFID applications. When β < ω/c, the
wave that propagates on the CRLH line leaks energy to the space around it: the structure
can be used to make leaky-wave antennas.
TEm0
a2 a1
0
z
b
b
Problem 5-1: A waveguide junction is formed from two rectangular waveguides, one of
which is 2 cm by 4 cm, and the other is 2.5 cm by 4 cm. Assume that a 5 GHz TE10 mode
is incident on the waveguide junction from the smaller waveguide. Use the transmission line
model, ascertain the approximate amplitudes of the reflection and transmission coefficients
for the TE10 mode.
Problem 5-2: To understand the number of modes required for convergence in the mode-
matching problem in a waveguide, consider the matching of the following two Fourier series:
P1
X
f1 (x) = an einπx/d1 , 0 < x < d1 , (5.8.25)
n=−P1
= 0, otherwise, (5.8.26)
P2
(
X 0 f1 (x), 0 < x < d1 ,
fˆ1 (x) = b0n ein πx/d2 = (5.8.27)
n0 =−P2
0, d 1 < x < d2 ,
where d2 > d1 . Find the coefficents b0n in terms of an . Show that in order for fˆ1 (x) to
approximate f1 (x) well, we require that P2 /d2 >> P1 /d1 .
Problem 5-3:
If a sum of TEm0 modes are incident at the discontinuity in the waveguide in Figure 5.30,
due to the symmetry of the problem, only TEm0 modes are reflected and transmitted. Given
that the incident modes are described by
X
mπx ikmz z
Hz = Hm0 cos e
m
a1
find the transmission and reflection operators that describe mode conversions at the discon-
tinuity.
2a 2b z
b 2
1 a
c 3
Problem 5-4:
A TE11 mode is propagating in a circular waveguide with radius a as shown in Figure
5.31. A circular diaphragm with radius b is placed at z = 0.
(a) Find the reflection and transmission operators due to the presence of this diaphragm.
(b) Give explicit expressions for the elements of the matrices involved in the description of
the reflection and transmission operators.
Problem 5-5:
A parallel plate waveguide is bifurcated as shown in Figure 5.32. Assume a bunch of TMn
mode incident from the left of the waveguide.
(a) Using the formulation of this Chapter, find the reflection operator in waveguide 1, and
transmission operators in waveguides 2 and 3.
The field at the waveguide discontinuity can either be approximated by the modes of
waveguide 1, or the modes of waveguides 2 and 3.
(b) If the field at the discontinuity is approximated by N modes of waveguide 1, how many
reflected modes are there in waveguide 1, and how many transmitted modes are there
in waveguides 2 and 3.
(c) If the field at the discontinuity is approximated by M modes in waveguide 2 and P
modes in waveguide 3, how many reflected modes are there in waveguide 1, and how
many transmitted modes are there in waveguides 2 and 3. In this latter method of
solving the problem, what should be the relative ratio of M and P in order to maintain
the same accuracy for the transmitted field in waveguides 2 and 3. (Hint: If only
1 a
b 2
c
1 mode is assumed in waveguide 2 while 10 modes are assumed in waveguide 3 at the

discontinuity, the accuracy of the transmitted field in waveguide 1 will not be as accurate
as that in waveguide 2. So, a certain ratio needs be maintained to achieve the same
order of accuracy. Think of this in terms of Fourier series expansions.)
Problem 5-6:
A parallel plate waveguide is shown in Figure 5.33. Assume a TM1 mode incident from
the left of the waveguide.
(a) Using the formulation of this Chapter, find the reflection operator in waveguide 1, and
transmission operator in waveguide 2.
(b) If the field at the discontinuity is approximated by N modes of a waveguide with the
same dimension as the aperture, how many reflected modes are there in waveguide 1,
and how many transmitted modes are there in waveguide 2 in theory, in order to obtain
a numerically accurate solution.
Problem 5-7: Show that a typical integral in Equation (5.2.39) looks like the expression in
Equation (5.2.40a).
Problem 5-8: Expand (5.4.10) in a geometrical series, and give a physical explanation of
each term of the geometrical series. Would this series always converge? What can you say
about the norm of the reflection matrices?
Problem 5-9: Explain how you would construct a time reversed solution to Maxwell’s
equations once you have found a time dependent solution. Is the time reversed solution for a
lossy waveguide a physical solution?
Problem 5-10: Show that when the Neumann boundary condition is satisfied for Hx in the
corrugated periodic waveguide, then it is equivalent to tangential electric field being zero on
the surface of the waveguide wall.
Problem 5-11: Evaluate Equation (5.6.23) and hence Equation (5.6.26) in closeform. Write
a computer program to solve for different values of kz for different k. Discuss how you would
truncate the series involved in the equations.
Problem 5-12: Derive Equations (5.7.15) and (5.7.16).

Problem 5-13: Write a computer program to plot kd versus kz d according to Equation

(5.7.24).
Problem 5-14: Derive the guidance condition for a transmission line that is periodically
loaded with a shunt capacitance. Plot kd versus kz d for this circuit.
Bibliography
[1] J. Helszajn, Waveguide Junction Circulators: Theory and Practice, John Wiley, NY, 1998.
[4] L. Matthaei, L. Young, and E. M. T. Jones, Microwave Filters Impedance Matching Net-
work, and Coupling Structures, New York, McGraw-Hill, 1964.
[5] R. Mittra and S.W. Lee, Analytical Techniques in the Theory of Guided Waves, New York,
Macmillan, 1971.
[6] E. Kühn, “A mode-matching method for solving field problem in waveguide and resonator
circuits,” Arch. Elek. Übertragung, vol. 27, 511518, 1973.
[7] L. Lewin, Theory of Waveguides, London: Newnes-Butterworths, 1975.
[8] M. Koshiba, M. Sato, and M. Suzuki, “Application of finite element method to H-plane
waveguide discontinuities,” Electron. Lett., Vol. 18, 364365, 1982.
[9] H. Patzelt and F. Arndt, “Double-plane steps in rectangular waveguides and their applica-
tion for transformers, irises and filters,” IEEE Trans. Microwave Theory and Techniques,
vol. MTT-30, pp. 771-776, May 1982.
[10] W.C. Chew, S. Barone, B. Anderson and C. Hennessy, “Diffraction of axisymmetric

waves in a borehole by bed boundary discontinuities,” Geophysics, vol. 49, no. 10, pp.
1586-1595, Oct. 1984.
[11] M. Koshiba and M. Suzuki, “Application of the boundary element method to waveguide
discontinuities,” IEEE Trans. Microwave Theory Tech., Vol. 34, 301307, 1986.
[12] J. D. Wade and R. H. MacPhie, “Scattering at circular-to-rectangular waveguide junc-

tions,” IEEE Trans. Microwave Theory Tech., vol. 34, pp. 10851091, Nov. 1986.
[13] F. Arndt, I. Ahrens, U. Papziner, U. Wiechmann, and R. Wilkeit, “Optimized E-plane T-

junction series power dividers,” IEEE Trans. Microwave Theory Tech., vol. 35, 10521059,
1987.
225
[14] F. Alessandri, G. Bartolucci, R. Sorrentino, “Admittance matrix formulation of waveg-

uide discontinuity problems: computer-aided design of branch guide directional couplers,”
IEEE Trans. Microwave Theory and Techniques, Vol. 36, No. 2, pp 394-403, Feb. 1988.
[15] R.R. Mansour, R.S.K. Tong, and R.H. McPhie, “Simplified description of the field dis-
tribution in finlines and ridge waveguides and its application to the analysis of E-plane
discontinuities,” IEEE Trans. Microwave Theory and Techniques, vol. MTT-36, pp. 1825-
1832, Dec. 1988.
[16] T. Itoh, Ed., Numerical Techniques for microwave and millimeter wave passive struc-
tures, New York, Wiley, 1989.
[17] Q.H. Liu and W.C. Chew, “Numerical mode-matching method for the multiregion, ver-
tically stratified media,” IEEE Trans. Antennas Propag., vol. AP-38, no. 4, pp. 498-506,
Apr. 1990.
[18] M. Guglielmi and C. Newport, “Rigorous, multimode equivalent network represntation

of inductive discontinuities,” IEEE Trans. Microwave Theory and Techniques, Vol. 38,
No. 11, pp 1651-1659, Nov. 1990.
[19] Q.H. Liu and W.C. Chew, “Analysis of discontinuities in planar dielectric waveguides:
An eigenmode propagation method,” IEEE Trans. Micro. Theory Tech., vol. 39, no. 3,
pp. 422-430, Mar. 1991.
[20] W. C. Chew, K. H. Lin, J. Friedrich and C. H. Chan, “Reflection and transmission

operators for general discontinuities in waveguides,” J. Elect. Waves Appl., vol. 5, no. 8,
pp. 819-834, 1991.
[21] K. Ise, K. Inoue, and M. Koshiba, “Three-dimensional finite-element method with edge
elements for electromagnetic waveguide discontinuities,” IEEE Trans. Microwave Theory
Tech., vol. 39, pp. 1289-1295, Aug. 1991.
[22] X. Liang, K. A. Zaki, and A. E. Atia, “Rigorous three plane mode-matching technique
for characterizing waveguide T-junctions, and its application in multiplexer design,” IEEE
Trans. Microwave Theory Tech., Vol. 39, 21382147, 1991.
[23] T. Sieverding and F. Arndt, “Field theoretic CAD of open or aperture matched T-
junction coupled rectangular waveguide structures,” IEEE Trans. Microwave Theory
Tech., vol. 40, 353362, 1992.
[24] R.R. Mansour and J. Dude, “Analysis of microstrip T-junction and its applications to
the design of transfer switches,” IEEE-MTT-S Dig., pp. 889-892, 1992.
[25] F. Alessandri, M. Mongiardo, and R. Sorrentino, “A technique for the fullwave automatic
synthesis of waveguide components: application to fixed phase shifters,” IEEE Trans.
Microwave Theory Tech., Vol. 40, 14841495, 1992.
[26] T. Sieverding and F. Arndt, “Modal analysis of the magic tee,” IEEE Microwave Guided
Wave Lett., vol. 3, pp. 150-152, May 1993.
[27] R. Keller and F. Arndt, “Rigorous modal analysis of the asymmetric rectangular iris in
circular waveguides,” IEEE Microwave Guided Wave Lett., vol. 3, pp. 185187, June 1993.
[28] J.H. Lee, H. J. Eom, J. W. Lee, and K. Yoshitomi, “Transverse electric mode scattering
from rectangular grooves in parallel plate,” Radio Science, vol. 29, 12151218, 1994.
[29] J.M. Rebollar, J. Esteban, and J. E. Page, “Fullwave analysis of three and four-port
rectangular waveguide junctions,” IEEE Trans. Microwave Theory Tech., vol. 42, 256263,
1994.
[30] F. Alessandri, M. Mongiardo, and R. Sorrentino, “Rigorous mode matching analysis
of mitered E-plane bends in rectangular waveguide,” IEEE Microwave and Guide Wave
Lett., vol. 4, 408410, 1994.
[31] J.W. Lee and H. J. Eom, “TE-mode scattering from two junctions in H-plane waveguide,”
IEEE Trans. Microwave Theory Tech., vol. 42, 601606, 1994.
[32] Z. Ma and E. Yamashita, “Port reflection coefficient method for solving multi-port mi-
crowave network problems,” IEEE Trans. Microwave Theory Tech., vol. 43, 331337, 1995.
[33] W. Pascher and R. Pregla, “Analysis of rectangular waveguide discontinuities by the
method of lines,” IEEE Trans. Microwave Theory Tech., vol. 43, 416420, 1995.
[34] R.H. MacPhie and K.L. Wu, “Scattering at the junction of a rectangular waveguide and
a larger circular waveguide,” IEEE Trans. Microwave Theory Tech., vol. 43, pp. 20412045,
Sept. 1995.
[35] Z. Shen and R.H. MacPhie, “Scattering by a thick off-centered circular iris in circular
waveguide,” IEEE Trans. Microwave Theory Tech., vol. 43, pp. 26392642, Nov. 1995.
[36] A. Weisshaar, M. Mongiardo, and V. K. Tripathi, “CAD-oriented equivalent circuit
modelling of step discontinities in rectangular waveguides,” IEEE Microwave and Guided
Wave Letters, vol. 6, no. 4, pp. 171-173, April 1996. 38
[37] A.A. Melcon, G. Connor, M. Guglielmi, “New simple procedure for the computation of
the multimode admittance or impedance matrix of planar waveguide junctions,” IEEE
Trans. Microwave Theory and Techniques, Vol. 44, No. 3, pp 413-416, March 1996
[38] C.T. Iatrou and M. Cavenago, “Field analysis of rectangular waveguide open junction,”
IEEE Trans. Microwave Theory Tech., vol. 45, 165172, 1997.
[39] P. Matras, R. Bunger, and F. Arndt, “Mode scattering matrix of the general step dis-
continuity in elliptical waveguides,” IEEE Trans. Microwave Theory Tech., vol. 45, pp.
453457, Mar. 1997.
[40] S.P. Yeo and S.G. Teo, “Thick eccentric circular iris in circular waveguide,” IEEE Trans.
Microwave Theory Tech., vol. 46, pp. 11771180, Aug. 1998.
[41] H. Jia, K. Yoshitomi, and K. Yasumoto, “Rigorous analysis of rectangular waveguide
junctions by Fourier transform technique,” Progress in Electromagnetics Research, PIER
20, 263282, 1998.
[42] S. Amari. J. Bornemann, A. Laisn, and R. Vahldieck, “Design and analysis of iris-coupled
and dielectric loaded 1/8-cut TE01-mode microwave bandpass filters,” IEEE Trans. Mi-
crowave Theory Tech. 49, 413-421, March 2001.
[43] K. Radhakrishnan and W.C. Chew, “Efficient analysis of waveguiding structures,” in
Fast and Efficient Algorithms in Computational Electromagnetics, W. Chew, J. Jin, E.
Michielssen, and J. Song, editors, Artech House, Boston, 2001.
[44] Z.X Shen, C.K. Law, C. Qian, “Hybrid finite-element-modal-expansion method for
matched magic T-junction,” IEEE Transactions on Magnetics, vol. 38, no. 2, pp. 385-
388, 2002.
[46] C. Kittel, Introduction to Solid State Physics, 7th Ed., John Wiley & Sons, 1995.
[48] H.A. Haus, Electromagnetic Noise and Quantum Optical Measurements, Berlin:
Springer-Verlag, 2000.
[49] V. G. Veselago, “The electrodynamics of substances with simultaneously negative values
of permittivity and permeability,” Soviet Physics USPEKI, Vol. 10, No. 509, 1968.
[50] J. B. Pendry, “Negative refraction makes a perfect lens,” Phys. Rev. Lett., Vol 85, No.
3966, 2000.
[51] R. A. Shelby, D. R. Smith, S. C. Nemat-Nasser, and S. Schultz, “Microwave transmission
through a two-dimensional, isotropic, left-handed metamaterial,” Applied Physics Letters,
Vol. 78, Issue 4, pp. 489-491, Jan. 22, 2001.
[52] J.R. Thomas and A. Ishimaru, “Transmission properties of material with relative per-
mittivity and permeability close to 1, Proc. SPIE, vol. 4806, pp. 167-175, 2002.
[53] W. C. Chew, “Some reflections on double negative materials,” PIER 51, pp. 1-26, 2005.
[54] A. Lai, T. Itoh, and C. Caloz, “Composite right/left-handed transmission line metama-
terials,” IEEE Microwave Magazine, 5 (3), 34-50, 2004.
[55] C. Caloz and T. Itoh, “Transmission line approach of left-handed (LH) structures and
microstrip implementation of an artificial LH transmission line, IEEE Trans. Antennas
Propagat., vol. 52, no. 5, pp. 11591166, May 2004.
Chapter 6
Optical Waveguides
Optical waveguides are some of the most important waveguides. Their importance stems
from the broad bandwidth and low loss deliverable by optical communication systems [1, 2].
As a result, optical fiber cables have replaced transmission line cables as submarine cables
throughout the world for global communication [3]. The use of optical fiber for communication
was first proposed by Kao and Hockham [4]. The success of Corning Glass Works on making
fiber of loss below 20 dB/km in the 1970s spurred tremendous interest in the use of optical
fiber. An interesting account of the history is given in Okoshi [5].
Optical fibers work by the physics of total internal reflection. Waves are confined within
a waveguide by total internal reflection due to the presence of a dielectric interface. Hence,
no metallic part is needed in the construction of such a waveguide. The presence of metallic
component is deleterious at high frequencies due to the loss it incurs. Optical fibers now can
have a loss as low as 0.2 dB/km [2].
In an optical waveguide, which usually is an open waveguide, a wave is guided along a
structure, but the field is not enclosed completely in the structure—the field extends to infin-
ity. However, outside the waveguide, the field is evanescent, an it decays exponentially away
from the guiding structure; hence, the energy of the wave is still localized around the guiding
structure. The guiding structure is often filled with inhomogeneous medium. Therefore, many
properties of inhomogeneously filled waveguides are also true in open, inhomogeneously filled
waveguides.
Due to the importance of optical waveguides, there has been a tremendous amount of
work on this subject. Much of the work can be found from the references for this chapter
and the references therein. This chapter only serves to provide a sampling of some topics
available in the vast literature. Optical waveguides are still under intensive research by many
workers (see reference list). A recent development is the use of photonic crystals for optical
waveguides. Artificial photonic crystals can generate stop bands (band gaps) in which the
wave has to be evanescent in the crystal. The band gap structures are then used to trap
waves inside the waveguide. An excellent overview of planar lightwave circuits is given by
Okamoto [14].
229
6.1 Surface Waveguides–Dielectric Slab Waveguides

An example of an open waveguide is a dielectric slab waveguide. The waveguide is made with
dielectric coating on a ground plane, or in the case of optical thin film waveguides, it is a
coating of an optically more dense medium on top of an optically less dense substrate [5,7–15].
Due to the symmetry of the geometry, we can decompose the field inside such a waveguide
into TM and TE types. The mode is guided by total internal reflection. This is only possible
if 1 > 0 and 1 > 2 . At total internal reflection, the fields in region 0 and 2 are evanescent,
and hence they decay exponentially away from the structure. Therefore, most of the energy
of the mode is still trapped and localized in the vicinity of the structure.
ε0 ε0
(a) ε1 ε1 (b)
ε2 ε0 < ε2 < ε1
Ground Plane
σ→∞
Figure 6.1: Geometry for dielectric slab waveguides.
Region 0 d
Region 1
z
Region 2 y
Figure 6.2: Bouncing waves in a dielectric slab waveguide.
If a TM wave is in a dielectric slab, we can write the field in region 1 as

H1 = ŷ A1 eik1x x + B1 e−ik1x x eikz z . (6.1.1)
Equation (6.1.1) has the physical meaning that the wave in region 1 is representable as
bouncing waves. At x = 0, the upgoing wave is the reflection of the downgoing wave; hence,
we have
TM
A1 = R12 B1 , (6.1.2)
TM
where R12 is the TM reflection coefficient at the 1-2 interface. If there are subsurface
TM
layers, R12 could be the generalized reflection coefficient that includes subsurface reflections.
Otherwise, it is just the single interface, Fresnel reflection coefficient for a TM wave. At the
upper interface at x = d, we require that the downgoing wave is a reflection of the upgoing
wave, i.e.,
B1 e−ik1x d = R10
T M ik1x d
e A1 . (6.1.3)
Optical Waveguides 231
For non-trivial A1 and B1 , Equations (6.1.2) and (6.1.3) imply that

T M T M 2ik1x d
1 − R10 R12 e = 0. (6.1.4)
The above is the guidance condition sometimes known as the transverse resonance condition
for TM modes in a dielectric slab, with
q
TM j kix − i kjx
Rij = , kix = ki2 − kz2 . (6.1.5)
j kix + i kjx
Hy Hy
TM0 Mode TM1 Mode
d/2
x x
–d/2 d/2 –d/2
Hy Hy
TM2 Mode TM3 Mode
d/2
x x
–d/2 d/2
–d/2
Figure 6.3: Field distributions for different modes in a dielectric slab waveguide.
Due to Equation (6.1.5), the guidance condition (6.1.4) can be expressed entirely as a
function of kz . We can either solve Equation (6.1.4) graphically or numerically on a computer.
Once the value of kz that satisfies (6.1.4) is found, it can be used in (6.1.2) or (6.1.3) to find
a relationship between A1 and B1 . The fields in region 0 and 2 can be found easily, i.e.,
TM
H0 = ŷT10 A1 eik1x d+ik0x (x−d)+ikz z , (6.1.6a)
TM
H2 = ŷT12 B1 e−ik2x x+ikz z . (6.1.6b)
In other words, the field in region 0 is a consequence of the transmission of the upgoing wave
in region 1, while the field in region 2 is a consequence of the transmission of the downgoing
wave in region 1. In the above, Tij is a transmission coefficient with
TijT M = 1 + Rij
TM
. (6.1.7)
In order for a mode to be trapped, the field has to decay exponentially in the x direction.
Therefore, k0x and k2x have to be pure imaginary. In other words, we can find the values of
kz for guidance only in the range kz > k0 and kz > k2 .
For a symmetric waveguide where regions 0 and 2 are the same, the first few modes of
the waveguides are as sketched in Figure 6.3. These modes are, in general, more well-trapped
when the contrast is high between region 1 and regions 0 and 2, or when the frequency is
high.
We can study the guidance of a TE mode which is the dual of a TM mode in a similar
fashion.
By a further manipulation of Equation (6.1.4), and using the definition of Fresnel reflection
coefficients, Equation (6.1.4) can be written as

d 0 d k1x d − mπ
α0x = k1x tan , (6.1.8)
2 1 2 2
p p
where α0x = kz2 − k02 , k1x = k12 − kz2 , and m = 1, 2, 3, 4, · · · . The left-hand side of (6.1.8)
can be expressed in terms of the k1x d variable, viz.,
s 2 2
q
2 2 d 2 2 d k1x d
kz − k0 = (k1 − k0 ) − , (6.1.9)
2 2 2
which is the equation of a circle. Equation (6.1.8) can hence be solved graphically by plotting
both sides of Equation (6.1.8) as a function of k1z d.
A mode in a dielectric slab waveguide has part of its energy outside the waveguide and
inside the waveguide as shown in Figure 6.5. Hence, the group velocity of the mode is between
that of the slab region and the outer region. When the frequency is very low, the mode is
weakly evanescent outside the dielectric slab and the group velocity of a mode is closer to
that of the outer region because most of the energy of the mode is outside the waveguide.
When the frequency is high, the mode is strongly evanescent outside and most of the energy
of the mode is trapped inside the slab. Hence, the group velocity of the mode is close to that
of the dielectric slab. Therefore, the dispersion curve of the mode is as shown in Figure 6.5.
From Figure 6.5, it is clear that the TM0 mode has no cut-off, since continuity of the slope
and field amplitude can be satisfied for all frequencies and yet the field is evanescent outside.
This is not true of the higher order modes. Figure 6.6 shows the profile of the TM1 mode
for different frequencies. For high frequencies, the mode is well-trapped inside the dielectric
waveguide. When the frequency is low such that λ/4 = d/2, the field ceases to be evanescent
outside the dielectric slab. In this case, the field is a constant outside, and this is the mode at
the cut-off frequency precisely. When λ/4 > d/2, the mode leaks energy to outside the slab,
it becomes a leaky mode and is not guided at all. Figure 6.7 shows the dispersion curve of
the TM1 mode in a dielectric slab waveguide. Close to cut-off, the group velocity of the mode
should be close to that of the outside region, and vice-versa for high frequencies. Figure 6.6
allows us to determine the cut-off frequency of a mode quite readily.
6.2 Circular Dielectric Waveguide

An optical fiber is a kind of circular dielectric waveguide. Usually, an optical fiber has a
protective cladding as well. In the ensuing analysis, we will ignore the protective cladding,
i.e., letting b → ∞ [11, 16, 17, 20].
In order for a mode to be guided, we require that 1 µ1 > 2 µ2 . In other words, the light
velocity in the core region has to be slower than the light velocity in the cladding. The field
outside the core region is evanescent for a guided mode. Therefore, letting b → ∞ affects the
guided mode slightly, especially if the mode is tightly bound to the core; hence, it is a good
approximation.
In the optical fiber modes, except for the axisymmetric modes, the TE and TM fields are
coupled to each other by the boundary conditions as in an inhomogeneously filled waveguide.
The z components of the field are solutions to the wave equation in cylindrical coordinates
for each of the homogeneous regions, i.e.,

1 ∂ ∂ 1 ∂2 ∂2 2 Eiz
ρ + + 2 + ω µi i = 0. (6.2.1)
ρ ∂ρ ∂ρ ρ2 ∂φ2 ∂z Hiz
The solution to (6.2.1) in region i is of the general form

 
 Jn (kiρ ρ)
 

e±inφ+ikz z , (6.2.2)

 (1) 

Hn (kiρ ρ)
2 (1)
where kiρ + kz2 = ki2 and ki is the wave number in region i. Hn (kρ ρ) is singular when ρ → 0
while Jn (kρ ρ) is regular. Therefore, for the einφ harmonic, the solution in the core region is
E1z = E1 Jn (k1ρ ρ)eikz z+inφ , (6.2.3a)
H1z = H1 Jn (k1ρ ρ)eikz z+inφ . (6.2.3b)

In the cladding region, when b → ∞, we should only have outgoing waves. Therefore,
E2z = E2 Hn(1) (k2ρ ρ)eikz z+inφ , (6.2.4a)
H2z = H2 Hn(1) (k2ρ ρ)eikz z+inφ . (6.2.4b)

By the phase matching condition, kz is the same in both regions. The choice of Hankel
functions of the first kind in (6.2.4a) and (6.2.4b) stems from the fact that
s
(1) 2 ikρ ρ−in π −i π
Hn (kρ ρ) ∼ e 2 4 , ρ → ∞, (6.2.5a)
πkρ ρ
s
2 nπ π
Jn (kρ ρ) ∼ cos(kρ ρ − − ), ρ → ∞. (6.2.5b)
πkρ ρ 2 4
Equation (6.2.5a) corresponds to an outgoing wave whilep(6.2.5b) corresponds to a standing
wave. When a mode is confined in the core, k2ρ = k22 − kz2 = iα2 is always positive
imaginary. Therefore, guidance is only possible if kz > k2 .
At this point, we introduce the modified Bessel function Kn (x) such that
2 −i πn
Hn(1) (iα2 ρ) = e 2 Kn (α2 ρ). (6.2.6)
πi
As such, Equations (6.2.4a) and (6.2.4b) can be written as
E2z = E20 Kn (α2 ρ)eikz z+inφ , (6.2.7a)
H2z = H20 Kn (α2 ρ)eikz z+inφ , (6.2.7b)

p
where α2 = kz2 − k22 . When ρ → ∞,
r
π −α2 ρ
Kn (α2 ρ) ∼ e , (6.2.8)
2α2 ρ
corresponding to an evanescent wave. A guided mode must satisfy the boundary conditions
at the core-cladding interface. Therefore, we need to find Eφ and Hφ , the other tangential
components of the field.
In the core region, we have

1 nkz
E1φ = 2 − E1 Jn (k1ρ ρ) − iωµ1 H1 k1ρ Jn0 (k1ρ ρ) einφ+ikz z , (6.2.9a)
k1ρ ρ

1 nkz
H1φ = 2 − H1 Jn (k1ρ ρ) + iω1 E1 k1ρ Jn (k1ρ ρ) einφ+ikz z .
0
(6.2.9b)
k1ρ ρ
In the cladding region, we have

1 nkz 0
E2φ = − 2 − E Kn (α2 ρ) − iωµ2 H2 α2 Kn (α2 ρ) einφ+ikz z ,
0 0
(6.2.10a)
α2 ρ 2

1 nkz 0
H2φ =− 2 − H Kn (α2 ρ) + iω2 E2 α2 Kn (α2 ρ) einφ+ikz z .
0 0
(6.2.10b)
α2 ρ 2
The continuity of the tangential components of the fields from (6.2.3a), (6.2.3b), (6.2.7a),
(6.2.7b), (6.2.9a), (6.2.9b), and (6.2.10a), (6.2.10b) implies that
E1 Jn (k1ρ a) = E20 Kn (α2 a), (6.2.11a)
H1 Jn (k1ρ a) = H20 Kn (α2 a), (6.2.11b)

1 nkz 0
2 − E1 Jn (k1ρ a) − iωµ1 H1 k1ρ Jn (k1ρ a)
k1ρ a
(6.2.11c)
1 nkz 0
=− 2 − E2 Kn (α2 a) − iωµ2 H20 α2 Kn0 (α2 a) ,
α2 a

1 nkz 0
2 − H1 Jn (k1ρ a) + iω1 E1 k1ρ Jn (k1ρ a)
k1ρ a
(6.2.11d)
1 nkz 0 0 0
=− 2 − H Kn (α2 a) + iω2 E2 α2 Kn (α2 a) .
α2 a 2
Substituting (6.2.11a) and (6.2.11b) into (6.2.11c) and (6.2.11d), and finding the determinant
of the resultant equations, we arrive at
!2
1 1
kz2 n2 2 a2 + α2 a2
k1ρ 2
0

2 µ1 Jn (k1ρ a) µ2 Kn0 (α2 a) (6.2.12)
=ω +
k1ρ aJn (k1ρ a) α2 aKn (α2 a)

1 Jn0 (k1ρ a) 2 Kn0 (α2 a)
+ .
p p
Since k1ρ = k12 − kz2 , α2 = kz2 − k22 , we can solve the above transcendental equation for
kz . Once the values of kz that satisfies (6.2.12) are found, we can find the ratios of E1 /H1
from (6.2.11a)–(6.2.11d). In particular,
! −1
E1 nkz 1 1 1 Jn0 (k1ρ a) 2 Kn0 (α2 a)
= 2 a2 + α2 a2 + . (6.2.13)
H1 iω k1ρ 2 k1ρ aJn (k1ρ a) α2 aKn (α2 a)
This is the ratio of the TM wave amplitude to the TE wave amplitude inside the core.
K 0 (x) J 0 (x)
In Equation (6.2.12), xKnn (x) is not rapidly oscillating while xJnn (x) is rapidly oscillating.
0
Jn (k1ρ a)
We can solve Equation (6.2.12) for k1ρ aJn (k1ρ a) giving

