Passive Open Planar Microwave Circuit Analysis by Enhanced Spectral Domain Method
Passive Open Planar Microwave Circuit Analysis by Enhanced Spectral Domain Method
Passive Open Planar Microwave Circuit Analysis by Enhanced Spectral Domain Method
December 1997
y
A thesis submitted to the University of Bristol in accordance with the requirements for
the degree of Doctor of Philosophy in the Faculty of Engineering, Department of Electrical
and Electronic Engineering
Abstract
Passive planar microwave circuits and planar antennas are essential components of
electronic communication systems utilising the microwave region of the electromag-
netic spectrum. This thesis is concerned with the development of a method which
allows ecient mathematical modelling to be obtained, giving the interactive design
tool capability of the simpler techniques but with the accuracy of the rigorous meth-
ods. The Spectral Domain Method (SDM), which is shown in the literature to be an
ecient technique for the analysis of components and discontinuities of the passive
planar microwave circuits, has been chosen to meet these requirements. SDM requires
a set of basis functions to be dened for the unknown current distribution and the
choice of this set is crucial to the eciency of the technique. The solution of this
problem is presented in this research in the form of precalculated basis functions and
of the application of sub{gridding.
The SDM is a frequency domain technique and requires repeated calculation of the
impedance matrix at each frequency point of interest. Special techniques to reduce
the number of the impedance matrix elements, which need to be calculated, and
to speed up the two{dimensional continuous integration have been introduced. The
adaptive integration technique, which consists of an adaptive integration range and
an adaptive integration step, has been introduced to limit the two dimensional con-
tinuous integration over an innite surface to nite computer resources. The use of
the asymptotic functions and of the location vector calculation are discussed. The
symmetry in both the Green's function and the impedance matrix has also been
exploited.
The method described in this thesis has been widely veried against both alternative
techniques and measured data for realistically complex circuits. The reduction in
computation time achieved is such as to make viable the interactive design of the open
planar microwave circuits and planar antennas with a full{wave rigorous technique.
i
In memory of my grandfather
Huseyin BALIK
1910{18 October 1997
ii
Acknowledgements
I would like to express my gratitude to all the people who have made completion
of this thesis possible. In particular I would like to thank Professor Joe McGeehan,
Professor Aydin Dumanoglu, Sefa Akpinar for the opportunity of undertaking a Ph.D.
at Bristol and their support and encouragement over the years. Financial support, for
which I am extremely grateful, has been provided by Karadeniz Technical University
in Turkey. I am also most grateful to my supervisor, Dr. Chris Railton, who has been
invaluable in the role of advisor throughout the course of this work
I have very much enjoyed working as part of the Centre for Communications Research
(CCR). I would especially like to acknowledge Steave Meade, Ian Craddock, K. M.
Krishnaiah. I would like to thank Naci Kayaoglu and Ross Wilkinson for reading my
thesis, and providing some valuable suggestions. I am also very gratwful to all my
friends specially Halil Ibrahim Okumus and Yusuf Baltaci for their friendships and
encouragements. I am very much indebted to Sefa Akpinar, Rifat Yazici and Cahit
Altan for being my nancial guarantors.
Last but not least, thanks to my wife, Dilek, for her unfailing love and support. I
express my deepest gratitude. I also thank my father, Osman, my mother Azime, my
sister, Hatice, and my brother, Ahmet, for their endless support over the years.
iii
Author's Declaration
Unless otherwise acknowledged, the content of this thesis is the original and sole work
of the author. No portion of the work in this thesis has been submitted by the author
in support of an application for any other degree or qualication, at this or any other
university or institute of learning. The views expressed in this thesis are those of the
author, and not necessarily those of the University of Bristol.
Copyright
Attention is drawn to the fact that the copyright of this thesis rests with the author.
This copy of the thesis has been supplied on condition that anyone who consults it is
understood to recognise that its copyright rests with the author and that no quotation
from the thesis and no information derived from it may be published without the prior
written consent of the author. This thesis may be made available for consultation
within the University Library and may be photocopied or lent to other libraries for
the purposes of consultation.
iv
Contents
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Author's Declaration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
Copyright . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
Table of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
List of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv
List of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
List of Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xvi
v
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.2 Available Full{Wave Analysis Techniques . . . . . . . . . . . . . . . . 18
2.2.1 Finite{Dierence Time{Domain Method (FDTD) . . . . . . . 18
2.2.2 Transmission Line Matrix (TLM) . . . . . . . . . . . . . . . . 19
2.2.3 Integral Equation Method (IEM) . . . . . . . . . . . . . . . . 20
2.3 Introduction to Spectral Domain Method . . . . . . . . . . . . . . . . 23
2.3.1 Choice of SDM . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.3.2 Historical Developments of Spectral Domain Method . . . . . 24
2.3.3 General Formulation . . . . . . . . . . . . . . . . . . . . . . . 26
2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
vi
3.5.6 Precalculated Discontinuity Function . . . . . . . . . . . . . . 49
3.6 Transfer Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.7 Two{Dimensional Numerical Examples . . . . . . . . . . . . . . . . . 51
3.7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.7.2 Dispersion Characteristics of Innite Microstrip Line . . . . . 52
3.8 Three{Dimensional Numerical Examples . . . . . . . . . . . . . . . . 52
3.8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.8.2 Precalculated Transverse Basis Function . . . . . . . . . . . . 53
3.8.3 Region Basis Function . . . . . . . . . . . . . . . . . . . . . . 57
3.8.4 Sub{gridding . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
vii
4.7 Inclusion of Loss in Dielectric Substrate . . . . . . . . . . . . . . . . . 84
4.8 Numerical Convergence . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.9 Loss in Dielectric Substrate . . . . . . . . . . . . . . . . . . . . . . . 87
4.10 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
viii
7.2 Derivation of Dyadic Green's Function . . . . . . . . . . . . . . . . . 128
7.3 Asymptotic Form of Green's function . . . . . . . . . . . . . . . . . . 133
7.4 Numerical Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
7.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
D Excitation 160
D.1 Fourier Transforms of Basis Functions for Feedlines . . . . . . . . . . 161
ix
List of Figures
1.1 The electromagnetic spectrum . . . . . . . . . . . . . . . . . . . . . . 2
1.2 The geometry of a microstrip line . . . . . . . . . . . . . . . . . . . . 7
1.3 Illustration of electromagnetic eld associated with a microstrip trans-
mission line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.4 Illustration of typical microstrip current distributions . . . . . . . . . 9
1.5 Illustration of microstrip singularities and discontinuities . . . . . . . 9
x
3.13 Plot of S{parameters's magnitude for the microstrip gap discontinuity
comparing basis function sets . . . . . . . . . . . . . . . . . . . . . . 57
3.14 Outlook of the microstrip line in nite length . . . . . . . . . . . . . 58
3.15 S{parameters' magnitude for the microstrip line . . . . . . . . . . . . 58
3.16 Microstrip edge{coupled lter . . . . . . . . . . . . . . . . . . . . . . 59
3.17 Current distributions for the rst four resonant modes of a strip resonator 61
3.18 Plot of S{parameters' magnitude for the edge{coupled lter, comparing
basis function sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.19 Microstrip step discontinuity . . . . . . . . . . . . . . . . . . . . . . . 63
3.20 Plot of S{parameters' magnitude for the step discontinuity . . . . . . 64
xi
5.6 General 3{D shielded microwave circuit . . . . . . . . . . . . . . . . . 104
5.7 Evaluation of ecient excitation modelling . . . . . . . . . . . . . . . 107
5.8 Finite microstrip line test structure . . . . . . . . . . . . . . . . . . . 109
5.9 Magnitude of S{parameters for the open microstrip line illustrating the
evaluation of the excitation modelling . . . . . . . . . . . . . . . . . . 110
5.10 Magnitude of S{parameters for the edge{coupled lter . . . . . . . . 111
xii
List of Tables
1.1 Frequency band designation . . . . . . . . . . . . . . . . . . . . . . . 3
xiii
List of Abbreviations
EFIE Electric Field Integral Equation
FDTD Finite Dierence Time Domain
FFT Fast Fourier Transform
HF High Frequency
IEM Integral Equation Method
MFIE Magnetic Field Integral Equation
MIC Microwave Integrated Circuit
MoM Method of Moments
MPIE Mixed Potential Integral Equation
MMIC Monolithic Microwave Integrated Circuit
SDM Spectral Domain Method
TE Transverse Electric
TEM Transverse Electromagnetic
TM Transverse Magnetic
TLM Transmission Line Matrix
UHF Ultra High Frequency
xiv
List of Symbols
E Electric led intensity
D Electric
ux intensity
H Magnetic eld intensity
B Electric
ux intensity
J Electric current density
Electric charge density
Permittivity
Permeability
Conductivity
kx; kz Fourier transform variables
G Dyadic Green's function
G1 Asymptotic form of the Green's function
Z Impedance matrix
V Excitation vector
Pi
Wavelength
kn Propagation constant or wave number
ko Free space wave number
w Weighting function
xv
List of Notations
Any function (possibly a vector) in the space domain
Any function (possibly a vector) in the spectral domain
A Any matrix in the space domain
A Any matrix in the spectral domain
AT Transpose of A
r Divergence of vector function
r Curl of vector function
xvi
List of Publications
Hasan H. Balik and C. J. Railton, \Sub{gridding in the spectral domain
method for the analysis of planar circuits and antennas," in Proceedings of
TELSIKS - 97, (Nis, Yugoslavia), p.592{595, University of Nis, October 1997.
Hasan H. Balik and C. J. Railton, \Adaptive numerical integration tech-
nique for the analysis of open planar circuits and antennas," in Proceedings
of 3rd High Frequency Postgraduate Student Colloquium," (Leeds, U.K),
p.100{105, University of Leeds, September 1997.
Hasan H. Balik and C. J. Railton, \Eective excitation mechanism for
open planar microwave circuits" in Proceedings of Spring Research Conference
- 97, (Bristol, UK), p.14, University of Bristol, April 1997.
xvii
Chapter 1
Introduction
1.1 Introduction Page 2
1.1 Introduction
Improvements in technology have always aected society as well as their standards of
living. The last few decades have witnessed advances in communication and computer
technology (referred to collectively as information technology). These developments
have transformed society from the industrial to the so-called information society [1,
chapter 1].
Radio
Microwave
Milimeter{wave
Infra{red
Visible light
Ultra{violet
X{rays
Gamma rays
10 3
106
10 9
10 12
10 15
10 18
1021
10 24
Frequency (Hz)
Figure 1.1: The electromagnetic spectrum
There are a number of classication schemes in use to designate frequency bands in
the electromagnetic spectrum. These classication schemes are summarised in table
1.1.
The lower frequency bands can provide only a limited amount of spectrum and are
already heavily used. The increase in demand for data transmission necessitated
communication systems to be moved upward to less{used frequency bands with more
available bandwidth. Although the frequencies above 3 MHz were considered useless
[2, chapter 6] in 1912 at the Radiotelegraph Conference in London, available cellular
mobile telephone services now operate in Europe at around 900 MHz and latest
generation of services (DCS1800 system) operates at 1.8 GHz.
The frequency region of interest in this thesis is the microwave region. The denition
of the microwave region varies. It is used for those wavelengths in the range 30 cm{0.3
mm which correspond to the frequency range 10 ; 10 Hz in [3, chapter 1], whereas
9 12
1.1 Introduction Page 4
in [4, chapter 1] microwaves are dened as those wavelengths in the range 1 cm{1
mm.
Communications, heating and remote sensing are common areas which utilise the
microwave frequency range. Microwave heating has become very popular in both
domestic and industrial environments. However, the communications industry has
been continuing a rapid growth in demand for the use of the microwave frequency
range. The spread of satellite communications and television broadcasting is an evid-
ent use of the microwave frequencies. Furthermore, mobile communications networks
are expected to continue to grow rapidly in coverage, with new systems using the
microwave frequencies.
The rapid increase of higher frequency usage over recent years has created a demand
for accurate design tools. A design tool which allows \right rst time" prediction is
invaluable to the realisation of cost{eective research and development. The accur-
acy of the design tool is often balanced against the computational eort required.
Existing packages allow rapid design, but their accuracy is not sucient for today's
applications. Alternatively there are highly{accurate analysis tools available, but
they require a large amount of computer resources. For example, software packages
such as Linmic and Sonnet are available, which utilise results from more rigorous
electromagnetic analysis [5,6] but are restricted by the inherent computational over-
heads.
The demands of the design engineer require a technique which is accurate, yet retains
the interactive design capabilities of the simpler techniques. High speed computers do
1.2 A Brief Review of Electromagnetism Page 5
not solve the problem, because the complexity of the circuits to be designed and the
computational overhead required for accurate modelling will outstretch any develop-
ment in computer technology. Therefore a design tool, which reverses the trend for
the requirement of ever{increasing computer resources, is required. Throughout this
thesis, fast accurate numerical microwave circuit analysis techniques will be sought.
Maxwell's equations, which are a set of fundamental equations governing all elec-
tromagnetic phenomena, can be written in both integral or dierential forms, but
here they are only presented in their dierential forms. For the time-varying elds,
Maxwell's equations in dierential forms can be written as
r E + @B
@t = 0 (Faraday's law) (1.1)
r H ; @D
@t = J (Maxwell{Ampere's law) (1.2)
r:D = (Gauss's law) (1.3)
r:B = 0 (Gauss's law{magnetic) (1.4)
where
E = electric eld intensity (volt/metre)
D = electric
ux intensity (coulomb/metre )2
r:J = ; @
@t (1.5)
Only three of (1.1){(1.5) are independent, and then called independent equations, the
other two equations can be derived from the chosen independent equations, and thus
are called auxiliary or dependent equations. The three selected independent equations
amongst Maxwell's four equations and the equation of continuity are less than the
number of unknowns. These equations are called indenite and must be transformed
to denite by seeking constitutive relations. These relations are,
D = E (1.6)
B = H (1.7)
J = E (1.8)
The electromagnetic eld at the boundary of two dierent media must satisfy a set
of equations called boundary conditions. In the electromagnetic problem involving
two or more media, Maxwell's equations together with the boundary conditions are
required to be simultaneously satised.
1.3 Basic Concepts of Microwave Circuits Page 7
1.3 Basic Concepts of Microwave Circuits
The physical implementations of modern microwave systems generally take the form
of MIC (Microwave Integrated Circuits) [7, 8]. This family can be subdivided into
hybrid MIC and MMIC (Monolithic Microwave Integrated Circuits). The latter are
completely integrated circuits where planar transmission lines, distributed elements
and active devices are formed on the same substrate. In contrast, the active devices
of the hybrid MIC are separate packages bonded to the integrated circuit. Although,
MMICs have many advantages such as the small package size and economics of scale
in manufacture, the hybrid MIC is inherently simpler to manufacture due to separate
fabrication of active devices.
Microstrip -z
The microwave integrated circuits of interest in this thesis are passive open planar
microwave circuits. The term passive implies that the circuits of interest contain no
active devices and the term planar implies that the circuit has a planar metallisation
surface with arbitrary shape. Moreover, the circuits of interest are completely open
and hence contain no side walls. Planar antennas may also be included within this
description. The basic element in the planar microwave circuit is a transmission line,
known as microstrip. This microstrip line consists of a grounded conducting plane,
1.3 Basic Concepts of Microwave Circuits Page 8
an insulating sheet, which is a dielectric substrate, and a conducting strip as shown
in gure 1.2.
Although microstrip is very simple and useful in the design of microwave miniature
and integrated circuits, its theoretical treatment is more complicated. Since the
dielectric material does not completely surround the conducting strip, consequently
the fundamental mode of propagation is not a pure TEM{mode [7, 9]. For a single
innite transmission line, a diagram of the E and H eld is shown in gure 1.3. The
fringing eld is evident in gure 1.3, and there is an inherent non-zero component
of E and H in the direction of propagation. Of course at low frequencies, the mode
can be approximated as a quasi{TEM, but in this research, this approximation is not
applicable to compact integrated circuits operating at microwave frequencies.
Metal strip
H
The corresponding surface current density on the microstrip has singularities at the
strip edges [10]. As the edge is approached, the current in the direction of propagation
is singular and the current in the direction perpendicular to the propagation tends to
zero, as illustrated in gure 1.4. The inclusion of the edge conditions is a fundamental
part of the research presented in this thesis.
Jz;x
Jz (x) Jx(x)
;w
2 x
w
2
Step
Jackson has calculated the current distribution for microstrip open{end and gap dis-
continuities in [11]. Re-entrant and protruding corners have been predicted the quasi{
1.4 Thesis Outline and Summary Page 10
static current distribution by Van Bladel et al [12] and Marchetti et al [13]. The actual
current distribution is more complex than that predicted by [12, 13]. For example,
frequency{dependent behaviour and coupling between discontinuities are not taken
into account. However, the inclusion of these current distributions in the numerical
analysis will greatly increase the convergence of the technique.
To reduce the computational overheads of circuit analysis is the core of this research.
Functions which include a priori knowledge of the current distribution of these corners
have been derived in [14, chapter 4] for shielded planar microwave circuits and, in
this research are adapted to open planar microwave circuits.
In chapter 1, the area of research for this thesis has been introduced and an outline
of the following chapters is given.
In chapter 2, a brief summary of alternative techniques is given and the basic philo-
sophy of each technique is highlighted together with its advantages and disadvant-
ages within the context of its application to passive planar microwave circuits. The
chosen Spectral Domain Method is introduced and discussed in the context of altern-
ative techniques. The historical development is reviewed and fundamental theory is
presented for a general formulation.
1.4 Thesis Outline and Summary Page 11
In chapter 3, the choice of current basis functions as introduced in chapter 2 is de-
scribed in detail. Variants of the present choice of basis functions are compared to
those commonly used by other researchers. A technique is presented for the derivation
of functions which allows the ecient inclusion of a priori knowledge of the current
distribution and sub{gridding of the metallisation due to discontinuity. Examples are
given to illustrate and verify the proposed enhancements.
Chapter 6 presents results for the analysis of realistically complex open microwave
circuits compared to measured data and numerical results of other techniques. Fea-
tures of the enhancements in this thesis are discussed in the context of the circuits
1.4 Thesis Outline and Summary Page 12
modelled.
In chapter 7, the derivation of the Green's function for multilayer structures is dis-
cussed and the immitance approach is given in detail. The asymptotic form of the
Green's function, which is proven to be eective to calculate the impedance matrix
in chapter 4 for two{layer circuits, is derived by the use of the immitance approach
for multilayer structures. To verify the accuracy of the immitance approach, a simple
microstrip line is modelled by the two dimensional version of the technique and com-
pared with published data. The eects of the order of dielectric substrates in the
microstrip structures are illustrated by modelling the multilayer microstrip line and
resonator.
Chapter 8 summarises the research implemented in the preceding chapters and sug-
gests potentially rewarding directions for future investigation.
