This action might not be possible to undo. Are you sure you want to continue?
additional work is needed to transform these experimen

tal devices into stable elements for standards applications.
Also, standard ﬁelds need to be developed, in order to cal
ibrate these probes of the future.
BIBLIOGRAPHY
1. M. Kanda, Standard probes for electromagnetic ﬁeld mea
surement, IEEE Trans. Anten. Propag. AP41:1349–1363
(1993).
2. M. Kanda, Analytical and numerical techniques for analyzing
an electrically short dipole with a nonlinear load, IEEETrans.
Anten. Propag. AP28:71–78 (1980).
3. M. Kanda, A relatively short cylindrical broadband antenna
with tapered resistive loading for picosecond pulse measure
ments, IEEE Trans. Anten. Propag. AP26:439–447 (1978).
4. M. Kanda and L. D. Driver, An isotropic electricﬁeld probe
with tapered resistive dipoles for broadband use, 100kHz to
18GHz, IEEE Trans. Microwave Theory Tech. MTT35:124–
130 (1987).
5. M. Kanda, An electromagnetic nearﬁeld sensor for simulta
neous electric and magnetic ﬁeld measurements, IEEE Trans.
Electromagn. Compat. EMC26:102–110 (1984).
6. L. D. Driver and M. Kanda, An optically linked electric and
magnetic ﬁeld sensor for Poynting vector measurements in
the near ﬁelds of a radiating source, IEEE Trans. Elect
romagn. Compat. 30:495–503 (1988).
7. M. Kanda and K. D. Masterson, Optically sensed EMﬁeld
probes for pulsed ﬁelds, Proc. IEEE 80:209–215 (1992).
8. J. C. Tippet, Model Characteristics of Rectangular Coaxial
Transmission Line, Ph.D. dissertation, Electrical Engineering
Dept., Univ. Colorado, Boulder, 1978.
9. S. A. Schelkunoff and H. T. Friis, Antennas, Theory and Prac
tice, Wiley, New York, 1952.
10. F. M. Greene, The nearzone magnetic ﬁeld of a small circu
larloop antenna, J. Res. Nat. Bur. Stand. U.S. C. Eng. In
strum. 71C(4) (1967).
11. E. C. Jordan and K. G. Balmain, Electromagnetic Waves and
Radiating Systems, PrenticeHall, Englewood Cliffs, NJ, 1968.
ELECTROMAGNETIC INVERSE PROBLEMS
ANYONG QING
YEOW BENG GAN
National University of
Singapore
Singapore
1. INTRODUCTION
1.1. Definition of Electromagnetic Inverse Problems
Electromagnetic inverse problems, also known as electro
magnetic inverse scattering problems, aim to recover in
formation on some inaccessible region from the scattered
electromagnetic ﬁelds measured in the exterior region [1].
This region (hereafter referred to as ‘‘scatterers’’) could be
homogeneous or inhomogeneous, of any dimension, with
or without electromagnetic scatterers embedded. The in
formation of interest is the location, shape, size, electrical
properties (penetrable or impenetrable), and electromag
netic constitutive parameters. Most of the electromagnetic
inverse problems can be considered as ‘‘active,’’ where a
known incident ﬁeld is artiﬁcially applied to illuminate
the inaccessible unknown scatterers to give rise to scat
tered ﬁelds, which is measured at some accessible area.
There are some cases, however, such as passive remote
sensing, in which the scattered electromagnetic ﬁelds
from the unknown scatterers are not due to such artiﬁ
cially applied incident electromagnetic ﬁelds. Usually, the
scattered electromagnetic ﬁelds are measured over limited
domain of aspect angle, frequency, and polarization, and
are contaminated by noise and measurement error.
Electromagnetic inverse scattering is concerned with
how we can obtain a large part of information about the
world surrounding us. An everyday example of electro
magnetic inverse problems is human vision; from the
measurements of scattered light that reaches our retinas,
our brains construct a detailed threedimensional map of
the world around us. This is a highly automated process,
and most of us do not stop to reﬂect on how difﬁcult this
problem is. In fact, a large part of the human brain is
devoted to such activities.
It should be pointed out that this definition of electro
magnetic inverse problems is made in a narrow sense. In
fact, all synthetic problems in electromagnetics, for exam
ple, the design of microwave ﬁlters and the synthesis of
antennas, and many others, can be regarded as electro
magnetic inverse problems. However, we will stick to this
narrow definition of the electromagnetic inverse problems,
unless otherwise stated.
Electromagnetic inverse problems and electromagnetic
scattering problems come in pairs. For a given electro
magnetic scattering problem, a priori information on the
size, shape, and material constituents of the scatterers
and the incident electromagnetic ﬁelds is provided, and
the scattered ﬁeld is calculated for a specific area and fre
quency domain. The electromagnetic community has em
braced scattering problems with a warmth that is not
generally extended to inverse problems. In fact, our train
ing on electromagnetics is dominated by direct problems
(in the general sense), while inverse problems continue to
be regarded as very new and challenging research topics.
Note that electromagnetic inverse problems belong to a
much wider class of inverse problems and are closely re
lated to inverse problems in acoustic and elastic waves. It
is also known that some techniques used in one ﬁeld are
identical, at least in principle, to those used in other, com
pletely different ﬁelds. These interdisciplinary applica
tions of the inversion techniques are drawing increasing
attention. Hence, although we focus on electromagnetic
inverse problems here, other ﬁelds of inverse problems
will be touched on slightly where necessary.
1.2. Some Mathematical Challenges in Electromagnetic
Inverse Problems
Electromagnetic inverse scattering problems and the as
sociated electromagnetic scattering problems are highly
1200 ELECTROMAGNETIC INVERSE PROBLEMS
Previous Page
mathematical. This is an ideal area for applied mathema
ticians. Electromagnetic inverse problems provide a rich
supply of challenging mathematical problems.
1.2.1. Electromagnetic Scattering. Studying an electro
magnetic inverse problem always requires a solid knowl
edge of the corresponding direct scattering problem.
Unfortunately, the study of electromagnetic scattering
problems is very far from complete. For example, the cur
rent level of understanding on wave propagation in com
plex media and random media remains very poor.
1.2.2. Nonlinearity. The scattering problem is linear in
the sense that for a known scatterer, the relationship be
tween the incident ﬁeld and scattered ﬁeld is linear. How
ever, for electromagnetic inverse problems, we are
interested in the relationship between the scatterer and
its action on the incident ﬁelds. This relationship is in
herently nonlinear.
1.2.3. IllPosedness. According to Hadamard [2], a
problem is wellposed if it has a unique solution that de
pends continuously on the given data. Problems that are
not wellposed are known as illposed. In general, electro
magnetic inverse problems are illposed. The illposedness
of electromagnetic inverse problems comes from the in
complete and contaminated measurement data, and the
existence of a nonradiating source.
For an illposed electromagnetic inverse problem, the
following questions must be addressed:
Existence: Is there any solution?
Uniqueness: Is the solution unique?
Stability: Is the solution stable? In other words, do
small perturbations of the measured scattered ﬁelds
always result in small perturbation of the solution?
1.3. Applications of Electromagnetic Inverse Problems
Although the electromagnetic inverse problem is a rela
tively new area of applied mathematical research, it has
been increasingly used in scientific, military, medical, in
dustrial, agricultural, and many other civil areas. Since
the mid1980s we have witnessed an explosion in the ap
plications of electromagnetic inverse problems.
1.3.1. Scientific Applications. Electromagnetic inverse
problems have been enormously inﬂuential in the devel
opment of natural sciences, with great advances in science
and technology made possible through their solutions.
Electromagnetic inverse problems could lead to the estab
lishment of physical laws via indirect observations. Their
solutions provide us with a wealth of scientific informa
tion: the discovery of DNA structure through solving X
ray diffraction problems and the structure of the atom and
its constituents from studies on the scattering phenomena
when materials are bombarded with particles. Over the
years, electromagnetic inverse problems have played an
increasingly important role in many scientific areas, such
as archaeology, seismology, geophysics, optics, material
science, and meteorology.
1.3.2. Military Applications. The area of electromagnet
ic inverse problems was strongly stimulated by the great
success in military applications during World War II,
which witnessed the invention of radar and sonar for de
tection and identiﬁcation of both friendly and hostile ob
jects. As the world is not devoid of violence and war,
military demands on electromagnetic inverse problems
continue to increase. As an example, for people living in
countries under the threat of landmines left over from
earlier wars, safe detection and removal of these land
mines using electromagnetic waves is a lifesaving method.
On the other hand, increasingly powerful radar, spy sat
ellite networks, and missile defense systems are being
built to defend against attacks from terrorists and hostile
countries.
1.3.3. Medical Applications. So far, besides military ap
plications, medical imaging is one of the most successful
application areas of electromagnetic inverse problems. In
fact, the Xray radiography machine is used in almost ev
ery hospital to diagnosis tuberculosis and other anomalies
in the human body. In 1972, G. N. Hounsﬁeld introduced a
new radiographic imaging procedure, Xray computer
assisted tomography (CAT). A picture of the CAT appara
tus in a hospital is shown in Fig. 1. It is replacing the
conventional and obsolete Xray radiography machines in
many hospitals. More recently, even some CAT machines
have been replaced by more advanced nuclear magnetic
resonance imaging machines for more accurate diagnosis.
1.3.4. Industrial Applications. The electromagnetic in
verse problems are of commercial value to the industries.
Oil companies, for example, determine the location of oil
through solving inverse scattering problems. A significant
fraction of the computational workload is performed by
the oil companies. In fact, much of the recent resurgence
Figure 1. CAT apparatus in a hospital.
(Source: http://encyclopedia.thefreedictionary.com/
Computer%20assisted%20tomography.)
ELECTROMAGNETIC INVERSE PROBLEMS 1201
in the oil industry is due to improvements in mathemat
ical algorithms that allow scientists to ‘‘see’’ through salt
layers to detect the oilbearing strata below.
Ground penetrating radar (GPR) is a highly commer
cialized solution for nondestructive evaluation (NDE) that
produces a continuous crosssectional proﬁle or record of
subsurface features, without drilling, probing, or digging.
The GPR for a shallow survey is shown in Fig. 2. Ground
penetrating radar proﬁles are used to evaluate the loca
tion and depth of buried objects such as pipes and cables,
and to investigate the presence and continuity of natural
subsurface conditions and features.
The commercial aviation industry has also beneﬁted
significantly from solving electromagnetic inverse prob
lems. For example, it is now mandatory for passengers to
receive security check, including metal detection, before
boarding. Other industries taking advantage of electro
magnetic inverse problems include nuclear energy, food,
construction, and marine enterprises. For example, the
nondestructive evaluation of steam generator tubes is a
major issue in the nuclear industry.
1.3.5. Agricultural Applications. Electromagnetic in
verse problems are also very important to agriculture. In
China, a large amount of effort has been put into predict
ing the national crop harvest using remote sensing tech
niques and monitoring the coverage of plants within
Chinese territory.
1.3.6. Other Civil Applications. Electromagnetic in
verse problems are also applicable to many other civil ar
eas. For example, the environment can be monitored using
remote sensing techniques.
1.4. Some Outstanding Research Groups
Electromagnetic inverse problems have been one of the
most challenging areas in electromagnetics. Many
research groups and researchers from both within and
outside the electromagnetic community have devoted
their efforts to study the fundamental theory of inverse
problems, develop inversion algorithms for particular
cases, produce real datasets to test the developed algo
rithms, develop prototypes for practical applications, and
commercialize the developed prototypes. Because of the
large number of researchers and groups involved in this
area, it is not possible to list all of them. However, special
tributes should be paid to the following researchers and
organizations for their outstanding work in this ﬁeld: In
stitut Fresnel (France), W. M. Boerner, R. E. Kleinman, K.
J. Langenberg, Rome Laboratory (renamed as Air Force
Research Laboratory in 1997), and A. N. Tikhonov.
Their wellestablished contributions to this ﬁeld will be
highlighted at the appropriate parts in this article.
1.5. Some Future Directions
The ﬁeld of electromagnetic inverse problems is a rela
tively new area with apparent potential for applications.
However, many problems remain to be solved. Besides the
topics mentioned in this article, the following topics in
electromagnetic inverse problems also require special at
tention.
1.5.1. ThreeDimensional Electromagnetic Inverse Prob
lems. Researchers in the ﬁeld of electromagnetic inverse
problems typically simplify their work by concentrating on
lowerdimensional problems. Electromagnetic inverse
problems that have been extensively studied usually as
sume geometry of only one or two dimensions, with a half
space or layered medium. However, such an assumption is
insufﬁcient for many practical problems, leading to failure
of the inversion algorithms to produce any meaningful
results.
Increasing attention has been devoted to threedimen
sional electromagnetic inverse problems. Researchers are
attempting to extend wellestablished lowerdimensional
inversion algorithms to threedimensional cases, as well
as to develop new inversion algorithms focusing on three
dimensional electromagnetic inverse problems.
1.5.2. Electromagnetic Inverse Problems of Complex
Scatterers. Most of the existing inversion algorithms as
sume that the scatterers are simple, that is, isotropic and
nondispersive. However, this assumption is not valid for
many real scatterers, such as human brains and organs.
1.5.3. Electromagnetic Inverse Problems in Complex
Media. Another important issue is the electromagnetic in
verse problem of scatterers in complex media. Some ex
amples of complex media are soil, foliage, sea ice, sea
surface, and the human body.
2. AN OVERVIEW OF RESEARCHES ON
ELECTROMAGNETIC INVERSE PROBLEMS
Because of the theoretical and practical significance of
electromagnetic inverse problems, extensive studies have
been performed on electromagnetic inverse problems.
Figure 2. Shallow groundpenetrating radar surveys. (Source:
http://www.geomodel.com./)
1202 ELECTROMAGNETIC INVERSE PROBLEMS
These studies include fundamental theories of electromag
netic inverse problems, measurement of scattering data,
development of inversion algorithms for specific electro
magnetic inverse problems and testing them with syn
thetic and/or real measurement data, development of
prototypes for practical applications, and commercializa
tion of the developed prototypes. Numerous monographs,
technical reports, and dissertations have been published.
The number of journals specializing in inverse problems
or with scope on inverse problems has increased signiﬁ
cantly, with rapid growth in the number of special issues
and papers on inverse problems. A large number of inter
national conferences, workshops, and symposia (e.g., APS,
MTTS, IGARSS, APMC, PIERS) on electromagnetic
inverse problems have also been organized.
2.1. A Brief History
The study on electromagnetic inverse problems can be
traced back to 1895 when X rays were discovered in the
ﬁrst body imaging tests conducted by Wilhelm Roentgen,
the discoverer of X rays. The ﬁrst major mathematical
breakthrough took place in 1917 when the radon trans
form, the basic mathematical framework common to a
large class of inverse problems, was proposed [3]; unfor
tunately, this work was not well appreciated in the appli
cation areas until the 1970s. The second stimulation to
electromagnetic inverse problems occurred during World
War II, when radar was invented by scientists and engi
neers to identify friendly and hostile aircraft. However,
the ﬁeld of electromagnetic inverse problems was not well
recognized until A. N. Tikhonov established the funda
mental theory of illposed problems in the mid1960s [4].
In 1972, G. N. Hounsﬁeld introduced CAT [5]. Since then,
an extensive and intensive study on electromagnetic in
verse problems and other related inverse problems has
been carried out.
The ﬁrst extensive survey covering many ﬁelds in
which inverse problems were applied was presented in a
NASA memorandum by Colin [6]. In 1981, the IEEE
Transactions on Antennas and Propagation published a
special issue on inverse methods in electromagnetics [7].
Subsequently, W. M. Boerner organized the NATO Ad
vanced Research Workshop on Inverse Methods in Elec
tromagnetic Imaging in 1983 [5]. Afterward, there was a
burst in publication of research results scattered in many
monographs [8], technical reports, dissertations, journals,
and conference proceedings.
Nevertheless, the ﬁeld of electromagnetic inverse prob
lems remains one of the most challenging problems of ap
plied electromagnetics. A review on the latest state of the
art is presented in Ref. 1.
2.2. Mathematical Fundamentals of Electromagnetic
Inverse Problems
Applied mathematicians such as D. Colton, R. Kress [8],
and P. Monk studied the fundamental mathematical
theories behind the electromagnetic inverse problems.
Such studies strengthened our understanding of the
electromagnetic inverse problems and gave conﬁdence
to developers of inversion algorithms. Apparently, these
fundamental mathematical theories are also directly ap
plicable—or can be applied with minor modiﬁcations—to
other ﬁelds of inverse problems.
2.2.1. Existence and Uniqueness. As mentioned before,
in any realistic situation, the measured data are not exact.
Hence, an exact solution to a real inverse problem seldom
exists in its original solution space. However, existence
can be forced by enlarging or reducing the solution space.
Therefore, the common practice is to assume that an ap
proximate solution does exist and is physically feasible.
The existence of an inverse problem is rarely considered
since it is of little theoretical and practical importance.
There are three famous uniqueness theorems in elec
tromagnetic inverse problems [9]: Tikhonov’s uniqueness
theorem for a onedimensional inverse problem, Weidelt’s
uniqueness theorem for a twodimensional inverse prob
lem with an electrical conductivity described by an ana
lytic function, and Gusarov’s uniqueness theorem for a
twodimensional inverse problem with a piecewise analyt
ic function of electrical conductivity. Later, Colton, Kress
[8], and P. Monk obtained the more general uniqueness
theorems for electromagnetic inverse problems.
2.2.2. Stability and Regularization. We will now consider
the issue of stability of inverse problems. This problem
reﬂects the practical fact that although two sets of obser
vation data may differ only in the noise level, the corre
sponding inversion results could be completely different.
Negligence or improper remedy for the illness could have
disastrous consequences. The common technique to re
store stability of an inverse problem is to employ regular
ization methods, which are discussed in detail in the
tutorial paper [10] as well as other topical review papers
[11–14] and monographs [4,8,9,15].
Consider a general inverse problem
Au¼f ð1Þ
where A:uAXfAYis an operator mapping space X into Y.
The wellknown Tikhonov regularization is designed to
minimize the Tikhonov functional deﬁned as
jjAu Àf jj
2
þgjjf jj
2
ð2Þ
2.3. Development of Inversion Algorithms
Development of inversion algorithms is always the key is
sue in the ﬁeld of electromagnetic inverse problems. Ex
tensive studies have been carried out on this subject, with
many inversion algorithms developed.
As relevant literature are scattered over a large num
ber of resources, a summary is given here for convenience
for future study. However, the author makes no claim that
this summary is exhaustive, since it is based only on the
author’s current knowledge.
As mentioned earlier, the ﬁeld of inverse problems is
interdisciplinary, as clearly demonstrated by the vast di
versity of literature on electromagnetic inverse problems
ELECTROMAGNETIC INVERSE PROBLEMS 1203
and the background of researchers involved in the area.
Some techniques used in one ﬁeld are identical, at least in
principle, to those used in completely different ﬁelds.
Therefore, the inversion algorithm is considered here re
gardless of its origin, if it is applicable to electromagnetic
inverse problems.
2.3.1. Linear Inversion Algorithms. The earliest at
tempts to solve the electromagnetic inverse problems sim
pliﬁed them by linearization through applying some
approximations, such as the Born approximation, the
Rytov approximation, and physical optics approximations
(or Kirchhoff approximation). Generally speaking, such
linear inversion algorithms are mathematically very ele
gant. For onedimensional (1D) electromagnetic inverse
problems, such linearization may lead to an analytical so
lution of the problem. Unfortunately, we seldom encounter
1D problems in practice. Moreover, the simplicity of the
1D case is rarely preserved in higher dimensions.
The most prominent inversion algorithms in this cate
gory include the layerstripping algorithm, the method of
characteristics, the Gelfand–Levitan–Marchenko integral
equation method, the Born iterative method, and the lin
ear diffraction tomography algorithm. K. J. Langenberg
established his reputation in developing linear inversion
algorithms. Readers are referred to Wirgin’s text [16] for
detailed contributions by Langenberg to linear inversion
algorithms.
2.3.2. Nonlinear Inversion Algorithms. As mentioned
previously, nonlinearity is one of the distinct features of
electromagnetic inverse problems. Developers of linear in
version algorithms usually ignore this basic feature of the
inverse problems. In addition, the approximations made
in linear inversion algorithms imposed too many restric
tions, thereby limiting their applications.
The nonlinear inversion algorithms retain the nonlin
earity of the electromagnetic inverse problems. Most of the
nonlinear algorithms are mathematically more difﬁcult
and computationally more expensive, as nonlinear elec
tromagnetic inverse problems are solved iteratively. On
the other hand, such algorithms are usually more versa
tile and accurate, hence attracting more attention from
practical applications.
In essence, electromagnetic inverse problems are opti
mization problems and can be solved by applying different
optimization algorithms. Accordingly, nonlinear inversion
algorithms can be further categorized in accordance with
the optimization algorithms applied: the local inversion
algorithms and the global inversion algorithm.
2.3.2.1. Local Inversion Algorithms. As the term im
plies, optimization algorithms involved in this category
are local, or gradientbased. The prominent local algo
rithms include the distorted Born iterative method (or
distorted wave Born iterative method, Newton–Kantorov
itch algorithm), the conjugategradient method, the local
shape function method, the modiﬁed gradient method, and
the contrast source inversion method.
R. E. Kleinman and P. M. van den Berg proposed the
modiﬁed gradient method and the contrast source inver
sion method, which have been actively applied. The two
algorithms have proved to be effective through experiment
using two sets of real data to be discussed later in this
article. Numerous papers have been published on their
application in the ﬁeld of electromagnetic inverse prob
lems and other ﬁelds of inverse problems. For details of
the two methods, please refer to their review paper [17].
2.3.2.2. Global Inversion Algorithms. The local algo
rithms rely critically on an initial solution to the electro
magnetic inverse problem of interest. Under some strict
hypotheses such as convexity, differentiability, and conti
nuity of the object function, the solution can be obtained
quickly from a good initial solution. Otherwise, the local
algorithms will be trapped in local minima or may even
diverge.
Global optimization algorithms have gained popularity
in the optimization community for their simplicity, versa
tility and, most of all, strong ability in searching. The
global optimization methods, including the genetic algo
rithms (GA), the evolution strategies, the Monte Carlo
method, the neural network, the simulated annealing
method, the memetic algorithm, and the particle swarm
algorithm, are in general stochastic.
For developers of the inversion algorithm, interest in
the global optimization algorithm has increased since the
1990s. Currently, S. Caorsi, C. C. Chiu, A. Massa, M. Pas
torino, and A. Qing are very active in this area.
2.3.2.3. Hybrid Inversion Algorithms. It is evident that
local and global inversion algorithms are complementary
to each other. It is therefore very logical to combine these
two types of inversion algorithms in an attempt to obtain
the global optimal solution within a shorter timeframe.
The common practice of this approach is to use a preset
threshold. Once the optimal objective function value ob
tained by the global inversion algorithm exceeds the
threshold, the inversion is transferred to the local inver
sion algorithm. A smarter hybrid inversion algorithm
combining the realcoded genetic algorithm (RGA) and
the Newton–Kantorovitch algorithm (NKA) is presented
in Ref. 18.
2.4. Veriﬁcations of Inversion Algorithms
Any developed inversion algorithm is subject to veriﬁca
tion. Such veriﬁcation includes tests on synthetic data
(scattered ﬁeld data obtained by numerical solution of the
corresponding scattering problem from a known scatter
er), tests on real data generated in controlled experiments,
and practical testing in the real world, where possible.
2.4.1. Synthetic Test and Inverse Crime. It is acceptable
and is almost a common practice to ﬁrst verify an inver
sion algorithm on synthetic data. The inversion algorithm
developer uses a synthetic test as a preliminary evalua
tion of the algorithm’s performance. The basic advantages
of this test are that it is costeffective and can be per
formed as and when required.
To ensure a reliable synthetic test of an inversion al
gorithm, inverse crime must be avoided. The inverse
1204 ELECTROMAGNETIC INVERSE PROBLEMS
crime, as deﬁned by Colton and Kress [8], implies that the
forward scattering solver used in generating the synthetic
data is also used in the inversion algorithm, and/or the
same discretization is used in both numerical procedures.
Synthetic test committing inversion crime usually leads to
surprisingly good results.
2.4.2. Test Using Controllable Experimental Data. As
compared to the synthetic test, verifying an inversion al
gorithm based on laboratorygenerated experimental
data, especially the wellestablished data generated inde
pendently by other researchers, is more convincing. The
algorithm developer can use such a test for an objective
performance evaluation of the algorithm. It is also very
beneﬁcial for the developer to reﬁne the algorithm on the
basis of experience gained during such test.
The Ipswich dataset generated by Rome Laboratory in
1995 is well known in the community of electromagnetic
inverse problems. Friendly contests among researchers
based on this dataset have been conducted and presented
in the IEEE Antennas and Propagation Magazine special
sessions [19–22]. The tested algorithms include the
Newton–Kantorovitch method, the modiﬁed gradient
method, the conjugategradient method, the local shape
function method, the backpropagation routine, the con
trast source inversion method, and the variable metric
method.
In 2001, Institut Fresnel provided a second set of real
data for friendly contests among algorithm developers
[23]. One interesting feature of this dataset comes from
the wide frequency range. The tested inversion algorithms
are the modiﬁed gradient method, the contrast source in
version method, the linear diffraction tomography algo
rithm, the realcoded genetic algorithm, the differential
evolution strategy (DES), the distortedwave Born itera
tive method, the linear spectral estimation technique, the
iterative and distorted Born method, and some other in
version algorithms.
Of course, some other researchers from both within and
outside the community of electromagnetic inverse prob
lems have produced experimental data on their own. How
ever, their works are more selffocused and their
experimental setups are more specific to their own inver
sion algorithms.
2.5. Prototype Development and Commercialization
The ultimate objective in studying electromagnetic in
verse problems is to apply the developed inversion algo
rithms and experimental setups in practical areas, and
perhaps commercialization as well. Besides the successful
commercialization of CATand GPR, many researchers are
building their own prototypes and testing them in differ
ent practical ﬁelds [1].
3. A STANDARD ELECTROMAGNETIC INVERSE PROBLEM
In this section, we consider a standard electromagnetic
inverse problem, namely, the reconstruction of multiple
perfectly conducting cylinders in free space.
The geometry of the problem is depicted in Fig. 3,
where O is the origin, O is a circle with radius R
meas
in
the measuring domain in which the scattered electric
ﬁelds are measured, the black dots on O are receivers,
and D is the imaging domain, which is usually chosen to
be circular or rectangular; K perfectly conducting cylin
ders within D are the objects to be reconstructed, and O
i
is
the local origin of the ith cylinder, which can be any point
within the cylinder contour C
i
.
A transverse magnetic (TM) incident ﬁeld is assumed
(time factor of e
jot
assumed and suppressed)
E
inc
ðrÞ ¼ ^ zzE
inc
z
¼ ^ zz expðÀjk
0
^
kk
.
rÞ ð3Þ
where o¼2pf is the angular frequency, r ¼x^ xx þy^ yy, k
0
¼
o/c is the wavenumber in free space, c is the speed of light,
^
kk ¼ cos j^ xx þ sin j^ yy is the incident wave unit vector, and j
is the incident angle; ^ xx, ^ yy, and ^ zz are the unit vectors in the
x, y, and z directions, respectively.
3.1. Problem Formulation
3.1.1. Direct Problem. Electric surface currents
J
j
ðrÞ ¼ ^ zzJ
zj
ðrÞ j ¼1, K are induced on the surface of cylin
ders, and the scattered electric ﬁeld E
scat
ðrÞ ¼ ^ zzE
scat
z
ðrÞ is
subsequently computed as follows
E
scat
z
ðrÞ ¼
K
j ¼1
À
om
0
4
_
C
j
J
zj
ðr
0
ÞH
ð2Þ
0
ðk
0
jr Àr
0
jÞdr
0
ð4Þ
where m
0
is the permeability of free space and H
ð2Þ
0
( Á ) is
the Hankel function of the second kind of zeroth order.
O
2
d
2
d
K
0
2
ç
2
0
−ç
K
ç
1
O
1
d
1
0
1
O
O
R
meas
ç
E
inc
D
Ω
Figure 3. Geometry of the problem.
ELECTROMAGNETIC INVERSE PROBLEMS 1205
On the surface of the cylinders, the tangential compo
nent of the total electric ﬁeld vanishes:
½E
scat
ðrÞ þE
inc
ðrÞ
.
^ zz ¼0 r 2
K
i ¼1
C
i
ð5Þ
Therefore
E
inc
z
ðrÞ ¼
K
j ¼1
om
0
4
_
C
j
J
zj
ðr
0
ÞH
ð2Þ
0
.
ðk
0
jr Àr
0
jÞdr
0
; r 2
_
K
i ¼1
C
i
ð6Þ
The distribution of surface current is obtained after solv
ing (6) by the pointmatching method [24] with pulse basis
function and the Dirac delta testing function.
3.1.2. Inverse Problem. The inverse problem is to locate
the unknown cylinders and reconstruct the cylinder con
tours, given the scattered electric ﬁelds measured on O,
E
scat
meas
ðr 2 OÞ. Usually, E
scat
meas
is measured at a ﬁnite num
ber of frequency points (N
f
) and a ﬁnite number of incident
angles for the incident plane waves (N
a
) by a ﬁnite num
ber of receivers (N
r
). Accordingly, E
scat
meas
becomes an (N
f
Â
N
a
Â N
r
)dimensional vector. Equations (4) and (6) are the
data and object (or state) equations of the present electro
magnetic inverse problem.
Mathematically, the inverse problem described above is
not well deﬁned. The locations and contours of the cylin
ders must be represented by some mathematical quanti
ties or functions. Usually, the location of a cylinder is
represented by any point (center) O
i
(d
i
,c
i
) within its con
tour, while the contour is represented by a local shape
function (LSF) r
i
¼F
i
(y
i
) in its local polar coordinate sys
tem, where d
i
and c
i
are the polar radius and angle of O
i
and r
i
and y
i
are the local polar radius and angle of a point
on the ith cylinder’s contour, respectively.
The inverse problem is cast into an optimization prob
lem by minimizing the object functional with respect to
the local origins and local shape functions as follows
f
_
K
i ¼1
½d
i
; c
i
; F
i
_ _
¼
jjE
scat
meas
ÀE
scat
simul
jj
jjE
scat
meas
jj
þa
K
i ¼1
K
j ¼1;jOi
_
C
i
F
i
ðy
i
ÞU
.
½F
j
ðy
j
Þ Àr
ij
ðy
j
Þdy
i
ð7Þ
where E
scat
simul
is an (N
f
Â N
a
Â N
r
)dimensional vector con
stituting the scattered electric ﬁeld simulated with the
latest reconstruction results of cylinder locations O
i
and
local shape functions F
i
and a is the intensity of penalty
imposed on infeasible reconstruction results with inter
secting cylinders as shown in Fig. 4:
jjE
scat
meas
jj ¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
N
f
ÂNa ÂNr
j ¼1
jðE
scat
meas
Þ
j
j
2
¸
¸
¸
_
jjE
scat
meas
ÀE
scat
simul
jj ¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
N
f
ÂN
a
ÂN
r
j ¼1
jðE
scat
meas
Þ
j
ÀðE
scat
simul
Þ
j
j
2
¸
¸
¸
_
UðtÞ ¼
0; to0
1; t ! 0
_
_
_
The ﬁrst term of the objective functional (7) is the rel
ative reconstruction error that gives a measurement on
how close the reconstruction results approach the true
proﬁle, while the second term is an artiﬁcial penalty
applied only in the case of intersecting cylinders.
3.1.3. Local Shape Functions
3.1.3.1. Trigonometric Local Shape Function. Most re
searchers approximate the LSF by a trigonometric series
of order N/2:
F
i
ðy
i
Þ % F
T
i
ðy
i
Þ ¼
N=2
n¼0
A
in
cosðny
i
Þ
þ
N=2
n¼1
B
in
sinðny
i
Þ
ð8Þ
Under such an approximation, the objective functional (7)
becomes the following objective function
f
_
K
i ¼1
½d
i
; c
i
; F
i
_ _
% f
T
ðx
T
Þ ð9Þ
where x
T
¼[x
1
? x
K
], x
i
¼½d
i
c
i
A
i0
Á Á Á A
iNÀ1
B
i1
Á Á Á
B
iNÀ1
.
A basic requirement on a LSF is that it be nonnegative.
However, it is hardly possible to guarantee the nonnega
tive definiteness of the trigonometric local shape function
(TLSF) when it is applied in the global inversion algo
rithms. The unrealistic initial proﬁle shown in Fig. 5 is the
O
j
O
i
j
i
j
j
ç
j
0
j
0
i
ç
i
j
ij
O
x
y
Figure 4. Intersecting cylinders.
1206 ELECTROMAGNETIC INVERSE PROBLEMS
best proﬁle in the initial population while reconstruct
ing proﬁle SP2, to be deﬁned later in this article using
RGA.
Some tricks can be used to guarantee the nonnegative
definiteness of the TLSF, for example, by restricting
the coefﬁcient A
i0
to be in a positive range with a big
positive average value, and other coefﬁcients in a much
smaller range. However, for the present electro
magnetic inverse problem, such tricks impose the unrea
sonable requirement of a priori knowledge of the cylin
der’s size.
3.1.3.2. Closed Cubic BSpline Local Shape Function. The
difﬁculty of the TLSF can be overcome by the closed cubic
Bspline local shape function (BLSF) of N control points
proposed by A. Qing [25]
F
i
ðy
i
Þ % F
B
i
ðy
i
Þ ¼R
N
2p
y
i
_ _
ð10Þ
where
Rðt
0
Þ ¼
NÀ1
n¼0
r
n
ðt
0
ÀnÞ
r
n
ðtÞ ¼C
modðnÀ1;NÞ
Q
0
ðtÞ
þC
modðn;NÞ
Q
1
ðtÞ þC
modðnþ1;NÞ
Q
2
ðtÞ
þC
modðnþ2;NÞ
Q
3
ðtÞ; t 2 ½0; 1
Q
0
ðtÞ ¼
1
6
ð1 ÀtÞ
3
Q
1
ðtÞ ¼
1
2
t
3
Àt
2
þ
2
3
Q
2
ðtÞ ¼ À
1
2
t
3
þ
1
2
t
2
þ
1
2
t þ
1
6
Q
3
ðtÞ ¼
1
6
t
3
One promising feature of the BLSF is its boundedness, ex
pressed mathematically as
min
0nNÀ1
C
in
F
B
i
ðy
i
Þ max
0nNÀ1
C
in
0 y
i
2p
ð11Þ
Therefore, the nonnegative definiteness of the BLSF can
be guaranteed if all control points C
in
are nonnegative.
Such natural condition does not impose any artiﬁcial re
strictions on the choice of the control points.
The corresponding objective function is
f
_
K
i ¼1
½d
i
; c
i
; F
i
_ _
% f
B
ðx
B
Þ ð12Þ
where x
B
¼[x
1
? x
K
], x
i
¼[d
i
c
i
C
i0
? C
iNÀ1
]
3.2. Applicable Inversion Algorithms
The present electromagnetic inverse problem has been
solved using different inversion algorithms, including the
linear diffraction tomography algorithm, the NKA, the GA,
the DES, the differential evolution strategy with individ
uals in groups (GDES), and hybrid algorithms. Reconstruc
tion using the linear diffraction tomography algorithm is
conducted by K. J. Langenberg and his colleagues. Readers
are referred to Ref. 23 for details, which will not be dis
cussed here. In addition, since the hybrid algorithm is ba
sically a combination of the NKA and one of the global
inversion algorithms, it is also not included here. Interest
ed readers are referred to Ref. 18 for more information.
3.2.1. Newton–Kantorovitch Algorithm. The NKA, a
generalization of the wellknown Newton method, was
proposed by Kantorovitch in 1948 and was applied to re
construct the shape of a perfectly conducting cylinder in
1981 with a priori knowledge of the exact location of the
cylinder [26]. The TLSF is used to approximate the cylin
der contour. Although mathematically complicated, it
is straightforward to generalize the formulation in Ref.
26 to reconstruct the locations and contours of multiple
y
(
m
)
x (m)
−1.0 −0.5 0.0 0.5 1.0 1.5
True
Initial
Reconstructed
−0.6
−0.4
−0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
Figure 5. An unrealistic initial proﬁle repre
sented by trigonometric local shape functions.
(This ﬁgure is available in full color at http://
www.mrw.interscience.wiley.com/erfme.)
ELECTROMAGNETIC INVERSE PROBLEMS 1207
perfectly conducting cylinders. The TLSF and the BLSF
can be regarded as two special cases of a more general LSF
in the form of expansion of basis functions.
The key idea in the NKA is to obtain the relationship
between small variation in the measured scattered elec
tric ﬁeld, dE
scat
z
ðrÞ, and small variation in the cylinder’s
proﬁle parameters dx:
dE
scat
z
ðrÞ ¼A
.
dx ð13Þ
This equation is usually overdetermined and illposed.
Different regularization measures can be applied to obtain
an update of the cylinder’s proﬁle parameters dx
dx¼A
À1
.
dE
scat
z
ðrÞ ð14Þ
where A
À1
, the inverse of operator (matrix) A, is in gen
eral obtained with regularization.
The cylinder’s proﬁle parameters are obtained itera
tively as follows:
1. Select an initial solution X
0
.
2. Solve the scattering problem for E
scat;k
simul
to obtain
dE
scat;k
z
¼E
scat;k
simul
ÀE
scat
meas
.
3. Update the cylinders’ proﬁle parameters
x
k þ1
¼x
k
þdx
k
¼x
k
þA
À1
k
.
dE
scat;k
z
.
4. Repeat steps 2 and 3 until the termination condi
tions are satisﬁed.
3.2.2. Genetic Algorithms. The concept of GA was ﬁrst
proposed by Holland [27]. It imitates the mechanism of
natural selection and evolution, and aims to solve an op
timization problem with object function f(x) where x is an
Ndimensional vector of optimization parameters. Usual
ly, the objective function f(x) is scaled to a ﬁtness function
ﬁt(x).
Genes and chromosomes are the basic building blocks
of GA. A gene can be the optimization parameter itself, or
a code in a code string of the encoded optimization param
eters. A chromosome is a concatenation of genes that takes
the form
Chromosome ¼ g
1
1
g
1
2
Á Á Á g
1
L
1
¸ﬄﬄﬄﬄﬄﬄﬄﬄ..ﬄﬄﬄﬄﬄﬄﬄﬄ¸
x
1
g
2
1
g
2
2
Á Á Á g
2
L
2
¸ﬄﬄﬄﬄﬄﬄﬄﬄ..ﬄﬄﬄﬄﬄﬄﬄﬄ¸
x
2
Á Á Á g
N
1
g
N
2
Á Á Á g
N
L
N
¸ﬄﬄﬄﬄﬄﬄﬄﬄﬄ..ﬄﬄﬄﬄﬄﬄﬄﬄﬄ¸
x
N
ð15Þ
where g
i
j
is a gene and L
i
is the length of the code string of
the ith optimization parameter.
The GA operates on a population of N
pop
individuals
simultaneously. Each individual is related to a ﬁtness val
ue, an optimization parameter vector, and a chromosome.
The GA starts with an initial population randomly gener
ated within the search space. On completing the initial
ization, the population enters the main GA loop and
performs a global optimization to search for the optimum
solution of the problem. The GA loop continues until the
termination conditions are fulﬁlled. A block diagram of the
GA is shown in Fig. 6.
In a GA loop, the three basic genetic operations—se
lection (or reproduction), crossover, and mutation—are
executed sequentially. The selection operator selects
good individuals on the basis of their ﬁtness values and
produces a temporary population, namely, the mating
pool. This can be achieved by many different schemes,
but the most common methods are the roulette wheel,
ranking, and stochastic binary tournament selection. The
selection operator is responsible for the convergence of
GA. The crossover operator is the main search tool. It
mates individuals in the mating pool by pairs and gener
ates candidate offspring by crossing over the mated pairs
with probability p
cross
. Many crossover schemes, such as
onepoint crossover and continuous or random multipoint
crossover, have been developed. After crossover, some of
the genes in the candidate offspring are subjected to mu
tation with a probability p
mut
. The mutation operator is
included to prevent premature convergence by ensuring
diversity in the population. There may be preprocessing
operation(s) before selection and postprocessing opera
tion(s) after mutation. However, these are not considered
as basic genetic operations.
The promising advantages of GA are the robust global
search ability, simplicity, and versatility, while the noto
rious disadvantage of GA is the long runtime.
The GA was appreciated by the electromagnetic com
munity in 1990s and has gained popularity since then.
Details of the genetic algorithms and its applications in
electromagnetics are comprehensively summarized in the
two tutorial papers [28,29] and a review paper [30].
3.2.2.1. Standard Genetic Algorithm. C. C. Chiu [31] ap
plied the conventional standard (or simple) genetic algo
rithm (SGA, also known as binary genetic algorithm,
BGA), to reconstruct the shape of a perfectly conducting
cylinder in free space with an exact a priori knowledge of
the cylinder location. Unfortunately, the results presented
in this pioneering work are dubious.
The SGA encodes the optimization parameters into a
string of binary codes. A gene in SGA is a binary code. The
Selection
Crossover
Mutation
Termination conditions fulfilled?
Output results
Yes
No
Generating Initial Population
Preprocessing
Postprocessing
Figure 6. Block diagram of genetic algorithms.
1208 ELECTROMAGNETIC INVERSE PROBLEMS
correspondence between a chromosome and the optimiza
tion parameters is given by
x
i
¼x
min
i
þ
x
max
i
Àx
min
i
2
L
i
À1
L
i
j ¼1
g
i
j
2
L
i
Àj
i ¼1; N ð16Þ
where ½x
min
i
; x
max
i
is the search range for the ith optimiza
tion parameter, g
i
j
2 f0; 1g.
3.2.2.2. RealCoded Genetic Algorithm. The SGA does
not operate directly on the optimization parameters but on
a discretised representation of them. Discretization error
will inevitably be introduced when encoding a real pa
rameter. The encoding and decoding operations also make
the algorithm more computationally expensive for prob
lems with real optimization parameters. The RGA, a GA
working directly on real optimization parameters, is more
desirable.
In the RGA, a gene is the optimization parameter itself,
and the chromosome of the ith individual in the nth pop
ulation takes the form
Chromosome
n;i
¼x
n;i
¼½ x
n;i
1
x
n;i
2
Á Á Á x
n;i
N
ð17Þ
Consequently, the crossover and mutation operators used
in the RGA are quite different from those in the SGA.
Commonly used crossover schemes in RGA include one
point mixture, continuous or random multipoint mixture,
and arithmetical onepoint crossover. The mutation in
RGA takes the form of random perturbation.
A. Qing has successfully applied the RGA to recon
struct the location and shape of multiple perfectly con
ducting cylinders [1,32,33]. It has also been veriﬁed by
real data reconstruction [34].
3.2.3. Differential Evolution Strategy. It is evident that
the genetic operators in the genetic algorithms could be
destructive: (1) the selection operation discriminates those
lessﬁt parents—hence, their chance to produce possibly
better babies is deprived; and (2) for crossover and muta
tion operations, all parents are replaced by their baby
children, even if their children are less ﬁt.
The DES [35] is a very simple but very powerful global
optimizer. Its apparent distinction from GA is in the order
of execution of the genetic operations. In the DES, the
three genetic operations are executed in the order of mu
tation, crossover, and selection (or mother–child competi
tion). Each parent individual is given equal chance to have
her child. Selection is conducted between the mother and
her child for a survival chance. In this way, destructive
genetic operations are avoided. The other key behind the
success of the DES is a scheme for generating trial pa
rameter vectors.
The DES was ﬁrst applied to reconstruct the location
and radius of buried circular conductors or tunnels [36]. It
was later applied to the present standard electromagnetic
inverse problem [37].
The block diagram of the DES is shown in Fig. 7.
(Please refer to Refs. 35–37 for a detailed explanation.)
3.2.4. Differential Evolution Strategy with Individuals in
Groups. Although an inexact a priori knowledge of the
number of cylinders in the imaging domain enables the
execution of the NKA, the GA, and the DES, in terms of
reconstruction time and quality of reconstruction results,
an exact a priori knowledge of the number of cylinders in
the imaging domain is crucial for these algorithms. Un
fortunately, it is never easy to obtain such an exact a priori
knowledge.
Such difﬁculty is overcome by the recently (in 2004)
proposed GDES [38]. The number of cylinders is treated as
an additional optimization parameter, to keep it ﬂexible
during the reconstruction. Therefore, no a priori knowl
edge on the number of cylinders is necessary. A crude
guess on the maximum number of cylinders in the imag
ing domain is sufﬁcient.
The key idea of the GDES is to organize the entire pop
ulation into different groups according to the number
of optimization parameters that they have; for instance,
individuals in the same group have the same number of
optimization parameters. This is possible since the max
imum number of cylinders is limited and because individ
uals with the same number of cylinders have the same
number of optimization parameters. The basic genetic op
erations of the DES (mutation, crossover, and mother–
child competition) are performed within each group to de
termine the optimal proﬁle in the group. An additional
operator, the group competition, is introduced to adjust
the group’s size during evolution.
Since the number of cylinders of the reconstructed pro
ﬁle matches the exact number of cylinders of the true pro
ﬁle, the quality of the reconstruction results are therefore
Yes
Yes
Yes
No
No
No
i=1
i=population size? i=i+1 n=n+1
Save results and stop
Generate initial population x
0
randomly
and uniformly in the search range, n=0
f (x
n,opt
)<c or n>n
max
?
Mutate to have a trial vector v
n+1,i
(v
n+1,i
)
j
= (x
n,opt
)
j
+ [
j
[(x
n,p
1
)
j
− (x
n,p
2
)
j
]
crossover v
n+1,i
with x
n,i
to deliver a baby y
n+1,i
(y
n+1,i
)
j
=
(v
n+1,i
)
j
¸
j
< crossover probability
(x
n,i
)
j
otherwise {
f (y
n+1,i
)<f (x
n,i
)?
x
n+1,i
=y
n+1,i
x
n+1,i
=x
n,i
Figure 7. Block diagram of differential evolution strategy
(e—convergence threshold for minimization problem; b
j
—random
number uniformly distributed in [0,1], g
j
—random number uni
formly distributed in [0,1]; 1rp
1
ap
2
airpopulation size).
ELECTROMAGNETIC INVERSE PROBLEMS 1209
much better, while the reconstruction time required is
correspondingly reduced.
3.3. Numerical Results
3.3.1. Optimization of a Mathematical Function
3.3.1.1. Function for Optimization. The mathematical
function to be optimized is
f ðxÞ ¼
K
k¼1
ðx
k
ÀkÞ
2
ð18Þ
The optimal solution is f(x
k
¼k) ¼0.
3.3.1.2. Search Range. The optimization parameters
are searched within the range [ À500,500], unless other
wise speciﬁed.
3.3.1.3. Termination Conditions. The two termination
conditions are
1. Normal termination: min
1iNpop
f
n
ðx
n;i
Þ e ¼0:01
2. Abnormal termination: n4N
max
¼1000
3.3.1.4. SGA versus RGA. Here, K¼14 is considered,
with N
pop
¼256, p
cross
¼1.0, p
mut
¼0.1 for RGA. The one
point arithmetic crossover scheme is chosen as the cross
over operator. The converging curves of RGA are shown in
Fig. 8, in which the applied selection schemes are also
shown.
For the SGA, N
pop
¼256, p
cross
¼0.8, p
mut
¼0.04 are
used. No test is successful for SGAwithin the same search
range, which is subsequently narrowed to [–20,20] for
SGA. The corresponding curves are shown in Fig. 9. The
0 50 100 150 200 250 300
Tournament
Roulettewheel
0.00
0.02
0.04
0.10
0.08
0.06
O
p
t
i
m
a
l
o
b
j
e
c
t
i
v
e
f
u
n
c
t
i
o
n
v
a
l
u
e
Generation
Figure 8. Optimization of function f ðxÞ ¼
14
k¼1
ðx
k
ÀkÞ
2
x
k
A[ À500,500] using RGA.
0 200 400 600 800 1000
0
2
4
6
8
10
O
p
t
i
m
a
l
o
b
j
e
c
t
i
v
e
f
u
n
c
t
i
o
n
v
a
l
u
e
n
Onepoint crossover
Randon multipoint crossover
Figure 9. Optimization of function f ðxÞ ¼
14
k¼1
ðx
k
ÀkÞ
2
x
k
A[ À20,20] using SGA.
1210 ELECTROMAGNETIC INVERSE PROBLEMS
convergence is not satisfactory even with the narrowed
search range.
This comparison clearly shows the advantages of RGA
over SGA.
3.3.1.5. RGA versus DES. Here, K¼20 is considered,
with N
pop
¼256, p
cross
¼1.0, p
mut
¼0.1 for RGA and
N
pop
¼100, p
cross
¼0.9, b ¼0.7 for DES. The convergence
performances are shown in Fig. 10.
The DES converges to the true solution with fewer
generations and a smaller population size. It is apparent
that the DES outperforms the RGA.
3.3.2. Synthetic Reconstruction
3.3.2.1. Synthetic Proﬁles. The following proﬁles are re
constructed using one or more of the abovementioned in
version algorithms:
SP1—a single cylinder: d¼0, c¼0, F(y) ¼0.3 þ
0.05sin2y
SP2—two parallel circular cylinders: d
1
¼0.5, c
1
¼0,
F
1
(y
1
) ¼0.3; d
2
¼0.5, c
2
¼p, F
2
(y
2
) ¼0.3
SP3—a single cylinder: d¼0, c¼0, F(y) ¼0.3 þ
0.06cosy þ0.03siny þ0.025cos2y
SP4—two parallel circular cylinders: d
1
¼0.8, c
1
¼0,
F
1
(y
1
) ¼0.3; d
2
¼0.8, c
1
¼1801, F
2
(y
2
) ¼0.3
SP5—a single circular cylinder: d¼0, c¼0, F(y) ¼0.3
3.3.2.2. Settings of Parameters. The incident frequency
is 300 MHz (l ¼1m). It illuminates the unknown objects
from eight incident directions, j
i
¼ip/4, i ¼1–8; 32 receiv
ers are located uniformly on O, with R
meas
¼10l.
For SGA, N
pop
¼256, p
cross
¼0.8, p
mut
¼0.04 are used.
For the RGA, N
pop
¼256, p
cross
¼1.0, p
mut
¼0.1 are cho
sen, and for DES, p
cross
¼0.9, b ¼0.7. The population size
is 5 times that of the number of optimization parameters
for DES. Finally, for the GDES, N
pop
¼300, p
cross
¼0.9,
b ¼0.7 are used. The local origins of the cylinders are
searched in a circle around the origin with a radius of 1l.
BLSF is used to approximate the cylinder’s contours.
The number of control points of a BLSF is 6. The search
range of control points is [0, l]
Settings that differ from the above speciﬁcation will be
pointed out explicitly.
3.3.2.3. Generation of Synthetic Data. The measured
scattering data are obtained from simulation through
solving the direct problem with ﬁner subdivision of the
cylinder contours. The synthetic data are assumed to be
noise free.
3.3.2.4. Search Ability and Robustness
3.3.2.4.1. NKA versus RGA. The NKA and the RGA are
applied to reconstruct SP1 and SP2. For SP1, an inexact a
priori knowledge of the cylinder’s location, d¼0.1, c¼p/4,
is assumed. The contour of SP1 is approximated by a
TLSF of order 4. The search ranges for the coefﬁcients of
the TLSF are [0,2] for A
0
and [ À0.5,0.5] for A
i
and B
i
,
i ¼1–4. Similarly, the locations of cylinders are known in
exactly in advance for SP2, with d
1
¼0.4, c
1
¼0, d
2
¼0.4,
and c
2
¼p. The order of the TLSF for SP2 is 3. The search
ranges are set to be similar. The initial guess for the NKA
is generated using the four schemes described in Ref. 1.
SP1 and SP2 are reconstructed successfully by the
RGA. However, most of the trials with NKA fail to recon
struct the proﬁles. On the other hand, the few successful
trials consume significantly less time.
3.3.2.4.2. SGA versus RGA. SP3, the proﬁle considered
in Ref. 31, is reconstructed using SGA and RGA. To be
consistent with Ref. 31, the cylinder location is assumed to
be known exactly in advance. The contour is approximated
by a TLSFof order 4. The search range of A
0
is [0,2]. For A
i
and B
i
, i ¼1–4, the search ranges of [ À0.5,0.5] and [0,2]
are tested. RGA successfully obtained the proﬁle with
the narrow search range, while SGA can obtain only a
0 50 100 150 200 250 300 350 400
10
−3
10
−2
10
−1
10
0
10
1
10
2
10
3
10
4
10
5
10
6
10
7
Generation
O
p
t
i
m
a
l
o
b
j
e
c
t
i
v
e
f
u
n
c
t
i
o
n
v
a
l
u
e
Figure 10. Optimization of function f ðxÞ ¼
20
k ¼1
ðx
k
ÀkÞ
2
x
k
A[ À500,500] using DES
(solid lines) and RGA (dotted lines). (This
ﬁgure is available in full color at http://
www.mrw.interscience.wiley.com/erfme.)
ELECTROMAGNETIC INVERSE PROBLEMS 1211
quasioptimal result (e ¼0.03) in affordable simulation
time. The corresponding convergence performances are
shown in Fig. 11. Both RGA and SGA failed to reconstruct
the true proﬁle with wide search range.
3.3.2.4.3. RGA versus DES. Proﬁle SP4 is reconstructed
using both RGA and DES. For the DES, N
pop
¼100, p
cross
¼0.9, b¼0.7 are used. A priori knowledge of the exact
locations of the cylinders is assumed for RGA. The con
vergence performances are shown in Fig. 12.
3.3.2.5. Effect of A Priori Knowledge of the Number of
Cylinders. A priori knowledge of the number of cylinders
in the imaging domain is absolutely required for all algo
rithms considered, with the exception of the GDES. How
ever, such a priori knowledge is rarely available. Here, its
effect on the reconstruction will be studied.
SP4 and SP5 are the proﬁles of interest. The inversion
algorithms studied are DES and GDES. The GDES re
quires a maximum number of cylinders in the imaging
domain, which is set to be 3 here. This requirement does
not impose too much inconvenience, since the maximum
value is not necessarily exact and is usually restricted
only by the available computational resources.
When reconstructing SP4 using DES, two cases of such
a priori knowledge are considered. The ﬁrst case is that
the number (2) is exactly known. The other case is that the
number is incorrectly assumed to be 3. The relationship
between the optimal objective function value and genera
tion is shown in Fig. 13. Apparently, the incorrect assump
tion on the number of cylinders significantly delays the
reconstruction. However, the quality of the reconstruction
result is not significantly affected.
It is worth noting that the difference in the reconstruc
tion time for these two cases is more significant, since it
takes much longer for the second case to complete one
generation due to the larger population size and longer
time required to evaluate an individual.
The GDES obtained the true proﬁle within a much
shorter time as shown in Fig. 14.
Similar results are observed for SP5 as shown in Fig. 15
and Fig. 16. However, in this case, the quality of the
0 50 100 150 200
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
SGA
O
p
t
i
m
a
l
o
b
j
e
c
t
i
v
e
f
u
n
c
t
i
o
n
v
a
l
u
e
Generation
RGA
Figure 11. Reconstruction of proﬁle SP3 using RGA (solid lines)
and SGA (dotted lines). (This ﬁgure is available in full color at
http://www.mrw.interscience.wiley.com/erfme.)
0 100 200 300 400 500 600
0.0
0.2
0.4
0.6
0.8
1.0
DES
RGA
n
f
n
(
x
)
Figure 12. Reconstruction of proﬁle SP4 using DES (black solid
line) and RGA (red dashed line).
0 500 1000 1500 2000
0.0
0.2
0.4
0.6
0.8
1.0
1.2
Exact (2) number of cylinders
Inexact (3) number of cylinders
Generation
O
p
t
i
m
a
l
o
b
j
e
c
t
i
v
e
f
u
n
c
t
i
o
n
v
a
l
u
e
Figure 13. Reconstruction of proﬁle SP4 using DES with exact
(black solid line) and inexact (red dashed line) a priori knowledge
of the number of cylinders.
0 300 600 900 1200 1500 1800
0
30000
60000
90000
120000
150000
180000
GDES, maximum number of cylinders: 3
DES, inexact (3) of number of cylinders
T
i
m
e
(
s
)
Generation
Figure 14. Reconstruction of proﬁle SP4 using GDES (black sol
id line) and DES (red dashed line) with inexact a priori knowledge
of the number of cylinders.
1212 ELECTROMAGNETIC INVERSE PROBLEMS
reconstruction results using DES deteriorates with in
exact assumption on the number of cylinders in the imag
ing domain, as shown in Fig. 17.
3.3.3. Reconstruction Using Fresnel Real Dataset
3.3.3.1. Experimental Setup. As mentioned earlier, the
set of data was measured at Institut Fresnel, France [23].
The experimental setup is shown in Fig. 18. Please refer to
Ref. 23 for more details.
3.3.3.2. Proﬁles Tested. The following two proﬁles are
considered here:
RP1—centered 12.7 Â 25.4mm rectangular cylinder
(center position known before reconstruction)
RP2—decentered 12.7 Â 25.4mm rectangular cylinder
(center position unknown before reconstruction)
3.3.3.3. Measurement Noise. The data are very noisy,
as shown in Table 1.
3.3.3.4. Reconstruction Results. Both RGA and DES are
applied to reconstruct the proﬁles.
The local origin is searched within a circle around the
origin with radius 100mm. Other parameters are identi
cal as speciﬁed earlier. The reconstruction results are
shown in Fig. 19 and 20. The cylinder is exactly located
in all reconstructions. For RP1, excellent reconstruction
results on the shape of RP1 are obtained at 8 and 16 GHz.
The shape of RP1 at 4 and 12 GHz are a bit poorer but
acceptable. The reconstructed shape of RP2 is good at 8,
10, 14, and 16 GHz, but a little bit poorer at 2, 4, 6, and
12 GHz. It is also observed that the reconstruction result
of the centered case is in general better than those of the
decentered case.
0 100 200 300 400 500
0.0
0.2
0.4
0.6
0.8
1.0
1.2
1.4
1.6
Exact (1) number of cylinder
Inexact (2) number of cylinder
Inexact (2) number of cylinder
O
p
t
i
m
a
l
o
b
j
e
c
t
i
v
e
f
u
n
c
t
i
o
n
v
a
l
u
e
Generation
Figure 15. Reconstruction of proﬁle SP5 using DES with exact
(black solid line) and inexact a priori knowledge of the number of
cylinders (red dashed line—2; blue dotted line—3).
0 100 200 300 400 500
0
10000
20000
30000
40000
50000
GDES, maximum number of cylinders: 3
DES, inexact (2) number of cylinders
DES, inexact (3) number of cylinders
T
i
m
e
(
s
)
Generation
GDES, maximum number of cylinders: 2
Figure 16. Reconstruction of proﬁle SP5 using GDES (black sol
id line—2; red dashed line—3) and DES with inexact a priori
knowledge of the number of cylinders (green dotted line—2, blue
dashed–dotted line—3).
(a)
−0.4 −0.2 0.0 0.2 0.4
−0.4
−0.2
0.0
0.2
0.4
True
Final Best
y
(
z
0
)
x (z
0
)
(b)
True
Final Best
−0.4 −0.2 0.0 0.2 0.4
−0.4
−0.2
0.0
0.2
0.4
y
(
z
0
)
x(z
0
)
Figure 17. Reconstruction of proﬁle SP5 using DES with inexact
assumption of the number of cylinders: (a) assumed number of
cylinders two; (b) assumed number of cylinders three.
ELECTROMAGNETIC INVERSE PROBLEMS 1213
It should be noted that the sample target, which is a
rectangular cylinder, has four sharp corners. From the re
construction results, it is noted that the cubic Bsplines
have difﬁculties in modeling these sharp corners.
Because of the strong noise in the experimental data,
the reconstructed proﬁle is quasioptimal. It is observed
that the number of generations required to obtain the
quasioptimal proﬁle is comparable between the DES and
the RGA. However, it is still reasonable to claim that the
DES converges faster than the RGA, since the population
size of the DES here (40) is much smaller than that of the
RGA (256).