Jn0 (k1ρ a) 1 µ2 2 Kn0 (α2 a)
=− +
k1ρ aJn (k1ρ a) 2 µ1 1 α2 aKn (α2 a)
" 2 2
1 µ2 2 Kn0 (α2 a)
± −
4 µ1 1 α2 aKn (α2 a) (6.2.14)
!2  21
n2 kz2 1 1  .
+ 2 2 a2 + α2 a2
k1 k1ρ 2
The plus and minus signs give rise to two classes of solutions. We next make use of the
recurrence relationship of Bessel functions
n
Jn0 (x) = −Jn+1 (x) + Jn (x), (6.2.15a)
x
n
Jn0 (x) = Jn−1 (x) − Jn (x), (6.2.15b)
x
to get

Jn+1 (k1ρ a) 1 µ2 2 Kn0 (α2 a)
= +
(6.2.16a)
n
+ −R , EH modes,
(k1ρ a)2

Jn−1 (k1ρ a) 1 µ2 2 Kn0 (α2 a)
=− +
(6.2.16b)
n
+ − R , HE modes,
(k1ρ a)2
where
 !2  21
2 2
1 µ2 2 Kn0 (α2 a) n2 kz2 1 1
R= − + 2 a2 + α2 a2
 . (6.2.16c)
4 µ1 1 α2 aKn (α2 a) k12 k1ρ 2
For the n = 0 case, or the axisymmetric case, the TE and TM fields are decoupled, and
the TE and TM modes are the dual of each other. The guidance conditions are
J1 (k1ρ a) 2 K1 (α2 a)
=− , TM, (6.2.17a)
k1ρ aJ0 (k1ρ a) 1 α2 aK0 (α2 a)
J1 (k1ρ a) µ2 K1 (α2 a)
=− , TE. (6.2.17b)
k1ρ aJ0 (k1ρ a) µ1 α2 aK0 (α2 a)
In the above, we have used K00 (x) = −K1 (x). Note that TE and TM waves are decoupled
from (6.2.11a)–(6.2.11c), since the TE and TM waves can satisfy the boundary conditions
separately. By writing
(α2 a)2 = (k1 a)2 − (k2 a)2 − (k1ρ a)2 = V 2 − (k1ρ a)2 , (6.2.18)
(6.2.17a) can be solved in terms of k1ρ a. To get a feeling for the behavior of the solution, it
can be solved graphically. Since α2 a > 0 for a guided mode, from (6.2.18), note that we need
only to consider the case where
2 2
0 < k1ρ a < (k12 − k22 )a2 = V 2 . (6.2.19)
The right-hand side of (6.2.17a) is always negative. For the TE case, it has a value of
−µ2 K1 (V )/µ1 V K0 (V ) at k1ρ a = 0, and it has a value of −∞, when k1ρ a → V . More
precisely,
µ2 K1 (α2 a) 2µ2
− ∼ 2 2 a2 ) , k1ρ a → V. (6.2.20)
µ1 α2 aK0 (α2 a) µ1 (V − k1ρ a ) ln(V 2 − k1ρ
2 2
The left-hand side of (6.2.17a) starts from 1/2 at k1ρ a = 0 and goes to infinity at the zeros
of J0 (k1ρ a), and goes to zero at the zeros of J1 (k1ρ ). A sketch of the left-hand side and the
right-hand
p side of (6.2.17a) is shown in Figure 6.9. The number of guided modes depends on
V = (k12 − k22 )a, the normalized frequency. V can be increased by increasing the contrast
between 1 and 2 , by raising the frequency, or by increasing a. For V < 2.405, there could
be no possible guided modes. Hence, all axisymmetric modes have a finite cut-off frequency.
When n = 1, Equations (6.2.16a), (6.2.16b), become

J2 (k1ρ a) 1 µ2 2 K10 (α2 a) 1
= + + − R , EH (6.2.21a)
k1ρ aJ1 (k1ρ a) 2 µ1 1 α2 aK1 (α2 a) (k1ρ a)2

J0 (k1ρ a) 1 µ2 2 K10 (α2 a) 1
=− + + − R , HE. (6.2.21b)
k1ρ aJ1 (k1ρ a) 2 µ1 1 α2 aK1 (α2 a) (k1ρ a)2
The left and the right-hand side of (6.2.21a) is shown in Figure 6.10 and Figure 6.11. The
right-hand side of (6.2.21a) goes to infinity at k1ρ a → V , while the left-hand side goes to zero
and infinity as before. Note that all the EH modes have finite cut-off frequencies while the
HE11 mode does not have a cut-off frequency. Therefore, the HE11 mode is the dominant
mode in an optical fiber. If (i) the optical fiber is small enough, or (ii) the frequency is low
enough, or (iii) when the contrast is very low, it is the only mode propagating for the single
mode operation of the optical fiber.
Usually, in a waveguide, an EH notation is used to denote a mode where Ez dominates
over Hz , or the TM component dominates over the TE component. However, due to a quirk in
the history of optical fibers, the EH notation is used to denote a mode whose TE component
dominates over its TM component, and vice versa for the HE notation.1
A parameter of interest is the axial wave number kz as a function of the normalized
frequency. In Figure 6.12, we plot kz /k0 versus V . At very low frequencies, or near the
cut-off of a mode, kz /k0 approaches the refractive index n2 of the cladding. This is because
the mode is not well confined, and most of the energy of the mode is in medium 2. Hence,
the mode phase velocity is close to that of medium 2. At higher normalized frequency, the
mode is well confined, and it propagates in medium 1. Hence, kz /k0 approaches n1 which is
the refractive index of medium 1.
From Equation (6.2.12), it is clear that when we find a solution, a dual solution also exists.
If µ1 /µ2 = 1 /2 , then a solution and its dual are degenerate. However, usually, µ1 /µ2 = 1;
therefore, a solution and its dual are not degenerate. However, when 1 /2 → 1 as in the case
of a weak-contrast optical fiber, a solution and its dual are near degenerate.
6.3 Weak-Contrast Optical Fiber

When the contrast between the cladding and the core of the fiber is small, the analysis of the
guided mode inside a fiber can be greatly simplified. In this case, polarization effect at the
interface of the fiber is unimportant, and scalar wave theory can be applied [11, 14, 18].
The vector wave equations governing electromagnetic fields in an inhomogeneous waveg-
uide are
∇ × µr−1 ∇ × E − ω 2 µ0 0 r E = 0, (6.3.1a)
∇ × −1 2
r ∇ × H − ω µ0 0 µr H = 0, (6.3.1b)
where µr = µ(r)/µ0 = f1 (r), and r = (r)/0 = f2 (r). If we find a solution to (6.3.1a), the
solution to the dual problem is obtained by letting E → −H, H → E, µr → r , r → µr .
However, if f1 (r) 6= f2 (r), the dual problem corresponds to a different waveguide with µr =
f2 (r) and r = f1 (r). In order for a dual problem to be itself, we require that f1 (r) = f2 (r).
If this is the case, then a mode and its dual are degenerate. Therefore, we can associate
every mode in a waveguide with a dual mode. However, f1 (r) = 1 usually, while f2 (r) 6= 1;
therefore, a mode is not degenerate with its dual.
1 The author has not verified this but read about it somewhere.
However, in the case of a weak-contrast optical fiber, r → 1; hence, a mode is near

degenerate with its dual. Furthermore, we can show that the vector nature of the wave is
unimportant. If µr = 1, we can rewrite (6.3.1a) and (6.3.1b) as
∇ × ∇ × E − ω 2 µ0 0 r E = 0, (6.3.2a)
∇ × ∇ × H − (∇ ln r ) × ∇ × H − ω 2 µ0 0 r H = 0. (6.3.2b)
∇r
In the above ∇ ln r = ris the effect of the polarization charges at the dielectric interface.
If the transverse and longitudinal components of (6.3.2a) and (6.3.2b) are extracted, the
following equations ensue:
∇2 Es + ∇s [∇s ln r · Es ] + k 2 Es = 0, (6.3.3a)
∇2 Ez + ikz (∇s ln r ) · Es + k 2 Ez = 0, (6.3.3b)

and
∇2 Hs + (∇s ln r ) × (∇s × Hs ) + k 2 Hs = 0, (6.3.4a)
2 2
∇ Hz + (∇s ln r ) × (∇s × Hz ) + (∇s ln r ) × (ẑikz × Hs ) + k Hz = 0. (6.3.4b)
In the limit when r → 1, the polarization charge terms in (6.3.3a), (6.3.3b), (6.3.4a), and
(6.3.4b) can be ignored with respect to the other terms, yielding
∇2 Es + k 2 Es = 0, (6.3.5a)
∇2 Ez + k 2 Ez = 0, (6.3.5b)
∇2 Hs + k 2 Hs = 0. (6.3.5c)
2 2
∇ Hz + k Hz = 0. (6.3.5d)
In other words, the wave guidance problem by a fiber of weak contrast reduces to a scalar
problem. However, when the contrast goes to 1, the guided mode in the fiber becomes a TEM
mode with k = kz , and Ez = Hz = 0. This represents the leading order solution when the
contrast r = 1.
When r > 1, we see from Equation (6.3.3b) that a nonzero Ez is induced by the presence
of Es . Moreover, by comparing terms, and assuming high frequency, Ez ∼ |∇ ln r · Es |/k.
Therefore, when ω → ∞, |Ez | |Es | (see Problem 6-8.). By the same token, |Hz | |Hs |
when ω → ∞. Therefore, when the contrast is very weak, and the frequency is very high, the
mode is quasi-TEM, and ∇s · r Es = −ikz r Ez ≈ 0 and ∇s · Hs = −ikz Hz ≈ 0.
When the contrast is such that r > 1, the polarization terms in (6.3.3a) and (6.3.4a) are
subdominant when ω → ∞. This is because the polarization terms in (6.3.3a) and (6.3.4a)
have only first derivative in space. When the frequency becomes large, the field becomes
rapidly varying. The first derivatives in space will be smaller than the second derivatives in
space. Therefore, the polarization term cannot match the k 2 term compared to the ∇2 term.
Consequently, (6.3.5a) and (6.3.5c) are still the equations to solve.
Equations (6.3.5a) and (6.3.5c) are equivalent to
(∇2 + k 2 )φ = 0 (6.3.6)
where φ is either Ex , Ey , Hx , or Hy .
For example, we can let
(
AJn (k1ρ ρ)einφ+ikz z , ρ < a,
φ= (6.3.7)
BKn (α2 ρ)einφ+ikz z , ρ > a,
The boundary conditions for φ at the interface where k 2 = ω 2 µ0 0 r displays a step discon-
tinuity is
φ1 = φ2 , (6.3.8a)
n̂ · ∇φ1 = n̂ · ∇φ2 . (6.3.8b)
These boundary conditions are derivable from Equation (6.3.6) alone.
Imposing the above boundary conditions at ρ = a for the weak contrast optical fiber,
whose field is given by (6.3.7), we have
AJn (k1ρ a) = BKn (α2 a), (6.3.9a)
Ak1ρ Jn0 (k1ρ a) = Bα2 Kn0 (α2 a). (6.3.9b)
The above yields
k1ρ Jn0 (k1ρ a) α2 Kn0 (α2 a)
= . (6.3.10)
Jn (k1ρ a) Kn (α2 a)
n
Using the recurrence relationship that Jn0 (x) = −Jn+1 (x) + x Jn (x) , and that Kn0 (x) =
−Kn+1 (x) + nx Kn (x), we can transform the above to
k1ρ Jn+1 (k1ρ a) α2 Kn+1 (α2 a)
= . (6.3.11)
n
Similarly, using the recurrence relationship that Jn0 (x) = Jn−1 (x) − x Jn (x), and Kn0 (x) =
−Kn−1 (x) − nx Kn (x), we have
k1ρ Jn−1 (k1ρ a) α2 Kn−1 (α2 a)
=− . (6.3.12)
Comparing with Equations (6.2.16a) and (6.2.16b), we note that now there are half as
many solutions as before. This is because when r → 1, HEn+1,m and EHn−1,m modes are
degenerate. We can solve (6.3.11) graphically as before. The modes thus found are designated
the LPnm mode. The lowest order mode is the LP01 mode which is the degenerate case of
2
the HE11 mode. LP here stands for “linearly polarized.” Figure 6.13 shows the dispersion
kz
curves of the LPnm modes. In the figure, b = k2 − n2 /(n1 − n2 ), and V is the normalized
frequency.
These modes are termed “weakly guided” modes, but it is a misnomer. When V is
large, a mode can still be tightly confined to the waveguide as demonstrated by the phase
velocity approaching that of a core. Also, Ex and Ey are not independent of each other since
∇s ·r Es ≈ 0. The same statement applies to Hx and Hy . Hence, these modes are not linearly
polarized.
2 This could be a misnomer as the fields are not linearly polarized.
It was pointed out by Erhan Kudeki that LP modes are possible in the
infinitely low contrast case. But if the contrast is not small, they break up
into their constituent modes which are not LP.
6.4 Perturbation Formula for Dielectric Waveguides

If we know the solution to a waveguide geometry, and wish to change the phase velocity of the
waveguiding mode, dielectric material can be added to achieve the purpose. A perturbative
approach can be used to analyze such problem [15, 21, 22].
For the unperturbed problem, the electromagnetic field satisfies
∇ × E0 = iωµ0 H0 , (6.4.1)
∇ × H0 = −iω0 E0 . (6.4.2)
Now, let us change the permittivity and the permeability of the waveguide. The new electro-
magnetic field satisfies
∇ × E = iωµH, (6.4.3)
∇ × H = −iωE, (6.4.4)
where µ and are new. Taking the divergence of E∗0 × H and E × H∗0 , we have
∇ · (E∗0 × H) = −iωµ0 H · H∗0 + iωE∗0 · E, (6.4.5a)
∇ · (E × H∗0 ) = iωµH · H∗0 − iω0 E∗0 · E. (6.4.5b)

Adding the above equations, we have
∇ · (E∗0 × H + E × H∗0 ) = iωδµH · H∗0 + iωδE∗0 · E, (6.4.6)
where δµ = µ − µ0 , and δ = − 0 . Since the unperturbed field has eik0z z dependence, while
the perturbed field has eikz z dependence, we have
∇ · (E∗0 × H + E × H∗0 ) = ∇s · (E∗0 × H + E × H∗0 )

(6.4.7a)
+ ẑi(kz − k0z ) · (E∗0 × H + E × H∗0 ).
Integrating Equation (6.4.6) over the cross-section of the waveguide, making use of (6.4.7a),
we have RR
dS[δµH · H∗0 + δE · E∗0 ]
S
kz − k0z = ω RR . (6.4.7b)
dS[E∗0 × H + E × H∗0 ]
S
Equation (6.4.7a) is exact at this point, but is not very useful because E and H are unknowns
on the right-hand side. However, when the perturbation is small, we can approximate E ' E0
and H ' H0 , and (6.4.7a) becomes
RR
dS[δµ|H0 |2 + δ|E0 |2 ]
S
kz − k0z ' ω RR . (6.4.8)
2<e dS[E0 × H∗0 ]
S
The time average power flow in the unperturbed waveguide is given by

ZZ
1
hPf i = <e dS[E0 × H∗0 ]. (6.4.9)
2
The time average energy stored per unit length is

ZZ
1
hW i = hWe i + hWm i = 2hWe i = dS|E|2 . (6.4.10)
2
S
In the case of a step-index optical fiber, where a perturbation is in the dielectric constant,
which is uniformly δ1 in the core region and is uniformly δ2 in the cladding region, we can
rewrite (6.4.8) as RR RR
dSδ1 |E1 |2 + dSδ2 |E2 |2
S1 S2
δkz ' ω , (6.4.11)
4hPf i
where E1 is the original field in the core region, and E2 is the original field in the cladding
region. Since δi is constant in region i, we can rewrite (6.4.11) as
δ1 −1 −1
1 hWe1 i + δ2 2 hWe2 i
δkz ' ω , (6.4.12)
hPf i
where hWe1 i is the time average energy stored in the electric field in region 1 while hWe2 i
is that for region 2. Since the total power flow in each region is 2vg hWei i, we can rewrite
(6.4.12) as
ω
δkz ' [δ1 −1 −1
1 Γ1 + δ2 2 Γ2 ], (6.4.13)
2vg
where Γi is the fraction of power flow in region i.
For a weak contrast optical fiber, we can further approximate the above with 1 ' 2 ,
vg ' c1 ' c2 ,
k1
δkz ' [δ1 Γ1 + δ2 Γ2 ]. (6.4.14)
21
Defining an effective refractive index n = kz /k0 , then δn = δkz /k0 , and the above can be
rewritten as
1
δn ' [δn21 Γ1 + δn22 Γ2 ], (6.4.15)
2n1
or that
δn ' [δn1 Γ1 + δn2 Γ2 ], (6.4.16)
for a weak contrast optical fiber. Hence, the change in the effective refractive index is pro-
portional to the change in the refractive index in each region, weighted by the fraction of the
power flow in each region.
6.5 Mode Dispersion in an Optical Fiber

Once the axial wave member kz of our optical fiber is found, we can define an effective
refractive index given by
kz
n= . (6.5.1)
k0
In other words,
ω
kz = n , (6.5.2)
c0
where n is a function of n1 , n2 and ω or n = n(n1 , n2 , ω). The signal in an optical fiber
travels at the group velocity. Hence, we can study how the group velocity depends on n1 , n2
and ω.
1 dkz ω ∂n dn1 ∂n dn2 ∂n n
= = + + + . (6.5.3)
vg dω c0 ∂n1 dω ∂n2 dω ∂ω c0
This first two terms in the square bracket come from the materials making up n1 and n2 ,
and the last term in the square bracket is a consequence of the waveguide geometry. From
(6.4.16), we deduce that
∂n ∂n
= Γ1 , = Γ2 , (6.5.4)
∂n1 ∂n2
This first two terms for a weak contrast optical fiber, and Γ1 and Γ2 are the fractions of power
flow in regions 1 and 2. With the approximation that

dn1 dn2 ∂n
' = , (6.5.5)
dω dω ∂ω m
where the subscript m stands for dispersion from material property, we can rewrite (6.5.3) as

1 dkz ω ∂n ∂n n
= = + + , (6.5.6)
vg dω c0 ∂ω m ∂ω w c0
where the subscript w stands for dispersion from waveguide geometry. With ω = 2πc0 /λ, we
have ω/dω = −λ/dλ, and the above can be rewritten as

1 λ ∂n ∂n n
=− + + . (6.5.7)
vg c0 ∂λ m ∂λ w c0
A measure of dispersion along an optical fiber of length L is the group velocity dispersion
D defined as
∂T
D = L−1 , (6.5.8)
∂λ
where T is the travel time taken by a pulse to traverse the length L of the optical fiber. Since
T = L/vg , we have 2 2
∂ 1 λ ∂ n ∂ n
D= =− 2
+ . (6.5.9)
∂λ vg c0 ∂λ m ∂λ2 w
In order to have least
2pulse
distortion, we should operate at a2 frequency
where both the
material dispersion, ∂∂λn2 , and the waveguide dispersion, ∂∂λn2 are small. For GeO2 -
m w
doped silica, the material dispersion passes through a minimum at λ = 1.3µm. The waveguide
dispersion can be altered by altering a as well as n1 and n2 . By choosing a core diameter
between 4 and 5 µm, and relative refractive index difference of (n1 − n2 )/n1 > 0.004, the
wavelength of the minimum group velocity dispersion can be shifted to 1.5 to 1.6 µm region
where the loss is lowest (see Figure 6.14 and Figure 6.15).
6.6 A Rectangular Dielectric Waveguide

When a dielectric waveguide is rectangular in shape, there is no closed form solution to the
problem. The eigenmodes of the waveguide has to be found numerically. We shall discuss
two methods of solving the rectangular dielectric waveguide problem.
6.6.1 Harmonic Expansion Method

In this method [24] 3 , we assume that the waveguide is piecewise homogeneous in each region.
Inside the waveguide, the field can be decomposed into TE and TM to z waves, each of which
satisfies the following equations,
TE : (∇2s + k1s
2
)H1z (r) = 0, r ∈ region 1, (6.6.1)
TM : (∇2s + k1s
2
)E1z (r) = 0, r ∈ region 1 (6.6.2)
2
where k1s = ω 2 µ1 1 − kz2 and ∇2s = ∂ 2 /∂x2 + ∂ 2 /∂y 2 . We assume all the fields to have
ikz z
e dependence due to the phase matching condition. In region 0, the fields satisfy similar
equations
TE : (∇2s + k0s
2
)H0z (r) = 0, r ∈ region 0, (6.6.3)
TM : (∇2s + k0s
2
)E0z (r) = 0, r ∈ region 0. (6.6.4)
The TE and TM waves will be coupled by the boundary conditions at the dielectric boundary.
In region 1, the general solution is
X∞
E1z an
= J (k ρ)einφ+ikz z
H1z bn n 1s
n=−∞
∞ (6.6.5)
X an
= <gψn (k1s , rs ),
bn
n=−∞
while in region 0, it is
X∞
E0z cn
= Hn(1) (k0s ρ)einφ+ikz z
H0z dn
n=−∞
∞ (6.6.6)
X cn
= ψ (k , r ),
dn n 0s s
n=−∞
(1)
where <gψn (ks , rs ) = Jn (ks ρ)einφ , and ψn (ks , rs ) = Hn (ks ρ)einφ . We need to derive the
transverse components of the fields in order to match boundary conditions. Using the follow-
ing equations,
i
Es = 2 [kz ∇s Ez − ωµẑ × ∇s Hz ], (6.6.7a)
ks
3 This method is similar to the method of Rayleigh hypothesis described in [25]. It eventually will have
ill-conditioning problem when the number of unknowns is large. Mercatili [26] also analyzed this problem
approximately, although with a different approach.
i
Hs = [kz ∇s Hz + ωẑ × ∇s Ez ], (6.6.7b)
ks2
we deduce that in region 1,
∞
X i
E1s = 2 [kz an ∇s <gψn (k1s , rs ) − ωµ1 bn ẑ × ∇s <gψn (k1s , rs )], (6.6.8a)
k
n=−∞ 1s
∞
X i
H1s = [kz bn ∇s <gψn (k1s , rs ) + ω1 an ẑ × ∇s <gψn (k1s , rs )], (6.6.8b)
k2
n=−∞ 1s
and in region 0,
∞
X i
E0s = 2 [kz cn ∇s ψn (k0s , rs ) − ωµ0 dn ẑ × ∇s ψn (k0s , rs )], (6.6.9a)
k
n=−∞ 0s
∞
X i
H0s = 2 [kz dn ∇s ψn (k0s , rs ) + ω0 cn ẑ × ∇s ψn (k0s , rs )]. (6.6.9b)
k
n=−∞ 0s
The boundary condition requires that tangential E and H be continuous across the dielectric
interface. Matching the z component of the field, we have
∞
X ∞
X
an <gψn (k1s , rs ) = cn ψn (k0s , rs ), rs ∈ C, (6.6.10a)
n=−∞ n=−∞
∞
X ∞
X
bn <gψn (k1s , rs ) = dn ψn (k0s , rs ), rs ∈ C. (6.6.10b)
n=−∞ n=−∞
By defining a unit vector ĉ to be pointing along the circumference of the waveguide C, we

can equate the tangential components in (6.6.8a) and (6.6.9a) on C. By doing so, we have
∞
X 1
k 2 [kz an ĉ · ∇s <gψn (k1s , rs ) − ωµ1 bn ĉ · ẑ × ∇s <gψn (k1s , rs )]
n=−∞ 1s
∞
(6.6.11a)
X 1
= [kz cn ĉ · ∇s ψn (k0s , rs ) − ωµ0 dn ĉ · ẑ × ∇s ψn (k0s , rs )] ,
k2
n=−∞ 0s
∞
X 1
k 2 [kz bn ĉ · ∇s <gψn (k1s , rs ) + ω1 an ĉ · ẑ × ∇s <gψn (k1s , rs )]
n=−∞ 1s
∞
(6.6.11b)
X 1
= 2 [kz dn ĉ · ∇s ψn (k0s , rs ) + ω0 cn ĉ · ẑ × ∇s ψn (k0s , rs )] .
k
n=−∞ 0s
Equations (6.6.10a) and (6.6.11a) can be written as

∞
X ∞
X
<gψn (k1s , rs )an = ψn (k0s , rs )cn , rs ∈ C, (6.6.12a)
n=−∞ n=−∞
∞
X
1 kz ĉ · ∇s <gψn −ωµ1 ĉ · ẑ × ∇s <gψn
· an
k 2 ω1 ĉ · ẑ × ∇s <gψn i
n=−∞ 1s
kz ĉ · ∇s <gψn
∞ (6.6.12b)
X 1 kz ĉ · ∇s ψn −ωµ0 ĉ · ẑ × ∇s ψn
= 2 · bn , rs ∈ C
k0s ω0 ĉ · ẑ × ∇s ψn kz ĉ · ∇s ψn
n=−∞
where atn = [an , bn ], ctn = [cn , dn ]. We can truncate the infinite summation to range from −N
to +N . In this case, the an , bn , cn and dn will constitute 4(2N + 1) unknowns. The method
of point matching can be used to convert (6.6.12a) and (6.6.12b) into matrix equations,
N
X N
X
<gψn (k1s , rms )an = ψn (k0s , rms )cn , m = −N, . . . , +N, (6.6.13a)
n=−N n=−N
2 N
X
k0s
2 <gψ n (k1s , rms ) · an
k1s
n=−N
(6.6.13b)
N
X
= ψ n (k0s , rms ) · cn , m = −N, . . . , N
n=−N
where
kz ĉ · ∇s ψn −ωµĉ · ẑ × ∇s ψn
ψn = . (6.6.14)
ωĉ · ẑ × ∇s ψn kz ĉ · ∇s ψn
Equations (6.6.13a) are matrix equations of the form
(1)
A·a=A · b, (6.6.15a)
(1)
B·a=B · b. (6.6.15b)
Eliminating a from the above yields
−1 (1) −1 (1)
[B ·B −A ·A ] · b = M · b = 0. (6.6.16)
p (1) (1)
Since
p A and B are functions of k 1s = k12 − kz2 , and A and B are functions of k0s =
k02 − kz2 , the matrix in (6.6.16) is a function of kz . Nontrivial solutions exist for b, and
hence a, (i.e., the field) only if
det M(kz ) = 0. (6.6.17)
Equation (6.6.17) allows us to solve for the wavenumber kz of a guided mode. The above
method is in general, applicable to waveguides of arbitrary shapes. For rectangular waveg-
uides, symmetry may be exploited to reduce the extend of the summation in (6.6.12a), and
hence the number of unknowns.
In this method, since we are only assuming standing wave in region 1 and outgoing wave
in region 0, it is not valid if the waveguide is of very distorted shapes. As mentioned earlier,
this method is similar to the Rayleigh’s hypothesis method of solving scattering problem [25].
Because of the hybrid nature of the modes, and the variation in the aspect ratio of a rect-
angular dielectric waveguide, the classification of modes in a rectangular dielectric waveguide
is a complex subject. The EHmn and HEmn notations have been adopted by some workers
to denote the TM-like and TE-like nature, respectively, of the modes. However, the EHmn
and HEmn notations do not indicate if a mode is x-polarized or y-polarized. Hence, another
x y
notation is Emn or Emn to denote if the mode’s electric field is x or y polarized. Yet, another
x y
notation, Hmn or Hmn , is used to denote if the magnetic field is x or y polarized, A combina-
x y x y
tion of Emn , Emn , and Hmn , Hmn has also been suggested to denote the TM-like or TE-like
nature of a mode. The subscripts mn in the above denote that the mode has m maxima in
the x direction and n maxima in the y direction.
Figure 6.17 shows the dispersion curves for a rectangular dielectric waveguide with unity
aspect ratio, a/b = 1, and when the aspect ratio, a/b = 2. In the figure, P 2 = [(kz /k0 )2 − 1]/[(k1 /k0 )2 − 1],
and B = 2b/λ0 [(k1 /k0 )2 −1]1/2 . The unity aspect ratio causes some of the modes to be degen-
erate. In Figure 6.18, the intensity field plot is displayed for several modes of a rectangular
dielectric waveguide.
6.6.2 Variational Method

The harmonic expansion method does not work if the rectangular waveguide is a part of a
substrate. In such a case, a more versatile method like the variational method,4 or the finite
element method [28] should be adopted for an inhomogeneous waveguide.5 A variational
expression for the propagation constant of a waveguide mode can be derived from the vector
wave equation governing the field of the waveguide. We have shown in Chapter 3 that the
equations governing the electromagnetic field in an inhomogeneously filled waveguide are
µ∇s × µ−1 ∇s × Es − ∇s −1 ∇s · Es − k 2 Es + kz2 Es = 0, (6.6.18)
∇s × −1 ∇s × Hs − ∇s µ−1 ∇s · µHs − k 2 Hs + kz2 Hs = 0. (6.6.19)

We can dot multiply (6.6.18) by ẑ × Hs and integrate the resultant equation over the cross-
section of the waveguide to yield
Z Z
dS ẑ × Hs · µ∇s × µ−1 ∇s × Es − dS ẑ × Hs · ∇s −1 ∇s · Es
S S
Z Z (6.6.20)
2
− dSk ẑ · (Hs × Es ) + kz2 dS ẑ · (Hs × Es ) = 0.
S S
Using the identity that ∇s · (A × B) = B · ∇s × A − A · ∇s × B, we have
ẑ × Hs · µ∇s × µ−1 ∇s × Es = ∇s · [(µ−1 ∇s × Es ) × (ẑ × µHs )]

+ ∇s × (ẑ × µHs ) · µ−1 ∇s × Es
(6.6.21)
= ∇s · [(µ−1 ∇s × Es ) × (ẑ × µHs )]
+ (∇s · µHs )ẑ · µ−1 ∇s × Es .
4 This section follows the analysis in [27].
5 Alternatively, it can be solved by the numerical mode matching method [29].
Using ∇s · φA = (∇s φ) · A + φ∇s · A, we have