Chapter 1: References
References
[1] I. J. Croddock, Enhanced Numerical Techniques for Time Domain Electromag-
netic Analysis. PhD thesis, University of Bristol, October 1995.
[2] M. Paetsch, Mobile Communications in the U.S. and Europe. Artech House,
1993.
[3] R. Chatterjee, Elements of Microwave Engineering. Ellis Horwood Ltd., 1986.
[4] R. E. Collin, Foundations for Microwave Engineering. McGraw-Hill, 1992.
[5] R. H. Jansen, \The spectral{domain approach for microwave integrated circuits,"
IEEE Transaction on Microwave Theory and Technique, vol. 33, pp. 1043{1056,
October 1985.
[6] J. C. Rautio and R. F. Harrington, \An electromagnetic time{harmonic analysis
of shielded microstrip circuits," IEEE Transaction on Microwave Theory and
Technique, vol. 35, pp. 726{729, August 1987.
[7] Y. Konishi, Microwave Integrated Circuits. Marcel Dekker Inc., 1991.
[8] D. B. Shorthouse, The CAD and Analysis of Passive Monolithic Microwave In-
tegrated Circuits by the Finite Dierence Time Domain Technique. PhD thesis,
University of Bristol, June 1992.
[9] P. Lorrain, D. R. Corson, and F. Lorrain, Electromagnetic Fields and Waves :
Including Electric Circuits. W. H. Freeman and Company, 1988.
[10] T. Itoh, Numerical Techniques for Microwave and Millimeter{Wave Passive
Structures. John Willey and Sons, 1989.
[11] R. W. Jackson and D. M. Pozar, \Full{wave analysis of microstrip open{end and
gap discontinuities," IEEE Transaction on Microwave Theory and Technique,
vol. 33, pp. 1036{1042, October 1985.
[12] R. D. Smedt and J. G. V. Bladel, \Field singularities near aperture corners," in
IEE Proceedings Pt., vol. 134, pp. 694{698, 1987.
Chapter 1: References Page 14
[13] S. Marchetti and T. Rozzi, \H{eld and J{current singularities at sharp edges
in printed circuits," IEEE Transaction on Antennas and Propagation, vol. 39,
pp. 1321{1331, September 1991.
[14] S. A. Meade, The Rapid and Rigorous Mathematical Modelling of Passive Planar
Microwave Circuits. PhD thesis, University of Bristol, March 1994.
Chapter 2
Most of the early numerical analysis methods for microwave integrated circuits were
based on either quasi{static or magnetic wall approximations, in which only the TEM{
mode of propagation was taken into account. At operating frequencies high enough to
full contemporary demands for spectrum space, it becomes necessary to use full{wave
analysis to include eects like coupling due to surface waves and discontinuities. This
chapter introduces modelling techniques which allow a rigorous full{wave analysis to
be achieved. In this context the descriptive term full{wave is used to dene the set
of numerical techniques which include all eld components and thus allow rigorous
analysis of dispersive structures, where coupling between circuit's elements cannot be
ignored in the microwave frequency band of interest.
Currently there are several available methods which full these requirements. The
Spectral Domain Method (SDM) has been chosen to analyse open planar microwave
circuits. It is a very popular method but has been criticised for being computationally
intensive, due to the need for the formation and inversion of a large matrix. Another
criticism is that SDM is severely limited in the complexity of the metallisation pat-
terns which it can handle. The results of this research overcome the limitations men-
tioned above and allow an interactive design tool to be developed for the computer
aided design of passive open planar microwave circuits as well as planar antennas.
To present a cross{section of the more commonly used methods with which SDM is
compared, a brief summary is given concerning alternative techniques in section 2.2.
2.1 Introduction Page 17
The basic philosophy of each technique is highlighted together with its advantages
and disadvantages. All criticism of alternative approaches is given within the context
of the application to passive planar microwave circuits. The techniques considered
are,
The above techniques do not include all available methods. FDTD and TLM have
minimal analytical preprocessing and are therefore applicable to a wide variety of
applications, but require a large amount of computer resources for complex circuits.
IEM is quite similar to the SDM, therefore it is included and given in relatively
more detail than the others to make SDM easy to understand. SDM and IEM are
both based on solving integral equations, but in dierent domains. The majority of
the calculations of IEM take place in the space domain, in contrast to the spectral
domain where all SDM calculations take place. Both require relatively large overheads
in analytical preprocessing, but the results may be yielded very quickly for certain
classes of problems with ecient implementations. It is proposed in this thesis that
open planar microwave circuits t into the class of problems suitable for analytical
preprocessing without restricting the application to unrealistically simple structures.
The basic philosophy of SDM is summarised in section 2.3 along with the historical
background of the present implementations. The choice of SDM for the fast rigorous
modelling of open planar microwave circuits is justied in comparison to the alternat-
ive techniques mentioned above. The general formulation of SDM is included for open
planar circuits. This chapter is intended to provide sucient background knowledge
to follow the enhancements in the following chapters.
2.2 Available Full{Wave Analysis Techniques Page 18
2.2 Available Full{Wave Analysis Techniques
2.2.1 Finite{Dierence Time{Domain Method (FDTD)
The Finite{Dierence Time{Domain Method is a full{wave numerical technique in
which Maxwell's time dependent electromagnetic curl equations are replaced by dif-
ference equations. The original FDTD algorithm proposed by Yee [1] in 1966 is
specically designed to solve electromagnetic problems in a rectangular co{ordinate
system and it assumes a lossless, isotropic and nondispersive medium. It uses the lin-
ear central dierence approximations for the time and space derivatives of the electric
and magnetic elds directly to replace the dierential operators of the curl equations.
The Finite Dierence Time Domain Method has been applied to a wide variety of elec-
tromagnetic problems including electromagnetic scattering [2], the analysis of various
planar transmission lines and discontinuities [3, 4], dielectric waveguides [5], planar
circuits [6, 7], microstrip lters and couplers [8], active [9] and passive [10] antennas
and medical electronics [11].
The calculation of an electric eld vector at a particular point in time and space
depends upon the value of that eld at the previous time, plus a function of the
surrounding perpendicular magnetic elds. Similarly, the magnetic eld depends
upon the previous value in time plus a function of the surrounding perpendicular
electric elds. As the electric eld and magnetic eld vectors are calculated in turn,
the electric elds are calculated a half time step out of sequence with the magnetic
values. Numerical boundary conditions are required for limit innite open structures
to nite computer resources. The frequency response can be obtained by taking the
Fourier transform of the resulting time{domain transient response.
The underlying principle of the Transmission Line Matrix Method is a mesh of dis-
crete nodes connected by transmission lines, with the solution derived in the time
domain. In common with FDTD, the main advantage of TLM is that it is possible to
include time{dependent parameters in the algorithm and the ease with which most
complicated structures can be evaluated. The main disadvantage of the method is
that it is limited by the amount of memory storage required, which depends on the
complexity of the structure and the nonuniformity of the eld [14]. The technique is
not optimised for planar structures thus it is computationally intensive compared to
the optimised Spectral Domain Method [18].
2.2 Available Full{Wave Analysis Techniques Page 20
2.2.3 Integral Equation Method (IEM)
The Integral Equation Method is one of the full{wave techniques used to analyse
microwave and millimetre{wave integrated circuits, and is reviewed by Mosig [14,
chapter 3]. It is based on solving the integral equation which is also used by the
Spectral Domain Method. The majority of the calculations take place in the space
domain, in contrast to the spectral domain. The spectral domain is used only during
the calculation of the Green's functions, then inverse transforms are taken and the
integral equation is solved in the space domain. Due to this similarity, this technique
is given here in relatively more detail to facilitate the comparison to SDM
Katethi and associated authors [19{22] used IEM to analyse microstrip discontinuities
for open and shielded structures. The Integral Equation Method has been enhanced
by several authors and hence called with some extra names to suit dierent applica-
tions. The Singular Integral Equation Method is used to determine the normal modes
of propagation in general transmission lines taking nlines as example in [23], whereas
in [24] the Boundary Integral Equation Method is proposed for the full{wave analysis
of suspended planar transmission line with nite metallisation thickness. Morisue et
al have launched the Lorentz Gauge vector potential formulation for the Boundary{
IEM in [25]. The application of the IEM to anisotropic inhomogeneous waveguides
has been presented by Urbach [26]. A Volume{Surface IEM is published in [27] to
enable the calculation of electromagnetic elds in complicated 3{D models.
The formulation of the integral equation can be written in dierent ways. The more
common one is the electric eld integral equation (EFIE) formulation where the funda-
mental unknown is the surface electric current
owing on the conductor. Alternative
formulations are the magnetic eld integral equation (MFIE) and mixed potential
integral equation formulations (MPIE) [14, chapter 3]. However the magnetic eld
integral equation is not suitable for microstrip integrated circuits. This is because, it
is numerically unstable when the embedded conductors are thin and it fails completely
2.2 Available Full{Wave Analysis Techniques Page 21
for zero{volume conductors [14, page 139].
To demonstrate the basic philosophy of the Integral Equation Method, the formula-
tion is summarised for the electric eld and surface current (Electric Field Integral
Equation, EFIE). An electric eld integral equation which relates the surface current
and electric eld is dened via a dyadic Green's function in the space domain.
ZZ
[Gzz (x ; x0 ; y; z ; z0 )Jz (x0 ; z0 ) + Gzx(x ; x0 ; y; z ; z0 )Jx(x0 ; z0 )]dx0 dz0 = Ez (x; z)
ZZ
[Gxz (x ; x0 ; y; z ; z0 )Jz (x0 ; z0 ) + Gxx(x ; x0 ; y; z ; z0 )Jx(x0 ; z0 )]dx0 dz0 = Ex(x; z)
(2.1)
No convenient form of the Green's function exits in the space domain for the planar
circuit congurations of interest [18]. The most ecient technique formulates the
Green's function in the spectral domain, and therefore is closely related to the SDM,
where x; z replaced by their spectral counterparts kx; kz according to the double Four-
ier transform. It is therefore the spectral domain which is used during the calculation
of the Green's functions, and then inverse transforms are taken. This procedure is
one of the main disadvantages of IEM as it is computationally intensive.
Equation 2.1 contains four unknowns, namely Jx; Jz ; Ex and Ez . At least two un-
knowns must be eliminated. In common practice, the Method of Moments (MoM) is
used to solve equation 2.1. The rst step of the analysis is to expand the unknown
current distribution on the metallisation of the circuit as a set of known basis func-
tions with unknown coecients. Basis functions which are dened as non{zero only
on the metallisation are given by;
N
X
Js(x; z) = anJsn(x; z); s = x; z (2.2)
n=1
The basis functions (Jsn in equation 2.2) must be chosen to approximate the unknown
current distribution on the metallisation of the circuit. Note that in equation 2.2, the
2.2 Available Full{Wave Analysis Techniques Page 22
Fourier transform of the basis functions is not required, as the algorithm operates in
the space domain.
After application of the Method of Moments using weighting functions, equation 2.1
becomes the matrix equation,
[Z ][I ] = [V ] (2.3)
where Z is the impedance matrix and I is the vector of the coecients (an in equa-
tion 2.2) and V is the excitation vector dened as zero everywhere except for sources.
If current basis and weighting functions are the same, then the procedure is called
Galerkin's.
By solving matrix equation 2.3 for the unknown current coecients, the unknown
current distribution of the metallisation of the circuit can be determined for a set of
excitation vectors and the S{parameters can be derived.
Equation 2.3 shows that the inverse of Z is required for the nal solution. The choice
of the current basis function is crucial, since if a large number of basis functions are
used, the size of the matrix to be inverted will be large. Thus special care must
be taken to minimise the size of the set of current functions for a given problem.
This problem is also shared by SDM. The major drawback of the method is that it
requires the inverse transform of the spectral domain Green's functions, which must
be performed numerically.
2.3 Introduction to Spectral Domain Method Page 23
2.3 Introduction to Spectral Domain Method
2.3.1 Choice of SDM
The purpose of the work described in this thesis is to develop an ecient full{wave
design tool for open planar microwave circuits. The Spectral Domain Method is
chosen as it is believed to be capable of meeting these requirements. Although FDTD
and TLM have minimal preprocessing, signicant analytical preprocessing is not their
basic philosophy, therefore redundancies often exist when applied to the planar struc-
tures of interest. These redundancies have been reduced to a minimum in SDM by
using signicant enhancements which will be explained in the following chapters.
Both SDM and IEM are based on solving integral equations, but in dierent domains.
No convenient form of the Green's functions for the planar circuits of interest exist
in the space domain, where the majority of the calculations of IEM take place. The
most ecient way to nd the Green's function is to use an inverse transform from
the spectral domain into the space domain. This is computationally intensive and
the major drawback of IEM. The principal advantage of using SDM is that the integ-
ral equation in the space domain is transformed into an algebraic formulation after
Fourier transformation of the convolution form.
The SDM is numerically ecient at the cost of signicant analytical preprocessing.
This feature imposes certain restrictions on the applicability of the technique. One of
the limitations is that SDM normally assumes an innitesimal thickness for the strip
conductor. It is also dicult to analyse circuits with a strip having nite conductivity.
No discontinuities are allowed in the substrate in the directions in which Fourier
transforms are taken. In spite of these limitations, SDM is one of the most popular
and widely used numerical techniques.
The eciency of SDM will be highlighted throughout the thesis. The existing SDM
2.3 Introduction to Spectral Domain Method Page 24
is enhanced to allow its use as interactive design tool. The computer run{time will
be reduced for circuits including those which are realistically complex. The following
sections are devoted to the description of the general formulation on which the present
implementation is based.
The analysis in the Fourier transform domain was rst introduced by Yamashita and
Mittra [28] in 1968 for the computation of the characteristic impedance and the phase
velocity of an open microstrip. It was based on a quasi{TEM approximation which
is a low frequency simplication that neglects the longitudinal electric and magnetic
elds supported by the microstrip.
Increasing of the operating frequency has led numerical techniques towards full{wave
analysis, because eects such as coupling due to surface waves and discontinuities be-
come signicant at high frequencies. Denlinger [29] solved the integral equation using
the Fourier transform technique. The solution by his method, however, is strongly de-
pendent on the assumed current distribution on the strip. To avoid this diculty and
allow systematic improvements of the solution for the current components to a desired
degree of accuracy, Itoh and Mittra [30] introduced the Spectral Domain Approach,
otherwise called the Spectral Domain Method in 1973 for open microstrip lines and
enhanced for shielded microstrip lines in [31]. The unknown current distribution on
the strip is dened as a set of known current basis functions with unknown coe-
cients and Galerkin's Procedure is used to yield a homogeneous system of equations
to determine the propagation constant and amplitude of current distribution.
2.3 Introduction to Spectral Domain Method Page 25
The application of the method to a general three{dimensional structure required the
introduction of a more advanced formulation. The move in this direction is presented
by Itoh [32] with the modelling of a microstrip resonator. In 1980, Itoh introduced
a formulation process called the immitance approach [33] for a easy formulation of
multilayer structures. This approach is based on the transverse equivalent circuit
concept as applied in the spectral domain. An application of the immitance approach
was presented by Itoh and Menzel [34] which allowed the analysis of planar resonator
antennas for the complex resonant frequency. The historical progress of the Spectral
Domain Method until 1984 was summarised by Jansen [35].
where
Gst is the dyadic Green's function (dependent on frequency, !),
Es is the electric eld at the air{dielectric interface,
Js is the current distribution on the metallisation.
Equation 2.4 can be solved if Gst , etc., are given . The Green's functions Gzz , etc.,
however, are not available in closed form for the nonhomogeneous structures. The
following algebraic equations (2.5), instead of the coupled integral equations (2.4), are
obtained in the spectral domain formulation. These equations are Fourier transforms
2.3 Introduction to Spectral Domain Method Page 27
of the coupled integral equations (2.4).
where the bold quantities are the Fourier transform of corresponding quantities and
d is the thickness of the dielectric substrate. The two dimensional Fourier transform
is dened as,
Z1Z1
(kx; kz ) = (x; z)ej kxx
( kz z) dxdz
+
(2.6)
;1 ;1
Derivations of the spectral domain forms of the Green's functions (Gzz , etc. in
equations 2.5 and 2.7) are widely available in the literature for single dielectric layer
[30, 31]. Derivation of the dyadic Green's function for multiple dielectric layers will
be discussed in detail in chapter 7.
2.3 Introduction to Spectral Domain Method Page 28
Equations 2.5 and 2.7 contain four unknowns which are Jz ; Jx; Ez and Ex. At least two
unknowns must be eliminated to solve the equations. The Method of Moments [18]
is applied to eliminate the Fourier transformed electric elds from the formulation.
The rst step to the solution is to expand the unknown surface current as a set of
known basis functions with unknown coecients.
N
X
Js(kx; kz ) = asnJsn(kx; kz ) s = x; z (2.8)
n=1
The basis functions (Jsn) must be chosen to approximate the unknown current distri-
bution on the metallisation of the circuit. The choice of basis functions is crucial to
the eciency of the technique. If they are not chosen to represent the actual current
distribution closely, then a large number of functions will be required for convergence.
The choice of current basis functions will be discussed in detail in chapter 3.
To eliminate the Fourier transformed electric elds (Ez ; Ex in equation 2.7), a set of
weighting functions (wt ) are required. The weighting functions (wt) are chosen to be
identical to the set of current basis functions. Two more weighting functions are also
required for the problem with excitation, the denition of these functions is described
in chapter 5.
By solving equation 2.9 for the unknown current coecients, the unknown current
distribution of the metallisation of the circuit can be determined. For the sourceless
eigenvalue problem, the excitation vector (Vt) is identically zero. This can be veried
using Parseval's theorem [14, page 341].
Note that the method is formulated for spot frequencies, that is the nal matrix
equation is repeatedly solved for each frequency of interest. Thus minimisation of the
order of the matrix equation is very important for the eciency of the technique. For
the Method of Moments, the order of the matrix is twice the total number of basis
functions plus two, therefore the critical choice of the basis functions is highlighted.
2.4 Summary
Three alternative numerical techniques have been discussed along with their advant-
ages and disadvantages. All criticism of above techniques were presented within the
context of the application of passive planar microwave circuits. Although these are
not inclusive, it is believed that they represent a cross{section of the more commonly
used methods. The choice of the Spectral Domain Method and highlighted reasons
why SDM is more suited to the application of passive planar circuits have also been
discussed in this chapter. The historical background of the SDM implementation has
been outlined and the general formulation has been presented.
Chapter 2: References
References
[1] K. S. Yee, \Numerical solution of initial boundary value problems involving
Maxwell's equations in isotropic media," IEEE Transactions on Antennas and
Propagation, vol. Ap{14, pp. 302{307, May 1966.
[2] A. Ta
ove and K. R. Umashankar, \The nite{dierence time{domain (FD{TD)
method for numerical modelling of electromagnetic scattering," IEEE Transac-
tions on Magnetics, vol. 25, pp. 3086{3091, July 1989.