3.4. General Remarks
The abovementioned reconstructions and numerical expe
riences show that
1. The NKA is mathematically much more complicat
ed. Its mathematical formulation is strongly related
to the problem under consideration.
2. A good initial solution is very critical to the NKA.
When such an initial solution is available, the NKA
obtains the optimal solution very quickly.
3. In practice, the TLSF is applicable in the NKA since
the NKA requires good a priori knowledge.
4. The BLSF applies equally well in all inversion algo
rithms considered.
5. The SGA applies equally well to problems with any
kinds of parameters (discrete, integer, real, or their
mixture). However, its evolution mechanism needs
to be improved.
10 5 0 −5 −10 −15 −20
15
10
5
0
−5
−10
−15
−20
−25
−30
y
(
m
m
)
x (mm)
True
Synthetic
f=4GHz
f=8GHz
f=12GHz
f=16GHz
Figure 19. Reconstruction results of RP1. (This ﬁgure is available
in full color at http://www.mrw.interscience.wiley.com/erfme.)
Receiver
2D Target
Target rotation
Emitter
(fixed)
0
T
0
R
Figure 18. Fresnel experimental conﬁguration.
Table 1. Measurement Error of RP1
Frequency (GHz) Relative Error (%)
4 59.89
8 38.97
12 75.40
16 29.23
60
50
40
30
20
−15 −10 −5 0
x (mm)
(a)
(b)
5 10 15
True
2GHz
4GHz
6GHz
8GHz
y
(
m
m
)
60
50
40
30
20
−15 −10 −5 0
x (mm)
5 10 15
True
10GHz
12GHz
14GHz
16GHz
y
(
m
m
)
Figure 20. Reconstruction results of RP2: (a) 2–8GHz; (b) 10–
16GHz. (This ﬁgure is available in full color at http://
www.mrw.interscience.wiley.com/erfme.)
1214 ELECTROMAGNETIC INVERSE PROBLEMS
6. For problems involving real parameters only, the
DES outperforms the RGA and the RGA outper
forms the SGA.
7. The GDES is the ﬁrst choice, if the number of cyl
inders in the imaging domain is unknown.
More recently, largescale numerical experiments on
the DES have been carried out. It is conﬁrmed that the
performance of the DES is dependent on the choice of
genetic parameters. It is also noted that the required
number of objective function evaluations is proportional
to the population size, while the success probability is
inversely proportional to the population size. Therefore,
a tradeoff between success probability and time must be
made.
We have developed a novel global inversion algorithm:
the dynamic differential evolution strategy. Preliminary
results show that the required number of objective func
tion evaluations is independent of the population size. De
tailed and complete results will be reported soon, when
the simulation is completed.
4. CONCLUSIONS
This article addresses the electromagnetic inverse prob
lem, beginning with a definition of this problem, followed
by detailed discussion of the mathematical challenges in
volved and the application areas. We then attempt to pre
dict some future directions in electromagnetic inverse
problems and present a survey of the research in electro
magnetic inverse problems. This survey includes a brief
history of electromagnetic inverse problems, an overview
of practical application ﬁelds, and a brief summary of the
fundamental theories involved in the electromagnetic in
verse problems and the inversion algorithms developed.
The survey also covers the wellknown real dataset and
the tested inversion algorithms. Subsequently, a standard
electromagnetic inverse problem—the reconstruction of
multiple perfectly conducting cylinders—is considered.
The inversion algorithms applicable to the cases present
ed in this article have been discussed. Comparisons be
tween different algorithms have been made. Remarks on
the advantages and disadvantages of each algorithm are
given. The most recent advancements in this area are also
briefly discussed.
Acknowledgment
The authors would like to thank Ms. Cassie Craig and
Prof. Kai Chang for their kind invitation to make this
contribution.
BIBLIOGRAPHY
1. A. Qing, Electromagnetic inverse scattering of twodimen
sional perfectly conducting objects by realcoded genetic al
gorithm, IEEE Trans. Geosci. Remote Sens. 39(3):665–676
(March 2001).
2. J. Hadamard, Lectures on the Cauchy Problem in Linear Par
tial Differential Equations, Yale Univ. Press, New Haven, CT,
1923.
3. S. R. Deans, The Radon Transform and Some of Its Applica
tions, Krieger, Malabar, FL, 1993.
4. A. N. Tikhonov and V. Y. Arsenin, Solutions of IllPosed Prob
lems, Wiley, New York, 1977.
5. G. Schwierz, Review of tomographic imaging methods
applied to medical imaging, in W. M. Boerner, et al., eds.,
NATO Advanced Research Workshop on Inverse Methods
in Electromagnetic Imaging, Bad Windsheim, Franconia,
FRG, Germany, Sept. 18–24, 1983, Reidel, Dordrecht,
1983.
6. L. Colin, Proc. Workshop on Mathematics of Proﬁle Inversion,
NASA Technical Memorandum x62, 150, Ames Research
Center, Moffett Field, CA, 1972.
7. W. M. Boerner, A. K. Jordan, and I. W. Kay, eds., Special issue
on inverse methods in electromagnetics, IEEE Trans. Anten.
Propag. 29(2):185–417 (March 1981).
8. D. Colton and R. Kress, Inverse Acoustic and Electromagnetic
Scattering Theory, 2nd ed., Springer, Berlin, 1998.
9. M. S. Zhdanov, Geophysical Inverse Theory and Regulariza
tion Problems, Elsevier, Amsterdam, 2002.
10. G. A. Deschamps and H. S. Cabayan, Antenna synthesis and
solution of inverse problems by regularization methods, IEEE
Trans. Anten. Propag. 20(3):268–274 (1972).
11. M. Nashed, Operatortheoretic and computational approach
es to illposed problems with applications to antenna theory,
IEEE Trans. Anten. Propag. 29(2):220–231 (1981).
12. C. de Mol, A critical survey of regularized inversion methods,
in M. Bertero and E. R. Pike, eds., NATO Advanced Research
Workshop, on Inverse Problems in Scattering and Imaging,
Cape Cod, April 14–19, 1991, Adam Hilger, Bristol–Philadel
phia–New York, 1992, pp. 345–370.
13. B. Hofmann, Regularization of nonlinear problems and the
degree of illposedness, in G. Anger, R. Gorenﬂo, and H.
Jockmann, eds., Inverse Problems: Principles and Applica
tions in Geophysics, Technology, and Medicine, Wiley, New
York, 1993, pp. 174–188.
14. F. Natterer, Regularization techniques in medical imaging, in
M. A. Viergever and A. ToddPokropek, eds., Mathematics and
Computer Science in Medical Imaging, SpringerVerlag, New
York, 1988, pp. 127–141.
15. A. Asanov, Regularization, Uniqueness and Existence of So
lutions of Volterra Equations of the First Kind, VSP, Utrecht,
The Netherlands, 1998.
16. K. J. Langenberg, M. Brandfass, R. Hannemann, C. Hofm
ann, T. Kaczorowski, J. Kostka, R. Marklein, K. Mayer, and A.
Pitsch, Inverse scattering with acoustic, electromagnetic, and
elastic waves as applied in nondestructive evaluation, in A.
Wirgin, ed., Waveﬁeld Inversion, Springer, New York, 1999,
pp. 59–118.
17. P. M. van den Berg and A. Abubakar, Contrast source inver
sion method: State of art, Progress in Electromagnetics Re
search, 2001, Vol. 34, Chap. 8, pp. 189–218.
18. A. Qing and C. K. Lee, A study on improving the conver
gence of the realcoded genetic algorithm for electro
magnetic inverse scattering of multiple perfectly conducting
cylinders, IEICE Trans. Electron. E85C(7):1460–1471
(2002).
19. R. V. McGahan and R. E. Kleinman, eds., Image reconstruc
tion using real data, special session, IEEE Anten. Propag.
Mag. 38(3):39–59 (June 1996).
ELECTROMAGNETIC INVERSE PROBLEMS 1215
20. R. V. McGahan and R. E. Kleinman, eds., Image reconstruc
tion using real data, second annual special session, IEEE An
ten. Propag. Mag. 39(2):7–32 (April 1997).
21. R. V. McGahan and R. E. Kleinman, eds., Image reconstruc
tion using real data, special session, IEEE Anten. Propag.
Mag. 41(1):34–51 (1999).
22. R. V. McGahan and R.E. Kleinman, eds., Image reconstruc
tion using real data, special session, IEEE Anten. Propag.
Mag. 41(2):20–36 (1999).
23. K. Belkebir and M. Saillard, eds., Special issue on testing in
version algorithms against experimental data, Inverse Prob
lems 17(6):1565–1701 (Dec. 2001).
24. R. F. Harrington, Field Computation by Moment Methods,
IEEE Press, New York, 1993.
25. A. Qing, Microwave imaging of parallel perfectly conducting
cylinders with transverse electric scattering data, J. Elect
romagn. Waves Appl. 15(5):665–685 (2001).
26. A. Roger, NewtonKantorovitch algorithm applied to an elec
tromagnetic inverse problem, IEEE Trans. Anten. Propag.
29(2):232–238 (1981).
27. J. H. Holland, Adaption in Natural and Artiﬁcial Systems,
Univ. Michigan Press, Ann Arbor, MI, 1975.
28. R. L. Haupt, An introduction to genetic algorithms for
electromagnetics, IEEE Anten. Propag. Mag. 37(2):8–15
(1995).
29. J. M. Johnson and V. RahmatSamii, Genetic algorithms in
engineering electromagnetics, IEEE Anten. Propag. Mag.
39(4):7–21 (1997).
30. D. S. Weile and E. Michielssen, Genetic algorithm optimiza
tion applied to electromagnetics: A review, IEEE Trans.
Anten. Propag. 45(3):343–353 (March 1997).
31. C. C. Chiu and P. T. Liu, Image reconstruction of a perfectly
conducting cylinder by the genetic algorithm, IEE Proc.
Microwave Anten. Propag. 143(3):249–253 (1996).
32. A. Qing, C. K. Lee, and L. Jen, Microwave imaging of parallel
perfectly conducting cylinders using realcoded genetic algo
rithm, J. Electromagn. Waves Appl. 13(8):1121–1143 (Aug.
1999).
33. A. Qing and C. K. Lee, Microwave imaging of a perfectly
conducting cylinder using realcoded genetic algorithm,
IEE Proc. Microwave Anten. Propag. 146(6):421–425 (Dec.
1999).
34. A. Qing, An experimental study on electromagnetic inverse
scattering of a perfectly conducting cylinder by using real
coded genetic algorithm, Microwave Opt. Technol. Lett.
30(5):315–320 (2001).
35. R. Storn and K. Price, Differential Evolution—a Simple and
Efﬁcient Adaptive Scheme for Global Optimization over Con
tinuous Spaces, TR95012, International Computer Science
Institute, March 1995 (ftp://ftp.icsi.berkeley.edu/pub/tech
reports/1995/tr95012.pdf).
36. K. A. Michalski, Electromagnetic imaging of circularcylin
drical conductors and tunnels using a differential evolution
algorithm, Microwave Opt. Technol. Lett. 27(5):330–334 (Dec.
2000).
37. A. Qing, Electromagnetic inverse scattering of multiple two
dimensional perfectly conducting objects by the differential
evolution strategy, IEEE Trans. Anten. Propag. 51(6):1251–
1262 (June 2003).
38. A. Qing, Electromagnetic inverse scattering of multiple per
fectly conducting cylinders by differential evolution strategy
with individuals in groups (GDES), IEEE Trans. Anten. Prop
ag. 52(5):1223–1229 (May 2004).
ELECTROMAGNETIC MATERIALS
P. K. CHOUDHURY
Multimedia University
Cyberjaya, Selangor, Malaysia
O. N. SINGH
Banaras Hindu University
Varanasi, India
1. INTRODUCTION
Greek philosophers Plato and Socrates were aware of the
polarities of lodestone, an iron ore, which could attract
iron. Pierre de Marricourt used this property of loadstone
to construct a compass to locate the North Pole. Gilbert
(1540–1603) postulated that Earth is like a giant spherical
magnet, and Kirchner (1601–1680) demonstrated that the
magnetic poles of a magnet are of equal strength. Isaac
Newton (1642–1727) made the ﬁrst attempt to determine
the law of force for a bar magnet, but the correct version
(i.e., the inverse square law) was determined in 1750 by
John Mitchell (1724–1793). In 1785 Charles de Coulomb
(1736–1806) used the same inverse square law for deter
mining the force between electric charges. Hans Christian
Oersted (1777–1851) discovered in 1820 that a current
carrying wire could deﬂect a compass, and later on Andre´
Ampe`re (1775–1836) discovered that two currentcarrying
wires exert force on each other. Then JeanBaptiste Biot
(1774–1862) and Fe´lix Savart (1791–1841) formulated
their famous Biot–Savart law, which quantiﬁes the force
between the currents.
The inverse square law for magnetism was more or less
established by the early nineteenth century. Benjamin
Franklin (1706–1790) proposed an experiment to postu
late the inverse square law for stationary electric charges,
similar to the law of gravitation. In succeeding years con
siderable amount of work was carried on by Cavendish (in
the early 1770s) and by Coulomb in 1785 to verify the in
verse square law for electric charges. Karl Frederick
Gauss (1777–1855) proposed the divergence theorem and
presented the famous Gauss law of electrostatics, specif
ically, that the integral of the normal component of the
electric ﬂux over a closed surface is directly related to the
net charge enclosed by that surface. In 1831 Michael Fara
day (1791–1867) discovered that a timevarying magnetic
ﬁeld produces an electric ﬁeld, and veriﬁed his law using
the battery invented in 1800 by Alessandro Volta (1745–
1827). In 1873, James Clerk Maxwell (1831–1879), a Scot
tish mathematician, published his famous treatise Elec
tricity and Magnetism, which uniﬁed not only electricity
and magnetism but also optics, and laid the foundation of
electromagnetics through his four equations. Excellent
descriptions of the groundwork of earlier researchers in
connection with the foundation of electromagnetics are
provided in Refs. 1–3.
On the application of electromagnetic (EM) ﬁelds, ma
terials react in variety of ways. The applied electric ﬁeld
would inﬂuence both free and bound electrons, whereas
the magnetic ﬁeld would change the orientation of tiny
1216 ELECTROMAGNETIC MATERIALS
atomic moments. However, these responses are mostly
linear (i.e., proportional to the applied ﬁeld) over a spec
iﬁed range of the ﬁeld. If the response to the ﬁeld is inde
pendent of the orientation of material with respect to the
direction of ﬁeld; the material is isotropic. By applying
timevarying EM ﬁelds to such linear, isotropic materials,
their response would greatly depend on the frequency of
the applied ﬁeld. A few examples of such linear, isotropic
materials are common dielectrics such as fused silica and
glass, and common conductors such as copper and alumi
num. Materials are classiﬁed as dielectric, conductor, and
magnetic on the basis of their dominant response; for ex
ample, ferromagnetic materials are mostly highly conduc
tive but are classiﬁed as magnetic materials because that
property is most significant and useful. In practice, all
materials have some response to magnetic ﬁelds but, ex
cept for ferromagnetic and ferrimagnetic, the response is
usually small.
2. DIELECTRIC MATERIALS
Dielectric materials do not possess free electric charges
and hence do not conduct electric current. The polar di
electric materials possess a dipole moment, but not the
nonpolar ones. In an external electric ﬁeld the molecules of
dielectric materials are deformed and an induced electric
moment appears. In some crystalline dielectric, spontane
ous orientation of the dipole moments occurs even in the
absence of electric ﬁeld, and these substances are called
ferroelectric materials (all ﬁelds are harmonic in time t).
For lossy nonmagnetic dielectric materials, permittivity
can be represented as
e ¼e
0
Àje
00
ð1Þ
In this equation both e
0
and e
00
are functions of frequency,
and for solid, liquid, and gaseous states, the various phys
ical phenomenons contributing to e
0
and e
00
essentially dif
fer [4–6].
It can be shown that polarized atoms or molecules in a
volume V present a dipole moment density deﬁned as
P
d
¼ lim
Dv!0
i
p
i
DV
Neglecting higherorder multipoles, this is identical to the
socalled polarization, which exists in the relationship be
tween D and E.
D¼e
0
EþP¼e
0
ð1þw
e
ÞE ð2Þ
where w
e
is the dielectric susceptibility. Here we have as
sumed that the polarization is linearly dependent on the
ﬁeld E. If N represents the number of like molecules per
unit volume, the induced polarization will have the form
P¼e
0
w
e
E¼NaE
l
¼NaOE ð3Þ
where a is the molecular polarizability, O is the ratio be
tween local ﬁeld E
l
acting on the molecule and the applied
ﬁeld E. It should be noted at this point that the local ﬁeld
E
l
differs from applied electric ﬁeld owing to the effect of
molecules in the surrounding region. The displacement
vector D is related to the applied electric ﬁeld E as
D¼eE¼e
0
e
r
E ð4Þ
Using Eqs. (2) and (4), the relative permittivity (e
r
¼e/e
0
)
can be written as
e
r
¼1þ
Nag
e
0
¼1 þw
e
ð5Þ
If the surrounding molecules have a spherically symmet
ric behavior, the factor g can then be written as
e
r
À1
e
r
þ2
¼
Na
3e
0
ð6Þ
This relationship is called the Mossotti–Clausius equa
tion. If the effects of frequency on the molecular polariz
ability a are included, this relationship is known to be
Debye equation.
Basically the molecular polarizability a consists of elec
tronic and ionic effects of molecules. The electronic effect
arises from the shift of the electron cloud (in each atom)
from its positive molecules, and the ionic part is due to the
displacement of positive and negative ions from their neu
tral positions. However, if the individual molecules have
permanent dipole moments, these also contribute to the
molecular polarizability, and the applied electric ﬁeld
tends to align these permanent dipoles. Thus the total
molecular polarizability can be written as
a
T
¼a
e
þa
i
þa
d
ð7Þ
with a
e
, a
i
and a
d
denote electronic, ionic, and permanent
dipole contributions, respectively.
According to the classical model of electronic polariza
tion, any displacement of the charge cloud from its central
ion produces a restoring force, and its interaction with
the inertia of the moving charge cloud produces a reso
nance. This phenomenon is the same as discussed for a
spring–mass system in mechanics. Similarly, the displace
ment of one ion from another produces resonances in
the ionic polarizability. The electronic polarizability a
e
is
given as
a
e
¼
ðe
2
=mÞ
ðo
2
0
Ào
2
Þ þjoG
ð8Þ
where G is a damping constant and o
0
is related to the
restoring force. This equation can be generalized to rep
resent all electronic and ionic resonant responses as
a
j
¼
F
j
ðo
2
j
Ào
2
Þ þjoG
j
ð9Þ
where F
j
measures the strength of the jth resonance; real
and imaginary parts of the equation contribute to e
0
ðoÞ and
ELECTROMAGNETIC MATERIALS 1217
e
00
ðoÞ, respectively. According to Debye model, e
0
ðoÞ can
take negative as well as positive values. The relationships
between e
0
ðoÞ and e
00
ðoÞ are known as Kramers–Kronig
relations [7], and are given by
e
0
ðoÞ ¼e
0
þ
2
p
P
_
1
0
o
0
e
00
ðo
0
Þ
ðo
02
Ào
2
Þ
do
0
ð10Þ
e
00
ðoÞ ¼ À
2o
p
P
_
1
0
½e
0
ðo
0
Þ Àe
0
ðo
02
Ào
2
Þ
do
0
ð11Þ
where P denotes that the principal value of the integral
that should be considered.
At optical frequencies one uses the refractive index
rather than permittivity, and this is also a complex quan
tity for materials with absorption. With m¼m
0
, the com
plex refractive index is given as
n
c
ðoÞ ¼n
r
ðoÞ Àjn
i
ðoÞ ¼
e
0
ðoÞ Àje
00
ðoÞ
e
0
_ _
1=2
ð12Þ
3. CONDUCTORS, SEMICONDUCTORS,
AND SUPERCONDUCTORS
Ohm’s law is truly obeyed in the case of good conductors.
Also, for such materials the displacement current is neg
ligible as compared to the conduction current, and there
fore, the current and the electric ﬁeld are in phase. The
displacement current becomes increasingly important at
higher frequencies. For example, at microwave frequen
cies semiconductors may have comparable conduction as
well as displacement currents.
Let n
e
be the density of free electrons in a medium with
a background of ﬁxed positive ions having the same den
sity; this model may be considered for a conductor, semi
conductor, or an ionized gas (plasma) with different values
of parameters. It can be shown that the components of the
complex permittivity deﬁned by Eq. (1) will have the forms
e
0
¼e
1
À
n
e
e
2
mðn
2
þo
2
Þ
ð13Þ
e
00
¼
n
e
e
2
n
omðn
2
þo
2
Þ
ð14Þ
where e
1
represents the effect of bound electrons of the
positive ion background, n is the collision frequency, and e
and m, respectively, are the electronic charge and mass. It
is to be noted that ðn
e
e
2
=mnÞ is the lowfrequency conduc
tivity s. In the case of materials with moderate to low
conductivities such as semiconductors, for microwave and
millimeterwave frequencies where o
2
bn
2
, the second
term in Eq. (13) is negligible. Thus, the complex permit
tivity ultimately becomes
e ¼e
1
Àj
s
o
ð15Þ
On the other hand, in the case of ionized gas with very low
density and negligible collision frequency, we can have
e
1
¼e
0
and e
00
¼0, and therefore, the permittivity takes
the form
e ¼e
0
1 À
o
p
o
_ _
2
_ _
ð16Þ
where o
p
is called the plasma frequency, and is given as
o
p
¼
n
e
e
2
e
0
m
_ _
1=2
ð17Þ
Thus, we observe from Eq. (16) that the permittivity
is negative for all frequencies below the plasma frequen
cy. Therefore, for ooo
p
, the intrinsic wave impedance
Z ( ¼O(m/e)) becomes imaginary. As such, when a wave
with ooo
p
incidents from free space on a region of ionized
gas, the wave is reﬂected. The o–b diagram for TEM
waves propagating in an ionized gas is shown in Fig. 1.
For perfect conductors, there is no electric ﬁeld at any
frequency, and also, there is then no timevarying mag
netic ﬁeld. A truly static magnetic ﬁeld is unaffected by
conductivity of any value. For a metallic electrode, the
‘‘perfect conductor’’ approximation ensures uniform po
tential over its surface. A perfect conductor can be ap
proximated as a conductor in which the collision frequency
approaches zero, and this was considered to be the correct
model for a superconductor in the period between its dis
covery (in 1911) by H. K. Onnes and the experiment (in
1933) by W. Meissner and Ochsenfeld [8]. A collision con
ductor can be modeled as a dense plasma, and for fre
quencies below the plasma frequency, the ﬁelds inside the
conductor can exist only near the interface where they are
excited. For extremely pure metallic crystals, very low
collision frequencies may occur at temperature near the
absolute zero. The ideal collisionless conductor excludes
Free
space
Ionized
gas
ω
0 4o
p
/ c 3o
p
/ c 2o
p
/ c o
p
/ c
4o
p
3o
p
2o
p
o
p
[
Figure 1. The o–b diagram for a plane wave propagating
through an ionized gas.
1218 ELECTROMAGNETIC MATERIALS
timevarying ﬁelds from the interior. However, the exper
iments conducted by Meissner and Ochenfeld on a super
conductor showed that constant magnetic ﬁelds are also
excluded from the interior. In 1935 London showed that
this result could be obtained if
=ÂJ¼ ÀLB ð18Þ
with L¼n
e
e
2
=m at T¼0. Substituting (18) in Maxwell’s
curl H equation, and neglecting the displacement current,
one ﬁnally obtains
=
2
BÀmLB¼0 ð19Þ
This last equation can be applied for the ﬁelds inside the
surface of a superconductor by considering a halfspace of
superconductor ﬁlling zZ0 (Fig. 2). If we assume that only
the x component of B exists with its variation along the z
direction only, then Eq. (19) ﬁnally becomes
d
2
B
x
dz
2
À
B
x
l
2
L
¼0 ð20Þ
where l
L
¼ðm=mn
e
e
2
Þ
1=2
at T¼0; l
L
is called the London
penetration depth. The solution of Eq. (20) will be of the
form
B
x
¼B
x0
e
Àz=l
L
ð21Þ
Experimentally determined values of penetration depths
typically lie in the range 50–200 nm.
The phenomenon of superconductivity requires that
the material be held at temperatures near absolute
zero. Superconductors are mostly metallic elements,
compounds, or alloys that make transitions into the su
perconducting state at critical temperatures below B23 K.
Among the others, niobium has the highest critical tem
perature about 9.2 K. Compounds of niobium with higher
critical temperatures are widely used in many applica
tions. A new family of oxidebased superconductors was
discovered in 1986, which have much higher critical tem
peratures about 125K [9]. However, the oxidebased su
perconductors are highly anisotropic as compared with
metallic superconductors.
Superconductive transmission lines are nearly nondis
persive in nature. Losses occur in superconductors at non
zero frequencies and nonzero temperatures (unlike the
case for the ideal collisionless conductor) because the con
duction electrons are not in the superconducting state and
the penetrating ﬁelds can cause them to have collisions,
resulting in heat dissipation. For frequencies throughout
the microwave range, these losses are much smaller than
in normal metals such as copper.
4. ANISOTROPIC DIELECTRICS
For isotropic dielectric materials, the quantities permit
tivity and permeability are scalar, and may be nonlinear
and frequencydependent. In fact, every permittivity is
frequencydependent because of the Kramers–Kronig re
lations. However, in some frequency bands, it may be pos
sible to ignore the frequency dependence. Materials that
respond differently to ﬁelds with different orientations are
termed anisotropic, and the property of anisotropy may be
in the response to either the electric or magnetic ﬁeld. The
anisotropy in the electric ﬁeld response is represented by a
matrix permittivity, which is an array of nine scalar quan
tities that may be nonlinear and frequencydependent.
Similarly, the anisotropy in the magnetic ﬁeld response is
represented by a matrix permeability [10,11]. Materials
that have both the permittivity and the permeability are
anisotropic, and are rarely found.
For anisotropic dielectrics, the relationship between D
and E [12–14] is given as
D
x
¼e
11
E
x
þe
12
E
y
þe
13
E
z
ð22aÞ
D
y
¼e
21
E
x
þe
22
E
y
þe
23
E
z
ð22bÞ
D
z
¼e
31
E
x
þe
32
E
y
þe
33
E
z
ð22cÞ
which may be presented in matrix form as
D
x
D
y
D
z
_
¸
¸
_
_
¸
¸
_
¼
e
11
e
12
e
13
e
21
e
22
e
23
e
31
e
32
e
33
_
¸
¸
_
_
¸
¸
_
E
x
E
y
E
z
_
¸
¸
_
_
¸
¸
_
ð23Þ
This equation can be written in a more compact form as
½D ¼½e½E ð24Þ
B
0
(∂B
x
/ ∂t)
z = 0
z
x
Collisionless
conductor
Superconductor
Figure 2. Fields at the surface of a halfspace of collisionless
conductor or superconductor.
ELECTROMAGNETIC MATERIALS 1219
For physically real materials, the permittivity matrix is
Hermitian, which is deﬁned as
e
ij
¼e
Ã
ji
ð25Þ
However, the permittivity is Hermitian only if a material
is lossless (no physical material is lossless, by virtue of
causality). For lossless crystals, Eq. (25) implies real and
symmetric matrices, which can be diagonalized by a rota
tion of coordinates. Thus, Eqs. (22) reduce to the form
D
x
¼e
11
E
x
; D
y
¼e
22
E
y
D
z
¼e
33
E
z
ð26Þ
where e
11
, e
22
, and e
33
are called as principal permittivities.
5. ELECTROOPTIC MATERIALS
Some certain materials show a change in permittivity on
application of an electric ﬁeld. If the change in permit
tivity is directly proportional to the applied electric ﬁeld,
this is known as Pockels effect. Certain noncentrosymmet
ric solids [13,15], such as lithium niobate and gallium ars
enide, are examples of materials showing the Pockels
effect. On the other hand, if the change in permittivity is
directly proportional to the square of the applied electric
ﬁeld, this is known as Kerr effect. Some liquids such as
carbon disulﬁde and nitrobenzene, and some centrosym
metric solids present the Kerr effect. Both the effects are,
however, relatively small for practical values of externally
applied electric ﬁeld.
The electrooptic effect can be described by considering
the change in permittivity element e
ij
on application of the
electric ﬁeld component E
k
. However, the effect is deﬁned
in terms of the reciprocal matrix. Let us deﬁne the matrix
1
n
2
_ _
¼
e
e
0
_ _
À1
ð27Þ
Now, when the electric ﬁeld is applied, the change in an
element of this matrix is
D
1
n
2
_ _
ij
¼
3
k ¼1
r
ijk
E
k
ð28Þ
where r is the distance from an arbitrary origin, where the
subscripts i,j,k each range over the three spatial coordi
nates x,y,z, denoted by the integers 1,2,3, respectively.
Since the matrices [e], and hence (1/n
2
), are symmetric,
one may implement contracted notations as 111, 222,
333, 23 ¼324, 13 ¼315, 12 ¼216, so that Eq. (27)
may be written in another form as
D
1
n
2
_ _
ij
¼
3
k ¼1
r
pk
E
k
ð29Þ
Here the subscript k has values ranging from 1 to 3 and
the subscript p ranges from 1 to 6. Table 1 shows the val
ues of r
pk
[as in Eq. (29)] for a few anisotropic crystals [15].
6. MAGNETIC MATERIALS
Under the inﬂuence of an external magnetic ﬁeld the in
ternal dipole moments align themselves. Also, a magnetic
moment is induced in the magnetic material. As such, in
the presence of a magnetic material, the resultant mag
netic ﬂux density will be different from its value in free
space. The effect of magnetization can be studied by in
corporating the equivalent volume current density, J
m
into the basic curl equation for the magnetic induction
B, given as
=ÂB¼m
0
J ð30Þ
We thus have
1
m
0
=ÂB¼JþJ
m
ð31Þ
where
J
m
¼=ÂM ð32Þ
It is assumed that an externally applied magnetic ﬁeld
causes the atomic circulating currents to align with it,
thereby magnetizing the material. In Eq. (32) M is the
magnetization vector which measures the strength of
magnetizing effect. From Eqs. (31) and (32), we ﬁnally get
=Â
B
m
0
ÀM
_ _
¼J ð33Þ
We now deﬁne the magnetic ﬁeld intensity H, which is a
new fundamental ﬁeld quantity, such that
H¼
B
m
0
ÀM ð34Þ
From Eqs. (33) and (34), we obtain the new equation
=ÂH¼J ð35Þ
where J is the volume density of free current. Equation
(35) along with the equation =
.
B¼0 (the divergence
equation for the magnetic induction) form the two funda
Table 1. Electrooptic Properties of a Few Materials
Materials
n
o
(l
0
¼550nm)
N
e
(l
0
¼550nm)
r
pk
( Â10
À12
mV
À1
)
KDP 1.51 1.47 r
41
¼8.6
r
63
¼10.6
GaAs (10.6mm) 3.34 r
41
¼1.6
LiNbO
3
2.29 2.20 r
33
¼30.8
r
13
¼8.6
r
22
¼3.4
r
42
¼28.0
BaTiO
3
2.437 2.365 r
33
¼23.0
r
13
¼8.0
r
42
¼820.0
1220 ELECTROMAGNETIC MATERIALS
mental governing differential equations for magnetostat
ics [16–18]. Clearly, the factor permeability of the medium
does not appear in both of these equations. For linear and
isotropic materials the magnetization is directly propor
tional to the magnetic ﬁeld intensity, and therefore
M¼w
m
H ð36Þ
where w
m
is a dimensionless quantity called the magnetic
susceptibility. Using Eqs. (34) and (36), we obtain
B¼m
0
ð1þw
m
ÞH¼m
0
m
r
H¼mH ð37Þ
or H¼ð1=mÞB, where
m
r
¼1 þw
m
¼
m
m
0
ð38Þ
m
r
is again a dimensionless quantity called as the relative
permeability of the medium. The parameter m¼m
0
m
r
is
said to be the absolute permeability (or the permeability)
of the medium, and the unit of m is henries per meter.
Linear, isotropic, and homogeneous media have constant
values of m and w
m
. At this point, it is to be noted that the
value of permeability for most materials is very close to
that of free space, namely, m
0
. For ferromagnetic materials
such as Fe, Ni, and Co, the value of m
r
could be very large
(e.g., 50–500 and Z10
6
for certain types of alloys). In fact,
the value of permeability depends not only on the magni
tude of magnetizing ﬁeld H but also on the previous his
tory of the material.
6.1. Behavior of Magnetic Materials
The classiﬁcation of magnetic materials can be roughly
made according to their relative permeability m
r
. In gen
eral, magnetic materials are classiﬁed into three main
groups; diamagnetic (m
r
r1; i.e., w
m
is a very small nega
tive number), paramagnetic (m
r
Z1; i.e., w
m
is a small pos
itive number), and ferromagnetic (m
r
b1; i.e., w
m
is a large
positive number) [19]. As such, the macroscopic magnetic
property of a linear, isotropic medium can be described by
deﬁning the magnetic susceptibility w
m
. Generally, all ma
terials have some response to magnetic ﬁelds. Except for
ferromagnetic and ferrimagnetic (a special class of ferro
magnetics) materials, the magnetic response is usually
very weak. The magnetic response can either decrease or
increase the ﬂux density for a given magnetizing ﬁeld H. If
B is decreased, the material is said to be diamagnetic, and
if it is increased, it is paramagnetic. However, a knowledge
of quantum mechanics is essentially required in order to
deal with the microscopic magnetic phenomena. A quali
tative description of the behavior of various types of mag
netic materials is given below, which is based on the
classical atomic model.
The property of diamagnetism arises primarily from
the orbital motion of electrons within an atom, and it is
present in almost all materials. However, in most of the
materials the effect of diamagnetism is too weak to be of
any practical significance. In the absence of an externally
applied magnetic ﬁeld, the atoms of diamagnetic materials
have vanishing net magnetic moment (due to the orbital
and spin motions of electrons). When an external magnet
ic ﬁeld is applied to diamagnetic materials, it produces a
force on the orbiting electrons, causing a perturbation in
angular velocities, which results into a net magnetic mo
ment. Now the induced magnetic moment always tends to
oppose the externally applied ﬁeld (according to Lenz law),
and therefore, the magnetic ﬂux density is ultimately re
duced. The macroscopic effect of this process is analogous
to that of a negative magnetization that can be described
by a negative magnetic susceptibility w
m
. Thus, the dia
magnetic response of atoms arises from the changes of
electron orbits when an external magnetic ﬁeld is applied.
According to Faraday’s law, an electric ﬁeld can be pro
duced by a changing magnetic ﬁeld. That ﬁeld induces
currents which, in turn, produce a magnetic ﬁeld that op
poses the change. The response of electron orbits in an
atom is of this kind. However, this kind of effect is gener
ally extremely small. The effect produces fractional chang
es in m from m
0
by only 10
À8
–10
À5
. The value of w
m
for
most of the known diamagnetic materials (Bi, Cu, Pb, Hg,
Ge, Au, silver, etc.) is of the order À10
À5
. Diamagnetic
materials do not exhibit permanent magnetism, and the
induced magnetic moment disappears when the applied
ﬁeld is withdrawn.
On the other hand, in some materials the magnetic
moments due to the orbiting and spinning electrons do not
cancel each other completely, and therefore, the atoms and
molecules possess a net magnetic moment. In such mate
rials the diamagnetic response is usually obscured. When
an external magnetic ﬁeld is applied, tiny molecular mag
netic moments tend to align in the direction of applied
ﬁeld, increasing the magnetic ﬂux density. In addition, a
very weak diamagnetic effect is also caused. The macro
scopic effect of this process is equivalent to that of a pos
itive magnetization that can be described by a positive
magnetic susceptibility. Materials with this kind of be
havior are said to be paramagnetic. Generally, such ma
terials (Al, Mg, Ti, and W) have very small positive values
of w
m
, of the order of 10
À5
.
The property of paramagnetism primarily arises be
cause of the magnetic dipole moments of spinning elec
trons. The externally applied ﬁeld acts on molecular
dipoles to align those, and the alignment forces are coun
teracted by the deranging effects of thermal agitation. The
property of paramagnetism is temperaturedependent,
unlike the case for diamagnetism (it is independent).
Therefore, paramagnetic effects are stronger at lower tem
peratures because there is less thermal agitation.
If there are induced magnetic dipoles in the atoms or
there is a nonzero average alignment of natural dipoles, in
that case, the term dipole density can be deﬁned as
M¼ lim
Dn!0
i
m
0i
DV
ð39Þ
This is identical to the socalled magnetization that enters
the relation between B and H as in Eq. (34). It can be
shown that the magnitude of magnetization can be ex
pressed in terms of molecular magnetic ﬁeld H
i
acting on
ELECTROMAGNETIC MATERIALS 1221
the atomic dipoles and the absolute temperature T [20] as
M¼Nm
0
coth
m
0
m
0
H
i
k
B
T
À
k
B
T
m
0
m
0
H
i
_ _
ð40Þ
where k
B
is Boltzmann’s constant, N is the number of di
poles per unit volume, and m
0
is the natural dipole mo
ment. Equation (40) is derived considering the statistical
distribution of the orientation of dipoles. In such materi
als, the direction of M is the same as that of the molecular
ﬁeld H
i
; that is, they are aligned parallel. If the external
magnetic ﬁeld H is not too strong and the temperature is
not too low, the terms in parentheses in Eq. (40) may be
approximated by the ﬁrst term only in its series expan
sion. In this case the magnetization is linearly related to
the applied ﬁeld:
M¼w
m
H ð41Þ
w
m
ﬃ
Nm
0
m
2
0
3k
B
T
ð42Þ
At room temperature, the values of w
m
for a variety of
materials are of the order of 10
À5
.
6.2. Materials with Residual Magnetization
If the temperatures of paramagnetic substances are re
duced below a certain value, which depends essentially on
materials, the magnetization M would be sufﬁcient en
ough to produce the ﬁeld necessary to hold the dipoles
aligned even after the removal of the external ﬁeld. The
molecular ﬁeld produced by the magnetization is given
as [4]
H
i
¼kM or M¼
1
k
H
i
ð43Þ
where k is the factor measuring the interaction of neigh
boring dipoles (it must be on the order of 1000). We see
that there exists another relationship between H
i
and M,
as observed in Eq. (40). By equating the Eqs. (43) and (40),
one may obtain the condition for spontaneous magnetiza
tion. The temperature below which a material exhibits
spontaneous magnetization is called its Curie tempera
ture.
For ferromagnetic materials, the effect of magnetiza
tion can be many orders of magnitude larger than that of
paramagnetics. The effect of ferromagnetism can be ex
plained on the basis of the model of magnetized domains,
according to which ferromagnetic materials (e.g., Co, Ni,
Fe) are composed of several tiny domains with their linear
dimensions ranging from a few micrometers to B1 mm.
Each domain contains B10
15
atoms, which are fully mag
netized; that is, they contain aligned magnetic dipoles
(even in the absence of an externally applied magnetic
ﬁeld) resulting from spinning electrons. In a molecular
domain, strong coupling forces exist between the magnetic
dipole moments (of the atoms). As a result, the different
dipole moments are aligned parallel to each other. There
exists a transition region between adjacent domains,
which is about 100 atoms thick. This region is called the
domain wall [21]. For ferromagnetic materials, when
these are not magnetized, the magnetic moments of the
adjacent domains have different directions (Fig. 3), and
the random nature of orientations of these domains re
sults in no net magnetization.
In ferromagnetic materials, under the inﬂuence of ex
ternal magnetic ﬁeld, the walls of domains having mag
netic moments in the direction of applied ﬁeld move in
such a way that the volumes of those domains grow, re
sulting thereby an increase in magnetic ﬂux density.
When the applied ﬁeld is weak (up to point P
1
in Fig. 4),
the movements of domain walls are reversible. However,
for stronger applied ﬁelds (i.e., after point P
1
is crossed),
domain wall movements are no longer reversible. Also,
domains rotate toward the direction of the externally ap
plied ﬁeld. If the applied strong ﬁeld is reduced to zero at
point P
2
, the B–H relationship will no longer follow the
solid curve path P
2
P
1
O, but will decrease from P
2
to P
0
2
instead, and the path followed will be along the broken
lines as shown in Fig. 4. This phenomenon is called hys
teresis, which is derived from a Greek word meaning ‘‘to
Figure 3. Domain structure of a polycrystalline ferromagnetic
specimen.
P
1
O
P
2
H
c
B
r
H
B
P
3
P ′
P ′
3
2
Figure 4. Hysteresis loop for ferromagnetic material.
1222 ELECTROMAGNETIC MATERIALS
lag’’; in this case we observe the lagging of magnetization
behind the applied ﬁeld. When the applied ﬁeld strength
ens (i.e., from point P
2
to P
3
in Fig. 4), domain rotation as
well as domain wall motion will cause a complete align
ment of the magnetic moments with the externally applied
ﬁeld. At this point, the ferromagnetic magnetic material is
said to be saturated. The curve OP
1
P
2
P
3
(on the B–H
plane) is called the normal magnetization curve. If, after
reaching point P
3
, the applied magnetic ﬁeld is reduced to
zero, the magnetic ﬂux density does not fall to zero, but
assumes the value at B
r
instead. This value of B is called
as the residual ﬂux density, and its unit is Webers per
square meter (Wb/m
2
). The value of B
r
depends on the
maximum intensity of the applied ﬁeld. The property of
the existence of a residual ﬂux density in ferromagnetic
materials is exploited in transforming them to permanent
magnets.
In order to render the magnetic ﬂux density of a ferro
magnetic material zero, it is necessary to apply a magnetic
ﬁeld intensity H
c
, called the coercive force (or the coercive
ﬁeld intensity), in the opposite direction. The unit of H
c
is
A/m, and its value also depends on the maximum value of
the intensity of the applied ﬁeld. It can be seen from the
Fig. 4 that the relationship between B and H ﬁelds is non
linear. Therefore, from the relationship B¼mH, the per
meability m is not a constant, and it is a function of the
magnitude of magnetizing ﬁeld H. The value of m also de
pends on the history of magnetization of the material.
Some important applications of ferromagnetic materi
als include permanent magnets, signal recording tapes
and disks, magnetic shielding, and cores for electromag
nets, transformers, electric motors, and generators [21].
These materials can be roughly divided into two different
categories called hard and soft; the designations refer to
permanence of magnetization when the external applied
ﬁeld is removed, with hard materials retaining a strong
magnetization. Ferromagnetic applications require that
the material should acquire a large magnetization for a
very small magnetic ﬁeld. As such, these materials should
have tall and narrow hysteresis loops. The properties of
some technically important materials for permanent mag
nets are listed in Table 2 [22]. Generally, the coercive ﬁeld
intensity of hard ferromagnetic materials (such as Alnico
alloys) would be about Z10
5
A/m, whereas that for soft
ferromagnetic materials, it is usually around r50 A/m.
The properties of some important soft materials are listed
in Table 3 [22,23].
It may thus be inferred that ferromagnetism is the re
sult of strong coupling effects between the magnetic dipole
moments of atoms in a domain. The atomic spin structure
of a ferromagnetic material is illustrated in Fig. 5a. If the
temperature of a ferromagnetic material is raised to such
a value that the thermal energy exceeds the coupling en
ergy, the tiny magnetized domains loose their organized
orientation, thus loosing the property of magnetization. As
Table 2. Properties of a Few Hard Magnetic Materials
Materials Composition (%)
Remanence
B
r
(T)
Coercive
Force H
c
(A/m)
Iron Bonded powder 0.6 0.0765
Iron–cobalt Bonded powder 1.08 0.0980
Alnico V Ni 14%;Al 8%,Co 24%;
Cu 3%,Fe 51%
1.27 0.0650
Ticonol II Ni 15%;Al 7%;Co 34%;
Cu 4%;Ti 5%;Fe 35%
1.18 0.1315
Table 3. Properties of a Few Soft Magnetic Materials
Materials Composition (%)
Initial
Permeability (m
r
)
0
Maximum
Permeability (m
r
)
max
Coercive Force
( Â10
4
) (A/m)
Saturation Induction
(T)
Iron Commercial (Fe 99%) 2 Â10
2
6Â10
3
0.9 2.16
Iron Pure (Fe 99.9%) 2.5Â10
4
3.5Â10
5
0.01 2.16
Hypersil Fe 97%;Si 3% 9 Â10
3
4Â10
4
0.15 2.01
78 permalloy Ni 78%;Fe 22% 4Â10
3
10
5
0.05 1.05
Mumetal Fe 18%;Ni 75%;Cu
5%;Cr 2%
2 Â10
4
10
5
0.05 0.75
Supermalloy Fe 15%;Ni 79%;Mo
5%;Mn 0.5%
9 Â10
4
10
6
0.004 0.8
Cryoperm Usable at cryogenic
temp.
6.5Â10
4
(a)
(b)
(c)
Figure 5. Schematic atomic spin structures for (a) ferro
magnetic, (b) antiferromagnetic, and (c) ferrimagnetic materials.
ELECTROMAGNETIC MATERIALS 1223
such, a critical value of temperature can be deﬁned, above
which a ferromagnetic material behaves like a paramag
netic material; the critical temperature is known as the
Curie temperature. For most of the ferromagnetic materi
als, the Curie temperature lies between a few hundred to a
thousand degrees Celsius. For iron, the Curie temperature
is about 7701C.
In the case of very small crystals, there are factors such
as crystal shape, energetically preferred crystal direc
tions, and the magnetic history of the sample that lead
to nonzero net magnetization [24]. In the presence of an
applied ﬁeld the existence of domain walls in thin ﬁlms is
exploited for the socalled bubble memory devices for com
puter memories [25].
Some elements, such as chromium and manganese,
also have strong coupling forces between the atomic di
pole moments. However, their coupling forces produce an
tiparallel alignments of electron spins (Fig. 5b), and the
spins alternate in direction from atom to atom, resulting
in no net magnetic moment. Such materials are referred to
as antiferromagnetics, and this property is also tempera
turedependent. An antiferromagnetic material, on heat
ing above its Curie temperature, exhibits random spin
directions, and the materials ultimately become paramag
netic.
Another class of magnetic materials that exhibit be
havior between ferromagnetism and antiferromagnetism
are called as ferrimagnetic materials. Here quantum
mechanical effects cause the directions of magnetic mo
ments in the ordered spin structure to alternate, and their
magnitudes to be unequal. As a result, a nonzero magnetic
moment exists for such materials (Fig. 5c). As such, in
ferrimagnetic materials, the neighboring dipoles are
aligned in an antiparallel arrangement, but different
types of atoms are present and the dipoles do not cancel.
As seen in Fig. 5c, because of the partial cancellation of
magnetic moments, the maximum magnetic ﬂux density
for ferrimagnetic materials is much lower than that for
ferromagnetics. Typically, its value is about 0.3 Wb/m
2
,
which is approximately onetenth the value of that for
ferromagnetic materials.
6.3. Ferrites
Ferrites are a subgroup of ferrimagnetic materials that
have low loss and strong magnetic effects at microwave
frequencies [26]. These materials have a particular type of
solid crystal structure made up of oxygen, iron, and an
other element such as lithium, magnesium, or zinc. These
mostly have the formula XOFe
2
O
3
, where X denotes a di
valent metallic ion such as Fe, Co, Ni, Mn, Mg, Zn, or Cd.
There are also some ceramiclike compounds with very low
conductivity as well as magnetic garnets which are used
as ferrites such as yttrium–iron–garnet (YIG; Y
3
Fe
5
O
12
).
Ferrites present relatively low losses at microwave fre
quencies. The dielectric constants of ferrites are relatively
high, and the anisotropic behavior of permeability comes
into existence when the material is subjected to a steady
magnetic ﬁeld [27].
In order to deﬁne the energy state of an atom, both the
orbital and spin quantum numbers must be speciﬁed. As
mentioned above, the electron spin generates a strong
magnetic moment. These magnetic moments are random
ly oriented in paramagnetic substances. However, a strong
coupling between spin magnetic moments of neighboring
atoms exists in ferromagnetics, antiferromagnetics, and
ferrites. This results in parallel or antiparallel alignment
of spin magnetic moments. The model of a spinning elec
tron is shown in Fig. 6. For a rotating body, the applied
torque T is the rate of change of angular momentum J:
dJ
dt
¼T ð44Þ
Since the magnetic moment m can be given as
m¼gJ ð45Þ
g being the gyromagnetic ratio, for a spinning electron, the
magnetic moment m can be related to the applied torque.
Now, the torque resulting from subjecting a magnetic mo
ment m to a magnetic ﬁeld B
i
is
T¼mÂB
i
ð46Þ
From Eq. (44), we would obtain
dm
dt
¼g
dJ
dt
¼gT¼gðmÂB
i
Þ ð47Þ
In a magnetic material, the magnetic ﬁeld B
i
acting on a
molecule can be represented as
B
i
¼m
0
HþM ð48Þ
where m
0
is the freespace permeability and M is the mag
netization vector. The vector M may also be treated as
magnetic dipole density, and can be given as N
0
m with N
0
as the effective number density. The relationship between
the intensity of magnetic ﬁeld H and the external applied
ﬁeld B depends on the shape of the ferrite body. The ma
trices [B] and [H] are related as
½B ¼½m½H ð49Þ
J
m
H
0
Figure 6. Model of a spinning electron in a magnetic ﬁeld.
1224 ELECTROMAGNETIC MATERIALS
or
B
x
B
y
B
z
_
¸
¸
_
_
¸
¸
_
¼
m
11
m
12
0
m
21
m
22
0
0 0 m
0
_
¸
¸
_
_
¸
¸
_
H
x
H
y
H
z
_
¸
¸
_
_
¸
¸
_
ð50Þ
Here the permeability components are
m
11
¼m
22
¼m
0
1þ
o
0
o
M
o
2
0
Ào
2
_ _
ð51Þ
m
12
¼m
Ã
21
¼j
mo
0
o
M
o
2
0
Ào
2
ð52Þ
with
o
M
¼ Àm
0
gM
0
and o
0
¼ Àm
0
gH
0
ð53Þ
where M
0
is the saturation magnetization and H
0
is the
strength of the DC magnetic ﬁeld. It is to be noted that the
value of magnetic intensity H is averaged over the space of
many molecules within the material, and therefore, the mag
netic ﬁeld within the ferrite differs from the applied ﬁeld.
It can be shown that linearly polarized waves cannot
propagate through ferrites, and therefore, it is not a nor
mal mode of propagation for the medium. In reality, cir
cularly polarized waves are normal modes for ferrites;
media having this property are called gyrotropic media. In
ferrite media, the propagation constants for clockwise
and counterclockwise polarized waves are different.
In 1845, Michael Faraday observed that when optical
waves are allowed to propagate through a dielectric with
an applied magnetic ﬁeld, there occurs a rotation of plane
of polarization of the optical wave, and the angle of rota
tion y can be given as
y / H
0
l ð54Þ
where H
0
is the DC magnetic ﬁeld strength and l is the
interaction length. The rotation of the plane of polariza
tion of optical waves is called Faraday rotation, and this
effect is very strong in ferrites. Equation (54) can be re
written as
y ¼VH
0
l ð55Þ
where the constant of proportionality V is known as the
Verdet constant.
Faraday rotation is also observed even in ordinary
dielectrics (e.g., silica), but the angle of rotation y is rel
atively small, and the value of Vis about 0.014 min/Oe Á cm
at the operating wavelength 600nm. In silica, the angle of
rotation decreases at longer wavelengths approximately
as l
À2
.
There are waveguides and stripline devices that em
ploy ferrites, and these devices have found enormous use
in microwave applications. Waveguide devices operate on
the Faraday effect principle. The operation of stripline de
vices is based on the asymmetry caused by an applied
steady magnetic ﬁeld. In waveguide devices, according to
the principle of Faraday rotation, for both the positive and
the negative traveling waves (as shown in Fig. 7), the lin
early polarized wave is oriented in the same direction rel
ative to the direction of the DC magnetic ﬁeld. The gyrator
[28] is the simplest microwave device that uses the effect
of Faraday rotation, and it produces a phase shift of 1801s
in one direction but no shift in the opposite direction.
The absorption isolator is another Faraday rotation de
vice where the wave traveling to one side is appreciably
attenuated (about 30 dB) and that traveling to the other
side suffers only a little attenuation (say, B0.5dB). Cir
culators also exploit the principle of Faraday rotation, and
these devices transmit a wave from one guide to the other.
These devices employ ferrite rods, and can also be used as
switches or modulators (by controlling the ﬁeld that mag
netizes the ferrite rod). The dimensional structure of fer
rite rods used in Faraday rotation devices is designed so
that the required rotation is easily achieved with reason
able magnetic ﬁelds, and the wave reﬂections are mini
mized with maximum powerhandling capacity. These
devices are useful only for low power.
The stripline Yjunction circulator is one among strip
line devices, and it can be used as a switch or isolator.
Such a device employs ferrite disks, and a DC magnetic
ﬁeld is applied perpendicular to the disks. Standingmode
patterns are formed in ferrite disks; a detailed description
[29,30] of the operating principle is beyond the scope of
this article.
7. BIANISOTROPIC MATERIALS
Since the mid1990s, intensive research on new artiﬁ
cially engineered physical systems has been reported
Direction of propagation
H
0
Direction of propagation
H
0
Figure 7. Rotation of waves in magnetized ferrite rods.
ELECTROMAGNETIC MATERIALS 1225
emphasizing the different aspects of nanostructure and
mesoscopic physics as well as physics of microwave com
posite materials. A good example of an artiﬁcially engi
neered physical system would be a small ferrite disk with
the magnetostatic wave oscillation spectrum. The mag
netostatic wave oscillations in a small ferrite disk
resonator are characterized by a discrete spectrum of en
ergy levels. Reports have been made on the different
mechanisms of excitation of magnetostatic wave oscilla
tions in small ferrite specimens by external RF magnetic
ﬁelds.
Since the mid1990s, rapid advancements in materials
science enabled the fabrication of complex materials in
the form of thin ﬁlms and particulate composites, thus
rendering the term bianisotropy quite fashionable
[31,32]. In bianisotropic materials the electromagnetic
ﬁeld vectors are coupled in a pattern more complex than
that in the usual type of anisotropy. Artiﬁcial bianisotro
pic materials, because of their potential usefulness for a
variety of applications, have attracted considerable at
tention of the R&D community devoted to research on the
electromagnetics of bianisotropic materials [33]. Artiﬁ
cial, isotropic chiral materials have been made (although
quite useless for RAM purposes). Artiﬁcial anisotropic
materials with a wide variety of properties have also been
made. Their properties are local when the longwave
length approximation is valid. With the use of compos
ites, a wide variety of novel response properties become
achievable.
There are several bianisotropic materials that possess
the property of chirality [34], specifically, optical activity
or handedness. Also, many bianisotropic materials have
the property of nonreciprocity, meaning that they exhibit
directiondependent responses. However, all materials
with directiondependent response properties are not non
reciprocal. Thus, bianisotropic materials have a wide va
riety of response properties owing to their complexity in
structure, and have found many different applications in
the areas of microwaves, electrical engineering, and op
tics. Sculptured thin ﬁlms which present nanoengineered
response properties, have been developed [35–40]. Many
new composite materials with nonlinear optical applica
tions are now emerging.
7.1. Linear Constitutive Equations
The reciprocity theorem due to Lorentz states that ﬁelds
E
a
, H
a
and E
b
, H
b
from two different sources at the same
frequency satisfy the condition
=ðE
a
ÂH
a
ÀE
b
ÂH
b
Þ ¼0 ð56Þ
This theorem is easily veriﬁed for isotropic media by sub
stituting Maxwell’s equations in complex form. It can also
be shown that this holds for an anisotropic medium pro
vided the permittivity and permeability matrices are sym
metric. The reciprocity theorem does not hold if the
matrices are asymmetric.
The theory of reciprocal bianisotropic materials can be
described on the basis of following equations [41]
D¼
"
" ee" ee
.
Eþj
"
" aa" aa
.
H ð57Þ
B¼
"
" mm" mm
.
HÀj
"
" aa" aa
T
.
E ð58Þ
where the electric and the magnetic ﬁelds are coupled.
These equations are valid in both the time and the fre
quency domains, and are called constitutive relations as
they contain information about the nature of the material
under consideration. It is to be noted that the constitutive
relations relate the matterderived ﬁelds D and H to the
basic ﬁelds E and B in any material medium. Both per
mittivity and permeability of a reciprocal material are
symmetric.
In Eqs. (57) and (58), the tensor
"
" aa" aa measures the optical
activity (or chirality) of the medium, the superscript ‘‘T’’
represents the transpose operation, and j stands for the
imaginary part. The fundamental properties of the mate
rial tensors
"
" ee" ee (the permittivity),
"
" mm" mm (the permeability), and
"
" aa" aa
have been determined [41–43]. Possible alternative cons
titutive relations have also appeared in the literature [44].
However, different constitutive equations are actually
equivalent (after appropriate redefinition of the ﬁeld vec
tors). Equations (57) and (58) provide the most rational
and convenient way to describe the effects of chirality,
especially in nonuniform media.
7.2. Electromagnetics of Bianisotropic Materials
Investigators have presented the study of spherical waves
in chiral bianisotropic materials with the scalar dielectric
permittivity and dyadic chirality parameter. The electro
magnetic behavior of bianisotropic materials with uniax
ial symmetry [45,46] have been discussed in the literature
with the appropriate constitutive equations as [45]
D¼
"
" ee" ee
.
EþjðÀa
"
"
II
"
II
t
þk
"
"
JJ
"
JJÞ
.
H ð59Þ
B¼
"
" mm" mm
.
Hþjða
"
"
II
"
II
t
þk
"
"
JJ
"
JJÞ
.
E ð60Þ
In these equations, the dielectric permittivity
"
" ee" ee and the
magnetic permeability
"
" mm" mm are uniaxial diadics given as
"
" ee" ee ¼e
t
"
"
II
"
II
t
þe
n
" zz
0
" zz
0
ð61Þ
"
" mm" mm ¼m
t
"
"
II
"
II
t
þm
n
" zz
0
" zz
0
ð62Þ
where " zz
0
represents the unit vector along the geometric
axis and
"
"
II
"
II
t
¼ " xx
0
" xx
0
þ " yy
0
" yy
0
is the transverse unit dyadic.
Further, in Eqs. (59) and (60),
"
"
JJ
"
JJð ¼ " zz
0
Â
"
"
II
"
II
t
¼ " yy
0
" xx
0
À " xx
0
" yy
0
Þ
is the 901 rotator in the transverse (x–y) plane, a is the
chirality parameter as deﬁned in Eqs. (57) and (58), and
the parameter k measures the coupling of orthogonal
transverse electric and magnetic ﬁelds. Composite mate
rials governed by the constitutive relations (59) and (60)
can be fabricated by embedding small Oshaped metal
elements in isotropic dielectric, together with chiral
1226 ELECTROMAGNETIC MATERIALS
elements. Such materials are useful in microwave appli
cations.
Bianisotropic omega materials can be described by the
constitutive relations [47]
D¼
"
" ee" ee
.
Eþj
"
" kk" kk
em
.
H ð63Þ
and
B¼
"
" mm" mm
.
HÀj
"
" kk" kk
me
.
E ð64Þ
In these equations, the dielectric permittivity
"
" ee" ee and the
magnetic permeability
"
" mm" mm are diagonal dyadics given as
"
" ee" ee ¼e
0
ðe
xx
" xx
0
" xx
0
þe
yy
" yy
0
" yy
0
þe
zz
" zz
0
" zz
0
Þ ð65Þ
"
" mm" mm¼m
0
ðm
xx
" xx
0
" xx
0
þm
yy
" yy
0
" yy
0
þm
zz
" zz
0
" zz
0
Þ ð66Þ
If the equivalent electric dipoles of molecules (or stems of
the omega particles in composite materials) are all aligned
with the z axis and the magnetic dipoles are in the y di
rection (the loops lie in the (x–z) plane), then
"
" kk" kk
em
and
"
" kk" kk
me
can be given as
"
" kk" kk
em
¼k" zz
0
" yy
0
ð67Þ
"
" kk" kk
me
¼k" yy
0
" zz
0
ð68Þ
The combination of omegashaped layers with biased fer
rites provides good microwave applications. For optical
applications, chiral bianisotropic materials can be manu
factured as multilayered structures formed by anisotropic
layers [48].