ẑ × Hs · ∇s −1 ∇s · Es = ∇s · [(ẑ × Hs )−1 ∇s · Es ] − ∇s · (ẑ × Hs )−1 ∇s · Es . (6.6.22)
Equations (6.6.21) and (6.6.22) can be used to simplify the first two integrals in (6.6.20).
After substituting them into (6.6.20), the divergence can be converted into a line integral
on C, the outermost domain of the waveguide. This integral becomes zero by virtue of the
boundary conditions on C, or when C → ∞. Consequently, Equation (6.6.20) becomes
Z Z
dS(∇s · µHs )ẑ · µ−1 ∇s × Es − dS ẑ · ∇s × Hs −1 ∇s · Es
S S
Z Z (6.6.23)
2
− dSk ẑ · (Hs × Es ) + kz2 dS ẑ · (Hs × Es ) = 0.
S
Applying the same operation to (6.6.19), or by duality, we have

Z Z
dS(∇s · Es )ẑ · −1 ∇s × Hs − dS ẑ · ∇s × Es µ−1 ∇s · µHs
S S
Z Z (6.6.24)
2
− dSk ẑ · (Es × Hs ) + kz2 dS ẑ · (Es × Hs ) = 0.
S
We note that Equations (6.6.23) and (6.6.24) are identical. If we write Equations (6.6.18)
and (6.6.19) as
Le · Es + kz2 Es = 0, (6.6.25)
Lh · Hs + kz2 Hs = 0 (6.6.26)
where Le and Lh are the differential operators in (6.6.18) and (6.6.19), then, Equations
(6.6.23) and (6.6.24) are the consequences of
hẑ × Hs , Le · Es i + kz2 hẑ × Hs , Es i = 0, (6.6.27)
hẑ × Es , Lh · Hs i + kz2 hẑ × Es , Hs i = 0. (6.6.28)
Hence (6.6.25) and (6.6.26) are transpose equation of each other. A variational expression for
kz2 is
hẑ × Hs , Le · Es i hẑ × Es , Lh · Hs i
kz2 = − =− . (6.6.29)
hẑ × Hs , Es i hẑ × Es , Hs i
R
In the above, hA, Bi = dSA · B. The above also imply that the (Hx , Hy ) formulation is
S
the same as the (Ex , Ey ) formulation.
We can take the first variation of (6.6.29) by letting Es = Ese + δEs , Hs = Hse + δHs ,
where Ese and Hse are the exact solutions. Then, after cross-multiplying the first equation
in (6.6.29) and taking its first variation, we have
hẑ × Hse , Le · Ese i + hẑ × δHs , Le · Ese i + hẑ × Hse , Le · δEs i
2
+ kze [hẑ × Hse , Ese i + hẑ × δHs , Ese i + hẑ × Hse , δEs i] (6.6.30)
+ δkz2 hẑ × Hse , Ese i + . . . = 0.
The leading order terms cancel as a consequence of (6.6.27). Similarly, as a consequence of

(6.6.25),
2
hẑ × δHs , Le · Ese i + kze hẑ × δHs , Ese i = 0. (6.6.31)
Because (6.6.23) and (6.6.24) are identical, we can show that
hẑ × A, Lh · Hs i = −hẑ × Hs , Le · Ai (6.6.32)
where A is an arbitrary vector satisfying the boundary conditions on C, or if it vanishes when

C → ∞. As a result,
2
hẑ × Hse , Le · δEs i + kze hẑ × Hse , δEs i = 0, (6.6.33)
and δkz2 = 0. Therefore, the first order variation in kz2 vanishes, implying the stationarity of
(6.6.29). The same thing can be shown for the other equation in (6.6.29).
Since (6.6.29) is variational, a Rayleigh-Ritz procedure can be adopted to obtain an opti-
mal solution for it. We let
N
X N
X
Es = an Ens , Hs = bm Hms . (6.6.34)
n=1 m=1
Substituting into (6.6.29), we have
P
N P
N
an bm hẑ × Hms , Le · Ens i
kz2 = − n=1Nm=1N , (6.6.35)
P P
an bm hẑ × Hms , Ens i
n=1 m=1
or
bt · A · a
kz2 = − (6.6.36)
bt · M · a
where the mn element of the matrices A and M are
Amn = hẑ × Hms , Le · Ens i, (6.6.37a)
Mmn = hẑ × Hms , Ens i. (6.6.37b)

The optimal values of a and b in (6.6.36) are obtained by requiring the first variation of kz2
to vanish, or that
bt0 · A · a0 + kz0
2
bt0 · M · a0 + δbt · A · a0 + kz0
2
δbt · M · a0
(6.6.38)
+ bt0 · A · δa + kz0
2
bt0 · M · δa = 0.
The leading order terms cancel each other by virtue of (6.6.36). The first order term vanishes
if
2
A · a0 + kz0 M · a0 = 0, (6.6.39a)
t 2 t
A · b0 + kz0 M · b0 = 0. (6.6.39b)
Equations (6.6.39a) and (6.6.39b) are matrix eigenvalue problems. They have the same set
of eigenvalues. Hence, we need only to solve one of them. If A and M are N × N matrices,
in general, there will be N eigenvalues and N eigenvectors a and b. It can be shown easily
that for two eigenvectors ai and bj corresponding to two distinct eigenvalues,
btj · M · ai = Di δij . (6.6.40)
In other words, they are M orthogonal.

Equations (6.6.39a) and (6.6.39b) are exactly the equations one would obtain if one applies
the Petrov-Galerkin method, or the method of weighted residuals to the differential Equations
(6.6.25) and (6.6.26) using Ens and Hns as expansion functions, respectively, and using
n̂ × Hms and n̂ × Ems as weighting functions, respectively. Equations (6.6.27) and (6.6.28)
are also the variational integrals to be used in the finite element method. In finite element, a
finite domain basis function is used. For example, the fields Ex and Ey can be written as a
linear superposition of pyramidal functions, with polygonal base. The pyramids overlap with
each other, and they form a piecewise linear approximation of the field between the nodal
values. Alternatively, edge elements can be used to model the electric field where tangential
components of the field are guaranteed to be continuous across edges [28]. Examples of finite
element grids and a pyramidal function are shown in Figure 6.19.
6.7 Discontinuities in Dielectric Waveguides

Discontinuities in dielectric waveguides have been studied by a number of workers [25, 29, 31–
36]. We have previously studied discontinuities in closed, hollow waveguides. There, we used
the method of mode matching to derive the solution of wave scattering by discontinuities. Be-
cause we have closed waveguides, only discrete modes exist [25]. However, when a waveguide
is open, the number of modes that a waveguide has is uncountably infinite. Furthermore,
there exists a set of modes which forms a continuum of modes. These modes carry energy
to infinity and hence, are called the radiation modes. We shall address the mode matching
problem for such an open waveguide. This problem is important, for instance, in ascertaining
reflection loss at the facet of a heterojunction laser [32, 36].
6.7.1 Reflection at a Laser Facet

Consider a laser facet as shown in Figure 6.20. This problem is important for the design of
lasers as the reflectivity at the facet of a laser cavity determines the Q of the laser cavity.
The reflectivity of such a facet can be found by mode-matching.
Consider a TE polarized mode with electric field polarized in the y direction. A part of
the energy of the mode will be transmitted yielding radiation modes for z > 0. For z < 0, the
mode will be reflected. Moreover, part of the energy of the reflected mode will be converted
into other reflected modes giving rise to “mode conversion,” just as discontinuities in uniform
waveguides (see Chapter 5).
Assume that the incident mode to be of the form
Eiy = am Ey (m, x)eikmz z (6.7.1)

where Ey (m, x) describes the transverse field distribution of the m-th mode, and kmz is its
corresponding wave number in the z direction. Then, the reflected modes can be expressed
as
∞
X
Ery = Rm0 m am Ey (m0 , x)e−ikm0 z z , (6.7.2)
m0 =1
where R m0 m is a reflection operator which is the unknown to be sought. Its off-diagonal
components account for the physics of mode conversion.
For z > 0, the field can be expressed as
Z∞
Ety = dkx eikx x+ikz z et (kx ), (6.7.3)
−∞
p
where kz = k 2 − kx2 . For a fixed x, the above is essentially a Fourier transform integral.
At this point, Rm0 m and et (kx ) are unknowns yet to be sought. They can be found by
matching boundary conditions at z = 0. Requiring that tangential E field be continuous at
z = 0, i.e., Eiy + Ery = Ety , we have
∞
X Z∞
0
Ey (m, x) + Rm0 m am Ey (m , x) = dkx eikx x et (kx ) (6.7.4)
m0 =1 −∞
The magnetic field is derived from Faraday’s law ∇ × E = iωµH. That is, the tangential
magnetic field is Hx and is derivable from
−1 ∂
Hx = Ey (6.7.5)
iωµ(x) ∂z
Consequently, in region I, we have
kmz
Hix + Hry = − am Ey (m, x)eikmz z
ωµ(x)
∞
X (6.7.6)
km0 z
+ Rm0 m am Ey (m0 , x)e−ikm0 z z ,
0
ωµ(x)
m =1
and in region II, we have

Z∞
kz ikx x+ikz z
Hty = − dkx e et (kx ). (6.7.7)
ωµ0
−∞
Matching boundary condition for magnetic field at z = 0, we have

∞
X
kmz km0 z
am Ey (m, x) − Rm0 m am Ey (m0 , x)
ωµ(x) 0
ωµ(x)
m =1
Z∞ (6.7.8)
kz ikx x
= dkx e et (kx ).
ωµ0
−∞
By Fourier inverse transforming (6.7.4),6 we obtain that

∞
X
Ẽy (m, kx ) + Rm0 m Ẽy (m0 , kx ) = et (kx ) (6.7.9)
m0 =1
where
Z∞
Ẽy (m, kx ) = dxe−ikx x Ey (m, x). (6.7.10)
−∞
The mode orthogonality relationship for an inhomogeneous waveguide is that

Z∞
− dxEy (n, x)Hx∗ (m, x) = Cn δnm (6.7.11)
−∞
Since Hx (m, x) = − kωµ

mz
Ey (m, x), this is equivalent to
Z∞ ∗
kmz
dx Ey (n, x)Ey∗ (m, x) = Cn δnm . (6.7.12)
ωµ(x)
−∞
For a lossless waveguide, µ is real, and we can normalize the modes such that
Z∞
Ey (n, x)Ey∗ (m, x)
dx = δnm . (6.7.13)
µ(x)
−∞
∗
Then, Cn in (6.7.12) is kmz /ω.
Multiplying (6.7.8) by Ey∗ (n, x) and integrating over x, we have
∞
X
am kmz δnm − Rm0 m am km0 z δnm0
m0 =1
 ∗
Z∞ Z∞
Ey (n, x)
= dkx kz et (kx )  dxe−ikx x 
(6.7.14)
µ0
−∞ −∞
Z∞
= dkx kz et (kx )ẽ∗y (n, kx )
−∞
where
Z∞
Ey (n, x)
ẽy (n, kx ) = dxe−ikx x . (6.7.15)
µ0
−∞
6 This is the same as testing the above equation with e−ikx x and integrate.
Using (6.7.9) for et (kx ) in (6.7.14), we have

Z∞
am kmz δnm − Rnm am knz = dkx kz Ẽy (m, kx )ẽ∗y (n, kx )
−∞
(6.7.16)
∞
X Z∞
+ Rm0 m am dkx kz Ẽy (m0 , kx )ẽ∗y (n, kx ).
m0 =1 −∞
The series summation in (6.7.16) can be truncated and (6.7.16) can then be solved as a matrix
equation for the unknown Rm0 m . By so doing, we obtain
Kz · a − Kz · R · a = A · a + A · R · a, (6.7.17)
where
Z∞
Anm = dkx kz Ẽy (m, kx )ẽ∗y (n, kx ).
−∞
Equation (6.7.17) can be solved easily for R · a or R.

The analysis above assumes that all the modes in the dielectric waveguide region are
discrete. In actual fact, continuum modes exist and the discrete summations in (6.7.2) will
have to be augmented by a continuous summation which is an integral. An analysis involving
such an integral is difficult and the continuum modes can be discretized by putting metallic
boundaries far away from the dielectric waveguide.
6.7.2 Determination of the Modes

The modes in the dielectric waveguide region are the natural solution of the wave equation [25]

∂ −1 ∂ 2 2
µr µr + k (x) − kz Ey = 0. (6.7.18)
∂x ∂x
Guidance Condition for TE m Modes Solution

4
εr = 2.56 Operating Frequency
f = 30 GHz
d = 0.25 in.
3.5
Normalized Attenuation Constant α 0x d

2 TE0
3
2.5 TE1
TE2
1.5
0.5
0
0.5 1 1.5 2 2.5 3 3.5 4
Normalized Transverse Wave Number k1x d
2
Guidance Condition for TE m Modes Solution
5 TE0 εr = 2.56
d = 0.25 in.
4.5
Normalized Attenuation Constant α 0x d
2
TE1
3.5
TE2
3
2.5
1.5
1
TE3
0.5 Operating Frequency

f = 60 GHz
0
0.5 1 1.5 2 2.5 3 3.5 4
4.5 5
Normalized Transverse Wave Number k1x d
2
Figure 6.4: Graphical solution of the guided modes of a dielectric slab waveguide.
Hy
TM0 mode
ω
x
d/2
0
Figure 6.5: For increasing frequency, a mode is better trapped inside the dielectric slab. Only
the field for x > 0 is sketched in the above.
Hy TM1 mode
ω = ωc
ω
x
0 d/2
Figure 6.6: The profile of the TM1 mode for different frequencies. When ω = ωc , the cutoff
frequency, the mode ceases to be evanescent outside the waveguide.
kz ω/c2
TM1 ω/c1
TM0
Figure 6.7: Dispersion of the modes in a dielectric slab waveguide.

2a
ε1, µ1 ε2, µ2
ε0, µ0
Core Cladding
b y
φ
x ρ
Figure 6.8: Geometry of an optical fiber—a circular dielectric waveguide.
TE modes
2
1.5
0.5
2.41 3.83 5.52 8.65
V
0
7.02 10.17 11.79

−0.5
TE
01
−1
−1.5
−2
0 2 4 6 8 10
k1ρ a
Figure 6.9: Graphical solution of the axially symmetric TE modes.

EH modes
256 2
1.5
0.5
3.83 5.13 7.02 10.17 11.61
V
0
8.47 13.32
−0.5
EH
11 EH
12
−1
−1.5
−2
0 2 4 6 8 10 12
k 1ρ a
Figure 6.10: Graphical solution of the EH modes.
HE modes
2
1.5
HE HE
0.5 12 13
HE
11 8.65 13.32
0
V
2.41
2.41 3.83 5.52
3.83 5.52 7.02 10.17 11.79
−0.5
−1
−1.5
−2
0 2 4 6 8 10 12
k1ρ a
Figure 6.11: Graphical solution of the HE modes.

Dispersion Curves of an Optical Fiber

ε 1 /ε 2 =1.0201 (n 1 /n 2 =1.01)
1.01
Relative Wave Number (k z /k 2 )

HE 21 , TE 01 , TM 01
1.008
HE11
1.006
HE 31, EH 11
1.004
HE 12
2a
1.002 ε1
ε2
1
0 1 2 3 4 5 6 7 8 9 10
Normalized Frequency (V= √ k12 -k 22 a)
Dispersion Curves of an Optical Fiber

ε 1 / ε 2 =2.25 (n 1 /n 2 =1.5)
1.5
1.45
Relative Wave Number (k z /k 2 )
TE 01
1.4
1.35 HE 11
HE 21
1.3 TM 01
1.25 HE 31
1.2
HE 12
1.15 2a
1.1
EH 11 ε1
ε2
1.05
1
0 1 2 3 4 5 6 7 8 9 10
Normalized Frequency (V= √ k -k a) 2

1
2
2
Figure 6.12: Dispersion curves for various modes of an optical fiber.

Figure 6.13: Dispersion curves for a weak contrast optical fiber [18].
Figure 6.14: Dispersion in a single-mode graded index optical fiber. (From W.A. Gambling
et al, “Zero total dispersion in graded-index single-mode fibers,” Electron. Lett., v. 15, p.
474, 1979.).
Figure 6.15: Different loss mechanisms in a germanosilicate single-mode optical fiber.
b/2
C ε0, µ0
ε1, µ1
x
–a/2 z 0 a/2
Region 1 Region 0
–b/2
Figure 6.16: The geometry of a rectangular dielectric waveguide.

Figure 6.17: Dispersion curves of different modes of a rectangular dielectric waveguide where
(1 /0 )1/2 = 1.5, and where (a) the aspect ratio a/b = 1, and (b) the aspect ratio a/b = 2 [24].
(a) (b) (c)
(d) (e) (f)
y
Figure 6.18: (Top row) Field intensity plots of the E11 for different degrees of mode confine-
2 2 2 y
ment: (a) P = 0.91, (b) P = 0.66, and (c) P = 0.08. (Bottom row) E12 for (d) P 2 = 0.85,
2 y 2
(e) P = 0.56, and E22 mode for (f) P = 0.66 (courtesy of K. Radhakrishnan [30]).
Figure 6.19: A basis function and a finite element grid.
x
ε0, µ0
ε0, µ0
ε1, µ1
ε2, µ2 z
y
I ε1, µ1 II
ε0, µ0
Dielectric Waveguide Free Space
Figure 6.20: A laser facet where a mode is reflected.

If the dielectric waveguide has a piecewise constant or step profile, these modes can be
found in closed form.
x
~
A0 R01 A0
z
µ0, ε0
d1
µ1, ε1
d2
µN–1, εN–1
dN–1
µN, εN
Figure 6.21: Waves in a layered medium.
For a particular mode, due to phase matching, the waves in all regions have eikz z depen-
dence. Hence, in the i-th layer,
Eiy = eiy (x)eikz z . (6.7.19)
Moreover, since each layer is homogeneous, eiy (x) is a linear superposition of upgoing and
downgoing waves. More specifically,
h i
eiy (x) = Ai e−ikix x + R̃i,i+1 e2ikix di +ikix x . (6.7.20)
The generalized Fresnel reflection coefficient can be found recursively via
Ri,i+1 + R̃i+1,i+2 e2iki+1,x (di+1 −di )

R̃i,i+1 = (6.7.21)
1 + Ri,i+1 R̃i+1,i+2 e2iki+1,x (di+1 −di )
where Rij is the local Fresnel reflection coefficient. The amplitudes Ai can be found via the
recursion relation
Ti−1,i Ai−1 eiki−1,z di−1
Ai eikiz di−1 = (6.7.22)
1 − Ri,i−1 R̃i,i+1 e2ikiz (di −di−1 )
A guided mode is defined as a solution to (6.7.18) without an external excitation. Hence,
it corresponds to a nonzero R̃01 even when A0 , the amplitude of the external excitation, is
zero. In other words, the guided modes by the layered region corresponds to the poles of R̃01 .
h i−1
Hence, they can be found by searching for the roots of R̃01 (kz ) .
The continuum modes are called radiation modes, so called because they carry energy to
infinity. There are two classes of radiation mode, one with real k0x , and the other with real
kN x . For the radiation modes with real k0x , their expression in region 0 is
h i
e0y (x) = A0 e−ik0x x + R̃01 e2ik0x d0 +ik0x x , (6.7.23)
2
where k0x + kz2 = k02 . Here, k0x has to be real in order for e0y (x) to be bounded when x → ∞.
Notice that in order for k0x to be real, kz lies along the locus as shown in the complex kz
plane. The field of the radiation mode in every layer can be found by using the recursion
relation as before.
Im[kz]
–k0
0
Re[kz]
k0
Figure 6.22: The distribution of kz corresponding to the radiation modes with real kox .
The other class of radiation modes with real kN x , has both upgoing and downgoing wave
in region N . Their expression in region N is
h i
eN y (x) = AN eikN x x + R̃N,N −1 e−2ikN x dN −1 −ikN x x (6.7.24)
Their field in every layer can be found by a similar recursive relation. Hence, in general, the
radiation modes of an open dielectric waveguide is expressible as
Z∞ Z∞
ikz z
Ey (x, y) = dk0x ey (x, k0x )e + dkN x ey (x, kN x )eikz z . (6.7.25)
0 0
6.8 Analyzing Weak Contrast Optical Fiber with WKB

method
When the profile of an optical fiber is slowly varying and the dimensions of the core is large
compared to wavelength, the WKB (Wentzel, Kramer and Brillouin) method can be applied
to the analysis of the optical fiber [1, 5]. The WKB method is often discussed in many books
on quantum mechanics, and also discussed in [25].
If φ in Equation (6.3.6) is written as
φ(r) = R(ρ)eikz z+inφ , (6.8.1)
then the equation for R(ρ) is
d2 1 d n2
R(ρ) + R(ρ) − R(ρ) + [k 2 (ρ) − kz2 ]R(ρ) = 0. (6.8.2)
dρ2 ρ dρ ρ2
By letting
√
R̂(ρ) = ρR(ρ), (6.8.3)
then

d2 R̂(ρ) d 1 −1 1
0
= ρ R(ρ) + ρ R (ρ)
2 2
dρ2 dρ 2 (6.8.4)
1 3 1 1
= − ρ− 2 R(ρ) + ρ− 2 R0 (ρ) + ρ 2 R00 (ρ).
4
1
Multiplying (6.8.2) by ρ 2 , and using (6.8.4) in the resultant (6.8.2), we have
d2 R̂(ρ)
+ [E − V (ρ)] R̂(ρ) = 0 (6.8.5)
dρ2
where
E = k 2 (∞) − kz2 ,

n2 − 1 (6.8.6)
V (ρ) = k 2 (∞) − k 2 (ρ) + 4
.
ρ2
Note that V (ρ) → 0, when ρ → ∞. Equation (6.8.5) is the same as the Schrödinger’s equation
for describing the motion of a particle in quantum mechanics. V (ρ) is the potential well, and
E is the energy of the particle. A particle is bound in the potential well when E < 0 , or
when kz is real andpthat kz > k(∞). This corresponds to a guided node in the optical fiber,
because kρ (∞) = k 2 (∞) − kz2 = iα(∞). When n = 0, the function V (ρ) is as shown in
Figure 1. When n is large, V (ρ) may not be negative at all, and no bound state or guided
mode can exist. When E > 0, kz < k(∞), the mode is not bound and it radiates energy to
infinity and becomes a radiation mode. This is because α(∞) is not real anymore.
6.8.1 The WKB Method

To analyze Equation (6.8.5) with the WKB method [25], we rewrite it as
R̂00 (ρ) + kρ2 (ρ)R̂(ρ) = 0 (6.8.7)
where kρ2 (ρ) = E − V (ρ). Note that kρ2 ∼ ω 2 when ω → ∞. Hence kρ2 becomes a large
parameter in the high frequency limit. By foreseeing that the solution of (6.8.7) may look
like a plane wave, we let
R̂(ρ) = Aeiωτ (ρ) . (6.8.8)
Then it follows that
R̂00 (ρ) = iωτ 00 (ρ) − [ωτ 0 (ρ)]2 Aeiωτ (ρ) , (6.8.9)
and (6.8.7) becomes
iωτ 00 (ρ) − [ωτ 0 (ρ)]2 + kρ2 (ρ) = 0. (6.8.10)
n=0 V(ρ) n=1
E for radiation mode

E for radiation mode
a ρ1
0 ρ 0 ρ
a
E for guided mode
E for guided mode
ρ1 ρ2
V(ρ)
(a) (b)
Figure 6.23: V (ρ) for n = 0 and n = 1.
Using the perturbation method, we expand τ (ρ) in a perturbation series, namely,

1
τ (ρ) = τ0 (ρ) + τ1 (ρ) + . . . , ω→∞ (6.8.11)
ω
where we have used 1/ω as the small parameter. Note that
iωτ000 (ρ) kρ2 (ρ), ω→∞ (6.8.12)
because kρ2 2
∼ ω . Consequently, using (6.8.11) in (6.8.10), and collecting leading order terms
when ω → ∞, we have
2
[ωτ00 (ρ)] = kρ2 (ρ) = ω 2 s2ρ (ρ) (6.8.13)
where we define sρ = kρ /ω to be the slowness of a wave. This is known as the eikonal
equation. Consequently, on solving (6.8.13), we arrive at
Zρ
τ0 (ρ) = ± dρ0 sρ (ρ0 ) + C0 . (6.8.14)
ρ0
Next, by collecting first order terms in (6.8.10) after substitution with (6.8.11), we have
iωτ000 (ρ) − 2ωτ00 (ρ)τ10 (ρ) = 0. (6.8.15)
This is known as the transport equation. Solving this yields
i i
τ1 (ρ) = ln τ00 (ρ) + C1 = ln sρ (ρ) + C1± . (6.8.16)
2 2
Consequently, using (6.8.14) and (6.8.16) in (6.8.11), we have

Zρ
i
τ (ρ) = ± dρ0 sρ (ρ0 ) + ln sρ (ρ) + C± , (6.8.17)
2ω
ρ0
or that the approximate solution to (6.8.5) is of the form

   
Zρ Zρ
A+ A −
R̂(ρ) ∼ √ exp iω sρ (ρ )dρ  + √ exp −iω sρ (ρ )dρ  .
0 0 0 0
(6.8.18)
sρ sρ
ρ0 ρ0
The first term in (6.8.18) corresponds to a right-going wave because its phase is increasing
with distance. By the same token, the second term in (6.8.18) is a left-going wave. Moreover,
the integral in the exponent elucidates the physical picture that the phase gained by a wave
gong from ρ0 to ρ is proportional to
Zρ
ω sρ (ρ0 )dρ0 (6.8.19)
ρ0
which is the integral summation of all the phases gained locally at ρ0 over the range from ρ0
to ρ. This physical picture is true only if the multiple reflections of the wave can be neglected
as it is propagating. Furthermore, Equation (6.8.12) shows that this physical picture, which
corresponds to the leading order solution, is correct only if
ωs0ρ (ρ) ω 2 s2ρ (ρ). (6.8.20)
Hence, this picture breaks down if the frequency is not high, or if sρ (ρ) ' 0. The factor of
√
1/ sρ in (6.8.18) is necessary for energy conservation. It is related to the wave impedance of
the wave, and hence, it alters the amplitude of the wave to conserve energy.
Note that when s2ρ < 0, corresponding to when E < V (ρ), or when the wave becomes
evanescent, the above analysis is still valid. Hence, the above analysis is valid for the field
of a guided mode in the region when E > V (ρ), and in the region where E < V (ρ), but not
in vicinity of the region where E = V (ρ). In the last case, sρ (ρ) ' 0. This happens, for
instance, in the guided mode case when ρ = ρ1 and ρ = ρ2 in Figure 6.23(b). In region II, we
have bouncing waves, and region I and III, the waves are evanescent. Hence, at ρ = ρ1 and
ρ = ρ2 , the waves are critically refracted, and they are also known as the turning points.
6.8.2 Solution in the Vicinity of a Turning Point

In the vicinity of ρ = ρ1 in Figure 1(a), kρ2 (ρ) can be approximated by a linear function, i.e.,
kρ2 ' ω 2 Ω(ρ1 − ρ), ρ → ρ1 , (6.8.21)
i.e., kρ2 is proportional to ω 2 . Hence, around ρ1 , Equation (6.8.7) becomes
R̂00 (ρ) + ω 2 Ω(ρ1 − ρ)R̂(ρ) = 0. (6.8.22)

2 1
Next by letting η = ω 3 Ω 3 (ρ1 − ρ), (6.8.22) becomes
2
d
+ η R̂(η) = 0, (6.8.23)
dη 2
which is the Airy equation. The general solution to the above equation is then
R̂(η) = C1 Ai (−η) + C2 Bi (−η) (6.8.24)

2 1
where Ai (−η) and Bi (−η) are special functions called the Airy functions. Since η = ω 3 Ω 3 (ρ1 −
ρ), η → ∞ when ρ ρ1 and η → −∞ when ρ ρ1 . The asymptotic expansions of Airy
functions can be used to approximate them when their arguments are large. Therefore, when
ρ ρ1 , η → −∞, and we have
1 −1 1 2
3
Ai (−η) ∼ π 2 (−η)− 4 e− 3 (−η) 2 , η → −∞, (6.8.25a)
2
3
1 1 2
Bi (−η) ∼ π − 2 (−η)− 4 e 3 (−η) 2 , η → −∞. (6.8.25b)
Ai (−η) corresponds to an exponentially decaying wave while Bi (−η) corresponds to an ex-
ponentially growing wave. Since we cannot have an exponentially growing wave to the right
of ρ1 , we must have C2 = 0. Hence, in the vicinity of the turning point ρ = ρ1 ,
R̂(η) = C1 Ai (−η). (6.8.26)
when ρ ρ1 , η → −∞, and we have

1 1 2 3 π
Ai (−η) ∼ π − 2 η − 4 sin η2 + , η → +∞, (6.8.27)
3 4
which corresponds to a standing wave resulting from a superposition of incident and reflected
waves on the left of the turning point.
6.8.3 Asymptotic Matching

The guidance conditions of the modes in Figure 6.23(b) can be found by asymptotic matching.
We shall illustrate asymptotic matching with the simpler case in Figure 6.23(a). In this
method, we seek the solutions in the region where 0 < ρ < ρ1 , ρ > ρ1 , and solutions in the
vicinity of ρ = 0 and ρ = ρ1 . For 0 < ρ < ρ1 and ρ > ρ1 , we can use the WKB solutions, while
for ρ in the vicinity of ρ = 0 and ρ = ρ1 , we need to use some special function solutions. Even
though the WKB solutions are not valid at ρ = 0 and ρ = ρ1 , but when ω → ∞, the WKB
solutions are valid in the vicinity of these points. By using the large argument expansions
of the special function solutions, overlapping regions of validity of the solutions exist, and
they can be matched to each other to find the unknowns and the guidance condition of the
waveguide.
The WKB solutions are also the geometrical optics solutions which are valid when the
frequency is high. Hence, they are similar to the ray-optics solutions. Ray optics solutions
break down at caustics where rays bunch together. It turns out that ρ = 0 and ρ = ρ1 are
caustics where rays bunch together. Hence, special function solutions are needed at these
points. It is worthwhile to notice that a ray undergoes a 90◦ phase shift at a caustic, and this
phenomenon will be observed in the later derivation.
The solution to the left of ρ = ρ1 is given by
Zρ Zρ
A+ 0 0A−
R̂(ρ) ∼ √ exp(iω sρ (ρ )dρ ) + √ exp(−iω sρ (ρ0 )dρ0 ) (6.8.28)
sρ sρ
0 0
When ρ → 0, assuming that k 2 (ρ) tends to a constant k 2 (0), then the solution
√
R̂(ρ) = A ρJ0 (ks ρ), ρ ' 0, (6.8.29)
p
where ks = k 2 (0) − kz2 . Hence ks is proportional to ω.
When the frequency is high, ks ρ 1 for ρ 6= 0, and (6.8.29) can be approximated by
r
2 π
R̂(ρ) ∼ A cos ks ρ − , k → ∞. (6.8.30)
πks 4
In the limit when ω → ∞, kρ2 (ρ) defined for (6.8.7) becomes
kρ2 ∼ k 2 (ρ) − kz2 , ω → ∞. (6.8.31)