[3] C. J. Railton and J. P. McGeehan, \An analysis of microstrip with rectangu-
lar and trapezoidal conductor cross sections," IEEE Transaction on Microwave
Theory and Technique, vol. 38, pp. 1017{1022, August 1990.
[4] X. Zhang and K. K. Mei, \Time{domain nite dierence approach to the calcu-
lation of the frequency{dependent characteristics of microstrip discontinuities,"
IEEE Transaction on Microwave Theory and Technique, vol. 36, pp. 1775{1787,
December 1988.
[5] C. J. Railton, E. M. Daniel, D. L. Paul, and C. J. McGeehan, \Optimized absorb-
ing boundary conditions for the analysis of planar circuits using nite dierence
time domain method," IEEE Transaction on Microwave Theory and Technique,
vol. 41, pp. 290{297, February 1993.
[6] P. Mezzanotte, M. Mongiardo, L. Roselli, R. Sorrentino, and W. Heinrich, \Ana-
lysis of packaged microwave integrated circuits by FDTD," IEEE Transaction on
Microwave Theory and Technique, vol. 42, pp. 1796{1801, September 1994.
[7] D. M. Sheen, S. M. Ali, M. D. Abdouzahra, and J. A. Kong, \Application of the
three{dimensional nite-dierence time{domain method of the analysis of planar
microstrip circuits," IEEE Transaction on Microwave Theory and Technique,
vol. 38, pp. 849{857, July 1990.
[8] D. L. Paul, E. M. Daniel, and C. J. Railton, \Fast nite dierence time do-
main method for the analysis of planar microstrip circuits," in 21st European
Microwave Conference, pp. 303{308, September 1991.
[9] B. Toland, J. Lin, B. Houshmand, and T. Itoh, \FDTD analysis of an act-
ive antenna," IEEE Microwave and Guided Wave Letters, vol. 3, pp. 423{425,
November 1993.
Chapter 2: References Page 31
[10] E. M. Daniel and C. J. Railton, \Fast nite dierence time domain analysis of
microstrip patch antennas," IEEE Antennas and propagation Symposium Digest,
pp. 414{417, 1991.
[11] N. M. Pothecary and C. J. Railton, \Finite dierence time domain modelling of
hyperthermia applicators for cancer therapy," IEEE MTT{S Digest, pp. 1151{
1154, 1993.
[12] D. B. Shorthouse and C. J. Railton, \The incorporation of static eld solu-
tion into the nite dierence time domain algorithm," IEEE Transaction on
Microwave Theory and Technique, vol. 40, pp. 986{994, May 1992.
[13] E. M. Daniel and C. J. Railton, \An improvement second order radiating bound-
ary condition for use with non{uniform grids in the nite dierence time domain
method," in 21st European Microwave Conference, pp. 547{552, September 1991.
[14] T. Itoh, Numerical Techniques for Microwave and Millimeter{Wave Passive
Structures. John Willey and Sons, 1989.
[15] J. L. Vetri and N. R. S. Simons, \A class of symmetrical condensed node TLM
methods derived directly from Maxwell's equations," IEEE Transaction on Mi-
crowave Theory and Technique, vol. 41, pp. 1419{1428, August 1993.
[16] M. Krumpholz and P. Russer, \On the dispersion in TLM and FDTD," IEEE
Transaction on Microwave Theory and Technique, vol. 42, pp. 1275{1279, July
1994.
[17] M. Celuch-Marcysiak and W. K. Gwarek, \Formal equivalence and eciency
comparison of the FD{TD, TLM and SN methods in application to microwace
CAD programs," in 21st European Microwave Conference, pp. 199{204, Septem-
ber 1991.
[18] S. A. Meade, The Rapid and Rigorous Mathematical Modelling of Passive Planar
Microwave Circuits. PhD thesis, University of Bristol, March 1994.
[19] P. B. Katehi and N. G. Alexopoulos, \Frequency{dependent characteristics of
microstrip discontinuities in millimeter{wave integrated circuits," IEEE Trans-
action on Microwave Theory and Technique, vol. 33, pp. 1029{1035, October
1985.
[20] L. P. Dunleavy and P. B. Katehi, \Shielding eects in microstrip discontinuities,"
IEEE Transaction on Microwave Theory and Technique, vol. 36, pp. 1767{1773,
December 1988.
[21] W. P. Harokopus and P. B. Katehi, \Characterisation of microstrip discontinu-
ities on multilayer dielectric substrates including radiation losses," IEEE Trans-
action on Microwave Theory and Technique, vol. 37, pp. 2058{2066, December
1989.
Chapter 2: References Page 32
[22] T. G. Livernois and P. B. Katehi, \A simple method for characterizing planar
transmission line discontinuities on dissipative substrate," IEEE Transaction on
Microwave Theory and Technique, vol. 39, pp. 368{370, February 1991.
[23] A. S. Omar and K. Schunemann, \Formulation of the singular integral equation
technique for planar transmission lines," IEEE Transaction on Microwave Theory
and Technique, vol. 33, pp. 1313{1321, December 1985.
[24] L. Zhu and E. Yamashita, \Full{wave boundary integral equation method for
suspended planar transmission lines with pedestals and nite metallisation thick-
ness," IEEE Transaction on Microwave Theory and Technique, vol. 41, pp. 478{
483, March 1993.
[25] T. Morisue and T. Yajima, \The Lorentz Gauge vector potential formulation
for the boundary integral equation method," IEEE Transaction on Magnetics,
vol. 30, pp. 3032{3035, September 1994.
[26] H. P. Urbach and E. S. A. M. Lepelaars, \On the domain integral equation
method for anisotropic inhomogeneous waveguides," IEEE Transaction on Mi-
crowave Theory and Technique, vol. 42, pp. 118{126, January 1994.
[27] J. Nadobny, P. Wust, M. Seebass, P. Deu
hard, and R. Felix, \A volume{surface
integral equation method for solving Maxwell's equations in electrically inhomo-
geneous media using tetrahedral grids," IEEE Transaction on Microwave Theory
and Technique, vol. 44, pp. 543{554, April 1996.
[28] E. Yamashita and R. Mittra, \Variational method for the analysis of microstrip
lines," IEEE Transaction on Microwave Theory and Technique, vol. 16, pp. 251{
255, April 1968.
[29] E. J. Denlinger, \A frequency dependent solution for microstrip transmission
lines," IEEE Transaction on Microwave Theory and Technique, vol. 19, pp. 30{
39, January 1971.
[30] T. Itoh and R. Mittra, \Spectral{domain approach for calculating the dispersion
characteristics of microstrip lines," IEEE Transaction on Microwave Theory and
Technique, pp. 496{499, July 1973.
[31] T. Itoh and R. Mittra, \A technique for computing dispersion characteristics of
shielded microstrip lines," IEEE Transaction on Microwave Theory and Tech-
nique, pp. 896{898, October 1974.
[32] T. Itoh, \Analysis of microstrip resonators," IEEE Transaction on Microwave
Theory and Technique, vol. 22, pp. 946{952, November 1974.
[33] T. Itoh, \Spectral domain immitance approach for dispersion characteristics of
generalized printed transmission lines," IEEE Transaction on Microwave Theory
and Technique, vol. 28, pp. 733{736, July 1980.
Chapter 2: References Page 33
[34] T. Itoh and W. Menzel, \A full{wave analysis method for open microstrip struc-
tures," IEEE Transactions on Antennas and Propagation, vol. 29, pp. 63{67,
January 1981.
[35] R. H. Jansen, \The spectral{domain approach for microwave integrated circuits,"
IEEE Transaction on Microwave Theory and Technique, vol. 33, pp. 1043{1056,
October 1985.
[36] R. W. Jackson and D. M. Pozar, \Full{wave analysis of microstrip open{end and
gap discontinuities," IEEE Transaction on Microwave Theory and Technique,
vol. 33, pp. 1036{1042, October 1985.
[37] J. C. Rautio and R. F. Harrington, \An electromagnetic time{harmonic analysis
of shielded microstrip circuits," IEEE Transaction on Microwave Theory and
Technique, vol. 35, pp. 726{729, August 1987.
[38] R. W. Jackson, \Full{wave, nite element analysis of irregular microstrip dis-
continuities," IEEE Transaction on Microwave Theory and Technique, vol. 37,
pp. 81{89, January 1989.
[39] N. K. Das and D. M. Pozar, \A generalised spectral{domain Green's function for
multilayer dielectric substrates with application to multilayer transmission lines,"
IEEE Transaction on Microwave Theory and Technique, vol. 35, pp. 326{335,
March 1987.
[40] C. J. Railton and T. Rozzi, \Complex modes in boxed microstrip," IEEE Trans-
action on Microwave Theory and Technique, vol. 36, pp. 865{873, May 1988.
[41] C. J. Railton and S. A. Meade, \Fast rigorous analysis of shielded planar lters,"
IEEE Transaction on Microwave Theory and Technique, vol. 40, pp. 978{985,
May 1992.
[42] S. A. Meade and C. J. Railton, \Ecient implementation of the spectral domain
method including precalculated corner basis functions," IEEE Transaction on
Microwave Theory and Technique, vol. 42, pp. 1678{1684, September 1994.
[43] M.-J. Tsai, D. Flaviis, O. Fordham, and N. G. Alexopoulos, \Modelling planar
arbitrary shaped microstrip elements in multilayered media," IEEE Transaction
on Microwave Theory and Technique, vol. 45, pp. 330{337, March 1997.
[44] J. Kuo and T. Itoh, \Hybrid{mode computation of propagation and attenuation
characteristic of parallel coupled microstrips with nite metalisation thickness,"
IEEE Transaction on Microwave Theory and Technique, vol. 45, pp. 274{280,
February 1997.
Chapter 3
The basic SDM formulation can be applied to simple structures in an ecient way,
but problems exist in the application of the method to complex circuits. An entire
domain basis function which is valid on the entire metallisation of the circuit can be
used to dene the unknown current distribution of simple structures such as an innite
microstrip line [1, 2], a microstrip resonator [3] or a hairpin resonator [4]. However
the implementation of an arbitrary metallisation pattern requires sub{domain basis
functions which are valid on a portion of the metallisation. It has been shown and
commonly used that the rooftop function [5,6] as a sub{domain current basis function
allows an irregular shaped microwave circuit's current to be dened. But, the use of
the rooftop functions to dene the unknown current distribution on the metallisation
of the circuit results in a large number of basis functions for convergence. As the size
of the impedance matrix to be calculated for each spot frequency is proportional to
double of the number of basis functions, it is therefore very important to minimise
the set of basis functions required.
It was proposed by Railton and Meade [4, 5] that precalculated basis functions be
used to reduce the number of basis functions required. They applied the proposed
technique to complex shielded planar microwave circuits showing that inclusion of
a priori knowledge of the current distribution for realistically complex structures
reduces the size of the impedance matrix to be calculated. The precalculated basis
3.2 Classes of Current Basis Functions Page 36
function is derived from the sub{domain basis functions, which are identical apart
from a shift in origin, to exploit the benet of using the FFT. In addition the number
of rooftop functions plus 1 in both directions must be an integer power of 2. These
conditions restrict the geometry of the circuit and the size of the rooftop function.
The present implementation applies all available enhanced current basis functions
in the literature. The eectiveness of rooftop functions is improved by introducing
sub{gridding in the spectral domain. The sub{gridding is used in the sense that sizes
of the rooftop functions are dened as functions of their locations. The precalculated
function is derived in this thesis from two dierent precalculations. It can be either
a linear combination of rooftop functions with precalculated coecients, or a current
wave with a precalculated wave number as explained in section 3.5.4. In both cases,
there is no restriction of geometry. This chapter also describes the combination of
sub{gridding, precalculation and re{mapping of the precalculated functions.
Throughout this thesis, the above classes of current basis functions are used to dene
the unknown current distribution of an arbitrary shaped metallisation pattern. Fur-
ther details of each class of basis function are given in following sections.
Arbitrary shaped planar microwave circuits have been modelled using rooftop func-
tions as sub{domain basis functions by numerous authors [4{6]. The geometry of a
rooftop function is illustrated in gure 3.1 and is dened in space domain by :
8
>
>
> xn ; lx x xn + lx
>
< 1 ; jx;lxxn j
>
Jxn(x; z) = >
>
zn ; lz z zn + lz (3.1)
>
>
>
: 0 otherwise
8
>
>
> xn ; lx x xn + lx
>
< 1 ; jz;lzzn j
>
Jzn(x; z) = > zn ; lz z zn + lz (3.2)
>
>
>
>
: 0 otherwise
where xn and zn are the coordinates of the centre of the nth rooftop and lx; lz are the
sizes of the grid where the rooftop function is located. The rooftop function is dened
by two separable functions: a triangle function in the direction of current
ow and
a step function in the direction perpendicular to current
ow. Figure 3.1 shows a
rooftop function to model current
ow in the z{direction as two separable functions,
the step function which is 2lx wide and the triangle function which is 2lz wide. Thus
the function overlaps in both directions.
3.4 Sub{domain Basis function Page 39
bn
2l x
2lz
Figure 3.1: Z {directed rooftop function
lx
Overlap of two rooftops
12
x Overlap of four rooftops
34
z lz
Figure 3.2: Illustration of sub{domain basis function gridding
3.4.2 Sub{gridding
The unknown current distribution on an arbitrary shaped metallisation can be dened
as a combination of the rooftop functions, but this approach results in a large number
of rooftop functions for convergence. The modelling of a corner singularity shown in
gure 3.3 requires even more rooftop functions in order to get accurate results for the
entire circuit if all rooftop functions are dened identical, apart from a shift in origin.
This is because, ne rooftops are required to model the singularity and the same
sized rooftops must be used to model the entire circuit. In this case, unnecessarily
ne rooftops would be used in the region where no discontinuity exists. This is a
major drawback of the technique and a reduction in the number of basis functions
required would be an improvement.
Fine grid
Coarse grid
Re-entrant corner
In contrast to the formulation for boxed structures [4], the FFT is not employed
for the calculation of the impedance matrix elements in this implementation. The
restrictions for the FFT usage no longer apply and alternative calculation methods
are introduced in chapter 4. The author has dened the size of the rooftop function
in [7] as a function of its location, allowing sub{gridding. Fine rooftops are used where
3.4 Sub{domain Basis function Page 41
rapid change in current distribution occurs and a coarse size where only slow changes
in current distribution occur, as illustrated for the re{entrant corner discontinuity
in gure 3.3. The ne grid is located next to the discontinuity and a coarse grid is
dened where the current distribution is less dependent on position.
It must be emphasised that the ratio of the ne and coarse grid size, unlike sub{
gridding in FDTD, can be any arbitrary number. It is not necessary to have this
ratio as an integer number. The discontinuity region is treated individually and only
ne rooftops are used to connect the ne grid region to adjacent coarse grid regions
as illustrated in gure 3.4.
A novel feature of this implementation is that, the size of the rooftop function is
dened as a function of its location as explained in section 3.4.2. To the author's
knowledge this method has not been used elsewhere, for example in [4, 5], the grid
sizes must be identical to exploit the benet of using the Fast Fourier transform
(FFT).
1 2
2 2 2
The division of a metallisation into regions is such that a region is dened between
each discontinuity. This is illustrated by example microstrip circuits in gure 3.5.
The rules of division are not strict and can be adjusted to t each specic example.
As shown in gure 3.5(a), the discontinuity has been sandwiched between two regions
for a microstrip step. If more than two regions are neighbours of a discontinuity,
a \T" junction is produced by the position of the stubs as shown in gure 3.5(b).
For a 90o bend discontinuity, three possible sub{divisions have been illustrated in
gure 3.5(c). A special case of basis functions is used to join two regions and can
be either rooftop functions, or precalculated current basis functions as described in
detail in section 3.6.
3.5 Region Basis Functions Page 44
3.5.2 Denition of Region Basis Function
The inclusion of a priori knowledge of the surface current distribution is essential to
the eciency of the technique. A technique which allows precalculated basis functions
to be used was presented by Railton in [4] for shielded microstrip structures and is
applied to the open planar structures in this present implementation of SDM.
Railton [4] introduced the concept of precalculated basis functions for the modes of
a microstrip resonator. This was extended by Meade [5] to allow a set of arbitrary
basis functions to be dened as,
M
X
J(r) = am m (r) (3.5)
m=1
where m(r) is the current distribution of mth region basis function. A precalculated
region basis function ( m(r) in equation 3.5) is dened in [4,5] as a linear combination
of the rooftop functions which are identical apart from a shift in origin as,
N
X
m (r) = bmnRn(r) (3.6)
n=1
lx 1
w ! w and l ! l
1 2 1 2
w1
rx = ww12 and rz = ll21
l1
lx = rxlx
lz 2
2 1
lx lz = rz lz
2 1
2
w 2
l 2
where L is the length of the microstrip line and Js(x)(s = z; x) is the current dis-
tribution along the line perpendicular to current
ow and precalculated by using the
two{dimensional version of the technique given in detail in appendix B. The Fourier
transform of the current basis function in the direction perpendicular to current
ow
(Js(kx)) is a sum of the Fourier transforms of the corresponding rooftop functions
with precalculated coecients. The Fourier transforms of the line modes in the dir-
ection of current
ow as dened in equations 3.7, 3.8 and derived in appendix A.2 are
given by:
!
Jz (kz ) = an k2(;;1)( nL) cos kz2L +
N
X 0 n n
2 2
n z L
=0
! (3.9)
X 00 2(;1)nkz
N k zL
j an k ; ( n ) cos 2
2 2
n z L
=1
!
X 0 2(;1) kz
N n k zL
Jx(kz ) =j bn k ; ( n ) cos 2 +
2 2
n z L
=0
! (3.10)
X 00 2(;1) L
N n n kzL
bn k ; ( n ) cos 2
2 2
n =1 z L
3.5 Region Basis Functions Page 47
Each line mode can be represented by a precalculated basis function with x and z
components. The number of modes required to fully describe a given length of line
are restricted by the minimum guide wavelength (min ) to be modelled. Therefore
the minimum line mode wavelength (N ) must be less than the minimum microstrip
guide wavelength required.
where N is the wavelength of the highest order line mode and the wavelength of the
nth line mode is given by
n = 2nL (3.12)
The microstrip guide wavelength (min in equation 3.11) is taken as the wavelength
for an innite microstrip line and calculated by using the two{dimensional version
of SDM (see appendix B). This allows the user to specify the maximum operating
frequency and from this, the number of modes and the related line mode current basis
functions are automatically calculated.