Apart from linear bianisotropic materials, there are
nonlinear bianisotropic materials as well, the crystals of
which combine the practical advantages of liquid crystals
possessing spiral structures with those of nonlinear ma
terials [49,50]. The property of chirality facilitates fre
quencyselective and polarizationsensitive reﬂection
together with polarization rotation, whereas the nonlin
earity makes the parametric interaction and frequency
transformation effects possible.
8. MAGNETOSTRICTION
In the early 1840s James Prescott Joule observed that the
magnetization of an iron sample induces a change in the
length of the sample. Since then this phenomenon, known
as the Joule effect, is often exploited in magnetostrictive
actuators. The converse of this effect, in which the mag
netization of the sample can be changed on application of
stress, known as the Villari effect, and also at times re
ferred to as the magnetostrictive effect, is usually exploited
in magnetostrictive sensors. There also exists another
pair À Wiedemann effect and its converse Matteuci ef
fect, which correlate, respectively, twist in the sample to
the generation of the helical magnetic ﬁeld, and vice ver
sa. Both of these effects are used in devices and torque
sensors. The Joule, the Wiedemann, the Villari and the
Matteuci effects lead to two modes of operation of magne
tostrictive transducers: transformation of magnetic ener
gy into mechanical energy and of mechanical energy into
magnetic energy. The former is exploited in actuators for
generating motion or force, and also, in sensors for detect
ing the states of magnetic ﬁeld; the latter is used in sen
sors to detect motion or force, passive damping devices,
and devices for inducing change in the state of magnetic
ﬁelds in materials.
Magnetoelasticity (or magnetostriction) is a phenome
non that describes the interaction between magnetism
and elasticity in ferromagnetic materials. In magnetoelas
ticity, Hooke’s law of elasticity is violated, since an
additional expansion/contraction results from the reorien
tation of tiny magnetic domains. Thus in magnetostriction
phenomenon strain is caused by the magnetic ﬁeld. When
an external stress is applied, the initially randomly ori
ented magnetic domains align themselves in the direction
of the resultant strain, and produce an irreversible change
in the dimension of the material. The total strain in the
magnetic material is the sum of pure elastic strain and the
strain produced by the displacement of domain walls. Al
though the elastic strain vanishes when the stress is re
moved, the magnetostrictive strain can be restored only
by applying a compressive stress. Such a behavior of
magnetic materials generates a hysteresis loop on the al
ternating stress–strain curve. The saturation magneto
striction coefﬁcient, represented by l
s
, a dimensionless
quantity, is deﬁned as the ratio of the change of length
to the original length, and is measured at the saturation
ﬂux density. If the magnetostriction is sufﬁciently large
and the applied ﬁeld is an alternating one (in the audible
frequency range of 50–60 Hz), it creates noise (sound). It
has been discovered that alloys of iron and rareearth el
ements (Sm, Tb, Dy, Ho, Er, Tm, etc.) cause intense mag
netostriction 100–1000 times greater than that generated
by CoFeNi alloys (the conventional magnetostrictive
materials). Thus, rareearth alloys are designated as su
permagnetostrictive materials. Magnetostrictive actua
tors, based on magnetostrictive effects, are driven by
external magnetic ﬁelds.
In the beginning, the magnetostrictive materials were
used in telephone receivers, hydrophones, magnetos
trictive oscillators, torque meters, and scanning sonars,
employing nickel and other magnetostrictive materials
that exhibit bulk saturation strains up to 100mm/m.
Some of the current applications of magnetostrictive de
vices include ultrasonic cleaners, highforce linear motors,
active vibration or noise control systems, medical and in
dustrial ultrasonics, pumps, and sonars. Ultrasonic mag
netostrictive transducers have been developed for
chemical and material processing, surgical tools, and un
derwater sonars.
Applications of amorphous magnetoelastic materials
for numerous tagging [51,52] and remote sensing applica
tions are now prevalent [53,54]. In sensors, these materi
als exhibit sufﬁciently large magnetoelastic effect [55];
specifically, the material effectively converts magnetic en
ergy into elastic energy and vice versa. Magnetoelastic
sensors are magnetostrictive; therefore, they launch mag
netic ﬂux oscillations when they become mechanically
ELECTROMAGNETIC MATERIALS 1227
deformed. A magnetoelastic sensor, when coated uniform
ly with a layer of an elastic material, its characteristic
resonant frequency changes owing to the loading of addi
tional mass as well as the inﬂuence of elasticity in the
coating [56]. Usually, in magnetoelastic sensors, the res
onant frequency is observed by using a pickup coil [53].
For example, in case of a thin longitudinally vibrating bar,
the frequency of magnetoelastic resonance coincides with
the resonance of acoustic vibration, and therefore, the res
onance (of the magnetoelastic sensor) can be described as
the mechanical resonance of the vibrating bar.
Magnetic shape memory (MSM) alloys are also gaining
importance as possible ‘‘smart’’ materials that can func
tion as both actuators and sensors. The MSM effect is due
to the magnetic ﬁeld induced by the redistribution of twin
variants in a martensitic phase, and it offers faster re
sponse compared to conventional temperaturedriven
shape memory alloys. In addition, the strains generated
are much larger than in ordinary magnetostrictive mate
rials [57].
8.1. Magnetoelastic Damping
Magnetoelastic (or magnetostrictive) materials are attrac
tive materials for the purpose of damping and for the ab
sorption of energy. TerfenolD is an example of such a
material, which, in a stress–strain cycle, will absorb up to
80% of the mechanical energy input to the system. How
ever, pure TerfenolD has low tensile strength, which lim
its its use in applications related to energy absorption. The
damping mechanism in TerfenolD arises from the jump
ing of the magnetization from one easy magnetic direction
to another through the process of magnetoelastic coupling,
which describes the relation between magnetic properties
of ferromagnetic materials and the applied mechanical
tension. The jumping requires a minimum amount of en
ergy that corresponds to a critical value of the applied
stress. When the material is returned to a lower level of
stress, the magnetization may not return to the original
state, and this indicates the absorption of energy. A sim
pliﬁed domain structure is illustrated in Fig. 8, which
shows the ‘‘jump’’ from a vertical orientation under ten
sion to a horizontal orientation under compression.
By incorporating TerfenolD powder into a resin ma
trix, the composite may be loaded in tension in order to
realize the damping properties. The mechanism of mag
netoelastic damping is not related to the rate of loading,
but it depends on the magnitude of loading. Thus, this
type of mechanism is completely different from the typical
damping mechanisms, which are highly ratedependent.
The aspect of magnetoelastic damping, along with the
composite material approach, allows the possibility of
nanoengineered hybrid dampers for certain applications
similar to impedance matching in transducers.
As mentioned earlier, in magnetostrictive materials,
vibrational energy is damped through the movement of
magnetic domains. Consequently, the dissipation of vibra
tional energy generates heat. This effect is not diminished
even at low temperatures, and therefore, magnetostrictive
materials can be expected to be effective in damping vi
brations at very low temperatures. In order to suppress
vibrations of large motors and transformers that operate
at temperatures below 100 K, initially viscoelastic damp
ers were in use, which are based on the principle of vis
coelasticity. Viscoelastic damping involves the movements
of atoms and, at low temperatures, these movements be
come very small. Therefore, such viscoelastic dampers
cannot work efﬁciently at low temperatures. As such, in
order to operate at such low temperatures, for such appli
cations, magnetoelastic dampers have also been proposed.
This type of damper is usually made of magnetostrictive
materials such Tb/Dy alloys [58,59]. At a temperature of
B77 K, a specimen of Tb/Dy alloy is found to dissipate
about 30% of the vibrational energy per stress–strain cy
cle. It has been observed that the dissipation factor can be
further enhanced by adding very small amounts of N, Ta,
or some other elements to a Tb/Dy alloy, and also, by
changing the processing conditions.
9. PIEZOELECTRICITY
Piezo is the Greek word for pressure. As such, piezoelec
tricity is also termed pressure electricity. The phenomenon
of piezoelectricity was discovered in the 1880s by Pierre
and Jacque Curie. They found that some crystalline ma
terials, when compressed, produce a voltage proportional
to the applied pressure, and that when an electric ﬁeld is
applied across the material, there occurs a corresponding
change in shape. This characteristic is called piezoelectric
ity. Thus, piezoelectric materials are those that generate
an electric charge when deformed mechanically. This is
commonly referred to as the ‘‘generator effect.’’ Conversely,
when an external electric ﬁeld is applied to piezoelectric
materials, they assume mechanical deformation or stress.
This is commonly referred to as the ‘‘motor effect.’’
Typical examples of the naturally occurring piezoelec
tric materials are quartz, tourmaline, human skin, and
Figure 8. Domain structure of magnetoelastic material.
1228 ELECTROMAGNETIC MATERIALS
human bone, although the latter two have very low cou
pling efﬁciencies. Some artiﬁcially produced piezoelectric
crystals are Rochelle salt, lithium sulfate, and ammonium
dihydrogen phosphate. Another class of materials possess
ing piezoelectric properties is polarized piezoelectric ce
ramic. They are typically referred to as ferroelectric
materials. In contrast to the naturally occurring piezo
electric crystals, ferroelectric ceramics are polycrystalline
in nature. The most commonly produced piezoelectric ce
ramics are barium titanate, lead titanate, lead metanio
bate, and lead zirconate titanate (PZT). Although
piezoelectricity is found in several types of natural mate
rials, most modern devices use PZT. Piezoelectric ceramics
respond rapidly to the changes in input voltage; noise due
to the power supply is the only limiting factor in the po
sitional resolution. Although high voltages are used to
produce piezoelectric effect, the power consumption is low,
and the energy consumption is minimal in maintaining a
ﬁxed position with a ﬁxed load.
Production of all piezoelectric ceramic materials in
volves detailed processing, and the material properties
may be altered by modifying the chemical composition and
the manufacturing process. This essentially provides the
designer with a means to tailor the material properties to
the application. The processing involves different stages
such as calcination, high ﬁring, polling, and aging. To dis
cuss the details of these processes is beyond the scope of
the present article.
Piezoelectric ceramic materials possess electrical, me
chanical, and electrochemical properties resulting from
the chemical formulation and the manufacturing process.
Typical electrical parameters are the dielectric constant
and the dissipation. High dielectric constants of the ma
terials are desirable as this property provides low im
pedance. Materials with low dissipations are required
because they result in low electrical losses. Typical elect
romechanical parameters are the electromechanical cou
pling constant k and the piezoelectric constants g and d.
High values of the electromechanical coupling result in a
more efﬁcient transfer of electrical energy to the mechan
ical energy. In fact, the constant k is a dimensionless
number related to the ratio of the energy stored in the
mechanical and electrical portion of the material. The
constant d relates to mechanical strain and applied elec
trical ﬁeld, and it is deﬁned as the ratio of the strain
to the ﬁeld, with the unit meters per volt. The constant
g relates the opencircuit voltage and mechanical
stress, and it is deﬁned as the ratio of the voltage to
stress. The unit of g is voltmeters per newton. Typical
mechanical parameters are the density and the elastic
constants.
Depolarization of piezoelectric ceramics would result if
these materials were exposed to excessive heat, electrical
drive, or mechanical stress or any combination of these.
The temperature at which piezoelectric ceramic is com
pletely depoled is known as the Curie point. Piezoelectric
ceramics also possess the properties of pyroelectricity, in
which a change in ceramic temperature results in a
change in mechanical dimensions, thereby producing
stress within the ceramic and a corresponding electrical
charge on the surface of electrodes.
An example of a piezoelectric device is ultrasonic
probes (transducers) employed to create underwater
Sonar. This area of exploitation of piezoelectric materials
is quite mature, and nowadays ultrasonic probes work al
most exclusively according to the piezoelectric effect. A
series of powerful piezoelectric materials are now avail
able to generate ultrasound in ultrasonic probes. These
materials differ in their physical properties, and depend
ing on the technical application and the required probe
characteristics, an appropriate selection of the piezoelec
tric material can be made. Knowledge of the construction
al details of the utilized piezoelectric materials, however,
not only contributes toward a better technical understand
ing of the basic action of the probe but also helps us un
derstand the behavior of the probe under certain
operation conditions.
For the construction of ultrasonic probes, the diameter
and the frequency of the piezoelectric element are the im
portant parameters to be considered. However, it is essen
tial to have a detailed knowledge of the physical and the
acoustic properties of the piezoelectric material used; the
relevant information will not only contribute to a better
technical understanding of the basic working principle of
probes but will also gives an insight into the behavior of
the probe under various operating conditions. Information
about the piezoelectric materials will also be helpful in
detecting the suitability of the type of probe under the
given circumstances. As such, it would be of worthwhile to
introduce the different types of piezoelectric materials
currently in use. We shall explain their strengths and
weaknesses and to show how they can be exploited for
various probe designs.
9.1. Piezoelectric Materials
In the early days of ultrasound technology, materials such
as quartz, lithium sulfate, and barium titanate were used.
However, such materials are rarely in use today; instead,
new powerful piezoelectric materials are available, which
have varieties of basic acoustic and electric characteris
tics. Table 4 lists the important physical characteristics of
different piezoelectric materials used today to generate
ultrasonic waves [60].
The acoustic impedances of piezoceramic materials are
usually high and comparable to those of many metallic or
ceramic solids. The coupling coefﬁcient for radial (or trans
verse) vibrations indicates to what extent a piezoelectric
element can convert energy into transverse modes, which
are perpendicular to the thickness mode. Transverse
modes should be suppressed as much as possible as they
cause undesired signal distortions. These modes are de
veloped at low frequencies, because they correspond to the
lateral dimensions of the piezoelectric element. Lead ti
tanate is the only piezoelectric material that has a negli
gibly small coupling coefﬁcient k. The electromechanical
coupling coefﬁcient k
t
is a measure of the fraction of elec
trical energy, which is converted into mechanical energy
in the thickness mode. It determines the efﬁciency of gen
erating ultrasonic waves in a particular vibrational mode
utilized in probes. The values of k
t
for piezoceramic mate
rials are usually high. However, if ultrasonic waves prop
ELECTROMAGNETIC MATERIALS 1229
agate into liquids or plastics, most of the acoustic energy
generated by the piezoceramic element is reﬂected at the
boundary between the piezoelectric material and the me
dium of propagation. Therefore, only a fraction of the
acoustic energy is transferred into the medium of propa
gation. Such piezoceramics are used for up to a maximum
of 30 MHz.
Piezoelectric plastics such as PVDF or related copoly
mers are better matched to the low acoustic impedance of
liquids or plastics. In addition, they are mechanically ﬂex
ible. It becomes possible to generate ultrasound at fre
quencies up to 160 MHz with very thin foils of PVDF.
However, it becomes evident from the coupling coefﬁcient
k
t
for the thickness mode that PVDF is less sensitive than
ceramics.
The socalled 1–3 composite materials [61] appear more
promising as they have high values of k
t
and low values of
k. Such composites perform better within the approximate
temperature range of 1001C. Table 4 shows typical values
for 1–3 piezocomposites. However, these values will vary
in a wide range by varying the piezoceramic material as
the design.
9.2. Applications of Piezoelectric Materials in Probes
Different properties of piezoelectric materials are very
useful for the design of different types of probes. The most
suitable material can be selected depending on the in
tended application of the probe. As we can see from
Table 4, lead metaniobate has the lowest acoustic imped
ance of all the piezoceramics. This is better suited to de
sign highresolution shockwave probes (shockwaves have
extremely short pulses). The relatively low impedance of
lead metaniobates is also useful for the design of broad
band angle probes and TR probes. Lead titanate has
small values of k, and therefore, radial vibrational modes
of this material are negligible. As such, lead titanate
can be utilized to produce probes with especially small
measurements. Lead zirconate titanate (PZT) is used
where large measurements and high ultrasonic absorp
tion are required. In such cases, it is important that the
probe should generate ultrasound of higher intensity
than usual, so that even echoes from the far end can
still be detected. For receiver probes, it is better to use a
bulk PZT element of appropriate thickness. This is be
cause, at a given input pressure, the voltage across the
electrodes of a piezoelectric element is proportional to its
thickness.
At high frequencies the use of piezoceramics is limited.
Since the resonant frequency of piezoelectric disks vibrat
ing in the thickness mode is inversely proportional to the
thickness, piezoeceramic disks become thinner than
0.1 mm above 20 MHz. Such piezoeceramic disks are brit
tle and difﬁcult to handle during the manufacturing pro
cess of the probe. However, at higher frequencies, thin
foils of PVDF are easier to handle [62]. Because of their
low acoustic impedance, an effective transfer of ultrasonic
waves is possible only into water or plastic materials.
Even at normal frequencies the high ﬂexibility of PVDF
and its low impedance, are useful for the design of special
linefocused immersion probes. Such linefocused PVDF
probes have been in use since 1990 in highspeed test sys
tems for steel pipes. These are described in more detail in
other publications [63,64].
Because of their relatively low acoustic impedance and
high sensitivity, 1–3 composites are also suitable for gen
erating ultrasound in liquids and plastics. However, the
manufacturing of such probes is less complicated than the
manufacturing process of the piezo element itself. 1–3
composites are also suitable for the manufacturing of
linefocused immersion probes. However, PVDF foils are
less expensive.
In conclusion, the currently available piezoelectric ma
terials complement each other in their physical charac
teristics. PVDF is very ﬂexible, and it is easy to handle
and shape. Further, it exhibits good stability over time,
and does not depolarize when subjected to very high al
ternating ﬁelds. But the coupling constant of PVDF is sig
nificantly lower than that of PZT. Shaping of PVDF can
reduce the effective coupling of mechanical and electrical
energies due to edge effects. In the design of ultrasonic
probes, it is very important to use a material that has the
best suitable characteristics. In addition, economy also
plays a vital role for the user, as well as for the manufac
turer. As a result, PZT has been considered as an economic
piezoelectric material over the years.
Acknowledgment
The authors are immensely grateful to Prof. Akhlesh La
khtakia of the Pennsylvania State University (USA) for
reviewing and making constructive suggestions to
improve the manuscript.
Table 4. Characteristics of Piezoelectric Materials
Physical Properties
Lead Zirconate
Titanate PZT5
Lead
Titanate PT
Lead
MetaniobatePbNb
2
O
6
Polyvinylidene
Fluoride PVDF
(Copolymer) 1–3 Composite
Acoustic impedance Z (10
6
kg m
À2
s
À1
)
33.7 33 20.5 3.9 9
Resonant frequency f (MHz)
o25 o20 o30 160–10 (55–2) o10
Coupling coefﬁcient (thickness mode) k
t
0.45 0.51 0.30 0.2 (0.3) 0.6
Coupling coefﬁcient (radial mode) k
p
0.58 o0.01 o0.1 0.12 (k
31
) B0.1
Relative permittivity e
r
1700 215 300 10 450
Maximum temperature (1C)
365 350 570 80 100
1230 ELECTROMAGNETIC MATERIALS
BIBLIOGRAPHY
1. R. S. Elliott, Electromagnetics, McGrawHill, New York, 1966.
2. P. F. Mottelay, Biographical History of Electricity and Mag
netism, Ayer Publishers, Salem, NH, 1975.
3. E. T. Whittaker, A History of the Theories of Aether and Elec
tricity, American Institute of Physics, New York, 1987.
4. C. Kittle, Introduction to Solid State Physics, 6th ed., Wiley,
New York, 1986.
5. B. K. P. Scaife, Principles of Dielectrics, Oxford Univ. Press,
Oxford, UK, 1989.
6. E. D. Palik, ed., Handbook of Optical Constants of Solids II,
Academic Press, Boston, MA, 1991.
7. J. D. Jackson, Classical Electrodynamics, 2nd ed., Wiley, New
York, 1975.
8. T. Van Duzer and C. W. Turner, Principles of Superconductive
Devices and Circuits, Elsevier, New York, 1981.
9. V. Z. Kresin and A. Wolf, Fundamentals of Superconductivity,
Plenum, New York, 1990.
10. J. A. Stratton, Electromagnetic Theory, McGrawHill, New
York, 1941.
11. R. E. Collin, Field Theory of Guided Waves, 2nd ed., IEEE
Press, Piscataway, NJ, 1991.
12. M. Born and E. Wolf, Principles of Optics, 6th ed., Pergamon
Press, New York, 1980.
13. A. Yariv and P. Yeh, Optical Waves in Crystals, Wiley, New
York, 1984.
14. H. A. Haus, Waves and Fields in Optoelectronics, Prentice
Hall, Englewood Cliffs, NJ, 1984.
15. A. Yariv, QuantumElectronics, 3rd ed., Wiley, NewYork, 1989.
16. M. Javid and P. M. Brown, Field Analysis and Electromag
netics, McGrawHill, New York, 1963.
17. P. Lorrian and D. Corson, Electromagnetic Fields and Waves,
2nd ed., Freeman, San Franscisco, CA, 1970.
18. J. D. Kraus, Electromagnetics, 3rd ed., McGrawHill, New
York, 1984.
19. D. T. Paris and F. K. Hurd, Basic Electromagnetic Theory,
McGrawHill, New York, 1969.
20. J. R. Reitz and F. J. Milford, Foundations of Electromagnetic
Theory, AddisonWesley, Reading, MA, 1960.
21. J. E. Parton, S. J. T. Owen, and M. S. Raven, Applied Elec
tromagnetics, 2nd ed., Macmillan, London, 1986.
22. A. H. Morrish, The Physical Principles of Magnetism, Krieger,
Malabar, FL, 1982.
23. R. Boll, ed., Soft Magnetic Materials, Heyden, London/Phila
delphia/Rheine, 1979.
24. R. Plonsey and R. E. Collin, Principles and Applications of
Electromagnetic Fields, 2nd ed., McGrawHill, New York,
1982.
25. H. Jouve, Magnetic Bubbles, Academic Press, San Diego, CA,
1987.
26. A. Goldman, Modern Ferrite Technology, Van NostrandRein
hold, New York, 1990.
27. R. S. Elliott, An Introduction to Guided Waves and Microwave
Circuits, PrenticeHall, Englewood Cliffs, NJ, 1993.
28. C. L. Hogan, Bell Syst. Tech. J. 31:1 (1952).
29. C. E. Fay and R. L. Comstock, IEEE Trans. Microwave Theory
Tech. MTT13:15 (1965).
30. E. F. Schloemann, Proc. IEEE 76:188 (1988).
31. A. Lakhtakia, Beltrami Fields in Chiral Media, World Scien
tific, Singapore, 1994.
32. A. Priou, A. Sihvola, S. Tretyakov, and A. Vinogradov, eds.,
Advances in Complex Electromagnetic Materials, Kluwer, Do
rdrecht, The Netherlands, 1997.
33. O. N. Singh and A. Lakhtakia, eds., Electromagnetic Fields in
Unconventional Materials and Structures, Wiley, New York,
2000.
34. W. S. Weiglhofer and A. Lakhtakia, Casuality and natural op
tical activity (chirality), J. Opt. Soc. Am. A 13:385–386 (1996).
35. A. Lakhtakia, R. Messier, M. J. Brett, and K. Robbie, Sculp
tured thin ﬁlms (STFs) for optical, chemical and biological
applications, Innovat. Mater. Res. 1:165–176 (1996).
36. A. Lakhtakia and R. Messier, Sculptured thin ﬁlms—I. Con
cepts, Mater. Res. Innovat. 1:145–148 (1997).
37. V. C. Venugopal and A. Lakhtakia, Dielectric sculptured ne
matic thin ﬁlms for rugatelike ﬁlters, Opt. Commun.
149:217–222 (1998).
38. R. Messier and A. Lakhtakia, Sculptured thin ﬁlms—II. Ex
periments and applications, Mater. Res. Innovat. 2:217–222
(1999).
39. A. Lakhtakia and M. McCall, Sculptured thin ﬁlms as nar
rowbandpass circular polarization ﬁlters, Opt. Commun.
168:457–466 (1999).
40. V. C. Venugopal and A. Lakhtakia, Sculptured thin ﬁlms:
Conception, optical properties, and applications, in O. N.
Singh and A. Lakhtakia, eds., Electromagnetic Fields in Un
conventional Materials and Structures, Wiley, New York,
2000.
41. B. V. Bokut, A. N. Serdyukov, and F. I. Fedorov, On the form of
constitutive equations in optically active crystals, Optika i
Spektroskopiya 37:288–293 (1974).
42. F. I. Fedorov, Theory of Gyrotropy, Nakua i Tekhinka (Science
and Technology Publishers), Minsk, 1976.
43. B. V. Bokut and A. N. Serdyukov, On the theory of optical ac
tivity of nonuniform media. Zh. Prikladnoi Spektroskopii
20:677–682 (1974).
44. L. D. Landau and E. M. Lifshitz, Electrodynamics of Contin
uous Media, 2nd ed., Pergamon Press, Oxford, 1984.
45. S. A. Tretyakov and A. A. Sochava, Eigenwaves in uniaxial
chiral omega media, Microwave Opt. Technol. Lett. 6:701–705
(1993).
46. S. A. Tretyakov and A. A. Sochava, Proposed composite ma
terial for nonreﬂecting shields and antenna radomes, Elec
tron. Lett. 29:1048–1049 (1993).
47. S. A. Tretyakov, Thin pseudochiral layers: Approximate
boundary conditions and potential applications, Microwave
Opt. Technol. Lett. 6: 112–115 (1993).
48. I. V. Semchenko, Gyrotropic properties of superlattices in the
longwave approximation, Kristallograﬁya 35:1047–1050
(1990).
49. N. I. Gheludyev and A. D. Petrenko, Physical mechanisms of
nonlinear optical activity in crystals, Kristallograﬁya
29:1045–1053 (1984).
50. I. N. Akhramenko, I. V. Semchenko, and A. N. Serdyukov,
Selective interaction of electromagnetic waves in crystals
with spiralmodulated nonstationary dielectric permittivity,
Optika i Spektroskopiya 66:618–622 (1989).
51. M. WunFogle, H. T. Savage, L. T. Kabacoff, K. B. Hathaway,
J. M. Merchant, and B. Beihoff, Magnetoelastic effects in
amorphous wires and amorphous ribbons with nonmagnetic
thinﬁlm coatings, J. Appl. Phys. 64:5405–5407 (1988).
52. T. H. O’Dell, Measurement of magnetomechanical coupling
factor in amorphous ribbons, Physica Status Solidi A 74:
565–572 (1982).
ELECTROMAGNETIC MATERIALS 1231
53. C. A. Grims, K. G. Ong, K. Loiselle, P. G. Stoyanov, D. Ko
uzoudis, Y. Liu, C. Tong, and F. Teﬁku, Magnetoelasic sensors
for remote query environmental monitoring, Smart. Mater.
Struct. 8:639–645 (1999).
54. M. K. Jain, S. Schmidt, K. G. Ong, C. Mungle, and C. Grimes,
Magnetoacoustic remote query temperature and humidity
sensors, Smart Mater. Struct. 9:502–510 (2000).
55. J. Gutie´rrez, J. M. Barandiara´ n, and O. V. Nielsen, Magneto
elastic properties of some Ferich FeCoSiB metallic glasses,
Physica Status Solidi A 111:279 (1989).
56. S. Schmidt and C. Grimes, Elastic modulus measurement of
thin ﬁlms coated onto magnetoelastic ribbons, IEEE Trans.
Magn. 37:2731–2733 (2001).
57. J. Enkovaara, A. Ayuela, L. Nordsto¨m, and R. M. Nieminen,
Structural, thermal, and magnetic properties of Ni
2
MnGa, J.
Appl. Phys. 91:7798–7800 (2002).
58. J. Dooley, B. Fultz, J. Voccio, and R. Chave, Magnetoelastic
vibration dampers, NASA Tech. Briefs 25:56–57 (2001).
59. J. Dooley, R. Chave, B. Fultz, A. Clark, N. Good, and J. Graetz,
Progress in magnetoelastic vibration dampers, NASA Tech.
Briefs 25:57 (2001).
60. M. Lach, M. Platte, and A. Ries, Piezoelectric materials for
ultrasonic probes, NDTnet 1(9):(1996).
61. L. Chofflet, M. Gauchet, and J. M. Tellier, Which piezoelectric
material for which transducer? Revue Annuelle LEP:37–38
(1990).
62. R. G. Swartz and J. D. Plummer, On the generation of high
frequency acoustic energy with polyvinylidene Fluoride,
IEEE Trans. Sonics Ultrason., SU 27, 295–303, 1980.
63. V. Deutsch, M. Platte, and P. Mo¨ller, Ultraschallpru¨ fko¨pfe aus
piezoelektrischen Hochpolymeren, Materialpru¨fung 32:
333–337 (1990).
64. M. Platte and P Mo¨ller, Automatisches Ultraschallpru¨ fen von
Blechen und Rohren, Ba¨nder Bleche Rohre 34:25–32 (1993).
ELECTROMAGNETIC MODELING
R. VAHLDIECK
Swiss Federal Institute of
Technology Zu¨ rich (ETHZ)
Zu¨ rich, Switzerland
WOLFGANG J.R. HOEFER
University of Victoria
Victoria, BC, Canada
Numerical modeling of electromagnetic (EM) ﬁelds or
computational electromagnetics is a combination of nu
merical methods and ﬁeld theory—a discipline in its own
right and of growing importance in such diverse areas as
microwave and RF (radiofrequency) engineering, antenna
design, EM ﬁeld scattering, semiconductor physics, bioe
lectromagnetics, and electromagnetic compatibility and
interference (EMC/EMI).
The starting point for modeling electromagnetic ﬁelds
is Maxwell’s equations considering a set of known bound
ary/initial conditions. In addition, certain theorems and
principles, such as superposition, equivalence, and duality
can be utilized. Entirely analytical solutions to Maxwell’s
equations are possible only for a narrow range of problems
and most of them were solved a long time ago. Classical
analytical methods are the methods of separation of
variables, conformal mapping, series expansion, and in
tegral methods (if the integrals have analytical solutions).
Today’s complexity of relevant EM ﬁeld problems pre
vent entirely analytical (closedform) solutions to Max
well’s equations. Rather, numerical approaches are
necessary that approximate the exact solution with any
desired degree of accuracy. Does this eliminate the need
for analytical methods? Certainly not! In fact, analytical
preprocessing is part of all numerical methods, and if the
analytical content is high, the method is often described as
semianalytical.
In general, the electromagnetic ﬁeld problem can be
conveniently formulated as a partialdifferential equation
while that of the source problem as an integral equation
(operator equation). The objective of numerical methods is
to transform an exact operator equation into a solvable
(discrete) matrix equation. To achieve this, virtually all
numerical methods follow the same principal steps:
1. To express the unknown function of the operator
equation by a sum of linear independent functions
with unknown expansion coefﬁcients. This is often
called a trial function which approximates the real
function if the number of terms in the sums goes to
inﬁnity.
2. The continuous solution domain is represented by a
set of discretized subdomains consisting of a ﬁnite
number of elements or nodes.
3. To determine the unknown expansion coefﬁcients in
the trial function, some form of error minimization is
chosen. Employing either variational principles or
the method of weighted residuals, the operator
equation is transformed into a matrix equation
that must be solved by appropriate techniques.
Depending on the kind of operator used, numerical
methods are in general categorized into domain methods
and boundary methods. More recently a new category has
been added, that of hybrid methods.
Domain methods solve the electromagnetic ﬁeld pro
blem described by partial differential equations. Examples
are the ﬁnite difference method and the ﬁnite element
method. Both methods are based on differential equations
and on discretization of the entire computational domain.
Boundary methods solve the electromagnetic source pro
blem described by integral equations, either volume inte
gral equations or boundary integral equations. Examples
are the boundary element method or the method of mo
ments. The latter can also be used to solve differential
equations. Hybrid methods are a combination of two or
more different methods. A hybrid numerical approach can
be a twostep procedure or an implicit hybrid algorithm. In
the twostep procedure, one part of the problem is solved
with one method, the results of which become the input
parameters for the next method. In the implicit hybrid
algorithm two or more modeling approaches are combined
into one new algorithm to exploit only the advantageous
features of each method. Hybrid methods are very useful
1232 ELECTROMAGNETIC MODELING
for problems that can either not be solved within the
framework of a single technique or in cases where a
combination of methods results in a computationally
more efﬁcient algorithm to model the electromagnetic ﬁeld.
Since the late 1970s the importance of modeling elec
tromagnetic ﬁelds has increased manifold. This refers not
only to the analysis of arbitrary electromagnetic ﬁeld
problems but also to the CAD (computeraided design) of
circuits and components. With rising operating frequency
and higher circuit density, quasistatic analysis methods
are replaced by fullwave analysis methods to account for
nearﬁeld effects or higherorder mode interaction. Sim
plifying assumptions about material properties have been
replaced by realistic material descriptions taking disper
sion and losses into account. With the steady increase in
computer power, research in electromagnetic modeling is
still advancing at a rapid pace, introducing new tech
niques, expanding the capability of existing methods or
simply improving their computational efﬁciency and
accuracy. Only some of the multitude of electromagnetic
modeling techniques available today can be outlined here.
This article will focus only on some mainstream tech
niques, most of which employ either the method of
weighted residuals or variational principles. We will begin
with modeling techniques in the frequency domain, most
importantly the method of moments (MoM). This section
is followed by timedomain methods and here in particular
the ﬁnitedifference timedomain method (FDTD) and the
transmissionline matrix (TLM) method. Finally, a brief
overview with respect to hybrid methods concludes this
article. A comparison between the various modeling ap
proaches as well as their advantages and disadvantages is
added where appropriate.
For a more detailed representation of the various
methods, the interested reader is referred to the books
by Sadiku [1] or Zhou [2]. Another excellent source of
information is the book edited by Itoh [3], which discusses
the various methods in the context of passive microwave
circuit modeling.
1. FREQUENCYDOMAIN METHODS
Most electromagnetic modeling techniques operate in the
frequency domain; that is, the time derivative d/dt is
replaced by jot. Therefore, all calculations are performed
at a single frequency only. In cases where the electromag
netic ﬁeld is of interest over a frequency band, repeated
calculations are necessary. This is in contrast to time
domain methods. Here a single run of the algorithm, after
a proper impulse excitation, followed by a Fourier trans
form of the impulse response, provides the information
over a wide frequency range. Which method to use de
pends on the problem at hand.
1.1. The Method of Moments
The method of moments (MoM) is a general form of
weighted residuals to solve integral, differential, and
integrodifferential equations. The method itself does not
provide any information about the derivation of the gov
erning equation, thereby allowing its applicability to a
wide range of physical phenomena.
The equations solved by MoM are normally an electric
ﬁeld integral equation (EFIE) or a magnetic ﬁeld integral
equation (MFIE) of the following form
EFIE : E¼f
e
ðJÞ
MFIE : H¼f
m
ðJÞ
whereby E and H are the incident ﬁeld quantities and J is
the induced current density. The form of integral equation
(EFIE or MFIE) depends on the problem. In most cases
these integrals are formulated in the frequency domain,
although the MoM can also be applied to solve problems in
the time domain.
The MoM as a mathematical technique was ﬁrst intro
duced by Mikhlin [4] and later popularized in the area of
numerical electromagnetics by Harrington [5]. The math
ematical formulation of the MoM is simple and general.
Consider a linear operator L such that
LJ ¼g ð1Þ
The function J (e.g., current density) is the response of the
system represented by the operator L to the excitation g
(magnetic or electric ﬁeld). When L and g are known,
Eq. (1) represents an analysis problem from which J is
determined. A synthesis problem is one in which both J
and g are known but L is to be determined.
The MoM solution of Eq. (1) takes advantage of the
linearity of the operator L to expand J in a series of the
form
J ¼
m
a
m
J
m
ð2Þ
The a
m
’s are unknown expansion coefﬁcients and the J
m
’s
are known functions in the domain of the operator L. The
functions J
m
, which are assumed to form a complete set,
are called basis functions. Substituting the expansion of J
in Eq. (1) and using the linearity of L, we get
LJ ¼L
m
a
m
J
m
¼
m
a
m
LJ
m
¼g ð3Þ
The original problem is now reduced to determining the
expansion coefﬁcients from this last equation. In the MoM,
both sides of the equation are projected onto the range of
the operator L. Let T
m
denote a complete set of functions
in the range of L. The T
m
’s are referred to as testing or
weighting functions. Taking the inner product of Eq. (3)
against the testing functions, yields
m
a
m
T
n
; LJ
m
h i ¼ T
n
; g
¸ _
n¼1; 2; . . .
ð4Þ
Here, a suitable inner product (a linear form) is assumed
to be known. Usually, the inner product is taken as the
ELECTROMAGNETIC MODELING 1233
integral of the product of the arguments
J; g
¸ _
¼
_
Jg ð5Þ
Equation (4) can be rewritten in a more convenient matrix
form
½A½a ¼½U ð6Þ
where
½A
mn
¼ T
m
; LJ
n
h i ð7Þ
½U
m
¼ T
m
; g
¸ _
ð8Þ
If the matrix [A] is not singular, its inverse [A]
À1
exists.
The expansion coefﬁcients a
m
are then given by
½a ¼½A
À1
½U ð9Þ
Once the expansion coefﬁcients are known, the solution is
given by Eq. (2). The MoM is most commonly implemented
in the form of Galerkin’s method in which the testing
functions are equal to the basis functions, T
m
¼J
m
.
In practice, the inﬁnite expansion Eq. (2) is truncated
at some upper value of m. The approximate solution, given
by Eq. (9), is then straightforwardly computed using a
computer. The efﬁciency of the technique depends heavily
on having basis functions which approximate well the
exact solution with only a few terms. Therefore, MoM
techniques are not very effective in the analysis of complex
geometries or inhomogeneous dielectrics. On the other
hand, MoM techniques are particularly efﬁcient for the
analysis of antennas and electromagnetic scattering pro
blems.
1.2. The ModeMatching Technique
The modematching technique (MMT), or modal analysis
method, can be viewed as a special case of the method of
moments and is a frequently used analysis tool for scatter
ing of electromagnetic waves at metal waveguide disconti
nuities. In this method, the ﬁelds immediately left and
right from the discontinuity are expanded in a series of
weighted eigenfunctions. From the matching condition or
continuity condition of the ﬁelds tangential to the discon
tinuity plane the coefﬁcients of the series expansions can
be determined. Details of the method can be found in Refs.
6 and 7. The MMT has been successfully applied to
eigenvalue as well as to scattering problems both in
homogeneous waveguides as well as partially dielectric
loaded waveguides. Numerous application examples for
eigenvalue problems are given, for example, in Refs. 8 and
9 and for scattering problems in Refs. 10 and 11.
The basic problem to be solved in the MMT is to ﬁnd the
coefﬁcients of the ﬁeld expansion to minimize the least
square error between the exact EM ﬁeld and the approx
imating series of eigenfunctions. If we postulate g(x) as the
exact ﬁeld in the aperture between two subregions, the
approximation by a series of orthogonal eigenfunctions
f
i
(x) is written
gðxÞ ﬃ
N
i
a
i
f
i
ðxÞ; x 2 x
1
; x
2
ð10Þ
where a
i
denote the weighting (expansion) coefﬁcients.
Multiplying both sides of Eq. (10) with a set of weighting
functions w
j
(x) and integrating over the domain of x
results in
_
x
2
x
1
w
j
ðxÞgðxÞ dx ¼
N
i
a
i
_
x
2
x
1
w
ji
ðxÞf
i
ðxÞ dx ð11Þ
We solve Eq. (11) such that its weighted residual is zero:
_
x
2
x
1
w
j
ðxÞ gðxÞ À
N
i
a
i
f
i
ðxÞ dx
_ _
dx ¼0;
j ¼1; 2; . . . ; N
ð12Þ
If the weighting function is a set of delta functions
w
j
ðxÞ ¼dðx Àx
j
Þ
Eq. (10) becomes Eq. (13) which is known as point match
ing Eq. (11):
gðx
j
Þ ﬃ
N
i
a
i
f
i
ðx
j
Þ; j ¼1; 2; . . . ; N ð13Þ
This equation is solved for a
i
. If the weighting functions
are the eigenfunctions themselves, we use the least
square error to select the expansion coefﬁcients such as
to minimize the integrated squared error
0¼
@
@a
i
_
gðxÞ À
N
i
a
i
f
i
ðxÞ
_ _
2
dx
By using the orthogonality of the eigenfunctions f
i
(x) we
ﬁnd the individual coefﬁcients
_
x
2
x
1
f
i
ðxÞgðxÞ dx
_
x
2
x
1
f
i
ðxÞ
2
dx
¼a
i
ð14Þ
It should be noted that the integral in the denominator of
Eq. (14) is a normalization constant while the integral in
the numerator (the coupling integral) can in most cases be
solved analytically, except in cases where the geometry of
the regions left and right of a discontinuity is described by
different coordinate systems.
The MMT becomes a powerful analysis tool in conjunction
with the generalized scattering matrix (GSM) approach. The
GSM relates all reﬂected wave amplitudes of fundamental
and higherorder modes at discontinuities to the incident
wave amplitudes. Wave amplitudes are in this context the
power normalized expansion coefﬁcients in Eq. (10).
1234 ELECTROMAGNETIC MODELING
A simple example illustrates the MMT in conjunction
with the GSM technique. A waveguide step discontinuity
is shown in Fig. 1. We assume that in both waveguide
sections only the fundamental TE
10
mode can propagate.
The TE
10
mode consists of three ﬁeld components of
which two, E
y
and H
x
, are tangential to the discontinuity
plane at z ¼0 (Fig. 1b). In regions 1 and 2, the electric ﬁeld
component can be expressed as a series of eigenfunctions
such as in Eq. (10). Applying the continuity condition at
the discontinuity plane (z ¼0), E
1
y
¼E
2
y
, results in
M
m¼1
T
1
m
sin
mp
a
x
_ _
ðA
1
m
þB
1
m
Þ
¼
N
n¼1
T
2
n
sin
np
a Àx
1
ðx Àx
1
Þ
_ _
ðA
2
n
þB
2
n
Þ
ð15Þ
The coefﬁcients T in Eq. (15) ensure power normalization
of the incident A and reﬂected B wave amplitudes for
fundamental and higherorder modes. The products TÁ A
or TÁ B are the expansion coefﬁcients of Eq. (10). Multi
plying both sides of Eq. (15) with sin(ip/a x) and integrat
ing over the cross section x ¼0 Àa yields
Ey : A
1
i
þB
1
i
¼
N
n¼1
T
2
n
_
a
x
1
sin
np
a
x
_ _
sin
ip
a Àx
1
ðx Àx
1
Þ
_ _
dx
_
a
0
sin
2
ip
a
x
_ _
dx
.ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ¸¸ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ.
LE
ðA
2
n
þB
2
n
Þ
ð16Þ
The normal modes satisfy the orthogonality relation, and
the integral in the denominator of Eq. (16) thus becomes
_
a
0
sin
2
ip
a
x
_ _
dx ¼
a
2
ð17Þ
A similar equation is obtained from the continuity condi
tion of the H
x
component
H
x
:
M
m¼1
T
2
m
Y
m
_
a
x1
sin
mp
a
x
_ _
sin
ip
a Àx
1
ðx Àx
1
Þ
_ _
dx
_
a
x
1
sin
2
ip
a Àx
1
x Àx
1
_ _
dx
.ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ¸¸ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ.
LH
A
1
i
ÀB
1
i
¼A
2
m
ÀB
2
m
ð18Þ
Both equations can be written in matrix notation:
E
y
: ½A
1
þ½B
1
¼½L
E
.
ð½A
2
þ½B
2
Þ
H
x
: ½L
H
.
ð½A
1
À½B
1
Þ ¼½A
2
À½B
2
Rearranging these equations leads to the generalized
scattering matrix of the waveguide discontinuity
½B
1
½A
2
_ _
¼
½S
11
½S
12
½S
21
½S
22
_ _
½A
1
½B
2
_ _
ð19Þ
A wellknown problem in the MMT is the slow conver
gence rate and the relative convergence phenomenon. The
latter can be alleviated to some degree by choosing the
number of modes on both sides of the discontinuity in the
same ratio as the waveguide dimensions left and right
from the discontinuity. The MMT is also not well suited for
problems in which the structure contains a mixedcoordi
nate system. Coupling integrals must then be solved
numerically, which makes the algorithm slow and not
very effective. In those cases the MMT can be combined
with other techniques (hybrid methods discussed later)
that are more appropriate for certain parts of the problem.
1.3. Coupled Integral Equation Technique
The problem of slow convergence in the MMT is due
mainly to the fact that the eigenfunctions of the wave
guides left and right from the discontinuity do not accom
modate the boundary conditions of the ﬁelds in the
discontinuity plane. This is not so in the coupled integral
equation technique (CIET). A major advantage of the
CIET is its ability to include a priori information, such
as the edge conditions, at multiple discontinuities simul
taneously. The salient features of the technique can be
found in Refs. 12 and 13; here we only show how the edge
condition at a single discontinuity is handled through the
CIET. For this we refer again to the example of Fig. 1.
The tangential electric ﬁeld within the discontinuity
region (the gap region) is expanded in a series of basic
functions that are chosen such that they include the edge
condition of the electric ﬁeld at x ¼x
1
:
E
gap
ðxÞ ¼
I
i
c
i
E
i
ðxÞ
E
gap
¼0; 0 x x
1
E
gap
O0; x
1
ox a
_
ð20Þ
A possible set of basic functions that satisfy the edge
condition is the following:
E
i
ðxÞ ¼
sin
ip
a Àx
1
ðx Àx
1
Þ
_ _
½ðx Àx
1
Þð2a Àx
1
ÀxÞ
1=3
Matching the tangential electric ﬁeld of both regions
(1 and 2) to the electric ﬁeld in the discontinuity plane
A
1
B
1
A
2
B
2
a
x
1
z = 0
x
y
b
z
(b) (a)
Figure 1. Hplane discontinuity in rectangular waveguide;
(a) perspective view; (b) scattered wave amplitudes.
ELECTROMAGNETIC MODELING 1235
E
gap
yields
E
1
y
ðxÞ ¼E
gap
ðxÞ; 0 x a
E
2
y
ðxÞ ¼E
gap
ðxÞ; x
1
x a
ð21Þ
From Eq. (21), using Eq. (20), one obtains
B
1
i
¼ Àd
i1
þ
2
a
_
a
x
1
E
gap
ðxÞ sin
ip
a
x
_ _
dx
B
2
i
¼
2
a Àx
1
_
a
x
1
E
gap
ðxÞ sin
ip
a Àx
1
x Àx
1
_ _
dx
ð22Þ
Utilizing the matching condition from the tangential
magnetic ﬁeld, H
1
x
¼H
2
x
, which is deﬁned in the interval
x
1
rxra and replacing the expansion coefﬁcients therein
by using Eq. (22), leads to the following integral equation
2Y
1
1
sin
p
a
x
_ _
¼
i
Y
2
i
2
a Àx
1
_
a
x
1
E
gap
ðxÞ sin
ip
a Àx
1
x
0
Àx
1
_ _
sin
ip
a Àx
1
x Àx
1
_ _
dx
0
þ
n
Y
1
n
2
a
_
a
x
1
E
gap
ðxÞ
sin
np
a
x
0
_ _
sin
np
a
x
_ _
dx
0
ð23Þ
The remaining step is to ﬁnd the expansion coefﬁcients c
i
in Eq. (20). This is accomplished by substituting Eq. (20)
in Eq. (23) and using the method of moments in the form of
Galerkin’s method. The number of terms in Eq. (20), the
value of I, is increased until convergence is achieved.
Typically only a few terms are needed to accurately
describe the discontinuity. Although the analytical con
tent of the CIET is higher than for the MMT, the numer
ical efﬁciency is signiﬁcantly better and no relative
convergence problems are encountered.
1.4. The SpectralDomain Method
The spectraldomain method (SDM) [14] utilizes the Four
ier transformation to eliminate all but one space variable
in the Helmholtz equation. The latter is then solved
analytically for the remaining space variable. The SDM
is a computationally very efﬁcient analysis tool for micro
wave transmission lines. The method has found numerous
applications, mainly in the analysis of electromagnetic
ﬁelds in planar transmission line structures where the
overall cross section can be divided into homogeneous
dielectric subregions. Although the method was originally
introduced for single conductor transmission lines with
inﬁnitely thin conductor [15], the SDM has been general
ized [16,17] to include ﬁnite metallization thickness and
multiple dielectric layers.
The method is best illustrated by considering the
example of a microstrip line (Fig. 2). In regions with
constant m and e the electromagnetic ﬁeld, represented
by a potential function f
e,h
, satisﬁes the Helmholtz
equation
@
@x
2
þ
@
@y
2
þ
@
@z
2
þk
2
_ _
f
e;h
ðx; y; zÞ ¼0
A twodimensional Fourier transform on f
e,h
(e.g., on x
and z)
~
ff
e;h
ða; y; bÞ ¼
__
e
Àjðax þbzÞ
f
e;h
ðx; y; zÞ dx dz ð24Þ
transforms the Helmholtz equation into one that contains
only one space variable
@
2 ~
ff
e;h
@y
2
Àða
2
þb
2
Àk
2
Þ
~
ff
e;h
¼0 ð25Þ
A solution to this equation is known in the form of
exponential functions or hyperbolic functions. In each
homogeneous subregion, f
e,h
can thus be transformed
from one boundary to the opposite. With respect to Fig. 2,
this implies that the known boundary condition of the
electromagnetic ﬁeld at planes y ¼0 and y ¼h can be
transformed into plane y ¼d. The ﬁnal step in the formula
tion of the SDM is then to satisfy the boundary condition of
the tangential ﬁelds at the interface y ¼d, which are in the
transformed domain
~
EE
z1
¼
~
EE
z2
;
~
EE
x1
¼
~
EE
x2
;
~
HH
x2
À
~
HH
x1
¼
~
JJ
z
;
~
HH
z2
À
~
HH
z1
¼ À
~
JJ
x
~
JJ
x
and
~
JJ
z
are the Fourier transforms of the unknown
currents on the strip. From this, a matrix equation can be
derived as follows
~
EE
z1
~
EE
x1
_ _
¼
~
ZZ
zz
~
ZZ
zx
~
ZZ
xz
~
ZZ
xx
_ _
~
JJ
z
~
JJ
x
_ _
ð26Þ
with
~
JJ
z
¼S
M
m
c
m
~
JJ
zm
ðaÞ and
~
JJ
x
¼S
M
m
d
m
~
JJ
xm
ðaÞ.
Using Galerkin’s technique,
~
EE
x
and
~
EE
z
in Eq. (26) can
be eliminated and a matrix equation for the expansion
h
d
y
x
W
Metal top plane
Metal ground plane
0
, = const Region 2
Region 1
µ c
Figure 2. Cross section of a microstrip transmission line.
1236 ELECTROMAGNETIC MODELING
coefﬁcients c
m
and d
m
is found
_
a
~
JJ
z
~
ZZ
zz
~
JJ
z
_
a
~
JJ
z
~
ZZ
zx
~
JJ
x
_
a
~
JJ
x
~
ZZ
xz
~
JJ
z
_
a
~
JJ
1
~
ZZ
xx
~
JJ
x
_ _
.ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ¸¸ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ.
½A
c
d
_ _
¼0 ð27Þ
The propagation constant b is contained in the matrix
elements of
~
ZZ. The transformed currents on the strip are
found from an educated guess of the current distribution
on the strip in the space domain. To describe the current in
the space domain as accurate as possible, basis functions
must be employed that are zero outside the strip area and
also model the singular behavior of the magnetic ﬁeld
components normal to the strip edge. Equation (27) can
then be solved for b by ﬁnding the zeros of the determinant
of [A] (i.e., det[A] ¼0).
A disadvantage of the SDM is that the algorithm is
developed for very speciﬁc transmissionline geometries. If
the geometry changes, for example, substrate regions
become inhomogeneous, or conductor contours do not ﬁt
into a rectangular coordinate system, the SDM algorithm
must be reformulated. This applies also to the previous
methods and can be avoided only if more general methods
are utilized that are based on ﬁniteelement or ﬁnite
difference discretization of Maxwell’s equations or the
Helmholtz equation.
1.5. The Generalized Multipole Method
The generalized multipole technique (GMT) is also based
on the weighted residual technique. It is a unique form of
the method of moments in that the expansion functions
are analytic solutions of the ﬁelds generated by sources a
distance away from the surface where the boundary
condition is being enforced. The GMT is a frequency
domain method for calculating electromagnetic ﬁelds
both in 2D and in 3D. The method is also known as the
multiplemultipole method (MMT) [19]. In the GMT the
ﬁeld domain is separated into a number of subdomains D
i
,
each with linear and homogeneous material. In each D
i
, a
separate expansion of the electromagnetic ﬁeld is given as
E
H
_ _
¼
E
0
H
0
_ _
þ
1
l ¼1
a
l
E
l
H
l
_ _
ð28Þ
In static cases, either the electric or magnetic ﬁeld is
replaced by a potential. E
0
, H
0
in Eq. (28) is a given
excitation. Any choice of the unknown coefﬁcients a
l
results in a correct solution of Maxwell’s equation, since
E
l
and H
l
is such a solution. Thus, all degrees of freedom in
this solution, that is, the set a
l
, may be used to satisfy the
boundary conditions. The GMT is often referred to as a
semianalytical method since the differential equations
(i.e., Maxwell’s equations) in each subdomain D
i
are solved
analytically ( ¼exactly), while the boundary conditions on
the boundaries qD
ij
between subdomains D
i
and D
j
are
solved numerically ( ¼approximately). In order to accel
erate convergence and to keep the number of unknowns as
low as possible, the expansion in each subdomain D
i
is
chosen such that it ﬁts best to the particular shape of D
i
and the particular excitation.
The boundary conditions are fulﬁlled numerically (ap
proximately), using the extended pointmatching techni
que. This is numerically equivalent to both a projection
technique using Galerkin’s choice of test functions and a
leastsquares error minimization. All the boundary condi
tions concerning ﬁelds may be taken into account. As a
special case, surface impedance boundary conditions are
also possible. Also in the GMT/MMT, a system of linear
equations is developed which is solved for the best coefﬁ
cients of the expansion functions. In order to save compu
ter memory and to improve the numerical stability, matrix
solvers like Givens plane rotation may be used.
The difference between the GMT and the MoM is that
the latter normally employs expansion functions that are
located on the boundary representing quantities such as
charge or current. The ﬁelds are then determined by
integrating these quantities over the entire surface. This
integration is not necessary in the GMT since the expan
sion functions are already ﬁeld solutions corresponding to
multiple poles.
Dielectric and conducting boundaries are treated with
the same efﬁciency in the GMT because the same expan
sion functions are used. Therefore, GMT models are quite
general and do not suffer from the limitations of most
MoM models. On the other hand, MoM models that
employ functions optimized for a particular problem, are
generally more efﬁcient than GMT models.
An overview article about GMT was published by
Ludwig [18]. Details of the GMTand application examples
are given in the text by Hafner [19] and Leuchtmann [20].
1.6. The Method of Lines
The method of lines (MoL) is also a semianalytical method
that was developed by Russian mathematicians [21,22] to
solve partialdifferential equations. In this scheme a set of
coupled differential equations is transformed into a set of
ordinary differential equations which can be solved ana
lytically. As such the MoL can also be classiﬁed as a hybrid
method since it combines an analytical approach with the
ﬁnitedifference method. For the microwave domain,
Pregla and coworkers were the ﬁrst to adopt this method,
mainly for the analysis of planar microwave circuits. A
very detailed description of the method is given by Pregla
and Pascher in Ref. 23. The MoL provides, in comparison
to the MMTor the SDM, more ﬂexibility in the analysis of
transmissionline geometries with almost arbitrary cross
section. The only restriction is that at least one space
direction must be amenable to an analytical solution,
which is always true for planar transmission lines.
To illustrate the MoL, it is best to choose a twodimen
sional problem, although threedimensional prob
lems can be solved as well. An example of a cross section
suitable for an MoL analysis is again the microstrip line as
shown in Fig. 2, only this time we consider also the ﬁnite
metallization thickness. The objective is to ﬁnd the effec
tive permittivity e
eff
. Assuming a symmetric structure, the
domain is bounded with a magnetic wall at the symmetry
plane, and with electric walls elsewhere. The discretization
ELECTROMAGNETIC MODELING 1237
of the cross section by lines is shown in Fig. 3 with three
homogeneous subdomains. In each subregion the wave
equation is discretized in x direction with stepsize h, and
analytical solutions are sought in y direction along the
solid and dashed lines shown in the ﬁgure. To be more
precise, analytical solutions for the electric ﬁeld compo
nents are deﬁned on the solid lines (x ¼kh, k¼0,1,2,y) in
order to fulﬁll the Dirichlet boundary condition at x ¼0 by
setting E
tangential
¼0. The magnetic ﬁeld will be deﬁned on
the dashed lines (x ¼khþh/2, k¼0,1,2,y), where the
Neumann boundary condition at x ¼0 can be fulﬁlled to
a ﬁrst order [H
tangential
( þh/2) ¼H
tangential
( Àh/2)]. Similar
considerations apply to the magnetic wall boundary at x ¼
a/2, as well as to subdomain 2 besides the strip. The
discretization in x direction is done by a ﬁnite difference
scheme, and the secondorder derivative at line i can be
written as
@
2
f
@x
2
¸
¸
¸
¸
i
%
f
iÀ1
À2f
i
þf
i þ1
h
2
The coupled differential equations in each subregion are
then found in matrix form as
@
2
/
@y
2
À
1
h
2
½P/þk
2
c
/¼0 ð29Þ
whereby the vector f represents either the longitudinal
electric (E
z
) or magnetic (H
z
) ﬁeld component. The second
order difference operator [P] couples the differential equa
tions of three neighboring lines. Because matrix [P] is
symmetric, decoupling of the differential equations above
is possible by an orthogonal transformation matrix [T]
([T]
À1
¼[T]
t
) such that
"
//¼½T
t
/
The eigenvectors of [P] are the columns of [T]. The dis
cretized wave equations in the transformed domain is
decoupled and written as
@
2 "
//
@y
2
þ k
2
c
À
l
k
h
_ _
2
_ _
"
//¼0 ð30Þ
where l
2
k
are the eigenvalues of [P]. Equation (30) can now
be solved analytically along each line by using trigono
metric functions. In other words, the EM ﬁelds in the
transformed domain can be transformed from one bound
ary of a subregion to the opposite. At the interfaces
between subregions (i.e., at y ¼d and at y ¼dþt, Fig. 3),
the tangential ﬁelds are matched at the points where the
lines cross the interface planes. Depending on the problem
at hand, ﬁeld matching is done either in the transformed
domain or in the space domain. The latter requires full
matrix inversions, which complicates the algorithm. Re
garding the example of the microstrip line, ﬁeld matching
in the space domain becomes necessary due to the ﬁnite
thickness of the strip (i.e., the number of lines to match is
different from one subregion to another). The assumption
of an inﬁnitely thin strip would make possible a ﬁeld
matching in the transformed domain and, therefore, lower
the computational cost of the algorithm by avoiding sev
eral matrix inversions. In any case, an eigenvalue problem
must be solved for k
c
. The size of the eigenvalue matrix
corresponds to the minimum number of lines in one layer:
the number of lines in region 2 besides the strip. The
effective permittivity of the microstrip line is then found
from the wavenumber k
c
.
A signiﬁcant advantage of the MoL over other space
discretization methods such as the ﬁnitedifference
method or the ﬁniteelement method is that a twodimen
sional problem requires only a onedimensional space
discretization. This feature can lower the computational
requirements signiﬁcantly at the cost of a higher analy
tical content.
1.7. The FiniteDifference Method
The ﬁnitedifference method (FDM) is an approximate
method to solve partialdifferential equations. In contrast
to the method of lines, the computational domain is
discretized in all three space directions. The derivative
operations, for example, for the space variable x, qf/qx and
q
2
f/qx
2
are approximated by Df(x)/Dx and D
2
f(x)/Dx
2
, and
thus the partialdifferential equation is reduced to a set of
algebraic equations. In electromagnetics as in other areas
of engineering, the FDM is one of the most important
methods to solve a wide range of problems. These include
linear and nonlinear problems, time and frequencydo
main problems, wave propagation in homogeneous and
inhomogeneous media and in media with different bound
ary conditions. An early example on the application of the
FDM to waveguide problems is given in Refs. 25 and 26.