This is even valid when ρ ' 0 as long as k 2 ρ2 → ∞. Hence,
kρ2 ' k 2 (0) − kz2 = ks2 , ω → ∞, ρ ' 0. (6.8.32)
In this limit, sρ = ks /ω. Then, Equation (6.8.28) in the vicinity of ρ ' 0 becomes
r r
ω iks ρ ω −iks ρ
R̂(ρ) ∼ A+ e + A− e . (6.8.33)
ks ks
Comparing (6.8.30) and (6.8.33), we require that
A π
A+ = √ e−i 4 , (6.8.34a)
2πω
A π
A− = √ e+i 4 . (6.8.34b)
2πω
The −90◦ phase shift between A− and A+ is reminiscent of an optical ray going through a
caustic. Consequently, (6.8.28) becomes
 ρ 
Z
A π
R̂(ρ) = p cos ω sρ (ρ0 )dρ0 −  . (6.8.35)
2πωsρ 4
0
In the vicinity of ρ = ρ1 , the phase integral can be approximated by

Zρ Zρ1 Zρ
0 0 0 0
sρ (ρ )dρ = sρ (ρ )dρ + sρ (ρ0 )dρ0 . (6.8.36)
0 0 ρ1
1 1
From (6.8.21), sρ (ρ0 ) ' Ω 2 (ρ1 − ρ) 2 when ρ ' ρ1 . Therefore,
Zρ (ρZ1 −ρ)
0 0 1 √ 2 1 3
sρ (ρ )dρ = Ω 2 xdx = Ω 2 (ρ1 − ρ) 2 , (6.8.37)
3
ρ1 0
and
A 2 1 3 π
R̂(ρ) ' 1 1 1 cos ωΩ (ρ1 − ρ) + φ −
2 2 , (6.8.38)
(2πω) 2 Ω 4 (ρ − ρ1 ) 4 3 4
where
Zρ1
φ=ω sρ (ρ0 )dρ0 . (6.8.39)
0
Using the definition of η in (6.8.26), and approximating Ai (−η) with (6.8.27), we have

1 2 1 3 π
R̂(ρ) ' C1 √ 1 1 1 sin ωΩ (ρ1 − ρ) +
2 2 , (6.8.40)
πω 6 Ω 12 (ρ1 − ρ) 4 3 4
1 3
when ωΩ 2 (ρ1 − ρ) 2 1.
Comparing (6.8.38) with (6.8.40), in order for the solution in region for which ρ < ρ1 and
the solution for which ρ ' ρ1 to agree with each other, we must have
(−1)m A
C1 = √ 1 1 , (6.8.41a)
2ω 3 Ω 6
Zρ1
φ=ω sρ (ρ0 )dρ0 = mπ, (6.8.41b)
0
where m = 1, 2, 3, . . . . Equation (6.8.41b) is the guidance condition for the n = 0 mode of

an optical fiber. The phase shift at the ρ = 0 caustic is −90◦ while the phase shift at the
ρ = ρ1 caustic is +90◦ . Hence, (6.8.41b) resembles the guidance condition for a parallel-
plate waveguide. Similar procedures can be used to find the guidance condition for the
n = 1, 2, 3, . . . modes. When n ∼ O(ka) then there will be two turning points both at ρ = ρ1
and ρ = ρ2 . The analysis will be slightly different from the above.
Notice that in the above, the WKB solution for ρ > ρ1 , was not used other than that
it is exponentially decaying. The reason is that if there is no turning point beyond ρ = ρ1 ,
the reflection of the bouncing waves in the optical fiber is determined by the turning point
at ρ = ρ1 only. A WKB solution presents a physical picture of a wave propagating without
reflection. Hence, a turning point is the only place where a wave is reflected. To obtain the
magnitude of the evanescent wave for ρ > ρ1 , asymptotic matching can be used.
6.9 Effective Index Method

In integrated optics waveguides, the refractive index profile is often obtained by doping.
Hence, the variation of the refractive index is weak. In such a case, an approximate method
n0
x
z n1
n2
Figure 6.24: A typical integrated optics waveguide cross-section.
called the effective index method may be used to analyze the waveguiding structure. The
method was proposed by Knox and Toulios [37], and has been used by many workers [38,
39]. Assuming that µr = 1, the equations governing the electromagnetic field in such an
inhomogeneous structure are exactly given by
∇2 E + ∇(∇ ln r · E) + k02 r E = 0, (6.9.1)

2
∇ H + ∇ ln r × ∇ × H + k02 r H = 0. (6.9.2)
6.9.1 Effective Index Concept

When a mode is propagating in a waveguide, say the optical fiber, with eikz z dependance, we
can define an effective index ne such that
kz = k0 ne (6.9.3)
The effective index is the propagation constant the wave will have if it is propagating in a
homogeneous medium with this index. This concept can be extended to other structures,
including a wave propagating in a slab or layered waveguide
Consider the geometry shown in Figure 6.9.1 where is a function of y only. Then if we
consider TE wave, which can be characterized by Hy , the governing equation is
∇2 Hy + k02 r (y)Hy = 0 (6.9.4)
The above can be solved by the separation of variables, by letting
Hy (x, y, z) = Y (y)h(x, z) (6.9.5)

2
∂2 ∂ ∂2
h(x, z) 2 Y (y) + Y (y) + 2 h(x, z) + k02 r Y (y)h(x, z) = 0 (6.9.6)
∂y ∂x2 ∂z
Figure 6.25: A layered medium waveguide.
Dividing by Y (y)h(x, z), we have

∂2 ∂2 ∂2
Y −1 Y + h−1 + h + k02 r = 0 (6.9.7)
∂y 2 ∂x2 ∂z 2
The above can be satisfied by letting
2
∂ ∂2
+ 2 h(x, z) = −k02 n2 h(x, z) (6.9.8)
∂x2 ∂z
∂2
Y (y) + k0 2 r (y)Y (y) = k02 n2 Y (y) (6.9.9)
∂y 2
where k02 n2 is the separation constant.
Equation (6.9.9) is an eigenvalue problem with eigenvalue k02 n2 and eigenfunction Y (y).
Here, Y (y) corresponding to guided mode or eigenmode in the layered medium. Equation
(6.9.8) represents a wave traveling in the (x, z) direction with effective index n, which is a
constant independent of (x, z).
6.9.2 Quasi-TE polarization

If we assume that E is polarized predominantly in the xy direction, and that r (x, y, z) is a
slowly varying function of x and z, then, ∇ ln r is predominantly ŷ directed. Extracting the
y component of (6.9.2) to characterize a TE to y wave, we have
∇2 Hy + k02 r Hy ≈ 0 (6.9.10)
Next, we assume that

Hy = Y (x, y, z)h(x, z), (6.9.11)
where ∂Y /∂x ' 0, ∂Y /∂z ' 0. In other words, Y (x, y, z) is a slowly varying function of x and
z. Equation (6.9.11) is motivated by the separation of variables as described in the previous
section. As noted before, the problem is completely separable if r is a function of y only.
But here, we assume that the problem is approximately separable.
Then, 2
∂2Y ∂ h ∂2h
∇ 2 Hy ' h 2 + Y + . (6.9.12)
∂y ∂x2 ∂z 2

∂2 ∂2 ∂2
h(x, z) 2 Y (y) + Y (y) + h(x, z) + k02 r Y (y)h(x, z) ≈ 0 (6.9.13)
∂y ∂x2 ∂z 2
Again, motivated by the separation of variables, we let
2
∂ ∂2
+ h(x, z) + k02 n2 (x, z)h(x, z) = 0 (6.9.14)
∂x2 ∂z 2
where n(x, z) is an effective index that is a slowly varying function of x and z only. Further-
more, we require that
∂2
Y (x, y, z) + k02 r (x, y, z)Y (x, y, z) = k02 n2 (x, z)Y (x, y, z). (6.9.15)
∂y 2
For every fixed x and z, the above is a one-dimensional eigenequation with eigenvalue k02 n2 (x, z).
In principle, there are infinitely many eigenvalues and eigenfunctions associated with (6.9.15).
The solutions of equation (6.9.15) are the modes of the structure. We can assume that only
the fundamental mode is important. If r (x, y, z) is piecewise constant in y, then transcen-
dental equations can be derived to yield k02 n2 (x, z). Hence, if r (x, y, z) is a slowly varying
function of x and z, Y (x, y, z) is in fact a slowly varying function of x and z validating the
assumption (6.9.11) and the self-consistency of the method.
6.9.3 Quasi-TM Polarization

In this case, we assume that the magnetic field is predominantly polarized in the xz plane.
Such a wave field can be characterized by the Ey component of the electric field. Extracting
the y component of (6.9.1), we have

∂ ∂
∇2 Ey + ln r Ey + k02 r Ey ≈ 0. (6.9.16)
∂y ∂y
In arriving at the approximate equation above, we assume that
∂
ln r ∇s ln r (6.9.17)
∂y
∂ ∂
where ∇s = x̂ ∂x + ẑ ∂z . In the above, (6.9.16) can be rewritten as
2
∂ ∂2 ∂ −1 ∂ 2
+ 2 + r r + ko r r Ey ≈ 0
∂x2 ∂z ∂y ∂y
to resemble that in [40, (2.1.7)].

Again, if we assume that
Ey = Y (x, y, z)e(x, z), (6.9.18)
where ∂Y /∂x ' 0 and ∂Y /∂z ' 0, then, motivated by the separation of variables, we let

∂2Y ∂ ∂ 2 2 2
+ ln r Y + k0 r Y ≈ k0 n Y. (6.9.19)
∂y 2 ∂y ∂y
The presence of the extra term in the equation for TM polarization compared to that for the
TE polarization is because in the TM polarization, the electric field has a component normal
to the interface that induces polarization charges at the interface.
The above could be rewritten as
∂ 1 ∂
r Y + k02 r Y ≈ k02 n2 Y. (6.9.20)
∂y r ∂y
For a fixed x and z, it is a one dimensional eigen-equation for the propagation of TM eigen-
modes in a layered medium. Here, k02 n2 is the eigenvalue of the problem.
Consequently, the equation governing e(x, z) is
2
∂ ∂2
+ 2 e(x, z) + k02 n2 (x, z)e(x, z) = 0, (6.9.21)
∂x2 ∂z
where n(x, z) is the effective index obtained by solving (6.9.20). Again, (6.9.21) is now reduced
to a two-dimensional equation.
The effective index method is equivalent to replacing the y variation of the field with
a single-mode approximation, and the propagation of this single mode in the xz plane is
governed by the two-dimensional equations (6.9.14) and (6.9.21), for different polarizations.
6.10 The Beam-Propagation Method

When the inhomogeneity of a waveguide is weakly varying, an efficient method of deriving a
solution is to use the beam-propagation method was proposed by Fleck, Morris, and Feit [41]
first for atmospheric wave propagation where the refractive index is often tenous. Later, it
was adapted for optical waveguide analysis [42,43]. Consider a scalar wave equation governed
by
2 ∂2 2 2
∇s + 2 + k0 n (x, y, z) φ(x, y, z) = 0, (6.10.1)
∂z
2 2
∂ ∂
where ∇2s = ∂x 2 + ∂y 2 . Assuming that n(x, y, z) is a weak function of space, then the above
equation can be factorized as

q q
∂ ∂
+ i k02 n2 + ∇2s − i k02 n2 + ∇2s φ(r) ≈ 0. (6.10.2)
∂z ∂z
Therefore, a solution to
q
∂
φ(r) = i k02 n2 + ∇2s φ(r) = i℘φ(r) (6.10.3)
∂z
p
is also an approximate solution to (6.10.1). In the above, ℘ = k02 n2 + ∇2s is to be interpreted
as an operator.
Here, (6.10.3) is also a one-way wave equation as it describes the propagation of the
wave in one direction only. Hence, multiply reflected wave is not accounted for in the above
approximation. This is a good approximation for weakly varying inhomogeneity. In arriving
at (6.10.2), we have assumed that the ∇ or the space derivative operator commutes with the
inhomogeneity n(x, y, z). This is only approximately true for slowly varying inhomogeneity.
Equation (6.10.3) is not any easier to solve compared to (6.10.1). To simplify it, we need
to make a paraxial approximation. This assumes that
||∇2s φ|| ||k02 n2 φ||, (6.10.4)
or that the transverse variation of φ is much smaller than its longitudinal variation. In
other words, the wave is propagating almost parallel to the axis of the waveguide. If we let
n = n0 + δn, then
q
k 2 n0 δn
℘∼= k02 n20 + ∇2s 1 + 2 02 + . . . . (6.10.5)
k0 n0 + ∇2s
With the assumption (6.10.4), we can approximate (6.10.5) as
q
℘∼= k02 n20 + ∇2s + k0 δn. (6.10.6)
The approximation (6.10.6) is judiciously tailored so that the first term is independent of
space, and the second term is independent of the operator ∇s .
Since ℘ is a function of z, Equation (6.10.3) still cannot be solved easily. However, if ℘ is
assumed to be independent of z within a small ∆z, then, we can write the solution to (6.10.3)
as
φ(x, y, z + ∆z) = ei℘∆z φ(x, y, z)
√ 2 2 2
∼
= eik0 δn∆z+i k0 n0 +∇s ∆z φ(x, y, z) (6.10.7)
√ 2 2 2
∼
= eik0 δn∆z ei k0 n0 +∇s ∆z φ(x, y, z).
Now, we can write ZZ ∞

1
φ(x, y, z) = dks eiks ·rs φ̃(ks , z) (6.10.8)
(2π)2 −∞
where ks = x̂kx + ŷky , rs = x̂x + ŷy. Substituting (6.10.8) into (6.10.7), we have
ZZ ∞ √ 2 2 2
1 ik0 δn∆z
φ(x, y, z + ∆z) = 2
e dks eiks ·rs ei k0 n0 −ks ∆z φ̃(ks , z). (6.10.9)
(2π) −∞
Equation (6.10.9) is the fundamental equation of the beam-propagation method. To imple-

ment it, one first takes the field φ(x, y, z) at a z = constant√plane and Fourier transform it.
2 2 2
Then one multiplies the result by a plane-wave propagator ei k0 n0 −ks ∆z in the Fourier space.
Next, a Fourier inverse transform is performed on the propagated result. Subsequently, the
field at each (x, y) location is added a phase of k0 δn∆z to yield the field at φ(x, y, z + ∆z).
Alternatively, one can write

q
ks2
k02 n20 − ks2 = k0 n0 − p , (6.10.10)
k02 n20 − ks2 + k0 n0
where the second term is much smaller than the first term if the wave is paraxial. Therefore,
we can let
φ(x, y, z) = w(x, y, z)eik0 n0 z , (6.10.11)
and the beam-propagation equation for w(x, y, z) is then
ZZ ∞
1 ik0 δn∆z
w(x, y, z + ∆z) = e dks eiks ·rs eip(ks )∆z φ̃(ks , z), (6.10.12)
(2π)2 −∞
where
ks2
p(ks ) = p . (6.10.13)
k02 n20 − ks2 + k0 n0
lens or phase screen
∆z ∆z ∆z ∆z
z
z=0
Figure 6.26: The physical interpretation of the beam-propagation method.
The Fourier transform and inverse transform in the beam-propagation method can be im-
plemented efficiently using Fast Fourier Transform (FFT) which requires O(N log N ) floating-
point operations. Hence, albeit approximate, the beam-propagation method can be imple-
mented efficiently. As it solves a first order equation (6.10.3), it only accounts for the forward
component of the wave, but no reflections.
The physical interpretation of (6.10.9) or (6.10.12) is that the wave is first propagated
through a homogeneous space using a homogeneous-space propagator. Then, the wave is
passed through a lens or a phase screen which alters the phase of the wave at each (x, y)
position. Therefore, Equation (6.10.9) or (6.10.12) represents physically the concatenation of
a series of lenses or phase screens in a homogeneous space.
6.11 Ray Tracing Method

Multimode fibers correspond to the case where the core size is much larger than the wave-
length. Hence, the fiber can be engineered by assuming that the light is a ray that bounces
around in the fiber. Hence, ray-tracing method can be used to solve the optical fiber problem.
This method is described by Okoshi [5] and Senior [1] and many references therein.
The vector wave equation for an inhomogeneous medium can be written as
∇2 E − ∇(∇ ln r · E) + k02 r E = 0. (6.11.1)
The above equation can be approximated when the frequency is high. When the frequency is
high so that the wavelength of the wave is much smaller than the lengthscale of the variation
of the inhomogeneity, we can assume that E(r) resembles a plane wave locally. In other
words, E(r) could be more aptly described with a ray picture. In this case, we let
E(r) = e(r)eik0 φ(r) , (6.11.2)
where φ(r) and e(r) are slowly varying but eik0 φ(r) is rapidly varying when k0 → ∞. There-
fore,
|∇E(r)| ∼ |k0 E(r)|, |∇2 E(r) ∼ |k02 E(r)|, (6.11.3)
when k0 → ∞. Hence, Equation (6.11.1) can be approximated by
∇2 E(r) + k02 r E(r) = 0. (6.11.4)
Hence, we are making the assertion that the polarization term in (1) is unimportant in the
high-frequency limit if the postulated form for E(r) given by (6.11.2) is true.
Taking the Laplacian of E, we have
∇2 E(r) ' e(r)∇2 eik0 φ(r) = −k02 (∇φ)2 E(r), (6.11.5)
after ignoring higher order terms. Using (6.11.5) in (6.11.4), we have
(∇φ)2 = r = n2 (6.11.6)
or
|∇φ| = n. (6.11.7)
∇φ is the direction in which the phase in the wave in (6.11.2) is varying most rapidly. Hence,
it is the direction of the wave in (6.11.2), and also the direction at which the ray is pointing.
A unit vector along the ray direction is given by
ŝ = ∇φ/|∇φ| = ∇φ/n. (6.11.8)
Defining a point in a ray by the position vector r, and that the ray is parametrized by the
variable s which is the path length along a ray, then ŝ = dr/ds, and (6.11.8) becomes
dr
nŝ = n = ∇φ. (6.11.9)
ds
Also, taking the gradient of (6.11.6), we obtain
2(∇∇φ) · ∇φ = 2n∇n. (6.11.10)
After making use of (6.11.9), we have
(∇∇φ) · ŝ = ∇n. (6.11.11)
Furthermore,

d d dx ∂ dy ∂ dz ∂
(nŝ) = (∇φ) = + + ∇φ
ds ds ds ∂s ds ∂y ds ∂z
(6.11.12)
dr
= · ∇∇φ = ŝ · ∇∇φ.
ds
But ŝ · ∇∇φ = (∇∇φ) · ŝ. Hence, from (6.11.11) and (6.11.12), we have

d d dr
(nŝ) = ∇n, n = ∇n, (6.11.13)
ds ds ds
which are the fundamental equations of ray tracing.

In addition to the above equations, if we assume that
H(r) = h(r)eik0 φ(r) , (6.11.14)
it can be shown easily from Maxwell’s equations that when k0 → ∞,
∇φ × h(r) ' ce(r), (6.11.15a)
∇φ × e(r) ' −cµh. (6.11.15b)

Hence,
∇φ · e ' ∇φ · h ' 0. (6.11.16)
The above equations indicate that the wave is locally a plane wave in the high frequency
limit.
6.11.1 Ray Tracing Equations in an Optical Fiber

In an optical fiber with axial symmetry and uniformity in the z direction, we can assume that
dn/dφ = 0, and dn/dz = 0. A point in a ray can be described by
r = ρ̂ρ + ẑz, (6.11.17)
in cylindrical coordinates. In the above, ρ̂ is a function of φ. Hence, r is a function of (ρ, φ, z).

Extracting the ρ̂ component of (6.11.13), we have

d dr d
ρ̂ · n = n. (6.11.18)
ds ds dρ
But
dr d dρ dρ̂ dz
= (ρ̂ρ + ẑz) = ρ̂ + ρ + ẑ , (6.11.19)
ds ds ds ds ds
and
dρ̂ d dφ
= (x̂ cos φ + ŷ sin φ) = −φ̂ . (6.11.20)
ds ds ds
Therefore,
d dr d dρ dφ
ρ̂ · n = ρ̂ · nρ̂ − nφ̂ρ . (6.11.21)
ds ds ds ds ds
But
dφ̂ d dφ
= (x̂ sin φ − ŷ cos φ) = ρ̂ . (6.11.22)
ds ds ds
Consequently,
2
d dr d dρ dφ d
ρ̂ · n = n − nρ = n. (6.11.23)
ds ds ds ds ds dρ
Similarly, we can extract the φ̂ component of (6.11.13) to obtain

d dr d dρ dφ
φ̂ · n = φ̂ · nρ̂ − nφ̂ρ
ds ds ds ds ds
(6.11.24)
dρ dφ d dφ
= −n − nρ = 0.
ds ds ds ds
Extracting the ẑ component of (6.11.13) yields

d dz
n = 0. (6.11.25)
ds ds
Equation (6.11.25) can be readily integrated to yield

n
ds = dz. (6.11.26)
n0 cos θi
where n0 is the value of n at the initial point, and θi is the angle of the ray with the z-axis
initially. Replacing n/ds in (6.11.24) with (6.11.26), we obtain

dρ dφ d dφ
+ ρ = 0, (6.11.27)
ds dz ds dz
which can be rewritten as

d 2 dφ
ρ = 0. (6.11.28)
ds dz
Equation (6.11.28) can be integrated to yield
dφ
ρ2 = C0 . (6.11.29)
dz
To integrate (6.11.23), we multiply it by n and use (6.11.26) to replace n/ds with n0 cos θi /dz
to obtain 2
d2 dφ 1 d 2
ρ − ρ = 2 n . (6.11.30)
dz 2 dz 2n0 cos2 θi dρ
dφ
Using (6.11.29) for dz , we obtain
d2 1 1 d 2
ρ = 3 C02 + 2 n . (6.11.31)
dz 2 ρ 2n0 cos2 θi dρ
The above equation can be integrated with respect to ρ. The left-hand side is
Zρ Zz Zz 2
d2 0 0 d2 ρ0 dρ0 0 1 d dρ0
ρ dρ = dz = dz 0
dz 02 02
dz dz 0 2 dz 0 dz 0
ρ0 0 0 (6.11.32)
2
1 dρ 1
= − D0 .
2 dz 2
2
dρ
where D0 = dz . The right hand side of (6.11.31) yields
z=0
ρ
1 2 1 2

− 2 C0 + 2 n
2ρ 2n0 cos2 θi ρ0
" 2 #
ρ0 C02 1 n2
= 1− + − 1 . (6.11.33)
ρ 2ρ20 2 cos2 θi n20
As a result, we have
2 " 2 #
dρ ρ0 C02 1 n2
= 1− + − 1 + D0 , (6.11.34)
dz ρ ρ20 cos2 θi n20
or
Zρ (" 2 # 2 )− 21
ρ0 C0 1 n2
z= dρ 1− + − 1 + D0 . (6.11.35)
ρ ρ0 cos2 θi n20
ρ0
Equation (6.11.35) is the basic equation for computing the ray path in an optical fiber, given
the initial condition C0 , D0 , cos θi , n0 and ρ0 at z = 0.
6.11.2 Determination of Initial Conditions

Let us assume that an optical ray enters an optical fiber at z = 0 at (x0 , y0 ) and the direction
of the ray is pointed at ŝ0 , where
ŝ0 = x̂ sin θi cos φi + ŷ sin θi sin φi + ẑ cos θi . (6.11.36)

If dφ/dz at (x0 , y0 , z = 0) of the ray can be found, then C0 in (6.11.29) can be found. The
φ̂0 component of ŝ0 can be found by ŝ0 · φ̂0 which is
ŝ0 · φ̂0 = ŝ0 · [−x̂ sin φ0 + ŷ cos φ0 ]

(6.11.37)
= − sin φ0 sin θi cos φi + cos φ0 sin θi sin φi .
Hence,
ρ0 dφ = ŝ0 · φ̂0 ds = ŝ0 · φ̂0 / cos θi dz. (6.11.38)
Therefore,
ŝ0 · φ̂0
C 0 = ρ0 . (6.11.39)
cos θi
If d ρ dz of the ray at (x0 , y0 , z = 0) is known, then D0 in (6.11.32) is known. But
ŝ0 · ρ̂0
dρ = ŝ0 · ρ̂0 ds = dz. (6.11.40)
cos θi
Therefore,
2
dρ (ŝ0 · ρ̂0 )2 sin2 θi (cos φi cos φ0 + sin φi sin φ0 )2
D0 = = =
dz z=0 cos2 θi cos2 θi (6.11.41)
2 2
= tan θi cos (φi − φ0 )
The above provides sufficient initial conditions to launch a ray pointing at ŝ0 from the
point (x0 , y0 , z = 0). A ray which is only propagating radially, i.e., ŝ0 · φ̂0 = C0 = 0, is known
as a meridional ray. A ray that propagates in both the ρ̂ and φ̂ directions is known as a skew
ray. A skew ray that propagates at a constant distance from the fiber axis is called a helical
ray.
The ray equation (6.11.35) can be used to derive index profile so that the axial velocity
of a ray is independent of the launch condition. In this manner, the dispersion of the fiber
will be minimized.
Problem 6-1:
Show that the guidance condition for a dielectric slab of thickness d, permittivity 1 and
permeability µ1 suspended in air can be simplified to

d µ0 d k1x d − mπ
α0x = k1x tan ,
2 µ1 2 2
p p
for the TE case. In the above, α0x = kz2 − k02 , k1x = k12 − kz2 and m is an integer. Plot
the right and the left hand side of the equation as a function of k1x d for m even and m odd
to obtain graphical solutions to the above equation.
Problem 6-2: Find the phase velocity of the TM1 mode of a symmetric dielectric slab
waveguide at cutoff. By analyzing the phase velocity in the vicinity of cutoff, find the group
velocity analytically. Explain the answers.
Problem 6-3: For a circular dielectric waveguide of radius a, find the cutoff frequencies of
the TE01 mode, the EH11 mode and the HE12 mode. Which is the next higher order mode to
the HE11 mode? If a = 10µ, n1 = 1.6 and n2 = 1.5, what is the bandwidth for single mode
propagation in the optical fiber? Explain why the usable bandwidth of an optical fiber is not
this bandwidth.
Problem 6-4: Find the phase velocity of the TE01 mode of an optical fiber near cutoff. Also,
find the group velocity near cutoff analytically. Explain what you have observed about the
answer.
Problem 6-5:
(a) Show that the ratio of Ez to Hz in the core region of an optical fiber is given by