At high frequencies, a large number of line modes are needed to accurately model the
unknown current distribution of a microstrip line. For this reason, a wave function is
proposed in the direction of current
ow as a current basis function for the microstrip
line of interest and given in the space domain by:
8
>
< e;jknz L z L
Jz ( z ) = > 2 2
(3.13)
: 0 otherwise
8
>
< e;jkn z; 2
( ) L z L
Jx ( z ) = > 2 2
(3.14)
: 0 otherwise
where kn is the precalculated wave number of an innite microstrip line, where the
width is identical to the microstrip line region, and L is the length of the microstrip
3.5 Region Basis Functions Page 48
line region. The precalculation of kn is carried out by the two{dimensional version of
the technique which is given in appendix B. The same precalculated transverse basis
functions (Jz (x); Jx(x) in equations 3.7 and 3.8) are used in the direction perpendic-
ular to current
ow. The Fourier transforms of equations 3.13 and 3.14 derived in
appendix A.3 are given by:
Note that the centre of the microstrip line region is located at the origin and can be
shifted by multiplying the Fourier transforms by ejkz zoff , where zoff is the oset from
the origin.
where L is the length of the microstrip resonator and Js(x)(s = z; x) is the transverse
current distribution, calculated by the two{dimensional version of the technique (see
appendix B). The Fourier transform of the current basis function in the direction
perpendicular to current
ow (Js(kx)) is also a sum of the Fourier transforms of
3.5 Region Basis Functions Page 49
the corresponding rooftop functions with precalculated coecients, as mentioned in
section 3.5.4. The Fourier transforms of the resonance modes in the direction of
current
ow as dened in equations 3.17, 3.18 are given by:
N 2(;1)n n !
Jz (kz ) = k ; ( n ) cos kz2L
X L
2 2
(3.19)
n z L
=1
!
X 2(;1)n kz
N k zL
Jx(kz ) = j k ; ( n ) cos 2
2 2
(3.20)
n
=1 z L
w -z
L
Dielectric Substrate (r ; r )
Ground Plane
x
Figure 3.7: Microstrip resonator
The maximum number of the resonant modes is determined by the criterion given for
line modes in equation 3.12. The origin is in the centre of the resonator and can be
shifted by multiplying by ejkz zoff , where zoff is the oset from the origin.
This technique has been developed to \extract" the required function from a spot
frequency solution for the current distribution of a region, by comparing this with
the current distribution for a similar region which does not contain any discontinu-
ity. It has been found by Meade [5] that a single function is required to model the
discontinuity over the entire frequency band of interest.
The precalculated discontinuity function depends only on the geometry of the corner
itself, not on the surrounding circuitry. Moreover, the function is also independent
of substrate thickness, permittivity and permeability. It is not limited to the exact
circuit for which it was originally dened, only the relative dimensions of the region
and the adjacent metallisation pattern of interest. This reduces the necessity to dene
new functions for new circuits.
1 2
Centre of transfer
function
3 Area of overlap
into neighbouring
region
Figure 3.8: Illustration of transfer functions for T{shaped junction
w Thickness 0
Ground plane
Figure 3.9: Innite microstrip line for two{dimensional version of SDM
Numerical considerations are introduced in this section with the aid of results from
the two{dimensional version of the technique, for which the formulation is given in
detail in appendix B. The two{dimensional version of SDM is formulated as an
eigenvalue problem to calculate the propagation constant (kz ) of innite microstrip
lines of the form in gure 3.9. From this solution, the eective permittivity, the current
distribution in the transverse direction, the guide wavelength and the characteristic
impedance can be derived. In this two{dimensional analysis, overlapping rooftop
3.8 Three{Dimensional Numerical Examples Page 52
functions (given section 3.4.1) are used.
g = 2k (3.22)
z
As an initial test of condence, the dispersion characteristics for a simple innite
microstrip line are calculated. The tested microstrip line is completely open and of
width 1.27 mm, on a substrate of thickness 1.27 mm and of relative permittivity 20.
Figure 3.10 is a plot of eective permittivity for a frequency range of 0.1 GHz to
20 GHz. The results are compared to the published results of Itoh [1]. A series of
curves are also plotted for dierent number of basis functions. A convergence to the
calculated reference data [1] can be noted.
Eective Permittivity
17
16
n=9
15
n=3
n=1
[1]
14
13
12
4 8 12 16 20
Frequency GHz
Figure 3.10: Dispersion characteristics for the open innite microstrip line
are presented in the forms of resonant frequencies and of scattering parameters. The
derivation of the resonant frequency is given in [3,12] and of scattering parameters is
described in chapter 5. In the three{dimensional analysis, non{overlapping rooftop
functions in the direction perpendicular to current
ow are not favoured and only
overlapping rooftop functions (see section 3.4.1) in both directions are used to analyse
complex metallisation patterns.
The microstrip resonator is the rst example used to test the similarity of the results
given by precomputed transverse basis function and a set of rooftop functions. This
is a useful integrated circuit component at microwave and millimetre{wave frequen-
cies for building lters, oscillators, etc. Before presenting the numerical results to
compare published results and experimental data, it should be noted that the struc-
tures used in [3] for both numerical computations and experiments are scale models
of millimetre{wave integrated circuits. Because in practice after a circuit is designed
and tested at low frequency, a miniature structure for millimetre{wave integrated
circuits may be obtained by reducing all dimensions of the circuit structure while
keeping the ratios of all the dimensions to the wavelength constant. The value of
the relative dielectric constant of the substrate is assumed to be unchanged in these
two frequency ranges. Since the eld distributions in both the scale model and the
actual structure are the same, the eld problem of millimetre{wave integrated circuit
structures is being solved even though the actual operating frequency is in the UHF
range.
To test the accuracy of the two approaches, the rst resonant frequency of the mi-
crostrip resonator on a substrate of thickness 1.27 cm and of relative permittivity
3.82 is calculated. The calculation of the resonant frequency is given in [3, 12]. In
the analysis, a set of rooftop functions and a precalculated transverse basis function,
in which the same number of rooftop functions are used for precalculation using the
3.8 Three{Dimensional Numerical Examples Page 55
two{dimensional version of the technique, are used in the direction perpendicular to
current
ow to test the similarity in accuracy. Both models are compared with exper-
imental and published data [3]. For the rst model, a total of 48 rooftop functions as
current basis functions are required, whereas the number of basis functions required
are reduced to 16 by the use of the precalculated transverse basis function for the
same denition (lx = w=4 and lz = L=9).
Resonant Frequency (MHz)
1000
900 Precalculated
Rooftops
+ Measured +
Itoh [3]
800 Quasi-TEM [3]
Paralel plate [3]
+
w = 6:78 cm
700
+
600
+
500
8 9 10 11 12 13 14
Strip Length (cm)
Figure 3.11: Comparison of the results from the present method with those obtained
from experiments and other methods
The rst resonant frequency versus length of the resonator is plotted in gure 3.11, in
which the results obtained by using the precalculated transverse basis function and
those obtained using a set of rooftop functions are almost identical. Other theoret-
ical results are based on the Spectral Domain Method by using entire domain basis
function, quasi{TEM and parallel{plate transmission line models. The experiments
were conducted by Itoh [3], using a strip of thickness 0.254 mm which is negligibly
small compared to other dimensions. As seen in gure 3.11, the results for the two
kinds of current basis functions are indistinguishable on the graphical gure and the
agreements between experimental data and the results computed by the proposed
3.8 Three{Dimensional Numerical Examples Page 56
technique are extremely good.
In order to further prove that a precalculated transverse basis function is valid for
a wide frequency band and gives similar results to a set of rooftop functions, the
microstrip gap discontinuity consisting of two identical end{coupled microstrip lines as
shown in gure 3.12 is modelled. The dimensions of the circuit are given in gure 3.12.
Port 1 Port 2
m
36m
w=1.272mm w=1.272mm
d=1.272mm
0.6
g=
r = 10; r = 1
L=5.088mm
Figure 3.12: Microstrip gap layout
72 basis functions in total (36 x and 36 z components) are required for the sub{
domain analysis for the denition (lx = lz = w=4 = 0:318 mm). The transverse basis
function is precalculated for the same denition (lx = w=4 = 0:318 mm) and used as
one basis function which is a sum of a number of rooftop functions with precalculated
coecients, as a result 24 basis functions become sucient for the same circuit. It is
emphasised that the plot of S{parameters in gure 3.13 clearly shows the agreement
of a set of rooftop functions to a precalculated transverse basis function on a wide
frequency band over the frequency range 1 GHz to 20 GHz, although the precalculated
basis function has been calculated at a single spot frequency.
3.8 Three{Dimensional Numerical Examples Page 57
S{Parameters (dB)
0
++
+
++++
++
+
+++
-10
++ + ++++
++++
++
+ +++
-20
+ +++
++ Precalculated (jS11 j)
+ Rooftop (jS11 j)
-30 + Precalculated (jS21 j)
Rooftop (jS21 j) +
+
-40
4 8 12 16 20
Frequency GHz
Figure 3.13: Plot of S{parameters's magnitude for the microstrip gap discontinuity
comparing basis function sets
The straight microstrip line with no discontinuity is the simplest region and a mi-
crostrip line in a nite length shown in gure 3.14 is the simplest microwave circuit to
test the proposed region basis function discussed in section 3.5.4. The microstrip line
in gure 3.14 is of length 6 mm, and of width 1.272 mm on a substrate of thickness
1.272 mm. The permittivity and permeability are 10 and 1 respectively.
3.8 Three{Dimensional Numerical Examples Page 58
Port 1 Port 2
w=1.272mm
d=1.272mm
L=6mm
r = 10; r = 1
Three sets of S{parameter results are plotted in gure 3.15. These are the results
calculated by using 36 rooftop functions for each current components, the results cal-
culated by utilising precalculated transverse basis functions in which case the number
of basis functions required are reduced to 6 for the same grid size (lx = w=4 and
lz = L=7) and the results calculated by using the single proposed microstrip line
region basis function given in section 3.5.4. As seen in equations 3.13 and 3.14, the
precalculated current wave function is an analytical function with a precalculated
wave number and thus dierent than the precalculated functions in [4,5].
S{Parameters (dB)
0+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
-10
-20
-30
-40
Region (jS11 j)
Transverse (jS11 j)
Rooftop(jS11 j)
-50 Region (jS21 j)
Transverse (jS21 j)
-60
Rooftop(jS21 j) +
8 12 16 20
Frequency GHz
As illustrated in gure 3.16, the lter consists of a central microstrip resonator and two
feedlines of length 12.72 mm and width 1.272 mm. The substrate has thickness 1.272
mm, relative permittivity 10 and relative permeability 1. The edge{coupled lter
is analysed using rooftop functions, precalculated transverse functions and resonant
mode region basis functions.
Port 1 L=25.44mm
l=12.72mm
5w=6.36mm
d=1.272mm
w
r = 10; r = 1 Port 2
Note that for model 1 and 2 only one basis function is required to fully describe the
strip resonator over the frequency range of interest. In total 114 basis functions (57
x and 57 z components) are required for the same structure, if only precalculated
transverse basis functions are used. In contrast to this, 342 rooftop functions (171
x and 171 z components) would be required in the basic method in which rooftop
functions are used as sub{domain basis functions for equivalent accuracy for the
same rooftop sizes (lx = 0:318 mm and lz = 0:636 mm). This allows a large saving
in computer memory and computational time per frequency. Figure 3.18 shows the
results for magnitude of S{parameters plotted against operating frequency. Model 2
agrees more closely with the basic rooftop, the precalculated transverse basis function
calculation and measured data [13].
3.8 Three{Dimensional Numerical Examples Page 61
n=1
n=2
n=3
n=4
Figure 3.17: Current distributions for the rst four resonant modes of a strip resonator
3.8.4 Sub{gridding
As mentioned in section 3.4.1, the grid sizes must be identical in order to exploit
the benet of using the Fast Fourier transform (FFT) for shielded planar microwave
circuits [4]. A major dierence between the boxed and the open case is that in
the former, the discrete Fourier transform is employed, whereas for the latter, the
continuous transform is required. Therefore the benets exploited by using the FFT
are not available. In this implementation sub{gridding is employed for the analysis of
complex metallisation patterns. The sub{gridding is used in the sense that the size of
the rooftop functions are dened as functions of their location. Fine rooftops are used
next to the discontinuity. Moreover the precalculated current basis functions are used
3.8 Three{Dimensional Numerical Examples Page 62
-10
-15
Rooftop
Transverse
-20 Model 1
Model 2
3 6 9 12 15
Frequency GHz
-10
-20
Insertion Loss (dB)
-30
-40
-50
Measured
-60 Rooftop
Transverse
Model 1
-70 Model 2
-80
3 6 9 12 15
Frequency GHz
Figure 3.18: Plot of S{parameters' magnitude for the edge{coupled lter, comparing
basis function sets
3.8 Three{Dimensional Numerical Examples Page 63
where only slow changes in the current distribution occur. As a test of condence, the
microstrip step discontinuity shown in gure 3.19 is taken as an example structure.
The planar circuit in gure 3.19 is completely open on a substrate of thickness 1.272
mm and relative permittivity 10. The other dimensions are given in gure 3.19.
w =2.544mm
w =0.636mm
2
1
1
l =3mm
1 l =5mm
2 l =3mm
1
Three sets of S{parameter results are plotted in gure 3.20. To illustrate the conver-
gence pattern, the circuit in gure 3.19 is analysed by using ne rooftop functions,
which are 162 in total (81 x and 81 z components) for the denition (lx = 0:318 mm
and lz = 0:5 mm), for the entire metallisation pattern. For the coarse grid analysis,
only three microstrip line region basis functions and two transfer current basis func-
tions which are ne rooftop functions overlap into neighbouring regions are required.
In the sub{grid analysis, the benets of using precalculated region basis functions
are exploited and almost same accuracy has been achieved with only 42 current basis
functions (21 x and 21 z components). For this analysis, ne rooftops are used to
model current distribution in the shaded area of length 1 mm for each region. The
current basis function required for an accurate analysis are reduced by 74% and the
results are found almost identical to the results by the ne grid analysis.
3.8 Three{Dimensional Numerical Examples Page 64
-20
-40
4 8 12 16 20
Frequency GHz
-10
Insertion Loss (dB)
-20
-40
4 8 12 16 20
Frequency GHz
References
[1] T. Itoh and R. Mittra, \Spectral{domain approach for calculating the dispersion
characteristics of microstrip lines," IEEE Transaction on Microwave Theory and
Technique, pp. 496{499, July 1973.
[2] T. Itoh and R. Mittra, \A technique for computing dispersion characteristics of
shielded microstrip lines," IEEE Transaction on Microwave Theory and Tech-
nique, pp. 896{898, October 1974.
[3] T. Itoh, \Analysis of microstrip resonators," IEEE Transaction on Microwave
Theory and Technique, vol. 22, pp. 946{952, November 1974.
[4] C. J. Railton and S. A. Meade, \Fast rigorous analysis of shielded planar lters,"
IEEE Transaction on Microwave Theory and Technique, vol. 40, pp. 978{985,
May 1992.
[5] S. A. Meade and C. J. Railton, \Ecient implementation of the spectral domain
method including precalculated corner basis functions," IEEE Transaction on
Microwave Theory and Technique, vol. 42, pp. 1678{1684, September 1994.
[6] R. W. Jackson, \Full{wave, nite element analysis of irregular microstrip dis-
continuities," IEEE Transaction on Microwave Theory and Technique, vol. 37,
pp. 81{89, January 1989.
[7] H. H. Balik and C. J. Railton, \Sub{gridding in the spectral domain method for
the analysis of planar circuits and antennas," in 3rd International conference on
telecommunications in modern satellite, cable and broadcasting services, pp. 592{
595, October 1997.
[8] J. C. Rautio and R. F. Harrington, \An electromagnetic time{harmonic analysis
of shielded microstrip circuits," IEEE Transaction on Microwave Theory and
Technique, vol. 35, pp. 726{729, August 1987.
[9] R. W. Jackson and D. M. Pozar, \Full{wave analysis of microstrip open{end and
gap discontinuities," IEEE Transaction on Microwave Theory and Technique,
vol. 33, pp. 1036{1042, October 1985.
Chapter 3: References Page 67
[10] W. P. Harokopus and P. B. Katehi, \Characterisation of microstrip discontinu-
ities on multilayer dielectric substrates including radiation losses," IEEE Trans-
action on Microwave Theory and Technique, vol. 37, pp. 2058{2066, December
1989.
[11] S. A. Meade, The Rapid and Rigorous Mathematical Modelling of Passive Planar
Microwave Circuits. PhD thesis, University of Bristol, March 1994.
[12] T. Itoh and W. Menzel, \A full{wave analysis method for open microstrip struc-
tures," IEEE Transactions on Antennas and Propagation, vol. 29, pp. 63{67,
January 1981.
[13] T. Shibata, T. Hayashi, and T. Kimura, \Analysis of microstrip circuits using
three{dimensional full{wave electromagnetic eld analysis in the time domain,"
IEEE Transaction on Microwave Theory and Technique, vol. 36, pp. 1064{1070,
June 1988.
Chapter 4
An ecient adaptive truncation for the impedance matrix integration shown in equa-
tion 2.10, which is over an innite surface due to open structure, is presented in this
chapter. This limits the numerical integration to nite computer resources by using
the features of the current basis function. In addition the integration step sizes of the
numerical integration are also dened as functions of the Fourier transform variables
as will be explained in section 4.2.2. The present implementation uses the asymptotic
4.1 Introduction Page 70
form of the Green's function [1] in the calculation of the frequency{independent part
of the impedance matrix, resulting in a much smaller two{dimensional numerical in-
tegration to be calculated for each spot frequency. It must be remembered that the
ecient truncation of the integration has already been employed, and as a result, spe-
cial care must be taken to determine the integration range of the frequency{dependent
part of the impedance matrix. The above enhancements are implemented to speed
up the two{dimensional numerical integration for the impedance matrix calculation.
The location vector calculation is an enhancement which reduces the number of im-
pedance matrix elements to be calculated. This technique is employed in the region
where identical grid sizes are used. In this region, the rooftop functions are dened
identical apart from a shift in origin. By this denition, the rooftop function can be
split into two parts: the rst part is identical for each rooftop and independent of its
location, and the second part is a vector which represents its location.
Because the sizes of rooftop functions are not identical in this implementation of
SDM, the above location vector calculation is not applicable to the entire impedance
matrix. For the entire impedance matrix elements which are the products of the
rooftop and the precalculated functions, the symmetry in both the impedance matrix
and the Green's function is exploited.
One enhancement presented here cannot completely be classied one way or the other:
it is actually a combination of two types of enhancements mentioned above. It reduces
the number of the impedance matrix elements in one dimension and hence speeds up
overall numerical integration. The eects of the loss in the dielectric substrate are
also included into the analysis, by replacing r by r (1 ; j tan ) where tan is the
loss tangent of the dielectric substrate.
The enhanced SDM is applied to the analysis of a microstrip line for the eective
permittivity to show the numerical convergence of the adaptive integration and the
use of the asymptotic form of the Green's function. The microstrip line has also been
4.2 Adaptive Integration Page 71
analysed for several loss tangents to observe the eects of the dielectric loss.
There are two possible ways to include such eects as described in the literature.