Detailed chapters on the FDM can be found in books by
Sadiku and Zhou [1,2].
y = d
∆x = h
3
2
1
x = a/2 x = 0
r
y = b
y = 0
y = d + t
w/2
c
Figure 3. Discretization scheme of the method of lines for a
microstrip transmission line with metallic enclosure of dimen
sions a Â b.
1238 ELECTROMAGNETIC MODELING
To illustrate the FDM, we choose a problem that can be
described by the twodimensional Laplace equation
@
2
f
@x
2
þ
@
2
f
@y
2
¼0 ð31Þ
Sampling the continuous electromagnetic ﬁeld in the
computational domain by a mesh of regular points sepa
rated by a constant distance h (Fig. 4), Eq. (31) is
approximated by the difference quotients at the adjacent
mesh points. Depending on the choice of the difference and
the difference quotient, different methods can be used to
derive the discretization formulations. Using a Taylor
expansion, the potentials surrounding the center node o
(Fig. 4) can be expressed as
f
xÀh
¼f
o
Àh
@f
@x
_ _
o
þ
h
2
2!
@
2
f
@x
2
_ _
o
À
h
3
3!
@
3
f
@x
3
_ _
o
þeðh
4
Þ
f
x þh
¼f
o
þh
@f
@x
_ _
o
þ
h
2
2!
@
2
f
@x
2
_ _
o
þ
h
3
3!
@
3
f
@x
3
_ _
o
þeðh
4
Þ
f
yÀh
¼f
o
Àh
@f
@y
_ _
o
þ
h
2
2!
@
2
f
@y
2
_ _
o
À
h
3
3!
@
3
f
@y
3
_ _
o
þeðh
4
Þ
f
y þh
¼f
o
Àh
@f
@y
_ _
o
þ
h
2
2!
@
2
f
@y
2
_ _
o
þ
h
3
3!
@
3
f
@y
3
_ _
o
þeðh
4
Þ
ð32Þ
where e(h
4
) is the remaining error. Adding these equations
and considering that the resulting term
h
2
@
2
f
@x
2
þ
@
2
f
@y
2
_ _
o
¼0
the approximation of f
o
at node center o becomes
1
4
ðf
x þh
þf
xÀh
þf
y þh
þf
yÀh
Þ ¼f
o
whereby it is assumed that e(h
4
) is negligible. This equa
tion shows that the value of f
o
is the average of the
potentials at the four neighboring points. The above
equation for the 2D problem is also said to be the ﬁve
point difference equation of the Laplacian problem. For a
3D problem the above equation expands to a sevenstar
node
1
6
ðf
x þh
þf
xÀh
þf
y þh
þf
yÀh
þf
z þh
þf
zÀh
Þ ¼f
o
Repeating this procedure over the whole computational
domain and considering the boundary conditions on f,
leads to the following matrix equation
½Af ¼X
Since the individual grid points are connected only to their
neighboring points, the coefﬁcient matrix [A] contains a
large number of zero elements (banded sparse matrix) and
only the diagonal and nearby elements are ﬁlled. f is a
vector of all potentials on the interior nodes and X con
tains the information about the boundary conditions (or
sources). Matrix [A] can be solved by the Gauss elimina
tion method but due to the sparsity of the matrix, iterative
methods such as the overrelaxation iteration are more
economical in terms of computer resources.
1.8. FiniteElement Method
Although the ﬁniteelement method (FEM) was used by
mechanical and civil engineers for many years [27], its
application to the electromagnetics area was not before
1967 by Winslow [28] and in 1970 by Silvester and co
workers [29]. Since then, the FEM has become a widely
used numerical simulation tool for electromagnetic ﬁelds
in structures with arbitrary boundary shape [30,31]. In
contrast to the ﬁnitedifference method, the ﬁnite element
method discretizes the computational domain with a
number of small interconnected subregions, called ele
ments (Fig. 5). The shape of these elements is typically
rectangular or triangular. This explains why there are
virtually no restrictions on the shape of the structures
that can be analyzed with the FEM. This feature is its
main advantage over other methods.
The FEM is based on the fact that the potential
function f
e
(superscript ‘‘e’’ denotes element) within each
element, can be approximated by an (often linear) inter
polation function which is zero outside the element.
Summation of f
e
over all elements Ngives an approximate
y
x – h
y – h
y + h
y
x x + h
x
r
W
O
c
Figure 4. Finitedifference discretization of a microstrip trans
mission line.
ELECTROMAGNETIC MODELING 1239
solution for the total potential in the computational
domain
f¼
N
p
f
e
ðx; yÞ ð33Þ
The simplest form of approximation for f
e
within a
triangular element (Fig. 5) is the following:
f
e
¼aþbx þcy ð34Þ
It is assumed that the ﬁeld strength is uniform within a
small element and that the potential varies linearly
depending on the coordinates x and y. The unknown
parameters a, b, c are found from the nodal parameters
f
e
p
; x
p
, and y
p
(p¼1, 2, 3) as
a
b
c
_
¸
¸
_
_
¸
¸
_
¼
1 x
1
y
1
1 x
2
y
2
1 x
3
y
3
_
¸
¸
_
_
¸
¸
_
À1
f
e
1
f
e
2
f
e
3
_
¸
¸
_
_
¸
¸
_
ð35Þ
Substituting the functions obtained from Eq. (35) into
Eq. (34) yields
f
e
¼
p
N
e
p
f
e
p
ð36Þ
N
e
p
are the socalled shape functions. The potential f
e
within each element is thus a linear combination of the
shape functions and the three nodal values of the triangle.
The next step in the FEM is to determine the potential
at the corners of all elements. This is usually done by
minimizing (or maximizing) a functional that is known to
be stationary about the true solution (variational method).
For a Laplacian problem, the equivalent functional for
each element is
I
e
ðf
e
Þ ¼
_
s
e
1
2
ejrf
e
j
2
dx dy ð37Þ
The functional over all elements is then the sum over
Eq. (37). From a physical point of view, I
e
is the energy per
unit length of the element e. Substituting the approxima
tion for f
e
into Eq. (37) yields
I
e
¼
1
2
3
p¼1
3
k¼1
ef
e
p
_
s
e
rN
e
p
.
rN
e
k
dx dy
_ _
f
e
k
.ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ¸¸ﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄﬄ.
C
pk
ð38Þ
or in matrix form
I
e
¼
1
2
e½f
e
T
½C
e
½f
e
ð39Þ
[C
e
] is the element coefﬁcient matrix (stiffness matrix in
structural analysis). The matrix elements represent the
coupling between the nodes. Summation of Eq. (39) over
all the elements and applying the extremum condition of
the functional
@IðfÞ
@f
i
¼0
yields the system matrix equation
½C½f ¼0 ð40Þ
[C] is a sparse, symmetric, and banded matrix of size
M Â M (M¼total number of nodes). Equation (40) can be
solved for the potentials of all nodes in the computational
domain.
The FEM is one of the most ﬂexible numerical modeling
approaches which can be applied to nonlinear problems
and also can be formulated in time and frequency domain.
Although its formulation is more involved than the ﬁnite
difference method the advantage of the FEM is that it can
be applied to almost arbitrarily shaped boundaries. Appli
cation to open boundaries is difﬁcult except if the FEM is
combined with other methods which are more suitable for
open boundary problems.
1.9. The BoundaryElement Method
The boundaryelement method (BEM) or boundary inte
gral element method is similar to the ﬁniteelement
method applied to the boundary only. As such the BEM
is also known as a form of weighted residual technique
that falls under the category of moment methods. The
main advantage of the BEM over the FEM is that it
reduces the dimension of the problem by one. For example,
in a threedimensional problem only the surface of the
computational domain needs to be discretized and not the
entire volume, which leads to a much smaller number of
algebraic equations. For twodimensional problems the
boundary elements are taken as straight line segments,
φ
X
Y
e
3
3
2 1
φ
e
2
φ
e
1
Figure 5. Finiteelement discretization of a ridged waveguide.
1240 ELECTROMAGNETIC MODELING
whereas for threedimensional problems triangular ele
ments are taken. Disadvantages of the BEM are that the
number of integrations required are great and that singu
larities must be considered. The calculation of the coefﬁ
cient matrix may require more time than in the FEM.
A very detailed account of the BEM may be found in
Ref. 2. In the BEM the quantity of interest f anywhere
within the computational domain is expressed in terms of
a functional, except that this functional now depends only
on the value of f at the boundary and its normal deriva
tive thereon. Again, for the Laplace equation in a source
free volume O bounded by a surface G, for example, the
potential f
i
inside the volume is given by the integral
equation
f
i
¼
1
4p
_
G
1
r
@f
@n
Àf
@
@n
1
r
_ _ _ _
dG
In most of the relevant EM problems, the governing
equation is not Laplacian and f on the boundary is
unknown. Then the boundary contour is discretized and
f for each boundary element is derived using the method
of weighted residuals where the expansion and weighting
functions are only deﬁned within the boundary cell. The
integral equation is thus transformed into a set of alge
braic equations at the nodes of the boundary, and the
value for f and its derivatives are found simultaneously
by solving a matrix equation [2].
2. TIMEDOMAIN METHODS
Timedomain methods are important if the timedomain
response of an electromagnetic structure is required.
Lately, timedomain methods have also gained momentum
over frequencydomain methods since they deliver, de
pending on the excitation, all frequencies of interest
with one computation run without the need for large
matrix inversions. This feature is attractive if a wide
band frequency response is required. One might also add
that a variety of problems are more naturally formulated
in the time domain than in the frequency domain. This is
in particular the case for nonlinear problems.
Some of the frequencydomain methods discussed be
fore can be formulated also in the time domain. This is a
great advantage if one is already familiar with a particu
lar method. Extending the formulation from one domain
into the other without leaving the framework of one
method not only minimizes the development effort but
also expands the application range of that method. Among
the many different techniques that can be formulated in
both domains, the ﬁnitedifference timedomain (FDTD)
method and the timedomain transmissionline matrix
(TDTLM) method are the most prominent ones. While
the ﬁnitedifference method in the frequency domain, the
FDM or FDFDM, has been discussed in a previous section,
the frequencydomain version of the TLM method, the
FDTLM method [32], is not discussed here. However, as in
the framework of the ﬁnitedifference methods the duality
between frequency and timedomain methods exists also
in the framework of the TLM method.
In the FDTD and TDTLM methods of electromagnetic
modeling the continuous ﬁeld functions that satisfy Max
well’s equations are approximated by samples of these
functions deﬁned only at discrete points in space and time.
In the most general sense, FDTD and TLM belong to the
methodofmoments (MoM) family. In FDTD the electro
magnetic ﬁeld is approximated by a set of local pulse
functions in space and time, while in TLM it is expressed
as a superposition of impulse waves traveling forward and
backward along the coordinate directions, their sum yield
ing the electric and their difference the magnetic ﬁeld
values, respectively. Thus, FDTD is formulated in terms of
total electric and magnetic ﬁeld samples in discretized
space, whereas the TLM formulation employs elementary
incident and reﬂected waves traveling on a mesh of
transmission lines (scattering formulation).
While both methods can be derived rigorously from
Maxwell’s equations using MoM formalism [33], a more
intuitive approach that is also historically authentic will
be used to formulate the basic FDTD and TLM algorithms.
Their properties and associated errors will be discussed,
and some more recent variations will be mentioned.
2.1. FiniteDifference TimeDomain Method
The ﬁnitedifference timedomain (FDTD) scheme is ob
tained by replacing the partial derivatives (space and
time) in Maxwell’s curl equations by ﬁnite differences.
The best approximation is obtained by central differencing
(trapezoidal rule), resulting in an error that is propor
tional to the square of the space and time step (second
order accuracy).
The ﬁrst FDTD formulation was proposed by Yee in
1966 [34] and subsequently applied and developed further
by Taﬂove and Brodwin [35]. Yee simply replaced the
partial derivatives in Maxwell’s curl equations by central
ﬁnite differences. Weiland [36] derived an equivalent
discretization approach using ﬁnite integration of Max
well’s equations in 1977. Fig. 6 shows a unit FDTD cell
(Yee cell) of a Cartesian space grid. Continuous space and
time coordinates (x,y,z,t) are replaced by discrete coordi
nates l Dx, mDy, n Dz, k Dt, where l,m,n,k are integers and
H
z
H
x
H
x
H
z
E
z
E
z
E
z
E
z
E
x
E
x
E
x
E
x
E
y
E
y
E
y
E
y
H
y
H
y
z/∆ z
x/∆ x
y/∆ y
(l,m + 1, n)
(l,m, n + 1)
(l,m,n)
(l + 1,m + 1, n) (l + 1,m,n)
Figure 6. Topology of the elementary FDTD cell (Yee cell).
Electric and magnetic ﬁeld components are interleaved in space
and time.
ELECTROMAGNETIC MODELING 1241
Dx, Dy, Dz, and Dt are the space/timesteps. Note that the
three electric ﬁeld components are deﬁned along the edges
of the cell, while the magnetic ﬁeld components are
normal to the cell faces. The staggering of the ﬁeld
components by onehalf of the cell dimensions is due to
the centraldifference approximation of the differential
operators. For the same reason, electric and magnetic
ﬁeld components are also staggered in time, the electric
ﬁeld components being deﬁned at time points k Dt and the
magnetic ﬁeld components at ðk þ
1
2
Þ Dt.
If we assume that Dx ¼p Dl; Dx ¼q Dl; Dx ¼r Dl, where
Dl is the unit reference length, and the scaling coefﬁcients
p, q, and r are all smaller or equal to unity, then the ﬁnite
difference equations for the electric and magnetic ﬁeld
components in each cell are given by
k þ1
E
x
ðl þ
1
2
; m; nÞ ¼
k
E
x
ðl þ
1
2
; m; nÞ
þsxf½
k þ1=2
H
z
ðl þ
1
2
; mþ
1
2
; nÞ
À
kþ1=2
H
z
ðl þ
1
2
; mÀ
1
2
; nÞ=q
þ½
k þ1=2
H
y
ðl þ
1
2
; m; n À
1
2
Þ
À
kþ1=2
H
y
ðl þ
1
2
; m; nþ
1
2
Þ=rg
kþ1
E
y
ðl; mþ
1
2
; nÞ ¼
k
E
y
ðl; mþ
1
2
; nÞ
þsyf½
kþ1=2
H
x
ðl; mþ
1
2
; nþ
1
2
Þ
À
k þ1=2
H
x
ðl; mþ
1
2
; n À
1
2
Þ=r
þ½
kþ1=2
H
z
ðl À
1
2
; m; nþ
1
2
; nÞ
À
k þ1=2
H
z
ðl þ
1
2
; mþ
1
2
; nÞ=pg
kþ1
E
z
ðl; m; nþ
1
2
Þ ¼
k
E
z
ðl; m; nþ
1
2
Þ
þszf½
kþ1=2
H
x
ðl; mÀ
1
2
; nþ
1
2
Þ
À
kþ1=2
H
x
ðl; mþ
1
2
; nþ
1
2
Þ=q
þ½
k þ1=2
H
y
ðl þ
1
2
; m; nþ
1
2
Þ
À
kþ1=2
H
y
ðl À
1
2
; m; nþ
1
2
Þ=pg
k þ1=2
H
x
ðl; mþ
1
2
; nþ
1
2
Þ ¼
kÀ1=2
H
x
ðl; mþ
1
2
; nþ
1
2
Þ
þsx
0
f½
k
E
y
ðl; mþ
1
2
; nþ1Þ
À
k
E
y
ðl; mþ
1
2
; nÞ=p
þ½
k
E
z
ðl; m; nþ
1
2
Þ
À
k
E
z
ðl; mþ1; nþ
1
2
Þ=qg
kþ1=2
H
y
ðl þ
1
2
; m; nþ
1
2
Þ ¼
kÀ1=2
H
y
ðl þ
1
2
; m; nþ
1
2
Þ
þsy
0
f½
k
E
x
ðl þ
1
2
; m; nÞ
À
k
E
x
ðl þ
1
2
; m; nþ1Þ=r
þ½
k
E
z
ðl þ1; m; nþ
1
2
Þ
À
k
E
z
ðl; m; nþ
1
2
Þ=pg
k þ1=2
H
z
ðl þ
1
2
; mþ
1
2
; nÞ ¼
kÀ1=2
H
z
ðl þ
1
2
; mþ
1
2
; nÞ
þsz
0
f½
k
E
x
ðl þ
1
2
; mþ1; nÞ
À
k
E
x
ðl þ
1
2
; m; nÞ=q
þ½
k
E
y
ðl; mþ
1
2
; nÞ
À
k
E
y
ðl þ1; mþ
1
2
; nÞ=rg
ð41Þ
where
sx ¼Z
0
cDt=ðe
rx
DlÞ
sy ¼Z
0
cDt=ðe
ry
DlÞ
sz ¼Z
0
cDt=ðe
rz
DlÞ
sx
0
¼cDt=ðm
rx
Z
0
DlÞ
sy
0
¼cDt=ðm
ry
Z
0
DlÞ
sz
0
¼cDt=ðm
rz
Z
0
DlÞ
ð42Þ
In these expressions, c and Z
o
are the velocity of light and
the wave impedance in vacuo, and e
rx
, e
ry
, e
rz
and m
rx
, m
ry
,
m
rz
are the diagonal elements of the relative permittivity
and permeability tensors of the medium, respectively. This
algorithm explicitly updates each ﬁeld component in a
leapfrog timestepping process. The change in each Eﬁeld
component is computed from the four Hﬁeld components
circulating around it, and vice versa.
2.1.1. Stability. The process is stable as long as the
timestep is smaller than a maximum value known as the
socalled Courant stability limit. For electrically and
magnetically isotropic media characterized by e
r
and m
r
the stability criterion is
Dt
Dl
ﬃﬃﬃﬃﬃﬃﬃﬃﬃ
m
r
e
r
p
c
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
1
p
2
þ
1
q
2
þ
1
r
2
_ ð43Þ
Since in anisotropic media the wave velocity depends on
the (generally unknown) polarization, it is prudent to
enter the smallest of the three m and e values of the
diagonal tensors into the stability condition. For free space
discretized into cubic cells (m
r
¼e
r
¼p¼q¼r ¼1), it
1242 ELECTROMAGNETIC MODELING
becomes
Dt
Dl
c
ﬃﬃﬃ
3
p ð44Þ
2.1.2. Initial and Boundary Conditions. At the start of a
computation the initial values of all electric and/or mag
netic ﬁeld components in the computational domain are
speciﬁed before the updating process can begin. By enfor
cing the ﬁeld values in certain regions at each timestep,
source functions with arbitrary time and space depen
dence can be modeled.
Boundary conditions must be enforced at each timestep
as well. Electric and magnetic walls are modeled by
extending the discretized space one cell beyond the bound
ary and imposing appropriate symmetry conditions on the
ﬁeld values on each side of the boundary. For example, the
tangential electric ﬁeld components must be identical on
either side of a magnetic wall (ideal open circuit) and
equal and opposite on either side of an electric wall (ideal
short circuit). A dual condition applies to the tangential
magnetic ﬁeld components. Lossy resistive boundary con
ditions call for a ﬁxed ratio between the tangential electric
and magnetic ﬁeld components at the boundary. More
complex boundary conditions such as wide band absorbing
walls or frequency dispersive boundaries call for special
algorithms, such as oneway absorbing boundary condi
tions [37,38] or Berenger’s perfectly matched layer [39]. A
detailed discussion of absorbing boundary conditions in
FDTD and TLM can be found in Ref. 40. Similar ap
proaches are required for the modeling of complex materi
als and devices.
The books by Kunz and Luebbers [41] and Taﬂove [42]
are excellent sources of information on all aspects of
FDTD modeling and contain extensive bibliographies on
the theory, implementation, and application of the FDTD
method. Together with Yee’s seminal paper [34] they are
good starting points for exploring the extensive literature
on FDTD theory and applications.
2.2. TransmissionLine Matrix Method
2.2.1. The Expanded Node. The transmissionline ma
trix (TLM) formulation of Maxwell’s equations was ﬁrst
proposed in 1971 by Johns and Beurle [43]. In their
seminal paper they describe a novel numerical technique
for solving twodimensional scattering problems. Inspired
by earlier network simulation techniques [44], they em
ploy a Cartesian mesh of shuntconnected twowire trans
mission lines as a discretized 2D propagation medium.
The nodes of this mesh act as scattering centers for short
voltage impulses. Johns and Akhtarzad [45] extended the
method to three space dimensions (the expanded node
TLM model) in 1974, by creating an intricate 3D lattice of
shunt and seriesconnected transmission lines, as shown
in Fig. 7. This model is, in many respects, similar to the
Yee cell in Fig. 6 since it yields identical solutions for the
six ﬁeld components when the time step in the Yee
algorithm is set to Dt ¼Dl/(2c) (free space, cubic cell).
However, in contrast to the strictly mathematical formu
lation of FDTD, the TLM model is a ‘‘hardwired’’ network
(albeit conceptual rather than material) to which all
known techniques of circuit and transmission line analy
sis can be applied in both frequency and time domains.
2.2.2. The Symmetric Condensed Node. One of the
shortcomings of these algorithms resides in the compli
cated topology of their unit cells and in the separate
locations of electric and magnetic ﬁeld components in
space and time. This makes the modeling of complex
boundary conditions and interfaces between materials
more difﬁcult and may introduce errors. To overcome these
drawbacks, Johns [46,47] introduced the symmetrical
condensed TLM node in 1986. This spawned the develop
ment of several new TLM formulations, from the hybrid
and supercondensed nodes to the alternating and rotated
alternating [48] TLM models. In the following, the basic
formulation proposed by Johns will be outlined. However,
the port numbering scheme proposed by Russer [48] will
be used since it allows a simpler and more compact
representation of the TLM algorithm than Johns’ original
numbering scheme.
2.2.3. The Symmetric Condensed Node TLM Algor
ithm. A unit cell of the symmetric condensed TLM model
is shown in Fig. 8. It contains a hybrid junction of 12
transmission lines (the node) that is characterized by a
12 Â 12 scattering matrix. The timedomain TLM algo
rithm is executed in two steps. First, 12 short voltage
pulses are simultaneously injected into the node ports 1–
12. The pulses are scattered and give rise to 12 reﬂected
voltage pulses. Second, the reﬂected pulses are trans
ferred to the neighboring nodes where they become in
cident pulses at the subsequent timestep, and the process
is repeated. This series of events can be described in
symbolic form as follows
½
k
v
r
¼S
.
½
k
v
i
; ½
k þ1
v
i
¼C
.
½
k
v
r
ð45Þ
H
x
H
z
E
z
E
z
E
z
E
x
E
x
E
x
E
y
E
y
E
y
H
y
z/∆ z
x/∆ x
y/∆ y
(l,m + 1, n)
(l,m, n + 1)
(l + 1,m + 1, n) (l + 1,m,n)
Figure 7. Topology of the expanded TLM node. Electric ﬁeld
components are modeled by the voltage across shunt connections,
while magnetic ﬁeld components are modeled by the loop current
in series connections of transmission lines. The positions of the
ﬁeld components in space and time are identical to those in the
Yee cell.
ELECTROMAGNETIC MODELING 1243
where ½
k
v
r
and ½
k
v
i
are the vectors of reﬂected and
incident pulses at the kth timestep, S is the impulse
scattering matrix of the node, and Cis a connection matrix
describing the topology of the network. It governs the
transfer of the reﬂected pulses to the connected ports of
the neighboring cells and/or the reﬂection from bound
aries. The subscripts k and k þ1 denote the discrete time
points at which the pulses are scattered at the nodes.
For a homogeneous, lossless, and isotropic medium, all
transmission lines of a cubic cell have the same character
istic impedance. The 12Â12 scattering matrix S is then
S¼
0 S
0
S
T
0
S
T
0
0 S
0
S
0
S
T
0
0
_
¸
¸
¸
_
_
¸
¸
¸
_
ð46Þ
where the submatrix S
0
is given by
S
0
¼
0 0
1
2
À
1
2
0 0 À
1
2
1
2
1
2
1
2
0 0
1
2
1
2
0 0
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
_
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
_
ð47Þ
Since the transit time Dt of the pulses is linked to the
space step Dl by the pulse propagation velocity along the
transmission lines, the TLM process is unconditionally
stable. The time step is automatically set to Dt ¼Dl/(2c).
2.2.4. Inhomogeneous Materials and Losses. Dielectric or
magnetic materials can be modeled by loading the nodes
situated inside these materials with reactive shunt stubs
of appropriate normalized characteristic admittance and a
length Dl/2 [46]. An opencircuited shunt stub will produce
the effect of additional capacitance at the node, while a
shortcircuited series stub creates additional inductance.
The resulting storage of reactive energy reduces the phase
velocity and alters the intrinsic impedance in the struc
ture. The interface conditions at the boundary between
different materials are automatically fulﬁlled. Each cell
can have a different set of stubs (three permittivity and
three permeability stubs), thus allowing the modeling of
inhomogeneous anisotropic materials with diagonal per
mittivity and permeability tensors. The six stubs add six
more ports to the node, and as a result, the S becomes an
18 Â 18 matrix. Losses can be modeled by connecting so
called loss stubs to the nodes. The loss stubs are matched
transmission line sections that extract a fraction of the
energy scattered at the node at each time step. Since no
pulses travel back into the nodes on these stubs, they only
modify the elements of S without increasing its size.
2.2.5. Initial and Boundary Conditions. At the start of a
computation the initial values of all pulses incident on all
ﬁeld components are uniquely determined in the center of
the nodes by a linear combination of these pulses at the
moment of scattering (47). When the pulses transit from
one cell to the next (t/Dt ¼k þ1/2) the tangential compo
nents of the ﬁelds are obtained in the cell boundaries as
well. By enforcing the pulse values (and hence the corre
sponding electric and magnetic ﬁeld values) in certain
regions at each timestep, source functions with arbitrary
time and space dependence can be modeled.
Boundary conditions can be imposed either in the
center of the nodes or in the cell boundaries. In the latter
case, boundaries are represented by means of impulse
reﬂection coefﬁcients. Electric walls reﬂect pulses with
‘‘aÀ1’’ reﬂection coefﬁcient, while magnetic walls have ‘‘a
þ1’’ reﬂection coefﬁcient. Lossy resistive boundaries have
impulse reﬂection coefﬁcients less than unity in magni
tude. More complex boundary conditions such as wide
band absorbing walls or frequency dispersive boundaries
are treated in the same way as FDTD boundaries with the
difference that the boundary operators are applied to
the incident pulses rather than to the ﬁeld quantities at
the boundaries. It is straightforward to implement
nonrecursive and recursive convolution techniques for
the modeling of frequency dispersive boundaries and for
partitioning large computational domains using time
domain diakoptics [49]. Similar approaches are required
for the modeling of complex materials and devices [50].
Johns’ seminal papers [43,45–47] are good starting points
for exploring the world of TLM modeling, as are an intro
ductory chapter on TLM by Hoefer [51] and a book by
Christopoulos [52]. They contain many references and de
scribe the implementation and applications of TLMin detail.
3. HYBRID METHODS
None of the previous methods is capable of solving all
electromagnetic modeling problems. The methods are
10
11
z
y
x
12
3
8
6
9
1
4
7
2
5
Figure 8. Topology of the symmetric condensed TLM node. All
six electric and magnetic ﬁeld components are deﬁned in the
center of the node, and tangential ﬁeld components are deﬁned in
the cell boundaries as well.
1244 ELECTROMAGNETIC MODELING
either limited by the available computer memory and/or
by computer runtime, or the numerical model can simply
not be applied to the structure at hand. For example, the
method of moments is not applicable to structures with
inhomogeneous or nonlinear dielectrics. The ﬁnite differ
ence method is difﬁcult to implement when ﬁne circuit
details must be resolved within a structure of large
dimensions. The discretization size chosen for the smallest
circuit detail determines the total number of discretiza
tion cells and thus the total matrix size to be handled may
become too large. This problem is even more pronounced
for the timedomain version of the ﬁnitedifference
method, the FDTD, since here also the time is discretized.
Similarly, the ﬁniteelement method cannot efﬁciently
model large radiation problems because of the large
computational space that must be discretized.
In practice, many of these—and more—complicating
factors are encountered. Not all appear in the same
problem and at the same time, thus making it possible
to choose one electromagnetic modeling approach over the
other. But there is a signiﬁcant number of problems (and
others are emerging) that cannot be solved within the
framework of any of the previous methods.
A solution to those problems is possible by combining
two or more techniques. The task is to apply each method
to the problem domain for which it is best suited. There
are two possible approaches. A two or more step procedure
is utilized in which one part of the problem is solved by one
method and the results are used as input data to solve
remaining parts of the problem with other techniques.
That requires that boundary conditions are established
that must be enforced at the interfaces between the
different regions. This approach is called an explicit
hybrid approach. An implicit hybrid approach is one in
which the advantageous features of one technique are
combined with those of another technique to form a new
standalone algorithm.
Several successful implementations of hybrid methods
have been reported in the literature. Although this re
search direction does not replace the effort to improve
existing single methods, hybrid methods offer electromag
netic modeling of a whole new class of problems and may
in particular be important in the area of CAD.
3.1. Combinations of FrequencyDomain Methods
3.1.1. GTD and MoM. The ﬁrst combinations of elec
tromagnetic modeling approaches appeared in the analy
sis of antenna problems and radar cross sections. The
geometric theory of diffraction (GTD) and the method of
moments were used to analyze antenna problems in Ref.
53. The GTD is an extension to geometric optics, which
includes the effect of diffraction. This method is only
accurate if the dimensions of the object being analyzed
is large compared to the wavelength of the ﬁeld. For that
reason this method is also called a highfrequency
method. In the combination of the GTD and the MoM,
the latter is used to solve the region close to the antenna,
while the GTD is used for the free space surrounding the
antenna.
3.1.2. FEM and GSM Technique. A combination of the
ﬁniteelement method (FEM) and the generalized scatter
ing matrix (GSM) technique was utilized in the study of
scattering from jet engines [54]. The FEM was applied to
the complex part of the scatterer to generate the GSM at
its boundary which can then be interfaced with high
frequency techniques for computation of the engine’s
scattered ﬁelds without reference to the geometry of the
jet engine. One of the methods alone would not have been
able to solve this complex problem.
3.1.3. FEM/FDM and MMT. For the analysis of micro
wave circuits the combination of the FEM and the mode
matching technique (MMT) has been proposed to study
large cavities [55]. In Ref. 56 the FEM was applied to
analyze waveguide discontinuities with arbitrary bound
ary shape. In Ref. 57 the FDM was employed instead of the
FEM to analyze segments of waveguide structures that
are not suitable for a MMTanalysis, for example a circular
stub in a rectangular waveguide. In all these papers the
MMT was used to characterize the uniform sections of the
waveguide while the FEM or FDM was employed to
analyze rounded corners or discontinuity shapes that do
not ﬁt into the coordinate system of the MMT. The latter
was then used to derive the scattering parameters of the
overall circuit.
3.1.4. Method of Lines and SDM. For the 3D analysis of
planar waveguide problems a combination of the method
of lines and the spectraldomain method was introduced in
Ref. 58. The purpose of this combination was to eliminate
some of the problems associated with the 2D MoL and the
2D SDM. The problem in the latter was the difﬁculty to
ﬁnd 2D basis functions that converge easily, while for the
2D MoL a 2D discretization may not always be able to
satisfy all boundary conditions simultaneously with effor
table computer memory. The combination of the computa
tionally very efﬁcient 1D SDM in transverse direction of
the propagating wave with the equally efﬁcient 1D MoL in
propagation direction eliminates these problems.
3.2. Combinations of Time and FrequencyDomain Methods
3.2.1. Hybrid FiniteDifference–TimeDomain Method.
The hybrid ﬁnitedifference–timedomain (HFDTD)
method is a combination of frequencydomain and time
domain concepts. In its widest sense, the technique uti
lizes a standard FDTD mesh in the areas of structure
inhomogenity and expansion into a known set of modes in
transversely homogeneous regions of the structure. This
provides substantial savings both in terms of computer
memory and CPU time as was ﬁrst demonstrated in the
eigenvalue analysis of planar transmission lines [60]. A
conventional FDTD analysis of such a structure requires a
3D mesh, which, depending on the space resolution re
quired, needs several thousand time iterations before a
Fourier transform can provide the results for the propaga
tion constant. By replacing the space discretization in
propagation direction (z) by a simple phase shift (note
that the ﬁeld at location l
z
is different from that at location
l
z
þz by only a factor e
Àjbz
), results in a 2D FDTD mesh
ELECTROMAGNETIC MODELING 1245
[59]. Multiplying the ﬁeld equations furthermore by a
factor j such that
E
x
; E
y
; H
z
¼jðE
x
; E
y
; H
z
Þe
Àjbz
H
x
; H
y
; E
z
¼ðH
x
; H
y
; E
z
Þe
Àjbz
leads to discretized Maxwell’s equations without complex
quantities. This feature accelerates the computation con
siderably. Exciting the 2D mesh with a time domain
impulse requires much less time iterations for the impulse
to settle than in the case of a 3D mesh. A Fourier trans
form provides the frequency at which the assumed value
of b is valid.
The same principle can be applied to the time domain
TLM method, and also here signiﬁcant savings in compu
ter resources are possible if only the propagation constant
and related quantities (characteristic impedance, losses)
are of interest.
3.2.2. TDTLM and MMT. In the TDTLM analysis of
complex cascaded discontinuity problems, diakoptics is
used to subdivide the problem into simpler subsections
that are then modeled individually. Interconnecting the
individual solutions requires a node to node convolution at
the interface. This approach tends to be computationally
quite demanding since the number of convolutions in
creases with N
2
(N is the number of branches of interest).
To reduce the computational effort, it was suggested in
Ref. 61 that the uniform sections of the problem domain
are modeled by modal functions treated in the time
domain (timedomain Green function), while the disconti
nuity region is represented by the TDTLM method for
which the incident ﬁelds are superpositions of those
modes. The response of that subvolume to an excitation
is obtained by convolution of the excitation with the time
domain Green function. This approach leads to a signiﬁ
cant reduction of the computational resources as com
pared to the analysis with only the TDTLM. Furthermore,
the complex discontinuity region is now represented by its
generalized scattering matrix (GSM), which makes it easy
to cascade discontinuities.
3.3. Combinations of TimeDomain Methods
3.3.1. FDTD and FEM. The FDTD method is well suited
for applications in Cartesian coordinates. However, as
soon as mixed coordinates are necessary to describe the
problem contour, a staircase approximation must be uti
lized. For example, a round structure within a rectangular
mesh layout can be described accurately only by a ﬁne
staircase approximation. This leads to a very ﬁne mesh
and consequently a small timestep to satisfy the stability
condition. The computational effort to calculate such a
structure with acceptable accuracy becomes prohibitive.
An alternative solution is to model the arbitrary boundary
with the ﬁniteelement method and incorporate this ap
proach in the FDTD method which is applied elsewhere in
the problem domain [62].
The list of methods that have been combined can be
continued and new combinations appear every month in
the various periodicals. A good starting point to ﬁnd out
more about hybrid methods and the rationale behind their
combinations is in Ref. 63.
4. SUMMARY
The tremendous increase of computer power since the late
1980s has inspired a new era in the ﬁeld of electromag
netic modeling or computational electromagnetics. Nu
merical codes that ran only on supercomputers
yesterday are running on workstation computers today.
This development will not stop here, and the electromag
netic modeling problem that appears to be inaccessible by
any of today’s available codes (because the required com
puter resources are just too large) will be solvable with
tomorrow’s computers and the then available (unlimited?)
computer memory.
Because of this rapid development in computer hard
ware, electromagnetic modeling has become commonplace
in the world of electrical engineers. Many of the methods
that we have brieﬂy described are already implemented in
commercial simulation tools used in practice to analyze a
wide variety of problems and to design (CAD) a wide range
of circuits and components that would not function other
wise.
We have divided the methods into frequencydomain,
timedomain, and hybrid methods. They all have advan
tages and disadvantages depending on the problem range
they are applied to. The aim of this article was not to
provide extensive information on all of these methods but
to introduce the reader to those numerical methods that
have a broad enough application as well as to provide key
references for further reading.
5. LITERATURE
The literature on electromagnetic ﬁeld modeling is quite
extensive, and only a few key references could be cited
here. For more information on the subject the reader is
referred to the books by R. Sorrentino [65] and E. Miller et
al. [66], who have assembled a collection of reprints of key
papers on the subject, and the book by E. Yamashita [64].
Furthermore, various journals and conferences are de
voted to the topic of numerical modeling of electromag
netic ﬁelds.
BIBLIOGRAPHY
1. Sadiku, Numerical Techniques in Electromagnetics, CRC
Press, Boca Raton, FL, 1992.
2. P. B. Zhou, Numerical Analysis of Electromagnetic Fields,
SpringerVerlag, New York, 1993.
3. T. Itoh, Numerical Techniques for Microwave and Millimeter
Wave Passive Structures, Wiley, New York, 1989.
4. G. S. Mikhlin, Variational Methods in Mathematical Physics,
Pergamon Press, New York, 1968.
5. R. F. Harrington, Field Computation by Moment Methods,
Krieger, Malabar, FL, 1968.
1246 ELECTROMAGNETIC MODELING
6. A. Wexler, Solution of waveguide discontinuities by modal
analysis, IEEE Trans. Microwave Theory Tech. MTT15:508–
517 (1967).
7. R. Mittra and W. W. Lee, Analytical Techniques in the Theory
of Guided Waves, Macmillan, New York, 1971.
8. R. Vahldieck, Accurate hybrid mode analysis of various ﬁnline
conﬁgurations including multilayered dielectrics, ﬁnite me
tallization thickness and substrate holding grooves, IEEE
Trans. Microwave Theory Tech. MTT32:1454–1460 (1984).
9. R. R. Mansour and R. McPhie, A uniﬁed hybridmode analysis
for planar transmission lines with multilayer isotropic/aniso
tropic substrates, IEEE Trans. Microwave Theory Tech. MTT
35:1382–1391 (1987).
10. Y. C. Shih, The modematching method, in T. Itoh, ed.,
Numerical Techniques for Passive Microwave and Milli
meterWave Structures, Wiley, New York, 1989, Chap. 9.
11. J. Bornemann and F. Arndt, Modal smatrix design of opti
mum, stepped, ridged and ﬁnned waveguide transformers,
IEEE Trans. Microwave Theory Tech. MTT35:561–567
(1987).
12. S. Amari, J. Bornemann, and R. Vahldieck, Accurate analysis
of scattering from multiple waveguide discontinuities using
the coupledintegral equation technique, J. Electromagn.
Waves Appl. 10:1623–1644 (1996).
13. S. Amari, J. Bornemann, and R. Vahldieck, Accurate and fast
analysis of waveguide ﬁlters by the coupledintegral equation
technique, IEEE Trans. Microwave Theory Tech. MTT
45:1611–1618 (1997).
14. T. Itoh and R. Mittra, Spectraldomain approach for calculat
ing the dispersion characteristics of microstrip lines, IEEE
Trans. Microwave Theory Tech. MTT21:496–499 (1973).
15. T. Uwaro and T. Itoh, Spectral domain approach, in T. Itoh,
ed., Numerical Techniques for Microwave and Millimeter
Wave Passive Structures, Wiley, New York, 1989, Chap. 5.
16. T. Kitazawa, Metallization thickness effect of striplines with
anisotropic media: quasistatic and hybrid mode analysis,
IEEE Trans. Microwave Theory Tech. MTT37:769–775
(1989).
17. T. Kitazawa and R. Mittra, Analysis of ﬁnline with ﬁnite
metallization thickness, IEEE Trans. Microwave Theory Tech.
MTT32:1484–1487 (1984).
18. A. Ludwig, A new technique for numerical electromagnetics,
IEEE APS Newslett. 31:40–41 (Feb. 1989).
19. C. Hafner and L. Bomholt, The 3D Electrodynamic Wave
Simulator, Wiley, New York, 1993.
20. P. Leuchtmann, The multiple multipole program (MMP):
theory, practical use and latest features, ACES, March 1995
(121 pp.; available from author at leuchtmann
@ifh.ee.ethz.ch.).
21. O. A. Liskovets, The method of lines, Diff. Uravneniya,
1:1662–1678 (1965) (review).
22. B. P. Demidowitsch et al., Numerical Methods of Analysis,
VEB Wissenschaften, Berlin, 1968, Chap. 5 (in German).
23. R. Pregla and W. Pascher, The method of lines, in T. Itoh, ed.,
Numerical Techniques for Microwave and Millimeter—Wave
Passive Structures, Wiley, New York, 1989, Chap. 6, pp. 381–
446.
24. S. Xiao, R. Vahldieck, and J. Hesselbarth, Analysis of cylind
rical transmission lines with the method of lines, IEEE Trans.
Microwave Theory Tech. MTT44:993–999 (1996).
25. G. Mur, Finite difference method for the solution of electro
magnetic waveguide discontinuity problem, IEEE Trans.
Microwave Theory Tech. MTT22:54–57 (1974).
26. H. E. Green, The numerical solution of some important
transmissionline problems, IEEE Trans. Microwave Theory
Tech. MIT13:676–692 (1965).
27. O. C. Zienkiewicz and Y. K. Cheung, The Finite Element
Method in Structural and Continuum Mechanics, McGraw
Hill, London, 1967.
28. A. M. Winslow, Numerical solution of the quasilinear pois
son’s equation in a nonuniform triangular mesh, J. Comput.
Phys. 2:149–172 (1967).
29. P. P. Silvester and M. V. K. Chari, Finite element solution of
saturable magnetic ﬁeld problems, IEEE Trans. Power Appar.
Syst. PAS89:1642–1651 (1970).
30. P. Silvester, Finite element analysis of planar microwave
networks, IEEE Trans. Microwave Theory Tech. MTT
21:104–108 (1973).
31. B. M. A. Rahman, A. Fernandez, and J. Davie, Review of ﬁnite
element methods for microwave and optical waveguides, Proc.
IEEE 79(10):1442–1448 (1991).
32. H. Jin and R. Vahldieck, The frequencydomain transmission
line matrix method—a new concept, IEEE Trans. Microwave
Theory Tech. MTT40:2207–2218 (1992).
33. M. Krumpholz, C. Huber, and P. Russer, A ﬁeld theoretical
comparison of FDTD and TLM, IEEE Trans. Microwave
Theory Tech. MTT43:1935–1959 (1995).
34. K. S. Yee, Numerical solutions of initial boundary value
problems involving Maxwell’s equations in isotropic media,
IEEE Trans. Anten. Propag. AP14:302–307 (1966).
35. A. Taﬂove and M. E. Brodwin, Numerical solution of steady
state electromagnetic scattering problems using the time
dependent Maxwell’s equations, IEEE Trans. Microwave
Theory Tech. MTT23:623–630 (1975).
36. T. Weiland, A discretization method for the solution of
Maxwell’s equations for sixcomponent ﬁelds, Electron. Com
mun. AEU
¨
31:116–124 (1977).
37. G. Mur, Absorbing boundary conditions for the ﬁnitediffer
ence approximation of the timedomain electromagnetic ﬁeld
equations, IEEE Trans. Eleclromagn. Compat. EMC23:
377–382 (1981).
38. T. G. Moore et al., Theory and application of radiation
boundary operators, IEEE Trans. Anten. Propag. AP
36:1797–1812 (1988).
39. J.P. Berenger, A perfectly matched layer for the absorption
of electromagnetic waves, J. Comput. Phys. 114:185–200
(1994).
40. C. Eswarappa and W. J. R. Hoefer, Absorbing boundary
conditions for timedomain TLM and FDTD analysis of
electromagnetic structures, Electromagnetics 16(5):489–519
(1996).
41. K. S. Kunz and R. J. Luebbers, Finite Difference Time Domain
Method for Electromagnetics, CRC Press, Boca Raton, FL,
1993.
42. A. Taﬂove, Computational Electrodynamics, the FiniteDiffer
ence TimeDomain Method, Artech House, Norwood, MA,
1995.
43. P. B. Johns and R. L. Beurle, Numerical solution of 2dimen
sional scattering problems using a transmissionline matrix,
IEE Proc. 118:1203–1208 (1971).
44. G. Kron, Equivalent circuit of the ﬁeld equations of Maxwell
I, Proc. IEEE 32:289–299 (1944).
45. S. Akhtarzad and P. B. Johns, Solution of 6compo
nent electromagnetic ﬁelds in three space dimensions
and time by the T.L.M. method, Electron. Lett. 10:535–537
(1974).
ELECTROMAGNETIC MODELING 1247
46. P. B. Johns, New symmetrical condensed node for three
dimensional solution of electromagnetic wave problems by
TLM, Electron. Lett. 22:162–164 (1986).
47. P. B. Johns, A symmetrical condensed node for the TLM
method, IEEE Trans. Microwave Theory Tech. MTT35:370–
377 (1987).
48. P. Russer and B. Bader, The alternating transmission line
matrix ATLM scheme, in 1995 IEEE MTT Int. Microwave
Symp. Digest, Orlando, FL, 1995, pp. 19–21.
49. W. J. R. Hoefer, The discrete time domain Green’s function or
Johns matrix—a new powerful concept in TLM, Int. J. Num.
Model. 2:215–225 (1989).
50. L. R. A. X. de Menezes and W. J. R. Hoefer, Modeling of
general constitutive relationships in SCN TLM, IEEE Trans.
Microwave Theory Tech. MTT44:854–861 (1996).
51. W. J. R. Hoefer, The transmission line matrix, TLM method, in
T. Itoh, ed., Numerical Techniques for Passive Microwave and
MillimeterWave Structures, Wiley, New York, 1989, Chap. 8.
52. C. Christopoulos, The TransmissionLine Modeling Method
TLM, IEEE Press, Piscataway, NJ, 1995.
53. G. A. Thiele and T. H. Newhouse, A hybrid technique for
combining moments method and geometrical theory of dif
fraction, IEEE Trans. Anten. Propag. 23:62 (1975).
54. D. C. Ross, J. L. Volakis, and H. T. Anastassiu, Hybrid ﬁnite
elementmodal analysis of jet engine inlet scattering, IEEE
Trans. Anten. Propag. 43:277–285 (1995).
55. M. S. Tharf and G. I. Costache, Finite element solutions of
ﬁeld distributions in large cavities, Int. J. Num. Model.
Electron. Networks, Device Fields 7(5):343–356 (1994).
56. R. Beyer and F. Arndt, Efﬁcient modal analysis of waveguide
ﬁlters including the orthogonal mode coupling elements by an
MM/FE method, IEEE Microwave Guided Wave Lett. 5(1):1–3
(1995).
57. M. Mogiardo and R. Sorrentino, Efﬁcient and versatile ana
lysis of microwave structures by combined mode matching
and ﬁnite difference methods, IEEE Microwave Guided Wave
Lett. 3(8):241–243 (1993).
58. K. Wu, M. Yu, and R. Vahldieck, Rigorous analysis of 3D
planar circuit discontinuities using the spacespectral domain
approach, IEEE Trans. Microwave Theory Tech. MTT
40:1475–1483 (1992).
59. S. Xiao and R. Vahldieck, An efﬁcient 2D FDTD algorithm
using real variables, IEEE Microwave Guided Wave Lett.
3:127–129 (1993).
60. F. Alimenti et al., Efﬁcient analysis of waveguide components
by FDTD combined with time domain modal expansion, IEEE
Microwave Guided Wave Lett. 5:351–353 (1995).
61. M. Righi and W. J. R. Hoefer, Efﬁcient 3DSCNTLM diakop
tics for waveguide components, IEEE Trans. Microwave The
ory Tech. MTT42:2381–2385 (1994).
62. R. Wu and T. Itoh, Hybridizating FDTD analysis with un
conditionally stable FEM for objects with curved boundary,
1995 IEEE MTT Int. Microwave Symp. Digest, Orlando, FL,
1995, pp. 833–836.
63. Session on hybrid methods, IEEE APS Int. Symp. Digest,
Newport Beach, CA, June 1995, pp. 1456–1483.
64. E. Yamashita, ed., Analysis Methods for Electromagnetic
Wave Problems, Artech House, Norwood, MA, 1990.
65. R. Sorrentino, ed., Numerical Methods for Passive Microwave
and Millimeter Wave Structures, IEEE Press, Piscataway, NJ,
1989.
66. E.Miller, L. MedgyesiMitschang, E. Newman, Computa
tional Electromagnetics, IEEE Press, Piscataway, NJ, 1991.
ELECTROMAGNETIC SHIELDING
SALVATORE CELOZZI
RODOLFO ARANEO
University of Rome ‘‘La
Sapienza’’
Rome, Italy
1. GENERAL ASPECTS
Electromagnetic shielding is an important constraint in
the design of RF and microwave devices consisting in the
mitigation of the emission levels of electromagnetic (EM)
sources and/or in the protection of people or electrical and
electronic apparatus and systems against possible effects
due to external EM ﬁelds. Very often the need for electro
magnetic shielding depends not only on functioning moti
vations but also on the compliance with standards ﬁxing
the limits of emission or immunity for various classes of
apparatus and systems. Such a limitation or protection is
generally obtained by means of a structure that is often,
but not necessarily, metallic called a shield. Shield perfor
mance depends on its geometric and electrical parameters
and on the characteristics of the unperturbed EMﬁeld (i.e.,
the EM ﬁeld that would exist without any shielding struc
ture, often referred to as incident). Various constructive
peculiarities affect the performance of the overall shielding
system. The most important factors in determining the
performance of a shielding structure are the geometric
conﬁguration and the thickness of the shield and its ma
terials, generally characterized by the values of conductiv
ity s, permeability m, and permittivity e. Also very
important are the socalled discontinuities of the shield
such as junctions, seams, gaps, and apertures, which are
always present in practical conﬁgurations and consider
ably affect the performance in the radio frequency range.
The performance of a shield conﬁguration is often ex
pressed synthetically in terms of shielding effectiveness
(SE), which is deﬁned by the IEEE as the ratio of electric
or magnetic ﬁeld strength (modulus of the vector) at a
point before (E
i
or H
i
) and after (E or H) the placement of
the shield between a given external source and the obser
vation point considered. In terms of Cartesian coordinates
x
0
,y
0
,z
0
, the following expressions apply:
SE
E
¼
E
i
ðx
0
; y
0
; z
0
Þ
Eðx
0
; y
0
; z
0
Þ
ð1aÞ
SE
H
¼
H
i
ðx
0
; y
0
; z
0
Þ
Hðx
0
; y
0
; z
0
Þ
ð1bÞ
These two ﬁgures of merit are usually expressed in deci
bels, as
SE
EðdBÞ
¼20log
10
E
i
ðx
0
; y
0
; z
0
Þ
Eðx
0
; y
0
; z
0
Þ
ð2aÞ
SE
HðdBÞ
¼20 log
10
H
i
ðx
0
; y
0
; z
0
Þ
Hðx
0
; y
0
; z
0
Þ
ð2bÞ
1248 ELECTROMAGNETIC SHIELDING
In case of sinusoidal sources the electric and the magnetic
ﬁelds, expressed as root meansquare (RMS) values in the
expressions above, do not account for polarization and are
local quantities.
Also very important is the evaluation of SE perfor
mance in terms of EM power (incident) P
i
and transmitted
power P:
SE
ðdBÞ
¼10 log
10
P
i
ðx
0
; y
0
; z
0
Þ
Pðx
0
; y
0
; z
0
Þ
ð3Þ
Figure 1 shows the conﬁguration leading to the evaluation
of SE.
2. SHIELDING THEORY AND PREDICTION OF
PERFORMANCE
Shielding theory, including analytical formulations, has
been well developed in order to obtain guidance in prac
tical problems and to ﬁrmly grasp the key factors on which
the theory of electromagnetic shielding is based. The ini
tial studies are due to Maxwell, although King presented
perhaps the ﬁrst work specifically oriented to the radio
frequency range [1]. For difﬁculty in handling real conﬁg
urations by means of exact analytical expressions,
approximate formulations have been developed and im
proved over the years, while the analysis of the SE of
closed structures, often referred to as enclosures, is gen
erally performed, in either the frequency or the time do
main, by using numerical techniques, and can require
significant computational effort, depending on the geomet
ric conﬁguration, the characteristics of the shield material,
and the source type; the latter aspect is the starting point
of every shielding study.
Among the analytical methods used to solve shielding
problems, the most relevant and successful are the direct
solution of the Maxwell equations governing the system
and the socalled transmissionline approach, proposed by
Schelkunoff [2].
2.1. Direct Solution of Maxwell Equations
The direct solution of the equations governing the shield
ing problem is available only in a few simple conﬁgura
tions. However, some relevant cases with reasonable
assumptions may be studied in this way. Moreover, such
exact solutions may serve as reference for the validation of
other methods and thus represent a fundamental tool in
shielding analyses. The system conﬁgurations for which
an analytical solution of the Maxwell equations has been
found include those characterized by planar shields of in
ﬁnite extent in the presence of straight or circular ﬁla
mental wire current sources [3–6]. The ideal conﬁguration
of a planar shield of inﬁnite extension and illuminated by
a plane wave provides an exact solution, which is
described in detail below.
2.2. TransmissionLine Approach
This method was initially introduced with reference to a
plane wave impinging with a normal angle of incidence on
an inﬁnite planar shield and successively extended to ac
commodate other source and shield conﬁgurations. The
method is based on the analogy existing between the
equations governing voltage and current in a transmis
sion line and those describing the electric and the mag
netic ﬁeld propagation inside a planar shield subjected to
a plane wave, as shown in Fig. 2. Assuming a sinusoidal
ﬁeld source, the problem may be solved in the frequency
domain, provided the shield material is linear. For a nor
mal angle of incidence of a plane wave having only the y
component of the electric ﬁeld and the z component of the
magnetic ﬁeld, Maxwell curl equations governing the ﬁeld
propagation through the shield become
dE
y
dx
¼ ÀjomH
z
ð4aÞ
dH
z
dx
¼ À s þjoe ð ÞE
y
ð4bÞ
It is readily recognized that this equation system de
scribes the electric and magnetic ﬁeld propagation as
well as the voltage and current propagation in a trans
mission line, when expressing voltage and current instead
of E
y
and H
z
, respectively, and substituting jom and
s þjoe ð Þ with the per unit length impedance and admit
tance of the line, respectively. The electric and magnetic
Real world
source
Incident
wave
Without
shield
Incident
wave
E
i
, H
i
E
t
, H
t
Observation
point
P(x, y, z)
Observation
point
P(x, y, z)
With
shield
Transmitted
wave
Figure 1. System conﬁgurations for the evaluation of the shield
ing effectiveness (SE).
ELECTROMAGNETIC SHIELDING 1249
ﬁelds inside the shield material, E
y
(x) and H
z
(x), at the
generic distance x from the ﬁrst interface between air and
shield, are expressed as functions of the corresponding
quantities at x ¼0
E
y
ðxÞ ¼E
y
ð0Þ coshðgxÞ þZH
z
ð0ÞsinhðgxÞ ð5aÞ
H
z
ðxÞ ¼
1
Z
E
y
ð0ÞsinhðgxÞ þH
z
ð0Þ coshðgxÞ ð5bÞ
where Z is the intrinsic impedance of the shield given by
Z ¼
jom
s þjoe
_ _
1=2
ð6Þ
and g is the propagation constant of the electromagnetic
ﬁeld inside the shield:
g ¼ jom sþjoe ð Þ ½
1=2
ð7Þ
The system is solved by applying the boundary conditions
at the interfaces. Usually, such boundary conditions are
applied considering the continuity of the tangential com
ponents of the electric and magnetic ﬁelds incident on
the shield surfaces, which are also the only components
propagating through the shield by virtue of the Poynting
theorem. In the conﬁguration considered, they are
expressed as
E
i
y
þE
r
y
¼E
t
y
at x ¼0 ð8aÞ
H
i
z
ÀH
r
z
¼H
t
z
at x ¼0 ð8bÞ
E
t
y
¼E
o
y
at x ¼d ð8cÞ
H
t
z
¼H
o
z
at x ¼d ð8dÞ
where superscripts i, r, t, and o respectively denote the
incident, reﬂected, transmitted, and outgoing ﬁelds and x
¼d is the abscissa of the second interface. The incident
ﬁeld is due to the source alone, the reﬂected ﬁeld is due to
the contribution of the induced currents on the shield sur
face, the transmitted ﬁeld is the ﬁeld propagating through
the shield medium, and the outgoing ﬁeld is the ﬁeld that
passes behind the shield.
2.2.1. Uniform PlaneWave Field Source. A uniform
plane wave is, by definition, a wave in which the electric
and magnetic ﬁelds are perpendicular to each other, their
direction of propagation does not vary in time, and their
amplitude is constant in space and in time. This is, of
course, an idealization because the EM ﬁeld produced by
real sources must decay in space, but if the observation
point is very far from the real source, that is, if its distance
is much greater than the wavelength of the electromag
netic ﬁeld, the uniform planewave assumption is quite
realistic. In a plane wave both incident and reﬂected elec
tric and magnetic ﬁelds remain in a ﬁxed ratio, which is
Shield E
i
E
r
E
i
H
i
H
r
H
i
n
x
y
n
i
n
i
n
r
E
H
Ground
R
s
E
s
V
i
V
r
V
t
V
I
r
I
i
I
t
R
L
z
I
Figure 2. Shielding of a planewave electromagnetic ﬁeld by means of a planar shield and analogy
with the transmissionline conﬁguration.
1250 ELECTROMAGNETIC SHIELDING
deﬁned as the wave impedance Z
w
of the EM ﬁeld, and its
value in vacuum is
Z
w
¼
E
i
y
H
i
z
¼
E
r
y
H
r
z
¼
E
o
y
H
o
z
¼
ﬃﬃﬃﬃﬃ
m
0
e
0
_
¼120p O ﬃ377 O ð9Þ
Equations (8b) and (8d), expressing the continuity of the
tangential components of the magnetic ﬁeld, are multi
plied by Z
w
, and the results are added to Eqs. (8a) and (8c),
respectively, in order to ﬁnd the relationships between the
incident and the transmitted ﬁelds at the two shield in
terfaces:
2E
i
y
¼E
t
y
þZ
w
H
t
z
at x ¼0 ð10aÞ
0¼E
t
y
ÀZ
w
H
t
z
at x ¼d ð10bÞ
In this way the reﬂected ﬁelds, which are generally un
known and not always easy to evaluate, do not appear in
the boundary conditions. It is evident that Eqs. (10) are
formally identical to those valid for a real voltage source
and a simple load impedance, allowing the use of the anal
ogy with a transmission line also with respect to the bound
ary conditions, as shown in Fig. 2. In case of a uniform
plane wave presenting an oblique angle of incidence on a
shield surface, the problem may be solved considering that
an arbitrarily oriented wave may be split up into various
waves, each formed by the orthogonal electric and magnetic
ﬁeld components propagating along different directions [7].
The use of the superposition principle, provided the shield
material is linear, allows one to easily accomplish all the
possible physical situations. It is worthy noting that, with
reference to the coordinate system shown in Fig. 2, the two
waves constituted by (E
y
, H
z
) and (E
z
, ÀH
y
) propagate to
ward the shield and must be accounted for in shielding an
alyses, as they are the other waves responsible for sliding in
directions parallel to the shield. Thus, two wave impedanc
es are considered to extend the previous formulation to the
oblique incidence case:
Z
w1
¼
E
y
H
z
ð11aÞ
Z
w2
¼ À
E
z
H
y
ð11bÞ
These two wave impedances are used to impose the bound
ary conditions in Eqs. (10) and allow the evaluation of the
total electric and magnetic ﬁeld components transmitted
beyond the shield.