−nkz 1 1
H1 /E1 = +
iω (k1ρ a)2 (α2 a)2
−1
µ1 Jn0 (k1ρ a) µ2 Kn0 (α2 a)
+
(b) Using the equation for the guidance conditions of the EH and HE modes, show that the
ratio E1 /H1 is in fact larger for the HE modes compared to the EH modes.
Problem 6-6:
(a) Starting with the equation for the guidance condition of the modes in a step-index
fiber (nonweakly guiding), by assuming that µ1 = µ2 and that 1 ≈ 2 , show that the
guidance condition for the weak contrast optical fiber mode can be derived as
k1ρ Jn±1 (k1ρ a) α2 Kn±1 (α2 a)
=±
Show that the above two equations are equivalent, and hence the modes they define are
degenerate.
(b) From the derivation, which of the EH mode is degenerate with the HE mode when the
contrast of the fiber tends to zero?
(c) Even though in the weak contrast fiber approximation, the equation indicates that the
x component and the y component of the electric field are decoupled, they are actually
weakly coupled in this limit. Hence, it is not possible to conceive some modes to have
Ex only or Ey only. The LP11 mode is such a mode. To test your physical insight,
sketch the electric field on the xy plane on the cross section of a weak contrast optical
fiber.
Problem 6-7: Write a computer program to solve for the kz of the HE11 , HE21 , TE01 ,
TM01 , EH11 , HE31 and HE12 modes. That is, produce the dispersion curve for the first three
families of modes shown in Figure 6.2.5 of the text. (a) First, generate the dispersion curves
when n1 /n2 = 1.5. (b) Second, generate the dispersion curves when n1 /n2 = 1.01.
Problem 6-8: Using 1/ω as a small parameter, expand Es and Ez as perturbation series in
a weak contrast optical fiber, and show that Ez ∼ |Es |/ω.
Problem 6-9: Explain why the Rayleigh scattering loss and ultraviolet absorption loss
diminish with wavelength in Figure 6.15, while the infrared absorption loss increases with
wavelength. Is it reasonable to assume that waveguide imperfection loss does not alter with
wavelength?
Problem 6-10: For the harmonic expansion method, why is the assumption of standing
wave inside the waveguide and outgoing wave outside the waveguide not a valid assumption
in Subsection 6.6.1?
Problem 6-11:
A circular waveguide is loaded with a circular dielectric rod of diameter d as shown in the
above figure. Find the change in the propagation constant kz of the TE11 mode due to the
presence of this rod using a perturbation calculation.
Problem 6-12: The vector wave equation governing the propagation of waves in an anisotropic
medium with reflection symmetry can be shown to be
ẑ × µs · ẑ × ∇s × µ−1 −1 2 2
zz ∇s × Es + ∇s zz ∇s · s · Es − ω ẑ × µs · ẑ × s · Es − kz Es = 0.
(a) Derive a variational expression for the guided wave number kz2 .
(b) Using Rayleigh-Ritz procedure, derive a matrix equation for the guided wave numbers
of the waveguide.
Problem 6-13:
The harmonic expansion method, in theory, can be used for dielectric waveguides of any
shapes. However, when the waveguide has symmetry about the x and the y axes as shown in
the above figure, the z components of the fields are either even or odd about the the x and
the y axes.
(a) Proof that if Ez (Hz ) is even about the x or the y axes, Hz (Ez ) has to be odd about
the x or the y axes, and vice versa.
(b) Because of this symmetry, we need not have to solve for the solution of the waveguide
in the full space, but only in a quadrant of the full space. Use the harmonic expansion
method, together with point matching, derive the guidance conditions for the modes
for which Ez is even about the x axis and odd about the y axis. Give a reason why this
method of solution is preferable.
Problem 6-14: Assume a parallel waveguide terminated abruptly so that the modes of
the waveguide will radiate into free space. Repeat the derivation of Subsection 6.7.1 for the
terminated parallel plate waveguide with a flange as shown.
Problem 6-15:
√
(a) In the WKB method, explain why the prefactor of 1/ sρ term is necessary for energy
conservation.
(b) Repeat the analysis of Subsection 6.8.1 for a planar geometry. Is the result of Equation
(6.8.18) much different compared to the planar geometry case? Explain why.
Problem 6-16:
A planar waveguide has a index profile as shown. It is terminated by a perfect electric
conductor at the x = 0 surface. Use the WKB method, write down the guidance condition
for such a waveguide.
Problem 6-17:
(a) If r (z) in Equation (6.9.15) of Subsection 6.9.1 for fixed x, y is described by a symmetric
dielectric slab. Find the transcendental equation from which the eigenvalues of Equation
(6.9.15) can be found.
(b) Repeat the same for Equation (6.9.20) of Subsection 6.9.2.
Problem 6-18: In the beam-propagation method, the equation of propagation is sometimes
written as √ 2 2 2
φ(x, y, z + ∆z) ∼
= eik0 δn∆z/2 ei k0 n0 +∇s ∆z eik0 δn∆z/2 φ(x, y, z).
as opposed to Equation (6.10.7) of Section 10. Explain if there is any advantage of writing
the propagation equation as shown above.
Problem 6-19: Describe how you would solve the ray-tracing equations (6.11.13) given in
Section 6.11 numerically.
Problem 6-20: Write a computer program to compute Equation (6.11.35) of Subsection
6.11.1 for ray tracing in an optical fiber.
−40 −40
−30 −30
−20 −20
−10 −10
0 0
10 10
20 20
30 30
40 40
0 50 100 150 200 250 300 350 400 450 500 0 50 100 150 200 250 300 350 400 450 500
−40 −40
−30 −30
−20 −20
−10 −10
0 0
10 10
20 20
30 30
40 40
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
−40 −40
−30 −30
−20 −20
−10 −10
0 0
10 10
20 20
30 30
40 40
0 500 1000 1500 0 500 1000 1500
Figure 6.27: Examples of numerical simulations using the beam-propagation method for
different kinds of optical waveguides. Some waveguide transitions give rise to much radiation
loss, while some do not. Also, depending on the mode profile, the radiation at the waveguide
transition is different. For instance, in the top figure, the fundamental mode (top left) couples
more smoothly than the higher-order mode (top right) with two minima (Courtesy of F.
Teixeira).
ρ0
y
ˆs0
φ0
(x0, y0)
x
Figure 6.28: Launching of a skew ray pointing at ŝ0 at (x0 , y0 , z = 0).
(a) (b) (c)
Figure 6.29: Cross-sectional pictures of (a) a meridional ray, (b) a complex skew ray, and (c)
a skew ray that is helical.
d µ0, ε0
µ1, ε1
z
µ0, ε0
µ0 , ε 0
µ0 , ε 1
b
d
ε0
ε1 ε1
x
ε1 ε1

εr
ε1
εs
ε0
x
0

Bibliography
[1] J.M. Senior, Optical Fiber Communications, Principles and Practice, 2nd Edition, Pub-
lisher: Prentice Hall International, Hertfordshire, UK, 1992.
[2] G.P. Agrawal, Fiber-Optic Communication Systems, Wiley-Interscience, 3rd edition, 2002.
[3] P.K. Runge, “Undersea lightwave systems,” AT&T Technical Journal, vol. 71, no. 1, pp.
5-13, 1992.
[4] K.C. Kao and G.A. Hockham, “Dielectric-fiber surface waveguides for optical frequencies,”
Proc. IEE (London), 113, pp. 1151-1158, 1966.
[5] T. Okoshi, Opitcal Fibers, Academic Press, New York, 1982.
[6] J.E. Midwinter, Optical Fibers for Transmission, John Wiley & Sons, NY, 1979.
[7] D. Marcuse, Ed., Integrated Optics, IEEE Press, NY, 1973.
[8] D. Marcuse, Theory of Optical Waveguides, Academic Press, NY, 1974.
[9] H.A. Haus, Waves and Fields in Optoelectronics, Prentice Hall, 1983.
[10] A.W. Snyder and J.D. Love, Optical Waveguide Theory, London U.K: Chapman and
Hall, 1983.
[11] A. Yariv, Optical Electronics, Holt, Rinehart, and Winston, New York, 1985.
[12] P.-A. Blanger, Optical Fiber Theory a Supplement to Applied Electromagnetism, World
Scientific, 1993.
[13] S.L. Chuang, Physics of Optoelectronic Devices, Wiley-Interscience, 1995.
[14] K. Okamoto, Fundamentals of Optical Waveguides, Academic Press, 2000.
[16] E. Snitzer, “Cylindrical dielectric waveguide modes,” J. Opt. Soc. Am., 51, pp. 491-498,
1961.
291
[17] A.W. Snyder, “Asymptotic expression for eigenfunctions and eigenvalues of dielectric
optical waveguides,” IEEE Trans. Microwave Theory Tech., MTT-17, pp. 1130-1138, 1969.
[18] D. Gloge, “Weakly guiding fibers,” App. Opt., 10, 2252-2258, 1971.
[19] D. Gloge and E.A.J. Mercatili, “Multimode theory of graded-core fibers,” Bell Syst.
Tech. J., 52, pp. 1563-1578, 1973.
[20] J.G. Dil and H. Blok, “Propagation of electromagnetic surface waves in a radially inho-
mogeneous optical waveguide,” Opto-Electronics (London), 5, pp. 415-428, 1973.
[21] M. Hashimoto, “A perturbation method for the analysis of wave propagation in inho-
mogeneous dielectric waveguides with perturbed media,” IEEE Trans. Microwave Theory
and Tech., vol. 24, no. 9, pp. 559-566, 1976.
[22] H.A. Haus, Electromagnetic Noise and Quantum Optical Measurements, Berlin:
Springer-Verlag, 2000.
[23] T. Miya, Y. Terunuma, T. Hosaka and T. Miyashita, “Ultimate low-loss single-mode

fiber at 1.55 µm,” Electron Lett., v. 15, p. 106, 1979.
[24] J.E Goell, “A circular-harmonic computer analysis of rectangular dielectric waveguides,”

Bell Syst. Tech. J., v. 48, pp. 2133-2160, 1969.
[26] E.A.J. Mercatili, “Dielectric rectangular waveguide and directional coupler for integrated
optics,” Bell Syst. Tech. J., 48, pp. 2071-2102, 1969.
[27] W. C. Chew and M. Nasir, “A variational analysis of anisotropic, inhomogeneous dielec-

tric waveguides,” IEEE Trans. Microwave Theory Techniques, vol. 37, no. 4, pp. 661-668,
Apr. 1989.
[28] J. Jin, The finite element method in electromagnetics, John Wiley & Sons, Inc., New
York, 1993.
[29] W.C. Chew, ”Analysis of optical and millimeter wave dielectric waveguide,” J. Elect.
Waves Appl., vol. 3, no. 4, pp. 359-377, 1989.
[30] K. Radhakrishnan and W.C. Chew, “An efficient Krylov subspace based algorithm to
solve the dielectric waveguide problem,” IEEE Trans. Micro. Theory Tech., vol. 49, no.
7, pp. 1345-1347, July 2001.
[31] C.M. Angulo, “Diffraction of surface waves by a semi-infinite dielectric slab,” IRE Trans.
Antennas Propag., vol. AP-7, pp. 261-274, 1957.
[32] T. Ikegami, “Reflectivity of mode of facet and oscillation mode in double heterostructure
injection lasers,” IEEE J. Quantum Electron., QE-8,470-476, 1972.
[33] T. Rozzi, “Rigorous analysis of the step discontinuities in planar dielectric waveguides,”
IEEE Trans. Microwave Theory Tech., vol. MTT-26, pp. 738-746, 1978.
[34] M. Pudensi and L. Ferreira, “Method to calculate the reflection and transmission of
guided waves,” J. Opt. Soc. Am., 72, 126-130, 1982.
[35] Q.H. Liu and W.C. Chew, “Numerical mode-matching method for the multiregion, ver-
tically stratified media,” IEEE Trans. Antennas Propag., vol. AP-38, no. 4, pp. 498-506,
Apr. 1990.
[36] C. M. Herzinger, C.C. Lu, T.A. De Temple and W.C. Chew, “Semiconductor waveguide
facet reflectivity problem,” IEEE J. Quantum Elec., 29, 2273-2281, 1993.
[37] R.M. Know and P.P. Toulios, “Integrated circuits for the millimeter through optical
frequency range,” Symposium on Submillimeter Waves, Polytechnic Institute of Brooklyn,
pp. 497-516, 1970.
[38] T. Tamir, Integrated Optics, Springer-Verlag, Berlin, 1975.
[39] J. Buus, “The effective index method and its application to semiconductor lasers,” IEEE
J. Quantum Electronics, vol. 18, no. 7, pp. 1083-1089, Jul 1982.
[41] J.A. Fleck, J.R. Morris, and M.D. Feit, “Time-dependent propagation of high energy
laser beams through the atmosphere,” Appl. Phys., 10, pp. 129-160, 1976.
[42] M.D. Feit and J.A. Fleck, “Light propagation in graded-index optical fibers,” Appl. Opt.,
vol. 17, pp. 3990-3997, 1978.
[43] W.P. Huang and C.L. Xu, “Simulation of three-dimensional optical waveguides by a full-
vector beam propagation method,” IEEE J. Quantum Electron., vol. 29, pp. 26392649,
Oct. 1993.
[44] A. Hochman and Y. Leviatan, “Calculation of confinement losses in photonic crystal

fibers by use of a source-model technique,” J. Opt. Soc. Am. B , vol. 22, no. 2, p. 474,
February 2005.
[45] T. Matsui, M. Ozaki, and K. Yoshino, “Tunable laser action in a dye-doped nematic
liquid-crystal waveguide under holographic excitation based on electric-field-induced TM
guided-mode modulation,” J. Opt. Soc. Am. B , vol. 21, no. 9, p. 1651, September 2004.
[46] J.K.S. Poon, J. Scheuer, Y. Xu, and A. Yariv, “Designing coupled-resonator optical
waveguide delay lines,” J. Opt. Soc. Am. B , vol. 21, no. 9, p. 1665, September 2004.
[47] Y. Lu, S. Liu, G. Zhang, R. Guo, N. Zhu, and L. Yang, “Waveguides and directional
coupler induced by white-light photovoltaic dark spatial solitons,” J. Opt. Soc. Am. B ,
vol. 21, no. 9, p. 1674, September 2004.
[48] N. Malkova and V. Gopalan, “Resonant light propagation through 90-bend waveguide
based on a strained two-dimensional photonic crystal,” J. Opt. Soc. Am. B , vol. 21, no.
9, p. 1679, September 2004.
[49] C. Kappel, A. Selle, M.A. Bader, and G. Marowsky, “Resonant double-grating waveguide
structures as inverted Fabry-Perot interferometers,” J. Opt. Soc. Am. B , vol. 21, no. 6,
p. 1127, June 2004.
[50] M. Mohebbi, “Dispersion of femtosecond laser pulses in hollow fibers,” J. Opt. Soc. Am.
B , vol. 21, no. 5, p. 893, May 2004.
[51] E. Miyai and S. Noda, “Structural dependence of coupling between a two-dimensional

photonic crystal waveguide and a wire waveguide,” J. Opt. Soc. Am. B , vol. 21, no. 1, p.
67, January 2004.
[52] P.E. Barclay, K. Srinivasan, and O. Painter, “Design of photonic crystal waveguides for
evanescent coupling to optical fiber tapers and integration with high-Q cavities,” J. Opt.
Soc. Am. B , vol. 20, no. 11, p. 2274, November 2003.
[53] P. Bienstman, S. Assefa, S.G. Johnson, J.D. Joannopoulos, G.S. Petrich, and L.A.
Kolodziejski, “Taper structures for coupling into photonic crystal slab waveguides,” J.
Opt. Soc. Am. B , vol. 20, no. 9, p. 1817, September 2003.
[54] O. Skorka, J. Salzman, and S. Zamir, “Coupled waveguides in GaN-based lasers,” J. Opt.
Soc. Am. B , vol. 20, no. 9, p. 1822, September 2003.
[55] T. Yang, Y. Sugimoto, S. Lan, N. Ikeda, Y. Tanaka, and K. Asakawa, “Transmission

properties of coupled-cavity waveguides based on two-dimensional photonic crystals with
a triangular lattice of air holes,” J. Opt. Soc. Am. B , vol. 20, no. 9, p. 1922, September
2003.
[56] A. Zakery, Y. Ruan, A. V. Rode, M. Samoc, and B. Luther-Davies, “Low-loss waveguides

in ultrafast laser-deposited As2 S3 chalcogenide films,” J. Opt. Soc. Am. B , vol. 20, no.
9, p. 1844, September 2003.
[57] J. Limeres, M. L. Calvo, J. M. Enoch, and V. Lakshminarayanan, “Light scattering by

an array of birefringent optical waveguides: theoretical foundations,” J. Opt. Soc. Am. B
, vol. 20, no. 7, p. 1542, July 2003.
[58] A.M. Ljungstrm and T.M. Monro, “Exploration of self-writing and photosensitivity in
ion-exchanged waveguides,” J. Opt. Soc. Am. B , vol. 20, no. 6, p. 1317, June 2003.
[59] M. Huang and X. Yan, “Thermal-stress effects on the temperature sensitivity of optical
waveguides,” J. Opt. Soc. Am. B , vol. 20, no. 6, p. 1326, June 2003.
[60] Z.-Y. Li and K.-M. Ho, “Waveguides in three-dimensional layer-by-layer photonic crys-
tals,” J. Opt. Soc. Am. B , vol. 20, no. 5, p. 801, May 2003.
[61] S. Hadjiloucas, R.K.H. Galvão, J.W. Bowen, R. Martini, M. Brucherseifer, H.P.M. Pelle-
mans, P.H. Bolvar, H. Kurz, J. Digby, G.M. Parkhurst, and J.M. Chamberlain, “Mea-
surement of propagation constant in waveguides with wideband coherent terahertz spec-
troscopy,” J. Opt. Soc. Am. B , vol. 20, no. 2, p. 391, February 2003.
Chapter 7
Microwave Integrated Circuits
Due to the advent of integrated circuits at microwave frequencies, microwave integrated cir-
cuits waveguides are omnipresent in microwave technologies. Their ease of fabrication, low
cost, conformal nature have made them extremely popular. Moreover, they are easily inte-
grated with other integrated circuits. Some examples are shown in Figure 7.1. More recently,
microwave integrated circuits have been used to study cavity QED (quantum electrodynam-
ics) that plays an important role in quantum computing [1]. Hence, they are being used at
the frontier of science investigations as well.
The microstrip line has a long history. Since its appearance before World War II, it has
been continuously worked on. The early work uses quasi-TEM approximation which does not
account for dispersive effects in the line [2–8]. The use of this waveguide at higher frequencies
call for the analysis accounting for dispersive effect [9–15,41]. Other works related to analysis
of microstrip integrated circuits are [17–25]. Recent typical works in this area are given
in [27–29].
7.1 Quasi-TEM Approximation

The integrated circuits waveguiding structures cannot support a TEM mode, for if they do,
the phase matching condition will be violated at the interface between the inhomogeneities.
This is because a TEM wave has the phase velocity of the medium in which the wave is
traveling.
However, if the wavelength under consideration is much larger than the transverse struc-
ture of the waveguide, we can show that the fundamental mode of such a structure is almost
TEM or quasi-TEM. A fundamental mode is the mode that is propagating when ω → 0 or
λ → ∞.
We can write Maxwell’s equations by separating out the transverse and longitudinal com-
ponents as
∇s × Es = ikηHz , (7.1.1)
297
y
x
z
εr εr
Microstrip Line Slot Line
εr εr
Coplanar Waveguide Coplanar Strips
Figure 7.1: Different kinds of microwave integrated circuits waveguides.
∇s × Hs = −ikη −1 Ez , (7.1.2)
∂
ẑ × Es − ẑ × ∇s Ez = ikηHs , (7.1.3)
∂z
∂
ẑ × Hs − ẑ × ∇s Hz = −ikη −1 Es , (7.1.4)
∂z
pµ
where η = . For a structure whose dominant transverse dimension is much smaller than
the wavelength, the transverse variation of the field would be more rapid than the longitudinal
variation. The transverse variation of the field has to be fast enough for the field to match
the boundary condition, namely, the x and y variations will vary on the length scale of δ. To
emphasize this fact, we can perform a coordinate stretching transformation [30] by letting
x = δx0 , y = δy 0 . (7.1.5)
Under such a coordinate stretching transformation,

1 0
∇s → ∇ , (7.1.6)
δ s
Equations (7.1.3) and (7.1.4) become
ikz δẑ × Es − ẑ × ∇0s Ez = ikηδHs , (7.1.7)
ikz δẑ × Hs − ẑ × ∇0s Hz = −ikη −1 δEs , (7.1.8)

ikz z
where p
we have assumed e dependence of the field. When δ/λ → 0, then, kδ → 0. Since
kz = k 2 − ks2 < k, we also expect kz δ → 0, when δ/λ → 0. Therefore, in the long
wavelength limit,
(Ez , ηHz ) ∼ O(kδ)(Es , ηHs ). (7.1.9)
Microwave Integrated Circuits 299
In other words, |Ez | |Es |, |Hz | |Hs |, implying that the field is quasi-TEM. Consequently,
we can write (7.1.1) and (7.1.2) as
∇s × Es = 0, (7.1.10)
∇s × Hs = 0. (7.1.11)
Taking the divergence of (7.1.3) and (7.1.4) and making use of (7.1.10) and (7.1.11), we have
∇s · Es = 0, (7.1.12)
∇s · µHs = 0. (7.1.13)
y
x
z εr
Figure 7.2: A generic geometry for the analysis of microwave integrated circuits waveguide.
Consequently, the transverse field of a quasi-TEM mode is essentially static. Because of

this, the waveguide can be analyzed as if a TEM mode is propagating on it. We can solve the
transverse electrostatic problem to find the line capacitance C of the line. The line inductance
L can be obtained by solving the magnetostatic problem. Then, kz , the axial wavenumber,
can be found by √
kz = ω LC. (7.1.14)
The above analysis indicates that when the wavelength is long, the axial variation of the
field is slow compared to the transverse variation. The transverse variation of the field has to
be such that the field can match the boundary condition on the metallic conductors, which is
much smaller in dimension than the wavelength. Hence, the transverse variation of the field
must balance itself resulting in equations (7.1.10) to (7.1.12), which are the static equations.
Please be noted that the reason for the quasi-TEM field here is quite different from that in
the weak-contrast optical fiber. In the weak-contrast optical fiber, the reason for quasi=TEM
is the paraxial nature of the wave as the frequency increases and the unimportance of the
polarization term comparatively. In the weak-contrast optical fiber, the transverse dimension
can be on the order or wavelength or larger, and yet the field is quasi-TEM.
7.1.1 Microstrip Line—Line Capacitance

We shall discuss how the line capacitance of a microstrip line can be found. There is no closed-
form solution for such a class of problem. When 0 = 1 , one may solve such problems by
conformal mapping. When h/w 1, the problem can be solved by asymptotic matching [31].
However, to get an accurate value of C for all w/h, a numerical analysis is more preferable.
To find the line capacitance of the microstrip line, one can first solve for the charge
distribution on the line. Then, the capacitance can be easily found from the equation Q = CV ,
where Q is the total charge per unit length on the line, and V is the voltage applied between
the strip and the ground plane. To find the charge distribution, we can first formulate an
integral equation governing the charge distribution. A potential φ can be defined such that
Es = −∇s φ and ∇2s φ = 0, because Es is an electrostatic field in the long wavelength limit.
y
φ=V
Region 0, ε0 W
x
z
Region 1, ε1 h
φ=0
Figure 7.3: Geometry for deriving the line capacitance of a microstrip line.
φ=V
ε0
ε0 x
W
Figure 7.4: A charged strip hanging in free space assumed to be infinitesimally thin.
Before formulating the integral equation for the geometry shown in Figure 7.3, let us first
consider the case of a charged strip suspended in free space shown in Figure 7.4. Using Fourier
transforms, which are the gist of the spectral domain approach, the potential can be written
as Z ∞
1
φ0± (x, y) = dλeiλx φ̃0± (λ, y). (7.1.15)
2π −∞
where the subscript + and − denote y > 0 and y < 0, respectively. Since ∇2s φ0± (x, y) = 0,
we deduce that
d2
φ̃0± (λ, y) − λ2 φ̃0± (λ, y) = 0, (7.1.16)
dy 2
or that
φ̃0± (λ, y) = A0± e−|λ|y + B0± e|λ|y . (7.1.17)
Furthermore, φ̃0+ (λ, y) = 0, y → ∞, and φ̃0− (λ, y) = 0, y → −∞, which implies that
B0+ = A0− = 0. But since φ0+ (x, y = 0) = φ0− (x, y = 0) for all x, we have φ̃0+ (λ, y = 0) =
φ̃0− (λ, y = 0) implying that A0+ = B0− . Consequently, we have
Z ∞
1 y>0
φ0± (x, y) = dλeiλx A0 (λ)e∓|λ|y , . (7.1.18)
2π −∞ y<0
The charge on the strip is given by

∂φ0+ ∂φ0−
σ(x) = −0 (x, y = 0) − (x, y = 0) , (7.1.19)
∂y ∂y
or that Z ∞
0
σ(x) = dλeiλx |λ|A0 (λ). (7.1.20)
π −∞
Defining the Fourier transform of σ(x) as σ̃(λ), it can be shown that
σ̃(λ)
A0 (λ) = , (7.1.21)
20 |λ|
or, in general, the potential becomes

Z ∞
1 σ̃(λ) y>0
φ0± (x, y) = dλeiλx∓|λ|y , . (7.1.22a)
4π0 −∞ |λ| y<0
or Z ∞
1 σ̃(λ)
φ0 (x, y) = dλeiλx−|λy| . (7.1.22b)
4π0 −∞ |λ|
φ=V
ε0
ε0 d x
ε1
Figure 7.5: A charged strip, assumed to be infinitesimally thin, over a dielectric half-space.
If we now place the charged strip over a dielectric half space, the potential in region 0
becomes Z ∞
1 σ̃(λ) iλx −|λy|
φ0 (x, y) = dλ e [e + r01 e−|λ|(y+d)−|λd| ], (7.1.23)
4π0 −∞ |λ|
where r01 is a reflection coefficient relating the reflected potential e−|λ|y to the incident
potential e−|λy| at y = −d. In region 1, the potential is
Z ∞
1 σ̃(λ) iλx
φ1 (x, y) = dλ e t01 e|λ|(y+d)−|λd| , (7.1.24)
4π0 −∞ |λ|
where t01 is a transmission coefficient relating the transmitted potentiale|λ|y to the incident
potential e−|λy| . The continuity of potential at y = −d requires that
1 + r01 = t01 . (7.1.25)
The continuity of normal electric flux at y = −d implies that
0 (1 − r01 ) = 1 t01 . (7.1.26)
Solving (7.1.25) and (7.1.26) yields

0 − 1 20
r01 = , t01 = . (7.1.27)
0 + 1 0 + 1
φ=V
ε0 d x
ε1 h
ε2
Figure 7.6: A charged strip over a dielectric slab.
For the geometry of Figure 7.6, we can find the total reflection coefficient using a geometric
series expansion [32, 33] to obtain
t01 r12 t10 e−2|λ|h

r̃01 = r01 + . (7.1.28)
1 − r10 r12 e−2|λ|h
Equation (7.1.28) can be written as a recursive relation if we have more subsurface layers. In
this case, the potential in region 1 can be written as
Z ∞
1 σ̃(λ)eiλx −|λy|
φ0 (x, y) = dλ [e + r̃01 e−|λ|(y+d)−|λd| ]. (7.1.29)
4π0 −∞ |λ|
For the geometry of Figure 7.3, d = 0, r12 = −1; therefore,

Z ∞
1 σ̃(λ)eiλx −|λ|y
φ0 (x, y) = dλ e (1 + r̃01 ), y > 0, (7.1.30)
4π0 −∞ |λ|
where
t01 t10 e−2|λ|h 1 − e−2|λ|h
1 + r̃01 = t01 − = t01 . (7.1.31)
1 − r01 e−2|λ|h 1 − r01 e−2|λ|h
Requiring that the potential be V on the strip, then the integral equation for σ̃(λ) can be
written as Z ∞
1 w
dλG̃(λ)σ̃(λ)eiλx = V, |x| < , (7.1.32a)
2π −∞ 2
Z ∞
1 w
σ(x) = dλσ̃(λ)eiλx = 0, |x| > , (7.1.32b)
2π −∞ 2
where σ̃(λ) is the Fourier transform of σ(x), and
1 1 − e−2|λ|h
G̃(λ) = . (7.1.33)
|λ|(0 + 1 ) 1 − 00 − 1 −2|λ|h
+1 e
Equations (7.1.32a) and (7.1.32b) are also known as dual integral equations. Dual integral
equations in general do not have closed-form solutions, except for semi-infinite structures. In
that case, they can be solved using Wiener-Hopf technique [38, 39].
In order to solve (7.1.32b), we use Galerkin’s method and let
N
X
σ(x) = an fn (x), (7.1.34)
n=0
where fn (x) = 0, |x| > w2 , and that the Fourier transform of fn (x) exists. Furthermore,
fn (x), n = 1, . . . , ∞ is complete for |x| < w2 . Then, we deduce that
N
X
σ̃(λ) = an fñ (λ), (7.1.35)
n=0
where fñ (λ) is the Fourier transform of fn (x). Substituting (7.1.35) into (7.1.32a), we have
N Z ∞
1 X
an dλG̃(λ)fñ (λ)eiλx = V. (7.1.36)
2π n=0 −∞
To remove the x dependence in (7.1.36), we multiply it by fm (x) and integrate over x, to

obtain
Z ∞ w/2
Z
XN
an ˜ ˜
dλfm (−λ)G̃(λ)fn (λ) = 2πV fm (x)dx. (7.1.37)
n=0 −∞
−w/2
The above is a matrix equation of the form

N
X
Amn an = bm , (7.1.38)
n=0
from which we can solve for an ’s. Once an ’s are known, we can find σ(x) from (7.1.34). Then
w/2
Z
Q= σ(x)dx,
−w/2
and the line capacitance is given by C = Q/V .

For the geometry considered in Figure 7.3, the quasi-TEM mode is even about x = 0.
Therefore, we need only to pick even functions for our basis functions in (7.1.34). Then, by
the theory of Fourier transform, fñ (λ) = fñ (−λ) if fn (x) is an even function. Furthermore,
since G̃(−λ) = G̃(λ), (7.1.37) becomes
Z∞ w/2
Z
N
X
an dλf˜m (λ)G̃(λ)fñ (λ) = 2πV fm (x)dx. (7.1.39)
n=0 0 0
Therefore, Amn is a symmetric matrix.

Equation (7.1.32a) has a different meaning if it is transformed back to x-space, i.e.,
w/2
Z
w
dx0 G(x − x0 )σ(x0 ) = V, |x| < , (7.1.40)
2
−w/2
where G(x) is the inverse Fourier transform of G̃(λ). G(x) could be thought of as the Green’s
function generating the potential due to a line of point surface charge at y = 0. Hence, the
convolution of G(x) with σ(x) gives the potential. However, closed-form expression does not
exist for G(x) in general. A more convenient method to calculate the integral (7.1.40) is in
the spectral domain as in (7.1.32a) where G̃(λ) exists in closed-form.
Once C is known, we can estimate kz via (7.1.28). To find L, we make use of the fact that
if 1 = 0 , a pure TEM mode propagates on the microstrip line. In this case,
p
kz = k0 = ω L0 C0 . (7.1.41)
Therefore, L0 can be found once C0 , the line capacitance with 1 = 0 , is known. C0 can
be found by solving the integral equation above. Since L0 , which is obtained by solving the
magnetostatic problem, remains unchanged when 1 6= 0 , we have

r
C
kz = k0 . (7.1.42)
C0
An effective relative dielectric constant can be defined such that

C
re = . (7.1.43)
C0
It is the dielectric constant with which one can fill the space homogeneously around a mi-
crostrip line to yield the same line capacitance as the inhomogeneously filled microstrip line.
Figure 7.7: Characteristic impedance of a microstrip line as a function of line dimension

w/h—quasi-static approximation (from H.A. Wheeler [2].)
A characteristic impedance can also be defined as

r r
L0 µ0 0 Za
Z0 = = = √0 , (7.1.44)
C C0 C r
where Z0a is the characteristic impedance of the air-filled microstrip line.

