05a

ELECTROMAGNETIC-BANDGAP-ASSISTED
BANDPASS FILTERS
NEMAI C. KARMAKAR
Monash University
Clayton, Victoria, Australia
MOHAMMAD NURUNNABI
MOLLAH
Khulna University of
Engineering and Technology
Khulna, Bangaladesh
1. INTRODUCTION
The new millennium has brought many broadband design
challenges, with the memory-hungry multimedia contents
and new-generation wireless mobile communications de-
mand for larger ever bandwidth to satisfy very high data
rates. The demand can be met with the design with high
functionality per unit volume. Naturally frequency agile
and unwanted frequency immune responses from RF and
microwave devices are needed. Electromagnetic bandgap
structures (EBGSs) are playing a vital role in enhancing
the performance of microwave devices. EBG materials are
periodic structures, that exhibit wide bandpass and band-
rejection properties at microwave frequencies. The intro-
duction of periodic perturbation such as dielectric rods,
holes, and patterns in waveguides and PCB substrates has
resulted in the development of EBGSs, which are also
termed photonic bandgap (PBG) materials. As in a photo-
nic crystal (PC), photon propagation is impeded by elec-
trons and the electromagnetic (EM) waves in a EBG
material are impeded by the periodic discontinuity, hence
resulting in a slow-wave structure. Because of their unique
properties of distinct passband and stopband characteris-
tics, planar PBG materials nd potential applications in
antennas, ampliers, waveguides, lters, power dividers/
combiners, phased arrays, electromagnetic compatibility
(EMC) measurements, and many other microwave devices
[118]. The passband characteristic of the PBG structure
(PBGS) is used as a slow wave medium, which reduces the
size of electronics boards. The wide stopband is applied to
suppress spurious transmission and leakage in guided
structures such as conductor-backed coplanar waveguides
(CPWs) and striplines. Careful design of the PBGS can
improve antenna matching and hence bandwidth, and can
suppress spurious harmonics, sidelobes, and backlobes of
antenna elements and arrays. In lters PBGSs are used for
harmonic suppressions such that in a lterbank the inter-
ference with other lters can be minimized.
This article investigates PBG-assisted asymmetric
coupled-line bandpass lters (BPFs). Both conventional
circular patterned PBGs and defected ground structures
(DGSs) are used to suppress the harmonics of the lter.
Since PBG is a periodic structure, the theory of periodic
structures is reviewed rst. The passband and stopband
phenomena are explained with the help of capacitively
loaded transmission line. The formulation for full-wave
analysis of 1D PBG structure is also reported. The trans-
mission-line model of a modied uniplanar compact PBG
structure (UCPBGS) is analyzed and a dispersion diagram
of the UCPBGS is presented. The calculated stopband and
passband properties of the modied UCPBGS validate the
model. This structure is used in harmonic suppression of a
hairpin lter [37].
Before presenting the PBG-assisted BPF design, we
will analyze a PBG-engineered 50-O transmission line
realized by 2D and 1D circular patterned uniform PBGSs
(UPBGSs). A comparison of the 1D and 2D UPBGSs shows
that there is no signicant difference in passband and
stopband properties between the two. Next, UPBGSs are
used to suppress the harmonics of a four-section asym-
metric coupled-line BPF. Different designs of UPBG-
assisted BPFs are investigated and the S-parameter
performance is described to investigate the effectiveness
of harmonic suppression. This comprehensive investiga-
tion reveals that a careful design procedure is required to
achieve the optimum suppression of harmonics without
affecting the passband performance of PBG-assisted
BPFs. The strong interaction between the poles and zeros
of PBGS and BPFs drastically distorts the passband
performance of BPF.
We investigated nonuniform PBGSs in 2003 [15]. Non-
uniform PBGSs provide better passband return loss (RL)
and minimum passband ripples compared with those for
the conventional UPBGSs. We also observe that the non-
uniform PBGSs yield better harmonic suppression com-
pared with the UPBGSs. In our investigation nonuniform
binomially distributed PBGSs are used for signicant
harmonic suppression. Few designs are investigated to
observe the effects of different numbers and positions of
the unit cells of binomially distributed PBGSs on the
harmonic suppression and passband performances.