2.2.2. NearField Sources. In order to analyze the per
formance of shields against nearﬁeld sources, it is useful
to introduce the elementary EM ﬁeld sources known as
the electric and the magnetic dipoles. Although these
sources are ideal (i.e., do not exist in the real world), the
EM ﬁeld that they produce is easy to determine and very
similar to that of some common real sources. The EM ﬁeld
produced by these sources depends strongly on the dis
tance r from the observation point. In fact, the amplitude
of the electric ﬁeld due to an electric dipole is expressed, in
spherical coordinates as the sum of three terms depending
on 1=r, 1=r
2
, and 1=r
3
, respectively, whereas the magnetic
ﬁeld is a function of 1=r and 1=r
2
only:
E
i
r
ðr; oÞ ¼
ﬃﬃﬃﬃﬃ
m
0
e
0
_
IðoÞL
4p
2 cosðyÞ
1
r
2
þ
1
jbr
3
_ _
e
Àjbr
ð12aÞ
E
i
y
ðr; oÞ ¼
ﬃﬃﬃﬃﬃ
m
0
e
0
_
IðoÞL
4p
sinðyÞ
jb
r
þ
1
r
2
þ
1
jbr
3
_ _
e
Àjbr
ð12bÞ
E
i
j
ðr; oÞ ¼0 ð12cÞ
H
i
r
ðr; oÞ ¼0 ð12dÞ
H
i
y
ðr; oÞ ¼0 ð12eÞ
H
i
j
ðr; oÞ ¼
IðoÞL
4p
sinðyÞ
jb
r
þ
1
r
2
_ _
e
Àjbr
ð12f Þ
where I(o) is the current ﬂowing in the elementary source
of length L. Therefore, at points far from an electric dipole
(farﬁeld region), at a distance r from the EM source much
greater than the wavelength l, the term proportional to
1=r prevails in the expressions for both the electric and
magnetic ﬁelds, so that their ratio, representing the wave
impedance Z
w
at a distance r, assumes the same value it
assumes in case of a uniform plane wave: 377 O. On the
contrary, in the nearﬁeld region, where ðr=lÞ51, the
terms proportional to 1=r
3
and 1=r
2
are the dominant
ones in the electric and magnetic ﬁeld expressions, re
spectively. Consequently, the wave impedance Z
w
is a
function of both the position and the frequency; its value
is generally greater than 377 O and, because of this char
acteristic, the electric dipole is termed a highimpedance
source. Magnetic dipoles present characteristics that are
dual with respect to electric ones. The amplitude of the
radiated magnetic ﬁeld is expressed as the sum of three
terms depending on 1=r, 1=r
2
, and 1=r
3
, respectively,
whereas the electric ﬁeld is a function of 1=r and 1=r
2
only
E
i
r
ðr; oÞ ¼0 ð13aÞ
E
i
y
ðr; oÞ ¼0 ð13bÞ
E
i
j
ðr; oÞ ¼ À
jom
0
IðoÞA
4p
sinðyÞ
jb
r
þ
1
r
2
_ _
e
Àjbr
ð13cÞ
H
i
r
ðr; oÞ ¼
jbIðoÞA
4p
2 cosðyÞ
1
r
2
þ
1
jbr
3
_ _
e
Àjbr
ð13dÞ
H
i
y
ðr; oÞ ¼
jbIðoÞA
4p
sinðyÞ
jb
r
þ
1
r
2
þ
1
jbr
3
_ _
e
Àjbr
ð13eÞ
H
i
j
ðr; oÞ ¼0 ð13f Þ
where I(o) is the current ﬂowing in the loop with surface
A. In the farﬁeld region, the EM ﬁeld is characterized by
the freespace wave impedance. Conversely, in the near
ﬁeld region, the terms proportional to 1=r
3
and 1=r
2
are
ELECTROMAGNETIC SHIELDING 1251
the dominant ones in the magnetic and electric ﬁelds, re
spectively, and the wave impedance Z
w
is lower than 377
O. For this reason, the magnetic dipole represents a low
impedance source.
In Fig. 3, the trends of the wave impedances of an elec
tric dipole and a magnetic dipole are reported, as functions
of the distance, normalized by l, of the observation point
from the source.
These considerations are of great importance in ex
tending the transmissionline approach to solve near ﬁeld
source problems. It should be noted that generally the
electric and the magnetic ﬁelds are not perpendicular to
each other and the various components directed toward
the shield, according to the Poynting theorem, may be ac
counted for separately, as described for the oblique inci
dence of uniform plane waves. According to the studies
carried out for the oblique incidence of a plane wave on a
planar shield, the wave impedances associated with the
various directions may be deﬁned and utilized to deter
mine the ﬁeld incident onto a preﬁxed shield surface. In
fact, the boundary conditions may be imposed following
the procedure described for the planewave source, pro
vided the electric and the magnetic ﬁeld are considered
space and frequencydependent.
2.2.3. Shielding Effectiveness Evaluation. In case of uni
form plane waves, the electromagnetic ﬁeld amplitudes
remain constant in space and the shielding effectiveness
may be evaluated as the ratio between the incident ﬁeld
and the outgoing ﬁeld. Moreover, SE
E
and SE
H
are nu
merically coincident. SE may be expressed in the following
very compact form
SE¼AþRþB ð14Þ
where the absorption loss A, reﬂection loss R, and multiple
reﬂections coefﬁcient B for a planar shield of thickness d
are respectively given (in decibels) by
A
dB
¼20 log
10
e
gd
¸
¸
¸
¸
ð15aÞ
R
dB
¼20 log
10
Z
w
þZ ð Þ
2
ð4Z
w
ZÞ
À1
¸
¸
¸
¸
ð15bÞ
B
dB
¼20 log
10
1 À
Z
w
ÀZ ð Þ
2
Z
w
þZ ð Þ
2
e
À2gd
¸
¸
¸
¸
¸
¸
¸
¸
¸
¸
ð15cÞ
The absorption loss coefﬁcient A
dB
is a function of the
shield characteristics only; the reﬂection loss coefﬁcient
R
dB
depends on the mismatch between the wave imped
ance and the intrinsic impedance of the shield. The mul
tiplereﬂection coefﬁcient B
dB
depends on both the
physical characteristics of the shield material and the in
cident ﬁeld. Coefﬁcient A
dB
can be expressed in the fol
lowing simple form:
A
dB
¼131:44
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
f m
r
s
r
_
d ð16Þ
where m
r
and s
r
denote respectively the relative magnetic
permeability and the relative conductivity (with respect to
the conductivity of copper s
Cu
¼5.810
7
S/m) of the material.
Different approximate expressions hold for coefﬁcient R
for planewave sources, highimpedance sources, and low
impedance sources, respectively:
R
dB
¼168:1 À20 log
10
ﬃﬃﬃﬃﬃﬃﬃ
f m
r
s
r
¸
ð17aÞ
R
dB
¼321:7 À10 log
10
s
r
f
3
m
r
r
2
_ _
ð17bÞ
R
dB
¼20 log
10
5:35
ﬃﬃﬃﬃﬃﬃﬃ
s
r
f
m
r
¸
r þ0:0117
ﬃﬃﬃﬃﬃﬃﬃ
m
r
s
r
f
_
1
r
þ0:5
_ _
ð17cÞ
Figure 4 shows the absorption loss coefﬁcient A
dB
as func
tion of the frequency for different thicknesses of a copper
shield. Figure 5 shows the frequency dependence of coef
ﬁcient R
dB
of copper against typical low or highimped
ance ﬁelds, respectively, both located at a distance of
30.48 cm from the shield surface.
It should be noted that in (and sometimes below) the
microwave band of the frequency spectrum, the SE is com
pletely dominated by the discontinuities, since the con
ductive materials behave like perfect electric conductors,
present at a skin depth that is much lower than the shield
thickness.
The coefﬁcient B
dB
in the radiofrequency range is often
negligible. In case of lowimpedance sources, the reﬂection
loss coefﬁcient exhibits values increasing with the fre
quency, for copper and aluminum, while for ferromagnetic
materials it initially decreases and then increases. For
highimpedance sources, the reﬂection loss coefﬁcient
always decreases for all the materials. The values of R
dB
are much lower in the lowimpedance source case, con
ﬁrming the difﬁculty in shielding magnetic ﬁelds in the
lowfrequency range. It should also be noted that the
previous expressions are rigorously valid for a plane
2000
1500
1000
500
377
0
0 0.2 0.4 0.6 0.8 1
r/z
Magnetic dipole wave impedance
Electric dipole wave impedance
W
a
v
e
i
m
p
e
d
a
n
c
e
[
Ω
]
Figure 3. Trend of the wave impedances associated with electric
and magnetic dipoles, as a function of the normalized distance
from the of the sinusoidal elemental source.
1252 ELECTROMAGNETIC SHIELDING
wave source impinging normally onto a shield that is con
ductive, inﬁnite, and planar. When extending the trans
missionlineapproach to waves with an oblique angle of
incidence, another hypothesis is necessary—the electro
magnetic ﬁeld propagation inside the shield must occur
normally to the shield surface; this assumption is gener
ally satisﬁed in consideration of the high values of con
ductivity, as stated by Snell’s law of refraction [8]
g
0
sinðy
i
Þ ¼g
s
sinðy
s
Þ ð18Þ
where g
0
and g
s
are respectively the propagation constant
in air and in the shield medium and y
i
and y
s
are the di
rections of propagation of the electromagnetic ﬁeld in the
two media, respectively.
2.3. Ideal Enclosures without Apertures
Real enclosures may be analyzed only by means of nu
merical methods because of their complex shape; however,
some formulations exist allowing an estimation of shield
ing performance under simplifying hypotheses and ap
proximations concerning the absence of constructive
defects and discontinuities in the structure and the shape
of the shield conﬁguration. The effect of the discontinuities
such as junctions and apertures may be accounted for at a
successive stage, as reported in the following. An analyt
ical expression valid, for example, for the SE against a
uniform magnetic ﬁeld incident transversely onto a cylin
drically shaped enclosure of inﬁnite length is
SE¼20 log
10
1
cosh g
.
d ð Þ þ
r
0
.
g
2m
r
þ
m
r
2r
0
.
g
_ _
cosh g
.
d ð Þ
ð19Þ
where r
0
is the radius of the cylindrical shield.
The SE of a sphericallyshaped enclosure against an
incident uniform magnetic ﬁeld is given by the following
expression, where r
0
represents the sphere radius:
SE¼20 log
10
1
cosh g
.
d ð Þ þ
r
0
.
g
3m
r
cosh g
.
d ð Þ
ð20Þ
More complex shapes have been also analyzed, and sev
eral useful expressions are reported in [9,10].
2.4. Shielded Cables
Cables are critical components because they may connect
apparatus sensitive to external radiofrequency EM ﬁelds,
or they may radiate unintentionally, becoming sources of
interference. With this twofold motivation, cables are of
ten shielded in various ways and either compact or perfo
rated braided shield may be effective in reducing EM
susceptibility or interference [11,12]. Generally, a synthet
ic parameter termed transfer impedance is introduced to
account for the penetration through the cable shield. The
transfer impedance may be regarded as the per unit
length voltage drop occurring on the external surface of
the cable shield when the inner conductor is driven by a
unit current or, by reciprocity, the effect on the inside of
the shield due to a current on the external shield surface.
Also of particular relevance are the status of ground con
nections at the ends of the cable, and the type of connec
tors, apart, of course, from the shield characteristics.
2.5. Numerical Methods for Shielding Analysis
The intrinsic limitations of analytical formulations have
lead to a wide literature concerning studies of the appli
cation of various numerical methods to shielding problems,
which would be an exhaustive, practically impossible re
port. However, examples of the application of the method of
moments to shielding conﬁgurations may be found in Refs.
13 and 14, while the ﬁnitedifference timedomain method
has been applied in Refs. 15 and 16, the boundary element
method in Ref. 17, the transmissionline method in Ref. 18,
and the ﬁniteelement method in Ref. 19.
A
[
d
B
]
1000
100
10
1
d = 25 µm
d = 75 µm
d = 10 µm
10
−2
10
−1
1 10 10
2
10
3
10
4
10
5
Frequency [MHz]
Figure 4. Frequency dependence of the absorption loss coefﬁ
cient A
dB
for different shield thicknesses. (This ﬁgure is available
in full color at http://www.mrw.interscience.wiley.com/erfme.)
R
[
d
B
]
200
100
50
20
10
−2
10
−1
1 10 10
2
10
3
10
4
10
5
Frequecny [MHz]
Highimpedance
source
Planewave
region
Lowimpedance
source
Figure 5. Frequency dependence of the reﬂection loss coefﬁcient
R
dB
of copper against lowimpedance and highimpedance
sources. (This ﬁgure is available in full color at http://www.mrw.
interscience.wiley.com/erfme.)
ELECTROMAGNETIC SHIELDING 1253
3. OTHER FACTORS AFFECTING SHIELDING
PERFORMANCE
The analytical methods described previously are based on
the assumption that no discontinuities (e.g., apertures,
holes, junctions, seams) are present in the shield; never
theless, such a hypothesis is quite unrealistic because
intentional apertures for various purposes and uninten
tional defects always exist and may considerably degrade
the performance of real shields. Moreover, the presence
of standing waves in closed enclosures may deteriorate
the shielding performance even in absence of disconti
nuities.
No general theory is available to account for the dis
continuities; however, some studies have been conducted
to quantify the EM ﬁeld transmitted through intentional
apertures of some regular shapes. In fact, when the ap
erture is electrically small and the shield is perfectly
conducting, its contribution to the EM ﬁeld beyond the
shield may be represented as that due to a combination of
appropriate elementar electric and magnetic dipoles, lo
cated in the centre of the aperture with the shield re
moved [20]. Periodically perforated shields have been
also studied to some extent [21]. Measured data and ap
proximate formulations have been also presented to
quantify the effect of various types of discontinuity in
EM penetrable shields, and, in particular, in case of sin
gle or multiple apertures two approaches can be used to
evaluate the overall effect of the aperture(s) in terms of
EM leakage whose value must be subtracted from the SE
value of the ideal shield [10], or to obtain a modiﬁed ex
pression for the SE that individually accounts for various
effects such as the number of openings per unit square,
coupling between apertures, and reduced absorption loss
terms [6].
3.1. Shielding Applications
Several practical rules and solutions are recommended for
the design and evaluation of the SE performance of real
shield conﬁgurations and may be found elsewhere [10]. It
is worth mentioning that care must be taken to bond
seams and joints according to their permanent or opera
tional nature; permanent joints should be riveted or
screwed, and SE performance is inﬂuenced by the spac
ing between the transverse components. Operational
joints should be ﬁnished with the socalled gaskets, ﬂex
ible and elastic conductive components capable of estab
lishing a good electrical continuity between ﬁxed and
mobile parts of the shield enclosure. Various commercial
gaskets are available, and their performance is generally
good enough to limit efﬁciently the performance deterio
ration. Another situation requiring a special treatment is
represented by windows for visualization; because of their
aperture dimensions, their presence may considerably re
duce the efﬁciency of an otherwise good structure. It is
highly advisable to apply on the required aperture a con
ductive optical substrate or a shield realized by means of a
thin wire mesh.
4. SHIELDING PERFORMANCE MEASUREMENT AND
STANDARDS
The most common documents describing the test
procedures recommended for assessment of the shielding
characteristics of enclosures are:
*
Military Standard 285: Method of Attenuation Mea
surements for Enclosures, Electromagnetic Shielding,
for Electronic Test Purposes.
*
National Security Agency (NSA) Speciﬁcation 656:
General Speciﬁcations for RF Shielded Enclosures for
Communications Equipment.
*
National Security Agency (NSA) Speciﬁcation 732:
General Speciﬁcations for Foil RF Shielded Enclo
sures for Communications Equipment.
*
IEEE 299: Standard Method of Measuring the Effec
tiveness of Electromagnetic Shielded Enclosures.
*
Several ASTM standards, including D493599, pro
vide the procedures for measuring the SE of a planar
material due to a planewave, farﬁeld EM wave, in
the frequency range from 30 MHz to 1.5 GHz.
All these documents describe antenna geometries and
conﬁgurations and also delineate some measurement
practices, classifying the source as a type of magnetic,
electric, or plane wave. It is important to mention that the
data obtained from any of the setups described above can
not be applied to source conﬁgurations different from
those used in the experiments. Typical experimental set
ups are shown in Figs. 6–9. Thus, they have only reference
value and other methods and experimental setups have
been also proposed and applied [22,23].
5. FINAL REMARKS
Some controversial aspects concerning electromagnetic
shielding have to be highlighted; the most important are
represented by the definition itself of the ﬁgure of merit
adopted for all the considerations and for design purposes.
Shield
Receiving
loop
2
R
0
2
R
0
2
R
0
2
R
0
Transmitting
loop
Figure 6. Experimental conﬁguration for the measurement of
magnetic ﬁeld shielding effectiveness, according to IEEE 299 and
MILSTD 285.
1254 ELECTROMAGNETIC SHIELDING
In fact, the SE, as deﬁned, is a local quantity, in the sense
that it represents a ﬁeld (or power) ratio at a specific point,
giving no information on the situation over a ﬁnite surface
or volume. For this reason, it is difﬁcult to extrapolate a
general evaluation of the real performance of the shield
from such limited information. Moreover, the SE measure
ment may be cumbersome or even impossible in some real
conﬁgurations because of either the shield dimensions
(e.g., too small) or the coupling between the enclosure
and the transmitting and receiving antennas. Other
controversial aspects concern the validity of the transmis
sionline analogy, which must be assessed case by case,
otherwise possibly leading to incorrect predictions.
6. CURRENT RESEARCH TOPICS
Various aspects of electromagnetic shielding are currently
under investigation. One of the most challenging research
topics is the design of real shielding structures. The need
for significant but measurable quantities is of paramount
importance in the shielding practice, making possible the
experimental veriﬁcation of predictions, the design of
shielding structures, as well as the promulgation of stan
dards; in this regard, the search for new ﬁgures of merit is
still underway, especially for small enclosures. Of course,
new, reliable, and simple ways to perform measurements
for the characterization of both materials and shielding
structures are always under investigation.
Another important ﬁeld of research is the development
of new synthetic materials providing good performance in
specific applications.
In addition, active shielding, realized by means of ad
ditional sources capable of generating an electromagnetic
ﬁeld opposite that undesired and to be mitigated, is a re
search topic of great interest, inasmuch as in the design of
optimal conﬁgurations.
Finally, since all the abovementioned research topics
necessitate the formulation of work hypotheses and their
successive experimental veriﬁcation, numerical modelling
and analysis methods must continuously deal with more
complex materials and conﬁgurations, and thus their im
provement in terms of accuracy and efﬁciency represents a
fundamental research topic in electromagnetic shielding.
BIBLIOGRAPHY
1. L. V. King, Electromagnetic shielding at radio frequencies,
Phil. Mag. 15(Ser. 7):201–223 (Feb. 1933).
2. S. A. Schelkunoff, Electromagnetic Waves, Van Nostrand, New
York, 1943.
3. J. A. Tegopoulos and E. E. Kriezis, Eddy Currents in Linear
Conducting Media, Elsevier, Amsterdam, 1985.
4. J. R. Moser, Lowfrequency lowimpedance electromagnetic
shielding, IEEE Trans. Electromagn. Compat. 30(3):202–210
(1988).
5. R. Araneo and S. Celozzi, On the exact solution of the low
frequency coplanar loops shielding conﬁguration, IEE Proc.
Sci. Meas. Technol. 149:37–44 (2002).
Shield
Receiving
loop
2R
0
2R
0
2
R
0
2
R
0
Transmitting
loop
Figure 7. Experimental conﬁguration for the evaluation of mag
netic ﬁeld shielding effectiveness, according to NSA 656.
Shield
Receiving
log antenna
d
d
Transmitting
log antenna
Figure 9. Typical NSA 656 setup for SE measurement in plane
wave source conﬁguration.
Shield
Receiving
rod antenna
Transmitting
rod antenna
d
d
Figure 8. Typical NSA 656 setup for SE measurement in high
impedance source conﬁguration.
ELECTROMAGNETIC SHIELDING 1255
6. R. B. Schulz, V. C. Plantz, and D. R. Brush, Shielding theory
and practice, IEEE Trans. Electromagn. Compat. 30(3):
187–201 (1988).
7. R. A. Adler, L. J. Chu, and R. M. Fano, Electromagnetic En
ergy Transmission and Radiation, Wiley, New York, 1960.
8. C. A. Balanis, Advanced Engineering Electromagnetics,
J Wiley, New York, 1989.
9. T. Rikitake, Magnetic and Electromagnetic Shielding, Terra
Scientific Publishing, Tokyo, 1987.
10. D. R. J. White and M. Mardiguian, Electromagnetic Shield
ing, Vol. 3, EMFEMI Control, Inc., Gainesville, FL, 1988.
11. A. Tsaliovich, Cable Shielding for Electromagnetic Compati
bility, J Wiley, New York, 1995.
12. F. M. Tesche, M. V. Ianoz, and T. Karlsson, EMC Analysis
Methods and Computational Models, J Wiley, New York,
1997.
13. H. Singer, H. D. Bruns, and G. Burger, State of the art in the
method of moments, Proc. IEEE 1996 Int. Symp. Electromag
netic Compatibility, Santa Clara, CA, 1996, pp. 122–127.
14. G. Burger, H. D. Bruns, and H. Singer, Simulation of thin
layers in the method of moments, Proc. 11th Int. Zurich
Symp., Zurich, Switzerland, 1995, pp. 339–344.
15. R. J. Luebbers, K. Kumagai, S. Adachi, and T. Uno, FDTD
calculation of transient pulse propagation through a nonlin
ear magnetic sheet, IEEE Trans. Electromagn. Compat.
33(1):90–94 (1993).
16. S. Celozzi and M. D’Amore, Magnetic ﬁeld attenuation of
nonlinear shields, IEEE Trans. Electromagn. Compat.
38(3):318–326 (1996).
17. J. Shen and A. Kost, Hybrid FEBE method for EMC prob
lems in cable systems, IEEE Trans. Magn. 32(3):1493–1496
(1996).
18. A. Mallik, D. P. Johns, and A. J. Wlodarczyk, TLM modelling
of wires and slots, Proc. 10th Int. Zurich Symp., Zurich, Swit
zerland, 1993, pp. 515–520.
19. S. Celozzi, FETD analysis of ferromagnetic shields against
near ﬁeld sources, Proc. 12th EMC Zurich Symp. Electromag
netic Compatibility, Feb. 18–20, 1997, Zurich, Switzerland,
pp. 281–286.
20. F. De Meulenaere and J. van Bladel, Polarizability of some
small apertures, IEEE Trans. Anten. Propag. 25(2):198–205
(1977).
21. K. F. Casey, Electromagnetic shielding behavior of wiremesh
screens, IEEE Trans. Electromagn. Compat. 30(3):298–306
(1988).
22. P. F. Wilson and M. T. Ma, Shielding effectiveness measure
ments with a dual TEM cell, IEEE Trans. Electromagn.
Compat. 27(3):137–142 (1985).
23. J. A. Catrysse and C. P. J. H. Borgmans, Measuring methods
and measuring setups for the characterisation of shielding
materials under different conditions, Proc. Int. Symp. Elec
tromagnetic Compatibility EMC ‘96 ROMA, Rome, Italy,
1996, pp. 562–568.
FURTHER READING
First Special Issue on Eelectromagnetic Shielding, IEEE Trans.
Electromagn. Compat. 10(1) (1968).
Second Special Issue on Eelectromagnetic Shielding, IEEE Trans.
Electromagn. Compat. 30(3) (1988).
L. H. Hemming, Architectural Shielding, IEEE Press, New York,
1992.
ELECTROMAGNETIC SUBSURFACE
REMOTE SENSING
S. Y. CHEN
W. C. CHEW
University of Illinois at Urbana–
Champaign
Urbana, Illinois
Subsurface electromagnetic (EM) methods are applied to
obtain underground information that is not available from
surface observations. Since electrical parameters such as
dielectric permittivity and conductivity of subsurface ma
terials may vary dramatically, the response of electromag
netic waves can be used to map the underground
structure. This technique is referred to as geological sur
veying. Another major application of subsurface EM meth
ods is to detect and locate underground anomalies such as
mineral deposits.
Subsurface EM methods include a variety of techniques
depending on the application, surveying method, system,
and interpretation procedure, and thus a ‘‘best’’ method
simply does not exist. Even though each system has its
own characteristics, they still share some common fea
tures. In general, each system has a transmitter, which
can be either natural or artiﬁcial, to send out the electro
magnetic energy that serves as an input signal. A receiver
is needed to collect the response signal. The underground
can be viewed as a system, which is characterized by the
material parameters and underground geometry. The task
of subsurface EM methods is to derive the underground
information from the response signal.
The EM transmitter radiates the primary ﬁeld into the
subsurface, which consists of conductive earth material.
This primary ﬁeld will induce a current, which in turn
radiates a secondary ﬁeld. Either the secondary ﬁeld or
the total ﬁeld will be detected by the receiver. After the
data interpretation, one can obtain the underground
information.
One of the most challenging parts of subsurface EM
methods is interpretation of the data. Since the incident
ﬁeld interacts with the subsurface in a very complex man
ner, it is never easy to subtract the information from the
receiver signal. Many definitions, such as apparent con
ductivity, are introduced to facilitate this procedure.
Data interpretation is also a critical factor in evaluat
ing the effectiveness of the system. How good the system is
always depends on how well the data can be explained. In
the early development of subsurface EM systems, data
interpretation largely depended on the personal experi
ence of the operator, due to the complexity of the problem.
Only with the aid of powerful computers and improve
ments in computational EM techniques is it possible to
analyze such a complicated problem in a reasonable time.
Computerbased interpretation and inversion methods
are attracting more and more attention. Nevertheless,
data interpretation is still ‘‘an artful balance of physical
understanding, awareness of the geological constraints,
and pure experience’’ [1].
In the following sections, we will use several typical
applications to outline the basic principles of subsurface
1256 ELECTROMAGNETIC SUBSURFACE REMOTE SENSING
EM methods. Physical insight is emphasized rather than
rigorous mathematical analysis. Details of each method
can be found in the references.
1. BOREHOLE EM METHODS
Borehole EM methods are an important part of well
logging methods. Since water is conductive and oil is an
insulator, resistivity measurements are good indicators of
oil presence. Water has an unusually high dielectic con
stant, and permittivity measurement is a good detector of
moisture content.
Early borehole EM methods consist of mainly electrical
measurements using very simple lowfrequency electrodes
like the short and the long normal. Then more sophisti
cated electrode tools were developed. Some of these tools
are mounted on a mandrel, which performs measurements
centered in a borehole. These tools are called mandrel
tools. Alternatively, the sensors can be mounted on a pad,
and the corresponding tool is called a pad tool.
One of the most successful borehole EM methods is in
duction logging. Since Doll published his ﬁrst paper in
1949 (2), this technique has been used widely with conﬁ
dence in the petroleum industry. Extensive research work
has been done in this area. The systems in use now are so
sophisticated that many modern electrical techniques are
involved. Nevertheless, the principles still remain the
same and can be understood by studying a simple case.
The induction logging technique, as proposed by Doll,
makes use of several coils wound on an isolating mandrel,
called a sonde. Some of the coils, referred to as transmit
ters, are powered with alternating current (AC). The
transmitters radiate the ﬁeld into the conductive forma
tion and induce a secondary current, which is nearly pro
portional to the formation conductivity. The secondary
current radiates a secondary ﬁeld, which can be detected
by the receiver coils. The receiver signal (voltage) is nor
malized with respect to the transmitter current and rep
resented as an apparent conductivity, which serves as an
indication of underground conductivity.
To obtain information from the apparent conductivity,
we need to understand how apparent conductivity and
true conductivity are related. According to Doll’s theory,
the relation in cylindrical coordinates is given by
s
a
¼
_
þ1
À1
dz
0
_
þ1
0
dr
0
g
D
ðr
0
; z
0
Þsðr
0
; z
0
Þ ð1Þ
where s
a
is the formation conductivity. The kernel g
D
(r, z)
is the socalled Doll geometric factor, which weights the
contribution of the conductivity from various regions in
the vicinity of sonde.
We notice that g
D
(r, z) is not a function of the true con
ductivity and hence is only determined by the tool conﬁg
uration. The interpretation of the data would be simple if
Doll’s theory were exact. Unfortunately, this is rarely the
case. Further studies showthat Eq. (1) is true only in some
extreme cases. The significance of Doll’s theory, however,
is that it relates the apparent conductivity and formation
conductivity, even though the theory is not exact. In the
early development of induction logging techniques, tool
design and data interpretation were based on Doll’s the
ory, and in most cases it gives reasonable answers.
To establish a ﬁrm understanding of induction logging
theory, we need to perform a rigorous analysis by using
Maxwell’s equations as follows:
rÂH¼ ÀioeEþJ
s
þsE ð2Þ
rÂE¼iomH ð3Þ
r
.
H¼0 ð4Þ
r
.
D¼r ð5Þ
where r
.
J
s
¼ior.
In the preceding equations, the time dependence e
Àiot
is assumed, and J
s
corresponds to the impressed current
source. Parameters m,e,s are the magnetic permeability,
dielectric permittivity, and electric conductivity, respec
tively. To simplify the analysis, we assume that both the
impressed source and geometry of the problem are axi
symmetric; consequently, all the ﬁeld components are in
dependent of the azimuthal angle. Furthermore, it can be
shown that there is no stored charge under the preceding
assumption. The working frequency of induction logging is
about 20 kHz, so the displacement current ÀioeE is very
small compared to the conduction current sE and hence is
neglected in the following discussion. After these simpli
ﬁcations, we have
rÂH ÀsE¼J
s
ð6Þ
rÂE ÀiomH¼0 ð7Þ
r
.
H¼0 ð8Þ
r
.
E¼0 ð9Þ
where we assume r
.
J
s
¼ior¼0.
For convenience, the auxiliary vector potential is intro
duced. Since r
.
H¼0 and r
.
(r Â ) ¼0, it is possible to
deﬁne H¼r Â A. To specify the ﬁeld uniquely, we choose
E¼iomA, which is true only when there is no charge ac
cumulation. Substituting these expressions into Eq. (6),
we have
rÂrÂA ÀiomsA¼J
s
ð10Þ
By using the vector identity, we have
r
2
Aþk
2
A¼ ÀJ
s
ð11Þ
where
k
2
¼ioms ð12Þ
To demonstrate how the apparent conductivity and
formation conductivity are related, we ﬁrst write down
the solution of Eq. (11) in a homogeneous medium as
ELECTROMAGNETIC SUBSURFACE REMOTE SENSING 1257
follows [3,4]:
Aðr; z; fÞ ¼
1
4p
_
V
0
J
s
ðr
0
; z
0
; f
0
Þ
" rr
1
e
ik" rr
1
dV
0
ð13Þ
where
" rr
1
¼fðz Àz
0
Þ
2
þr
2
þr
02
À2rr
0
cosðf Àf
0
Þg
1=2
ð14Þ
The volume integration is evaluated over regions contain
ing the impressed current sources and the coordinate sys
tem used in Eq. (13), as shown in Fig. 1. Usually, a small
current loop is used as an excitation, which implies that
only A
f
exists. Hence, Eq. (13) can be furthermore simpli
ﬁed as
A
f
ðr; zÞ ¼
1
4p
_
V
0
J
f
ðr
0
; z
0
Þ cosðf Àf
0
Þ
e
ik" rr
1
" rr
1
dV
0
ð15Þ
When the radius of the current loop becomes inﬁnitely
small, it can be viewed as a magnetic dipole, and thus the
preceding integration can be approximated as
A
f
¼
m
4p
r
r
3
1
ð1 Àikr
1
Þe
ikr
1
ð16Þ
where m¼N
T
I(pa
2
) is the magnetic dipole moment and N
T
is the number of turns wound on the mandrel. At the re
ceiver point, the voltage induced on the receiver with N
R
turns can be represented as
V ¼2paN
r
E
f
¼
2N
T
N
R
ðpa
2
Þ
2
I
4p
iomð1 ÀikLÞ
e
ikL
L
3
ð17Þ
where
E
f
¼iomA
f
ða; LÞ ð18Þ
and L is the distance between the transmitter and the re
ceiver. Since the voltage is a complex quantity, it can be
separated into real and imaginary parts and expanded in
powers of kL as follows [3]
V
R
¼ ÀKs 1 À
2
3
L
d
þ Á Á Á
_ _
ð19Þ
V
X
¼Ks
d
2
L
2
1 À
2
3
L
2
d
3
þ Á Á Á
_ _
ð20Þ
where
K ¼
ðomÞ
2
ðpa
2
Þ
2
4p
N
T
N
R
I
L
ð21Þ
and
d ¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃ
2
oms
¸
ð22Þ
The quantity K is known as the ‘‘tool constant’’ and is to
tally determined by the conﬁguration of the tool, and s is
the socalled skin depth, which describes the attenuation
of a conductor in terms of the ﬁeld penetration distance.
The quantity V
R
is called the R signal. The apparent con
ductivity is deﬁned as [3]
s
a
¼ À
V
R
K
ﬃs 1 À
2
3
L
d
_ _
ð23Þ
In the preceding analysis, there are some important
facts that need to be mentioned. In Eq. (19), we see that
the apparent conductivity is a nonlinear function of the
true conductivity, even in a homogeneous medium. The
lower the working frequency or lower the true conductiv
ity, the more linear it will be. The difference between true
conductivity and apparent conductivity is deﬁned as the
skin effect signal:
s
s
¼s Às
a
ð24Þ
The leading term of the imaginary part V
X
is not a func
tion of true conductivity. In fact, it corresponds to the di
rect coupling ﬁeld, which does not contain any formation
information. What remains in V
X
is the socalled X signal.
Since the direct term is much larger than the residual part
including V
R
, it is difﬁcult to separate the X signal. The
r
s
r
s
r
1
r
1
r
2
r
2
(j′ , Z ′, [′)
(j′ , Z′ , [′ )
Formation
element
Transmitter
Receiver
a
a
X
Z
Y
Figure 1. Induction logging tool transmitter and receiver coil
pair used to explain the geometric factor theory. (Redrawn from
Ref. 4.)
1258 ELECTROMAGNETIC SUBSURFACE REMOTE SENSING
importance of the X signal is seen by comparing Eqs. (19)
and (20), from which we ﬁnd that the X signal is the ﬁrst
order approximation of the nonlinear term in V
R
, the R
signal. This fact can be used to compensate for the skin
effect.
So far we have introduced the concept of apparent con
ductivity by studying the homogeneous case. In practice,
the formation conductivity distribution is far more com
plicated. The apparent conductivity and formation con
ductivity are related through a nonlinear convolution. As
a proof we derive the solution in an integral form, instead
of directly solving the differential equations. To this end,
we ﬁrst rewrite Eq. (11) as
r
2
A¼ ÀJ
s
ÀJ
i
ð25Þ
where J
i
¼ Àk
2
A is the induced current. The solution of
Eq. (25) can be written in the integral form as
A¼
1
4p
_
V
0
J
s
" rr
s
dV
0
þ
1
4p
_
V
J
i
" rr
2
dV ð26Þ
The ﬁrst integral is evaluated over the regions contain
ing the impressed sources, and the second one is per
formed over the entire formation. Under the same
assumption as we have made in the preceding analysis,
the receiver voltage can be written as [4]
V ¼
i2paN
R
om
4p
_
V
0
J
f
" rr
s
dV
0
À
2paN
R
o
2
m
2
4p
_
V
sðr
0
; z
0
ÞA
f
ðr
0
; z
0
Þ
" rr
2
dV
ð27Þ
The vector potential can also be separated into real and
imaginary parts:
A
f
¼A
fR
þiA
fI
ð28Þ
Substituting Eq. (28) into Eq. (27) and separating out the
real part of the receiver voltage, we have
V
R
¼
ÀðomÞ
2
ð2paN
R
Þ
4p
_
1
À1
dz
0
_
1
0
dr
0
sðr
0
; z
0
Þ A
fR
_
2p
0
cosðf Àf
0
Þ
" rr
2
df
0
ð29Þ
Applying the same procedure, we obtain the apparent
conductivity as
s
a
¼ À
V
R
K
¼
_
1
0
dr
0
_
1
À1
dz
0
sðr
0
; z
0
Þg
P
ðr
0
; z
0
Þ
ð30Þ
where
g
P
¼
2pLr
0
ðpaÞ
3
N
T
I
A
fR
_
2p
0
cosðf Àf
0
Þ
" rr
2
df
0
ð31Þ
The function g
P
is the exact definition of the geometric
factor. In comparison with Doll’s geometric factor, g
P
de
pends not only on the tool conﬁguration but also on the
formation conductivity, since the vector potential depends
on the formation conductivity. The integral form solution
does not provide any computational advantage, since the
differential equation for the vector potential A
f,R
must
still be solved. But it is now clear from Eq. (30) that the
apparent conductivity is the result of a nonlinear convo
lution. Equation (30) also represents the starting point of
inverse ﬁltering techniques, which make use of both the R
and X signals to reconstruct the formation conductivity.
Finding the vector potential A is still a challenge. An
alytic solutions are available only for a few simple geom
etries. In most cases, we have to use numerical techniques
such as the ﬁniteelement method (FEM), ﬁnitedifference
method (FDM), numerical mode matching (NMM), or the
volume integral equation method (VIEM). Interested
readers may ﬁnd Refs. 5 through 8 useful.
Previously, we mentioned that Doll’s geometric factor
theory is valid only under some extreme conditions. In
fact, it can be derived from the exact geometric factor as a
special case [4]. In a homogeneous medium, the vector
potential A
f,R
can be calculated as
A
fR
ﬃ
ðpa
2
ÞN
T
I
4p
r
0
r
3
1
Refð1 Àikr
1
Þe
ikr
1
g ð32Þ
The integration with respect to f
0
in Eq. (31) can also be
performed for " rr
2
ba. The ﬁnal result is
s
a
¼
_
1
À1
_
1
0
sg
D
ðr
0
; z
0
ÞRefð1 Àikr
1
Þe
ikr
1
gdr
0
dz
0
ð33Þ
where
g
D
ðr
0
; z
0
Þ ¼
L
2
r
03
r
3
1
r
3
2
ð34Þ
It is now clear that Doll’s geometric factor and the exact
geometric factor are the same when the medium is homo
geneous and the wavenumber approaches zero.
So far we have discussed the basic theory of induction
logging. We now use a simple example to show some prac
tical concerns and briefly discuss the solutions. In Fig. 2,
we show an apparent resistivity (the inverse of apparent
conductivity) response of a commercial logging tool 6FF40
(trademark of the Schlumberger Company) in the Okla
homa benchmark. The black line is the formation resis
tivity, and the red line is the unprocessed data of 6FF40.
We notice that the apparent resistivity data roughly indi
cate the variation of the true resistivity, but around 4850ft
the apparent resistivity R
a
is much higher than the true
resistivity R
t
, which results from the ‘‘skin effect’’ [9].
From 4927 to 4955 ft, R
a
is substantially lower than R
t
,
ELECTROMAGNETIC SUBSURFACE REMOTE SENSING 1259
which is caused by the socalled shoulder effect. The shoul
der effect arises when two adjacent lowresistance layers
generate strong signals, even though the tool is not in
these two regions. Around 5000 ft, there are a number of
thin layers, but the tool’s response fails to indicate them.
This failure results from the tool’s limited resolution,
which is represented in terms of the smallest thickness
that can be identiﬁed by the tool.
The blue line is the processed 6FF40 data after skin
effect boosting and a threepoint deconvolution. Skin effect
boosting is based on Eq. (19), which is solved iteratively
for the true conductivity from the apparent conductivity.
The threepoint deconvolution is performed under the as
sumption that the convolution in Eq. (30) is almost linear
[10]. These two methods do improve the ﬁnal results to
some degree, but they also cause spurious artifacts ob
served near 4880ft, since the two effects are considered
separately. The green curve is the response of the HRI
(highresolution induction) tool (trademark of Hallibur
ton) [11]. A complex coil conﬁguration is used to optimize
the geometric factor. After the raw data are obtained, a
nonlinear deconvolution based on the X signal is per
formed. The improvement in the ﬁnal results is signiﬁ
cant.
In 1991 Schlumberger Company released its AIT (array
induction image tool), which uses eight induction coil ar
rays operating at different frequencies [12]. The deconvo
lutions are performed in both radial and vertical
directions, and a quantitative twodimensional image of
formation resistivity is possible after a large number of
measurements [13,14].
The aforementioned data processing techniques are
based on the inverse deconvolution ﬁlter, which is compu
tationally effective and easily run in real time on a logging
truck computer. An alternative approach is to use inverse
scattering theory, which is becoming increasingly practi
cal and promising with the development of highspeed
computers [8,15].
Besides the induction method, there are other methods,
such as electrode methods and propagation methods.
Induction methods are suitable for the freshwater mud,
oilbase mud, or airﬁlled boreholes, since the little or no
conductivity in the borehole has a lesser effect on the
measurement. If the mud is very conductive, it will gen
erate a strong signal at the receiver and hence seriously
degrade the tool’s ability to make a deep reading. In such a
case, electrode methods are preferable, since the conduc
tive mud places the electrodes into better electrical con
tact with the formation. In the electrode methods, very low
frequencies (51000Hz) are used and Laplace’s equation is
solved instead of the Helmholtz equation. The typical tools
are DLL (dual laterolog) and SFL (spherical focusing log),
both from Schlumberger. The dual laterolog is intended
for both deep and shallow measurements, while the SFL is
for shallow measurements [16–19].
In addition, there are many tools mounted on pads to
perform shallow measurements on the borehole wall.
These may be just button electrodes mounted on a metal
lic pad. Due to their small size, they have high resolution
but a shallow depth of investigation. Their highresolution
capability can be used to map out ﬁne stratiﬁcations on
the borehole wall. When four pads are equipped with these
button electrodes, the resistivity logs they measure can be
correlated to obtain the dip of a geologic bed. An example
of this is the SHDT (stratigraphic highresolution dip
meter tool), also from Schlumberger [20].
When an array of buttons are mounted on a pad, they
can be used to generate a resistivity image of the borehole
10
3
10
2
10
1
10
0
10
−1
R
e
s
i
s
t
i
v
i
t
y
(
Ω
.
m
)
True resistivity
Raw data of 6FF40
With DC & SB
HRI data
HRI LOG and 6FF40 LOG
4850 4900 4950 5000
Depth (ft)
Figure 2. Apparent resistivity respons
es of a different tool in the Oklahoma
benchmark. The improvement of resolu
tion ability of the HR1 tool is significant.
1260 ELECTROMAGNETIC SUBSURFACE REMOTE SENSING
wall for formation evaluation, such as dips, cracks, and
stratigraphy. Such a tool is called a formation microscan
ner (FMS) and is available from Schlumberger [21].
For oilbased mud the SHDT does not work well, and
microinduction sensors have been mounted on a pad to
dipping bed evaluation. Such a tool is known as the
OBMDT (oilbased mud dipmeter tool) and is manufac
tured by Schlumberger [22,23].
Sometimes information is needed relating not only to
the conductivity but also to the dielectric permittivity. In
such cases, the EPT (electromagnetic wave propagation
tool), from Schlumberger can be used. The working fre
quency of EPT can be as high as hundreds of megahertz
to 1 GHz. At such high frequencies, the real part of e
0
is
dominant, as follows:
e
0
¼e þi
s
o
ð35Þ
EPT measurements provide information about dielectric
permittivity and hence can better distinguish fresh water
from oil. Water has a much higher dielectric constant
(80e
0
) compared to oil (2e
0
). Phase delays at two receivers
are used to infer the wave phase velocity and hence the
permittivity. Interested readers can ﬁnd materials on
these methods in Refs. 24 and 25.
Other techniques in electrical well logging include the
use of borehole radar. In such a case, a pulse is sent to a
transmitting antenna in the borehole, and the pulse echo
from the formation is measured at the receiver. Borehole
radar ﬁnds application in salt domes where the electro
magnetic loss is low. In addition, the nuclear magnetic
resonance (NMR) technique can be used to detect the per
centage of free water in a rock formation. The NMR signal
in a rock formation is proportional to the spin echos from
free protons that abound in free water. An example of such
a tool is the PNMT (pulsed nuclear magnetic resonance
tool), from Schlumberger [26].
2. GROUND PENETRATING RADAR
Another outgrowth of subsurface EM methods is ground
penetrating radar (GPR). Because of its numerous advan
tages, GPR has been widely used in geological surveying,
civil engineering, artiﬁcial target detection, and some
other areas.
The GPR design is largely application oriented. Even
though various systems have different applications and
considerations, their advantages can be summarized as
follows: (1) because the frequency used in GPR is much
higher than that used in the induction method, GPR has a
higher resolution; (2) since the antennas do not need to
touch the ground, rapid surveying can be achieved; (3) the
data retrieved by some GPR systems can be interpreted in
real time; and (4) GPR is potentially useful for organic
contaminant detection and nondestructive detection
[27–31].
On the other hand, GPR has some disadvantages, such
as shallow investigation depth and sitespecific applica
bility. The working frequency of GPR is much higher than
that used in the induction method. At such high frequen
cies, the soil is usually very lossy. Even though there is
always a tradeoff between the investigation depth and
resolution, a typical depth is no more than 10 mand highly
dependent on soil type and moisture content.
The working principle of GPR is illustrated in Fig. 3a
[28]. The transmitter T generates transient or continuous
EM waves propagating in the underground. Whenever a
change in the electrical properties of underground regions
is encountered, the wave is reﬂected and refracted. The
receiver R detects and records the reﬂected waves. From
the recorded data, information pertaining to the depth,
geometry, and material type can be obtained. As a simple
example, we use Figs. 3b and 3c to illustrate how the data
are recorded and interpreted. The underground contains
one interface, one cavity, and one lens. At a single position,
the receiver signals at different times are stacked along
the time axis. After completing the measurement at one
position, the procedure is iterated at all subsequent posi
tions. The ﬁnal results are presented in a twodimensional
map, which is called an echo sounder–type display. To
locate objects or interfaces, we need to know the wave
speed in the underground medium. The wavespeed in a
medium of relative dielectric permittivity e
r
is
C
s
¼
C
0
ﬃﬃﬃﬃ
e
r
p ð36Þ
where C
0
¼3 Â10
8
m/s. Usually, the transmitter and the
receiver are close enough, and thus the wave’s path of
propagation is considered to be vertical. The depth of the
(a)
(b)
(c)
Distance
Distance
Interface
Interface
T
i
m
e
T
i
m
e
Lens
Cavity
R T R T R
R
T
T
T
T
o
t
a
l
Figure 3. Working principle of the GPR. (Redrawn from Ref. 20.)
ELECTROMAGNETIC SUBSURFACE REMOTE SENSING 1261
interface is approximated as
D¼0:5 ÂðC
s
ÂT
total
Þ ð37Þ
where T
total
is the total wave propagation time.
A practical GPR system is much more complicated, and
a block diagram of a typical baseband GPR system is
shown in Fig. 4. Generally, a successful system design
should meet the following requirements [27]: (1) efﬁcient
coupling of the EM energy between antenna and ground,
(2) adequate penetration with respect to the target depth,
(3) sufﬁciently large return signal for detection, and
(4) adequate bandwidth for the desired resolution and
noise control.
The working frequency of typical GPR ranges from a
few tens of megahertz to several gigahertz, depending on
the application. The usual tradeoff holds—the wider the
bandwidth, the higher the resolution but the shallower
the penetration depth. A good choice is usually a tradeoff
between resolution and depth. Soil properties are also
critical in determining the penetration depth. It is ob
served experimentally that the attenuation of different
soils can vary substantially. For example, dry desert and
nonporous rocks have very low attenuation (about
1 dBm
À1
at 1 GHz) while the attenuation of sea water
can be as high as 300 dBm
À1
at 1 GHz. Some typical ap
plications and preferred operating frequencies are listed
in Table 1 [27].
To meet the requirements of different applications, a
variety of modulation schemes have been developed and
can be classiﬁed in the following three categories: ampli
tude modulation (AM), frequencymodulated continuous
wave (FMCW), and continuous wave (CW). We will briefly
discuss the advantages and limitations of each modulation
scheme.
There are two types of AM transmission used in GPR.
For investigation of lowconductivity medium, such as ice
and fresh water, a pulse modulated carrier is preferred
[32,33]. The carrier frequency can be chosen as low as tens
of megahertz. Since the reﬂectors are well spaced, a rel
atively narrow transmission bandwidth is needed. The re
ceiver signal is demodulated to extract the pulse envelope.
For shallow and highresolution applications, such as the
detection of buried artifacts, a baseband pulse is preferred
Source and
modulation
Ground (soil,
water, ice, etc.)
Signal sampling
and digitization
Receive
antenna
Transmit
antenna
Data
storage
Targets
Signal
processing
Display
Figure 4. Block diagram showing operation of a typical baseband GPR system. (Redrawn from
Ref. 19.)
Table 1. Desired Frequencies for Different Applications
a
Material
Typical Desired
Penetration
Depth
b
Approximate Maximum
Frequency at Which
Operation May Be
Usefully Performed
Cold pure fresh
water ice
10 km 10MHz
Temperate pure ice 1 km 2MHz
Saline ice 10m 50MHz
Fresh water 100m 100MHz
Sand (desert) 5 m 1 GHz
Sandy soil 3 m 1 GHz
Loam soil 3 m 500MHz
Clay (dry) 2 m 100MHz
Salt (dry) 1 km 250MHz
Coal 20m 500MHz
Rocks 20m 50MHz
Walls 0.3 m 10GHz
a
Redrawn from Ref. 19.
b
The ﬁgures used under this heading are the depths at which radar prob
ing gives useful information, taking into account the attenuation normally
encountered and the nature of the reﬂectors of interest.
1262 ELECTROMAGNETIC SUBSURFACE REMOTE SENSING
to avoid the problems caused by high soil attenuation,
since most of the energy is in the lowfrequency band. A
pulsetrain with a duration of 1 to 2 ns, a peak amplitude of
about 100V, and a repetition rate of 100kHz is applied to
the broadband antenna. The received signal is downcon
verted by sampling circuits before being displayed. There
are three primary advantages of the AM scheme: (1) it
provides a realtime display without the need for subse
quent signal processing, (2) the measurement time is
short, and (3) it is implemented with small equipment
but without synthesized sources and hence is costeffec
tive. But for the AM scheme, it is difﬁcult to control the
transmission spectrum, and the signaltonoise ratio
(SNR) is not as good as that of the FMCW method.
For the FMCW scheme, the frequency of the transmit
ted signal is continuously swept, and the receiver signal is
mixed with a sample of transmitted signals. The Fourier
transform of the received signal results in a timedomain
pulse that represents the receiver signal if a time domain
pulse were transmitted. The frequency sweep must be lin
ear in time to minimize signal degradation, and a stable
output is required to facilitate signal processing. The ma
jor advantage of the FMCW scheme is easier control of the
signal spectrum; the ﬁlter technique can be applied to ob
tain better SNR. A shortcoming of the FMCW system is
the use of a synthesized frequency source, which means
that the system is expensive and bulky. Additional data
processing is also needed before the display [34,35].
A continuouswave scheme was used in the early de
velopment of GPR, but now it is mainly employed in syn
thetic aperture and subsurface holography techniques
[36–38]. In these techniques, measurements are per
formed at a single or a few wellspaced frequencies over
an aperture at the ground surface. The wave front extrap
olation technique is applied to reconstruct the under
ground region, with the resolution depending on the size
of the aperture. Narrowband transmission is used and
hence highspeed data capture is avoided. The difﬁculty of
the CW scheme comes from the requirement for accurate
scanning of the twodimensional aperture. The operation
frequencies should be carefully chosen to minimize reso
lution degradation [27].
Antennas play an important role in the system perfor
mance. An ideal antenna should introduce the least dis
tortion on the signal spectrum or else one for which the
modiﬁcation can be easily compensated. Unlike the an
tennas used in the atmospheric radar, the antennas used
in GPR should be considered as loaded. The radiation pat
tern of the GPR antenna can be quite different due to the
strong interaction between the antenna and ground. Sep
arate antennas for transmission and reception are com
monly used, because it is difﬁcult to make a switch that is
fast enough to protect the receiver signal from the direct
coupling signal. The direct breakthrough signals will se
riously reduce the SNR and hence degrade the system
performance. Moreover, in a separateantenna system, the
orientation of antennas can be carefully chosen to reduce
further the crosscoupling level.
Except for the CW scheme, other modulation types re
quire wideband transmission, which greatly restricts the
choice of antenna. Four types of antennas, including ele
ment antennas, travelingwave antennas, frequencyinde
pendent antennas, and aperture antennas, have been
used in GPR designs. Element antennas, such as mono
poles, cylindrical dipoles, and biconical dipoles, are easy to
fabricate and hence widely used in GPR systems. Orthog
onal arrangement is usually chosen to maintain a low
level of crosscoupling. To overcome the limitation of nar
row transmission bandwidth of thin dipole or monopole
antennas, the distributed loading technique is used to
expand the bandwidth at the expense of reduced efﬁ
ciency [39–42].
Another commonly used antenna type is traveling
wave antennas, such as longwire antennas, Vshaped an
tennas, and rhombic antennas. The travelingwave anten
nas distinguish themselves from standingwave antennas
in the sense that the current pattern is a traveling wave
rather than a standing wave. Standingwave antennas,
such as halfwave dipoles, are also referred to as resonant
antennas and are narrowband, while travelingwave an
tennas are broadband. The disadvantage of traveling
wave antennas is that half of the power is wasted at the
matching resistor [43,44].
Frequencyindependent antennas are often preferred
in the impulse GPR system. It has been proved that if the
antenna geometry is speciﬁed only by angles, its perfor
mance will be independent of frequency. In practice, we
have to truncate the antenna, due to its limited outer size
and inner feeding region, which determine the lower
bound and upper bound of the frequency, respectively. In
general, this type of antenna will introduce nonlinear
phase distortion, which results in an extended pulse
response in the time domain [27,45]. A phase correction
procedure is needed if the antenna is used in a high
resolution GPR system.
A wire antenna is a onedimensional antenna that has
a small effective area and hence lower gain. For some GPR
systems, higher gain or a more directive radiation pattern
is sometimes required. Aperture antennas, such as horn
antennas, are preferred because of their large effective
area. A ridge design is used to improve the bandwidth and
reduce the size. Ridged horns with gain better than
10 dBm over a range of 0.3–2 GHz and VSWR lower than
1.5 over a range of 0.2–1.8 GHz have been reported [46].
Since many aperture antennas are fed via waveguides, the
phase distortion associated with the different modulation
schemes needs to be considered.
Generally, antennas used in GPR systems require
broad bandwidth and linear phase in the operating fre
quency range. Since the antennas work in close proximity
to the ground surface, the interaction between them must
be taken into account.
Signal processing is one of the most important parts in
the GPR system. Some modulation schemes directly give
the timedomain data while the signals of other schemes
need to be demodulated before the information is avail
able. Signal processing can be performed in the time do
main, frequency domain, or space domain. A successful
signal processing scheme usually consists of a combina
tion of several kinds of processing techniques that are ap
plied at different stages. Here, we outline some basic
signal processing techniques involved in the GPR system.
ELECTROMAGNETIC SUBSURFACE REMOTE SENSING 1263
The ﬁrst commonly used method is noise reduction by
time averaging. It is assumed that the noise is random, so
that the noise can be reduced to 1/Nt by averaging N iden
tical measurements spaced in time t. This technique only
works for random noise but has no effect on the clutter.
Clutter reduction can be achieved by subtracting the
mean. This technique is performed under the assumption
that the statistics of the underground are independent of
position. A number of measurements are performed at a
set of locations over the same material type to obtain the
mean, which can be considered as a measure of the system
clutter.
The frequency ﬁlter technique is commonly used in the
FMCW system. Signals that are not in the desired infor
mation bandwidth are rejected. Thus the SNR of the
FMCW scheme is usually higher than that of the AM
scheme.
In some very lossy soils, the return signal is highly at
tenuated, which makes interpretation of the data difﬁcult.
If the material attenuation information is available, the
results can be improved by exponentially weighting the
time traces to counter the decrease in signal level due to
the loss. In practice, this is done by using a specially de
signed ampliﬁer. Caution is needed when using this meth
od, since the noise can also increase in such a system [27].
3. MAGNETOTELLURIC METHODS
The basic idea of the magnetotelluric (MT) method is to
use natural electromagnetic ﬁelds to investigate the elec
trical conductivity structure of Earth. This method was
ﬁrst proposed by Tikhonov in 1950 [47]. In his paper, the
author assumed that Earth’s crust is a planar layer of ﬁ
nite conductivity lying on an ideally conducting substrate,
such that a simple relation between the horizontal com
ponents of the E and H ﬁelds at the surface can be found
[48]
im
0
oH
x
ﬃE
y
g coshðglÞ ð38Þ
where
g ¼ðism
0
oÞ
ð1=2Þ
ð39Þ
The author used the data observed at Tucson (Arizona)
and Zui (former USSR) to compute the value of conduc
tivity and thickness of the crust that best ﬁt the ﬁrst four
harmonics. For Tucson, the conductivity and thickness
were about 4.0 Â 10
À3
S/m and 1000km, respectively. For
Zui, the corresponding values are 3.0 Â 10
À1
S/m and
100km.
The MT method distinguishes itself from other subsur
face EM methods because verylowfrequency natural
sources are used. The actual mechanisms of natural sourc
es have been under discussion for a long time, but now it is
well accepted that the sources of frequency above 1 Hz are
thunderstorms while the sources below 1 Hz are due to the
current system in the magnetosphere caused by solar ac
tivity. In comparison with other EM methods, the use of a
natural source is a major advantage. The frequencies used
range from 0.001 Hz to 10
4
Hz, and thus investigation
depth can be achieved from 50 to 100m to several kilome
ters. Installation is much simpler and has less impact on
the environment. The MT method has also proved very
useful in some extreme areas where conventional seismic
methods are expensive or ineffective. The main shortcom
ings of the MT method are limited resolution and difﬁculty
in achieving a high SNR, especially in electrically noisy
areas [49].
In MT measurements, the timevarying horizontal elec
trical and magnetic ﬁelds at the surface are recorded si
multaneously. The data recorded in the time domain are
ﬁrst converted into frequencydomain data by using a fast
Fourier transform (FFT). An apparent conductivity is then
deﬁned as a function of frequency. To interpret the data,
theoretical apparent conductivity curves are generated by
the model studies. The model whose apparent conductivity
curve best matches the measurement data is taken as an
approximate model of the subsurface.
Since it is more convenient and meaningful to repre
sent the apparent conductivity in terms of skin depth, we
ﬁrst introduce the concept of skin depth by studying a
simple case. The model we use is shown in Fig. 5, which
consists of a homogeneous medium with conductivity s
and a uniform current sheet ﬂowing along the x direction
in the xy plane. If the density of current at the ground
(z ¼0) is represented as [50]
I
x
¼ cos ot; I
y
¼I
z
¼0 ð40Þ
then the current density at depth z is
I
x
¼e
Àz
ﬃﬃﬃﬃﬃﬃﬃﬃ
2oms
p
=2
cosðot Àz
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
2oms
_
Þ; I
y
¼I
z
¼0 ð41Þ
When z increases, we notice that the amplitude of the
current decreases exponentially with respect to z; mean
while the phase retardation progressively increases. To
describe the amplitude attenuation, we introduce the skin
depth p as [50]
p¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃ
2
oms
¸
ð42Þ
where the current amplitude decreases to e
À1
of the cur
rent at the surface. Since the unit in Eq. (42) is not
σ
O
Y
Z
X
Conductivity:
Current sheet
Figure 5. Current sheet ﬂowing on Earth’s surface, used to ex
plain magnetotelluric method.
1264 ELECTROMAGNETIC SUBSURFACE REMOTE SENSING
convenient, some prospectors like to use the following
formula
p¼
1
2p
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
10rT
_
ð43Þ
where Tis the period in seconds, r is the resistivity in O/m,
and the unit for p is km. The skin depth indicates the
depth the wave can penetrate the ground. For example, if
the resistivity of the underground is 10 S/m and the period
of the wave is 3 s, the skin depth is 2.76 km. Subsurface
methods seldom have such a great penetration depth.