S1 +Q
ε0 φ (r)
^n
V εr
S2
–Q
Figure 7.8: Geometry for the analysis of bounds for capacitance.
7.1.2 Variational Expressions and Bounds for Capacitance

There exists bounds for the capacitance between two conductors due to the minimum energy
principle [34–37]. Due to the prevalence use of capacitance extraction software in the computer
chip industry, this could be of importance.
The exact solution of Laplace’s equation minimizes the energy in the potential as shall be
shown. The capacitance can be related to the energy stored in the potential or chage in two
ways, i.e.
1 1 Q2
W = CΦ2 = , (7.1.45)
2 2 C
where W is the energy stored in the potential Φ, the voltage difference between the two
conductors. Then if W is not accurately known, the exact capacitance Ce is bounded by
2W Q2
≥ Ce ≥ (7.1.46)
Φ2 2W
if either Φ is kept constant of if Q is kept constant on the metallic conductor. The above
assertion shall be proved in the following.
In the first inequality in (7.1.46), the potential satisfies the boundary condition exactly,
but does not satisfy Laplace’s equation exactly in the space in between the conductors. It
can be rewritten as
R R
|E|2 dV |∇φ|2 dV
V
=V ≥ Ce . (7.1.47)
Φ2 Φ2
If we let φ = φe + δφ such that δφ = 0 on S1 and S2 , because φ satisfies the boundary
condition exactly, then
Z Z Z Z
|∇φ|2 = |∇φe |2 dV + 2 ∇φe · ∇δφW + |∇δφ|2 dV. (7.1.48)
V V V V
But
Z Z Z
∇φe · ∇δφdV = ∇ · (δφ∇φe )dV − δφ(∇ · ∇φe )dV
V V V
Z Z (7.1.49)
=− n̂ · (δφ∇φe )dS − δφ(∇ · ∇φe )dV.
S1 +S2 V
The first term in (7.1.49) vanishes because δφ = 0 on S1 and S2 , and the second term vanishes
because ∇ · ∇φe = 0, in V . Therefore, if φ = φe + δφ where δφ is the error field in V , then
Z Z Z Z
|∇φ|2 dV = |∇φe |2 dV + |∇δφ|2 dV > |∇φe |2 dV, (7.1.50)
V V V V
implying the bound in (7.1.47) for a constant Φ.

The second inequality in (7.1.46) can be written as
R R
|E|2 dV |∇φ|2 dV
V V
= ≥ Ce−1 (7.1.51)
Q2 Q2
where Q is constant but φ satisfies Laplace’s equation exactly in V but does not satisfy the
boundary condition. Similar to (7.1.48), we let φ = φe + δφ, but now, δφ 6= 0 on S1 and S2 .
Similar to (7.1.49), we obtain that the second term in (7.1.48) is
Z Z Z
∇φe · ∇δφdV = ∇ · (φe ∇δφ)dV − φe ∇ · ∇δφdV
V V V
Z Z (7.1.52)
=− n̂ · (φe ∇δφ)dS − φe ∇ · ∇δφdV.
S1 +S2 V
The second term in (7.1.52) vanishes because ∇ · ∇δφ = 0. Since φe = Φ on S1 and

φe = 0 on S2 , the first term becomes
Z
−Φ n̂ · ∇δφdS. (7.1.53)
S1
Since −n̂ · ∇δφ = δσ, the error surface charge density on S1 , it integrates to zero because we
assume that Q is a constant so that δQ = 0. Therefore, the second term in (7.1.48) vanishes
too for this case. Consequently, we obtain the bound in (7.1.51).
Equation (7.1.51) is also Thompson’s theorem [35] which says that a set of charge Q
always adjusts itself on a metallic conductor such that the energy stored in the electric field is
minimized. The inequality (7.1.51) assumes that the potential φ satisfies ∇ · ∇φ = 0 exactly
in V but does not satisfy the boundary condition. Hence, it is more appropriate to write
(7.1.51) in terms of surface integrals. To this end, we express using integration by parts that
Z Z Z
|∇φ|2 dV = − dSφ n̂ · ∇φ = dSφσ. (7.1.54)
V S1 +S2 S1 +S2
In this manner, (7.1.51) becomes

R
dSφσ
S1 +S2
Ce−1 ≤ . (7.1.55)
Q2
A potential which satisfies ∇ · ∇φ = 0 exactly in V can be obtained by the Green’s function
method if the Green’s function is known exactly. Hence,
Z
φ(r) = dS 0 G(r, r0 )σ(r0 ), (7.1.56)
S1 +S2
or that (7.1.55) can be expressed as

Z Z
−1
Ce ≤ dSσ(r) dS 0 G(r, r0 )σ(r0 )/Q2 , (7.1.57)
S1 +S2 S1 +S2
or
hσ, G, σi
Ce−1 ≤ . (7.1.58)
Q2
where Z Z
hσ, G, σi = dSσ(r) dS 0 G(r, r0 )σ(r0 )
S1 +S2 S1 +S2
The expressions that Z

C= dV |∇φ|2 /Φ2 (7.1.59)
V
and
C −1 = hσ, G, σi/Q2 (7.1.60)
are both variational expressions. Rayleigh-Ritz procedure can be applied to (7.1.59) to solve
for φ which is the same as applying the finite element method to solve ∇ · ∇φ = 0. Applying
Rayleigh-Ritz procedure to solve (7.1.60) for σ is the same as applying Galerkin’s method to
solve the integral equation
Z (
0 0 0 V, r ∈ S1
dS G(r, r )σ(r ) = . (7.1.61)
0, r ∈ S2
S1 +S2
7.2 Microstrip Line—A Frequency Dependent Theory

The quasi-TEM model of the microstrip line triumphs in predicting the value of kz or the
phase velocity in the long wavelength limit. However, kz is in general frequency dispersive.
In order to have a kz that is valid at high frequencies as well, we need to solve the full wave
solution to the microstrip line problem. To do this, an integral equation can be formulated
from Maxwell’s equation with no approximation. From the integral equation, the guidance
condition of the microstrip line can be solved for.
z x
Figure 7.9: An infinitesimally thin strip hanging in free space.
7.2.1 Derivation of the Integral Equation

Before deriving an integral equation for the microstrip line, let us consider first a metallic strip
carrying a current suspended in free space. The field around the strip can be decomposed
into TE to y and TM to y fields by H0y and E0y , respectively. Using Fourier transforms, the
fields can be written as
Z ∞
1
E0y (x, y, z) = eikz z dkx eikx x Ẽ0y (kx , y), (7.2.1)
2π −∞
Z ∞
1
H0y (x, y, z) = eikz z dkx eikx x H̃0y (kx , y). (7.2.2)
2π −∞
In the above, we assume that the fields have eikz z variation. Since (∇2 + k02 )E0y = 0, and
(∇2 + k02 )H0y = 0, it follows that

d2 2
+ k y Ẽ0y (kx , y) = 0, (7.2.3a)
dy 2

d2 2
+ ky H̃0y (kx , y) = 0, (7.2.3b)
dy 2
where ky2 = k02 − kx2 − kz2 . This implies that
Ẽ0y (kx , y) = e0+ eiky y + e0− e−iky y , (7.2.4a)
H̃0y (kx , y) = h0+ eiky y + h0− e−iky y . (7.2.4b)

However, for y > 0, only upward going waves exist, while for y < 0, only have downward
going waves exist. Therefore, e0− = h0− = 0, y > 0, e0+ = h0+ = 0, y < 0. Consequently,
(7.2.1) and (7.2.2) become
Z ∞
eikz z y>0
E0y (x, y, z) = dkx eikx x e0± e±iky y , , (7.2.5a)
2π −∞ y<0
Z ∞
eikz z y>0
H0y (x, y, z) = dkx eikx x h0± e±iky y , . (7.2.5b)
2π −∞ y<0
On the other hand, if the strip is carrying electric current, then E0y (x, y+, z) = −E0y (x, y−, z)
and H0y (x, y+, z) = H0y (x, y−, z) for all x. Therefore, e0+ = −e0− = e0 , h0+ = h0− = h0 ,
and we have Z
eikz z ∞
E0y (x, y, z) = ± dkx eikx x+iky |y| e0 (kx ), , (7.2.6a)
2π −∞
Z
eikz z ∞
H0y (x, y, z) = dkx eikx x+iky |y| h0 (kx ).. (7.2.6b)
2π −∞
Figure 7.10: Frequency dependent effective dielectric constant of the quasi-TEM mode on
single microstrip lines (from Jansen [14], reproduced by G. Papadopoulous.)
If now, the strip is on top of a stratified half space as shown in Figure 7.10, the field in
region 0 can be written as
Z h i
eikz z ∞
E0y (x, y, z) = dkx eikx x e0 (kx ) ±eiky |y| − R̃T M eiky (y+d)+iky |d| , (7.2.7a)
2π −∞
Z h i
eikz z ∞
H0y (x, y, z) = dkx eikx x h0 (kx ) eiky |y| + R̃T E eiky (y+d)+iky |d| . (7.2.7b)
2π −∞
In order to relate e0 and h0 to the current on the strip, we need to derive the transverse
components of the field. Since each spectral component in (7.2.7a) and (7.2.7b) consists of
waves with e±iky y dependence, the fields transverse to y for each spectral component are

1 ∂
Ês = 2 ∇s Êy − iωµ0 ŷ × ∇s Ĥy , , (7.2.8a)
ks ∂y

1 ∂
Ĥs = 2 ∇s Ĥy + iω0 ŷ × ∇s Êy , . (7.2.8b)
ks ∂y
∂ ∂
where the hat denotes the spectral component, ks2 = k02 − ky2 , and ∇s = x̂ ∂x + ẑ ∂z . Applying
(7.2.8b) to (7.2.7a) and (7.2.7b), we have
Z
eikz z ∞ eikx x
Hs = dkx 2 {−h0 (kx )ky ks [±eiky |y| + R̃T E eiky (y+d)+iky |d| ]
2π −∞ ks (7.2.9)
iky |y| T M iky (y+d)+iky |d|
− ω0 e0 (kx )ŷ × ks [±e − R̃ e ]}.
The discontinuity in Hs at y = 0 equals the current on the strip. More precisely, Js =

ŷ × [Hs (y = 0+) − Hs (y = 0−)]. Hence,
Z
eikz z ∞ eikx x
Js = dkx 2 [−2h0 ky ŷ × ks + 2ω0 e0 ks ]
2π −∞ ks
Z
eikz z ∞ eikx x 2ω0 e0 /ks
= dkx [ks , −ŷ × ks ] (7.2.10)
2π −∞ ks 2ky h0 /ks
Z
eikz z ∞
= dkx F(kx , x) · Ks (kx ),
2π −∞
ikx x
where F(kx , x) = e ks [ks , −ŷ × ks ], Kts (kx ) = 2[ω0 e0 , ky h0 ]/ks .
Similarly, Es can be derived to be
Z
eikz z ∞ eikx x
Es = dkx 2 {−ky ks e0 [eiky |y| − R̃T M eiky (y+d)+iky |d| ]
2π −∞ ks (7.2.11)
iky |y| T E iky (y+d)+iky |d|
+ ωµ0 h0 ŷ × ks [e + R̃ e ]}.
We can rewrite (7.2.11) as

Z
eikz z ∞ eikx x
Es = dkx [ks , −ŷ × ks ]
2π −∞ ks

−ky e0 [eiky |y| − R̃T M eiky (y+d)+iky |d| ]/ks
(7.2.12)
−ωµ0 h0 [eiky |y| + R̃T E eiky (y+d)+iky |d| ]/ks
Z
eikz z ∞
= dkx F(kx , x) · G(kx , y) · Ks (kx ),
2π −∞
where
G(kx , y) =
" k #
− 2ωy 0 [eiky |y| − R̃T M eiky (y+d)+iky |d| ] 0 (7.2.13)
.
0 − ωµ
2ky [e
0 iky |y|
+ R̃T E eiky (y+d)+iky |d| ]
For a microstrip line, d = 0, and from the boundary condition Es = 0 for |x| < w/2 and
y = 0. Therefore, the integral equation is
Z
eikz z ∞ w
Es (x, y = 0, z) = F(kx , x) · G0 (kx ) · Ks (kx )dkx = 0, |x| < , (7.2.14)
2π −∞ 2
where Ks (kx ) is the unknown to be sought, and

" k #
y
[1 − R̃T M ] 0
G0 (kx ) = − 2ω0 ωµ0 TE . (7.2.15)
0 2ky [1 + R̃ ]
7.2.2 Vector Fourier Transform (VFT)

With F(kx , x) defined in (7.2.10), there exists a vector Fourier transform pair given by [40,41].
Z ∞
1
h(x) = dkx F(kx , x) · H(kx ), (7.2.16a)
2π −∞
Z ∞
H(kx ) = dx0 F(kx , −x0 ) · h(x0 ), (7.2.16b)
−∞
where
eikx x eikx x kz kx
F(kx , x) = [ks , −ŷ × ks ] = . (7.2.16c)
ks ks kx −kz
This can be proven by substituting (7.2.16b) into (7.2.16a) to obtain
Z ∞ Z ∞
1
h(x) = dkx dx0 F(kx , x) · F(kx , −x0 ) · h(x0 )
2π −∞ −∞
Z ∞ Z ∞ (7.2.17)
1 0 0
= dx dkx eikx (x−x ) h(x0 ) = h(x).
2π −∞ −∞
A similar substitution of (7.2.16a) into (7.2.16b) yields similar results. Therefore, (7.2.16a)
and (7.2.16b) constitute a vector Fourier transform pair.
From the above, note that in Equation (7.2.10), if Js (x, z) = eikz z ks (x), then Ks (kx ) is
the vector Fourier transform of ks (x). In order to solve (7.2.14), we let
XN XN
ksz (x) an knz (x)
ks (x) = = = kn (x) · an , (7.2.18a)
ksx (x) bn knz (x)
n=1 n=1
where
a knz (x) 0
an = n kn (x) = . (7.2.18b)
bn 0 knx (x)
kn (x) is chosen so that its vector Fourier transform exists. Then
N
X
Ks (kx ) = Kn (kx ) · an , (7.2.19)
n=1
where Kn (kx ) is the VFT of kn (x). Substituting (7.2.19) into (7.2.14), we have
N Z
X ∞
dkx F(kx , x) · G0 (kx ) · Kn (kx ) · an . (7.2.20)
n=1 −∞
0
To remove the x-dependence in the above equation, we multiply it by km (x) and integrate
over x, where
0 k (x) 0
km (x) = mz . (7.2.21)
0 −kmx (x)
Then (7.2.20) becomes
N Z
X ∞
0
dkx Km (kx ) · G0 (kx ) · Kn (kx ) · an , (7.2.22)
n=1 −∞
where Z ∞
0 0
Km (kx ) = dxkm (x) · F(kx , x). (7.2.23)
−∞
Equation (7.2.22) is of the form

N
X
Mmn · an = 0, m = 1, . . . , N. (7.2.24)
n=1
The above is expressible in terms of a matrix equation
M · a = 0, (7.2.25)
where  
  a1
M11 M12 ... a2 
M21 M22   
M= , a = a3  . (7.2.26)
.. ..  
. . ..
.
In order for a to be non-trivial, i.e., for a mode to exist, it is necessary that

det M(kz ) = 0. (7.2.27)

Since det M(kz ) is a function of kz , the roots of (7.2.27) can be solved for numerically.
The roots are values of kz that satisfy the guidance condition of the strip. These include the
higher order modes plus the fundamental modes.
With the weighting function as defined by (7.2.21), M in (7.2.25) is a symmetric ma-
trix. The reflection coefficients in (7.2.15) are generalized reflection coefficients for a layered
medium. For a substrate backed by a ground plane, we have
TM
R01 + e2ik1y h TE
R01 − e2ik1y h
R̃T M = , R̃T E = . (7.2.28)
1 + R01M e2ik1y h
T 1 − R01E e2ik1y h
T
In the above, h is the thickness of the substrate, and
pj kiy − pi kjy
Rij = , (7.2.29)
pj kiy + pi kjy
Figure 7.11: Dispersion characteristics of the fundamental and the first two higher order
modes on a wide single microstrip line (1 = 9.7, w = 9.15mm), dashed curve: cover height
h = 3d = 1.92mm (from Jansen [14], reproduced by G. Papadopoulos).
p
where pi = µi for TE waves, pi = i for TM waves, kiy = ki2 − kx2 − kz2 .
In general, for guidance to be possible, we require that
k0 < kz < k1 , (7.2.30)
where k1 is the wave number of the substrate region and k0 is the free-space wave number.
Hence, the solution of (7.2.27) is sought only for kz within the window defined by (7.2.30).
However, this window can be made even smaller with the following consideration.
A substrate backed by a ground plane has a fundamental TM0 mode with no cutoff. In
order for a mode to be guided by a microstrip line, the mode on the microstrip line must
have aqslower phase velocity than the TM0 mode [16]. If this TM0 mode has a wave number
ksp = 2 + k 2 , then
kxp zp
k0 < ksp < kz < k1 . (7.2.31)
ksp > k0 , because TM0 mode is a guided mode, and its phase velocity must be slower
than that of air. Equation (7.2.31) leads to a narrower window of search for kz . Equation
(7.2.31) ensures that the microstrip mode does not leak energy to the fundamental TM0
mode. Moreover, there are guided modes by a ground-plane backed dielectric substrates.
These modes can potentially take energy away from the modes guided by the microstrip line.
The above condition also prevents the leakage of energy from the guided microstrip line modes
to the dielectric substrate modes.
Since a microstrip line falls into the class of inhomogeneously filled waveguides, the varia-
tional method developed in Chapters 3 and 6 together with the finite element method can be
used to solve for the wavenumber kz ’s of the guided mode. In this case, the partial differential
equation is converted into a matrix eigenvalue problem where the eigenvalues are from kz2 ,
and they can be found explicitly, rather than through a root searching method above.
7.3 Microstrip Patch Revisited

Armed with the spectral domain technique, we are well equipped to analyze the microstrip
patch over a layered medium [26]. To this end, we let (here, we assume that the z axis to be
vertical as is usually the case).
Z ∞Z ∞
E0z (r) = dkx dky eikx x+iky y Ẽ0z (kx , ky , z) (7.3.1)
−∞ −∞
Z ∞ Z ∞
H0z (r) = dkx dky eikx x+iky y H̃0z (kx , ky , z) (7.3.2)
−∞ −∞
Substituting the above into the wave equation, we deduce that

2
d 2 2
+ kz Ẽ0z (kx , ky , z) = 0 (7.3.3)
dz 2

d2 2 2
+ k z H̃0z (kx , ky , z) = 0 (7.3.4)
dz 2
where kz2 = k02 − kx2 − ky2 . Hence, we can write

Z ∞ Z ∞
E0z (r) = ± dkx dky eiks ·rs +ikz |z| e0z (ks ) (7.3.5)
−∞ −∞
Z ∞ Z ∞
H0z (r) = dkx dky eiks ·rs +ikz |z| h0z (ks ) (7.3.6)
−∞ −∞
The ± sign indicates the odd symmetry of E0z (r) about z = 0 plane. When the patch is
placed on top of a layered medium, the above becomes
Z ∞Z ∞

E0z (r) = dks eiks ·rs e0z (ks ) ± eikz |z| − R̃T M eikz (z+d)+ikz |d| (7.3.7)
−∞ −∞
Z ∞ Z ∞
H0z (r) = dks eiks ·rs h0z (ks ) eikz |z| + R̃T E eikz (z+d)+ikz |d| (7.3.8)
−∞ −∞
The phase factors for the reflected waves are appropriately chosen so that R̃T M and R̃T E
represent the reflection coefficients.
Using the fact that

1 ∂
Ês (r) = 2 Êz − iωµ0 ẑ × ∇s Ĥz (7.3.9)
ks ∂z

1 ∂
Ĥs (r) = Ĥz + iω0 ẑ × ∇ Ê
s z (7.3.10)
ks2 ∂z
and applying the above to the integrands of spectral integrals, we obtain
Z ∞Z ∞
1
E0s (r) = dks eiks ·rs 2 − kz ks e0z (ks ) + ωµ0 ẑ × ks hoz (ks ) ekz |z| (7.3.11)
−∞ −∞ ks
Z ∞ Z ∞
1
H0s (r) = dks eiks ·rs ∓ kz ks h0z (ks ) ∓ ω0 ẑ × ks eoz (ks ) ekz |z| (7.3.12)
−∞ −∞ ks2
Using the fact that
J0s = ẑ × (H0s+ − H0s− )|z=0 (7.3.13)
the above can be written as
ZZ ∞
1
J0s (rs ) = dks eiks ·rs [2ω0 ks e0z − 2kz ẑ × ks h0z ] (7.3.14)
−∞ ks2
The above can be written using vector Fourier transform [40]
ZZ ∞
J0s (rs ) = dks F(ks , rs ) · K0 (ks ) (7.3.15)
−∞
where

1 1 kx ky
F(ks , rs ) = [ks , −ẑ × ks ]eiks ·rs = eiks ·rs (7.3.16)
ks ks ky −kx

1 2ω0 e0z
K0 (ks ) = (7.3.17)
ks 2kz h0z
An inverse vector Fourier transform exists as
ZZ ∞
1
K0 (ks ) = drs F(ks , −rs ) · J0s (rs ) (7.3.18)
(2π)2 −∞
We can also write

ZZ ∞
E0s = dks F(ks , rs ) · G(kz , z) · K0 (ks ) (7.3.19)
−∞
where
kz
0
G(kz ) = − 2ω0
ωµ0 eikz |z| (7.3.20)
0 2kz
The above constitutes a compact way to express the field and current in terms of their spectral
domain quantities.
7.3.1 Integral Equation for the Resonance Case

The resonance of a microstrip patch has been calculated before [17, 18, 22, 26]
ZZ ∞
E0s (rs , z = 0) = 0 = dks F(ks , rs ) · G(kz , z = 0) · K0 (ks ), rs ∈ P (7.3.21)
−∞
J0 (rs ) = 0, rs ∈
/P (7.3.22)
where P is the patch surface. By using Galerkin’s method, the above can be converted into
a matrix equation
A(ω) · a = 0 (7.3.23)
A non-trivial a exists only if

det A(ω) = 0 (7.3.24)
The above equation can be solved with a zero-searching method to obtain the resonance
frequencies of the microstrip patch.
7.3.2 Integral Equation for the Excitation Case

If the patch is excited by an incident field, say due to a probe source, then the integral
equation for excitation is [7, 19]:
n̂ × [E0s (rs , z = 0) + Einc (rs , z = 0)] = 0 (7.3.25)
or after using (7.3.21)

ZZ ∞
−n̂ × Einc (rs , z = 0) = n̂ × dks F(ks , rs ) · G(kz , z = 0) · K0 (ks ) (7.3.26)
−∞
By Galerkin’s method, the above can be converted to a matrix equation:
A·a=b (7.3.27)
one can solve the above to obtain the current on the patch, and then calculate the field
everywhere.
7.3.3 Far Field Calculation

The far field of the Fourier integral given in (7.3.7) and (7.3.8) can be found using the
stationary phase method [32]. We will assume that h is small so that when (x, y, z) → ∞, the
dominant variation of the integrand comes from eiks ·rs and eikz z . To this end, we can write
(7.3.7) with d = 0 as
ZZ ∞
E0z (r) = dks eiks ·rs +ikz z eoz (ks ) 1 − R̃T M (7.3.28)
−∞
θ
y
The exponential function is given by
eikx x+iky y+ikz z (7.3.29)
The stationary phase point is given by
∂ ∂
[kx x + ky y + kz z] = 0, [kx x + ky y + kz z] = 0 (7.3.30)
∂kx ∂ky
yielding
kx ky
x− q z = 0, y− q z=0 (7.3.31)
k 2 − kx2 − ky2 k 2 − kx2 − ky2
If we set
kx = k0 sin θ cos φ ky = k0 sin θ sin φ (7.3.32)
then
kz = k0 cos θ (7.3.33)
and (7.3.31) is satisfied. Therefore the above is the stationary phase point of the integrand.
Most of the contribution of the integral to (7.3.28) will come from this point. Hence, we can
approximate the slowly varying part of the integrand by its value at this point and rewrite
(7.3.28) approximately as
h iZ Z +∞
eiks ·rs +ikz z
E0z (r) ∼ kzs e0z (kxs , kys ) 1 − R̃T M (kxs , kys ) dks (7.3.34)
−∞ kz
where kxs , kys , and kzs are given by (7.3.32) and (7.3.33). Making use of the Weyl identity,
Z Z +∞
eik0 r i eiks ·rs +ikz |z|
= dks (7.3.35)
r 2π −∞ kz
we can rewrite (7.3.34) as
2π h i eik0 r
E0z (r) ∼ kzs e0z (kxs , kys ) 1 − R̃T M (kxs , kys ) (7.3.36)
i r
By the same token,
2π h i eik0 r
H0z (r) ∼ kzs h0z (kxs , kys ) 1 + R̃T E (kxs , kys ) (7.3.37)
i r
The far field appears as a spherical wave. Hence, the electric field of the far field is of the
form
E ≈ θ̂Eθ + φ̂Eφ (7.3.38)
with
1
H ≈ −r̂ × E/η = φ̂Eθ − θ̂Eφ (7.3.39)
η
From the above, we gather that
1
E0z ≈ −Eθ sin θ, H0z ≈ Eφ sin θ (7.3.40)
η
Hence, we can get (Eθ , Eφ ) from (E0z , H0z ) in the far field and compute the total radiation
power.
7.4 Edge Condition

In many numerical and analytic methods, it is useful to know how the charge and the current
behave near the sharp edge of a geometry. This will help in the choice of the correct basis
functions to improve the convergence of the numerical methods. Also, the edge condition has
to be imposed for a boundary value problem in order to guarantee uniqueness of the solution.
This problem has been studied by Rayleigh, Meixner, Maue, Jones, and Silver and Heins,
Hayashi, and many more. References can be found in Heins and Silver [43], and Hayashi [44].
The field near a sharp edge is singular because of charge accumulation into a singular point.
The singular behavior can be ascertained by solving Laplace’s equation in the vicinity of the
sharp edge. This is because the singular behavior of the field is entirely a local phenomenon.
Consider a wedge with an angle γ as shown. The Φ that satisfies Laplace’s equation
∇2 Φ = 0 is given by
Φ = Aρ±α e±inα , α>0 (7.4.1)
Using the fact that
1 ∂ ∂ 1 ∂2
∇2 = ρ + 2 2 (7.4.2)
ρ ∂ρ ∂ρ ρ ∂φ
in cylindrical coordinates, it can be easily shown that ∇2 Φ = 0 by direct substitution. A
solution that satisfies the boundary condition that Φ = V on the wedge surface is
Φ = V + aρα sin (αφ) (7.4.3)

y
ρ

 V x

γ α
0 1/2
π/2 2/3
π 1
3π/2 2
Table 7.1: The singularities associated with different wedge angles.
with that requirement that sin [α(2π − γ)] = 0, or that

π
α(2π − γ) = π, α= (7.4.4)
2π − γ
for the minimal α and that sin (αφ) > 0 for 0 < φ < 2π − γ.
We ignore the ρ−α term in (7.4.1) since |∇Φ|2 has to be square integrable because it
represents the finite stored energy in the field. Moreover, we pick minimal α so that the
electric field E = −∇Φ given by (7.4.3) is simple and does not have scallop patterns.
Hence for the different wedge angles, we have the singularities as shown in Table 7.1. For
a thin wedge with γ = 0,
1 φ
Φ = V + aρ sin
2 (7.4.5)
2
the corresponding electrostatic field

a 1 φ a 1 φ
E = −∇Φ = −ρ̂ ρ− 2 sin + φ̂ ρ− 2 cos (7.4.6)
2 2 2 2
Since the charge is proportional to n̂ · E, the surface charge density σ is such that
1
σ∼x− 2 (7.4.7)
near the edge. If a current flows in the x direction, from ∇ · J = iω%, the current
1
Jx ∼x 2 (7.4.8)
near the edge. If the current is flowing in the z direction into the paper, then
Jz = σvz (7.4.9)
Hence,
1
Jz ∼x− 2 (7.4.10)
7.5 Discontinuities in Microstrip Lines

Microstrip lines are some of the easiest waveguides to fabricate. However, they are also some of
the hardest to analyze. Due to the ease in their fabrication, microwave integrated circuits have
come in a rich variety of shapes. One example is the variety of microstrip line discontinuities.
These discontinuities are in general hard to analyze. Usually, numerical methods have to
be employed for their analysis. This is especially true for the full-wave analysis whereby all
electrodynamic effects are accounted for. Much work on this topic can be found in [45–55]
Microstrip line discontinuities are often modeled by lumped elements in the transmission
line model. In the following, we shall discuss some typical discontinuities encountered in
microwave integrated circuits.
Figure 7.12: An open-end discontinuity.
7.5.1 An Open-End Discontinuity

A simple way to make an open circuit in a microstrip line is to let the microstrip line terminate
in an open end. Open circuits are often used as a stub tuner in microwave integrated circuits.
The abrupt termination of a microstrip line intercepts the flow of current on the line. This
gives rise to charge accumulation, which can be modeled by a capacitance. The equivalent
circuit for an open end microstrip line is then a transmission line terminated by a capacitor,
or an open circuited transmission line whose length is slightly longer than the physical length
of the microstrip line. The capacitor is also called a fringing field capacitor, as the excess
charge at the termination gives rise to excess fringing field at the end of the line. These
fringing capacitance can be found by numerical calculations which are sometimes replaced by
analytical or semi-empirical formulas.
7.5.2 A Step Discontinuity

Another commonly encountered discontinuity in a microstrip line is a step discontinuity. A
step discontinuity has excess charges on its excesss edges, and hence can be modeled by a
Figure 7.13: Equivalent circuit models for an open end microstrip line.
Figure 7.14: A step discontinuity in a microstrip line.
shunt capacitance. This kind of discontinuity is encountered in the design of a quarter-wave

transformer for instance. A more sophisticated model consists of two series inductors as well
as a shunt capacitor.
Z Z
Figure 7.15: Equivalent circuit models for a step discontinuity.
7.5.3 A Gap Discontinuity

A gap discontinuity is deliberately introduced in a microstrip line, for example, in filter design.
A gap discontinuity can be thought of as two open end discontinuities brought close together.
The proximity of two open end discontinuities results in capacitive coupling between the two
open ends. Therefore, the equivalent circuit consists of two shunt capacitors and a series
capacitor.
7.5.4 A Slit Discontinuity

A slit discontinuity has very little effect on a low frequency signal, but it will scatter a high
frequency signal. At low frequency, the current can flow continuously along the line, and sees
Figure 7.16: A gap discontinuity.
Figure 7.17: The equivalent circuit for a gap discontinuity.
a matched line at the other end. At higher frequency, the restricted channel is felt and the
slit reflects a signal impinging on it. Hence, it acts as a low-pass filter. it can be modeled by
an T equivalent circuit with two series inductances and a shunt capacitance.
7.5.5 A Microstrip Bend

A microstrip bend can be a corner bend or a chamfered bend. Both microstrip bends can
be modeled by a T circuit with two series inductors and a shunt capacitor. A corner bend
has a sharp corner favoring the build up of excessive charges giving rise to a larger shunt
capacitance. A chamfered bend reduces this capacitance and mismatch, even less so than a
rounded bend.
All guided waves around a bend in an open waveguide radiate. A chamfered bend has
been found to reduce the radiation loss as well.
7.5.6 A T Junction
A T junction in microwave integrated circuits is commonly encountered, e.g., in a stub tuner.
It can be modeled by series inductances plus a shunt capacitance. A cross can also be similarly
modeled.
Figure 7.18: A slit discontinuity.