The rst way is to determine the exact pole location by using one of the numerical
root{nding techniques such as the Newton{Rhapson, which is well{explained in any
numerical analysis text book, then to skip the poles and to include the eect of
the poles as explained in [4]. The main disadvantage of this technique is the need
to calculate the location of the poles for each spot frequency. A second and more
ecient technique is introduced in [5] for the polar co{ordinate systems. This has
been adopted to t the cartesian co{ordinate system used in this implementation.
In a line integration, the result is independent of the chosen path [6, page 409]. The
technique used here is based on changing the integration path to a new path on
which no poles exist. As the poles are located between jkoj and jpr r k j as shown in
0
gure 4.1, a contour integration path which skips this region obtained by shifting the
integration path by 5% of k , as shown in gure 4.1. The value of 5% has been chosen
0
from the author's experience, and the eects of poles are included in the analysis.
In a numerical integration procedure, there are two parameters which have the greatest
eects. These parameters are discussed in the following sections and enhancements
are described.
4.2 Adaptive Integration Page 72
Im(kz;x)
j 0:05k0
Magnitude
2lz
1:5lz Triangle
Step
lz
0:5lz
0
;0:5lz
; lz4
; lz
3
; lz
2
; lz 0
lz
2
lz
3
lz
4
lz
Transform variable (kz;x )
In gure 4.3, max is the truncation position which is described in section 4.2.1 and
the integration step as a function of the transform variables is given by;
hmin jkx;z j
h(kx;z ) = hmax e ln( hmax ) max (4.1)
As seen in equation 4.1 and gure 4.3, ne integration steps are used for the small
values of the Fourier transform variables and coarse steps are used for large values of
4.3 Asymptotic Form of Green's Function Page 74
Integration Step
hmax
h = h(kx;z )
hmin
kx;z
-max max
Transform Variables (kx;z )
Figure 4.3: Adaptive integration step
The spectral domain form of the Green's function (G in equation 2.5) is a function
of the Fourier transform variables (kz and kx in equation 2.6), thickness of dielectric
substrate and substrate parameters (r and r ) as well as operating frequency. There-
fore any improvement in eciency in calculating the terms of the Green's function
will result in a signicant reduction in the run{time required.
4.3 Asymptotic Form of Green's Function Page 75
The calculation of the full Green's function is relatively computationally intensive,
therefore the derivation of a simple asymptotic form of the function is desirable. This
is a function which is valid for large values of kz and kx. Such a function is presented
by Railton and Rozzi [7]
G1(kx; kz ) = kx;klim
z !1
G(kx; kz ) (4.2)
The asymptotic form of the Green's function is similar to the one which has been
presented by Railton [8] for shielded planar structures, but modied to be applicable
to open planar microwave circuits. For large values of kx and kz , two approximations
can be made. These are
q q
i = kx + kz ; ki kx + kz
2 2 2 2 2
(4.3)
Coth
d 1 2 (4.4)
The asymptotic form of the Green's function for large kx and kz derived in ap-
pendix C.1 is given by:
where
h = ; ! (1j+ )
1 h = (1j!
2
r
+ r )
0
0 r
K1zz = q kz K1zz = q kz
2 2
1
kx + kz
2 2
2
(kx + kz )
2 2 3
Kxz = Kzx
1 1 Kxz = Kzx
2 2
K1xx = q kx K1xx = q kx
2 2
1
kx + kz
2 2
2
(kx + kz )
2 2 3
4.3 Asymptotic Form of Green's Function Page 76
with ! = 2f and r ; r the relative permittivity and permeability of the electric
substrate respectively.
The functions K1 sti (s; t = z or x; i = 1; 2) are just functions of kx and kz and thus
are independent of frequency and metallisation pattern. The constants hi (i = 1; 2)
need only be calculated once per frequency, as they are independent of kx and kz .
Thus the integration over an innite surface is replaced by a double integration over
just a few cycles of Fourier transform of the rooftop function components. The
truncation of the innite integration is given in detail in section 4.2.1.
Using the asymptotic form of the Green's function as given by equation 4.5 the
impedance matrix in equation 3.8 is now split into two parts:
Z max0z Z max0x
Zst = ;max0z ;max0x
wt(kx; kz ) [G(kx; kz ; d; !) ; G1(kx; kz )] Js(kx; kz )dkxdkz
+ Z1st (4.8)
where
Z maxz Z maxx
Z1st1 = wt(kx; kz )K1st (kx; kz )Js(kx; kz )dkxdkz (4.11)
Z;maxz Z;maxx
1
maxz maxx
Z1st =
2
;maxz ;maxx
wt(kx; kz )K1st (kx; kz )Js(kx; kz )dkxdkz
2 (4.12)
hi and K1sti (s; t = x or z ; i = 1; 2) are dened for the asymptotic form of the Green's
function in equation 4.5.
Following section 4.3.1, Zst and Zst are only functions of Fourier transform vari-
1 2
ables and independent of the operating frequency and dielectric substrate parameters
and therefore need to be calculated only once per metallisation pattern. Thus the
algorithm allows the frequency, permittivity, permeability and substrate thickness
to be changed without requiring Zst and Zst to be re{calculated. Moreover only
1 2
the rst integration in equation 4.8 and simple hi (i = 1; 2) are repeatedly calcu-
lated for each spot frequency, but it is emphasised that the integration range has
been eciently truncated by using the features of the rooftop function's components.
Therefore special care must be taken to determine the truncation location of the nu-
merical integration of the dyadic Green's function by satisfying equations 4.3 and 4.4.
Since it has been found and illustrated in section 4.2.1 that the numerical integration
of the impedance matrix elements over a just a few cycles of the Fourier transform
of the rooftop function's components gives accurate results, instead of integrating
4.4 Location Vector Calculation Page 78
over an innite surface, the conditions given in equations 4.3 and 4.4 then become
inapplicable. Therefore the conditions 4.3 and 4.4 must be checked for each spot
frequency and terminate the use of the asymptotic form of the Green's function when
anyone of equations 4.3 and 4.4 is not satised.
where Pn is the function of the location of the nth rooftop, whereas Jn is a function
of the dimensions of the rooftop, but independent of its location as follows,
Pn = ej kxxn
( +kz zn ) = ej kxplx
( kz qlz )
+
lx; lz are grid sizes and xn; zn, which are functions of p; q are the origin of the nth
rooftop function. The set of functions Pn(r) are referred to as the location vector. The
maximum values of p; q are P; Q which are the total number of the rooftop functions
required for the region of interest. The Method of Moments is the technique used
to nd the unknown coecients of the current basis functions. A set of weighting
functions, which are identical to the set of basis functions for the metallisation of the
circuit, are required. Therefore the impedance matrix elements for the region where
4.4 Location Vector Calculation Page 79
the sizes of the grids are identical are given by
Z1Z1
Zst (p; q) = ;1 ;1
(JsJt Gst)PsPtdkxdkz s; t = x; z (4.15)
To facilitate the solution and speed up the impedance matrix calculation, the imped-
ance matrix of the region of interest is divided into four quadrants as,
2 3
Z (p; q) Zzx(p; q) 7
Z = 64 zz 5 (4.16)
Zxz (p; q) Zxx(p; q)
In equation 4.16, only half of the zz, of xx and zx are required to be calculated if
symmetry is used alone (given in detail in section 4.5.1). After substitution of 4.14
into 4.15, the impedance matrix elements ( in equation 4.15) dier from each other
by their integer numbers (p and q in equation 4.14). It has been found that the rst
line alone of each quarter of the matrix is sucient to nd whole region of interest in
the impedance matrix by using re{dened rooftops. The impedance matrix elements,
as a function of the integer numbers (p; q) can be written as,
Z1Z1 ;1)
PX
( ( ;1)
QX
Zst (p; q) = ;1 ;1
(JsJtGst ) ejkxplx ejkz qlz (4.17)
p=0 q=0
Zst (;p; q) = Zst (p; q) Zst (p; ;q) = Zst (p; q) (4.18)
Zst (;p; q) = ;Zst (p; q) Zst (p; ;q) = ;Zst (p; q) (4.19)
Equation 4.18 is used to calculate the rst and fourth quadrants of the impedance
matrix and equation 4.19 is used for the second quadrant. This technique gives great
savings in computer run{time, especially for complex circuits. The total number of
impedance matrix elements for the region of interest are (2PQ) whereas the number 2
4.5 Symmetry Page 80
of impedance matrix elements required to be calculated by using the method proposed
here are only 3PQ.
4.5 Symmetry
4.5.1 Symmetry in Impedance Matrix
The minimisation of the number of the impedance matrix elements required to be
calculated is crucial to achieve an ecient implementation. Therefore symmetries in
the impedance matrix calculation are taken advantage of the algorithm in a similar
manner to [3]. As illustrated in gure 4.4 the square impedance matrix is divided
into four quadrants which are Zzz ; Zzx; Zxz and Zxx. As a result of the Method of
Moments, the following symmetries apply:
where Zst(n; m) represents the product of nth weighting function with mth basis func-
tion.
Therefore the algorithm requires half of the impedance matrix shown in gure 4.4
plus the n = m diagonal to be calculated. The rest of the impedance matrix elements
are found by using the symmetries.
Extra symmetries can be applied to Zzx, if the procedure is Galerkin's and the basis
functions are rooftop functions with identical dimensions. In this case, the loca-
tion vector calculation given in detail in section 4.4 is more ecient than the use of
symmetry. The symmetry in the impedance matrix is favoured when precalculated
current basis functions and sub{gridding of the metallisation are used to dene the
4.5 Symmetry Page 81
Calculated region
Zzz Zzx
Z= Extra symmetry
Zxz Zxx
where s; t = x; z. These features are exploited for the calculation of Zst in equa-
tion 4.8. In this implementation the Green's function is calculated for only one
quarter of the integration range (kx 0 and kz 0) and the rest are derived by using
the equations 4.20, 4.21.
4.6 Enhancement for Numerical Integration Page 82
4.6 Enhancement for Numerical Integration
In the previous sections, modications to speed up the impedance matrix calculation
and to reduce the number of impedance matrix elements have been discussed. The
improvement introduced in this section is actually a combination of these two kinds
of enhancements. The speed of the calculation of the impedance matrix elements is
increased by reducing the number of the impedance matrix elements to be calculated
in one dimension. As mentioned previously, the rst step of the analysis is to expand
the unknown current distribution as a set of known current basis functions with
unknown coecients, as given in equation 2.8. The known current basis functions
which can be either sub{domain or precalculated functions are dened as two separate
functions for each direction as:
N
X
Js(kx; kz ) = asnJsn(kx; kz )
n=1 (4.22)
Nx Nz !
X X
= asn Jsnx(kx) Jsnz (kz ) n = 1::N
nx=1 nz=1
and the Method of Moments requires a set of weighting functions, which is identical
to the set of current basis functions if the procedure is Galerkin's, dened by:
N
X
wt(kx; kz ) = wtn(kx; kz )
n=1 (4.23)
Nx
X Nz
X
= wtnx(kx) wsnz (kz ) n = 1::N
nx=1 nz=1
The integration thus splits into two dependent integrations, but the number of integ-
rands are independent for each integration in one dimension. The run{time for the
impedance matrix calculation is a function of the number of current basis functions
used in both directions and can be formulated as;
Nz (2NxNz + 1) (4.26)
(2Nx + 1)
To illustrate the eciency of the proposed technique, a simple microstrip line of nite
length as shown in gure 4.5 is taken as an example. The dimensions of the circuit
are not considered because the basic idea of the enhancement is discussed instead of
presenting numerical results.
Port 1 Port 2
w
L
d r ; r
Figure 4.5: Microstrip line
A total of 57 rooftop functions are required for the analysis which does not use any pre-
calculated current basis functions in either direction for the denition (lx = w=4; lz =
L=20). Therefore N in equations 4.22 and 4.23 is 57. The size of the impedance
matrix is double of the number of current basis functions and N (2N + 1), which is
6555, the impedance matrix elements need to be calculated for each spot frequency
4.7 Inclusion of Loss in Dielectric Substrate Page 84
if symmetries in the impedance matrix are used. But the rooftop functions are two
separable functions and the integration can be split into two dependent parts. In this
case (N 1(2N 1 + 1), which is 21, sub{impedance matrix elements are required to be
calculated and 6555 impedance matrix elements are derived by facilitating 21 sub{
impedance matrix elements. If precalculated current basis functions are used, the
savings in computer memory and run{rime are even greater. In the author's exper-
ience, this enhancement gives up to 99% saving for the one dimensional integration
and up to 90% for overall integration.
It must be noted that as the current basis functions are dened as two separable
functions, the values of the current basis functions at each integral point are only
calculated once and used during numerical integration for the entire frequency range.
where 00r + ! may be considered as the imaginary part of the permittivity or !00r +
as the total conductivity. The loss tangent of a dielectric medium is dened by:
00r + !
tan = 0 (4.28)
r
and the complex relative permittivity of a dielectric substrate can be written by using
4.8 Numerical Convergence Page 85
the loss tangent (tan ) as;
r = r (1 ; j tan )
loss (4.29)
Any measurement of tan always includes the eects of the nite conductivity. At
microwave frequencies, however 00r is usually much larger than ! . In this implement-
ation, the dielectric losses are included by replacing r by r (1 ; j tan ).
18
This Technique
[10]
17 f=10 GHz
16
15
To illustrate the convergence pattern, results are presented for the example of two{
dimensional open microstrip line of width 1.27 mm on a substrate of thickness 1.27
mm and of permittivity 20. In gure 4.6, the eective permittivity versus the number
of cycles of the Fourier transform of the rooftop function included in the numerical
integration is plotted. It is evident that relative convergence for this example has
4.8 Numerical Convergence Page 86
been reached when Nc 1 (Nc is the number of cycles). The exact value of Nc
required for convergence is problem dependent, but is usually of the order of 1 even
for more complicated three{dimensional structures.
Eective Permittivity
7.5
7
Adaptive
6.5 Uniform
[10]
6
4 8 12 16 20
Frequency GHz
The convergence pattern for the adaptive integration step size is shown in gure 4.7
for the same structure. First the microstrip line has been modelled using the adaptive
integration step as a function of the Fourier transform variables. Latter, it has been
modelled using uniform integration steps while keeping the number of integration
steps unchanged. Both results are compared with the published results by Itoh [10].
Even though identical number of integration steps are used, the accuracy is improved
by using the adaptive integration technique as illustrated in gure 4.7.
Eective Permittivity
8
7.95
7.8
7.75
7.7
0.2 0.4 0.6 0.8 1
max0 =max
permittivity versus the ratio of max0 and max, given in equations 4.7, 4.8 and 4.9, is
plotted. Since the numerical integration is over one cycle, max is given by;
where lx = w=4. max0 is the integration range of the asymptotic part of the im-
pedance matrix and a function of the operating frequency. It is evident that relative
convergence for this example has been reached when ratio 0:2, that is a small
percentage error compared to large values of the ratio. The exact value of ratio is
problem dependent and determined by the conditions given in equations 4.3 and 4.4.
Eective Permittivity
11.5
11
10.5
10
9.5 tan = 0
tan = 0:5
9 tan = 1
8.5
8
7.5
5 10 15 20 25 30 35 40
Frequency (GHz)
Figure 4.9: Eective permittivity of the microstrip line for dierent loss tangents of
the dielectric substrate
Secondly, S{parameter results are presented for a simple microstrip line, which has
been modelled in section 3.8.3.1, for dierent loss tangents. In this model, a precalcu-
lated transverse basis function is used in the direction perpendicular to current
ow,
whereas a current wave with a precalculated wave number is used in the direction
of propagation. As illustrated in gure 4.10, S is almost unity for the substrate
21
without any loss, because there is neither discontinuity nor loss in the circuit. This
is no longer valid for the microstrip line with dielectric loss. Even though there is
no singularity in the microstrip line, the S is not around unity, this is because the
21
energy is lost in the lossy microstrip line. The loss is proportional to the loss tangent
4.10 Summary Page 89
of the dielectric substrate as illustrated in gure 4.10. The energy loss also increases
as the operating frequency increases.
S{Parameters (dB)
0
-10
-20
-30
-40
-50 tan = 0 (jS j)
tan = 0:1 (jS
11
j)
-60 tan = 0:5 (jS11
j)
tan = 0 (jS11
j)
-70 tan = 0:1 (jS
21
j)
tan = 0:5 (jS21
21 j)
-80
5 10 15 20
Frequency GHz
Figure 4.10: S{parameters' magnitude for the microstrip line with dierent loss tan-
gents of the dielectric substrate
4.10 Summary
In this chapter, enhancements to improve the calculation of the impedance matrix
have been described. The adaptive integration technique, which includes an adaptive
truncation and an adaptive numerical integration step size has been proposed, to
speed up the impedance matrix integration. The asymptotic form of the Green's
function, which splits the impedance matrix into frequency dependent and asymptotic
parts has been adopted for open planar microwave circuits. The location vector
calculation has been introduced and applied in the region where identical grid sizes are
used, to reduce the number of the impedance matrix elements to be calculated. The
nature of the basis function which is to be dened as two separable functions for each
4.10 Summary Page 90
direction has been exploited to improve the numerical two{dimensional integration.
Moreover the loss in the dielectric substrate has been included into the analysis and
its eect has been illustrated on the simple microstrip structures.
Chapter 4: References
References
[1] C. J. Railton and S. A. Meade, \Fast rigorous analysis of shielded planar lters,"
IEEE Transaction on Microwave Theory and Technique, vol. 40, pp. 978{985,
May 1992.
[2] T. Itoh, Numerical Techniques for Microwave and Millimeter{Wave Passive
Structures. John Willey and Sons, 1989.
[3] R. W. Jackson, \Full{wave, nite element analysis of irregular microstrip dis-
continuities," IEEE Transaction on Microwave Theory and Technique, vol. 37,
pp. 81{89, January 1989.
[4] D. M. Pozar, \Input impedance and mutual coupling of rectangular microstrip
antennas," IEEE Transaction on Antennas and Propagation, vol. 30, pp. 1191{
1196, November 1982.
[5] E. H. Newman and D. Forrai, \Scattering from a microstrip patch," IEEE Trans-
action on Antennas and Propagation, vol. 35, pp. 245{251, March 1987.
[6] E. Kreyszing, Advanced Engineering Mathematics. Jhon Willy, 5. ed., 1983.
[7] C. J. Railton and T. Rozzi, \Complex modes in boxed microstrip," IEEE Trans-
action on Microwave Theory and Technique, vol. 36, pp. 865{873, May 1988.
[8] C. J. Railton and J. P. McGeehan, \A rigorous and computationally ecient
analysis of microstrip for use as an electro{optic modulator," IEEE Transaction
on Microwave Theory and Technique, vol. 37, pp. 1099{1103, July 1989.
[9] R. E. Collin, Foundations for Microwave Engineering. McGraw-Hill, 1992.
[10] T. Itoh and R. Mittra, \Spectral{domain approach for calculating the dispersion
characteristics of microstrip lines," IEEE Transaction on Microwave Theory and
Technique, pp. 496{499, July 1973.