The data interpretation of the MT method is based on
the model studies. Earth is modeled as a two or three
layer medium. For a twolayer model as shown in Fig. 6,
the general expression for the ﬁeld can be written as [50]
0rzrh:
E
z
¼Ae
a
ﬃﬃﬃﬃ
s
1
p
z
þbe
Àa
ﬃﬃﬃﬃ
s
1
p
z
ð44aÞ
H
y
¼e
ip=4
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
2s
1
T
_
½ÀAe
a
ﬃﬃﬃﬃ
s
1
p
z
þBe
Àa
ﬃﬃﬃﬃ
s
1
p
z
ð44bÞ
hrzrN:
E
x
¼e
Àa
ﬃﬃﬃﬃ
s
2
p
z
ð45aÞ
H
x
¼e
ip=4
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
2s
2
T
_
e
Àa
ﬃﬃ
s
p
z
ð45bÞ
where h is the thickness of the upper layer, and s
1
, s
2
are
the conductivities of the upper and lower layers, respec
tively. Matching the boundary conditions at z ¼h, we have
A¼
ﬃﬃﬃﬃﬃ
s
1
p
À
ﬃﬃﬃﬃﬃ
s
2
p
2
ﬃﬃﬃﬃﬃ
s
1
p e
Àahð
ﬃﬃﬃﬃ
s
1
p
þ
ﬃﬃﬃﬃ
s
2
p
Þ
ð46Þ
B¼
ﬃﬃﬃﬃﬃ
s
1
p
þ
ﬃﬃﬃﬃﬃ
s
2
p
2
ﬃﬃﬃﬃﬃ
s
1
p e
ahð
ﬃﬃﬃﬃ
s1
p
À
ﬃﬃﬃﬃ
s2
p
Þ
ð47Þ
Since we are interested in the ratio between the E and H
ﬁeld on the surface, Eq. (44) can be rewritten for z ¼0 as
E
x
H
y
¼
1
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
2s
1
T
p
M
N
e
Àiðp=4þfþcÞ
ð48Þ
where M, N, f, and c satisfy the following equations:
M cos f¼
1
p
1
cosh
h
p
1
þ
1
p
2
sinh
h
p
1
_ _
cos
h
p
1
ð49aÞ
M sin f¼
1
p
1
sinh
h
p
1
þ
1
p
2
cosh
h
p
1
_ _
sin
h
p
1
ð49bÞ
N cos c¼
1
p
1
sinh
h
p
1
þ
1
p
2
cosh
h
p
1
_ _
cos
h
p
1
ð49cÞ
N sin c¼
1
p
1
cosh
h
p
1
þ
1
p
2
sinh
h
p
1
_ _
sin
h
p
1
ð49dÞ
where p
1
, p
2
are the skin depths of upper and lower layers,
respectively.
For a multilayer medium, after applying the same pro
cedure, we can obtain exactly the same relation between
E
x
and H
y
as shown in Eq. (48) except that the expressions
for M, N, f, and c are much more complicated. Because of
this similarity, we have
E
x
H
y
¸
¸
¸
¸
¸
¸
¸
¸
¼
1
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
2s
a
T
p ¼
M
N
1
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
2s
1
T
p ð50Þ
where s
a
is deﬁned as the apparent conductivity. If the
medium is homogeneous, the apparent conductivity is
equal to the true conductivity. In a multilayer medium
the apparent conductivity is an average effect of all layers.
To obtain a better understanding of the preceding for
mulas, we ﬁrst study two twolayer models and their cor
responding apparent conductivity curves, as shown in
Fig. 7 (51). At very low frequencies, the wave can easily
penetrate the upper layer, and thus its conductivity has
little effect on the apparent conductivity. Consequently,
the apparent resistivity approaches the true resistivity of
the lower layer. As the frequency increases, less energy
can penetrate the upper layer due to the skin effect, and
thus the effect from the upper layer is dominant. As a re
sult, the apparent resistivity is asymptotic to r
1
. Compar
ing the two curves, we note that both of them change
smoothly, and for the same frequency, case A has lower
h
Z = h
Z = 0
Z
First layer o
1
Second layer o
2
O
Figure 6. Twolayer model of Earth’s crust, used to demonstrate
the responses of the magnetotelluric method.
1
ρ
2
ρ
1
ρ
1
ρ
A
A
B
B
Conductive sediments
Resistive basement
A
p
p
a
r
e
n
t
r
e
s
i
s
t
i
v
i
t
y
Low High Frequency
Figure 7. Diagrammatic twolayer apparent resistivity curves
for the models shown. (Redrawn from Ref. 43.)
ELECTROMAGNETIC SUBSURFACE REMOTE SENSING 1265
apparent resistivity than case B, since the conductive sed
iments of case B are thicker.
Our next example is a threelayer model as shown in
Fig. 8 [51]. The center layer is more conductive than the
two adjacent ones. As expected, the curve approaches r
1
and r
2
at each end. The existence of the center conductive
bed is obvious from the curve, but the apparent resistivity
never reaches the true resistivity of the center layer, since
its effect is averaged by the effects from the other two
layers.
So far we have only discussed the horizontally layered
medium, which is a onedimensional model. In practice,
twodimensional or even threedimensional structures are
often encountered. In a 2D case, the conductivity changes
not only along the z direction but also along one of the
horizontal directions. The other horizontal direction is
called the ‘‘strike’’ direction. If the strike direction is not
in the x or y direction, we obtain a general relation be
tween the horizontal ﬁeld components as [51]
E
x
¼Z
xx
H
x
þZ
xy
H
y
ð51aÞ
E
y
¼Z
yx
H
x
þZ
yy
H
y
ð51bÞ
Since E
x
, E
y
, H
x
, and H
y
are generally out of phase, Z
i, j
are
complex numbers. It can also be shown that Z
i, j
have the
following properties:
Z
xx
þZ
yy
¼0 ð52Þ
Z
xy
ÀZ
yx
¼constant ð53Þ
A simple vertical layer model and its corresponding
curves are shown in Fig. 9 [51]. In Fig. 9b, the apparent
resistivity with respect to E

changes slowly from r
1
to r
2
because of the continuity of H
>
and E

across the inter
face. On the other hand, the apparent resistivity corre
sponding to E
>
has an abrupt change across the contact,
since the E
>
is discontinuous at the interface. The relative
amplitude of H
>
varies significantly around the interface
and approaches a constant at a large distance, as shown in
Fig. 9d. This is caused by the change in current density
near the interface, as shown in Fig. 9f. We also observe
that H
z
appears near the interface, as shown in Fig. 9c.
The reason is that the partial derivative of E

with re
spect to > direction is nonzero.
We have discussed the responses in some idealized
models. For more complicated cases, their response curves
can be obtained by forward modeling. Since the measure
ment data are in the time domain, we need to convert
them into the frequency domain data by using a Fourier
transform. In practice, ﬁve components are measured.
There are four unknowns in Eqs. (51a) and (51b), but
only two equations. This difﬁculty can be overcome by
making use of the fact that Z
i,j
changes very slowly with
frequency. In fact, Z
i,j
is computed as an average over a
frequency band that contains several frequency sample
points. A commonly used method is given in Ref. 52,
according to which Eq. (51a) is rewritten as
E
x
A
Ã
h i ¼Z
xx
H
x
A
Ã
h i þZ
xy
H
y
A
Ã
¸ _
ð54Þ
and
E
x
B
Ã
h i ¼Z
xx
H
x
B
Ã
h i þZ
xy
H
y
B
Ã
¸ _
ð55Þ
1
ρ
3
ρ
2
ρ
1
ρ
3
ρ
Resistive sediments
2
ρ Conductive sediments
Resistive basement
A
p
p
a
r
e
n
t
r
e
s
i
s
t
i
v
i
t
y
Low High Frequency
Figure 8. Diagrammatic threelayer apparent resistivity curve
for the model shown. (Redrawn from Ref. 43.)
ρ
0
1
2
ρ
1
ρ
1
ρ
2
1
0
1
0
2
(> )
Model
Distance
E
⊥
E
⊥

Current flow E
⊥
E
Current flow E

Apparent
resistivity
Tipper
Relative
(a)
(b)
(c)
(d)
(e)
(f)
D
e
p
t
h
D
e
p
t
h
0
D
e
p
t
h
 H
z

 H
⊥

 H
⊥

H
⊥
ρ
Figure 9. Diagrammatic response curves for a simple vertical
contact at frequency f. (Redrawn from Ref. 43.)
1266 ELECTROMAGNETIC SUBSURFACE REMOTE SENSING
where A* and B* are the complex conjugates of any two of
the horizontal ﬁeld components. The crosspowers are de
ﬁned as
AB
Ã
h iðo
1
Þ ¼
1
Do
_
o
1
þðDo=2Þ
o
1
ÀðDo=2Þ
AB
Ã
do ð56Þ
There are six possible combinations, and the pair (H
x
, H
y
)
is preferred in most cases due to its greater degree of
independence. Solving Eqs. (54) and (55), we have
Z
xx
¼
E
x
A
Ã
h i H
y
B
Ã
¸ _
À E
x
B
Ã
h i H
y
A
Ã
¸ _
H
x
A
Ã
h i H
y
B
Ã
¸ _
À H
x
B
Ã
h i H
y
A
Ã
¸ _ ð57aÞ
and
Z
xy
¼
E
x
A
Ã
h i H
x
B
Ã
h i À E
x
B
Ã
h i H
x
A
Ã
h i
H
y
A
Ã
¸ _
H
x
B
Ã
h i À H
y
B
Ã
¸ _
H
x
A
Ã
h i
ð57bÞ
Applying the same procedure to Eq. (51b), we have
Z
yx
¼
E
y
A
Ã
¸ _
H
y
B
Ã
¸ _
À E
y
B
Ã
¸ _
H
y
A
Ã
¸ _
H
x
A
Ã
h i H
y
B
Ã
¸ _
À H
x
B
Ã
h i H
y
A
Ã
¸ _ ð57cÞ
and
Z
yy
¼
E
y
A
Ã
¸ _
H
x
B
Ã
h i À E
y
B
Ã
¸ _
H
x
A
Ã
h i
H
y
A
Ã
¸ _
H
x
B
Ã
h i À H
y
B
Ã
¸ _
H
x
A
Ã
h i
ð57dÞ
After obtaining Z
i,j
, they can be substituted into Eqs.
(51a) and (51b) to solve for the other pair (E
x
, E
y
), which is
then used to check the measurement data. The difference
is due either to noise or to measurement error. This pro
cedure is usually used to verify the quality of the mea
sured data.
4. AIRBORNE ELECTROMAGNETIC METHODS
Airborne EM methods (AEMs) are widely used in geologic
surveys and prospecting for conductive ore bodies. These
methods are suitable for large area surveys because of
their speed and costeffectiveness. They are also preferred
in some areas where access is difﬁcult, such as swamps or
icecovered areas. In contrast to ground EM methods, air
borne EM methods are usually used to outline largescale
structures while ground EM methods are preferred for
more detailed investigations [53].
The difference between airborne and ground EM sys
tems results from the technical limitations inherent in the
use of aircraft. The limited separation between transmit
ter and receiver determines the shallow investigation
depth, usually from 25 to 75 m. Even though greater pen
etration depth can be achieved by placing the transmitter
and receiver on different aircraft, the disadvantages are
obvious.
The transmitters and receivers are usually 200–500ft
above the surface. Consequently, the amplitude ratio of
the primary ﬁeld to the secondary ﬁeld becomes very small
and thus the resolution of airborne EM methods is not
very high. The operating frequency is usually chosen from
300 to 4000Hz. The lower limit is set by the transmission
effectiveness, and the upper limit is set by the skin depth.
Based on different design principles and application
requirements, many systems have been built and operated
all over the world since the 1940s. Despite the tremendous
diversity, most airborne EM systems can be classiﬁed in
one of the following categories according to the quantities
measured: phase component measuring systems, quad
rature systems, rotating ﬁeld systems, and transient re
sponse systems [54].
For a phase component measuring system, the inphase
and quadrature components are measured at a single
frequency and recorded as parts per million (ppm) of the
primary ﬁeld. In the system design, vertical loop arrange
ments are preferred, since they are more sensitive to the
steeply dipping conductor and less sensitive to the hori
zontally layered conductor [55]. Accurate maintenance of
transmitterreceiver separation is essential and can be
achieved by ﬁxing the transmitter and receiver at the
two wingtips. Once this requirement is satisﬁed, a sensi
tivity of a few ppm can be achieved [54]. A diagram of the
phase component measuring system is shown in Fig. 10
[55]. A balancing network associated with the reference
loop is used to buck the primary ﬁeld at the receiver.
The receiver signal is then fed to two phasesensitive
demodulators to obtain the inphase and quadrature com
ponents. Lowpass ﬁlters are used to reject veryhigh
frequency signals that do not originate from the earth.
The data are interpreted by matching the curves obtained
from the modeling. Some response curves of typical struc
tures are given in Ref. 56.
Reference
loop
Receiving
loop
Bucking
loop
Filter
Amplifier
Amplifier Amplifier
Amplifier Amplifier
Filter
Amplifier
90° phase
shifter
Demodulator Demodulator
Transmitting
loop
Transmitter
Balance
network
Preamplifier
Recorder Recorder
Integrator
and filter
Integrator
and filter
Figure 10. Block diagram showing operation of a typical phase
component measuring system. (Redrawn from Ref. 43.)
ELECTROMAGNETIC SUBSURFACE REMOTE SENSING 1267
The quadrature system employs a horizontal coil
placed on the airplane as a transmitter and a vertical
coil towed behind the plane as a receiver. The vertical coil
is referred to as a ‘‘towed bird.’’ Since only the quadrature
component is measured, the separation distance is less
critical. To reduce the noise further, an auxiliary horizon
tal coil, powered with a current 901 out of phase with re
spect to the main transmitter current, is used to cancel the
secondary ﬁeld caused by the metal body of the aircraft.
Since the response at a single frequency may have two
interpretations, two frequencies are used to eliminate the
ambiguity. The lower frequency is about 400Hz, and the
higher one is chosen from 2000 to 2500 Hz. The system
responses in different environments can be obtained by
model studies. Reference 57 gives a number of curves for
thin sheets and shows the effects of variation in depth,
dipping angle, and conductivity.
In an airborne system, it is hard to control the relative
rotation of receiver and transmitter. The rotating ﬁeld
method is introduced to overcome this difﬁculty. Two
transmitter coils are placed perpendicular to each other
on the plane, and a similar arrangement is used for the
receiver. The two transmitters are powered with current
of the same frequency shifted 901 out of phase, so that the
resultant ﬁeld rotates about the axis, as shown in Fig. 11
[58]. The two receiver signals are phaseshifted by 901
with respect to each other, and then the inphase and
quadrature differences at the two receivers are ampliﬁed
and recorded by two different channels. Over a barren
area, the outputs are set to zero. When the system is
within a conducting zone, anomalies in the conductivity
are indicated by nonzero outputs in both the inphase and
quadrature channels. The noise introduced by the ﬂuctu
ation of orientation can be reduced by this scheme, but it is
relatively expensive and the data interpretation is com
plicated by the complex coil system [58].
The fundamental problem of airborne EM systems is
the difﬁculty in detecting the relatively small secondary
ﬁeld in the presence of a strong primary ﬁeld. This difﬁ
culty can be alleviated by using the transient ﬁeld method.
A wellknown system based on the transient ﬁeld method
is INPUT (induced pulsed transient) [59], which was de
signed by Barringer during the 1950s. In the INPUT sys
tem, a large horizontal transmitting coil is placed on the
aircraft and a vertical receiving coil is towed in the bird
with the axis aligned with the ﬂight direction.
The working principle of INPUT is shown in Fig. 12
[60]. A half–sine wave with a duration of about 1.5 ms and
quiet period of about 2.0 ms is generated as the primary
ﬁeld, as shown in Fig. 12a. If there are no conducting
zones, the current in the receiver is induced only by the
primary ﬁeld, as shown in Fig. 12b. In the presence of
conductive anomalies, the primary ﬁeld will induce an
eddy current. After the primary ﬁeld is cut off, the eddy
current decays exponentially. The duration of the eddy
current is proportional to the conductivity anomalies, as
shown in Fig. 12c. The higher the conductivity, the longer
the duration time. The decay curve in the quiet period is
sampled successively in time by six channels and then
displayed on a strip, as shown in Fig. 13. As we can see,
Flight
direction
Transmitter
Receiver
Output
π
π
2
2
Figure 11. Working principle of the rotary ﬁeld AEM system.
(Redrawn from Ref. 50.)
(c)
(b)
(a)
Primary current
(and magnetic field)
2.0 ms
1.5 ms
Time
Current induced in R
by primary field alone
Total field
induced in R
Decay curve due to
good conductors
Channel sample
periods
1 3 5
Decay curve due to
poor conductor
Figure 12. Working principle of the INPUT system. (Redrawn
from Ref. 52.)
1268 ELECTROMAGNETIC SUBSURFACE REMOTE SENSING
the distortion caused by a good conductor appears in all
the channels, while the distortion corresponding to a poor
conductor only registers on the early channels.
Since the secondary ﬁeld can be measured more accu
rately in the absence of the primary ﬁeld, transient sys
tems provide greater investigation depths, which may
reach 100 m under favorable conditions. In addition,
they can also provide a direct indication of the type of
conductor encountered [58].
On the other hand, this system design gives rise to
other problems inherent in the transient method. Since
the eddy current in the quiet period becomes very small, a
more intense source has to be used in order to obtain the
same signal level as that in the continuouswave method.
The circuitry for the transient system is much more com
plicated, and it is more difﬁcult to reject the noise due to
the wideband property of the transient signal.
BIBLIOGRAPHY
1. C. M. Swift, Jr., Fundamentals of the electromagnetic meth
od, in M. N. Nabighian, ed., Electromagnetic Methods in
Applied Geophysics: Theory, Society of Exploration Geophys
icists, Tulsa, OK, 1988.
2. H. G. Doll, Introduction to induction logging and application
to logging of wells drilled with oil based mud, J. Petroleum
Technol. 1:148–162 (1949).
3. J. H. Moran and K. S. Kunz, Basic theory of induction logging
and application to study of twocoil sondes, Geophysics
27(6):829–858 (1962).
4. S. J. Thandani and H. E. Hall, Propagated geometric factors
in induction logging, Trans. SPWLA, Vol. 2, Paper WW, 1981.
5. B. Anderson, Induction sonde response in stratiﬁed medium,
The Log Analyst XXIV(1):25–31.
6. W. C. Chew, Response of a current loop antenna in an invaded
borehole, Geophysics 49:81–91 (1984).
7. J. R. Lovell, Finite Element Method in Resistivity Logging,
Schlumberger Technology Corp., Ridgeﬁeld, CT, 1993.
8. W. C. Chew and Q. H. Liu, Inversion of induction tool mea
surements using the distorted Born iterative method and CG
FFHT, IEEE Trans. Geosci. Remote Sens. 32(4):878–884
(1994).
9. S. Gianzero and B. Anderson, A new look at skin effect, The
Log Analyst 23(1):20–34 (1982).
10. L. C. Shen, Effects of skineffect correction and threepoint
deconvolution on induction logs, The Log Analyst (July–Aug.)
217 (1989).
11. R. Strickland et al., New developments in the high resolution
induction log, Trans. SPWLA, Vol. 2, Paper ZZ, 1991.
12. T. D. Barber and R. A. Rosthal, Using a multiarray induction
tool to achieve high resolution logs with minimum environ
mental effects, Trans. SPE, Paper SPE 22725, 1991.
13. G. P. Grove and G. N. Minerbo, An adaptive borehole correc
tion scheme for array induction tools, presented at the 32nd
Ann. SPWLA Symp., Midland, TX, 1991.
14. Schlumberger Educational Services, AIT Array Induction
Image Tool, 1992.
15. R. Freedman and G. N. Minerbo, Maximum entropy inversion
of induction log data, Trans. SPE 5:381–394, Paper SPE
19608, 1989.
16. S. J. Grimaldi, P. Poupon, and P. Souhaite, The dual laterolog
R
xo
tool, Trans. SPE 2:1–12, Paper SPE 4018, 1972.
17. R. Chemali et al., The shoulder bed effect on the dual laterolog
and its variation with the resistivity of the borehole ﬂuid,
Trans. SPWLA, Paper UU, 1983.
18. Q. Liu, B. Anderson, and W. C. Chew, Modeling low frequency
electrodetype resistivity tools in invaded thin beds, IEEE
Trans. Geosci. Remote Sens. 32(3):494–498 (1994).
19. B. Anderson and W. C. Chew, SFL interpretation using high
speed synthetic computer generated logs, Trans. SPWLA, Pa
per K, 1985.
20. Y. Chauvel, D. A. Seeburger, and C. O. Alfonso, Application of
the SHDTstratigraphic high resolution dipmeter to the study
of depositional environments, Trans. SPWLA, Paper G, 1984.
21. A. R. Badr and M. R. Ayoub, Study of a complex carbonate
reservoir using the Formation MicroScanner (FMS) tool, Proc
6th Middle East Oil Show, Bahrain, March, 1989, pp. 507–
516.
22. R. L. Kleinberg et al., Microinduction sensor for the oilbased
mud dipmeter, SPE Formation Evaluation, Dec. 1988, Vol. 3,
pp. 733–742.
23. W. C. Chew and R. L. Kleinberg, Theory of microinduction
measurements, IEEE Trans. Geosci. Remote Sens. 26:707–719
(1988).
24. W. C. Chew and S. Gianzero, Theoretical investigation of the
electromagnetic wave propagation tool, IEEE Trans. Geosci.
Remote Sens. GE19:1–7 (1981).
25. W. C. Chew et al., An effective solution for the response of
electrical welllogging tool in a complex environment, IEEE
Trans. Geosci. Remote Sens. 29:303–313 (1991).
26. D. D. Grifﬁn, R. L. Kleinberg, and M. Fukuhara, Lowfre
quency NMR spectrometer measurement, Sci. Technol.
4:968–975 (1993).
27. D. J. Daniels, D. J. Gunton, and H. F. Scott, Introduction to
subsurface radar, IEEE Proc. 135:278–320 (1988).
28. D. K. Butler, Elementary GPR overview, Proc. Government
Users Workshop on Ground Penetrating Radar Application
and Equipment, 1992, pp. 25–30.
29. W. H. Weedon and W. C. Chew, Broadband microwave inverse
scattering for nondestructive evaluation, Proc. 20th Annual
Review Progress Quantitative Nondestructive Evaluation,
Drunswick, ME, 1993.
30. F. C. Chen and W. C. Chew, Timedomain ultrawideband mi
crowave imaging radar system, Proc. IEEE Instrum. Meas.
Technol. Conf., St. Paul, MN, 1998, pp. 648–650.
31. F. C. Chen and W. C. Chew, Development and testing of the
timedomain microwave nondestructive evaluation system,
Poor conductor anomalies
appear on early channels
Good conductor anomalies
appear on early channels
Channel 1
Channel 2
Channel 3
Channel 4
Channel 5
Channel 6
Figure 13. Responses of different anomalies appearing on differ
ent channels. (Redrawn from Ref. 52.)
ELECTROMAGNETIC SUBSURFACE REMOTE SENSING 1269
Review of Progress in Quantitative Evaluation, Vol. 17,
Plenum, New York, 1998, pp. 713–718.
32. M. Walford, Exploration of temperate glaciers, Phys. Bull.
36:108–109 (1985).
33. D. K. Hall, A review of the utility of remote sensing in Alas
kan permafrost studies, IEEE Trans. Geosci. Remote Sens.
GE20:390–394 (1982).
34. A. Z. Botros et al., Microwave detection of hidden objects in
walls, Electron. Lett. 20:379–380 (1984).
35. P. Dennis and S. E. Gibbs, Solidstate linear FM/CW radar
systems—their promise and their problems, Proc. IEEE MTT
Symp., 1974, pp. 340–342.
36. A. P. Anderson and P. J. Richards, Microwave imaging of sub
surface cylindrical scatters from crosspolar backscatter, Elec
tron. Lett. 13:617–619 (1977).
37. K. Lizuka et al., Hologram matrix radar, Proc. IEEE,
64:1493–1504 (1976).
38. N. Osumi and K. Ueno, Microwave holographic imaging
method with improved resolution, IEEE Trans. Anten. Prop
ag. AP32: 1018–1026 (1984)
39. M. C. Bailey, Broadband halfwave dipole, IEEE Trans.
Anten. Propag. AP32:410–412 (1984).
40. R. P. King, Antennas in material media near boundaries with
application to communication and geophysical exploration,
IEEE Trans. Anten. Propag. AP34:483–496 (1986).
41. M. Kanda, A relatively short cylindrical broadband antenna
with tapered resistive loading for picosecond pulse measure
ments, IEEE Trans. Anten. Propag., AP26:439–447 (1978).
42. C. A. Balanis, Antenna Theory, Wiley, New York, 1997,
Chap. 9.
43. P. Degauque and J. P. Thery, Electromagnetic subsurface
radar using the transient radiated by a wire antenna, IEEE
Trans. Geosci. Remote Sens. GE24:805–812 (1986).
44. C. A. Balanis, Antenna Theory, Wiley, New York, 1997,
Chap. 10.
45. C. A. Balanis, Antenna Theory, Wiley, New York, 1997,
Chap. 11.
46. J. L. Kerr, Short axial length broadband horns, IEEE Trans.
Anten. Propag. AP21:710–714 (1973).
47. A. N. Tikhonov, Determination of the electrical characteristics
of the deep strate of the earth’s crust, Proc. Acad. Sci. USSR
83(2):295–297 (1950).
48. J. R. Wait, Theory of magnetotelluric ﬁeld, J. Res. Natl. Bur.
Stand. D, Radio Propagation, 66D:509–541 (1962).
49. K. Vozoff, The magnetotelluric method, in M. N. Nabighian,
ed., Electromagnetic Methods in Applied Geophysics:
Application, Society of Exploration Geophysicists, Tulsa,
OK, 1988.
50. L. Cagniard, Basic theory of the magnetotelluric method of
geophysical prospecting, Geophysics 18:605–635 (1952).
51. K. Vozoff, The magnetotelluric method in the exploration of
sedimentary basins, Geophysics 37:98–141 (1972).
52. T. Madden and P. Nelson, A defense of Cagniard’s magneto
telluric method, Geophysics Reprinted Series 5: Magnetotell
uric Methods, Society of Exploration Geophysicists, Tulsa,
OK, 1985, pp. 89–102.
53. G. J. Palacky and G. F. West, Airborne electromagnetic meth
ods, in M. N. Nabighian, ed., Electromagnetic Methods in
Applied Geophysics: Theory, Society of Exploration Geophys
icists, Tulsa, OK, 1988.
54. J. C. Gerkens, Foundation of Exploration Geophysics, Else
vier, New York, 1989.
55. G. V. Keller and F. Frischknecht, Electrical Methods in
Geophysical Prospecting, Pergamon, New York, 1966.
56. D. Boyd and B. C. Roberts, Model experiments and survey
results from a wing tipmounted electromagnetic prospecting
system, Geophys. Proc. 9:411–420 (1961).
57. N. R. Patterson, Experimental and ﬁeld data for the dual
frequency phaseshift method of airborne electromagnetic
prospecting, Geophysics 26:601–617 (1961).
58. P. Kearey and M. Brooks, An Introduction to Geophysical
Exploration, Blackwell Scientific, Boston, 1984.
59. A. R. Barringer, The INPUT electrical pulse prospecting
system, Min. Cong. J. 48:49–52 (1962).
60. A. E. Beck, Physical Principles of Exploration Methods,
Macmillan, New York, 1981.
ELECTROMAGNETIC SURFACE WAVES
ZHIPENG WU
UMIST
Manchester, United Kingdom
1. INTRODUCTION
In electromagnetic waves, the term ‘‘surface wave’’ has
been used to describe the nonradiating electromagnetic
waves propagating along the interface between two media
[1,2]. It has also been used to describe the electromagnetic
waves propagating along the air–ground interface pro
duced by an antenna on or near the ground surface [3–10].
Both types of surface wave will be introduced in this ar
ticle, and the ‘‘surface wave’’ is broadly deﬁned as the
wave propagating along an interface separating two me
dia, or an interface between air and a layered structure.
The surface waves considered include
1. The Zenneck surface wave
2. Surface wave on a conductorbacked dielectric slab
3. Radial cylindrical surface wave on a ﬂat surface
4. Axial cylindrical surface wave
5. Surface wave on a dielectriccoated conducting cyl
inder
6. Norton and trapped surface waves on ﬂat ground
interface
7. Surfacewave propagation over an inhomogeneous
ﬂat surface
8. Surfacewave propagation over a spherical ground
The governing equations and properties of these surfaces
waves are presented in the following section.
2. SURFACE WAVES
2.1. Zenneck Wave
The Zenneck wave is a solution of Maxwell’s equations
obtained by Zenneck in 1907 [11]. It is a plane wave that
1270 ELECTROMAGNETIC SURFACE WAVES
travels along an interface between two dielectric media.
Consider the air–dielectric interface with the established
rectangular coordinates shown in Fig. 1. The properties
of the dielectric medium (medium 1) is characterized
by conductivity s
1
, permeability m
0
, and permittivity
e
1
. The vertically polarised Zenneck wave is also a trans
verse magnetic (TM) plane wave with zero magnetic
ﬁeld in the propagation direction, z. The ﬁeld components
of the Zenneck wave above the surface (xZ0) in air
(medium 2) with permeability m
0
and permittivity e
0
, are
given by
H
y2
¼Ae
Àu
2
x
e
Àgz
E
z2
¼ ÀA
u
2
joe
0
e
Àu
2
x
e
Àgz
E
x2
¼A
g
joe
0
e
Àu
2
x
e
Àgz
ð1Þ
where A is a constant, and u
2
and g are the propaga
tion constants in x and z directions, respectively. The
electromagnetic ﬁeld components below the surface
(xo0) are
H
y1
¼Ae
u
1
x
e
Àgz
E
z1
¼A
u
1
s
1
þjoe
1
e
u
1
x
e
Àgz
E
x1
¼A
g
s
1
þjoe
1
e
u
1
x
e
Àgz
ð2Þ
where u
1
is the propagation constant along the x direction
in the dielectric medium. The propagation constants sat
isfy the following equations
g
2
þu
2
2
¼ Ào
2
m
0
e
0
g
2
þu
2
1
¼jom
0
ðs
1
þjoe
1
Þ
u
1
ðs
1
þjoe
1
Þ
¼ À
u
2
joe
0
ð3Þ
where o is the angular frequency of the electromagnetic
wave. Solving these equations gives
u
1
¼a
1
þjb
1
¼
jom
0
ðs
1
þjoe
1
Þ
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
jom
0
ðs
1
þjoe
1
Þ Ào
2
m
0
e
0
_
u
2
¼a
2
Àjb
2
¼
o
2
m
0
e
0
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
jom
0
ðs
1
þjoe
1
Þ Ào
2
m
0
e
0
_
g ¼a þjb¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
Àjom
0
ðs
1
þjoe
1
Þo
2
m
0
e
0
jom
0
ðs
1
þjoe
1
Þ Ào
2
m
0
e
0
¸
ð4Þ
The Zenneck wave propagates along the surface and in the
direction normal to the surface with two different propa
gation constants. The wave attenuates in both directions,
particularly in the direction normal to the surface. For
example, the attenuation constants a
2
, a
1
, and a of a sur
face wave at 300MHz propagating over a dry ground of
s
1
¼4 and e
1
¼4e
0
are 0.167, 1.008, and 0.083 Np/m
[napers (napiers) per meter], respectively. The equalam
plitude contour of the H
y
component above and below the
dry ground surface with A¼1 is shown in Fig. 2 over the
range À5 moxo5m and 0ozo20 m. The ﬁeld attenuates
more rapidly inside the ground. The surface wave above
the surface has less attenuation. The phase velocity of the
Zenneck wave is greater than the speed of light. Hence,
the Zenneck surface wave is a fast wave.
On the surface, the ratio of the tangential electric ﬁeld
and magnetic ﬁeld gives the surface impedance [1,2]:
Z
s
¼R
s
þjX
s
¼
E
z2
ðx ¼0Þ
H
y2
ðx ¼0Þ
ð5Þ
Y
Z
X
0
H
y2
E
x2
E
z2
H
y1
E
x1
E
z1
Air (j
0
c
0
)
Medium 1
Medium 2 lossy ground (o
1
j
0
c
1
)
Figure 1. Zenneck wave on a planar surface separating two
dielectric media.
0 5 10 15 20
15
10
5
0
−5
0.728
0.818
0.546
0.637
0.455
0.455
0.364
0.274
0.274
0.274
0.274
0.183
0.183
0.183
0.183
0.092
0.092
0.092
0.092
Hy_amp
Figure 2. Contour plot of the H
y
 component of the Zenneck
wave above and below a dry ground of s
1
¼4 and e
1
¼4e
0
with
A¼1 in range À5 moxo5m and 0ozo20 m.
ELECTROMAGNETIC SURFACE WAVES 1271
The surface impedance and u
2
are related by
Z
s
¼R
s
þjX
s
¼ À
u
2
joe
0
¼
1
oe
0
ðb
2
þja
2
Þ ð6Þ
Hence, the attenuation constant of the surface wave along
the x direction above the surface is determined by the
surface reactance, and the phase constant by the surface
resistance as
a
2
¼oe
0
X
s
; b
2
¼oe
0
R
s
ð7Þ
In order to support a Zenneck surface wave, the dielectric
medium has to be lossy so that a
2
40, which is a feature of
a Zenneck wave.
2.2. Surface Wave on a ConductorBacked Dielectric Slab
A conductorbacked dielectric slab structure, as shown in
Fig. 3, is a typical example of multilayered structures.
This structure is commonly used in microwave printed
circuits. For simplicity, the conductor is treated as a ‘‘per
fect’’ conductor, and the dielectric of thickness d is as
sumed to be lossless with a real permittivity e
1
, or a
relative dielectric constant e
r
. With these assumptions, a
surface wave can propagate along the surface with
g ¼jb; u
1
¼jk
x
; u
2
¼h ð8Þ
For a TM surface wave, the solutions of the ﬁeld compo
nents above the dielectric surface (xZd) are [12,13]
H
y2
¼
joe
0
h
A sinðk
x
dÞe
ÀhðxÀdÞ
e
Àjbz
E
x2
¼ À
jb
h
A sinðk
x
dÞ e
ÀhðxÀdÞ
e
Àjbz
E
z2
¼A sinðk
x
dÞe
ÀhðxÀdÞ
e
Àjbz
:
ð9Þ
where A is a constant, and the ﬁelds in the dielectric
slab are
H
y1
¼ À
joe
1
k
x
A cosðk
x
xÞe
Àjbz
E
x1
¼ À
jb
k
x
A cosðk
x
xÞe
Àjbz
E
z1
¼A sinðk
x
xÞe
Àjbz
ð10Þ
The propagation constants satisfy the following equations
Àh
2
þb
2
¼k
2
0
¼o
2
m
0
e
0
k
2
x
þb
2
¼e
r
k
2
0
ð11Þ
where k
0
is the wavenumber in air and k
x
and h are re
lated by the transcendental equation
k
x
tanðk
x
dÞ ¼e
r
h ð12Þ
with
k
2
x
þh
2
¼ðe
r
À1Þk
2
0
ð13Þ
There exists a sequence of solutions of k
x
and h, which give
rise to a set of TM
n
modes. The nth mode, where
n¼0,1,2y, has a cutoff frequency of
f
c;TM
n
¼
nc
2d
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
e
r
À1
p ð14Þ
Similarly, for a transverse electric (TE) surface wave,
the ﬁeld components above the dielectric surface (xZd)
are [12]
E
y2
¼
jom
0
h
A cosðk
x
dÞe
ÀhðxÀdÞ
e
Àjbz
H
x2
¼ À
jb
h
A cosðk
x
dÞe
ÀhðxÀdÞ
e
Àjbz
H
z2
¼A cosðk
x
dÞe
ÀhðxÀdÞ
e
Àjbz
ð15Þ
where A is a constant, and the ﬁelds in the dielectric slab
are
E
y1
¼ À
jom
0
k
x
A sinðk
x
xÞe
Àjbz
H
x1
¼
jb
k
x
A sinðk
x
xÞe
Àjbz
H
z1
¼A cosðk
x
xÞe
Àjbz
ð16Þ
The propagation constants also satisfy Eq. (11), but k
x
and
h are related by the transcendental equation
k
x
cotðk
x
dÞ ¼ Àh ð17Þ
with
k
2
x
þh
2
¼ðe
r
À1Þk
2
0
ð18Þ
There exists, again, a sequence of solutions of k
x
and h,
which give rise to a set of TE
n
modes. The nth mode, where
n¼1, 2y, has a cutoff frequency of
f
c;TE
n
¼
ð2n À1Þc
4d
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
e
r
À1
p ð19Þ
Y
Z
X
Air (j
0
c
0
)
Good conductor (o
M
)
Slab (j
0
c
1
) d
0
Figure 3. Surface wave on a conductorbacked dielectric slab.
1272 ELECTROMAGNETIC SURFACE WAVES
Hence, the surface wave may propagate above the dielec
tric slab in a sequence of TM
0
,TE
1
,TM
1
,TE
1
,y modes if
the frequency of the wave is greater than their cutoff fre
quencies. The phase velocities of these modes are less than
the speed of light. Hence, the surface wave on a conductor
backed dielectric slab is a slow wave.
For example, a surface wave of 10 GHz on a microwave
printed circuit board of e
r
¼4.4 and d¼0.6mm can prop
agate in TM
0
mode as the cutoff frequency of the next
mode is 67.8 GHz. The k
x
and h values of the mode are
385.6 rad/m and 20.6 Np/m, respectively. The surface wave
attenuates slowly along the x axis at a rate of 0.0124 Np
per slab thickness or 0.6 mm. The electromagnetic ﬁelds
are not conﬁned within the dielectric slab, as shown in
Fig. 4 for the plot of the equalamplitude contour of the H
y
component with A¼1000 over the range 0oxo20d and
0ozo20d. However, at 50 GHz, the ﬁelds are more con
ﬁned within the dielectric slab with k
x
¼1783 rad/m and h
¼740.5 Np/m or 0.444 Np per slab thickness. The equal
amplitude contour plot of the H
y
component with A¼1000
over the range 0oxo3d and 0ozo20d is shown in Fig. 5.
The ﬁelds of the surface wave attenuate more rapidly as
the frequency approaches the cutoff frequency of TE
1
mode.
2.3. Radial Cylindrical Surface Wave on a Flat Surface
The Zenneck wave in Section 2.1 is a plane surface wave
on a ﬂat surface with ﬁelds in x, y, and z directions. A
similar radial cylindrical surface wave can be set up on a
ﬂat surface with propagation and attenuation in r and x
directions, as shown in Fig. 6. The electromagnetic ﬁelds
above the surface (xZ0) are [1]
H
f2
¼Ae
Àu
2
x
H
ð2Þ
1
ðÀjgrÞ
E
r2
¼A
u
2
joe
0
e
Àu
2
x
H
ð2Þ
1
ðÀjgrÞ
E
x2
¼A
g
oe
0
e
Àu
2
x
H
ð2Þ
0
ðÀjgrÞ
ð20Þ
where A is a constant, and H
0
(2)
and H
1
(2)
are the
Hankel functions of the second kind and orders 0 and 1,
0 5 10 15 20
20
15
10
5
0
113.671 113.671 113.671
113.685 113.685 113.685
113.699 113.699 113.699
113.713 113.713 113.713
113.727 113.727 113.727
113.741 113.741 113.741
113.755 113.755 113.755
113.769 113.769 113.769
113.783 113.783 113.783
113.797 113.797 113.797
113.811 113.811 113.811
113.825 113.825 113.825
113.838 113.838 113.838
113.852 113.852 113.852
113.866 113.866 113.866
113.88 113.88 113.88
113.894 113.894 113.894
Hy_amp
Figure 4. Contour plot of the H
y
 component of the surface
wave at 10GHz above and inside the dielectric slab of e
r
¼4.4
and d¼0.6 mm with A¼1000 in the range 0o(x/d)o20 and
0o(z/d)o20.
0 5 10 15 20
3
2
1
0
2.373 2.373 2.373
2.279 2.279 2.279
2.184 2.184
2.09 2.09 2.09
1.996 1.996 1.996
1.902 1.902 1.902
1.807 1.807 1.807
1.713 1.713 1.713
1.619 1.619 1.619
1.524 1.524 1.524
1.43 1.43 1.43
1.336 1.336 1.336
1.242 1.242
1.147 1.147 1.147
1.053 1.053 1.053
0.959 0.959 0.959
0.864 0.864 0.864
0.77 0.77 0.77
0.676 0.676 0.676
0.581 0.581 0.581
Hy_amp
Figure 5. Contour plot of the H
y
 component of the surface
wave at 50GHz above and inside the dielectric slab of e
r
¼4.4
and d¼0.6 mm with A¼1000 in the range 0o
(x/d)o3 and 0o(z/d)o20.
r
X
0
E
x2
H
02
E
r2
E
x1
H
01
E
r1
(j
0
c
0
) Air
Lossy ground (o
1
j
0
c
1
)
Figure 6. Planar surface with radial cylindrical surface wave.
ELECTROMAGNETIC SURFACE WAVES 1273
respectively. The ﬁelds in the dielectric medium are
H
f1
¼Ae
u
1
x
H
ð2Þ
1
ðÀjgrÞ
E
r1
¼ ÀA
u
1
s
1
þjoe
1
e
u
1
x
H
ð2Þ
1
ðÀjgrÞ
E
x1
¼A
jg
s
1
þjoe
1
e
u
1
x
H
ð2Þ
0
ðÀjgrÞ
ð21Þ
where the propagation constants are determined in the
same way as the Zenneck wave. For large r, we obtain
H
ð2Þ
n
ðÀjgrÞ )
e
Àgr
ﬃﬃﬃ
r
p ð22Þ
Hence, the radial surface wave is a cylindrical wave with
electric and magnetic ﬁelds proportional to r
À(1/2)
at large
distances. The radial cylindrical surface wave is also a fast
wave with phase velocity greater than the speed of light.
2.4. Axial Cylindrical Surface Wave
The axial cylindrical surface wave is also called the So
mmerfeld–Goubau surface wave [1]. For a cylindrical di
electric structure, as shown in Fig. 7, with the radius a,
conductivity s
1
, permeability m
0
and permittivity e
1
, an
axial surface wave may propagate along the air–dielectric
interface in the z direction. The surfacewave ﬁelds of a
TM wave outside the surface (rZa) are [1,7]
H
y2
¼A
oe
0
u
2
e
Àgz
H
ð1Þ
1
ðju
2
rÞ
E
z2
¼Ae
Àgz
H
ð1Þ
0
ðju
2
rÞ
E
r2
¼A
g
ju
2
e
Àgz
H
ð1Þ
1
ðju
2
rÞ
ð23Þ
where A is a constant and H
0
(1)
and H
1
(1)
are the Hankel
functions of the ﬁrst kind and orders 0 and 1, respectively.
The ﬁelds in the dielectric cylinder are
H
y1
¼ ÀA
s
1
þjoe
1
ju
1
e
Àgz
J
1
ðju
1
rÞ
E
z1
¼Ae
Àgz
J
0
ðju
1
rÞ
E
r1
¼A
g
ju
1
e
Àgz
J
1
ðju
1
rÞ
ð24Þ
where J
0
and J
1
are Bessel functions of the ﬁrst kind, or
ders 0 and 1, respectively, and the propagation constants
are determined in the same way as the Zenneck wave. The
surface impedance of the axial cylindrical surface wave is
Z
s
¼
u
2
oe
0
H
ð1Þ
0
ðju
2
aÞ
H
ð1Þ
1
ðju
2
aÞ
ð25Þ
For a large cylinder in which u
2
ab1, the surface imped
ance can be approximated as
Z
s
%
u
2
joe
0
ð26Þ
which is the same as that of the Zenneck wave for a ﬂat
surface. Hence, for large cylinders, the axial cylindrical
surface wave has propagation characteristics similar to
those of the Zenneck wave on a ﬂat surface. However, for
small cylinders, the axial cylindrical surface wave may
have characteristics different from those of the Zenneck
wave [1].
2.5. Surface Wave on a Dielectric Coated
Conducting Cylinder
Like the conductorbacked dielectric slab structure de
scribed in Section 2.2, a dielectriccoated conducting cyl
inder or wire as shown in Fig. 8 can support a surface
wave [13]. The conductor of radius a is treated as a perfect
conductor. The dielectric layer of thickness d is assumed to
be lossless with a real permittivity e
1
, or a relative dielec
tric constant e
r
. The propagation constants, g in the
H
θ1
E
z1
E
r1
H
θ2
E
z2
E
r2
Air (j
0
c
0
)
Lossy cylinder (o
1
j
0
c
1
)
0
a
Z
0
r
Figure 7. Cylindrical dielectric structure supporting axial cylin
drical surface wave.
H
θ1
E
z1
E
r1
H
θ2
E
z2
E
r2
Air (j
0
c
0
)
Good conductor
0
a
Z
0
r
b
Dielectric coating
(j
0
c
1
)
Figure 8. Dielectriccoated conducting cylinder.
1274 ELECTROMAGNETIC SURFACE WAVES
z direction and u
1
and u
2
in the r direction in the dielectric
layer and air medium, can be expressed as
g ¼jb; u
1
¼jk
r
; u
2
¼h ð27Þ
For a TM surface wave, the ﬁeld components outside the
dielectric surface [rZ(aþd)] are
E
z2
¼
A
K
0
ðhbÞ
J
0
ðk
r
bÞ À
J
0
ðk
r
aÞ
Y
0
ðk
r
aÞ
Y
0
ðk
r
bÞ
_ _
K
0
ðhrÞe
Àjbz
E
r2
¼
jbA
hK
0
ðhbÞ
J
0
ðk
r
bÞ À
J
0
ðk
r
aÞ
Y
0
ðk
r
aÞ
Y
0
ðk
r
bÞ
_ _
K
0
0
ðhrÞe
Àjbz
H
f2
¼
k
0
bZ
0
E
r2 ð28Þ
where A is a constant; b¼aþd, Y
0
and Y
1
are the Bessel
functions of the second kind, orders 0 and 1, respectively;
and K
0
is the modiﬁed Bessel function of the second kind
and order 0. The ﬁeld components in the dielectric coating
layer are
E
z1
¼A J
0
ðk
r
rÞ À
J
0
ðk
r
aÞ
Y
0
ðk
r
aÞ
Y
0
ðk
r
rÞ
_ _
e
Àjbz
E
r1
¼ À
jb
k
r
A J
0
0
ðk
r
rÞ À
J
0
ðk
r
aÞ
Y
0
ðk
r
aÞ
Y
0
0
ðk
r
rÞ
_ _
e
Àjbz
H
f1
¼
e
r
k
0
bZ
0
E
r1
ð29Þ
where A is a constant, and
Àh
2
þb
2
¼k
2
0
¼o
2
m
0
e
0
k
2
r
þb
2
¼e
r
k
2
0
ð30Þ
The parameters k
x
and h are the solutions of the tran
scendental equation
K
0
0
ðhbÞ
K
0
ðhbÞ
¼ À
e
r
h
k
r
J
0
0
ðk
r
bÞY
0
ðk
r
aÞ ÀY
0
0
ðk
r
bÞJ
0
ðk
r
aÞ
J
0
ðk
r
bÞY
0
ðk
r
aÞ ÀY
0
ðk
r
bÞJ
0
ðk
r
aÞ
ð31Þ
with
k
2
r
þh
2
¼ðe
r
À1Þk
2
0
ð32Þ
There exists a sequence of solutions of k
x
and h, corre
sponding to TM
n
modes where n¼0,1,2y. The cutoff fre
quency of the TM
0
mode is 0.
Similarly, for a TE surface wave, the ﬁeld components
outside the dielectric surface (xZd) are
H
z2
¼
A
K
0
ðhbÞ
J
0
ðk
r
bÞ À
J
0
0
ðk
r
aÞ
Y
0
0
ðk
r
aÞ
Y
0
ðk
r
bÞ
_ _
K
0
ðhrÞe
Àjbz
H
r2
¼
jbA
hK
0
ðhbÞ
J
0
ðk
r
bÞ À
J
0
0
ðk
r
aÞ
Y
0
0
ðk
r
aÞ
Y
0
ðk
r
bÞ
_ _
K
0
0
ðhrÞe
Àjbz
E
f2
¼ À
k
0
Z
0
b
H
r2
ð33Þ
and those inside the dielectric layer are
H
z1
¼A J
0
ðk
r
rÞ À
J
0
0
ðk
r
aÞ
Y
0
0
ðk
r
aÞ
Y
0
ðk
r
rÞ
_ _
e
Àjbz
H
r1
¼ À
jb
k
r
A J
0
0
ðk
r
rÞ À
J
0
0
ðk
r
aÞ
Y
0
0
ðk
r
aÞ
Y
0
0
ðk
r
rÞ
_ _
e
Àjbz
E
f1
¼ À
e
r
k
0
Z
0
b
H
r1
ð34Þ
The parameters k
x
and h of the TE wave are the solutions
of the following transcendental equation
K
0
0
ðhbÞ
K
0
ðhbÞ
¼ À
e
r
h
k
r
J
0
0
ðk
r
bÞY
0
0
ðk
r
aÞ ÀY
0
0
ðk
r
bÞJ
0
0
ðk
r
aÞ
J
0
ðk
r
bÞY
0
0
ðk
r
aÞ ÀY
0
ðk
r
bÞJ
0
0
ðk
r
aÞ
ð35Þ
There exists again a sequence of solutions of k
x
and h,
which give rise to a set of TE
n
modes, such as that in a
conductorbacked dielectric slab. As an approximation, the
cutoff frequencies of the TE
n
and TM
n
modes can be esti
mated using the same formulas as those for the conductor
backed dielectric slabs.
For example, a surface wave of 50 GHz on a dielectric
coated cylinder with d¼0.6 mm, e
r
¼4.4, and a¼10 mm
can propagate in TM
0
mode with k
r
¼1762 rad/m and
h¼789.4 Np/m or 0.473 Np per dielectric thickness. The
equalamplitude contour plot of the H
f
component with A
¼1000 over the range 0oro3d and 0ozo20d is shown in
Fig. 9. The ﬁelds of the surface wave attenuate rapidly
with the increase of radial distance r at this frequency.
The plot is very similar to that for a conductorbacked
dielectric slab shown in Fig. 5.
2.6. Norton and Trapped Surface Waves on Flat
Ground Interface
When a short antenna radiates at a point above the
ground surface of Earth, the general solution of electro
magnetic ﬁelds in air consists of both ‘‘space wave’’ and
‘‘surface wave’’ [7–10]. The properties of the ground can be
characterized either by conductivity s
1
, permeability m
0
,
and permittivity e
1
for a homogenous ground, or by the
surface impedance Z
s
, or the normalized surface imped
ance D
s
¼Z
s
/Z
0
¼constant D
0
, where Z
0
¼120p for loaded
ground, corrugated ground [2], and multilayered struc
tures [7,9]. The surface impedance can be obtained by
ELECTROMAGNETIC SURFACE WAVES 1275
considering a planewave incidence onto the ground sur
face.
The historical development of the solution to such a
radiation problem has not been straightforward since the
work of Sommerfeld in 1909 [1,5,7,14–17]. But the subject
is now well understood [10,17], and the electromagnetic
wave propagation over the ground surface can be accu
rately predicted [18]. The ‘‘surface wave’’ in the solution
dominates when the antenna is on or close to the ground
surface. For simplicity, the short antenna is considered to
be on the ground surface as shown in Fig. 10. In this case,
the ‘‘space wave’’ component vanishes. The short antenna
is further considered to be vertically orientated so that it
produces a TM wave with a vertical electric ﬁeld compo
nent [7–10]
E
z
¼A
e
Àjk
0
R
0
R
0
FðwÞ ð36Þ
where A is a constant, k
0
is the wavenumber in free space,
R
0
is the distance from the antenna to the receiving point
B at height z, and F(w) is the Sommerfeld attenuation
function given by
FðwÞ ¼1 ÀjðpwÞ
1=2
e
Àw
erfcðjw
1=2
Þ ð37Þ
with the numerical distance
w¼ À
jk
0
R
0
2
D
2
s
1 þ
z
D
s
R
0
_ _
2
¼jwje
jf
w
ð38Þ
and the complementary error function,
erfcðjw
1=2
Þ ¼
2
ﬃﬃﬃ
p
p
_
1
jw
1=2
e
Àu
2
du ð39Þ
The modulus of the numerical distance w is proportion
al to the physical distance R
0
and Z
s

2
. Its argument is
related to the argument of Z
s
as listed below for different
types of ground [8]:
For the range 0rRe(w)r2.4 and À1rIm(w)r3, the
equalamplitude contour plots of F(w) in dB and arg(F(w))
in degrees are shown in Fig. 11. Both F(w) and
arg(F(w)) have smooth variations when f
w
o0. However,
when f
w
o0, several dips appear in the vicinity of those
w values that make F(w) small or zero. The amplitudes
of F(w) in dB are again shown in Fig. 12 for
10
À2
rwr10
3
and À901rf
w
r801. The variation of
F(w) is now examined in terms of series expansion for
large values of w.
For À2prf
w
r0, or highly capacitive, capacitive, and
inductive ground surfaces, the Sommerfeld attenuation
function for large w can be expanded to
FðwÞ ¼ À
1
2w
À
1
.
3
ð2wÞ
2
À
1
.
3
.
5
ð2wÞ
3
À
1
.
3
.
5
.
7
ð2wÞ
4
ÀÁ Á Á
¼F
Norton
ðwÞ
ð40Þ
The wave associated with the attenuation function given
in this series [F
Norton
(w)] is referred to as the Norton sur
face wave [7–10]. The Norton surface wave or its ﬁeld
0 5 10 15 20
3
2
1
0
0.506 0.506 0.506
0.485 0.485 0.485
0.464 0.464 0.464
0.443 0.443 0.443
0.422 0.422 0.422
0.401 0.401 0.401
0.38 0.38 0.38
0.36 0.36 0.36
0.339 0.339 0.339
0.318 0.318 0.318
0.297 0.297 0.297
0.276 0.276 0.276
0.255 0.255 0.255
0.234 0.234 0.234
0.213 0.213 0.213
0.192 0.192 0.192
0.171 0.171 0.171
0.15 0.15 0.15
0.129 0.129 0.129
0.109 0.109 0.109
H[ amp
Figure 9. Contour plot of the H
f
 component of the surface
wave at 50GHz above and inside the dielectric coating layer of
e
r
¼4.4 and d¼0.6 mm on a conducting cylinder of radius 10mm
with A¼1000 in the range 0o(r/d)o3 and 0o(z/d)o20.
Y
Z
X
0
Air (j
0
c
0
)
Homogeneous ground
(o
1
j
0
c
1
)
Vertical
antenna
B
d
Surface wave
Z
R
0
Figure 10. Short antenna radiation on a homogeneous ﬂat
ground surface.
Highly Capacitive Capacitive Inductive (Normal Ground) Highly Inductive
Àp/2rarg(Z
s
)oÀp/4
À3p/2rf
w
oÀp
Àp/4rarg(Z
s
)o0
Àprf
w
oÀp/2
0rarg(Z
s
)op/4
Àp/2rf
w
o0
p/4rarg(Z
s
)rp/2
0rf
w
rp/2
1276 ELECTROMAGNETIC SURFACE WAVES
components decay as the inverse square of the distance.
However, for 0rf
w
rp/2 or a highly inductive surface,
F(w) has a different expansion for large w
FðwÞ ¼F
trapped
ðwÞ þF
Norton
ðwÞ ð41Þ
where
F
trapped
ðwÞ ¼ À2jðpwÞ
1=2
e
Àw
ð42Þ
An additional term [F
trapped
(w)] has appeared in the se
ries. This term dominates at short distances, but it atten
uates exponentially as the distance increases. F
trapped
ðwÞ
has a maximum modulus value at w¼0.5cos(f
w
). The
wave associated with this additional term of attenuation
function is referred to as the trapped surface wave [7–10].
The dips in Fig. 11 are therefore the result of interference
between the trapped and Norton surface waves. Unlike
the Norton surface wave, the trapped wave or its ﬁeld
components decay inversely with the product of the
square root of the distance and the exponential of this
distance. A highly inductive surface impedance can be ob
tained by loading a metal surface with a thin dielectric
layer or by corrugating a metal surface [2]. Such surfaces
can therefore support a trapped surface wave.
For a normal dry ground with s
1
¼0.01 and e
r
¼4, the
predicted electric ﬁeld E
z
of the surface wave at 10 MHz on
the ground surface with transmitted power of 1 kW and
antenna gain G
t
¼1 using the software package in de
scribed in Ref. 18 is shown in Fig. 13 over a horizontal
distance of 1–60km. The amplitude of the electric ﬁeld
attenuates monotonically against the distance for the ho
mogeneous (inductive) ground. The surface wave is thus a
Norton surface wave. For comparison, a prediction of the
electric ﬁeld E
z
of the surface wave over a highly inductive
ground with D
0
¼0.01þj0.09 at the same frequency with
the same transmitted power and antenna gain is shown in
Fig. 14. It can be seen that E
z
 does not decrease mono
tonically, but varies against distance as a result of the in
terference of Norton and trapped surface waves.
Comparison between Figs. 13 and 14 shows that the sur
face wave on the ground surface is much stronger when
the surface impedance is highly inductive.
The surface wave on the ground surface has a height
gain factor when the receiving point is considered to move
from the ground surface to a height z. The height gain
Figure 11. F(w) and arg(F(w)) contour plots for 0rRe(w)r2.4
and À1rIm(w)r3. (This ﬁgure is available in full color at http://
www.mrw.interscience.wiley.com/erfme.)
20
10
−10
−20
M
a
g
n
i
t
u
d
e
o
f
F
(
w
)
i
n
d
B
−30
−40
−50
−60
−3 −2 −1 0 1 2 3
0
LOG 10
Amplitude of the numerical distance w
[
w
= −90°
[
w
: argument of w
[
w
= 80°
70°
Figure 12. Amplitudes of F(w) in dB for 10
À2
rwr10
3
and
À901rf
w
r801.
10
Scale/
Ref.
−50
Start 0 Stop 60000
Figure 13. Prediction of surfacewave propagation at 10MHz
over a dry ﬂat ground surface with s
1
¼0.01 and e
r
¼4: E
z
 in dB
versus horizontal distance.
ELECTROMAGNETIC SURFACE WAVES 1277
factor is given by
G¼1þjk
0
D
0
z ð43Þ
For a highly capacitive or capacitive ground, the electric
ﬁeld tends to increase when the receiving point is raised.
On the other hand, for a normal homogeneous ground or a
highly inductive ground, the electric ﬁeld tends to
decrease as the receiving height is increased.
2.7. Surface Wave Propagation over an Inhomogeneous
Flat Surface
If the ﬂat ground surface is inhomogeneous along the path
of propagation as shown in Fig. 15, the normalised surface
impedance D
s
will not be a constant, but a function of po
sition. For the TM wave considered in Section 2.6 with the
antenna on the ground surface, the Sommerfeld attenua
tion function in Eq. (36) has to be replaced by an attenu
ation factor F(d), where d is the distance to the antenna,
which takes into account the effect of boundary disconti
nuities along the propagation path. The vertical electric
ﬁeld on the inhomogeneous ground surface then becomes
[7,9,10]
E
z
¼A
e
Àjk
0
R
0
R
0
FðdÞ ð44Þ
The attenuation factor is given by the integral equation
FðdÞ ¼1 À
jk
0
d
2p
_ _
1=2
_
d
0
D
s
ðyÞFðyÞ
dy
½yðd ÀyÞ
1=2
ð45Þ
where y is the distance from a point on the propagation
path to the antenna. The integral equation can be solved
numerically [10]. F(d) can also be expressed in other forms
of the integral equation, or in closed forms for special cases
[10]. The propagation of the surface wave along the inho
mogeneous path can thus be predicted.
Figure 16 shows an example prediction of the surface
wave propagation along a land–sea–land–sea–land mixed
path over a range of 30 km using the software package in
described in Ref. 18. The surface wave has a frequency of
10 MHz. Each mixedpath section has a length of 6km.
The electrical properties of the land are s¼0.01 and e
r
¼4,
and those of the sea are s ¼4 and e
r
¼80. The wellknown
‘‘sea gain’’ phenomenon in groundwave propagation can
be seen from the prediction.