L L
Z1 1
Figure 7.19: The equivalent circuit for a slit discontinuity.
(a) (b)
Figure 7.20: (a) A corner bend. (b) A chamfered bend.
7.6 Directional Coupler Using Microstrip Line

Due to the simplicity of the microstrip waveguide, a new class of directional coupler has been
proposed. Some of these papers can be found in [56–58]. When two microstrip lines are put
side by side in parallel with respect to each other, electromagnetic coupling exists between the
two lines. Since one line has one fundamental mode, two lines would have two fundamental
modes. For symmetric lines, these two fundamental modes are the odd and the even modes.
For two identical lines, the electric field for the odd mode is odd-symmetric about the plane
of symmetry while for the even mode, it is even-symmetric about the same plane. Hence,
we can use the image theorem to analyze such a problem: A PMC wall can be placed at the
plane of symmetry for the even symmetric mode, while a PEC wall can be placed at the same
plane for the odd symmetric mode.
Figure 7.21: A T junction in microwave integrated circuits.

Figure 7.22: Equivalent circuit for a T junction.
+ 1 2 –
V1 V2
0 l 0
– Plane of symmetry
V1
1
A A′
1
–
V3 0 –
0 V4
3 4
Figure 7.23: A directional coupler made of microstrip lines.
The directional coupler is a four-port network. We can analyze it as a linear superposition

of two two-port network, one with a PMC at AA0 , and another one with a PEC at AA0 . The
PMC case is equivalent to an even-mode excitation of the geometry. This is the same as, say,
an incident wave of V + at both ports 1 and 3 of the network. The PEC case corresponds
to an odd-mode excitation, which is the same as a V + incident wave at port 1 and a −V +
incident wave at port 3. The linear superposition of these two excitations correspond to a
2V + excitation at port 1 and none at the other ports.
The PMC case corresponds to only even mode propagation, and we can use the bounce
diagram approach to write down the reflected waves at ports 1 and 2. We assume that the
fundamental mode from the input port will undergo a reflection when it enters the coupled
e
mode region. Hence, the reflection coefficient at port 1 can be expressed as the S11 coefficient
given as
e 2ike l e e
e e T01 e R10 T10
S11 = R01 + e . (7.6.1)
1 − (R10 ) e e l
2 2ik
The transmission coefficient from port 1 to port 2 can be expressed as
e e ike l
e T01 T10 e
S12 = e . (7.6.2)
1 − (R10 )2 e2ike l
In the above
e Ze − Zc e e
R01 = , R10 = −R01 , (7.6.3a)
Ze + Zc
e e e e
T01 = 1 + R01 , T10 = 1 + R10 , (7.6.3b)
where Ze is the characteristic impedance of the even mode while Zc is the characteristic
impedance of the feedline.
For the PEC case, a similar analysis shows that
o 2iko l o o
o o T01 e R10 T10
S11 = R01 + o )2 e2iko l , (7.6.4)
1 − (R10
o o iko l
o T01 T10 e
S12 = o )2 e2iko l , (7.6.5)
1 − (R10
where
o Zo − Zc
R01 = , (7.6.6)
Zo + Zc
and so on.
The key to the design of the directional coupler is to let Zc be the geometric mean of Zo
and Ze , i.e., p
Zc = Zo Ze . (7.6.7)
o e
In this case, R01 = −R01 . Furthermore, if we make ke = ko , then it can be shown that
o e
S11 = −S11 . (7.6.8)
Superposing the even and the odd solutions, we obtain that V1+ = 2V + , and V1− = (S11
e
+
o +
S11 )V = 0. Similarly, the transmitted wave at port 4 is given by
V4− = S12
e
V + − S12
o
V +, (7.6.9)
e e o o
when ke = ko , and that (7.6.7) is satisfied, it is clear that T01 T10 = T01 T10 . Therefore,
e o −
S12 = S12 and from (7.6.9), we see that V4 = 0. At port 3, we have
1 e
V3− = S11
e
V + − S11
o
V+ = o
(S − S11 )V1+ = S13 V1+ , (7.6.10)
2 11
where
1 e o e
S13 = (S − S11 ) = S11 . (7.6.11)
2 11
Similarly, we have
1 e o e
S12 = (S + S12 ) = S12 . (7.6.12)
2 12
Therefore, the above works as a perfect directional coupler, where nothing is coupled to
port 4 from port 1, while some signal is transmitted to port 2 and port 3. If the coupled
line is in a homogeneous region, it is clear that both the odd and the even modes are TEM
modes, and hence ke = ko = k. However, if the coupled line is fabricated on a substrate as
in the case of a microstrip line, then ke 6= ko because the field distributions of the two modes
are different. However, ke can be made to be close to ko by covering the microstrip line with
a superstrate with the same permittivity as the substrate.
The directivity of a directional coupler is the ratio of the | V3− | to | V4− |. In the ideal
case, it is infinite. The coupling coefficient is the ratio of | V3− | to | V1+ |. In this case, it is

e e 2ike l e
T01 e e
R10 T10
e
C =| S11 |= R01 +
e )2 e2ike l .
1 − (R10
(7.6.13)
e e e 2 e
Using the fact that T01 T10 = 1 − (R01 ) , we can rewrite S11 as
e 2ike l e
e e R10 e R01 (1 − e2ike l )
S11 = R01 + e = e )2 e2ike l . (7.6.14)
1 − (R10 )2 e2ike l 1 − (R10
e
| S11 | achieves a maximum at ke l ' π/2 or l ' λ/4.
7.7 A Branch Line Directional Coupler
B
+
V1 Y1
1 2
Yc
–
V1 Y2
A A′
3 4
B′
Figure 7.24: A branch line directional coupler.
A branch line coupler, like the microstrip line directional coupler, has two planes of sym-
metry. Even though a microstrip line directional coupler has two planes of symmetry, we
have analyzed it using only one plane of symmetry. However, for a branch line coupler, it is
simpler to analyze using two planes of symmetry.
On the plane AA0 , we can put either a PMC or a PEC. A PMC yields an open circuit
at AA0 or an even symmetric electric field or voltage about the AA0 . A PEC yields a closed
circuit at AA0 or an odd symmetric electric field or voltage about AA0 . The same statement
applies to the plane BB 0 . Therefore, there are altogether four possible excitations of the
above geometry.
Bibliography
[1] J. Koch, T. M. Yu, J. Gambetta, A. A. Houck, D. I. Schuster, J. Majer, A. Blais, M. H.

Devoret, S. M. Girvin, and R. J. Schoelkopf, “Charge-insensitive qubit design derived
from the Cooper pair box,” Phys. Rev. A, 76, no. 4 (2007): 042319.
[2] H. A. Wheeler, “Transmission-line properties of parallel strips separated by a dielectric

sheet,” IEEE Trans. Microwave Theory Tech., vol. MTT-13. pp. 172-185, Mar. 1965.
[3] T. G. Bryant and J. A. Weiss, “Parameters of microstrip transmission lines and of coupled
pairs of microstrip lines,” IEEE Trans. Microwave Theory Tech., vol. MTT-16. pp. 1021-
1027. Dec. 1968.
[4] M. V. Schneider, “Microstrip lines for microwave integrated circuits,” Bell Syst. Tech. J.,
vol. 48, pp. 1421-1444. May/June 1969.
[5] H. E. Green, “The numerical solution of some important transmission-line problems,”

IEEE Trans. Microwave Theory Tech., vol. MTT-13, pp. 676-692, Sept. 1965.
[6] E. Yamashita and R. Mittra, “Variational method for the analysis of microstrip lines,”
IEEE Trans. Microwave Theory Tech., vol. MTT-16. PP- 251-256, Apr. 1968.
[7] E. Yamashita, “Variational method for the analysis of microstrip-like transmission lines,”
IEEE Trans. Microwave Theory Tech., vol. MTT-16. pp. 529-535, Aug. 1968.
[8] H. E. Stinehelfer, Sr., “An accurate calculation of uniform microstrip transmission lines,”
IEEE Trans. Microwave Theory Tech., vol. MTT-16, pp. 439-444, July 1968.
[9] G. I. Zysman and D. Varon, “Wave propagation in microstrip transmission lines.” IEEE
G-MTT Int. Microwave Symp. Dig., pp. 3-9, 1969.
[10] E. J. Denlinger, “A frequency dependent solution for microstrip transmission lines,”

IEEE Trans. Microwave Theory Tech., vol. MTT-19, pp. 30-39, Jan. 1971.
[11] J. S. Hornsby and A. Gopinath, “Numerical analysis of a dielectric-loaded waveguide with

a microstrip line—Finite difference methods,” IEEE Trans. Microwave Theory Tech., vol.
MTT-17, pp. 684-690, Sept. 1969.
329
[12] H. J. Schmitt and K. H. Sarges, “Wave propagation in microstrip,” Nachrichtentech. Z.,

vol. 24, pp. 260-264, May 1971.
[13] R. Mittra and T. Itoh, “A new technique for the analysis of dispersion characteristics
of microstrip lines,” IEEE Trans. Microwave Theory Tech., vol. MTT-19, pp. 47-56, Jan.
1971.
[14] R. H. Jansen, “High-speed computation of single and coupled microstrip parameters

including dispersion, high-order modes, loss and finite strip thickness,” IEEE Trans. Mi-
crowave Theory Tech., vol. MTT-26, pp. 75-82, Feb. 1978.
[15] M. Kobayashi and F. Ando, “Dispersion characteristics of open microstrip lines,” IEEE
Trans. Microwave Theory Tech., vol. MTT-35, pp. 101-105. Feb. 1987.
[16] L. Gürel and W. C. Chew, “Guidance or resonance conditions for strips or disks embed-
ded in homogeneous and layered media,” IEEE Trans. Microwave Theory Tech., vol. 36,
no. 11, pp. 1498-1506, Nov. 1988.
[17] W. C. Chew and J. A. Kong, “Resonance of the axial-symmetric modes in microstrip

disk resonators,” J. Math. Phys., vol. 21, no. 3, pp. 582-591, Mar. 1980.
[18] W. C. Chew and J. A. Kong, “Resonance of nonaxial symmetric modes in microstrip

disk antenna.” J. Math. Phys., vol. 21, no. 10, pp. 2590-2598. Oct. 1980.
[19] W. C. Chew and J. A. Kong, “Analysis of a circular microstrip disk antenna with a thick
dielectric substrate,” IEEE Trans. Antennas Propagat., vol. AP-29, pp. 68-76, Jan. 1981.
[20] R. H. Jansen. “The spectral-domain approach for microwave integrated circuits,” IEEE
Trans. Microwave Theory Tech., vol. MTT-33. pp. 1043-1056. Oct. 1985.
[21] T. Itoh and R. Mittra, “Spectral-domain approach for calculating the dispersion char-
acteristics of rnicrostrip lines,” IEEE Trans.Microwave Theory Tech., vol. MTT-21. PP.
496-499. July 1973.
[22] T. Itoh, “Analysis of microstrip resonators,” Microwave Theory and Techniques, IEEE
Transactions on, 22, no. 11 (1974): 946-952.
[23] R. H. Jansen, “Microstrip lines with partially removed ground metallization, theory and
applications,” Arch. Elek. Uberrragung, vol. 32, pp. 485-492, Dec. 1978.
[24] T. Itch, “Spectral domain immittance approach for dispersion characteristics of general-
ized transmission lines,” IEEE Trans.Microwave Theory Teoh., vol. MTT-28, pp. 733-736,
July 1980.
[25] D. Pozar, “Input impedance and mutual coupling of rectangular microstrip antennas.”
Antennas and Propagation, IEEE Transactions on, 30, no. 6 (1982): 1191-1196.
[26] W. C. Chew and Q. Liu. “Resonance frequency of a rectangular microstrip patch,” IEEE
Trans. Antennas Propagat., vol. 36, Aug.1988.
[27] G. Kergonou, M. Drissi, T. Zak, and C. Xavier, “Frequency dependence in high speed
interconnections,” Electromagnetic Compatibility, 2001. EMC. 2001 IEEE International
Symposium on, vol. 1, pp. 632-634. IEEE, 2001.
[28] R. A. York, R. C. Compton, and B. J. Rubin. “Experimental verification of the 2-D

rooftop approach for modeling microstrip patch antennas,” Antennas and Propagation,
IEEE Transactions on 39, no. 5 (1991): 690-694.
[29] S. Sun and L. Zhu, “Compact dual-band microstrip bandpass filter without external
feeds,” Microwave and Wireless Components Letters, IEEE 15, no. 10 (2005): 644-646.
[30] C. M. Bender and S. A. Orszag, Advanced mathematical methods for scientists and en-
gineers I: Asymptotic methods and perturbation theory, Vol. 1. Springer, 1999.
[31] S. Y. Poh, W. C. Chew. and J. A. Kong, “Approximate formulas for line capacitance
and characteristic impedance of microstrip line,” IEEE Trans. Microwave Theory Tech.,
vol. MTT-29, pp.135-142, Feb. 1981.
[32] W. C. Chew, Waves and Fields in Inhomogeneous Media, Van Nostrand Reinhold, New
York, 1990. Reprinted by IEEE Press, 1995.
[33] J. A. Kong, Electromagnetic Wave Theory. New York: Wiley, 1986.
[34] J. D. Jackson, Classical Electrodynamics, Wiley, New York, NY, 1962, p. 51.
[35] J. J. Thomson, footnote on p. 154 of Maxwells Treatise On Electricity and Magnetism,

(Dover, New York, NY, 1954).
[36] Y. L. Chow, Y. F. Lan, and D. G. Fang, “Capacitance and its upper and lower bounds by
the method of optimized simulated images,” J. Appl. Phys., 53, no. 11 (1982): 7144-7148.
[37] W. E. Parr, “Upper and lower bounds for the capacitance of the regular solids,” J. Soc.
Indust. Appl. Math., 9, no. 3 (1961): 334-386.
[38] N. Wiener and E. Hopf, “Ueber eine Klasses singulärer Integralgleichungen,” S.-B.
Deutsch. Akad. Wiss. Berlin KI. Math. Phys. Tech., pp. 696-706 1931.
[39] W. C. Chew, “Mixed boundary value problems in microstrip and geophysical probing
applications,” Ph.D. Thesis, Massachusetts Institue of Technology, Cambride, MA, 1980.
[40] W. C. Chew and T. M. Habashy, “The use of vector transforms in solving some electro-
magnetic scattering problems,” IEEE Trans. Antennas Propagat., vol. AP-34, pp. 871-879,
July 1986.
[41] W. C. Chew and L. Gürel, “Reflection and transmission operators for strips or disks
embedded in homogeneous and layered media,” IEEE Trans. Microwave Theory Tech.,
vol. MTT-36, no. 11, pp. 1488-1497, Nov. 1988.
[42] K. C. Gupta, R. Garg, and I. J. Bahl, Microstrip lines and slotlines, Artech House,
Norwood, MA, 1979.
[43] A. E. Heins and S. Silver, “The edge conditions and field representation theorem in the
theory of electromagnetic diffraction, Proc. Cambridge Phil. Soc., Vol. 51, 149161, 1955.
[44] Y. Hayashi, “Electromagnetic theory based on integral representation of fields and anal-
ysis of scattering by open boundary,” Progress In Electromagnetics Research, PIER 13,
186, 1996.
[45] R. Garg and I. J. Bahl, “Microstrip discontinuities,” International Journal of Electronics,

45:1, 81-87, 1978.
[46] T. S. Chu and T. Itoh. “Generalized scattering matrix method for analysis of cascaded
and offset microstrip step discontinuities,” IEEE Trans. MTT, 34-2: 280-284, 1986.
[47] R. W. Jackson, “Full-wave, finite element analysis of irregular microstrip discontinuities,”

Microwave Theory and Techniques, IEEE Trans. MTT, 37-1: 81-89, 1989.
[48] X. Zhang and K. K. Mei. “Time-domain finite difference approach to the calculation of
the frequency-dependent characteristics of microstrip discontinuities,” IEEE Trans. MTT,
36-12: 1775-1787, 1988.
[49] M. Kirschning, R. H. Jansen, and N. H. L. Koster, “Measurement and computer-aided

modeling of microstrip discontinuities by an improved resonator method,” IEEE MTT-S,
International Microwave Symposium Digest, 1983.
[50] C. J. Railton and T. Rozzi. “The rigorous analysis of cascaded step discontinuities in
microstrip,” IEEE Trans. MTT, 36-7, 1177-1185, 1988.
[51] A. K. Verma, H. Singh, and Y. K. Awasthi, “Circuit model of multilayer microstrip step
discontinuity using single-layer reduction formulation,” Electromagnetics, 29-6: 483-498,
2009.
[52] R. J. P. Douville and D. S. James, “Experimental study of symmetric microstrip bends

and their compensation,” IEEE Trans. MTT, 26, no. 3: 175-182, 1978.
[53] J. Moore and H. Ling, “Characterization of a 90 microstrip bend with arbitrary miter via
the time-domain finite difference method,” IEEE Trans. MTT,38, no. 4 (1990): 405-410.
[54] R. Mehran, “Calculation of microstrip bends and Y-junctions with arbitrary angle,”
IEEE Trans. Microwave Theory and Techniques, 6, no. 6 (1978): 400-405.
[55] P. H. Harms and R. Mittra, “Equivalent circuits for multiconductor microstrip bend
discontinuities,” IEEE Trans. Microwave Theory Tech., 41, no. 1 (1993): 62-69.
[56] F. C. De Ronde, “A new class of microstrip directional couplers,” In Microwave Sympo-

sium, G-MTT 1970 International, pp. 184-189. IEEE, 1970.
[57] M. Dydyk, “Microstrip directional couplers with ideal performance via single-element
compensation,” Microwave Theory and Techniques, IEEE Transactions on 47, no. 6
(1999): 956-964.
[58] C. Caloz and T. Itoh, “A novel mixed conventional microstrip and composite right/left-
handed backward-wave directional coupler with broadband and tight coupling character-
istics,” Microwave and Wireless Components Letters, IEEE, 14, no. 1 (2004): 31-33.
Chapter 8
Solitons
8.1 Optical Solitons

Dispersions in an optical fiber is the prime reason for limiting its bandwidth. The dispersion
causes an optical pulse to distort or spread as it propagates over long distances. Moreover,
the loss of the optical fiber also causes the pulse to spread. Therefore, repeaters are needed
every several tens of kilometers to rejuvenate the pulses. When the pulses are narrower in
order to facilitate higher transmission rates, dispersion effects become more pronounced and
repeaters have to be closely space. The spacing of the repeaters is then inversely proportional
to the square of the pulse width.
An optical soliton makes use of the nonlinear effect in an optical fiber to propagate a pulse
with no distortion. The nonlinear effect is used to counter the dispersion effect so that the
pulse propagates with little or no distortion. The nonlinear effect is proportional to the field
strength of a mode in the optical fiber. Hence, loss in the fiber will eventually distort the pulse.
However, using the Raman effect in the fiber, optical solitons can be continuously pumped
by a pumping lightwave which is simultaneously transmitted through the fiber together with
the solitons. Hence, an all optical system without repeaters is possible.
The nonlinear property of an optical fiber comes from the Kerr effect which produces a
change in the refractive index of glass due to the deformation of electron orbits by the electric
field of light. The refractive index of glass is then n = n0 + n2 |E|2 where n2 is the Kerr
coefficient, and E is the electric field. n2 is usually of the order of 10−22 (m/V )2 and E is of
the order of 106 V /m. Therefore, the change in the refractive index is 10−10 . Even though
this is a small change, the high operating frequency of an optical fiber makes this change
significant.
8.2 The Korteweg de Vries Equation

A soliton or a solitary wave is a result of nonlinear phenomena. A nonlinear partial differential
equation that admits a soliton as a solution is the KdV (Korteweg de Vries) equation. We
shall motivate the equation here.
335
A wave solution to a linear wave equation has the following form

φ(x, t) = φ(x − vt) (8.2.1)
where v is the velocity of the wave. The above describes a right-going wave and it satisfies
the equation
∂φ ∂φ
+v = 0. (8.2.2)
∂t ∂x
If one observes this wave in a moving coordinate system such that x0 = x − v0 t, then equation
(8.2.1) becomes
φ(x0 , t) = φ[x0 − (v − v0 )t]. (8.2.3)
and
∂φ ∂φ
+ δv 0 = 0, (8.2.4)
∂t ∂x
where δv = v − v0 . If v0 = v, then in the moving coordinate system,
∂
φ = 0, (8.2.5)
∂t
or that the field remains stationary, and φ = φ(x0 ) only.
In Equation (8.2.4), δv denotes the “extra” velocity that the wave is moving with respect
to the moving coordinates. If we are in a moving coordinates such that this “extra” velocity
is dependent on the amplitude of the wave, then (8.2.4) can be rewritten as
∂φ ∂φ
+ δ1 φ 0 = 0. (8.2.6)
∂t ∂x
which is a nonlinear equation. The above equation says that the field with a higher amplitude
moves faster than the field with a lower amplitude. The above expression also implies that
the field cannot remain stationary in this moving coordinates. In other words, φ is a function
of both x0 and t. It further means that if we have a symmetric pulse to begin with, the pulse
will start to lean over as shown in Figure 1, generating a sharp shock wavefront. The shock
wavefront has higher spectral components. Consequently, the high frequency component of
the pulse has increased due to nonlinearity.
A pulse shape can also be distorted by a dispersive effect. A medium is dispersive if the
wave speed is a function of the wavenumber of the wave. In a dispersive medium, the wave
velocity has a higher order term that is proportional to −δ2 kx2 . Therefore, a dispersive term
can be added to (8.2.6) resulting in
∂φ ∂φ ∂3φ
+ δ1 φ 0 + δ2 03 = 0, (8.2.7)
∂t ∂x ∂x
since ∂ 2 /∂x02 = −kx2 . The dispersive effect implies that the waves with high wave numbers
travel at a smaller velocity. Therefore, it tends to counter the pulse sharpening effect due to
nonlinearity. In fact a solution exists for (8.2.7) for a pulse that propagates without distortion.
1 1
If we let x0 = δ23 ξ, φ = δ23 δ1−1 ψ, then (8.2.7) becomes
∂ψ ∂ψ ∂ 3 ψ
+ψ + = 0, (8.2.8)
∂t ∂ξ ∂ξ 3
Solitons 337
Figure 8.1: The leaning over of a symmetric pulse due to the nonlinearity in Equation (8.2.6).
which is the KdV (Korteweg de Vries) equation. It can be shown to have the solitary wave
solution √
2 v0
ψ(t, ξ) = 3v0 sech (ξ − v0 t) . (8.2.9)
2
This soliton has a velocity v0 with respect to the moving frame ξ which is related to x0 via
1 1
x0 = δ23 ξ. Therefore it has a velocity of δ23 v0 with respect to the moving frame x0 and a
1
velocity of v + δ23 v0 with respect to the stationary frame x. Note that since sech2 x decays
√
exponentially for large arguments, the width of the soliton is 2/ v0 and its amplitude is 3v0 .
A larger amplitude soliton moves faster and has a narrower width.
8.3 Derivation of the Nonlinear Schrödinger Equation

The propagation of solitons in optical fiber is governed by the nonlinear Schrödinger (NLS)
equation. The nonlear Schrödinger equation is due to nonlinear effect on electromagnetic
wave propagation in an optical fiber. Starting with the Maxwell’s equations and assuming
that ∂ /∂x = ∂ /∂y = 0, for an electric field polarized in the y direction, we have
∂ ∂
E y = Bx , (8.3.1)
∂z ∂t
∂ ∂
Bx = µ0 Dy , (8.3.2)
∂z ∂t
where µ0 is assumed to be a constant. Eliminating Bx , we have
∂2 ∂2
Ey = µ0 Dy . (8.3.3)
∂z 2 ∂t2
For a medium, Dy = 0 n2 Ey , where n is the refractive index. For a Kerr medium, n is
nonlinearly related to the electric field , or
2
n = n0 + n2 |Ey | . (8.3.4)
2
The Kerr effect is a very small effect such that n2 |Ey | << n0 . But at optical frequencies,
a small change in the refractive index can have a significant effect on the phase shift of the
wave due to the short wavelengths involved. We can approximate
2 2
n2 Ey = (n0 + n2 |Ey | )2 Ey ≈ (n20 + 2n0 n2 |Ey | )Ey . (8.3.5)
Consequently, we can write (8.3.3) as
∂2 1 ∂2 n2 ∂ 2 2n0 n2 ∂ 2 2
Ey = 2 2 n2 Ey = 20 2 Ey + |Ey | Ey . (8.3.6)
∂z 2 c0 ∂t c0 ∂t c20 ∂t2
Next, we assume that
Ey = φ(z, t)ei(k0 z−ω0 t) , (8.3.7)
where φ(z, t) is a slowly varying function of z and t. It is also the envelope function of a wave.
2
∂2 ∂ ∂ 2
Ey = φ − 2iω 0 φ − ω 0 φ ei(k0 z−ω0 t) , (8.3.8)
∂t2 ∂t2 ∂t

∂2 ∂2 ∂ 2
Ey = φ + 2ik 0 φ − k 0 φ ei(k0 z−ω0 t) . (8.3.9)
∂z 2 ∂z 2 ∂z
We assume that envelope function φ(z, t) is slowly varying compared to exp[i(k0 z − ω0 t)] so
that
∂2
φ ω02 φ, (8.3.10)
∂t2
∂2
φ k02 φ. (8.3.11)
∂z 2

∂2 ∂
Ey ≈ −2iω0 φ − ω0 φ ei(k0 z−ω0 t) ,
2
(8.3.12)
∂t2 ∂t

∂2 ∂
Ey ≈ 2ik0 φ − k02 φ ei(k0 z−ω0 t) . (8.3.13)
∂z 2 ∂t
We assume that the rapidly varying solution satisfies k0 = n0 ω0 /c0 so that

∂2 n20 ∂ 2 ∂ n20 ∂
Ey − 2 2 Ey = 2i k0 φ + 2 ω0 φ ei(k0 z−ω0 t)
∂z 2 c0 ∂t ∂z c0 ∂t

∂ n0 ∂
= 2ik0 φ+ φ ei(k0 z−ω0 t) . (8.3.14)
∂z c0 ∂t
Solitons 339
The above can be used in (8.3.6) to arrive at

∂ n0 ∂ n0 n2 ∂ 2 2
ik0 φ+ φ ei(k0 z−ω0 t) = 2 |Ey | Ey . (8.3.15)
∂z c0 ∂t c0 ∂t2
2 2
Since |Ey | = |φ| is slowly varying compared to Ey itself, we have
∂2 2 2 ∂
2
2
2
|Ey | Ey ≈ |Ey | Ey = −ω02 |φ| φei(k0 z−ω0 t) . (8.3.16)
∂t ∂t2
Using (8.3.16) in (8.3.15), we have

∂ n0 ∂ k0 n2 2
φ+ φ =i |φ| φ. (8.3.17)
∂z c0 ∂t n0
The above is the equation for the envelope function φ(z, t) when no dispertion is assumed
in the medium. So when nonlinearity is absent, the pertinent equation is

∂ n0 ∂
φ+ φ = 0, (8.3.18)
∂z c0 ∂t
which is the advective equation governing the propagation of a distortionless pulse. In the
above, the dispersion relation is
n0
k= ω, (8.3.19)
c0
i.e., there is a linear relationship between k and ω. As a result, both the phase velocity (ω/k)
and the group velocity (dω/dk) are independent of frequencies.
8.3.1 Dispersive effect

Dispersion occurs when n0 is a function of ω so that k is not linearly proportional to ω
anymore. In this case, Dy = 0 n2 ? Ey , where “?” means “convolves”. We can let
Z ∞ Z ∞
Ey (z, t) = dk dω Ẽy (k, ω)ei(kz−ωt) , (8.3.20)
−∞ −∞
and assuming the absence of Kerr effect or the nonlinear term, (8.3.3) becomes
ω2 2
k 2 Ẽy = n (ω)Ẽy . (8.3.21)
c20 0
n0 (ω)
k= ω. (8.3.22)
c0
We can rewrite (8.3.22), using Taylor expansions of its righthand side, as

1
k = k0 + k00 (ω − w0 ) + k000 (ω − ω0 )2 + . . . , (8.3.23)
2
where we have assumed that

n0 (ω0 ) dk d2 k
k0 = ω0 , k00 = , k000 = . (8.3.24)
c0 dω dω 2
ω=ω0 ω=ω0
Squaring (8.3.23), and putting it back into (8.3.21), we have
[k 2 − k02 − 2k0 k00 (ω − ω0 ) − (k00 )2 (ω − ω0 )2 − k0 k000 (ω − ω0 )2 + . . . ]Ẽy = 0. (8.3.25)
Furthermore, we can write k 2 −k02 = (k−k0 )2 k0 +(k−k0 )2 and, using the fact that (k−k0 )2 ≈
(k00 )2 (ω − ω0 )2 from (8.3.23), we have
[(k − k0 )2k0 − 2k0 k00 (ω − ω0 ) − k0 k000 (ω − ω0 )2 + . . . ]Ẽy = 0. (8.3.26)
We can rewrite (8.3.20) as
Z ∞ Z ∞
Ey (z, t) = ei(k0 z−ω0 t) dk dω Ẽy (k, ω)e[(k−k0 )z−(ω−ω0 )t] (8.3.27)
−∞ −∞
= φ(z, t)ei(k0 z−ω0 t) . (8.3.28)
Fourier inverse transforming (8.3.25) using (8.3.20), which is the same as Fourier inverse
transforming (8.3.21), we have

∂ ∂ ∂2
−2ik0 − 2ik0 k00 + k0 k000 2 φ(z, t) = 0 (8.3.29)
∂z ∂t ∂t
In the above analysis, we have assumed that ω ' ω0 and k ' k0 . This is the same as assuming
that φ(z, t) is slowly varying since it has only low frequency components. If k000 = 0, the above
is an alternative way of deriving (8.3.18). Equation (8.3.29) can be rewritten as

∂ 1 ∂ i 00 ∂ 2
+ + k φ(z, t) = 0 (8.3.30)
∂z vg ∂t 2 0 ∂t2
where vg = 1/k00 .
For the Kerr effect, we can assume that n2 is frequency independent, we can add the
nonlinear part to the above equation to obtain
∂ 1 ∂ k0 n2 2 i ∂2
φ+ φ=i |φ| φ − k000 2 φ (8.3.31)
∂z vg ∂t n0 2 ∂t
The above is the nonlinear Schrödinger equation.