Chapter 5
For the shielded planar microwave circuits, the S{parameters at a spot frequency
are derived by Meade [1, chapter 5] from the solution of the matrix equation (equa-
tion 2.9), i.e. the set of coecients dene the unknown surface current distribution
for an excitation conguration dened by the excitation vector (Vt in equation 2.9).
This implementation cannot simply be applied to open structures without further
modications, because there are no side walls where the tangential electric eld is
identically zero.
A method was introduced by Jackson [2, 3] in 1985 for open planar structures, but
his technique, in contrast to the method described here, is not ecient and requires
a large number of extra sub{domain basis functions at relatively low frequencies,
therefore a method is needed to yield accurate results in the whole microwave region
and must be eective in remedying this deciency.
To simplify the excitation, the circuit is represented as a black box and feedlines are
connected to the ports of the circuit as shown in gure 5.1. The excitation problem
now becomes the denition of the unknown current distribution on the metallisation
of the entire system, which consists of a microwave circuit with feedlines connected to
the ports. Denition of the unknown surface current distribution on the metallisation
of the circuit (a black box in gure 5.2) is discussed in detail in chapter 3. A set of basis
5.1 Introduction Page 94
functions are required to dene the unknown current distribution on the feedlines, in
order to complete the algorithm.
Microwave Circuit
Port 1 Port 2
r ; r
Feedlines
Figure 5.1: Illustration of excitation of the two port microwave circuit
The length of the feedline is theoretically innite, but in practice is an integer num-
ber of half wavelengths of the travelling current wave. In this implementation, the
travelling current wave which is directed into the input port is called the incident
current wave and that directed out of the input port is called the re
ected current
5.1 Introduction Page 95
wave. The transmitted current wave is the travelling current wave emanating from
the output port as illustrated in gure 5.2.
The basic philosophy of the excitation, shown in gure 5.2, is that the incident current
wave with a unit amplitude comes from innity and goes into the input port. Some
of the current travels through the circuit towards the output port, and some of the
current re
ects from the input port.
Incident Microwave
Re
ected Circuit Transmitted
Connect the feedlines to the ports of the circuit. The width of the feedline must
be identical to the width of the corresponding port
Dene the unknown current distribution on the metallisation of the circuit as a
set of known current basis functions with unknown coecients
Use the sum of an incident current wave with unit amplitude and a re
ected
current wave with unknown amplitude for the feedline connected to the input
port and a transmitted current wave with unknown amplitude for the feedline
connected to output port as current basis functions
Use the compensation function at the interface between the black box and ad-
jacent feedline in order to transmit the full eects of the excitation into the
circuit
5.2 Excitation Modelling Page 96
Apply the Method of Moments to eliminate the electric eld components and
solve the matrix equation, i.e. the set of unknown current coecients which
denes the unknown current distribution for the incident current wave with
unit amplitude
Calculate the S{parameters using the re
ected and transmitted current wave
coecients.
Repeat this procedure for each spot frequency, to nd the S{parameters of the
circuit for the frequency range of interest .
In the following sections, the ecient excitation modelling for open planar circuits is
described in detail and compared with available techniques. Examples are also given
to highlight the eciency of the proposed technique.
Circuits in this thesis are excited by connecting semi{innite feedlines whose widths
are identical to the joining ports, therefore the excitation has been simplied to
the denition of the unknown current distribution of the system. The fundamental
microstrip mode is assumed to propagate on each feedline and thus a travelling current
wave is chosen as a current basis function. Derivations of the incident, re
ected and
transmitted current waves by using the travelling current wave in the direction of
5.2 Excitation Modelling Page 97
Port 1 Port 2
Ji(z) = >
: 0 otherwise
8
>
< ;ar ejkn z;zi ;L + zi z zi
( )
Jt (z) = >
: 0 otherwise
Jinput = Ji + Jr
(5.2)
= (1 ; ar ) cos(kn(z ; zi)) ; j (1 + ar ) sin(kn(z ; zi ))
Joutput = Jt
(5.3)
= at [cos(kn(z ; zo)) ; j sin(kn(z ; zo ))]
Note that these functions are in the space domain and Fourier transforms are required
in the Spectral Domain Method. The Fourier transforms of the current basis functions
of the feedlines as dened in equation 5.1 and derived in appendix D.1 are given by;
Ji(kn; kz ) = k ;2 k sin (kz ; kn) L2 e;j kz ;kn L2 ejkz zi
( )
(5.4)
z n
2 L
Jr (kn; kz ) = k + k sin (kz + kn) 2 e;j kz kn L2 ejkz zi
( + )
(5.5)
z n
Jt (kn; kz ) = k ;2 k sin (kz ; kn) L2 ej kz ;kn L2 ejkz zo
( )
(5.6)
z n
where L is the length of the feedline, which theoretically extends to innity, but will
yield accurate results if it is an integer number of half wavelengths [2,3]. The current
basis functions for the feedlines in both directions in the spectral domain are dened
by;
where functions gzin; gxin; gzo and gxo are the sum of the Fourier transforms of the rooftop
functions with previously calculated coecients by the two{dimensional version of the
technique (given in detail in appendix B).
The denition of the current basis functions for the feedlines connected to the ports
of the circuit is described as shown in gure 5.4. As shown in equations 5.2, 5.3 and in
5.2 Excitation Modelling Page 99
gure 5.4, the current basis functions of the feedlines have real (cos knz) and imaginary
(sin knz) parts. The real part causes current discontinuities in the direction of current
ow, because the triangle function as a rooftop function component is identically zero
at the interface between the port and the adjacent feedline, whereas cos(knz) has
a nite value. The problem is the same in the direction perpendicular to current
ow, because in this case the step function as a rooftop function component has
a nite value, whereas sin(knz) is identically zero. Therefore extra basis functions
are still required in each interface between the port and adjacent feedline for the
correct solution. These functions are introduced and referred to by the author as
compensation functions which are shown in gure 5.5. The compensation functions
are dened for the input port in the space domain by:
8
>
< 1 ; z;lzzi zi z zi + lz
Jz (z) = > (5.9)
: 0 otherwise
8
>
< z;zi zi z zi + lz
Jx(z) = > lz (5.10)
: 0 otherwise
5.2 Excitation Modelling Page 100
and for the output port;
8
>
< 1 + z;l zo zo ; lz z zo
Jz (z) = > z (5.11)
: 0 otherwise
8
>
< ; z;l zo zo ; lz z zo
Jx(z) = > z (5.12)
: 0 otherwise
Jz (z ) Jx(z)
(Zi+lz) (Zi+lz)
Zi Zi
Zo Zo
(Zo-lz) (Zo-lz)
(b) Output port
Figure 5.5: Compensation functions for the input and output ports
Jz (z) in equation 5.9 and gure 5.5(a) transfers the eect of the cosine portion of the
incident current wave in the direction of current
ow, whereas Jx(z) in equation 5.10
and gure 5.5(a) transfers the eect of the sine portion of the incident current wave
in the direction perpendicular to the current
ow. Similarly Jz (z) in equation 5.11
and gure 5.5(b) transfers the current wave to the output port in the direction of
5.2 Excitation Modelling Page 101
propagation whereas Jx(z) in equation 5.12 and gure 5.5(b) transfers the current
wave in the transverse direction.
where Jleft and Jright indicate the Fourier transform of the left and right hand side of
a rooftop function. The derivation of the compensation functions from the left and
right hand side of the rooftop function for the input port are given by;
The identical precalculated basis functions (gzin; gxin; gzo and gxo in equation 5.8) are
used for the components of the compensation functions in the transverse direction.
Including all current basis functions of the entire system, the total current is expressed
5.2 Excitation Modelling Page 102
as:
N
X
Jtotal = Jbox(kx; kz ) + (apnJportn (kx; knkz ) + acnJcompn (kx; kz ))
n=1
(5.19)
where N is the number of ports, Jbox refers to the basis functions of the actual open
circuit and Jportn is either the sum of the incident and re
ected current wave or the
transmitted current wave. To calculate the S{parameters of the circuit, apn must be
known. The Method of Moments is employed to eliminate the electric eld compon-
ents and to nd the unknown coecients. Weighting functions are then required: Jbox
and Jcompn are used. A total of N weighting functions must be dened to complete
the algorithm. A triangle function which straddles the lines separating the port and
the feedline, is used in the direction of current
ow and precalculated basis functions
are applied in the direction perpendicular to current
ow. After application of the
Method of Moments, the matrix equation yields,
Because two travelling waves, which are the incident current wave with unit amplitude
and the re
ected current wave with unknown amplitude, are used for the feedline con-
nected to the input port of the circuit, Z is not a square matrix and contains unknown
coecients, as well as the known unit amplitude. The column of the impedance mat-
rix corresponding to the incident current wave products has been moved to the right
hand side of the equation and the equation 5.20 becomes,
where Zn consists of the elements related to unknown current basis functions and
Zi is the column matrix containing the elements related to incident current waves.
Any root{nding procedure can be applied to solve equation 5.21 and the unknown
coecients can be found. The S{parameters of the N{port circuit are actually the
5.3 Comparison with other Excitation Modellings Page 103
coecients of the re
ected and the transmitted current waves.
It is emphasised that the asymptotic form of the Green's function given in detail
in section 4.3.1 is utilised to split the impedance matrix into frequency{dependent
and frequency{independent parts, but the current basis function for the feedline in
equation 5.1 is a function of wave number, which is a function of operating fre-
quency. Therefore the asymptotic form of the Green's function can only be applied
to the impedance matrix elements related to the products of current basis functions
of the circuit. However, numerical enhancements introduced in section 4.6 are used
to reduce the number of impedance matrix elements required and to speed up the
two{dimensional integration.
where apq is the relative weighting coecient of the qth component of the pth basis
function, lz is the grid size in z direction and a and b are the dimensions of the box.
The 4(lz ) term was derived from the width of the weighting function and the ab term
2
The matrix equation 2.9 is then solved for the set of current coecients and the em-
bedded Y{parameters are calculated. The S{parameters are derived by the standard
5.3 Comparison with other Excitation Modellings Page 104
Shield
Port 1 Port 2
S = (Y + I); (Y ; I)
1
(5.23)
It is not possible to apply this technique to the planar open structure, because no side
walls exist in the circuit, and so the tangential electric eld can not be assumed to
be identically zero. Moreover, the eects of port/feedline arrangement are dominant
to the resulting S{parameters, therefore Meade has also introduced a de{embedding
algorithm to minimise the eects of the port/feed arrangement [4]. The extraction
of the S{parameters from the a priori knowledge of the surface current distribution
is achieved by applying transmission line theory to the feedlines. The precalculated
line mode basis functions are used to model the current standing wave set up on the
feedlines. The de{embedding algorithm relies on the presence of a current standing
wave minimum on the input line. This can only be guaranteed when the line length
L g =2 so the frequency range of the technique has a lower limit which is a function
of the transmission line length and the guide wavelength. Thus, the de{embedding
algorithm is problem dependent and hence not valid for the wide frequency range of
interest.
5.4 Evaluation of Ecient Excitation Modelling Page 105
Jackson [2] has introduced a method to calculated S{parameters of gap{discontinuities
in 1985 and applied to irregular shaped microstrip structures [3] in 1989. The tech-
nique described here was initially inspired by his work, but his technique is not ecient
at relatively low frequencies, just like Meade's technique mentioned above. Because
the cosine portion of the travelling wave is believed to cause numerical diculties
and it is truncated one{quarter of a guide wavelength from a zero of the sine. The
length of the truncation is a function of the operating frequency. At low frequencies,
its length becomes larger than the entire circuit's dimension, and a large number of
extra rooftop functions are required otherwise the cosine portion of the current wave
re
ects before reaching the input port.
These are the limitations and drawbacks of existing excitation modellings. The tech-
nique introduced here does not have either lower limit or truncation, and therefore it
is valid over a wide microwave frequency band and does not require a large number
of rooftop functions at low frequencies.
The second step of the evaluation is to use a travelling current wave without trunca-
tion. The termination is still in the middle of the nite element grid, instead of at
port as shown in gure 5.7(b). In this case, extra rooftop functions are not required,
because there is no truncation of the current wave. But, at the port, the coecient
of the rooftop function must be identical to the sum of the incident and re
ected
current waves, because there is no discontinuity in the interface between the port and
feedline. However the shaded area is dened twice which causes numerical diculties
and a current discontinuity.
The solution to the problem outlined above is to shift the current wave back, so that
it terminates at the port of the circuit. In this case, as shown in gure 5.7(c) the
cosine portion of the current wave has a nite value, whereas the rooftop function in
the direction of current
ow is identically zero. Therefore, the cosine portion of the
incident current wave re
ects from the input port, because the current at the input
port must be identically zero (shaded area in gure 5.7(c)). So this is not the solution
to the problem.
The nal step of the evaluation and ecient excitation mechanism is shown in g-
ure 5.7(d). There is neither truncation of the cosine portion of current wave nor
shifting the current wave to terminate in the middle of the nite element grid. In
order to transmit the eect of the cosine portion of the current wave in the direction
of current
ow and the eect of the sine portion of the current wave in the direction
5.4 Evaluation of Ecient Excitation Modelling Page 107
sin kn z cos kn z
undened region
(a)
Microwave Circuit Sub{domain Basis Functions (rooftops)
sin kn z cos kn z
sin kn z cos kn z
Current Discontinuity
(b)
Sub{domain Basis Functions (rooftops)
Microwave Circuit sin kn z cos kn z
sin kn z cos kn z
Current Discontinuity
(c)
Sub{domain Basis Functions (rooftops)
Microwave Circuit sin kn z cos kn z
sin kn z cos kn z
Compensation functions
(d)
Figure 5.7: Evaluation of ecient excitation modelling
5.5 Numerical Results Page 108
perpendicular to current
ow, the compensation functions are used at the interface
between the feedline and port of the circuit.
The microstrip line in gure 5.8 is of length 2.8 mm, of width 0.7 mm on the substrate
thickness 0.381 mm, relative permittivity 11.7 and relative permeability 1. The di-
mensions of the circuit have been intentionally chosen by the author to be relatively
small, in order to highlight the deciency of the Jackson's technique and to illustrate
the improvement in the accuracy.
Four sets of S{parameter results are plotted in gure 5.9. These are the results
calculated by using 42 rooftop functions in total for the denition (lx = w=4 = 0:157
mm and lz = L=8 = 0:35 mm). First the test structure is analysed by using truncated
travelling current waves for excitation. The minimum operating frequency in the
analysis is 1 GHz and the corresponding length of the truncated cosine portion is
26.25 mm, which is 75 times bigger than the overlapping rooftop size, which is 0.35
mm in the direction of propagation. For the maximum operating frequency of 20
5.5 Numerical Results Page 109
Port 1 Port 2
w=0.7mm
d=0.381mm
L=2.8mm
r = 11:7; r = 1
GHz, the corresponding length of the truncated portion is 1.25 mm, which is still
more than 3 times bigger than the rooftop size. Therefore the cosine portion of the
incident current wave never reaches to the input port of the circuit in this frequency
range and the circuit behaves like an open{end discontinuity as shown in gure 5.9.
To get the same accuracy by Jackson's technique for this frequency range, up to
888 extra rooftop functions are required. So Jackson's technique is not ecient for
this type of small circuit in this frequency range. Even though the second and third
steps of the evaluation of the excitation give slight improvement without requiring
extra rooftops, they are still not suitable excitation modelling techniques for the open
structures over the entire microwave region.
In the end, the introduced excitation modelling is used for the analysis of the test
structure, and the improvement in the accuracy is highlighted in gure 5.9. In this
model, the eects of excitation are transferred into the circuit without any loss. The
S is almost zero, whereas S is negative large value as they have been predicted.
21 11
-10
-30
-40
Truncated
Step 2
-50 Step 3
This technique
2 4 6 8 10 12 14 16 18
Frequency GHz
-5
-15
Insertion Loss (dB)
-25
-35
Truncated
Step 2
-45 Step 3
This technique
2 4 6 8 10 12 14 16 18
Frequency GHz
Figure 5.9: Magnitude of S{parameters for the open microstrip line illustrating the
evaluation of the excitation modelling
5.5 Numerical Results Page 111
-10
-20
-40
improvement in the accuracy, because in the literature, results are available from
measurements performed by Shibata et al [5], from FDTD performed by Paul et al [6]
and from SDM shielded performed by Railton et al [7]. The dimensions of the circuit
are given in section 3.8.3.2.
As seen in gure 5.10, there is a clear agreement between the introduced excitation
modelling and the published results for SDM shielded and FDTD as well as with meas-
ured data. Furthermore, problems caused by box mode broadband using the method
of [7] have been eliminated. The use of travelling current waves with truncation yields
completely meaningless results, this is because the minimum operation frequency is 1
GHz with the corresponding length of the truncated cosine portion being 29.14 mm.
This is almost 46 times bigger than the size of the rooftop in the direction of the
propagation for the denition (lx = w=4 = 0:318 mm and lz = l=20 = 0:636 mm).
The maximum operating frequency, which is 15 GHz, is still not sucient to cover
the missing portion without extra rooftops. For the maximum operation frequency
5.6 Summary Page 112
the length of the truncated portion is 1.78 mm and still almost 3 times bigger than
the rooftop size. Again for the same accuracy by Jackson's technique requires up to
540 extra rooftops (270 x and 270 z components) to complete the missing portions
It must be noted that, the Jackson technique was used as described in his paper [2,3]
and shown to be accurate. However his technique is accurate only if the missing
portion is completed with a large number of rooftops. At low frequencies, a large
number of rooftop functions are required to complete the missing portion.
5.6 Summary
In this chapter, an ecient excitation algorithm to extract the circuit S{parameters
from the known surface current distribution has been discussed. It has been shown
that the introduced excitation modelling is eective in remedying all deciencies of
available excitation modellings in the literature. The accuracy of the technique was
compared with published results with the aid of numerical examples of a microstrip
line and the edge{coupled lter. It is emphasised that, the number of extra rooftops
required to complete the missing portion are not \several" as quoted in [3], but a huge
number of rooftop functions are required to yield accurate solution at low frequencies.
Chapter 5: References
References
[1] S. A. Meade, The Rapid and Rigorous Mathematical Modelling of Passive Planar
Microwave Circuits. PhD thesis, University of Bristol, March 1994.
[2] R. W. Jackson and D. M. Pozar, \Full{wave analysis of microstrip open{end
and gap discontinuities," IEEE Transaction on Microwave Theory and Technique,
vol. 33, pp. 1036{1042, October 1985.
[3] R. W. Jackson, \Full{wave, nite element analysis of irregular microstrip dis-
continuities," IEEE Transaction on Microwave Theory and Technique, vol. 37,
pp. 81{89, January 1989.
[4] S. A. Meade and C. J. Railton, \Ecient extraction of s{parameters from the
spectral domain method for passive planar microwave circuits," in 24 th. European
Microwave Conference, pp. 1084{1089, September 1994.