2.8. SurfaceWave Propagation over a Spherical Ground
If the curvature of Earth’s surface is taken into consider
ation, the surface wave guided along the curvature of
Earth as shown in Fig. 17 can also be modeled [7,10,19–
26]. For the TM wave considered in Section 2.6 with the
antenna and the receiving points both on the ground sur
face, the electric ﬁeld normal to the surface of the spher
ical Earth model with radius a at angle y away from the
antenna at y ¼0 can be expressed as
E
r
ðyÞ ¼A
e
Àjk
0
ay
ay
FðyÞ ð46Þ
where F(y) is the attenuation factor given by the residue
series [7,10]
FðyÞ ¼e
Àjðp=4Þ
ðpx
y
Þ
1=2
1
s ¼1
e
Àjxts
t
s
ð47Þ
Scale/
Ref.
Start 0 Stop 60000
10
−50
Figure 14. Prediction of surfacewave propagation at 10MHz
over a highly inductive ground surface with D
0
¼0.01þj0.09:
E
z
 in dB versus horizontal distance.
Y
Z
X
0
Air
Inhomogeneous ground
Vertical
antenna
B
d
1
Surface wave
d
2
d
3
d
Figure 15. Short antenna radiation on an inhomogeneous ﬂat
ground surface.
Scale /
Ref.
Start 0 Stop 30000
10
−40
Figure 16. Prediction of surfacewave propagation at 10MHz
over a land–sea–land–sea–land mixed path: E
z
 in dB versus
horizontal distance.
1278 ELECTROMAGNETIC SURFACE WAVES
In the equation above, the electrical distance x
y
is deﬁned
as
x
y
¼
k
0
a
2
_ _
1=3
y ð48Þ
and t
s
(s ¼1,2,y) are given by
t
s
¼jt
s
je
Àjp=3
ð49Þ
with
jt
1
j ¼1:01879; jt
2
j ¼3:24820; jt
3
j ¼4:82010;
jt
4
j ¼6:16331; jt
5
j ¼7:37218
ð50Þ
and
jt
s
j ¼½1:5ðs À0:75Þp
2=3
ð51Þ
for s45. The correction due to the effect of Earth’s curva
ture becomes necessary if the distance from the receiving
point to the antenna is greater than 5 Â10
3
(l
0
)
1/3
or
10
4
(l
0
)
1/3
meters [24].
Figure 18 shows an example prediction of the surface
wave propagation along the surface of a spherical Earth
with s¼0.01 and e
r
¼4 at 10MHz over a range of
1–500 km. The electric ﬁeld normal to the surface atten
uates at 26 dB per 100km, as the surface wave is guided
along the curvature of the spherical Earth, compared with
6 dB per 100km for a ﬂat Earth model shown in Figs. 18
and 19. Earth’s curvature therefore causes the surface
wave to attenuate more quickly as it propagates.
3. DISCUSSIONS
In this article, the ‘‘surface waves’’ guided along an inter
face between two media, or an interface between air and a
multilayered structure in planar, cylindrical, and spheri
cal geometries have been introduced. Several examples
have been used to illustrate the propagation of surface
waves in these geometric structures. In some cases,
only TM surface waves are considered, but TE waves
can be treated in a similar way. For propagation over a
ﬂat or spherical Earth, TE Norton surface waves tend to
attenuate much more rapidly than do TM waves. Details
of the formulation of both TM and TE waves over ﬂat,
spherical, homogeneous, and inhomogeneous ground
can be found in Ref. 10, and the prediction of such prop
agation can be made using the software package described
in Ref. 18.
The propagation of radio surface waves along Earth’s
surface can be utilized for longdistance wireless commu
nications and remote sensing. On the other hand, the sur
face waves on microwave printed circuits and microstrip
antenna systems should be avoided to minimize the elec
tromagnetic coupling between circuit elements, or power
loss in the systems. The understanding and control of sur
face waves are thus important in radio systems.
Z
a
B(r, 0, [)
A
0
Spherical Earth
Antenna
Figure 17. Short antenna radiation on a homogeneous spherical
ground surface.
Scale/
Ref.
Start 0 Stop 500000
30
−90
Figure 18. Prediction of surfacewave propagation at 10MHz
over a dry spherical ground surface with s
1
¼0.01 and e
r
¼4: E
z

in dB versus arc distance.
Scale/
30
Ref.
Start 0 Stop 500000
−90
Figure 19. Prediction of surfacewave propagation at 10 MHz;
the same path as Fig. 18, but without curvature correction.
ELECTROMAGNETIC SURFACE WAVES 1279
BIBLIOGRAPHY
1. H. M. Barlow and A. L. Cullen, Surface waves, Proc. IEE
100(Part III):329–341 (1953).
2. H. M. Barlow and J. Brown, Radio Surface Waves, Oxford
Univ. Press, Oxford, 1962.
3. K. A. Norton, The physical reality of space and surface waves
in the radiation ﬁeld of radio antennas, Proc. IRE 25:
1192–1202 (1937).
4. K. A. Norton, The propagation of radio waves over the surface
of the earth and in the upper atmosphere, Proc. IRE 25:
1203–1236 (1937).
5. W. H. Wise, The physical reality of Zenneck’s surface wave,
Bell Syst. Tech. J. 16:35–44 (1937).
6. H. Bremmer, The surfacewave concept in connection with
propagation trajectories associated with the Sommerfeld
problem, IRE Trans. Anten. Propag. S175–S182 (1959).
7. J. R. Wait, Electromagnetic surface waves, in J. A. Saxton, ed.,
Advances in Radio Research, Academic Press, New York,
1964, Vol. 1, pp. 157–217.
8. R. J. King, Electromagnetic wave propagation over a constant
impedance plane, Radio Sci. 4:255–268 (1969).
9. R. J. King and J. R. Wait, Electromagnetic ground wave prop
agation theory and experiment, Symp. Math. 18:107–208
(1976).
10. T. S. M. Maclean and Z. Wu, Radiowave Propagation over
Ground, Chapman & Hall, London, 1993.
11. J. Zenneck, Uber die Fortpﬂanzung ebener elektromagnetisc
her Wellen langs einer ebener Leiterﬂache und ihre Bezie
hung zur drahtlosen Telegraphie, Annal. Phys. 23:846–866
(1907).
12. D. M. Pozar, Microwave Engineering, AddisonWesley, New
York, 1990.
13. R. E. Colin, Foundations for Microwave Engineering,
McGrawHill, New York, 1966.
14. A. Sommerﬂed, Partial Differential Equations in Physics,
Academic Press, New York, 1949.
15. A. Sommerﬂed, Uber die Ausbeitung der Wellen in der
drahtlosen Telegraphie, Annal. Phys. 28:655–736 (1909).
16. A. Sommerﬂed, Uber die Ausbeitung der Wellen in der
drahtlosen Telegraphie, Annal. Phys. 81:1135–1153 (1926).
17. A. Banos, Dipole Radiation in the Presence of a Conducting
Halfspace, Pergamon Press, Oxford, 1966.
18. Z. Wu and T. S. M. Maclean, Radiowave Propagation over
Ground Software, Chapman & Hall, London, 1998.
19. G. N. Watson, The transmission of electric waves round the
Earth, Proc. Roy. Soc. Lond. A 95:546–563 (1919).
20. B. von der Pol and H. Bremmer, The diffraction of electro
magnetic waves from an electrical point source round a ﬁ
nitely conducting sphere, Phil. Mag. Ser. 7 24:141–176 (1937).
21. B. von der Pol and H. Bremmer, The diffraction of elec
tromagnetic waves from an electrical point source round a
ﬁnitely conducting sphere, Phil. Mag. Ser. 7 24:825–864
(1937).
22. B. von der Pol and H. Bremmer, The diffraction of electro
magnetic waves from an electrical point source round a ﬁ
nitely conducting sphere, Phil. Mag. Ser. 7 25:817–834 (1938).
23. B. von der Pol and H. Bremmer, The diffraction of electro
magnetic waves from an electrical point source round a ﬁ
nitely conducting sphere, Phil. Mag. Ser. 7 26:261–275 (1939).
24. H. Bremmer, Terrestrial Radio Waves, Elsevier, New York,
1949.
25. J. R. Wait, Recent analytical investigations of electromagnetic
ground wave propagation over inhomogeneous earth models,
Proc. IEEE 62:1061–1071 (1974).
26. V. A. Fock, Electromagnetic Diffraction and Propagation
Problems, Pergamon Press, Oxford, 1965.
ELECTROMAGNETIC WAVE PROPAGATION
FERNANDO L. TEIXEIRA
The Ohio State University
Columbus, Ohio
FERNANDO J. S. MOREIRA
ODILON M. C. PEREIRA
FILHO
Federal University of Minas
Gerais
Belo Horizonte, Brazil
1. INTRODUCTION
1.1. Historical Perspective
Electromagnetic waves play a major role in remote sen
sing and communication systems. The equations govern
ing electromagnetic wave propagation were ﬁrst
established by J. C. Maxwell (1831–1879) in 1864. How
ever, one should not overlook previous contributions of
many scientists over the centuries, especially in optics.
Speculations regarding the nature of light date since
ancient Greece, when philosophers were already ac
quainted with the rectilinear propagation, reﬂection, and
refraction of light [1]. The major contributions to the ﬁeld,
however, initiated during the Renaissance, on the founda
tions of the experimental method introduced by Galileo
(1564–1642). Some examples are the law of refraction
discovered by Snell (1580–1626) in 1621, the principle of
least time established by Fermat (1601–1665) in 1657, the
observation of light diffraction by Grimaldi (1618–1663)
and Hooke (1635–1703) circa 1665, and Huygen’s (1629–
1695) envelope construction (leading to the principle
named after him) in 1678 [1]. Such observations and
experiments lead to the development of a wave theory to
explain the nature of light as luminous sources vibrating
in adjacent portions of an ethereal medium, an interpreta
tion similar to that of acoustic waves. The wave theory of
light was, however, later rejected by Newton (1642–1727),
who proposed a corpuscular interpretation instead [1].
Many years passed until new experiments reinforced
the wave theory of light. Among the leading scientists
responsible for that is Fresnel (1788–1827), who developed
the theory of light reﬂection and refraction in a more
quantitative way and as well as a ﬁrmer mathematical
basis for the interpretation of light as a transverse wave.
Fresnel’s work temporarily obscured the corpuscular the
ory of light, which regained strength only after Planck
(1858–1947) and Einstein (1879–1955) started to unveil
the quantum aspects of light in the beginning of the 1900s.
The nineteenth century also witnessed the efforts of many
1280 ELECTROMAGNETIC WAVE PROPAGATION
physicists to experimentally determine the speed of light,
such as Michelson (1852–1931).
Apart from the developments in optics, many scientists
dedicated efforts to establish the physical nature of
magnetism and electricity. Among them, it is worth
mentioning Coulomb (1736–1806), who demonstrated
the inverse square law for electrical forces in 1785;
Oersted (1777–1851), who was perhaps the ﬁrst to observe
experimentally the effect of electrical currents on the
magnetic ﬁeld in 1820; and Ampe`re (1775–1836), who
formulated the circuit force law and postulated magnet
ism as an electrical phenomenon. The milestone experi
ment for the uniﬁcation of magnetism and electricity,
however, was carried out by Faraday (1791–1867), who
discovered electromagnetic induction in 1831. The
electromagnetic theory known at that time was put
on analytical ground by Maxwell (1831–1879) in 1855/56.
He further developed a model to explain electromagne
tics as a mechanical phenomenon, which lead into the
concept of displacement currents and the consequent
generalization of Ampe`re’s law in 1861. The mechanical
model was ﬁnally abandoned and Maxwell, in 1864,
published his third paper on the subject, establishing
the basis of the electromagnetic theory [2]. Besides
postulating the displacement current, Maxwell was the
ﬁrst to predict the propagation of electromagnetic waves
and to postulate light itself as an electromagnetic radia
tion. This provoked strong opposition from scientists at
that time, as no evidence of such waves had been observed
by experiments.
Maxwell did not survive to see his ideas been accepted
by the majority of the scientists, which came just after the
ﬁrst experimental observation of electromagnetic wave
propagation by Hertz, published in 1888. After that, the
use of electromagnetic waves for practical purposes was
just a matter of time, and in 1901 Marconi (1874–1937)
achieved the ﬁrst transmission of radio signals across the
Atlantic. Today, electromagnetic waves permeate most of
modern technologies.
In this article, we will brieﬂy discuss some fundamen
tal aspects of the electromagnetic wave propagation.
Further details on this vast subject can be found in several
books, such as Refs. 1 and 3–11.
1.2. The Electromagnetic Wave Spectrum
Electromagnetic waves can propagate at different fre
quencies. It is common to classify electromagnetic waves
according to their frequency range as 3–30Hz, ELF (extra
lowfrequency or extremelyLF) waves; 30–300 Hz, SLF
(superlowfrequency) waves; 300 Hz–3kHz, ULF (ultra
lowfrequency) waves; 3–30kHz, VLF (verylowfre
quency) waves; 30–300kHz, LF (lowfrequency or long)
waves; 300kHz–3 MHz, MF (mediumfrequency or med
ium) waves, which include most AM radiowaves; 3–
30 MHz, HF (highfrequency or short) waves, which in
clude most of shortwave radio; 30–300MHz, VHF waves,
which include FM radio and TV signals; 300MHz–3 GHz,
UHF waves, which include TV signals, radar waves at L
and S bands, and microwave oven radiation; 3–30GHz,
SHF (centimeter) waves, which include radars at C, X, Ku,
and K bands, satellite communication links, and aircraft
landing systems; and 30–300 GHz, EHF (millimeter)
waves, which include radars at Ka band.
Electromagnetic waves can also be classiﬁed by their
wavelength. For an electromagnetic wave propagating in
air or vacuum, the wavelength l and frequency f are
related by l ¼c/f, where cE3Â10
À8
m/s is the speed of
light in vacuum. Microwaves correspond to electromag
netic waves with l around 1 cm–1 m (300 MHz–30 GHz).
Millimeter and submillimeter waves have l around 1 mm–
1 cm (30–300 GHz) and just below it, respectively. Visible
light is a form of electromagnetic wave with l ¼0.38–
0.72 mm. Wavelengths just below visible light correspond
to ultraviolet waves, while wavelengths just above visible
light correspond to nearinfrared waves (0.72–1.3mm) and
thermal infrared waves (7–15 mm). At even smaller wave
lengths (higher frequencies) one encounters X and gamma
rays [10].
2. MAXWELL’S EQUATIONS AND THE WAVE EQUATION
The propagation of electromagnetic waves is governed by
Maxwell’s equations, which in SI units and for macro
scopic ﬁeld quantities are
rÂE¼ À
@B
@t
r
.
D¼q
rÂH¼Jþ
@D
@t
r
.
B¼0
ð1Þ
These four equations are supplemented by constitutive
relations that relate D and B to E and H. We shall
assume propagation in simple media, with vacuum as a
special case. The term simple media in this context
denotes linear, homogeneous, and isotropic materials,
which will be assumed lossless for the time being. The
constitutive relations in this case are expressed as (see
MAXWELL’S EQUATIONS article)
D¼eE and B¼mH ð2Þ
where e is the permittivity (e ¼e
0
E8.85418782 Â10
À12
F/m in vacuum) and m is the permeability (m ¼m
0
¼4p Â
10
À7
H/m in vacuum), constant scalar numbers for simple
media. The electric current J and charge q densities can
be interpreted as the sources of the electromagnetic ﬁeld
and are interrelated by the continuity equation:
r
.
J¼ À
@q
@t
ð3Þ
which is implicit in (1) (see MAXWELL’S EQUATIONS article).
From (1) one can also derive a law for the conservation of
electromagnetic energy, the Poynting theorem (see MAX
WELL’S EQUATIONS article). For our purposes, we just need to
emphasize the Poynting vector, deﬁned as
S¼EÂH ð4Þ
ELECTROMAGNETIC WAVE PROPAGATION 1281
This vector gives the magnitude and direction of power
ﬂow density (watts per square meter).
We will discuss the relations between sources ðJÞ and
ﬁelds (E and H) later. For now, we shall only investigate
simple characteristics of the electromagnetic propagation.
In regions of a simple media where no source is present
(i.e., J¼q¼0) the (homogeneous) wave equations below
can be derived from (1) (see MAXWELL’S EQUATIONS article):
r
2
E À
1
c
2
@
2
E
@t
2
¼0
r
2
HÀ
1
c
2
@
2
H
@t
2
¼0
ð5Þ
where c ¼1=
ﬃﬃﬃﬃﬃ
me
p
is the speed of light in the medium (c ¼
c
0
¼299,792,458 m/s in vacuum, by deﬁnition). From (5)
one can readily show that each Cartesian component of E
and H also satisﬁes the following homogeneous (scalar)
wave equation
r
2
c À
1
c
2
@
2
c
@t
2
¼0 ð6Þ
where c represents one of such components. This wave
equation governs the behavior of c with respect to both
position (x,y,z) and time (t). For simplicity, let us consider a
onedimensional problem (i.e., with no variation in x or y),
such that (6) simpliﬁes to
@
2
c
@z
2
À
1
c
2
@c
@t
¼0 ð7Þ
It can be easily veriﬁed that a solution to (7) is
c¼f ðz ÀctÞ þgðz þctÞ ð8Þ
where f and g are arbitrary real functions [12]. These are
called D’Alembert’s solutions, where f(z Àct) represents an
arbitrary waveform propagating in the positive z direction
with speed equal to c, while g(z þct) represents another
waveform with the same speed but propagating in the
negative z direction.
Although (8) represents the solution of an idealized
problem, it illustrates one fundamental property of the
solutions of (6); namely, for unbounded simple media they
represent a wave traveling with the speed of light. In real
life situations, obstacles (ground, vegetation, humanpro
duced constructions, etc.) are present and impose addi
tional constraints (boundary conditions) on the solutions
of (6). In general, the presence of obstacles complicates the
problem and simple analytical solutions can be found only
for simple geometries.
2.1. TimeHarmonic Regime
Most RF and microwave applications deal with time
invariant linear media [6]. In this case, the use of a
timeharmonic representation is often more convenient
for dealing with electromagnetic wave phenomena. As
suming a e
jot
time variation, where o is the angular
frequency, Maxwell’s equations (1) are rewritten as [6]
rÂE¼ ÀjoB r
.
D¼r
rÂH¼J þjoD r
.
B¼0
ð9Þ
where the functions (of space only) in (9) are now complex
phasor representations of the corresponding quantities in
(1), according to
E¼ReðEe
jot
Þ ð10Þ
Here Re denotes the real part. A similar relation holds for
the other ﬁelds (and sources). For simple media and from
(2), the constitutive relations are then
D¼eE and B¼mH ð11Þ
The wave equations (5) also assume a simpler represen
tation [6]:
r
2
Eþk
2
E¼0
r
2
Hþk
2
H¼0
ð12Þ
known as the (homogeneous) Helmholtz equations, where
k ¼o
ﬃﬃﬃﬃﬃ
me
p
is the wave number.
It is useful to deﬁne timeaverage powers when dealing
with timeharmonic ﬁelds. For instance, from (4) and the
deﬁnition in (10) one can deﬁne the complex Poynting
vector
S¼
1
2
EÂH
Ã
ð13Þ
The real part of the complex Poynting vector is the time
average of S [4]. Equation (13) stresses an important
characteristic of electromagnetic wave propagation. In
phase components of E and H contribute to ReðSÞ, repre
senting the power density propagating in the direction of
ReðSÞ [6]. On the other hand, components of E and H in
phase quadrature contribute to the imaginary part of S,
representing a (stationary) reactive power density.
In the timeharmonic regime, (7) becomes
d
2
C
dz
2
þk
2
C¼0 ð14Þ
where C is the complex representation of c according to
(10). The solution of (14) is
C¼p
1
e
jðkz þf
1
Þ
þp
2
e
Àjðkz þf
2
Þ
ð15Þ
Consequently, from (10) and observing that for simple
media k ¼o/c, we obtain
c¼p
1
cos½kðz þctÞ þf
1
þp
2
cos½kðz ÀctÞ þf
2
ð16Þ
which is a particular timeharmonic solution of (7) and (8)
known as a plane wave.
1282 ELECTROMAGNETIC WAVE PROPAGATION
3. PLANE, TEM, AND STATIONARY WAVES
The plane wave is the simplest solution of the homoge
neous wave equation. Many important properties of elec
tromagnetic wave propagation can be inferred from this
fundamental solution. Moreover, an arbitrary electromag
netic wave in sourceless linear media can be decomposed
in terms of plane waves (e
7jkz
in one dimension) in a
manner similar to the decomposition of a timedomain
signal in terms of its spectral components e
jot
[13].
The complex representation of a plane wave in (15)
speciﬁes an amplitude (p) and phase (kz þf). Moreover,
since C represents a Cartesian component of E or H, it
also indicates the orientation (i.e., the polarization) of the
vector ﬁeld.
The planewave solution of C in three dimensions is
written as e
Àjðk
x
x þk
y
y þk
z
zÞ
, where k
x,y,z
/k can be understood
as the cosine directors of the wavefront. By deﬁning the
propagation vector k¼k
x
^ xxþk
y
^ yy þk
z
^ zz and putting the
components of E and H back together, we arrive at a
more general complex representation for plane waves [6]
E¼E
0
e
Àjk
.
r
and H¼H
0
e
Àjk
.
r
ð17Þ
where r ¼x ^ xxþy ^ yy þz ^ zz denotes the obervation point, and
amplitudes and phases of each component are incorpo
rated in the constant complex vectors E
0
and H
0
(which
also determine the wave polarization). Substituting (17)
into (12), one obtains the characteristic equation or dis
persion relation (for simple media):
k
2
¼o
2
me ¼k
2
x
þk
2
y
þk
2
z
ð18Þ
The planewave relations for E and H in simple media are
obtained by substituting (11), (17), and (18) into (9) with
no sources present
kÂE¼omH k
.
E¼0
kÂH¼ ÀoeE k
.
H¼0
ð19Þ
from which one observes that for sourceless simple media
Ampe`re’s and Faraday’s laws are independent equations.
Actually, this is also true for any wave solution in such
media, which can be inferred from (9) and (11) [6].
Note that k
x,y,z
can be complex and still satisfy (18).
Consequently, it is useful to regard k as a complex vector
k¼b–ja ð20Þ
where a and b are real vectors [6]. Since both a and b are
real, they can be interpreted geometrically. Substituting
(20) into (17), one veriﬁes that the planewave spatial
variation in complex notation is of the form e
Àa
.
r
e
Àjb
.
r
,
representing the variations of the amplitude (ﬁrst term)
and phase (second one) of the wave as it propagates
through the medium. Consequently, the constant ampli
tude and phase surfaces are planes (hence the name plane
waves), with a and b pointing to their normal directions,
respectively. If such planes coincide (i.e., ajjb), then the
plane wave is denoted as a uniform plane wave. Other
wise, it is denoted as nonuniform. Furthermore,
^
bb (i.e.,
unit normal vector to the equiphase surface) denotes the
direction of propagation of the plane wave.
For uniform plane waves, one can show from (18) and
(20) that k¼k
^
kk, where
^
kk¼
^
bb is the direction of propaga
tion. Now k has a geometrically deﬁned direction and one
can inspect from (19) that Ampe`re’s and Faraday’s laws
reduce to
H¼
k
om
^
kkÂE¼
1
Z
^
kkÂE
E¼
Àk
oe
^
kkÂH¼ ÀZ
^
kkÂH
ð21Þ
where Z ¼
ﬃﬃﬃﬃﬃﬃﬃ
m=e
_
is the intrinsic impedance of the medium
(Z ¼Z
0
E376.730313 O in vacuum). So, a uniform plane
wave has E, H, and
^
kk mutually orthogonal to each other
(which is not the case for nonuniform plane waves). Waves
with such characteristics, that is, obeying the relations in
(21), are called transverse electromagnetic (TEM) waves,
the simplest example of which is the plane wave.
3.1. Wavelength and Phase Velocity
For simplicity, let us consider a uniform plane wave
propagating in the
^
kk¼ ^ zz direction. Consequently,
k
.
r ¼kz. Since
^
kk
.
E¼0, let us further consider
E¼E
0
e
Àjkz
^ xx with E
0
¼pe
jf
. Consequently, from (21),
H¼ðE
0
=ZÞe
Àjkz
^ yy. Note that this is a onedimensional pro
blem; speciﬁcally, the components of E and H satisfy (14),
with constant amplitude and phase planes perpendicular
to ^ zz.
To obtain corresponding expressions in time domain,
we apply the deﬁnition in (10):
E¼p cos ðot Àkz þfÞ ^ xx
H¼
p
Z
cos ðot Àkz þfÞ ^ yy
ð22Þ
So, for a certain instant of time, the spatial variation of E
and H is sinusoidal with a period equal to l ¼2p/k, the
wavelength of the electromagnetic wave. Since k ¼o/c,
then
l ¼
c
f
ð23Þ
where f is the frequency. For such sinusoidal variation,
o/k ¼c is called the phase velocity (v
p
) of the TEM wave.
3.2. Polarization
In the previous example we have assumed Ejj ^ x x, but any
combination between x and y components also satisﬁes
^
kk
.
E¼0 for
^
kk¼ ^ zz. In this situation, such components are
said to be orthogonal to each other and their phase and
amplitude relationships describe the nature of the wave
polarization. Wave polarization plays an important role in
RF and microwave systems, as the interaction between
ﬁelds and obstacles can strongly depend on it. In principle,
ELECTROMAGNETIC WAVE PROPAGATION 1283
it is possible to transmit and receive orthogonal polariza
tions in a communication channel independently, thus
doubling the capacity of the channel. In practice, depolar
ization effects (partial conversion of energy in one polar
ization to another) often occur. This effect can cause, for
instance, cochannel interference in wireless communica
tions. Orthogonal polarizations can also be used in closely
spaced radiolinks to minimize cochannel interference.
According to (21), TEM wave polarization can be
fully characterized by E and
^
kk. So, let us assume
E¼ðp
x
e
jf
x
^ xxþp
y
e
jf
y
^ yyÞe
Àjkz
. Then, from (10), we obtain
E¼p
x
cosðot Àkz þf
x
Þ ^ xxþp
y
cosðot Àkz þf
y
Þ ^ yy ð24Þ
If the x and y components are in phase (i.e., f
x
¼f
y
), then
(24) can be rewritten as
E¼ðp
x
^ xxþp
y
^ yyÞ cosðot Àkz þf
x
Þ ð25Þ
and, for any position and time, E is always parallel to the
plane containing ^ zzð ¼
^
kkÞ and p
x
^ xxþp
y
^ yy. Then, the wave is
said to have a linear polarization, as depicted in Fig. 1.
Note that f
x
¼f
y
7p also provides a linear polarization.
Now, let us assume that p
x
¼p
y
¼p and f
y
Àf
x
¼7p/2.
Then, from (24), we have
E¼p½cosðot Àkz þf
x
Þ ^ xx Ç sinðot Àkz þf
x
Þ ^ yy ð26Þ
and the wave has a circular polarization, as the tip of E
describes a circular helix in space with an axis in the
direction of ^ zz. This is depicted in Fig. 2. For
^
kk¼ ^ zz, f
y
À
f
x
¼p/2 ( ¼ Àp/2) deﬁnes a left (right)hand circular
polarization. For any other combination between f
x
,f
y
and p
x
,p
y
, the wave polarization is elliptical. Finally, from
the discussion conducted here it should be clear that any
circular or elliptical polarization can be decomposed into
two orthogonal linear polarizations with appropriate am
plitude and phase relations. For pictorial depictions of the
different wave polarizations, the reader is referred to, for
instance, the text by Balanis [14].
3.3. Complex Poynting Vector and Stationary Waves
We will now investigate in further detail the behavior of S
for uniform plane waves.
From (13) and (21) we can verify that for a TEM plane
wave
S¼
jEj
2
2Z
^
kk¼
ZjHj
2
2
^
kk ð27Þ
indicating that (in a lossless simple media) S is real (i.e., a
pure active power density) and, consequently, the direc
tion of propagation of a uniform plane coincides with the
direction of energy ﬂux. For instance, in the examples of
Sections 3.1 and 3.2, one immediately observes that
ReðSÞjj ^ zz.
So, let us now extend the example of Section 3.1 to
investigate the case of a stationary wave, which is de
scribed here as a superposition of two TEM plane waves
propagating in opposite directions (i.e.,
^
kk¼ Æ ^ zz)
E¼ E
þ
0
e
Àjkz
þE
À
0
e
jkz
_ _
^ xx
H¼
1
Z
E
þ
0
e
Àjkz
ÀE
À
0
e
jkz
_ _
^ yy
ð28Þ
where H was directly obtained from (21), noting that each
individual wave has its
^
kk pointing in the opposite direction
of the other. Substituting (28) into (13), we obtain
S¼
jE
þ
0
j
2
2Z
À
jE
À
0
j
2
2Z
þj
jE
þ
0
jjE
À
0
j
Z
sinð2kz Àf
þ
þf
À
Þ
_ _
^ zz
ð29Þ
where f
7
are the corresponding phases of E
Æ
0
. So, (29)
indicates that S has real (corresponding to the net ﬂux of
power density) and imaginary (corresponding to the sta
tionary reactive power density) parts. This is the case, for
instance, of ﬁelds on a transmission line terminated by a
load that does not match the line impedance. If jE
þ
0
j ¼jE
À
0
j
the wave of (28) is purely stationary and there is no
average power ﬂux [i.e., Re(S) ¼0, that occurs when the
line is terminated, for example, by a short circuit].
p
y
p
x
y
z
x
z
c
Figure 1. Snapshot of the electric ﬁeld vector of a linearly
polarized wave traveling in the z direction.
p
y
z
x
z
c
Figure 2. Snapshot of the electric ﬁeld vector of a circularly
polarized wave traveling in the z direction.
1284 ELECTROMAGNETIC WAVE PROPAGATION
4. WAVES AND SOURCES
Until this point, we have focused on wave propagation
through regions of simple media where no source is
present. We shall now present relations describing ﬁelds
produced by elementary electric current sources in simple
media. The most common way to accomplish this objective
is by obtaining the socalled Green’s functions. For a
systematic treatment, the reader is referred to Collin’s
text [7] (for guidedwave examples) and Silver’s treatise
[9] (for radiation examples with applications to antenna
theory). The approach adopted here is more restricted and
can be found in most of the Bibliography.
Observing from (1) that r
.
B¼0, one can deﬁne the
magnetic vector potential A such that rÂA¼B. In
simple media, and after applying the Lorenz gauge condi
tion given by c
2
rÁ A¼ À@F/@t, where F is the electric
scalar potential, it can be shown that A satisﬁes a wave
equation as follows [4]:
r
2
AÀ
1
c
2
@
2
A
@t
2
¼ ÀmJ ð30Þ
For timeharmonic ﬁelds, this equation can be represented
in complex notation as
r
2
Aþk
2
A¼ ÀmJ ð31Þ
A solution of these equations can be written in a generic
form as [4]
A¼m
___
V
Jðr
0
ÞGðr; r
0
Þ dv
0
ð32Þ
where V denotes the volumetric region encompassing
J; r
0
and r denote source and observation points, respec
tively; and Gðr; r
0
Þ is the Green’s function. Gðr; r
0
Þ can be
interpreted physically as the ﬁeld produced by a point
source, namely, the solution of
r
2
Gþk
2
G¼ Àdðr Àr
0
Þ ð33Þ
satisfying the appropriate boundary conditions [4]. For
instance, for bounded sources in unbounded simple media
(free space), it can be shown that
Gðr; r
0
Þ ¼
e
ÀjkjrÀr
0
j
4pjr Àr
0
j
ð34Þ
which is known as the freespace Green’s function.
Furthermore, the timeharmonic electromagnetic ﬁeld
can be written in terms of A as [4]
H¼
1
m
rÂA
E¼ Àjo Aþ
1
k
2
rðr
.
AÞ
_ _
ð35Þ
Use of the Lorenz condition in the derivation of (30) has
an important consequence—one does not need to explicitly
consider the charge densities to obtain the ﬁeld. That
should come as no surprise, since charges are related to
currents by the continuity equation (3). In the time
harmonic regime their effects become implicit in (35). In
any event, it is important to mention that (in a classical
sense) gauge conditions other than those due to Lorenz
can be applied [4].
As a simple example, we consider a uniform time
harmonic electric current distribution over a small seg
ment (wire) ‘ along the z axis, centered at the origin
(z ¼0). The length ‘ is assumed very small, such that
‘5l. The electric current ﬂows along ^ zz , with a constant
phasor I
0
for jzj ‘=2 (I
0
¼0 for jzj > ‘=2). This source
distribution is called inﬁnitesimal electric dipole. If the
dipole radiates in free space, then the resulting vector
potential A is given by (32) and (34). Since r
0
¼z
0
^ zz with
jz
0
j ‘=25l, then kjr Àr
0
j % kr and, consequently,
Gðr; r
0
Þ % e
Àjkr
=ð4prÞ. The integral in (32) can then be
readily evaluated to give [9]
A¼
mI
0
‘
4p
e
Àjkr
r
^ zz ð36Þ
where I
0
‘ is the electric dipole moment. The expressions for
the associated electric and magnetic ﬁelds are obtained by
substituting (36) into (35). The result can be easily ex
tended to arbitrary dipole’s locations and orientations [9].
Note from (35) and (36) that the dipole radiation
depends on the length ‘ multiplied by k, that is, on the
ratio ‘=l (also called the electrical length). Any bounded
source distribution in a simple medium can be written as a
superposition of inﬁnitesimal dipoles. As a result, the
freespace radiation properties of any bounded source
distribution depend on its electrical dimension, namely,
its dimension relative to the wavelength. This is also the
case for scattering by bounded obstacles immersed in
simple media.
We note that the integral (32) is generally difﬁcult to
evaluate except for simple current distributions and sim
ple Green’s functions (such as the freespace Green’s
function described above).
4.1. FarField (Radiation) Region
In free space and for observation points located sufﬁciently
far away from (bounded) sources, (34) and, consequently,
(32) can be simpliﬁed by means of a Taylor expansion on
jr Àr
0
j for small values of jr
0
j with respect to r. By
keeping just the ﬁrst few terms on this expansion, one
ends up with a simpliﬁed relation valid for the socalled
radiation (or farﬁeld) region [9]
A %
m
4p
e
Àjkr
r
___
V
J ðr
0
Þe
jk^ rr
.
r
0
dv
0
ð37Þ
ELECTROMAGNETIC WAVE PROPAGATION 1285
where r ¼jrj, in which case relations (35) simplify to
E % Àjo½A ÀðA
.
^ rrÞ ^ rr
H %
1
Z
^ rr ÂE
ð38Þ
4.2. Spherical and Cylindrical TEM Wavefronts
A closer look at (37) and (38) reveals some interesting
properties of the farﬁeld radiation of bounded sources in
simple media. The ﬁeld dependence on the radial distance
r is of the form e
–jkr
/r. This indicates that the surfaces of
constant phase are concentric spheres and that the ﬁeld
intensity decays with 1/r. Also, the propagation direction
(normal to the equiphase surface) is ^ rr. Furthermore, E, H,
and ^ rr are mutually orthogonal. This characterizes a
spherical TEM wave, for which the basic TEM relations
of Section 3 still hold, but now with
^
kk¼ ^ rr. For instance, a
complex Eﬁeld notation of the form E¼pðy; fÞ
ð
^
yy Æj
^
ffÞ expðÀjkrÞ=r characterizes a spherical TEM wave
with a circular polarization, according to the basic deﬁni
tions of Section 3.2. For such a wave, the complex Poynting
vector is given by S¼
^
kkjEj
2
=ð2ZÞ ¼ ^ rrjpðy; fÞj
2
=ðZr
2
Þ, accord
ing to (27).
Such observations show that the average power density
of a spherical TEM wave decays with 1/r
2
in lossless
simple media. This result is also expected from simple
considerations about energy conservation. Since total
radiated power in a lossless medium is conserved and
the spherical area of the wavefront increases with r
2
, the
power density of the spherical wavefront must decay with
1/r
2
. Furthermore, note that a spherical TEM wave be
haves locally as a plane wave when krb1, so that, in free
space, the ﬁeld close to a receiver antenna located sufﬁ
ciently far away from the transmitter antenna can be
locally approximated as a plane wave.
Apart from plane and spherical waves, another simple
kind of wavefront is the cylindrical one. It is useful to
picture a cylindrical wave as the ﬁeld produced by an
inﬁnitely long line current [6]. For observation points far
away from the current axis, the corresponding wavefronts
can be approximated as cylindrical surfaces with a cross
sectional area that increases linearly with distance (the
cylindrical rcoordinate). Consequently, from considera
tions on energy conservation, the radiating ﬁeld decays
with r
À1/2
[6]. It can also be shown that such cylindrical
wave is also a TEM ﬁeld obeying those basic relations of
Section 3, now with
^
kk¼ ^ qq. Furthermore, for krb1 the
cylindrical wavefront can also be locally approximated as
a plane wave.
4.3. Ray Optics Limit
The discussion in the previous section leads toward the
picture of TEM wavefronts propagating far away from the
radiating sources, with directions of propagation normal
to the corresponding equiphase surfaces. As stressed
previously, distances and dimensions are to be considered
large or small visa` vis the wavelength of operation. In the
limit of l0 or, alternatively, kN, the electromagnetic
ﬁeld is expected to behave locally as a TEM wave (i.e., far
away from sources). This is often called the geometric
optics (GO) limit, where diffraction effects are neglected
and a number of simpliﬁcations of the nature of electro
magnetic wave propagation can be made, such as those in
(21) [15].
The GO principles are directly related to the classical
treatment of light. The developments date from ancient
Greece and are intrinsically related to those of Euclidean
geometry [1]. Obviously, such studies were not based on
Maxwell’s equations, but, as expected, the GO principles
can be derived from Maxwell’s equations. The usual start
ing point for this derivation is to assume a monochromatic
wave (i.e., a timeharmonic ﬁeld) governed by (12) in
complex notation. Adopting the notation of Section 2.1,
we let C be a Cartesian component of E or H(a function of
position only). Consequently
r
2
Cþk
2
C¼0 ð39Þ
From the TEM waves discussed up to here, it can be
assumed, as a ﬁrstorder ansatz, that [1]
C % pðrÞe
Àjk
0
FðrÞ
ð40Þ
where p represents the (complex) amplitude variation, F
represents the phase variation with position only, and k
0
is
the wavenumber for vacuum. For other simple media one
can deﬁne the index of refraction
n¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
ðmeÞ=ðm
0
e
0
Þ
_
ð41Þ
such that k¼nk
0
. In (40), n is accounted for by FðrÞ.
Furthermore, the wavefront is given by the surfaces of
constant FðrÞ (equiphase surfaces). Consequently, the
direction of propagation is
^
kk¼rF=jrFj. The equation
that establishes the optical path (trajectory) in terms of
the wavefront properties is accomplished by substituting
(40) into (39) and, in the k
0
Nlimit, assuming that any
variation of p(r) with position is negligible with respect to
that of k
0
F(r). After some algebraic manipulations, one
arrives at the socalled eikonal equation [1]:
jrFj
2
¼n
2
ð42Þ
This equation can be used to derive Fermat’s principle [1]
Fðr
2
Þ ÀFðr
1
Þ ¼
_
r
2
r
1
nd‘ ¼
_
tðr
2
Þ
tðr
1
Þ
c dt ð43Þ
where Fðr
2
Þ ÀFðr
1
Þ is the optical arc length (along the
optical path ‘) between points r
1
and r
2
, as illustrated in
Fig. 3. This principle states that light (or, more precisely,
electromagnetic waves in the GO limit) follows the path
corresponding to the shortest travel time [1]. For simple
media, both n and c are constants, but (43) holds for
inhomogeneous media as well. From (43) one can also
derive the laws of reﬂection and refraction [1], or establish
approximate trajectories of radiowave propagation
through Earth’s atmosphere [16].
1286 ELECTROMAGNETIC WAVE PROPAGATION
GO principles are useful (although approximate) tools
for the characterization of electromagnetic wave propaga
tion not only at optical frequencies but also often at RF
and microwave frequencies. For instance, such approxi
mation can be used (possibly augmented by small correc
tions) for characterization of some radio channels at UHF
or higher frequencies [17]. The range of validity of such
approximation can depend on many factors, but it is fair to
say that GO can be used as long as the characteristic
lengths of the problem (obstacle feature sizes, mutual
distances, etc.) are much larger than l.
5. WAVES IN LOSSY AND DISPERSIVE MEDIA
So far, we have considered the propagation of electromag
netic waves in simple media without losses or dispersion.
This assumption implies that the permittivity and perme
ability in the constitutive relations (2) and (11) are con
stant scalar parameters. In general, however, constitutive
relations are not that simple. For example, ferroelectric
and ferromagnetic materials are an example of nonlinear
media, where the constitutive parameters at a point
depend on the ﬁeld strength at that point [4]. Even
dielectrics can exhibit nonlinear effects under sufﬁciently
large ﬁeld strengths (e.g., breakdowns inside capacitors or
transmission lines). On the other hand, crystals present a
wellorganized atomic structure, where the ﬁeld response
is highly dependent on wave polarization. Crystals are an
example of anisotropic materials, where the permittivity
(or permeability) has a tensorial nature [18]. The iono
sphere is another example of anisotropic medium at radio
frequencies, although due to very different reasons [4].
Wave propagation on nonlinear or anisotropic media
will not be considered here; The interested reader may
consult Refs. 4 and 18 for further details. We shall con
tinue to assume simple (linear, homogeneous, and isotro
pic) media, but now accounting for the presence of losses
(lossy simple media).
In simple media and for a timeharmonic regime, losses
in the medium cause a phase delay between the electric
displacement ðDÞ and the electric ﬁeld ðEÞ, such that, in
complex notation, we obtain
D¼eðoÞE¼½e
0
ðoÞ Àje
00
ðoÞE ð44Þ
where e
0
and Àe
00
are the real and imaginary parts of the
complex permittivity e, respectively. The negative imagin
ary part of e indicates the phase delay on D. Here, we still
assume that (11) applies for the magnetic relation, since
for conductors and dielectrics mEm
0
at radiofrequencies,
although, in general, a complex permeability of the form
m(o) ¼m
0
(o) Àjm
00
(o) can also be deﬁned [6].
For dielectrics, the ratio e
00
/e
0
deﬁnes the loss tangent,
where tan
À1
(e
00
/e
0
) is the phase difference between E and D
due to the polarization inertia of the atomic structure
(macroscopic interpretation). For conductors, net free
charges are present and generate a conduction current
J
c
whenever an external ﬁeld is applied. For most con
ductors, J
c
is given by Ohm’s law [4]
J
c
¼sE ð45Þ
where s is the material conductivity of the medium
(approximately constant for frequencies below the infra
red region). For conductive media, J
c
and the correspond
ing net charge density r
c
can be subtracted fromJ and r in
(9), respectively, and Maxwell’s equations in complex
notation can be rearranged in the same form as in (9),
but now with the permittivity replaced by the complex
permittivity of (44), where e
0
is the real permittivity and
e
00
¼s/o [6]. After that, J and r in (9) correspond to the
impressed (external) sources only, as J
c
and r
c
are im
plicitly taken into account by the complex e.
It is important to note that, once the complex permit
tivity in (44) is adopted, k ¼o
ﬃﬃﬃﬃﬃ
me
p
and Z ¼
ﬃﬃﬃﬃﬃﬃﬃ
m=e
_
are
complex quantities as well. These quantities now depend
on o but are still spatially uniform (for homogeneous
media). Therefore, all the results derived from Maxwell’s
equations (in complex notation) in previous sections still
hold true, except for those where one had to deal with
conjugate or absolute values (i.e., relations regarding
average energy and power densities). For instance, (4) is
still valid, since it is a general deﬁnition. However, for a
TEM wave propagating in a lossy simple medium, (27)
must account to the fact that Z is now complex and hence
generalizes to
S¼
jEj
2
2Z
Ã
^
kk¼
ZjHj
2
2
^
kk ð46Þ
5.1. Wave Attenuation and Frequency Dispersion
One of the important consequences of losses is the at
tenuation of electromagnetic waves inside the medium. To
illustrate this, we revisit the problem of Section 3.1, where
a uniform plane wave propagates in the ^ zz direction, but
now in a lossy simple medium. From (20), and as the wave
is considered uniform, then akbk^ zz and, consequently,
k¼k^ zz ¼k
^
kk, where k is a complexvalued quantity. So, if
one deﬁnes
k¼b Àja ð47Þ
Wavefront
Optical
path
r
2
r
1
k
Origin
∧
Figure 3. Wavefront and optical path in the geometric optics
approximation.
ELECTROMAGNETIC WAVE PROPAGATION 1287
where a and b are nonnegative real quantities, then a ¼a
^
kk
and b¼b
^
kk. Such deﬁnition is valid for any TEM wave
propagating through a lossy simple medium in the direc
tion
^
kk. The expressions for a and b are obtained from the
deﬁnition k¼o
ﬃﬃﬃﬃﬃ
me
p
and (44):
a
b
_ _
¼o
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
me
0
2
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
1þ
e
0 0
e
0
_ _
2
¸
Æ1
_
_
_
_
¸
¸
¸
¸
_ ð48Þ
The planewave complex representation of Section 3.1 now
becomes
E¼E
0
e
Àaz
e
Àjbz
^ xx
H¼
E
0
Z
e
Àaz
e
Àjbz
^ yy
ð49Þ
Note that Z is now complex. We observe from (49) that the
phase factor b controls the phase variation with distance,
whereas the attenuation factor a controls the amplitude
attenuation as the wave propagates (i.e., as z increases).
Despite the attenuation effect, the surfaces of constant
amplitude and phase are still planes perpendicular to ^ zz,
characterizing a uniform plane wave.
Note that the oscillatory behavior is controlled by e
Àjbz
,
and this factor deﬁnes the wavelength and phase velocity,
now represented as
l ¼
2p
b
and v
p
¼
o
b
ð50Þ
To have a better picture of the wave behavior, we apply the
deﬁnition in (10) to (49) in order to obtain E in time
domain. Assuming E
0
¼pe
Àjf
, we obtain
E¼pe
Àaz
cosðot Àbz þfÞ ^ xx ð51Þ
So, the picture here is that of a sinusoidal function (with
spatial period l) multiplied by an envelope variation given
by pe
Àaz
. The ﬁeld behavior is sketched in Fig. 4 for a
ﬁxed instant of time.
By extending the present analysis to the spherical wave
solution of Section 4.1, one arrives at a ﬁeld dependence
with the radial distance r according to e
Àar
e
Àjbr
/r. The
deﬁnitions in (50) still apply in this case. Besides the
original (algebraic) amplitude attenuation caused by the
spherical spreading factor 1/r, the presence of losses pro
duce an additional (exponential) attenuation factor e
Àar
.
Equation (48) indicates that both b and a have a non
linear variation with o, due to e
0
(o) and e
00
(o). Conse
quently, one observes from (50) that v
p
varies with o, that
is, different frequency components travel at different
phase velocities. This causes the dispersion of an electro
magnetic wave composed by several frequencies. In RF
and microwave applications, electromagnetic waves often
have a certain frequency bandwidth, and dispersion may
cause waveform distortion as the wave propagates. In
general, losses can signiﬁcantly decrease the signalto
noise ratio (SNR) over RF and microwave links, and need
to be minimized and/or compensated (by using, e.g.,
repeaters through longdistance links).
5.2. Group Velocity
In a dispersive (simple) medium, the phase velocity v
p
depends on o. In this case, it becomes necessary to deﬁne
the velocity of a wavepacket (i.e., a wave composed by
different spectral components) for v
p
is no longer ade
quate. To better understand this aspect, let us assume a
wavepacket composed of two timeharmonic TEM compo
nents, propagating in a lossy simple medium in the ^ zz
direction and with the same linear polarization, such that
E¼½p
1
e
Àa
1
z
cosðo
1
t Àb
1
z þf
1
Þ
þp
2
e
Àa
2
z
cosðo
2
t Àb
2
z þf
2
Þ ^ xx
ð52Þ
For the sake of simplicity, we assume p
1
¼p
2
¼p, f
1
¼f
2
¼
0, and small losses, so that a
1
Ea
2
E0. Furthermore, we
deﬁne o
c
¼(o
1
þo
2
)/2, such that o
1,2
¼o
c
8d
o
, where
d
o
¼(o
2
Ào
1
)/2. By so doing, (52) can be rewritten as
E¼2p cos o
c
t Àðb
2
þb
1
Þ
z
2
_ _
cos d
o
t Àðb
2
Àb
1
Þ
z
2
_ _
^ xx ð53Þ
which corresponds to the envelope distribution depicted in
Fig. 5. The ﬁrst cosine (representing, e.g., a carrier with
angular frequency o
c
) propagates with a velocity equal to
2o
c
/(b
2
þb
1
). In the limit d
o
0, (b
2
þb
1
)/2Eb
c
(where b
c
is
the phase constant at o
c
) and, consequently, the carrier’s
velocity tends to the phase velocity at o
c
. The second
cosine in (53) modulates the amplitude of the carrier (see
Fig. 5) and propagates with a velocity equal to d
o
/d
b
, where
d
b
¼(b
2
Àb
1
)/2 corresponds to the variation of b around o
c
.
We then deﬁne the group velocity
v
g
¼ lim
d
o
!0
d
o
d
b
¼
@b
@o
_ _
À1
¸
¸
¸
¸
¸
o¼oc
ð54Þ
as the velocity of the wavepacket envelope, which can also
be interpreted as the velocity of the signal (information)
Vector Field
z
v
p
= o/[
pe
−:z

z
Figure 4. Uniform plane wave propagating in a lossy medium.
1288 ELECTROMAGNETIC WAVE PROPAGATION
that modulates the carrier. A spectral analysis will de
monstrate that (54) still holds for more arbitrary but
bandlimited wavepackets propagating through a medium
with little dispersion (i.e., with relatively small variations
of b) [4]. As the wave energy is intrinsically related to the
ﬁeld strength (amplitude), for bandlimited signals v
g
is
also deﬁned as the velocity of energy ﬂow [4]. Obviously,
v
g
¼v
p
¼c for a TEM propagating through a lossless
simple medium. More detailed discussions on wave velo
cities can be found in Refs. 4 and 5.
6. ELECTROMAGNETIC WAVE INTERACTION WITH
OBSTACLES
6.1. Boundary Conditions
Boundary conditions are needed for the proper solution of
Maxwell’s equations at the interface between two dissim
ilar media (due to the discontinuity on their physical
properties) or at discontinuous source distributions.
One boundary condition is associated with each one of
(1) [6]. For a timeinvariant interface between two regions
1 and 2, the four boundary conditions read as
^ nnÂðE
2
ÀE
1
Þ ¼0 ^ nn
.
ðD
2
ÀD
1
Þ ¼q
s
^ nnÂðH
2
ÀH
1
Þ ¼J
s
^ nn
.
ðB
2
ÀB
1
Þ ¼0
ð55Þ
where ^ nn is the unit normal to the interface pointing
toward region 2 and J
s
and q
s
denote surface (or line)
current and charge distributions over the interface, re
spectively. In (55), the ﬁeld components are evaluated
right next to each side of the interface, at region 1 or 2,
respectively. In the timeharmonic regime, the complex
representation of the boundary conditions have the same
form as (55). If both regions represent simple media, then
(11), (44), and the complex representation of (55) gives
^ nnÂðE
2
ÀE
1
Þ ¼0 ^ nn
.
ðe
2
E
2
Àe
1
E
1
Þ ¼r
s
^ nnÂðH
2
ÀH
1
Þ ¼J
s
^ nn
.
ðm
1
H
2
Àm
1
H
1
Þ ¼0
ð56Þ
where m
j
and e
j
refer to region j (j ¼1, 2).
In simple media and if no surface currents or charges
are present at the interface, both (55) and (56) indicate
that the tangential (to the interface) components of the
electric and magnetic ﬁelds must be continuous across the
interface, and these are the only boundary conditions that
must be imposed for the proper wave solution, as at
regions of a simple medium without sources only Fara
day’s and Ampe`re’s laws are needed [6]. A particular case
of interest is that at the interface of a perfect electric
conductor, which will be discussed in Section 6.3.
6.2. Reﬂection and Refraction of Plane Waves
The planewave reﬂection and refraction at a planar
interface is a canonical problem in electromagnetic wave
theory and provides many useful insights into the beha
vior of waves in the presence of more general obstacles
[19–22]. Consider a plane interface between two simple
media (lossy or not), characterized by their (complex)
z
Vector Field Envelope
v
p
v
g
Figure 5. Group and phase velocities for a wave with two
spectral components with distinct frequencies.
(a)
H
i
y
H
t
k
t
E
t
H
r
k
i
E
i
E
r
k
r
0
i
0
r
0
t
∧
x
∧
z
∧
(b)
H
i
y
H
t
k
t
E
t
H
r
k
i
E
i
E
r
k
r
0
i
0
r
0
t
∧
x
∧
z
∧
j
1
,c
1
j
2
,c
2
j
1
,c
1
j
2
,c
2
Figure 6. Planewave incidence on a plane interface: (a) perpen
dicular and (b) parallel polarizations.
ELECTROMAGNETIC WAVE PROPAGATION 1289
permeabilities (m
1
and m
2
) and permittivities (e
1
and e
2
), as
shown in Fig. 6. An incident plane wave (index i) impinges
on the interface from medium 1, generating a reﬂected
plane wave (index r), propagating back into medium 1,
and a transmitted plane wave (index t) into medium 2. The
exact reﬂected and transmitted energies of the corre
sponding waves depend on the media properties and on
the direction of propagation and polarization of the
incident wave.
Any arbitrary polarization can be decomposed into two
linear and orthogonal ones in simple media (see Section
3.2). For the present analysis, the most commonly used are
the perpendicular (no Eﬁeld component normal to the
interface) and parallel (no Hﬁeld component normal to
the interface) linear polarizations, as depicted in Figs. 6a
and 6b, respectively. The formulas to be presented are
valid for any incident plane wave, uniform or not. Accord
ing to the discussion in Section 3, the geometric inter
pretation illustrated in Fig. 6 (i.e., with realvalued y
angles) is valid only for uniform waves. Depending on
the case, some y angles may come out complex, indicating
that the corresponding plane wave is nonuniform. In any
event, Fig. 6 is indeed useful for establishing the tangen
tial (and normal) components of the ﬁelds at the interface,
necessary for the application of the pertinent boundary
conditions and, consequently, the solution of the problem.
For uniform waves, Fig. 6 also provides a nice picture of
the problem.
We will assume a timeharmonic regime (i.e., mono
chromatic waves) and use the complex phasor notation.
According to (19), we can assume, without loss of general
ity, that the incident propagation vector ðk
i
Þ has no
component in the ^ yy direction. The boundary conditions
will further prove that the same occurs for the reﬂected
ðk
r
Þ and transmitted ðk
t
Þ propagation vectors. With that
we can deﬁne the xz plane in Fig. 6 as the plane of
incidence. For uniform plane waves, that is in conformity
with Fermat’s principle; speciﬁcally, the trajectories are
rectilinear (in simple media) and corresponding to the
least travel time (see Section 4.3).
Starting with the perpendicular polarization and with
the help of (17) and Fig. 6a, we deﬁne the electric ﬁelds of
the plane waves as
E
i
¼E
0
e
Àjk
i
.
r
^ yy
E
r
¼G
?
E
0
e
Àjkr
.
r
^ yy
E
t
¼T
?
E
0
e
Àjk
t
.
r
^ yy
ð57Þ
where G
>
and T
>
are Fresnel’s reﬂection and transmis
sion coefﬁcients for the perpendicular polarization, respec
tively, relating the amplitude and phase of the
corresponding ﬁeld to those of the incident one. The
propagation vectors are written as
k
i
¼k
1
ðsiny
i
^ xxþ cos y
i
^ zzÞ
k
r
¼k
1
ðsiny
r
^ xx Àcos y
r
^ zzÞ
k
t
¼k
2
ðsiny
t
^ xxþ cos y
t
^ zzÞ
ð58Þ
where k
j
¼o
ﬃﬃﬃﬃﬃﬃﬃﬃ
m
j
e
j
p
(with j ¼1,2). Note that (58) is in agree
ment with (18). From (19), (57), and (58) one immediately
obtains the Hﬁeld expressions, observing that
Z ¼Z
j
¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
m
j
=e
j
_
with j ¼1 for the incident and reﬂected
waves and ¼2 for the transmitted wave.
Enforcing the boundary conditions (i.e., the continuity
of the tangential E and Hﬁeld components) at the inter
face plane z ¼0, one comes up with Snell’s law
k
1
sin y
i
¼k
1
sin y
r
¼k
2
sin y
t
ð59Þ
and also with
G
?
¼
Z
2
cos y
i
ÀZ
1
cos y
t
Z
2
cos y
i
þZ
1
cos y
t
T
?
¼
2Z
2
cos y
i
Z
2
cos y
i
þZ
1
cos y
t
ð60Þ
Snell’s law (59) comes from the matching of the phase
variation of the ﬁelds at the interface. For a uniform
incident wave (with a realvalued y
i
), (59) imposes that
y
r
¼y
i
(i.e., the reﬂected wave is also uniform), which is the
law of reﬂection known for centuries in optics and in
conformity with Fermat’s principle [1]. The Snell law for
refraction is given by the second equality in (59), which
can be rewritten with the help of (41)
n
1
sin y
i
¼n
2
sin y
t
ð61Þ
where n
1,2
are the indices of refraction at regions 1 and 2,
respectively. Equations (59) and (61) provide the value of
y
t
with respect to y
i
and the physical properties of the
medium. If one obtains a complexvalued y
t
, this indicates
simply that the transmitted wave is nonuniform. That
may happen even for a uniform incident wave if losses are
present in one of the media (resulting in a complex index
of refraction) or at total reﬂection.
The analysis of the parallel polarization follows along a
line similar to that of its perpendicular counterpart and a
comparison between Figs. 6a and 6b provides the neces
sary insights. After enforcement of the boundary condi
tions at the interface, one comes up with (59) and (61) once
more (i.e., Snell’s law is valid for any planewave polari
zation) and
G
jj
¼
Z
2
cos y
t
ÀZ
1
cos y
i
Z
2
cos y
t
þZ
1
cos y
i
T
jj
¼
2Z
2
cos y
i
Z
2
cos y
t
þZ
1
cos y
i
ð62Þ
which are the Fresnel’s reﬂection and transmission coefﬁ
cients for the parallel polarization, respectively.
For simplicity, we shall assume next that both media
are lossless simple dielectrics (i.e., m
1
Em
2
Em
0
and real
valued e
1
and e
2
) and that the incident plane wave is
uniform (y
i
is real). The incidence is deﬁned external if
e
2
4e
1
(e.g., the incidence of a radiowave on ground in a RF
link). For realvalued y
t
, (61) provides that y
t
oy
i
for
1290 ELECTROMAGNETIC WAVE PROPAGATION
external incidences. Otherwise, the incidence is internal
(e.g., a refraction from water into air).
The behavior of G and T with respect to y
i
is
sketched in Figs. 7a and 7b for external and internal
incidences, respectively. From the ﬁgures one observes
two particular angles of incidence: the Brewster (or polar
izing) angle y
B
, for which jG
k
j ¼0, and the critical angle y
C
,
such that, at internal incidence and for y
i
Zy
C
, G¼1 and
total reﬂection occurs. For lossless and simple media, the
Brewster angle just occurs for the parallel polarization [1].
Its value can be determined from (61) and (62) by setting
G
k
¼0:
tan y
B
¼
n
2
n
1
ð63Þ
If losses are present, then G
k
is complex and jG
k
j ¼0 is
never met for any y
i
. Instead, G
8
 passes through a
minimum, and that deﬁnes an effective Brewster angle.
In any event, one can infer from Fig. 7 that G
8
rG
>

for 0oy
i
op/2. For this reason, polarizing glasses are
designed to block horizontally polarized light reﬂected by
the ground. For this reason also, vertically polarized RF
waves are preferred when attempting to obtain a uniform
coverage of a urban cell in mobile communications at UHF
(due to reﬂections from vertical buildings).