Solitons 341
If we are
in the moving coordinate system such that vg z = z − vg t, then φ(z, t) =
z
φ z, vg − τ = φ̃(z, τ ). Moreover,
∂ ∂ φ̃ ∂ φ̃ ∂φ 1 ∂
φ=− , = + φ (8.3.32)
∂t ∂τ ∂z ∂z vg ∂t
or
∂ φ̃ k0 n2 2 i ∂2
=i |φ̃| φ̃ − k000 2 φ̃ (8.3.33)
∂z n0 2 ∂τ
The above is similar to the Schrödinger equation
∂ ~2 ∂ 2
−i~ ψ=− ψ+Vψ (8.3.34)
∂t 2m ∂x2
if we identify z with t and τ with x, and |φ̃|2 with V .
When V is negative to form a potential well, Equation (8.3.34) admits solution of the
form
ψ = Ae−iωt g(x) (8.3.35)
When g(x) is a function localized in x corresponding to bound states or trapped modes in he
potential well which are stationary states. By the same token, by k0 00 is negative, Equation
(8.3.33) admits solution of the form
φ̃ = Beikz f (vg τ )
(8.3.36)
= Beikz f (z − vg t)
where f (vg τ ) is a function localized in τ . Equation (8.3.36) corresponds to a solution.

Since the potential well is created by |φ̃|2 ,which is proportional to the field itself, this is a
self-trapping phenomenon. Conservation of energy requires that the pulse becomes narrower
when the amplitude of φ̃ becomes larger. On the other hand, the dispersive effect tends to
spread the pulse more when it becomes narrower. A final equilibrium is reached where the
pulse propagates without distortion.
If k0 00 is positive, Equation (30) cannot trap a mode. However, it can admit a solution
signified by the absence of light, hence the name dark solutions.
8.3.2 Solution of the Nonlinear Schrödinger Equation

Via a change of variables and coordinates, the nonlinear Schrödinger equation can be written
in dimensionless form as
∂ψ 1 ∂2ψ 2
i + + |ψ| ψ = 0. (8.3.37)
∂z 2 ∂T 2
A solution is a stationary solution to the above which is stationary in amplitude with respect
to z and T . Therefore, we seek a solution of the form
p
ψ(T, z) = ρ(T )eiα(z−z0 ) . (8.3.38)
√ 1 ∂2 √ 3
−α ρ + ρ + ρ 2 = 0, (8.3.39)
2 ∂T 2
where we have assumed ρ to be real-valued. The above is the same as
" #
1 1 0 2 ρ00
−α − (ρ ) − 2 + ρ = 0, (8.3.40)
8 ρ2 ρ
dρ d2 ρ
where ρ0 = dT , ρ00 = dT 2 . It can be further simplified to
1 d 1 0 2
−α + (ρ ) + ρ = 0. (8.3.41)
8 dρ ρ
Integrating the above, we have
ρ2 11 0 2
−αρ + + (ρ ) + C = 0. (8.3.42)
2 8ρ
Since ρ and ρ0 are independent of T , C is also independent of T . But since ρ → 0 when
T → ∞, and we assume that ρ0 ∼ ρ → 0 when T → ∞, then C = 0. When ρ is maximum,
we expect ρ0 = 0. Therefore, (8.3.42) has to be of the form
(ρ0 )2 = −4ρ2 (ρ − ρ0 ), (8.3.43)
where ρ0 is the maximum of ρ. Therefore α = ρ0 /2. We can let ρ0 y 2 = ρ − ρ0 in (8.3.43) to

obtain
√
dy = (1 + y 2 ) ρ0 idT. (8.3.44)
√
Letting τ = i ρ0 T , we have
dy
= dτ, (8.3.45)
1 + y2
y = tan(τ − τ0 ), (8.3.46)
or
ρ = ρ0 (y 2 + 1) = ρ0 sec2 (τ − τ0 ), (8.3.47)
or
√
ρ = ρ0 sech2 ρ0 (T − T0 ) . (8.3.48)
Consequently,
√ √ ρ0
ψ(T, z) = ρ0 sech ρ0 (T − T0 ) ei 2 (z−z0 ) . (8.3.49)
Solitons 343
Since T → (z − vg t), this pulse travels with the same speed irrespective of the amplitude of
the pulse.
Schödinger equation (8.3.34) is Galilean invariant. Hence, if a solution ψ(x, t) is found, a
new solution ψ(x + vt, t)eiθ(x,t) is also a solution. Using this fact, a new solution
1
ψ1 (T, z) = ψ(T + kz, z)e−ik(T + 2 kz) (8.3.50)
can be constructed that is also a solution of the nonlinear Schrödinger equation.

Equation (8.3.49) represents a stationary solution in the moving coordinate system, or it
is a soliton solution that moves at the group velocity. Equation (8.3.50) represents a soliton
solution that moves at arbitrary velocity since k is arbitrary. The solution has the property
that the larger its amplitude, the narrower is the pulse. However, unlike the KdV solution,
the NLS solution’s velocity does not increase with its amplitude.
8.4 Solution of the KdV Equation via Inverse Scattering

Transform
Via a change of variable, the KdV equation can be written as
∂q ∂q ∂3q
− 6q + = 0. (8.4.1)
∂t ∂x ∂x3
It turns out that if q(x, t) is a potential in a Schrödinger eigenvalue equation, i.e.,
∂2u
+ [λ − q(x, t)]u = 0, (8.4.2)
∂x2
then u(x, t) evolves in time according to the following equation,
∂u ∂q ∂u
= − u + (4λ + 2q)
∂t ∂x ∂x
(8.4.3)
∂3u ∂q ∂u
= −4 3 + 3 u + 6q .
∂x ∂x ∂x
Equation (8.4.2) can be expressed as
Lu = λu (8.4.4)
2
∂
where L = − ∂x 2 + q(x, t), while Equation (8.4.3) is expressible as
∂u
= Mu (8.4.5)
∂t
where
∂q ∂ ∂3 ∂q ∂
M =− + (4λ + 2q) = −4 3 + 3 + 6q . (8.4.6)
∂x ∂x ∂x ∂x ∂x
The critical point here is that λ is assumed independent of time. So if we differentiate (8.4.4)
with respect to time, we have
Lt u + LM u = λM u = M λu = M Lu, (8.4.7)
or that
(Lt + LM − M L)u = 0. (8.4.8)
It can be shown that in order for (8.4.8) to be satisfied, q satisfies (8.4.1), the KdV equation.
The above gives an alternative way of solving (8.4.1). Given the initial value q(x, 0), one
can solve for u(x, 0) from (8.4.2). Then if one can determine u(x, t) from u(x, 0), then inverse
scattering theory can be used to find q(x, t) from u(x, t) in (8.4.2).
8.4.1 Inverse Scattering

Consider the Schrödinger equation
∂2
− u(x) + q(x)u(x) = λu(x), (8.4.9)
∂x2
and that q(x) → 0 when |x| → ∞. The above equation has two independent solutions u1 (x, k)
and u2 (x, k) such that u1 ∼ eikx , x → +∞, and that u2 ∼ e−ikx , x → −∞, where k 2 = λ.
Define the function A(x, y) as
Z∞
ikx
u1 (x, k) = e + A(x, t)eikt dt. (8.4.10)
x
We can substitute (8.4.10) into (8.4.9) to obtain an equation for A(x, t). First we notice that
Z∞ Z∞
∂2 ikt ∂2
A(x, t)e dt = [h(t − x)A(x, t)]eikt dt, (8.4.11)
∂x2 ∂x2
x 0
where h(t) is a Heavyside step function. Evaluating the above, we have
Z∞
∂2
A(x, t)e ikt
dt = −At (x, t) e ikx
− ikA(x, x)e ikx
− 2Ax (x, t) eikx
∂x2 t=x t=x
x
(8.4.12)
Z∞
ikt
+ Axx (x, t)e dt.
x
Solitons 345
Furthermore, since λ = k 2 ,
Z∞ Z∞
2 ikt ∂ 2 ikt
− k A(x, t)e dt = + A(x, t) e dt
∂t2
x x
∞ Z∞
∂ ikt ∂
=A(x, t) e − At (x, t) eikt dt
∂t x ∂t
x
∞ ∞ Z∞

ikt

ikt
=A(x, t)ike − At (x, t)e + Att (x, t)eikt dt
x x
(8.4.13)
x
= lim [ikA(x, t)eikt − At (x, t)eikt ]
t→∞

− ikA(x, x)e + At (x, t) eikx
ikx
t=x
Z∞
+ Att (x, t)eikt dt.
x
Consequently,
Z∞
∂2
u1 − q(x)u1 + k 2 u1 = [Axx (x, t) − Att (x, t) − q(x)A(x, t)]eikt dt
∂x2
x
+ lim [At (x, t) − ikA(x, t)]eikt (8.4.14)
t→∞

d
− 2 A(x, x) + q(x) eikx ,
dx
where
d ∂A(x, t) ∂A(x, t)
A(x, x) = + . (8.4.15)
dx ∂x t=x ∂t t=x
d
If lim At (x, t) = lim A(x, t) = 0, and that q(x) = −2 dx A(x, x), then,
t→∞ t→∞
Axx − Att − q(x)A = 0. (8.4.16)
If a wave is incident on the potential q(x) from x = ∞ as e−ikx , the solution for x → ∞
must be e−ikx + R(k)eikx , x → ∞, where R(k) is the reflection coefficient. For x → −∞, the
solution is T (k)e−ikx where T (k) is the transmission coefficient. This solution, by matching
its large |x| behaviors, must be expressible as R(k)u1 (x, k) + u1 (x, −k) or T (k)u2 (x, k) since
u1 (x, k) ∼ eikx , x → ∞, and u2 (x, k) ∼ e−ikx , x → −∞. Therefore, we conclude that
T (k)u2 (x, k) = R(k)u1 (x, k) + u1 (x, −k). (8.4.17)
Furthermore, when k → ∞ or that the energy of the wave tends to infinity, the potential
barrier has a negligible effect on the wave, and T (k) → 1 . Moreover, T (k) and u2 (x, k)
are the Fourier transforms of causal signals and they have to be analytic for =m[k] > 0.
Consequently, we can Fourier inverse transform (8.4.17) along an inversion contour C which
is above the singularities of T (k) and u2 (x, k), or
Z Z
T (k)u2 (x, k)e−ikt dk = R(k)u1 (x, k)e−ikt dk
C
ZC (8.4.18)
+ u1 (x, −k)e−ikt dk.
C
1

Since T (k)u2 (x, k)e−ikt ∼ e−ik(t+x) + O k e−ik(t+x) , k → ∞, a fact easily proven from Born
approximation, we have
Z Z
1
T (k)u2 (x, k)e−ikt dk = 2πδ(t + x) + O e−ik(t+x) . (8.4.19)
k
C C
Using Jordan’s lemma for the second term in (8.4.19), we conclude that
Z
T (k)u2 (x, k)e−ikt dk = 0, t + x < 0. (8.4.20)
C
Using the form in Equation (8.4.10) for u1 (x, k) in (8.4.18), we have
Z Z∞ Z
0
0= R(k)eik(x−t) dk + dt0 A(x, t0 ) R(k)eik(t −t) dt
C x C
(8.4.21)
Z Z∞ Z
−ik(x+t) 0 0 −ik(t0 +t)
+ e dk + dt A(x, t ) e dt, t + x < 0.
C x C
The third term is zero because t + x < 0, and defining

Z
1
r(t) = R(k)e−ikt dk, (8.4.22)
2π
C
we have
Z∞
0 = r(t − x) + dt0 A(x, t0 )r(t − t0 ) + A(x, −t), t + x < 0. (8.4.23)
x
The above is known as the Gelfand-Levitan-Marchenko equation. Given r(t), one can solve
for A(x, t) and obtain q(x) via
d
q(x) = −2 A(x, x). (8.4.24)
dx
Solitons 347
As an example, we can consider a case where r(t) = me+µt . Using it in (8.4.23), we obtain
that
Z∞
0
+µt−µx +µt
0 = me + me dt0 A(x, t0 )e−µt + A(x, −t), t + x < 0. (8.4.25)
x
Letting A(x, t) = a(x)e−µt , we have

Z∞
0
0 = me−µx + ma(x) dt0 e−2µt + a(x), t+x<0 (8.4.26)
x
or that
−me−µx
a(x) = m −2µx . (8.4.27)
1 + 2µ e
It follows then that
−me−2µx −m
A(x, x) = m −2µx = m . (8.4.28)
1 + 2µ e 2µ + e2µx
Differentiating the above with respect to x, we have
−4mµe2µx 2 e2φ+2µx
q(x) = 2 = −8µ 2φ
m 2µx (e + e2µx )2
2µ + e (8.4.29)
2 2
= −2µ sech (µx − φ).
1 m
where φ = 2 ln 2µ .
8.4.2 Solution of the KdV Equation

If the solution to the KdV equation, q(x, t), is a potential to the Schröd-inger equation
−uxx +q(x, t)u = k 2 u, then the eigenfunction evolves according to ut = −4uxxx +3qx u+6qux
according to (8.4.3). Therefore, if u is known when |x| → ∞, we can use the inverse scattering
theory to reconstruct q(x, t), and hence, the solution to the KdV equation.
As is shown in (8.4.17) the fundamental solutions to the Schrödinger equation satisfy
u2 (x, k) = c11 (k)u1 (x, k) + c12 u1 (x, −k). (8.4.30)
where c11 (k) = R(k)/T (k), c12 (k) = 1/T (k). Since u2 (x, k) ∼ e−ikx , x → −∞, irrespective
of q(x, t), and so independent of t, we assume that u = h(t)e−ikx , x → −∞. If q and qx tend
to zero when |x| → ∞, then
ut = −4uxxx . (8.4.31)
Hence for x → −∞, we have
ht = −4ik 3 h, (8.4.32)
or that 3
h = h0 e−4ik t . (8.4.33)
For x → +∞, we have

u(x) ∼ h(t)(c11 eikx + c12 e−ikx ).
Therefore, from (8.4.31), we have
(ht c11 + hc11t )eikx + (ht c12 + hc12t )e−ikx

(8.4.34)
= −4h[c11 (ik)3 eikx + c12 (ik)3 e−ikx ],
Comparing left and right-hand sides, and making use of (8.4.32), we have
c11t = 8ik 3 c11 , c12t = 0 (8.4.35)
or 3
c11 = co11 e8ik t , c12 = c012 . (8.4.36)
Consequently, we can find the time evolution of c11 and c12 . We can obtain R(k) = c11 (k)/c12 (k).
Since c11 and c12 are evolving with time, R(k) also evolves with time. At this point, the no-
tation is rather confusing since there is a time variable also in (8.4.22) and (8.4.23). We shall
call the time variable in the aforementioned equation τ to avoid the confusion. Hence,
Z
1
r(τ, t) = R(k, t)e−ikτ dk. (8.4.37)
2π
C
From Equation (8.4.36), we notice that

3
R(k, t) = R0 (k)e8ik t . (8.4.38)
For a simple case, we consider the case where R0 (k) has one simple pole at k = iµ. Then,
the above integral can be evaluated so that
3
r(τ, t) = me+8µ t+µτ
. (8.4.39)
According to (8.4.29), q(x, t) is
q(x, t) = −2µ2 sech2 (µx − φ) (8.4.40)
where !
3
1 me8µ t
φ = ln = 4µ3 t + φ0 , (8.4.41)
2 2µ
or
q(x, t) = −2µ2 sech2 (µx − 4µ3 t + φ0 ) (8.4.42)
2 1
Letting µ = 4 c, the above becomes

c 2 1√
q(x, t) = − sech c(x − ct) + φ0 . (8.4.43)
2 2
The above method can be used to find the multiple soliton solution to the KdV equation by
assuming more poles in the reflection coefficient. It addition, the inverse scattering transform
method can be used to solve nonlinear equations like the nonlinear Schrödinger equation.
Solitons 349
8.4.3 Inverse Scattering with Schrödinger Equation

Consider the Schrödinger equation
d2
− u(x) + q(x)u(x) = λu(x), (8.4.44)
dx2
where q(x) → 0 when |x| → ∞. The above equation has two independent solutions. We can
define them as u1 (x, k) and u2 (x, k) such that u1 (x, k) ∼ eikx , x → +∞, and that u2 (x, k) ∼
e−ikx , x → −∞, where k 2 = λ. Clearly, u1 and u2 are independent of each other when q = 0.
When q(x) 6= 0, the scattering potential will generate more waves but u1 and u2 remain
independent of each other. Therefore, any solution to (8.4.44) can be written as a linear
superposition of u1 and u2 .
The general solution to (1) can be written in terms of an integral equation
Z ∞
u(x) = u0 (x) + g(x − x0 )q(x0 )u(x0 )dx0 , (8.4.45)
−∞
where g(x), the Green’s function, is a solution to

!
d2 2
+ k g(x) = δ(x), (8.4.46)
dx2
and u0 (x) is a solution to

!
d2 2
+ k u0 (x) = 0. (8.4.47)
dx2
If physical condition is not imposed, there could be many solutions to (3). The possible
solutions to (8.4.46) are
eik|x|
g1 (x) = , (8.4.48)
2ik
e−ik|x|
g2 (x) = , (8.4.49)
2ik


0, x < 0,
g3 x = (8.4.50)

 sin kx
k , x < 0.

 sin kx
 k , x > 0,
g4 (x) = (8.4.51)


0, x < 0.
If e−ikt time convention is used, only g1 (x) is physical because it corresponds to outgoing
waves, while gi (x), i > 1 are unphysical. However, they can be used in (8.4.45) to provide
bonnefide solution to (8.4.44).
To construct u1 (x, k), we let u0 in (8.4.45) to be eikx , and g(x) in (8.4.45) to be g4 (x).
Then u1 (x, k) satifies
Z ∞
ikx
u1 (x) = e + g4 (x − x0 )q(x0 )u1 (x0 )dx0 . (8.4.52)
−∞
Due to the property of g4 (x), clearly, u1 (x) ∼ eikx , x → +∞.

To construct u2 (x, k), we let u0 in (8.4.45) to be e−ikx and g(x) to be g3 (x). Then, u2 (x, k)
satisfies
Z ∞
u2 (x) = e−ikx + g3 (x − x0 )q(x0 )u2 (x0 )dx0 . (8.4.53)
−∞
Clearly, u2 (x) ∼ e−ikx , x → −∞. Due to the independence of g3 (x) and g4 (x), u1 (x, k) and
u2 (x, k) are independent of each other. Since u1 (x, −k) is independent of u1 (x, k), they are
related by
T (k)u2 (x, k) = R(k)u1 (x, k) + u1 (x, −k) = u3 (x, k). (8.4.54)
Equation (8.4.54) corresponds to a scattering solution where it becomes Reikx + e−ikx , x →

+∞, and T e−ikx , x → −∞. Therefore, R and T physically correspond to reflection and
transmission coefficients respectively. Equation (8) can be used to derive the Gelfand-Levitan-
Marchenko equation.
8.4.4 Time-Domain Solutions

If we express
Z ∞
1
u(x, τ ) = dke−ikτ u(x, k), (8.4.55)
2π −∞
where k 2 = λ, and Fourier inverse transform (1) accordingly, we have
∂2 ∂2
− 2
u(x, τ ) + q(x)u(x, τ ) = − 2 u(x, τ ). (8.4.56)
∂x ∂τ
We shall call (8.4.55) the Schrödinger-like equation since it has a second derivative in time
while Schrödinger equation has a first derivative in time. Then (8.4.45) becomes
Z ∞
u(x, τ ) = u0 (x, τ ) + g(x − x0 , τ ) ? u(x0 , τ )q(x0 )dx0 , (8.4.57)
−∞
were g(x, τ ) is the Fourier inverse transform of the Green’s function g(x, k), given in (8.4.48)–
(8.4.51). We define the Fourier inversion contour to be above the singularity at the origin in
(8.4.48) and (8.4.49) so that only (8.4.48) is causal while the rest of the Green’s functions
Solitons 351
are not causal. The supports of the various Green’s functions are shown in Figure 1 on a
space-time diagram.
Consequently, the time-domain equation of (6) can be written as
Z ∞
u1 (x, τ ) = δ(x − τ ) + g4 (x − x0 , τ ) ? u1 (x0 , τ )q(x0 )dx0 . (8.4.58)
−∞
To determine the support of u1 (x, τ ), we can express (8.4.54) as a Neumann series which is
also a multiple scattering series - the first term corresponds to single scattering, the second
term corresponds to double scattering and so on. Then it is seen that the support of u1 (x, τ )
is given in Figure 2, when we assume that q(x) 6= 0, |x| < a. Similarly, the support of u2 (x, τ )
is given in Figure 3. Equation (8.4.54) corresponds to a physical scattering case where a wave,
e−ikx , is incident from x → +∞. The support of u3 (x, τ ) should be as shown in Figure 4,
whith a wave δ(x + τ ) incident fom x → +∞. The time domain equivalence of (8.4.54) is
R(τ ) ? u1 (x, τ ) + u1 (x, −τ ) = u3 (x, τ ). (8.4.59)
Notice that u1 (x, −k) is the Fourier inverse transform to u1 (x, −τ ). ¿From Figure 4, one can
see that R(τ ) 6= 0, τ > 2a. Consequently, the support of R(τ ) ? u1 (x, τ ) is as shown in Figure
5.
The support of u1 (x, −τ ) is shown in Figure 6. Since R(τ )?u1 (x, τ )+u1 (x, −τ ) = u3 (x, τ )
whose support is shown in Figure 4, we require that
R(τ ) ? u1 (x, τ ) + u1 (x, −τ ) = 0, τ < −x. (8.4.60)
The above is the key to the derivation of the Gelfand-Levitan-Marchenko equation. We can
rewrite Equation (8.4.58) as
u1 (x, τ ) = δ(x − τ ) + A(x, τ ), (8.4.61)
where A(x, τ ) = 0, τ < x. Then (8.4.60) becomes
R(τ − x) + R(τ ) ? A(x, τ ) + A(x, −τ ) = 0, τ < −x (8.4.62)
or
Z ∞
R(τ − x) + dτ 0 R(τ − τ 0 )A(x, τ ) + A(x, −τ ) = 0. τ < −x (8.4.63)
x
The above is the Gelfand-Levitan-Marchenko equation. Given the reflection coefficient R(τ ),
one can solve for A(x, τ ). It can be shown that the scattering potential q(x) can be derived
from A(x, τ ). We can express Equation (8.4.61) as
u1 (x, τ ) = δ(x − τ ) + A(x, τ )H(τ − x), (8.4.64)
where H(τ ) is a Heaviside step function. Substituting (8.4.64) into (8.4.56), we have
∂2 ∂2
− 2
A(x, τ )H(τ − x) + 2 A(x, τ )H(τ − x) +q(x)δ(x − τ ) (8.4.65)
∂x ∂τ
+ q(x)A(x, τ )H(τ − x) = 0. (8.4.66)
Since
∂2
A(x, τ )H(τ − x) = Hxx (τ − x)A(x, τ ) (8.4.67)
∂x2
+ 2Hx (τ − x)Ax (x, τ ) + H(τ − x)Axx (x, τ ), (8.4.68)
∂2
A(x, τ )H(τ − x) = Hτ τ (τ − x)A(x, τ ) (8.4.69)
∂τ 2
+ 2Hτ (τ − x)Aτ (x, τ ) + H(τ − x)Aτ τ (x, τ ). (8.4.70)
Since Hxx (τ − x) = Hτ τ (τ − x) = δ 0 (τ − x), Hx (τ − x) = −Hτ (τ − x) = −δ(x − τ ), using

(8.4.68) in (8.4.66) and matching terms of the same singularity, we have

2Ax (x, τ ) + 2Aτ (x, τ ) + q(x) = 0. (8.4.71)
τ =x τ =x
Consequently,
d
q(x) = −2 A(x, x), (8.4.72)
dx
where

d

A(x, x) = Ax (x, τ ) + Aτ (x, τ ) . (8.4.73)
dx τ =x τ =x

Chew Microwave and Optical Fiber PDF

Uploaded by

Copyright:

Available Formats

Chew Microwave and Optical Fiber PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chew Microwave and Optical Fiber PDF

Uploaded by

Copyright:

Available Formats

Lectures on

Theory of Microwave and Optical Waveguides

2.2.2 Lossy Transmission Lines . . . . . . . . . . . . . . . . . . . . . . . . . 43

3 Inhomogeneously Filled Waveguides 91

3.9.4 Variable Attenuator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

4 Coupling of Waveguides and Cavities 135

5 Discontinuities in Waveguides 179

6 Optical Waveguides 229

7 Microwave Integrated Circuits 297

7.7 A Branch Line Directional Coupler . . . . . . . . . . . . . . . . . . . . . . . 327

This monograph is an outgrowth of many years of teaching at the University of Illinois at

Weng Cho CHEW

1.2 History of Electricity and Magnetism

1.3 Maxwell-Heaviside Equations

COAX Twinax Rectangular Finline

Figure 1.1: Examples of closed waveguides.

Optical Thin-Film Periodic Dipole Array

Figure 1.2: Examples of open waveguides.

∇ × E(r) = iωB(r), (1.3.6)

∇ × H(r) = −iωD(r) + J(r), (1.3.7)

D = E, B = µH. (1.3.14)

In free-space,  = 0 = 8.854 × 10−12 farad/m, µ = µ0 = 4π × 10−7 henry/m. The constant

1.4 Wave Equation

∇ × µ−1 · ∇ × E(r) − ω 2  · E(r) = iωJ(r). (1.4.5)

Similarly, we can show that

∇ × −1 · ∇ × H(r) − ω 2 µ · H(r) = ∇ × −1 · J(r). (1.4.6)

∇ × µ−1 ∇ × E(r) − ω 2 E(r) = iωJ(r), (1.4.7)

∇ × −1 ∇ × H(r) − ω 2 µH(r) = ∇ × −1 J(r). (1.4.8)

1.5 Boundary Conditions

Figure 1.3: Boundary conditions at an interface.

1.6 Reciprocity Theorem

hE2 , ∇ × µ−1 · ∇ × E1 i − ω 2 hE2 ,  · E1 i = iωhE2 , J1 i, (1.6.4)

hE1 , ∇ × µ−1 · ∇ × E2 i − ω 2 hE1 ,  · E2 i = iωhE1 , J2 i. (1.6.5)

we can use the identity

and Gauss’ theorem to get

(µ−1 · ∇ × E1 ) × E2 |rs = iµ−1 −1

Figure 1.4: Proof of Reciprocity.

hE1 , J2 i − hH1 , M2 i = hE2 , J1 i − hH2 , M1 i (1.6.14)

hE2 , DE1 i = hE1 , DE2 i (1.6.17)

The above is the analogue of

1.6.1 Lorentz Reciprocity Theorem

When the volume does not enclose the sources, we arrive at

1.7 Energy Conservation

1.7.1 Time Domain Poynting Theorem

∇ · [E(r, t) × H(r, t)] = H · ∇ × E − E · ∇ × H

S (r, t) = E (r, t) × H (r, t) (1.7.2)

1.7.2 Frequency Domain Poynting Theorem

Defining the complex Poynting vector

the above becomes

1.7.3 Complex Power

1.7.4 Lossless Conditions

A sufficient condition for arbitrary V is to require that H∗ · µ · H and E · ∗ · E∗ to be purely

(H∗ · µ · H)∗ = H · µ∗ · H∗ = H∗ · µ† · H = H∗ · µ · H. (1.7.15)

Therefore, µ† = µ. Similarly the condition on E · ∗ · E∗ to be purely real is † = .

1.8 Energy Density in Dispersive Medium

∇ · [E(t) × H∗ (t)] = H∗ (t) · ∇ × E(t) − E(t) · ∇ × H∗ (t)

Next, we let ω = ω 0 + iω 00 where ω 0 and ω 00 are real numbers. Then

∇ · [E(t) × H∗ (t)] = i(ω 0 + iω 00 )H∗ (t) · µ(ω 0 + iω 00 ) · H(t)

Collecting leading order and first order terms, we have

D = E, B = µH. (1.3.14)

In free-space, = 0 = 8.854 × 10−12 farad/m, µ = µ0 = 4π × 10−7 henry/m. The constant

∇ × µ−1 · ∇ × E(r) − ω 2 · E(r) = iωJ(r). (1.4.5)

∇ × −1 · ∇ × H(r) − ω 2 µ · H(r) = ∇ × −1 · J(r). (1.4.6)

∇ × µ−1 ∇ × E(r) − ω 2 E(r) = iωJ(r), (1.4.7)

∇ × −1 ∇ × H(r) − ω 2 µH(r) = ∇ × −1 J(r). (1.4.8)

hE2 , ∇ × µ−1 · ∇ × E1 i − ω 2 hE2 , · E1 i = iωhE2 , J1 i, (1.6.4)

hE1 , ∇ × µ−1 · ∇ × E2 i − ω 2 hE1 , · E2 i = iωhE1 , J2 i. (1.6.5)

(µ−1 · ∇ × E1 ) × E2 |rs = iµ−1 −1

A sufficient condition for arbitrary V is to require that H∗ · µ · H and E · ∗ · E∗ to be purely

Therefore, µ† = µ. Similarly the condition on E · ∗ · E∗ to be purely real is † = .

By using the fact that ∇ × ∇ × E = −∇2 E + ∇∇ · E, and that ∇ · E = 1 ρ = 1

∇ × µ −1 · ∇ × E2 (r) − ω 2 · E2 (r) = S(r), (1.11.11)

where k 2 = ω 2 µ. By the same token, we have