Finally, DGS has been used in the BPF. The wide
stopband property of the DGS is exploited to suppress
both second and third harmonics of the BPF. The uniform
and nonuniform PBGSs are designed at a single stopband
frequency, which is the second- or third-harmonic fre-
quency of BPFs. Therefore, two sets of PBGSs are needed
to suppress the second and third harmonics. The new
dumbbell-shaped DGS reported in the open literature [19]
is more compact and yields very wide stopband perfor-
mance [20]. These DGS assisted BPFs are investigated for
simultaneous second- and third-harmonic suppressions of
BPF. Detailed investigations into DGS-assisted BPFs are
presented toward the end of the article, followed by the
conclusion.
2. THEORY OF PBG STRUCTURES
PBGSs have occupied signicant portions of the micro-
wave engineering literature. PBGSs enhance the perfor-
mance of many microwave devices and components. The
acronym PBG is actually used in optical communication.
E
1117
More recently, PBG-related terminology has been scaled
down for RF engineering applications. PBGSs are old
concepts because they generate stopband- and passband-
like periodic structures. To understand the stopband and
passband phenomena of PBGSs, we rst review the con-
ventional periodic structure [21].
Waveguides and transmission lines loaded at periodic
intervals with identical reactive elements are referred to
as periodic structures [21]. This type of period structure
yields two distinct properties:
1. Passbandstopband characteristics
2. Stemming waves with phase velocity lower than the
velocity of light in free space
In the passband EM waves does not attenuate along the
structure. There is some incidental conductor loss only. On
the other hand, in the stopband the EM waves are totally
attenuate so that they cannot propagate throughout the
structure. Stopband characteristics are very important to
suppress surface waves. EM waves having the velocity
lower than the velocity of light in free space are called slow
waves. Periodic perturbation in the ground plane provides
periodic discontinuity. Thus the slow-wave property of the
EM waves is achieved. Slow-wave structures (SWSs) are
promising candidates for compact design.
2.1. Periodic Structures
2.1.1. Capacitively Loaded Transmission Line. A simple
capacitively loaded transmission line can be analyzed to
conceive the idea of periodic structures. The velocity of EM
wave in a physically smooth transmission line can be
written as
V
p
LC
p
1
m
0
e
0
e
r
p 1
where V
p
is the phase velocity of EM wave, L is the series
inductance per unit length, C is the shunt capacitance per
unit length, e
r
is the dielectric constant of the medium
surrounding the conductor, and e
0
and m
0
are free-space
values of the permeability and permittivity, respectively.
From Eq. (1) it is seen that with the value of the
dielectric constant (e
r
), the phase velocity of EM waves
reduces. One problem arises from this simple way of
reducing the phase velocity. If we increase the value of
dielectric constant, then higher-order modes propagate. To
avoid this propagation, the cross-sectional dimensions of
the line must be reduced accordingly. This is the limitation
of increasing the value of dielectric constant to get the
reduced value of the phase velocity of EM waves.
We know LCm
0
e
r
e
0
for dielectric media, so any at-
tempt of increasing the value of C to reduce the phase
velocity is restricted here, because if we increase the value
of C, the value of L will automatically reduce to maintain
the relation LCm
0
e
r
e
0
in a physically smooth transmis-
sion line. Under this circumstance, we can relax the
restriction of a physically smooth transmission line in-
stead of an electrically smooth line. An effective increase
in the shunt capacitance per unit length (C) can be
achieved without disturbing the value of inductance per
unit length (L) by loading lumped shunt capacitance at
periodic intervals where the spacings between the loaded
shunt capacitance are small compared with the wave-
length. At this stage, although the line is not physically
smooth, it will be an electrically smooth line. Under this
condition the capacitance will increase, which can be
observed from the following equation of the phase velocity
V
p
CC
0
=dL
_
o
b
2
where C
0
/d is the loaded lumped capacitance per unit
length, o is the angular frequency, and b is the phase
constant.
There are many ways of obtaining periodic structures.
One of the simplest ways is to load a thin diaphragm at a
regular interval in a coaxial transmission line. The dia-
phragm may be machined as the integral part of the
center conductor. The fringing electric eld in the vicinity
of the diaphragm increases the local storage of electric
energy and gives more extra shunt capacitance.