In lossless simple media, the critical angle y
C
occurs
just at internal incidence. This is established from (61)
and from the fact that n
1
4n
2
for internal incidence,
yielding y
t
4y
i
for y
i
ry
C
with y
t
¼p/2 when y
i
¼y
C
by
deﬁnition. Besides, y
t
becomes complex for y
C
oy
i
rp/2,
indicating that the transmitted wave is nonuniform for
incidence angles greater than y
C
. The primary conse
quence is G¼1 for y
i
Zy
C
, as depicted in Fig. 7b.
To better understand the behavior of the transmitted
wave at total reﬂection, let us recall the propagation
vector k
t
of (58). For the present scenario and from (59)
one will verify that the x component of k
t
is realvalued
while its z component is purely imaginary, such that the
transmitted wave propagates along the ^ xx direction while
its intensity dies off exponentially from the interface. So,
the transmitted wave is strongly localized near the inter
face and propagates parallel to it, characterizing a surface
wave. Also, if one obtains the E and Hﬁeld compo
nents—with the help of (21) and (58)—and substitutes
them into (13), it will be shown that the complex Poynting
vector has a purely imaginary ^ zz component and, conse
quently, there is no average ﬂux of active (real) power
through the interface (i.e., the incident power is totally
reﬂected by the interface). In practical situations, losses
will prevent such idealized conditions from occuring and a
(small) amount of energy will eventually be transmitted
through the interface [4]. The critical angle and the
consequent total reﬂection helps in providing a nice
picture of how light is guided throughout an optical
ﬁber [23].
6.3. Waves on Good Electric Conductors
The equations of Section 6.2 can also be used to investi
gate the behavior of a plane wave incident on the surface
of a good electric conductor. Let us assume that region 1 in
Fig. 6 is a lossless simple medium (i.e., m
1
and e
1
are real
valued constants), while region 2 is a good conductor, in
which case m
2
Em
0
and e
2
is complex and according to (44).
By deﬁnition, for good conductors sboe, such that
e
2
EÀjs/o for our purposes, where s is the conducti
vity of region 2. Consequently, k
2
%
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
Àjm
0
s=o
_
and
Z
2
%
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
jom
0
=s
_
.
Applying such deﬁnitions to (59), one should observe
that y
i
¼y
r
, as expected, and that y
t
Ep/2 as sN. Be
sides, from (58) it is veriﬁed that k
t
% k
2
^ zz %
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
Àjm
0
s=o
_
^ zz.
So, we come to the conclusion that the transmitted plane
wave tends to behave as a uniform one inside the good
conductor, propagating in the direction normal to the
conductor’s surface. The corresponding amplitude factor
a
2
%
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
ðm
0
sÞ=ð2oÞ
_
has a considerably large value, so that
the ﬁeld intensity dies off very rapidly away from the
surface (tending to zero when sN, as expected). Thus, a
skin depth (d) is deﬁned as the propagation distance
Perpendicular Pol.
Parallel Pol.
1
0
0
B
0
B
0
C
0
i
0
i
¬
¬
/2
2p
2
p
2
+p
1
p
2
+p
1
p
2
−p
1
2p
2
p
2
+p
1
p
2
+p
1
p
2
−p
1
T
Γ
(a)
(b)
Perpendicular Pol.
Parallel Pol.
2
1
0 /2
2p
2
p
1
T
Γ
Figure 7. G and T as a function of y
i
: (a) external and (b)
internal incidences.
ELECTROMAGNETIC WAVE PROPAGATION 1291
needed for the ﬁeld intensity to decay by e
À1
, namely,
d ¼1=a
2
%
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
ð2oÞ=ðm
0
sÞ
_
.
As a consequence, E
t
is highly concentrated nearby and
parallel to the surface. So, from (45) we end up concluding
that there will be a volumetric conduction current highly
concentrated nearby and ﬂowing parallel to the conduc
tor’s surface. In the limit of a perfect electric conductor
(i.e., sN), such a current behaves as a surface one ðJ
s
Þ,
in which case (55) and (56) apply with null ﬁelds inside
region 2 (note that ^ nn¼ ^ zz in Fig. 6).
It is interesting to stress that the results presented
here are valid for any angle of incidence y
i
and for any
wave polarization. So, according to the discussion in the
beginning of Section 3, the results are valid for any
electromagnetic wave impinging on the surface of a good
electric conductor.
The limiting case of a perfect electric conductor is not
useful just for good conductors. As an example, let us
assume an horizontally polarized radiowave propagating
over ground and at grazing incidence (i.e., with y
i
p/2,
which is generally the case in a longdistance UHF radi
olink). This corresponds to a perpendicular polarization
and, according to (60), G
>
À1. Note also from (60) that
G
>
À1 as well for incidences on a good electric conduc
tor, as Z
2
0 in this case. So, it is often used to
approximate the ground as a perfect (or good) electric
conductor to simplify the analysis [16]. For vertical polar
ization the approximation to be adopted also depends on
other factors (such as the value of y
B
) [16].
6.4. Scattering and Diffraction
The scattering of an electromagnetic wave by an arbitrary
obstacle is a difﬁcult problem to solve using purely analy
tical techniques. Usually, numerical methods are em
ployed. Among these, one can cite the method of
moments [24], the ﬁniteelement method [25], and the
ﬁnitedifference timedomain method [26].
However, if the incidence wave is locally TEM and if the
obstacle is immersed in a simple medium and has a
smooth surface, the concepts discussed in Sections 4.3
and 6.2 can be adopted (as approximations), to the extent
that the obstacle’s dimensions are large compared to the
wavelength. Corrections to account for the curvatures of
the incident wavefront (in case it is not a plane wave) and
of the obstacle surface can be derived from GO principles
and included in G of (60) and (62), according to the wave
polarization [15,27]. The procedure is then conducted by
tracing rays from the transmitter point to the receiver,
such that any reﬂection on the obstacle must obey Fer
mat’s principle, according to which the trajectories are
rectilinear (in simple media) and y
i
¼y
r
with respect to the
surface’s normal. The difﬁculty of such technique gener
ally appears in the determination of the specular points
(where reﬂection occurs) over the surface of the obstacle.
If the obstacle presents curvature discontinuities (such
as, e.g., at the edge of a wedge or at the border of a reﬂector
antenna), then the propagation mechanisms associated
with the incidence on such regions is classiﬁed as a
diffraction (as the diffraction of a laser beam by a metallic
slit). Asymptotic (in the sense of kN) techniques based
on the geometric theory of diffraction (GTD) can be applied
to account for diffraction [27]. Such techniques are also
based on ray tracing and can be accommodated together
with GO principles to characterize the wave propagation
through regions with several obstacles, such as a urban
scenario in mobile communications [18].
7. GUIDED WAVES
Electromagnetic waves can propagate either in open space
or through guiding structures such as transmission lines
and waveguides. The choice of transmission lines depends
on characteristics such as frequency of operation, band
width, powerhandling capability, and losses. Some of the
most usual transmission lines are twowire lines, coaxial
cables, rectangular and circular waveguides, microstrips,
and striplines. These structures are discussed in more
detail elsewhere in this encyclopedia in the article titled
HIGH FREQUENCY TRANSMISSION LINES.
Some authors employ the waveguide denomination for
guiding structures that allow propagation only of trans
verse electric (TE) and/or transverse magnetic (TM) waves,
as described below, while the term transmission lines is
used for guiding structures that allow propagation of TEM
waves as well. Other authors use these terms interchange
ably, as we will do here. We discuss TE and TM waves
next.
7.1. TE and TM Waves
We will assume that the waves are guided in the
z direction. TE and TM waves are a class of solutions of
Maxwell’s equations. As we are looking for propagating
ﬁelds, their z dependence will be assumed on the form
e
Àjk
z
z
. In this case, the transverse components of the
electric and magnetic ﬁelds can be written as [20]
E
t
¼
1
k
2
Àk
2
z
½Àjomðr
t
H
z
Þ Â ^ zz Àjk
z
r
t
E
z
H
t
¼
1
k
2
Àk
2
z
½joeðr
t
E
z
Þ Â ^ zz Àjk
z
r
t
H
z
ð64Þ
where r
t
stands for the transverse (to the z direction)
portion of the r operator. The electric ﬁeld of the TE
component is entirely transverse (E
z
¼0), while the mag
netic ﬁeld has a longitudinal component (H
z
a0). From
(64), the TE ﬁeld is given by H
z
and the corresponding
transverse components E
t
and H
t
. On the other hand, the
magnetic ﬁeld of the TM component is entirely transverse
(H
z
¼0), while the electric ﬁeld has a longitudinal compo
nent (E
z
a0). From (64) the TM ﬁeld is given by the
longitudinal electric ﬁeld E
z
and the corresponding trans
verse components.
TE and TM waves are supported by waveguides con
taining one or more perfect conductors and a homoge
neous dielectric. In this case, each wave (or mode) satisﬁes
the boundary conditions at the waveguide walls, and these
waves are decoupled from each other. A useful example of
such structure is the rectangular waveguide, formed by
four metallic walls (perfect conductors), as shown in Fig. 8.
1292 ELECTROMAGNETIC WAVE PROPAGATION
In general, a rectangular waveguide may be empty inside
or loaded, that is, partially or completely ﬁlled with a
dielectric. We will consider homogeneous (i.e., empty or
completely ﬁlled with a uniform dielectric) rectangular
waveguides next.
For TE modes (also known as H modes), the long
itudinal component H
z
(x,y,z) is given by the solution of
the scalar wave equation that satisﬁes the appropriate
boundary conditions at the four metallic walls (in this
case, @H
z
/@n¼0)
H
z
¼H
0
cos
mpx
a
_ _
cos
npy
b
_ _
e
Àjk
z
z
ð65Þ
where k
z
¼
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
k
2
Àðmp=aÞ
2
Àðnp=bÞ
2
_
, m¼0,1,2y, n¼
0,1,2y (but not m¼n¼0). The transverse components
of the electric and magnetic ﬁelds can be obtained directly
from (64). These modes are denoted TE
mn
, in reference to
the indices m and n of (65). According to the frequency of
operation, the TE
mn
modes may propagate or not inside
the waveguide. A propagating mode is characterized by a
real k
z
. The parameter k
z
is real only if k
2
4(mp/a)
2
þ
(np/b)
2
. Otherwise, the mode is evanescent (exponentially
decaying). The threshold frequency (where k
z
¼0) is called
cutoff frequency and is given by
f
mn
¼
1
2
ﬃﬃﬃﬃﬃ
me
p
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
m
a
_ _
2
þ
n
b
_ _
2
_
ð66Þ
For TM modes (also known as E modes), the longitudi
nal component of the electric ﬁeld is obtained from the
solution of the scalar wave equation satisfying the bound
ary conditions at the four metallic walls (in this case,
E
z
¼0)
E
z
¼E
0
sin
mpx
a
_ _
sin
npy
b
_ _
e
Àjk
z
z
ð67Þ
where m¼1,2y, n¼1,2y, and k
z
is the same as above.
The transverse components of the ﬁelds are obtained by
substituting (67) into (64). Again the TM
mn
modes propa
gate only for frequencies above the cutoff frequency f
mn
given by (66).
Assuming that a4b, the ﬁrst mode that propagates
(i.e., the one with smallest cutoff frequency) is the TE
10
,
which is called the dominant (or fundamental) mode.
Usually, waveguides are designed to operate in a fre
quency range where only the dominant mode can propa
gate. This avoids intermodal dispersion [7] that results
from the different phase velocities of two or more propa
gating modes. Note that if the waveguide is ﬁlled with a
lossy dielectric, the modes are attenuated even above
cutoff.
Mathematically, TE and TM modes are orthogonal
to each other and form a complete set. This means that
any ﬁeld distribution inside the waveguide can be repre
sented as a superposition of TE and TM modes
[7]. However, when the ﬁnite conductivity of the metallic
walls are considered, this orthogonality is no longer valid,
and the modes become coupled. Also, when the dielectric
is not homogeneous, the propagating modes become
a combination of TE and TM ﬁelds, also called hybrid
modes [7].
7.2. Surface and Leaky Waves
Surface waves were introduced in Section 6.2, under
the condition of total internal reﬂection. This type of
wave exhibits an exponential decay away from a guiding
interface, while propagating in a direction parallel to it.
Such a wave is also supported by dielectric waveguides,
where no conductor is needed to guide the ﬁelds. A simple
example of this kind of waveguide is the dielectric slab
waveguide, formed by a dielectric layer (inﬁnite in ^ x x and ^ zz
directions), surrounded by air, as shown in Fig. 9. In the
air, surfacewave modes are evanescent, and there is no
average power ﬂow from the dielectric to the air. TE and
TM modes can be obtained for dielectric waveguide by
following a procedure similar to the one described in
Section 7.1 [6]. For TE and TM modes, there is also a
possibility of even and odd modes. For example, even TM
modes are given by [6]
E
z
¼
E
d
cosðb
y
yÞe
Àjkzz
for jyj
d
2
E
0
e
Àayjyj
e
Àjkzz
for jyj !
d
2
_
¸
¸
_
¸
¸
_
ð68Þ
The associated transverse ﬁeld components are obtained
by substituting (68) into (64). The characteristic equations
a
Metallic wall
b
Z
Y
X
Figure 8. Rectangular waveguide geometry.
Dielectric
d
2
d
2
X
Y
Z
c
o
c
o
c
d
Figure 9. Dielectric slab waveguide.
ELECTROMAGNETIC WAVE PROPAGATION 1293
in the dielectric and in air are given by k
2
z
þb
2
y
¼o
2
m
0
e
d
and k
2
z
Àa
2
y
¼o
2
m
0
e
0
. The boundary conditions at the
dielectric interfaces provides the relation between
the amplitudes ½E
d
cosðb
y
d=2Þ ¼E
0
e
Àa
y
d=2
and require
that [6]
b
y
cot
b
y
d
2
¼
Àa
y
e
d
e
0
ð69Þ
which is the transcendental equation for k
z
, and conse
quently b
y
and a
y
. Multiple solutions of this transcenden
tal equation are often identiﬁed by a subscript n (TM
n
mode).
These modes present surface waves properties when a
y
is real and positive. The cutoff frequency of the TM
n
mode
is the lowest frequency for which it propagates with no
attenuation ½a
y
¼0; b
y
¼o
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
m
0
ðe
d
Àe
0
Þ
_
. In this case, (69)
results in
f
n
¼
n
2d
ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ
m
0
ðe
d
Àe
0
Þ
_ ; with n¼1; 3; 5; . . . ð70Þ
For frequencies below the cutoff, the power is no longer
conﬁned within the dielectric, and part of it is radiated
into air. As a consequence of this radiation loss, the modes
propagate with attenuation, having complex k
z
¼b
z
Àja
z
.
This kind of wave is called a leaky wave [7]. A simple
interpretation of these waves comes from the planewave
expansion of a ﬁeld within the dielectric. As mentioned
earlier, any ﬁeld can be written as a superposition of
plane waves in a sourceless linear medium. From the
dielectric interface problem, it is known that the plane
waves with incidence angles above the critical angle will
be totally reﬂected back to the dielectric, forming a sur
face wave. But those plane waves with incidence angles
below the critical angle will be refracted into the air
(radiation), forming a leaky wave. Odd TM modes and
both even and odd TE modes can be obtained similarly,
and present the same properties as described for even TM
modes.
An important example of dielectric waveguide is the
optical ﬁber [23], which is extensively used in longdis
tance and highbandwidth communications. An optical
ﬁber is formed by a dielectric rod, called core, and one or
more surrounding cylindrical dielectric layers, called
claddings. In the simplest format, only one cladding is
used. Usually the refraction index of the core is slightly
above that of the cladding, allowing the propagation
of surface waves. When the refraction index of the core
is constant, the conﬁguration is known as a steppedindex
ﬁber. Otherwise, it is known as a gradedindex ﬁber.
The modes that propagate in the ﬁber are, in general,
hybrid (combination of TE and TM), and the domi
nant mode is called the HE
11
mode, with no cutoff
frequency [23].
8. COHERENT AND INCOHERENT WAVES
Electromagnetic waves can also be classiﬁed as coherent
or incoherent [1]. Coherent waves have deﬁnite phase
fronts (for each wavelength), while incoherent waves do
not. Coherent waves are produced by sources that emit
energy through a collectively dependent process such as
antennas. Most electromagnetic waves produced by syn
thetic devices at RF and microwave frequencies are co
herent, as we have been considering here. Lasers are also
an example of coherent wave sources.
On the other hand, incoherent waves result from the
radiation of many collectively independent sources. In this
case, the wavefront is not well deﬁned and random phase
ﬂuctuations occur across space (spatial incoherence) and
with varying wavelengths at random intervals (temporal
incoherence). Incoherent sources include most natural
sources of radiation such as the sun, light from ﬂuorescent
lamps and lightbulbs, and LEDs (lightemitting diodes). In
general, electromagnetic waves exhibit some partial de
gree of coherence. As such, this classiﬁcation corresponds
to two ideal extremes. In reality, electromagnetic waves
are at most highly incoherent (in one extreme) or mostly
coherent (in the other extreme).
9. FURTHER REMARKS
In this article, we have focused mainly on classical
aspects of electromagnetic wave propagation in simple
media. For the vast majority of RF and microwave appli
cations, this classical description is sufﬁcient. However,
the detailed interaction and propagation of electromag
netic waves in material media, including molecular and
atomic effects, depends on quantum aspects that are not
covered by a strictly classical description. The theory
of electromagnetic interaction that takes into account
the laws of quantum mechanics is called quantum
electrodynamics (QED) [28,29]. The development
of QED was the basis for the 1965 Nobel Prize in Physics,
shared by Tomonaga (1906–1979), Schwinger (1918–
1994), and Feynmann (1918–1988), who followed earlier
developments by Dirac (1902–1984). When averaged
at a macroscopic level, QED reduces to Maxwell’s equa
tions augmented by phenomenological equations that
can be expressed in terms of macroscopic constitutive
laws.
As a ﬁnal side remark, it should also be noted here that
Maxwell’s equations in their classical form are invariant
to Lorentz transformations [30] and already incorporate
special relativity. Indeed, the need to preserve the form of
Maxwell’s equations in any nonaccelerated frames of
reference (special relativity principle) was a major moti
vation to the development of special relativity theory by
Einstein [31].
BIBLIOGRAPHY
1. M. Born and E. Wolf, Principles of Optics, 6th ed., Pergamon
Press, Oxford, 1980.
2. J. C. Maxwell, Electricity and Magnetism, Academic Press,
New York, 1935.
3. R. S. Elliott, Electromagnetics: History, Theory, and Applica
tions, IEEE Press, Piscataway, NJ, 1993.
1294 ELECTROMAGNETIC WAVE PROPAGATION
4. J. D. Jackson, Classical Electrodynamics, 3rd ed., Wiley, New
York, 1999.
5. J. A. Stratton, Electromagnetic Theory, McGrawHill, New
York, 1941.
6. R. F. Harrington, TimeHarmonic Electromagnetic Fields,
McGrawHill, New York, 1961.
7. R. E. Collin, Field Theory of Guided Waves, 2nd ed., IEEE
Press, Piscataway, NJ, 1991.
8. L. B. Felsen and N. Marcuvitz, Radiation and Scattering of
Waves, IEEE Press, Piscataway, NJ, 1994.
9. S. Silver, ed., Microwave Antenna Theory and Design,
McGrawHill, New York, 1949.
10. J. A. Kong, Electromagnetic Wave Theory, EMW Publishing,
Cambridge, MA, 2000.
11. W. Chew, Waves and Fields in Inhomogeneous Media, IEEE
Press, Piscataway, NJ, 1995.
12. P. M. Morse and H. Feshbach, Methods of Theoretical Physics,
Vol. 1, New York: McGrawHill Book Co., 1953.
13. P. C. Clemmow and J. Wait, The Plane Wave Spectrum
Representation of Electromagnetic Fields, Oxford Univ. Press,
1996.
14. C. A. Balanis, Antenna Theory Analysis and Design, 2nd ed.,
Wiley, New York, 1997.
15. G. A. Deschamps, Ray techniques in electromagnetics, Proc.
IEEE 60(9):1022–1035 (Sept. 1972).
16. D. E. Kerr, ed., Propagation of Short Radio Waves, McGraw
Hill, New York, 1949.
17. J. D. Parsons, The Mobile Radio Propagation Channel, 2nd
ed., Wiley, New York, 2000.
18. A. Yariv and P. Yeh, Optical Waves in Crystals, Wiley, New
York, 1984.
19. C. A. Balanis, Advanced Engineering Electromagnetics, Wi
ley, New York, 1989.
20. D. K. Cheng, Field and Wave Electromagnetics, 2nd ed.,
AddisonWesley, Reading, MA, 1992.
21. D. M. Pozar, Microwave Engineering, 2nd ed., Wiley, New
York, 1998.
22. S. Ramo, J. R. Whinnery, and T. Van Duzer, Field and Waves
in Communication Electronics, 2nd ed., Wiley, New York,
1984.
23. J. A. Buck, Fundamentals of Optical Fibers, Wiley, New York,
1995.
24. J. J.H. Wang, Generalized Moment Methods in Electromag
netics, Wiley, New York, 1991.
25. J. Jin, The Finite Element Method in Electromagnetics, 2nd
ed., Wiley, New York, 2002.
26. A. Taﬂove and S. C. Hagness, Computational Electrody
namics: The FiniteDifference TimeDomain Method, 2nd
ed., Artech House, Boston, 2000.
27. D. A. McNamara, C. W. I. Pistorius, and J. A. G. Malherbe,
Introduction to the Uniform Geometrical Theory of Diffrac
tion, Artech House, Boston, 1990.
28. V. B. Berestetskii, E. M. Lifshitz, and L. P. Pitaevskii,
Quantum Electrodynamics, 2nd ed., Pergamon Press, Oxford,
1982.
29. J. Schwinger, ed., Selected Papers on Quantum Electrody
namics, Dover, New York, 1958.
30. C. W. Misner, K. S. Thorne, and J. A. Wheeler, Gravitation,
Freeman, San Fransisco, 1973.
31. A. Einstein and L. Infeld, Evolution of Physics, Simon &
Schuster, New York, 1966.
ELECTROMAGNETIC WAVE SCATTERING
RANDALL L. MUSSELMAN
USAF Academy
Colorado Springs, Colorado
1. TYPES OF ELECTROMAGNETIC SCATTERING
Fundamentally, there are three types of electromagnetic
scattering mechanisms: reﬂection, refraction, and diffrac
tion. These scattering mechanisms can radiate specularly
or diffusely. Specular scattering, caused by electrically
large ﬂat objects, means that electromagnetic reradiation
travels in parallel rays. Diffuse scattering, caused by ir
regular or electrically small objects, means that the elec
tromagnetic reradiation spreads as it propagates away
from the scattering object.
1.1. Specular Scattering
Of the three fundamental scattering mechanisms, the
most familiar are specular reﬂection and refraction. If
any corners or bends that exist at the boundary are very
gradual compared to the wavelength of the incident ﬁeld,
then the boundary tends to cause specular scattering. Op
tical scattering is often assumed to be specular because
most obstructing bodies that are smooth are electrically
large compared to optical wavelengths. Specular scatter
ing can be modeled with the specular law of reﬂection and
Snell’s law of refraction.
1.1.1. Reﬂection. A familiar example of specular reﬂec
tion is the common reﬂection of a visible image in a mirror,
since the dimensions of the mirror would be huge com
pared to the wavelength of visible light. Other examples
are the radar return from a large ﬂat target, cellular tele
phone multipath from a water tower, or a parabolic dish
used to reﬂect microwave energy to the antenna at its fo
cal point. The ratio of the reﬂected electric ﬁeld E
r
, to the
incident electric ﬁeld E
i
, is called the reﬂection coefﬁcient:
G¼
E
r
E
i
ð1Þ
If the electric ﬁeld in Fig. 1 is parallel to the plane of
incidence containing all three propagation paths (i.e., in
cident, reﬂected, and transmitted paths), then it is has
parallel polarization. When the electric ﬁeld is perpendic
ular to the plane of incidence (the x–z plane in Fig. 1), then
it has perpendicular polarization. The reﬂection coefﬁ
cients for parallel and perpendicular polarizations are
G
jj
¼
Z
2
cos y
t
ÀZ
1
cos y
i
Z
2
cos y
t
þZ
1
cos y
i
ð2aÞ
G
?
¼
Z
2
cos y
i
ÀZ
1
cos y
t
Z
2
cos y
i
þZ
1
cos y
t
ð2bÞ
ELECTROMAGNETIC WAVE SCATTERING 1295
where Z ¼ðm=eÞ
1=2
is the intrinsic impedance for each me
dium.
1.1.2. Refraction. The energy that is not reﬂected from
the electrically large boundary is transmitted through the
boundary. The ratio of the transmitted ﬁeld strength to
the incident ﬁeld strength is called the transmission coef
ﬁcient:
t ¼
E
t
E
i
ð3Þ
The transmission coefﬁcients for parallel and perpen
dicular polarizations are
t
jj
¼
2Z
2
cos y
i
Z
2
cos y
t
þZ
1
cos y
i
ð4aÞ
t
?
¼
2Z
2
cos y
i
Z
2
cos y
i
þZ
1
cos y
t
ð4bÞ
Equations (2) and (4) are known as Fresnel’s equations.
In the process of propagating from one electrical medi
um to the next, the speed of propagation changes. This
change in speed causes a change in the propagation angle
at the boundary, known as refraction. This phenomenon is
commonly seen with visible light at the surface of a calm
pool of water. The fact that an object extending from the
air into the water appears bent is due to the increase in
the propagation velocity of light as it leaves the water and
enters the air. This phenomenon can be modeled with
Snell’s law of refraction.
1.2. Diffuse Scattering
The laws for specular scattering are valid only for electri
cally large scattering bodies. If the object causing the elec
tromagnetic scattering is small compared to the
wavelength of the incident electromagnetic ﬁeld, the in
duced currents would tend to radiate around the contour
of the object, creating diffuse scattering. Unlike specular
scattering, diffuse scattering results when the electromag
netic energy spreads outward as it radiates from the scat
tering object. The smaller the object, the more the energy
will spread as it reradiates.
A simple example of diffuse scattering is an electro
magnetic ﬁeld incident on a very thin cylindrical conduc
tor of length L along the z axis. At frequencies at or below
the microwave region of the electromagnetic spectrum
(fr20 GHz), a thin copper wire will have an electrically
small radius. A ﬁrst approximation is to assume that the
cylinder is a uniform line source [1], with a current dis
tribution of
IðzÞ ¼
I
0
e
jb
0
z
À
L
2
ozo
L
2
0 elsewhere
_
ð5Þ
assuming that the source of the incident ﬁeld is sufﬁcient
ly far away that the induced current magnitude is approx
imately constant across the entire cross section and along
the length of the wire. The phase angle b
0
z along the
length of the wire, is a result of the angle of incidence,
where b
0
is the phase shift per unit length along the wire.
The scattered or reradiated ﬁeld from this wire is similar
to the ﬁeld radiated from a wire antenna and will radiate
as
E¼jom
e
Àjbr
4pr
sin y
_
L=2
ÀL=2
I
0
e
jb
0
z
e
jbz cos y
dz ð6Þ
The radiation pattern of this scattered ﬁeld is of the
form ðsin uÞ=u, where u¼ðb
0
þb cos yÞL=2. The total ﬁeld
around the wire is the superposition, or vector sum at each
point in space, of the scattered ﬁeld and the original inci
dent ﬁeld that would have existed without the wire
present. The scattered ﬁeld added to the incident ﬁeld
creates a pattern with constructive reinforcement in some
directions, and destructive cancellation in other direc
tions. This is the function of the passive elements found
on the Yagi–Uda antenna, common in television and other
VHF and UHF communications.
In the Yagi–Uda antenna, only one set of elements is
active. A halfwavelength dipole antenna usually makes
up the active (or driven) element. The other elements are
simply conductive cylinders or wires that reradiate some
of the energy incident on them from the active element.
Depending on the relative lengths of these passive ele
ments, each of their reradiated ﬁelds will add to the in
cident ﬁeld of the active element, to create an overall
pattern of power ﬂow [2]. This focusing of energy is called
directivity or antenna gain. Other forms of diffuse scatter
ing by electrically small bodies are not so intentional.
1.3. Diffraction
Another form of scattering that cannot be accounted for by
reﬂection or refraction is diffraction. For electrically large
x
E
i
θ
t
θ
i
θ
r
E
t
H
t
E
r
H
r
H
i
z
µ
1
, ε
1
, σ
1
µ
2
, ε
2
, σ
2
Figure 1. Illustration of Snell’s law. A planewave electromag
netic ﬁeld obliquely incident onto a plane boundary, separating
medium 1 (m
1
,e
1
,s
1
) from medium 2 (m
2
,e
2
,s
2
). A reﬂected ﬁeld and
a transmitted ﬁeld scatter from this discontinuity in electrical
constants.
1296 ELECTROMAGNETIC WAVE SCATTERING
scattering bodies, diffraction appears to occur at geometric
discontinuities such as edges and corners. A ﬁrst approx
imation is that currents induced only at these discontinu
ities reradiate electromagnetic energy. Diffraction is the
scattering mechanism that accounts for radiation ﬁlling in
the region that would have been completely blocked (shad
owed) by an opaque obstruction.
Diffraction and reﬂection are important scattering
mechanisms in communications. Whether in an urban or
rural environment, a cellphone user will rarely have di
rect line of sight to the cell base station. Often, the com
munication link can be established only by energy
reﬂecting off of a nearby building, or energy diffracted
around a building or over a hill. Since these obstructions
are typically electrically large, ray tracing techniques,
which incorporate the laws for specular reﬂection and dif
fuse diffraction at edges, are often used to model the prop
agation characteristics of the communication channel.
2. THE LAWS OF SPECULAR REFLECTION AND
REFRACTION
In many applications, an electromagnetic ﬁeld can be as
sumed to be a plane wave. A plane wave is a convenient
approximation amounting to the assumption that the elec
tromagnetic ﬁeld does not vary over the plane perpendic
ular to the direction of propagation. This approximation is
similar to assuming that over small geographic areas, the
planet Earth is ﬂat. For wave propagation, this assump
tion is valid for a small observation area at a great dis
tance from the source of a spherically propagating wave.
Figure 1 shows a planewave electromagnetic ﬁeld in
cident on a boundary in the x–y plane. The generalized
electric ﬁeld will have components in the x, y, and z direc
tions, that is
E¼ð ^ xxE
x
þ ^ yyE
y
þ ^ zzE
z
Þe
Àjbðsin yx þ cos yzÞ
ð7Þ
where b ¼oðmeÞ
1=2
is the phase constant or wavenumber in
units of radians per meter, m is the permeability in henries
per meter, and e is the permittivity of the material in far
ads per meter. Imposing the tangential boundary condi
tion for the electric ﬁeld, the sum of the tangential
components of the incident and reﬂected ﬁelds must equal
that of the transmitted ﬁeld [3–5]:
ð ^ xxE
i
x
þ ^ yyE
i
y
Þe
Àjb
1
sin y
1
x
þð ^ xxE
r
x
þ ^ yyE
r
y
Þe
Àjb
1
sin yrx
¼ð ^ xxE
t
x
þ ^ yyE
t
y
Þe
Àjb
2
sin y
t
x
ð8Þ
This equality can be true for all x only when the expo
nents, or phases, are equal:
b
1
sin y
i
x ¼b
1
sin y
r
x ¼b
2
sin y
t
x ð9Þ
Equation (9) proves the wellknown specular law for re
ﬂection, namely
y
i
¼y
r
ð10Þ
which simply states that the angle of reﬂection equals the
angle of incidence. Equation (9) also leads to Snell’s law of
refraction:
sin y
i
sin y
t
¼
m
2
e
2
m
1
e
1
_ _
1=2
ð11Þ
For most materials, the permeability is the same as that
of free space, m¼m
0
. Assuming m
1
¼m
2
, Eq. (11) reduces to
n
1
siny
1
¼n
2
siny
2
, where n¼ðe
r
Þ
1=2
is the index of refrac
tion and e
r
is the relative permittivity or dielectric con
stant.
3. ELECTROMAGNETIC THEOREMS
Few electromagnetic scattering problems lend themselves
to the simple application of the laws of reﬂection and re
fraction. To develop more sophisticated analysis tools, a
discussion of some basic electromagnetic theorems will be
useful.
3.1. Uniqueness Theorem
Knowledge of the sources (currents) induced on the sur
face of a scattering body S enables unique solutions of the
ﬁelds reradiated by those induced sources. Conversely, the
known ﬁelds allow a unique calculation of the induced
sources. The electric ﬁeld E and magnetic ﬁeld H are
uniquely determined if [6,7]
1. ^ nnÂ
~
EE, the tangential component of E, is speciﬁed
on S.
2. ^ nnÂ
~
HH, the tangential component of H, is speciﬁed
on S.
3. ^ nnÂ
~
EE is speciﬁed on part of S, and ^ nnÂ
~
HH is speciﬁed
on the remaining part of S.
3.2. Induction Theorem
In general, sources, such as conduction, displacement, and
polarization currents, are induced at electrical disconti
nuities in the medium through which the incident ﬁeld is
propagating. Figure 2 shows a typical discontinuity rep
resented by region 2. Region 2 represents a scattering
body of material consisting of a permittivity e, permeabil
ity m, and conductivity s, different from those of the sur
rounding region 1. Region 2 is bounded by a surface S.
Assume that an electric ﬁeld E
1
and its associated mag
netic ﬁeld H
1
, originate from a source current density J
0
.
This source could simply be the current oscillating in a
transmitting antenna. These ﬁelds propagate undis
turbed, through region 1, until they become incident
upon the scattering body of region 2. As the ﬁelds cross
the boundary between regions 1 and 2, they will be per
turbed, that is, E
2
and H
2
in region 2 will generally not be
equal to the ﬁelds E
1
and H
1
propagating in region 1.
This abrupt change or discontinuity in electric and mag
netic ﬁeld strength results in currents that are induced
at the discontinuity. In general, these currents will be
ELECTROMAGNETIC WAVE SCATTERING 1297
distributed through the volume of regions 1 and 2, de
pending on their electrical constants.
According to the induction theorem, whenever there is
a discontinuity of the E and Hﬁelds crossing a boundary S
between two media with different electrical constants, one
can assume that induced currents at S cause the discon
tinuities in the ﬁelds. The induced current can be an elec
tric current sheet [6–8]
~
JJ
s
¼ ^ nnÂð
~
HH
s
À
~
HH
t
Þ ¼ À ^ nnÂ
~
HH
i
ð12aÞ
or a ﬁctitious, but mathematically useful, magnetic cur
rent sheet
~
MM
s
¼ À ^ nnÂð
~
EE
s
À
~
EE
t
Þ ¼ ^ nnÂ
~
EE
i
ð12bÞ
The superscripts ‘‘i,’’ ‘‘s,’’ and ‘‘t’’, pertain to the
incident, outwardly scattered (reﬂected), and transmitted
ﬁelds, respectively, and ^ nn is the normal unit vector
pointing out of the scattering body. If the scattering ob
ject is a perfect conductor, the transmitted ﬁelds vanish,
leaving
~
JJ
s
¼ À ^ nnÂ
~
HH
i
¼ ^ nnÂ
~
HH
s
ð13aÞ
and
~
MM
s
¼ ^ nnÂ
~
EE
i
¼ À ^ nnÂ
~
EE
s
ð13bÞ
The induction theorem alleviates the problem of
knowing the exact distribution of current densities
throughout the volume of the scattering body. The
assumed currents exist only on the boundary S between
the two media. Furthermore, the induced currents can be
calculated directly from knowledge of the incident ﬁeld
that would have existed in the absence of any scattering
object.
3.3. Equivalence Principle
If two different sources produce the same radiating ﬁeld
within a region, these sources are equivalent [8]. If both
regions have the same electrical constants, only an in
wardly scattered (transmitted) ﬁeld exists. It follows from
Eq. (13) that the ﬁelds that are incident on the boundary S
can be replaced by the equivalent current sheets
~
JJ
s
¼ ^ nnÂ
~
HH
s
¼ ^ nnÂ
~
HH
i
ð14aÞ
and
~
MM
s
¼ À ^ nnÂ
~
EE
s
¼ À ^ nnÂ
~
EE
i
ð14bÞ
where in this case, ^ nn is pointing in the direction of the
transmitted or scattered wave. The equivalence theorem
is useful for modeling radiation through apertures, such
as a slot in a conductive plane or a horn antenna.
4. DIFFRACTION
Diffraction is the scattering mechanism that neither re
ﬂects off nor transmits through an obstruction. Even with
opaque material allowing no transmission, diffraction ac
counts for radiation into the geometric shadow region.
This scattering mechanism cannot be modeled with the
specular laws of reﬂection and refraction, such as Snell’s
law. To analyze diffraction exactly would require more
knowledge about the induced current distribution around
the scattering structure than would typically be feasible.
Therefore approximations must be made to simplify
the analysis. Two common approaches to analyzing dif
fraction are the use of geometrical optics and physical
optics.
4.1. Geometrical Optics
Geometrical optics (GO) is a ray tracing technique that
assumes that the electromagnetic energy travels in
straight parallel lines or rays that are perpendicular to
the wavefront. These rays travel from the point of reradi
ation to the observation point. While relatively easy to
implement [9,10], GO is an approximation that relies on
some important assumptions, primarily that the wave
length of the electromagnetic ﬁeld must approach zero.
Clearly, GO is an asymptotic technique valid only for suf
ﬁciently high frequencies, such that the wavelength is in
ﬁnitesimal compared to the dimension of the obstruction.
Since GO assumes inﬁnite frequency, it ignores the wave
nature of the electromagnetic scattering ﬁeld, thus ignor
ing diffraction. The GO model creates an abrupt change in
energy at the transition from the illuminated region to the
shadow region. The abrupt change in ﬁeld strength, with
out currents or charges to account for this discontinuity,
violates boundary conditions. Therefore, GO provides only
the crudest model, accounting only for reﬂection and re
fraction, but not for diffraction.
The geometrical theory of diffraction (GTD) extends GO
to account for diffraction, by introducing a diffraction co
efﬁcient D, analogous to G for reﬂection and to t for trans
mission [11–13]. The total electric ﬁeld E
T
around the
obstruction is
E
T
¼E
g
þE
d
ð15Þ
E
2
, H
2
E
1
, H
1
nˆ
S
µ
1
, ε
1
, σ
1
µ
2
, ε
2
, σ
2
J
s
M
s
M
o
J
o
Figure 2. A scattering body. The incident ﬁeld propagates from
the source current, J
0
, through region 1. As this ﬁeld strikes re
gion 2, the currents J
s
and M
s
are induced at the surface S of the
scattering body.
1298 ELECTROMAGNETIC WAVE SCATTERING
where E
g
is the electric ﬁeld predicted by GO and is zero in
the shadow region. The diffracted ﬁeld for a plane wave of
incidence is given by [14]
E
d
ðrÞ ¼DE
0
ðrÞ
À1=2
e
Àjbr
ð16Þ
Figure 3 shows a plane wave incident on a perfectly ab
sorbing halfscreen. The diffraction coefﬁcient can be quite
involved even for this simple scattering structure. How
ever, away from the shadow boundary [15]
E
d
¼ À
1
2
E
0
ðlÞ
1=2
r
px
e
Àjbr
ð17Þ
In the shadow ( Àx), the diffracted ﬁeld given in Eq.
(17) is the only ﬁeld present. In the region of GO illumi
nation ( þx), the magnitude of the diffracted ﬁeld of Eq.
(17) subtracts from the incident ﬁeld. Figure 4 illustrates
the sum of the diffracted ﬁeld and the GO incident ﬁeld for
(a) z ¼2l and (b) z ¼20l. Clearly, there is a discontinuity
at the transition between the GO illumination and shadow
regions, around x ¼0. This is an obvious limitation of the
GTD, since there should be a smooth transition. One crude
solution would be to simply draw a smooth curve connect
ing each side of the discontinuity through the point x ¼0,
E¼E
0
/2. A more sophisticated method is the uniform the
ory of diffraction (UTD), which is an extension of GTD that
forces a smooth transition between the GO illumination
and the shadow boundary [16].
Many common diffraction problems, such as hilltops
and buildings, can be modeled with this halfscreen or
knifeedge approximation. However, the GTD still relies
on several assumptions. The diffracted ray is assumed to
depend entirely on the incident ray and the characteristics
of the discontinuity itself, such as an edge of a scattering
structure [17]. The GTD is still a highfrequency asymp
totic approximation because it assumes that the structure
is electrically large and conductive [18]. Furthermore,
GTD suffers from the unrealistic discontinuity problem
at the GO illumination–shadow boundary.
4.2. Physical Optics
The edge diffraction problem of Fig. 3 can also be analyzed
using the concept of physical optics (PO), which relies on
Huygen’s principle. Huygen’s principle states that each
point of a primary wavefront acts as a secondary point
source. Each of these secondary sources radiates a spher
ical wave [14]. The primary difference between PO and
GTD is that GTD assumes rays connect from the geomet
ric discontinuity to the observation point, while PO as
sumes that secondary spherical waves radiate from the
unobstructed primary wavefront. Figure 5 shows Huygen
sources radiating into the GO shadow region behind the
absorbing halfscreen. The elemental electric ﬁeld due to
z
r
wavefront
Conducting
Halfplane
GO shadow
x
Figure 3. Halfscreen diffraction using GTD. The currents
induced at the edge, of the conducting halfscreen, radiate into
the GO shadow region.
Figure 4. Simulation of halfscreen diffraction by GTD. The solid
straight line represents the GO incident ﬁeld. The oscillating
curve is the diffracted ﬁeld, calculated by GTD, added to the GO
ﬁeld, at a distance behind the screen of (a) z ¼2 wavelengths and
(b) z ¼20 wavelengths. (This ﬁgure is available in full color at
http://www.mrw.interscience.wiley.com/erfme.)
ELECTROMAGNETIC WAVE SCATTERING 1299
each secondary point source is
dE¼
E
0
r þd
e
Àjbðr þdÞ
ð18Þ
where E
0
is the electric ﬁeld incident on the half plane, r is
the distance from the observation point to the absorbing
half plane, and d is the additional distance to the second
ary sources. From Fig. 5, we obtain
ðr þdÞ
2
¼r
2
þ2rd þd
2
¼x
2
þr
2
ð19Þ
Clearly, the secondary sources closest to the halfscreen
will dominate the amplitude termin Eq. (18). Therefore, one
could assume that r > d in the amplitude termand r
2
> d
2
in
the phase term of Eq. (18). Thus, Eq. (19) reduces to
d ¼
x
2
2r
ð20Þ
and Eq. (18) for the total electric ﬁeld becomes
E¼
E
0
r
e
Àjbr
_
1
xo
e
Àjbðx
2
=2rÞ
dx ð21Þ
where r is a constant of integration. Letting u¼ð2=lrÞ
1=2
x,
and u
0
¼ð2=lrÞ
1=2
x
0
, Eq. (21) becomes
E¼
l
2r
_ _
1=2
E
0
e
Àjbr
_
1
u
0
e
Àjðpu
2
=2Þ
du ð22Þ
The limits of integration can be split into two terms:
E¼
l
2r
_ _
1=2
E
0
e
Àjbr
_
1
0
e
Àjðpu
2
=2Þ
du À
_
u
o
0
e
Àjðpu
2
=2Þ
du
_ _
ð23Þ
which has the form of Fresnel cosine and sine integrals.
Equation (23) can be written as
E¼
l
2r
_ _
1=2
E
0
e
Àjbr
1
2
þj
1
2
À Cðu
0
Þ þjSðu
0
Þ ½
_ _
ð24aÞ
where C(u
0
) and S(u
0
) are the Fresnel sine and cosine in
tegrals, respectively [19–21]. The solution to Eq. (24a) is
similar to the GTD solution found in Fig. 3, with the ex
ception that Eq. (24a) for PO does not suffer the disconti
nuity of Eq. (17) for GTD. In fact, Eq. (18a) has an analytic
solution in the GO illumination–shadow transition region.
The total electric ﬁeld at x
0
¼0 is
Eðx
0
¼0Þ ¼
l
2r
_ _
1=2
E
0
e
Àjbr
1
2
þj
1
2
_ _
ð24bÞ
and has a magnitude of ðE
0
=2Þðl=rÞ
1=2
.
5. DIFFRACTION THROUGH AN APERTURE
The equivalence principle can be combined with PO to
analyze scattering through an aperture. Let an electro
magnetic plane wave be incident normal to the a Âb ap
erture in a thin conducting screen of inﬁnite extent, shown
in Fig. 6. While this assumption of a screen with inﬁnite
extent may not be realistic, it can provide a good approx
imation for an aperture in an electrically large conductive
plane. From the equivalence principle, the reradiated ﬁeld
appears to be generated by the current sheets described in
Eq. (12). Starting from Maxwell’s equations, the electric
and magnetic ﬁelds radiated from the electric and mag
netic current sources are
~
EE¼ Àj
om
4p
__
S
0
~
JJ
0
e
ÀjbR
R
dx
0
dz
0
À
rÂ
1
4p
__
S
0
~
MM
0
e
ÀjbR
R
dx
0
dz
0
_ _
ð25aÞ
and
~
HH¼ Àj
oe
4p
__
S
0
M
0
e
ÀjbR
R
dx
0
dz
0
þ
rÂ
1
4p
__
S
00
~
JJ
0
e
ÀjbR
R
dx
0
dz
0
_ _
ð25bÞ
where the primed terms refer to the source, rather than
the ﬁeld. The equivalence principle allows the electric ﬁeld
in the aperture to be replaced by the magnetic current
sheet M
s
over a continuous conducting screen [7]. The ap
erture is essentially shortcircuited, which cancels J
s
.
x
δ
r
wavefront
Absorbing
Halfplane
GO shadow
Secondary
sources
Figure 5. Halfscreen diffraction using PO. The unblocked sec
ondary sources radiate into the GO shadow region, accounting for
diffraction.
1300 ELECTROMAGNETIC WAVE SCATTERING
From image theory, it appears as though an identical im
age of M
s
lay on the opposite side of the screen. Since these
two current sheets nearly coincide, the entire problem can
be replaced with 2M
s
at the aperture location, and no
screen at all. Then Eq. (25b) becomes
~
HH¼ Àj
oe
2p
__
S
0
M
0
e
ÀjbR
R
dx
0
dz
0
ð26Þ
Rather than solving the integrodifferential equation
(25a), Ampere’s law can be used to obtain directly the
scattered electric ﬁeld in the sourcefree region:
rÂ
~
HH¼joe
~
EE ð27Þ
The distance R from each elemental source to the ﬁeld
point can be obtained from the law of cosines
R¼½r
2
þðr
0
Þ
2
À2rr
0
cos c
1=2
ð28Þ
where r
0
cos c¼x
0
siny cos fþy
0
siny sinf. Equation (26)
would be difﬁcult to integrate with a direct substitution of
Eq. (28). However, if the scattered ﬁeld is observed in the
farﬁeld Fraunhofer region, R and r will be virtually par
allel. The farﬁeld limit is usually taken to be
r !
2D
2
l
ð29Þ
where D is the largest dimension of the aperture, in this
case, the length of the diagonal [19]. The farﬁeld assump
tion allows for the approximation R % r Àr
0
cos c for the
phase, and R % r in the amplitude. Furthermore, in the far
ﬁeld, E¼ZH, where Z ¼ðm=eÞ
1=2
is the intrinsic impedance
of the surrounding medium. This eliminates the need to
solve Eq. (25a) or (27). Since in this case the incident plane
wave is normal to the aperture, it will not vary over the
aperture. Therefore, it can be brought out of the integral.
Then the equation for the scattered magnetic ﬁeld
becomes
H¼ Àj
oe
2pr
E
0
_
a=2
Àa=2
_
b=2
Àb=2
e
Àjbðx
0
sin y cos fþy
0
sin y sin fÞ
dx
0
dz
0
ð30Þ
While appearing messy, Eq. (30) is a straightforward
integral. After integrating the two exponential terms, sub
stituting the limits, and applying the identity
sin a ¼
e
ja
Àe
Àja
j2
the scattered magnetic ﬁeld in Eq. (30) becomes
H¼j
abe
Àjbr
Zlr
sinðXÞ
X
_ _
sinðYÞ
Y
_ _
ð31aÞ
where
X ¼
ba sin y cos f
2
ð31bÞ
and
Y ¼
bb sin y sin f
2
ð31cÞ
Figure 7 is a plot of Eq. (31), with the amplitude nor
malized. The x dimension is a¼6l, the y dimension
is b ¼3l, and the observation screen is z ¼100l from the
aperture.
The normal angle of incidence was chosen for this prob
lem to illustrate the concept, while keeping the mathe
matics simple. However, Eq. (31) can be extended to
oblique incidence by modifying the current source M
s
. As
suming that the source of the incident ﬁeld is far from the
aperture, the amplitude will not vary significantly across
the aperture. However, the phase of each differential ele
ment of M
s
will vary. The procedure is the same as for this
analysis, except that some angle terms for the incident
ﬁeld will be added to X and Y in Eq. (25). The integration
then follows in a similar manner [22].
6. BABINET’S PRINCIPLE
Scattering from a conductive plate can be modeled in a
manner that virtually parallels the preceding solution to
the aperture. In the case of scattering from a conductive
plate, the current sources are obtained using the induction
theorem. In fact, scattering through the aperture is the
exact complement to the scattering off of the conductive
plate that was essentially cut out of the conductive screen
to create the aperture. If every electric parameter and ev
ery magnetic parameter were swapped, the solutions
would be identical. Babinet’s principle originally stated
θ
r’
r
z
R
y
x
φ
Figure 6. Coordinate system for the aperture diffraction prob
lem. R is the vector pointing from the differential element dx dy to
the ﬁeld point, and is the resultant vector sum of r and r
0
. The
position vector from the origin to the secondary source in the
aperture plane is r
0
¼ ðx
0
Þ
2
þðy
0
Þ
2
_ ¸
1=2
.
ELECTROMAGNETIC WAVE SCATTERING 1301
that the sum of the intensities from an obstruction and its
complement (i.e., a similarly shaped aperture in the inﬁ
nite screen) is equal to the intensity that would have ex
isted if no obstruction existed at all
S
a
þS
c
¼S
0
ð32Þ
While this relationship works for optics, it does not take
into account polarization. To apply Babinet’s principle to
vector ﬁelds, it must be modiﬁed to [23]
H
a
H
i
þ
E
c
E
i
¼1 ð33Þ
The ﬁrst term in this equation is the ratio of the ﬁeld
diffracted by the aperture to the ﬁeld with no screen
present at all, and the second term is the ratio of the ﬁeld
produced by the complementary screen to the conjugate
source. The conjugate source refers to the opposite ﬁeld
rotated by 901. In vector form, Eq. (33) can be rewritten as
E
c
¼E
i
ÀZH
a
ð34Þ
which indicates that the electric ﬁeld scattered from a
conductive plate can be calculated from the ﬁeld scattered
from the aperture, by subtracting the latter from the
incident ﬁeld [22].
7. SPECIAL CASES OF ELECTROMAGNETIC WAVE
SCATTERING
7.1. Rayleigh Scattering
If the scattering object is much smaller than a wavelength,
its scattered energy varies inversely as the fourth power of
the wavelength [1,24,25]. Therefore, for a given subwave
length object, higher frequencies will scatter more than
will lower frequencies. This is the basis behind the concept
of Rayleigh scattering for small scatterers. In fact, Ray
leigh scattering answers the proverbial question: Why is
the sky blue? Since the blue end of the visible spectrum
has the shortest wavelength, blue light scatters more than
does the rest of the visible spectrum from dust, water, and
even air molecules. As the scattering objects become larg
er, they fall into the category called Mie scattering.
7.2. Radar Cross Section
Electromagnetic wave scattering is the basis by which ra
dar signals are returned to the radar receiver from a target.
Since the typical radar system employs a collocated trans
mit/receive antenna, the source and observation points are
the same. This scenario is a specific case of electromagnetic
wave scattering, known as monostatic scattering.
As the transmitted power P
T
propagates through space,
it spreads over an increasing surface area A, resulting in
decreased power density S
T
¼P
T
=A. If P
T
spreads spheri
cally as with a point source or isotropic radiator, then
A¼4pd
2
, where d is the distance from the transmitter. A
target can intercept part of the transmitted power and
scatter it in various directions. The radar cross section
(RCS) is the effective area of the target that would return
the monostatic power density back to the source, if this
target scattered the power isotropically [1,26,27]. The
amount of power returned to the source is
P
R
¼
P
T
G
2
l
2
ðRCSÞ
ð4pÞ
3
d
4
ð35Þ
where G is the antenna gain and l is the wavelength of the
radar signal. The RCS is related to the physical crosssec
tional area of the target but also depends on factors such
as the frequency and polarization of the radar signal as
well as the target’s shape, material, and orientation with
respect to the transmitter.
Bistatic radar differs from monostatic radar in that the
receiver and transmitter are not collocated. In the case of
bistatic radar, we obtain
P
R
¼
P
T
G
T
G
R
l
2
ðRCSÞ
ð4pÞ
3
d
2
1
d
2
2
ð36Þ
where G
T
and G
R
are the gains of the transmitting and
receiving antennas, respectively, and d
1
and d
2
are the
distances from the transmitter to the scatterer and the
distance from the scatterer to the receiver, respectively.
This can be a useful model for determining the power
density and thus the electric ﬁeld scattered off of an object
when the transmitter power and antenna gain are known.
[The product of transmitter power and antenna gain is
known as effective isotropic radiated power (EIRP).] The
scattered power density (in Watts per square meter) is
SðinðW=m
2
Þ ¼
P
T
G
T
ðRCSÞ
4p ð Þ
2
d
2
1
d
2
2
ð37Þ
and the RMS electric ﬁeld (in volts per meter) is
E¼ðZSÞ
1=2
. If the EIRP is not known, Eq. (37) can be
modiﬁed by substituting a measured power density S
0
Figure 7. Normalized scattering pattern through the aperture.
The dimensions of the aperture are 6Â3 wavelengths; the obser
vation screen is 100 wavelengths from the plane of the aperture.
The scattering pattern is wider in the x direction, since the x di
mension of the aperture is twice that of the y dimension. (This
ﬁgure is available in full color at http://www.mrw.interscience.
wiley.com/erfme.)
1302 ELECTROMAGNETIC WAVE SCATTERING
measured at a distance d
0
from the original source. In
terms of the measured power density, the scattered power
density is
SðinðW=m
2
Þ ¼S
0
d
0
d
1
_ _
2
ðRCSÞ
4pd
2
2
ð38Þ
8. SUMMARY
Since electromagnetic scattering perturbs the incident
ﬁeld, it can create interference, both constructive and
destructive. The three main mechanisms of scattering
are reﬂection, refraction, and diffraction. Reﬂection and
refraction are the most common, and since these scatter
ing mechanisms tend to be specular, they are easiest to
analyze. Diffraction is much more difﬁcult to analyze, and
is typically not as dominant as the other two. Various
approximations can lead to solutions of diffraction
problems. The two main approximations covered were
the GTD, which takes advantage of ray tracing, and
PO, which relies on Huygen secondary sources. Slight
modiﬁcations to the PO solution for the aperture problem
can lead to solutions for backscattering, and forward
scattering from a conductive plate. This simple structure
can serve as a building block for more complicated struc
tures, which can be modeled as composites of conductive
plates.
BIBLIOGRAPHY
1. W. L. Stutzman, Antenna Theory and Design, 2nd ed., Wiley,
New York, 1998.
2. J. D. Kraus, Antennas, 2nd ed., McGrawHill, New York,
1988.
3. C. R. Paul and S. A. Nasar, Introduction to Electromagnetic
Fields, McGrawHill, New York, 1987.
4. C. T. A. Johnk, Engineering Electromagnetic Fields and
Waves, Wiley, New York, 1988.
5. G. G. Skitek and S. V. Marshall, Electromagnetic Concepts
and Applications, PrenticeHall, Englewood Cliffs, NJ, 1982.
6. C. A. Balanis, Advanced Engineering Electromagnetics, Wiley,
New York, 1989.
7. E. C. Jordan and K. G. Balmain, Electromagnetic Waves and
Radiating Systems, PrenticeHall, Englewood Cliffs, NJ,
1968.
8. R. F. Harrington, TimeHarmonic Electromagnetic Fields, Mc
GrawHill, New York, 1961.
9. G. E. Corazza, V. DegliEsposti, M. Frullone, and G. Riva, A
characterization of indoor space and frequency diversity by
raytracing modeling, IEEE J. Select. Areas Commun. 14(3):
411–419 (April 1996).
10. M. Kimpe, V. Bohossian, and H. Leib, Ray tracing for
indoor radio channel estimation, Proc. IEEE, 2nd Int. Conf.
Universal Personal Communication (ICUPC), Oct. 1993, pp.
64–68.
11. J. B. Keller, A geometrical theory of diffraction, in L. M.
Graves, ed., Calculus of Variations and its Applications,
Proc. Symp. Applied Mathematics, Vol III, McGrawHill,
New York, 1958, pp. 27–52.
12. G. E. Athanasiadou, A. R. Nix, and J. P. McGeehan, Indoor 3D
ray tracing predictions and their comparison with high res
olution wideband measurements, Proc. IEEE, 46th Vehicle
Technology Conf., April 1996, Vol. 1, pp. 36–40.
13. O. Landron, M. J. Feuerstein, and T. S. Rappaport, A com
parison of theoretical and empirical reﬂection coefﬁcients for
typical exterior wall surfaces in a mobile radio environment,
IEEE Trans. Anten. Propag. 44(3):341–351 (March 1996).
14. R. H. Clarke and J. Brown\com}, Diffraction Theory and An
tennas, Wiley, New York, 1980.
15. J. D. Kraus, Electromagnetics, 4th ed., McGrawHill, New
York, 1992, pp. 622–627.
16. R. G. Kouyoumjian and P. H. Pathak, A uniform geometrical
theory of diffraction for an edge in a perfectly conducting sur
face, Proc. IEEE 62(11):1448–1461 (Nov. 1974).
17. J. B. Keller, Geometrical theory of diffraction, J. Opt. Soc. Am.
52(2):116–130 (Feb. 1962).
18. J. B. Keller and A. Blank, Diffraction and reﬂection of pulses
by wedges and corners, in The Theory of Electromagnetic
Waves, Interscience Publishers, New York, 1951, pp. 139–158.
19. C. A. Balanis, Antenna Theory, Analysis and Design, Harper
& Row, New York, 1982, pp. 92–94, 448.
20. E. Hecht, Optics, 2nd ed. AddisonWesley, Reading, MA, 1987,
p. 393.
21. M. Spiegel, Advanced Mathematics for Engineers and Scien
tists, Schaum’s Outline Series, McGrawHill, New York, 1971.
22. R. L. Musselman, Analytic scattering model for indoor prop
agation, Proc. IEEE Antennas & Propagation Society Int.
Symp., 2001, Vol. 4, pp. 734–737.
23. H. G. Booker, Slot aerials and their relation to complementa
ry wire aerials, J. Inst. Electric. Eng. (Part IIIA): 620–626
(1946).
24. A. Ishimaru, Wave Propagation and Scattering in Random
Media, Academic Press, San Diego, 1978.
25. H. C. van de Hulst, Light Scattering by Small Particles, Dover
Publications, New York, 1981.
26. M. I. Skolnik, Introduction to Radar Systems, 2nd ed.,
McGrawHill, New York, 1980.
27. S. A. Hovanessian, Radar System Design and Analysis,
Artech House, Norwood, MA, 1984.
ELECTROMAGNETIC WAVES IN IONOSPHERE
ROBERT D. HUNSUCKER
RP Consultants
Klamath Falls, Oregon
Although the emphasis of this encyclopedia is on the high
er frequencies in the radio spectrum, the importance of the
solar–terrestrial environment on radiowaves of all fre
quencies is of considerable importance to engineers de
signing radio systems for various services. An idealized
representation of the solar–terrestrial environment is
shown in Fig. 1.
It is interesting to note that EM waves over a very large
spectral range interact with the terrestrial ionosphere and
with radio receivers on Earth’s surface or in space: (1) so
lar radiation, including the extreme ultraviolet (EUV:
ELECTROMAGNETIC WAVES IN IONOSPHERE 1303
Next Page