[5] T. Shibata, T. Hayashi, and T. Kimura, \Analysis of microstrip circuits using
three{dimensional full{wave electromagnetic eld analysis in the time domain,"
IEEE Transaction on Microwave Theory and Technique, vol. 36, pp. 1064{1070,
June 1988.
[6] D. L. Paul, E. M. Daniel, and C. J. Railton, \Fast nite dierence time domain
method for the analysis of planar microstrip circuits," in 21st European Microwave
Conference, pp. 303{308, September 1991.
[7] C. J. Railton and S. A. Meade, \Fast rigorous analysis of shielded planar lters,"
IEEE Transaction on Microwave Theory and Technique, vol. 40, pp. 978{985,
May 1992.
Chapter 6
In this chapter, the total improvement in the complex circuit analysis will be dis-
cussed. To do this, the improvement gained by each enhancement will be given in
percentage, instead of run{time requirement on the computer. Because the run{time
is computer dependent and the speed of the analysis is a function of CPU and memory
of the computer as well as computer load.
4mm
0.636mm
2.54
5mm
1.272mm 11mm
r = 10; r = 1
1 2 3 4 5 6 7
Figure 6.2: Plan evaluation of the microstrip step discontinuity shoving regions
The set of basis functions for the model in gure 6.2 is summarised below:
The total number of current basis functions required to model the unknown current
on the metallisation of the circuit is 42 (21 x and 21 z components) in contrast to
162 (81 x and 81 z components) as given in detail in section 3.8.4. The current basis
functions required for accurate modelling are reduced by 74%. A truncated current
wave and an extra set of rooftop functions to complete this truncated portion as used
by Jackson [2] are employed to excite the open circuit of interest. The minimum
operating frequency is 1 GHz and the corresponding length of the truncated portion
is 27.7 mm, which is 55 times bigger than the rooftop function size. For the maximum
operating frequency of 20 GHz, the corresponding length of the truncation is 1.25 mm,
which is still 2.5 times bigger than the rooftop size. To complete the missing portion
in order to get accurate results, up to 1512 extra rooftop functions (756 x and 756 z
components) are required. In this thesis, however a novel compensation function has
been introduced and the eect of excitation has been successfully transferred into the
circuit of interest by using two for each port. Therefore, 4 compensation functions
(2 x and 2 z components) for 2{port circuit are enough in contrast to 1512 extra
rooftops and the saving is 99.7% at 1 GHz.
The adaptive integration technique given in detail in section 4.2 is used to calculate
the impedance matrix elements. A coarse integration step which is 4 times bigger
than the ne integration step is chosen. With this choice, the speed of the calcu-
lation is increased 4 times compared with the use of uniform integration steps. In
addition, in section 4.6 the feature of the current basis function which is to be two
6.3 Edge{Coupled Filter Page 118
separable functions in each direction is exploited. 903 impedance matrix elements
must be integrated in each direction if the enhancement in section 4.6 is not used. 18
basis functions (9 x and 9 z components) are used in x direction and 903 impedance
matrix elements can be derived from 171 sub{impedance matrix elements. Therefore
the impedance matrix elements needed to be integrated are reduced by 81% in one
dimension. The improvement by using the symmetry of the Green's function also
gives 10% improvement. The S{parameter results were plotted in gure 3.20.
1 2
4 5
Figure 6.3: Plan evaluation of the microstrip edge{coupled lter shoving regions
The edge{coupled lter has been divided into 5 regions as shown in gure 6.3. The
set of basis functions for the model in gure 6.3 is summarised below:
6.3 Edge{Coupled Filter Page 119
Regions 1 and 5 :
2 precalculated current waves as dened by equations 3.13 and 3.14
Regions 2 and 4 :
18 precalculated transverse basis functions in the direction perpendicular to
current
ow and rooftop functions in the direction of propagation as given in
detail in section 3.8.2
Region 3 :
1 precalculated transverse basis function in the direction perpendicular to cur-
rent
ow and resonance mode current basis function in the direction of propaga-
tion as given in detail in section 3.8.3.2 and shown in gure 3.17
Ports :
The sum of an incident and re
ected current wave for the input port and a
transmitted current wave for the output port as dened by equations 5.2 and 5.3
Transfer :
{ 2 compensation functions in the interfaces between ports and feedlines as
dened by equations 5.15, 5.16, 5.17 and 5.18
{ 2 rooftop functions to join two regions (lx = 0:318 mm and lz = 0:636 mm)
The total number of current basis functions required for this metallisation is 46 in
total in contrast to 342. The current basis functions required for accurate modelling
are reduced by 86%. The truncated current wave as given in detail in section 5.3
and an extra set of rooftop functions to complete this truncated portion are used to
excite the edge{coupled lter. The minimum operating frequency is 1 GHz and the
corresponding length of the truncated portion is 29.14 mm, which is almost 46 times
bigger than the rooftop size. For the maximum operating frequency of 15 GHz, the
corresponding length of truncation is 1.78 mm, which is still almost 3 times bigger
than the rooftop size. To complete the missing portion in order to get accurate results,
540 extra rooftop functions are required at 1 GHz. In this model, the compensation
6.4 Simple Low{Pass Filter Page 120
functions have been used and the eect of excitation has been successfully transferred
into the circuit.
The adaptive integration technique has been employed to calculate the impedance
matrix elements. A coarse integration step is 4 times longer than the ne integration
step so, the speed of the calculation is increased 4 times compared with the use of
uniform integration step. 6 basis functions are used in x direction and 1081 impedance
matrix elements can be derived from 21 sub{impedance matrix elements. Therefore
the impedance matrix elements needed to be integrated are reduced by 98% in one
dimension. The improvement by using the symmetry of the Green's function also
gives 10% improvement. The S{parameter results were shown in gures 3.18 and 5.10.
The run{time on HP700 workstations is around 1 minute to calculate S{parameters
at each spot frequency. This is slightly bigger than the calculation time reported
in [4] for the shielded edge{coupled lter. This is because of having more current
basis functions and of excitation problem. The current wave in a length of a number
of half wavelength which is a function of operating frequency. This length is bigger
at low frequency and its Fourier transform is ner. Therefore ne integration steps
are required to model.
The low{pass lter has been divided into 7 regions as shown in gure 6.5. The set of
basis functions for the model in gure 6.5 is summarised below:
6.4 Simple Low{Pass Filter Page 121
43.561mm
2.413mm
mm
mm
20.32
2.54
2.413mm
m
5.65m
0.794mm
r = 2:2; r = 1
Regions 1 and 7 :
2 precalculated current waves as dened by equations 3.13 and 3.14
Regions 2 and 6 :
18 rooftop current basis functions (lx = 0:60325 mm and lz = 0:60325 mm)
Region 3 :
57 rooftop current basis functions (lx = 0:635 mm and lz = 0:635 mm)
Regions 4 and 5 :
12 rooftop current basis functions (lx = 0:635 mm and lz = 1:27 mm)
Ports :
The sum of an incident and re
ected current wave for the input port and a
transmitted current wave for the output port as dened by equations 5.2 and 5.3
Transfer :
{ 2 compensation functions in the interfaces between ports and feedlines as
dened by equations 5.15, 5.16, 5.17 and 5.18
6.4 Simple Low{Pass Filter Page 122
3.81mm
4
1 2
18.0975mm 3
6 7
Figure 6.5: Plan evaluation of the microstrip low{pass lter shoving regions
The total number of current basis functions required for this analysis is 214 (107
x and 107 z components) in total in contrast to 594. The current basis functions
required for accurate modelling are reduced by 64%. The truncated current wave
as given in detail in section 5.3 and an extra set of rooftop functions to complete
this truncated portion are used to excite the open circuit of interest. The minimum
operating frequency is 2 GHz and the corresponding length of the truncated portion is
27.42 mm, which is 45 times bigger than the rooftop size. For the maximum operating
frequency of 20 GHz, the corresponding length of truncation is 2.6925 mm, which is
still 4 times bigger than the rooftop size. To complete the missing portion in order
to get accurate results, 528 extra rooftop functions are required to analyse at 1 GHz.
With the use of the compensation functions, 4 functions are enough instead of using
the truncated current wave and 528 extra rooftop functions.
The adaptive integration given in detail in section 4.2 is also used to calculate the
impedance matrix elements. A coarse integration step is 4 times longer than the
ne integration step. With this choice, the speed of the calculation is increased 4
6.5 Summary Page 123
times compared with the use of uniform integration steps. 66 basis functions are
used in x direction and 23005 impedance matrix elements can be derived from 2211
sub{impedance matrix elements. Therefore the impedance matrix elements needed
to be integrated are reduced by 90.4% in one dimension. The improvement by using
the symmetry of the Green's function also gives 5% improvement. The S{parameter
results are plotted in gures 6.6. As seen in gure 6.6, the calculated results and
measurements are in very good agreement.
6.5 Summary
This chapter has presented results for the SDM implementation compared to calcu-
lated and measured data available in the literature for open planar passive microwave
circuits to illustrate the total improvement in the analysis. The eciency of the
method is highlighted by computation time improvement in percentage for spot fre-
quency S{parameters.
6.5 Summary Page 124
-5
-10
Reflection Loss (dB)
-15
-20
-25
-30
-35 Calculated
Measured
-40
-45
2 4 6 8 10 12 14 16 18 20
Frequency GHz
-5
-10
Insetion Loss (dB)
-15
-20
-25
-30 Calculated
Measured
-35
-40
2 4 6 8 10 12 14 16 18 20
Frequency GHz
References
[1] R. W. Jackson and D. M. Pozar, \Full{wave analysis of microstrip open{end
and gap discontinuities," IEEE Transaction on Microwave Theory and Technique,
vol. 33, pp. 1036{1042, October 1985.
[2] R. W. Jackson, \Full{wave, nite element analysis of irregular microstrip dis-
continuities," IEEE Transaction on Microwave Theory and Technique, vol. 37,
pp. 81{89, January 1989.
[3] H. H. Balik and C. J. Railton, \Sub{gridding in the spectral domain method for
the analysis of planar circuits and antennas," in 3rd International conference on
telecommunications in modern satellite, cable and broadcasting services, pp. 592{
595, October 1997.
[4] C. J. Railton and S. A. Meade, \Fast rigorous analysis of shielded planar lters,"
IEEE Transaction on Microwave Theory and Technique, vol. 40, pp. 978{985,
May 1992.
[5] D. M. Sheen, S. M. Ali, M. D. Abdouzahra, and J. A. Kong, \Application of
the three{dimensional nite-dierence time{domain method of the analysis of
planar microstrip circuits," IEEE Transaction on Microwave Theory and Tech-
nique, vol. 38, pp. 849{857, July 1990.
Chapter 7
Metallisation
o ; o
Port 1 Port 2
r2 ; r2
r3 ; r3
r(N ;1) ; r(N ;1)
rN ; rN
Figure 7.1: General 3{D multilayered microwave circuit
The Green's function in the spectral domain is commonly derived for the multilayer
circuit of interest by using one of two dierent formulation precesses. First, the
Green function can be directly derived by solving Maxwell's equations in the spectral
domain, with suitable boundary conditions [1{3], but extending this procedure for
multiple dielectric layers becomes too complicated [2]. In [4], a generalised spectral
7.2 Derivation of Dyadic Green's Function Page 128
domain Green's function is derived in terms of suitable components of vector electric
and magnetic potentials. With these vector potentials, the boundary conditions were
simplied into equivalent transmission line problems in the spectral domain, as in [5].
The solutions were then simplied to solving a standard form containing only two
layers, by the use of iterative solutions. The eect of other layers is taken into account
via the re
ected and transmitted coecients, which are functions of the parameters
of the other layers (thickness, dielectric permittivity and permeability). Despite these
simplications, the technique is found by the author to be still complicated and not
favourable to this implementation.
Itoh [5] presented a simple method, called the immitance approach, for deriving the
dyadic Green's functions for a microstrip line and applied this to the analysis of a
microstrip resonator [6] in 1981. Itoh used the transverse equivalent circuit concept
in the spectral domain, in conjunction with a simple coordinate transformation rule.
The author has chosen Itoh's immitance approach to derive Green's function for
multilayer structures and the formulation process is given in detail in this chapter.
The asymptotic form of the Green's function for the multilayer structure is derived
for the large value of Fourier transform variables by using the immitance approach.
Numerical results are plotted for multilayer structures to verify the accuracy of the
derivation and the eect of multilayering with the aid of a microstrip line and a
microstrip resonator.
y x
Q(x,y,0) P(x,y,z)
o ; o r
z
d2 r2 ; r2
d3 r3 ; r3
dN ; 1 r(N ;1) ; r(N ;1)
dN rN ; rN
Figure 7.2: Cross{section of a multilayered microstrip resonator
The basic concept of the immitance approach can be understood if the inverse trans-
form of the Fourier transform in equation 7.1 for the eld is examined.
Z1Z1
(x; z) = (21) (kx; kz )e;j kxx ( kz z) dk
+
xdkz (7.1)
2
;1 ;1
From this expression, all eld components are a superposition over kx and kz of
inhomogeneous (in y) plane waves
are propagating in the direction of from
which
the z{axis , where = cos; pkk2z k2 . For each , waves may be decomposed into
1
x z +
TM{to{y (Ey ; Ev ; Hu), and TE{to{y (Hy ; Hv ; Eu) where the coordinates v and u are
7.2 Derivation of Dyadic Green's Function Page 130
x
kx u cos = pkk2z k2
x+ z
y
kz sin = pkk2x k2
z
x+ z
v
Figure 7.3: Co{ordinate transformation
The current Jv creates only the TM elds, because it is concerned with Hu and
likewise Ju creates the TE elds. Therefore, an equivalent circuit for the TM and
TE elds can be drawn, as shown in gure 7.4. The characteristic impedance in each
region is given by;
Ev =
i
ZTMi = H (7.4)
u j!i
E
ZTEi = ; Hu = j! i
(7.5)
q v i
i = kx + kz ; ki 2 2 2
(7.6)
where i = 1::N and ki = ! ii . The
i is the propagation constant in the y direction
2 2
in the ith region. All boundary conditions for the TM and TE waves are incorporated
in the equivalent circuits. The electric elds Ev and Eu are continuous at y = 0 and
7.2 Derivation of Dyadic Green's Function Page 131
given by;
Ze
1
y=0 Zh
1
Ze
2 Jv Ju Zh
2
Ze 3 Zh3
ZhN ;
( 1)
ZhN ;
( 1)
ZeN ZhN
y = ; PNn dn=1
Ze and Zh are the input impedances looking into the equivalent circuits at y = 0 and
are given by;
; !
Ze (kx; kz ; d) = Z1e + Z1e
1
(7.9)
!;
1 2
Z (k ; k ; d) = 1 + 1 1
(7.10)
h x z
Zh Zh
1 2
where Ze and Ze are input impedances looking into the corresponding regions at
1 2
Ze = ZTM1
1 Zh = ZTE1
1 (7.11)
7.2 Derivation of Dyadic Green's Function Page 132
As seen in gure 7.4, a transmission line is terminated by another transmission line of
dierent characteristic impedance. Therefore conventional transmission line theory
can be used to nd the Ze and Zh as
2 2
TE2 coth
d + Zh
2
2 2 3
where
ZTMN
ZeN = coth ZTEN
ZhN = coth
d N N
d N N
The nal part of the immitance approach formulation is to map from the (u; v) to
(x; z) co{ordinate system for the spectral wave corresponding to each given by kx
and kz . Because of the coordinate transform in equations 7.2 and 7.3, Ex and Ez are
linear combination of Eu and Ev . Similarly Jx and Jz are superpositions of Ju and
Jv . When the above notations are applied, the Green's function elements are found
to be:
Gzz = Nz Ze + Nx Zh
2 2
(7.13)
Gzx = NxNz (;Ze + Zh ) (7.14)
Gxz = Gzx (7.15)
Gxx = Nx Ze + Nz Zh
2 2
(7.16)
7.3 Asymptotic Form of Green's function Page 133
where Nx and Nz are transforming ratios given by;
Nx = q kx Nz = q kz (7.17)
kx + kz
2 2
kx + kz
2 2
Note that Ze and Zh are functions of kx + kz and the ratio of kx and kz enters through
2 2
Nx and Nz .
So, for the large Fourier transform variables, all multilayer structures are equivalent
to the structure containing the immediate two layers, extended to innity, on each
side of the strip. The reason for this is that large values of transform variables account
for the reactive eld of the source, which is a localised eect.
With the above argument, the asymptotic form of the Green's function of multilayered
structures can be directly obtained from the asymptotic form of the Green's function
7.3 Asymptotic Form of Green's function Page 134
for two layer structure, which is given in detail in section 4.3.1 as;
where
h = ; ! (1j+ )
1 h = (1j!
2
r
+ r2 )
0
r2
0
K1zz = q kz K1zz = q kz
2 2
1
kx + kz
2 2
2
(kx + kz )
2 2 3
Kxz = Kzx
1 1 Kxz = Kzx
2 2
K1xx = q kx K1xx = q kx
2 2
1
kx + kz
2 2
2
(kx + kz )
2 2 3
with ! = 2f and r ; r the relative permittivity and permeability of the electric
2 2
substrate next to the strip respectively. It must be noted that the rst layer is free{
space because of the open structure.
The functions K1 sti (s; t = z or x; i = 1; 2) are just functions of kx and kz and thus
are independent of frequency and metallisation pattern. The constants hi (i = 1; 2)
need only be calculated once per frequency, as they are independent of kx and kz .
An important point to be considered for the spectral domain integration of the im-
pedance matrix elements is the treatment of the surface wave poles. It can be easily
proven that all the poles of the Green's function lie between ko and prmax rmaxko.
The same technique given in section 4.2 is used for the impedance matrix elements
integration.
7.4 Numerical Examples Page 135
7.4 Numerical Examples
To verify the accuracy of the Green's function which has been derived by the use
of the immitance approach, the simple two layer microstrip line of width 1.27 mm
on a substrate of thickness 1.27 mm and of permittivity 8.875 is taken as a test
structure. The microstrip line is analysed for the eective permittivity by the existing
two dimensional version of the technique, then the substrate is divided into three,
assuming that the structure is now a four layer microstrip line with identical substrate
parameters. The eective permittivities are plotted in gure 7.5 and compared with
published data [7].
Eective Permittivity
+
+
7
+
+
+
+
+ two layer
four layer
+ [7] +
+
6
+
5 10 15 20
Frequency (GHz)
To further test the accuracy of the Green's function which has been derived by the
immitance approach, results are compared with the simple closed{form expression
for the eective dielectric constant of a microstrip on a two layer substrate [8]. The
geometry of the two substrate layer microstrip used in the present test is shown in
gure 7.6.
7.4 Numerical Examples Page 136
w
r 1 d 1
r 2 d 2
Following [8], the total height of the substrate layers, dtot = d + d , is held constant
1 2
the eective dielectric permittivities of the single layer cases, reff (dr = 0) and
reff (dr = 1)
Figure 7.7 shows the eective dielectric permittivity as a function of the height ratio
for dierent dielectric constants of the lower substrate. The results are compared
with published data [8]. In gure 7.7, a linear normalisation is used, described by;
reff norm (dr ) = reff (dr ) ; reff (0) ; dr [reff (1) ; reff (0)]
( ) (7.22)
where reff (0) and reff (1) are the eective permittivity for the single substrate cases
given by;
As shown in gures 7.5 and 7.7, the immitance approach is shown to be an accurate
7.4 Numerical Examples Page 137
Normalized reff
This technique
r = 10 [8]
0.5 r = 8 [8]
r = 4 [8]
0.25
0
-0.25
-0.5
-0.75
-1
0 0.2 0.4 0.6 0.8 1
Substrate height ratio d2 =dtot
Figure 7.7: reff of a microstrip on a two layer substrate normalised using equa-
tion 7.22 as a function of the substrate height ratio for dierent values of r 2
(w=dtot = 1 and r = 6)
1
To observe the eects of the order of dielectric substrates with dierent parameters,
the microstrip line of width 1 mm is modelled and a series of results are plotted in
gure 7.8. In the analysis, each layer thickness has been chosen to be 0.25 mm.
As shown in gure 7.8, the order of the dielectric layers is very eective. At low
frequencies the value of the eective permittivity is determined by the relative per-
mittivity of the substrate next to the strip. It is clearly shown in gure 7.8 that the
parameters of substrate next to the ground plane is eective on the increment of the
eective permittivity from one frequency point to another.
Further illustrate the eects of multilayering, a simple three layer microstrip resonator
of width 1 mm and of length 8 mm as shown in gure 7.9 is modelled and the table 7.1
has been prepared. The calculation of the resonant frequency is given in [1, 6]. As
7.4 Numerical Examples Page 138
Eective Permittivity
11
r r r
1 2 3
10 20 10 5
10 5 20
5 20 10
9 5 10 20
7
6
5
10 20 30 40 50
Frequency (GHz)
shown in table 7.1 that, the order of the dielectric layers is dominant on the resonant
frequency. The resonant frequency is inversely proportional to the value of parameters
of dielectric substrate next to the conducting strip.
r 1
d =0.636mm1
r 2 20 10 5
20 3.05+j1.7010; 3.93+j6.7110; 5.13+j1.8510;
3 3 2
d =0.636mm
w = 1mm
L = 8mm
d 1
r ; r
1 1
d 2 r ; r
2 2
7.5 Summary
This chapter discusses the derivation of the Green's function for the multi dielectric
layer structures. To verify the derivation, the simple microstrip line which has been
previously analysed in the literature is modelled by the two dimensional version of the
technique. The eects of multilayering on the eective permittivity of a microstrip
line and the resonant frequency of a microstrip resonator have been illustrated by
modelling corresponding structure.
Chapter 7: References
References
[1] T. Itoh, \Analysis of microstrip resonators," IEEE Transaction on Microwave
Theory and Technique, vol. 22, pp. 946{952, November 1974.
[2] J. B. Davies and D. Mirshekar-Syahkal, \Spectral domain solution of arbitrary
coplanar transmission line with multilayer substrate," IEEE Transaction on Mi-
crowave Theory and Technique, vol. 25, pp. 143{146, February 1977.
[3] D. M. Pozar, \General radiations for a phased array of printed antennas derived
from innite current sheets," IEEE Transaction on Antennas and Propagation,
vol. 33, pp. 498{504, May 1985.
[4] N. K. Das and D. M. Pozar, \A generalised spectral{domain Green's function for
multilayer dielectric substrates with application to multilayer transmission lines,"
IEEE Transaction on Microwave Theory and Technique, vol. 35, pp. 326{335,
March 1987.
[5] T. Itoh, \Spectral domain immitance approach for dispersion characteristics of
generalized printed transmission lines," IEEE Transaction on Microwave Theory
and Technique, vol. 28, pp. 733{736, July 1980.
[6] T. Itoh and W. Menzel, \A full{wave analysis method for open microstrip struc-
tures," IEEE Transactions on Antennas and Propagation, vol. 29, pp. 63{67,
January 1981.
[7] T. Itoh and R. Mittra, \Spectral{domain approach for calculating the dispersion
characteristics of microstrip lines," IEEE Transaction on Microwave Theory and
Technique, pp. 496{499, July 1973.
[8] J. P. K. Gilb and C. A. Balanis, \Closed{form expressions for the design of mi-
crostrip lines with two substrate layers," in IEEE MTT{S INT. Microwave Symp.,
pp. 1005{1008, 1993.
Chapter 8
The lower frequency bands can provide only limited amount of spectrum for radio{
borne services such as cellular telephones and wireless local area networks and are
already heavily used. As the operating frequency is moved upward to less used fre-
quency bands with more available bandwidth, the approximations, which simplify the
complicated equations describing the operation of microwave circuits, are no longer
valid. Therefore it is necessary to use full{wave analysis in order to full present
demands.
Currently several methods are reported in the literature applied to planar transmis-
sion line problems and discontinuity analysis, as discussed in chapter 2. The Spectral
Domain Method (SDM) which is based on solving the coupled integral equation in
the spectral domain has been chosen by the author to meet these requirements. The
SDM is very popular but has been criticised for being computationally intensive, due
to the need for the analytical preprocessing. This feature imposes certain restrictions
on the applicability of the technique, but allows the development of an interactive
design tool. The general formulation of SDM has been given in detail in section 2.3.3.
The choice of basis functions, which dene the unknown current distribution on the
metallisation of the circuit, is crucial to the eciency of the technique. The current
basis functions must be chosen to approximate the unknown current distribution
as closely as possible, otherwise a large number of basis functions are required for
convergence. Chapter 3 discussed the class of current basis functions used in this
thesis. The implementation of an arbitrary metallisation pattern requires sub{domain
basis functions which are the rooftop functions described in section 3.4. As has been
discussed, the rooftop function diers from that commonly implemented. The use of
8.1 Summary Page 143
the rooftop function has been improved by introducing sub{gridding in the spectral
domain. The sub{gridding is used in the sense that the rooftop functions are dened
as functions of their locations.
The precalculated basis function which was introduced by Railton and Meade [1, 2]
has been dened either as a linear combination of rooftop functions with precalculated
coecients, or as a current wave function with a precalculated propagation constant.
It has been noted in chapter 3 that there is no restriction of geometry and the size of
the rooftop function which is necessary in [1,2] to exploit the benets of using FFT.
The SDM method is a frequency domain technique and requires the repeated im-
pedance matrix calculation at each frequency point of interest. This is in contrast
to methods such as the FDTD and TLM, which give the time domain response of
the circuit, which can then be transformed to a wide band frequency response for
a single run of the algorithm. In chapter 4, the enhancements in order to reduce
the number of impedance matrix elements calculated at each spot frequency and in
order to speed up the two{dimensional continuous integration have been introduced.
The adaptive integration technique, which consists of an adaptive truncation of the
integration range and an adaptive integration step has been introduced to limit the
two{dimensional continuous integration over an innite surface to nite computer
resources.
Throughout this thesis, the circuits have been represented by N{port network para-
meters, i.e. S{parameters. Chapter 5 has presented an improved form of excita-
tion modelling which remedies the problems in the currently available methods, and
provides a more ecient implementation. The current wave function without any
form of truncation has been used to excite the open microwave circuit of interest.
The compensation function has been introduced in section 5.2, to be used at the
interface between the port and the feedline, in order to transmit the full eect of
excitation.
In chapter 7, the derivation of the Green's function for multi dielectric layer structures
has been given in detail and its asymptotic forms have been derived. The loss in the
dielectric substrate has been included into analysis by replacing r by r (1 ; j tan )
where tan is the loss tangent of the dielectric substrate.
The scope for future work to further increase the versatility of the SDM enhanced
in this thesis includes the following: inclusion of conductor loss and nite conductor
thickness, plus the capability to handle multiple conductor layers and 3{D metallisa-
tion.
8.3 Concluding Remarks Page 145
In this implementation, the conductor has been assumed to be perfect and of inn-
itesimal thickness. Most of the applications meet above requirements. The inclusion
of conductor loss and nite conductor thickness have been investigated by other re-
searchers [3{8]. The 3{D metallisation structures have been analysed by Becks and
Wol [9] in the spectral domain. The technique to analyse circuits with multiple
conductor layers has also been introduced in [10]. However the enhancements in this
thesis have reduced the computational overheads so that interactive and the present
implementation can be widened in order to include the above applications.
References
[1] C. J. Railton and S. A. Meade, \Fast rigorous analysis of shielded planar lters,"
IEEE Transaction on Microwave Theory and Technique, vol. 40, pp. 978{985,
May 1992.
[2] S. A. Meade and C. J. Railton, \Ecient implementation of the spectral domain
method including precalculated corner basis functions," IEEE Transaction on
Microwave Theory and Technique, vol. 42, pp. 1678{1684, September 1994.
[3] W. Heinrich, \Full{wave analysis of conductor losses on MMIC transmission
lines," IEEE Transaction on Microwave Theory and Technique, vol. 38, pp. 1468{
1472, October 1990.
[4] C. Shih, R. Wu, S. Jeng, and C. H. Chen, \Frequency{dependent chacterist-
ics of open microstrip lines with nite strip thickness," IEEE Transaction on
Microwave Theory and Technique, vol. 37, pp. 793{795, April 1989.
[5] K. M. Rahman and C. Nguyen, \Full{wave analysis of coplanar strips considering
the nite strip metalisation thickness," IEEE Transaction on Microwave Theory
and Technique, vol. 42, pp. 2177{2179, November 1994.
[6] Z. Ma, E. Yamashita, and R. Mittra, \Hybrid{mode analysis of planar trans-
mission lines with arbitrary metalisation cross section," IEEE Transaction on
Microwave Theory and Technique, vol. 41, pp. 491{497, March 1993.
[7] J. Kuo and T. Itoh, \Hybrid{mode computation of propagation and attenuation
characteristic of parallel coupled microstrips with nite metalisation thickness,"
IEEE Transaction on Microwave Theory and Technique, vol. 45, pp. 274{280,
February 1997.
[8] G. B. Stracca, \A simple evaluation of losses in thin microstrips," IEEE Transac-
tion on Microwave Theory and Technique, vol. 45, pp. 281{283, February 1997.
[9] T. Becks and I. Wol, \Analysis of 3{D metallisation structures by a full{wave
spectral domain technique," IEEE Transaction on Microwave Theory and Tech-
nique, vol. 40, pp. 2219{2227, December 1992.
Chapter 8: References Page 147
[10] M.-J. Tsai, D. Flaviis, O. Fordham, and N. G. Alexopoulos, \Modelling planar
arbitrary shaped microstrip elements in multilayered media," IEEE Transaction
on Microwave Theory and Technique, vol. 45, pp. 330{337, March 1997.
Appendix A
where Jxn(x; z) and Jzn(x; z) are the nth rooftop functions for current
ow in x and z
directions respectively. The case for the x{directed current is considered rst. Taking
Fourier transforms in the x and z direction with the transform
Z1Z1
Jxn(kx; kz ) = ;1 ;1
Jxn(x; z)ej kxx
( + kz z) dxdz (A.3)
Since the rooftop functions are dened as two separable functions, equation A.3 can
be split onto two separate integrals of the form
Z1 Z1
Jxn(kx; kz ) = Jxn(kx)Jxn(kz ) = ;1
Jxn(x)ejkxxdx
;1
Jxn(z)ejkz z dz
(A.4)
Considering the rst integral in equation A.4, the triangle function Jxn(x) in g-
ure A.1(a) is only dened over the interval xn ; lx to xn + lx equal to a triangle
function and is otherwise zero, therefore can be written as,
Z xn +lx
Jxn(kx) = 1 ; jx ;l xnj ejkxxdx
xn ;lx x (A.5)
= k 2l (1 ; cos(kxlx))ejkxxn
xx
2
A.1 Fourier Transform of Rooftop Function Page 150
Jxn(x) Jxn(z)
1 1
x z
xn ; lx xn xn + lx zn + lz zn zn + lz
(a) Triangle function Jxn(x) (a) Step function Jxn(z)
Figure A.1: Components of a rooftop function (x{directed rooftop)
Considering the second integral in equation A.4, the step function Jxn(z) in g-
ure A.1(b) is only dened over the interval zn ; lz to zn + lz equal to unity and
is otherwise zero.
Z zn +lz
Jxn(kz ) = zn ;lz
1ejkz z dz
(A.6)
= k2 sin(kz lz )ejkz zn
z
Combining the two transforms with equation A.4, the two{dimensional Fourier trans-
form of the x{directed rooftop basis function Jxn(kx; kz ) is
where L is the length of the microstrip line. The case for the z{directed current is
considered rst. Taking a Fourier transform in z direction with the transform
Z1
Jz (kz ) = ;1
Jz (z)ejkz z dz
N Z1 nz N Z1
sin( nz
X jk z X
= cos( L )e dz + z
L )ejkz z dz (A.11)
n ;1
=0 n ;1 =1
Considering the st integral in equation A.11, the cosine function is only dened over
the interval ; L to L and is otherwise zero, therefore can be written as
2 2
N Z1 nz N Z L2
cos( nz )ejkz z dz
X jk z X
cos( )e dz =
z
n=0 ;1 L n ;2
L L
=0
N 2(;1)n n (A.12)
= (k ; ( nL) ) cos( kz2L )
X
2 2
n z L
=0
Considering the second integral in equation A.11, the sine function is only dened
over the interval ; L to L and is otherwise zero, therefore can be written as
2 2
N Z1 nz N Z L2
sin( nz )ejkz z dz
X jk z X
sin( )e dz =
z
n=1 ;1 L n ;2
L L
=1
(A.13)
= j (k2(;
XN 1)nkz cos( kz L )
z ;(L) )
n 2
22
n =1
A.3 Fourier Transform of Current Wave Page 152
Combining the two transforms with equation A.11, the Fourier transform of Jz is
N 2(;1)n n k N 2(;1)nkz cos( kz L )
X zL X
Jz (kz ) = L
(kz ; ( nL ) )
2
cos(
2
2
) + j (kz ; ( nL ) ) 2 2 2
n=0 n =1
(A.14)
where L is the length of the microstrip line and kn is the precalculated wave constant.
The case for the z{directed current considered rst. Taking a Fourier transform in z
direction with the transform
Z1
Jz (kz ) = ;1
Jz (z)ejkz z dz
Z L (A.18)
= 2
ej(kz ;kn)z dz
; L2
Considering the st integral in equation A.11, the current wave function is only dened
over the interval ; L to L and is otherwise zero. The Fourier transform yields as,
2 2
where
8
>
< 1 ; jx;l xn j xn ; lx x xn + lx
Jxn(x) = > x (B.4)
: 0 otherwise
8
>
<1 xn ; lx x xn + lx
Jzn(x) = > (B.5)
:0 otherwise
The electric eld can be related to the surface current (Js(x)) via the following integral
equation in the spectral domain
2 32 3 2 3
64 Gzz Gzx 7 6 Jz 7 6 Ez 7
54 5=4 5 (B.6)
Gxz Gxx Jx Ex
where
B.1 Two{Dimensional Version of SDM Page 156
G(kx; d; kz ; !) is the dyadic Green's function in the spectral domain,
Jz is the Fourier transform of the z{directed surface current density,
Jx is the Fourier transform of the x{directed surface current density,
E is the Fourier transform of the electric eld.
Because the calculations take place in the spectral domain, current basis functions
(equation B.3) must be transformed into the spectral domain by using the Fourier
transform equation which is dened as
Z1
(kx) = (x)ejkxxdx (B.7)
;1
To solve equation B.1, the Method of Moments is applied. This eliminates the un-
known Fourier transform of the electric eld (E) from the formulation. Inner products
are now taken by using a set of weighting functions which are identical to the set of
current basis functions (Procedure of Galerkin's).
2 32 3
64 Zzz Zzx 7 6 az 7
54 5 = 0 (B.8)
Zxz Zxx ax
the elements of the impedance matrix Z are given by
Z1
Zst = ;1
Jt (kx)G(kx; kz ; d; !)Js(kx)dkx (B.9)
The matrix equation B.8 is solved for the propagation constant kn in the z direction
by an iterative search for kz , solving for the determinant of the impedance matrix
equal to zero. The set of unknown current coecients (an) and thus the precalculated
surface current distribution in the direction perpendicular to current
ow are obtained
as elements of an eigenvector for the wave constant (kn). The set of coecients and
kn are stored to use in the complex microwave circuit calculation.
Appendix C
Coth
d 1 2
q q
i = kx + kz ; ki kx + kz
2 2 2 2 2
(C.1)
x z
The asymptotic form of the Green's function thus obtained can be arranged in a
closed form as follows,
h = ; ! (1j+ )
1 h = (1j!
2
0r
+ r )
0 r
K1zz = q zk 2
K1zz = q kz
2
1
kx + kz
2 2
2
(kx + kz )
2 2 3
Kxz = Kzx
1 1 Kxz = Kzx
2 2
K1xx = ; q kx K1xx = ; q kx
2 2
1
kx + kz
2 2
2
(kx + kz )
2 2 3
Appendix D
Excitation
D.1 Fourier Transforms of Basis Functions for Feedlines Page 161
D.1 Fourier Transforms of Basis Functions for Feed-
lines
The travelling current wave has been chosen in this thesis to be current basis functions
of the feedlines. The derivations of incident re
ected and transmitted current waves
from the travelling wave in the direction of propagation are given in the space domain
by;
8
>
< e;jkn z;zi ;L + zi z zi
( )
The case for the incident current wave in equation D.1 is considered rst. Taking the
Fourier transform in the direction of propagation (z direction) with the transform
Z1
Ji(kx; kz ) = ;1
Ji(z)ejkz z dz (D.4)
2
+
L L
(D.5)
= k ; k sin (kz ; kn) 2 e;j kz ;kn 2 ejkz zi ( )
z n
D.1 Fourier Transforms of Basis Functions for Feedlines Page 162
The Fourier transform of the re
ected current wave in equation D.2 is calculated by;
Z1
Jr (kx; kz ) = ;1
Jr (z)ejkz z dz (D.6)
The re
ected current wave is dened over the identical interval to incident current
wave, thus can be written as,
Z zi
Jr (kz ) = ;L+zi
ejkn z;zi ejkz z dz
( )
2 L (D.7)
;j kz kn ) L2 ejkz zi
=
kz + kn sin (kz + kn) 2 e
( +
Similarly, the Fourier transform of the transmitted current wave in equation D.3,
dened over an interval zo to L + zo , can be written as
Z L+zo
Jt(kz ) = z e;jkn(z;zo) ejkz z dz
o
2 L (D.8)
=
kz ; kn sin (kz ; kn) 2 ej(kz ;kn) L2 ejkz zo
Copy of Publications