2.1.2. Circuit Analysis of a Periodic Structure. We may
assume a transmission line to be a combination of the
nite unit cell of the structure. Figure 1a illustrates the
equivalent circuit of a basic unit cell of a capacitively
loaded coaxial line, and Fig. 1b shows the complete
transmission line composed of the basic unit cells.
The unit cell may be divided into three parts: two
transmission lines of length d/2 on either side of a norma-
lized shunt succeptance B.
2.1.2.1. Relationship between Input and Output Varia-
bles. Let V
n
and I
n
be the input voltage and current
variables, respectively, and let V
n1
and I
n1
be the
output voltage and current variables, respectively.
The relationship can be found by using the ABCD
transmission matrix. Overall, the ABCD parameters of a
unit cell can be written as
A B
C D
_
_
_
_
cos y=2 j sin y=2

j sin y=2 cos y=2
_
_
_
_
1 0
j
"
BB 1
_
_
_
_
cos y=2 j sin y=2
j sin y=2 cos y=2
_
_
_
_
3
where y k
0
d and k
0
is the free-space wavenumber.
Figure 1. (a) Equivalent-circuit model of a unit cell of capaci-
tively loaded line; (b) a transmission line cascaded by unit cells.
1118 ELECTROMAGNETIC-BANDGAP-ASSISTED BANDPASS FILTERS
Our unit cells are symmetric where AD. When an
EM wave propagates through periodic structures, the
voltage and current at the (n1)th terminal are equal
to the values of the voltage and current of the nth terminal
except for the phase delay. Please note that we assume the
transmission lines to be lossless. Therefore, the relations
of V and I are as follows:
V
n1
e
gd
V
n
4
I
n1
e
gd
I
n
5
Using Eqs. (4) and (5), we nd that Eq. (6) generates a
matrix eigenvalue equation for the solution of V
n1
and
I
n1
:
V
n
I
n
_ _
A B
C D
_ _
V
n1
I
n1
_ _
e
gd
V
n1
I
n1
_ _
6
Nontrivial solutions for V
n1
and I
n1
are obtained if the
determinant of this matrix vanishes:
A e
gd
B
C De
gd
_
_
_
_
0 or
ADBCe
2gd
e
gd
AD 0
Assuming reciprocal circuit and letting ADBC1, we
have 1 e
2gd
e
gd
(AD) 0 or
cosh gd
AD
2
7
Expansion of (3) shows that AD cos y
"
BB=2 sin y.
Substituting these values in Eq. (7), we have
cosh gd cos y
"
BB
2
sin y 8
Equation (8) is a very vital equation for understanding
passband and stopband phenomena created by periodic
structure.
2.1.2.2. Explanation. We can explain the phenomena
with the following cases.
Case 1. When the right-hand side (RHS) of Eq. (8) is less
than 1, cosh gd cos y
"
BB=2 sin yo1, then, under these
circumstances g jb and
cos bd cos y
"
BB
2
sin y 9
Case 2. When cos y
"
BB=2 sin y > 1, we have cosh gd
cos y
"
BB=2 sin y > 1. Under these circumstances g a,
b 0, and
cosh ad cos y
"
BB
2
sin y 10
Case 3. When cosh gd cos y
"
BB=2 sin yo1, then
gdjp a and coshgdcosh(jp ad) coshad or
cosh ad cos y
"
BB
2
sin yo1
cosh ad cos y
"
BB
2
sin y
_ _ 11
Equations (9)(11) are very important for understanding
the propagation of EM waves through periodic structures.
It is apparent that there will be frequency bands for which
unattenuated wave propagation is possible, separated by
frequency bands in which the wave is attenuated. Fully
attenuated waves yield stopband, and unattenuated
waves generate passband. Thus these equations help us
understand stopband and passband phenomena of PBGS.
Let us consider d5l
0
in our case. So the electrical
length y k
0
d will be small and bd will also be small.
Under these conditions
cos y 1
y
2
2
12
cos bd1
b
2
d
2
2
13
Equation (9) then assumes either of the following forms:
)1
b
2
d
2
2
1
k
2
0
d
2
2

Bk
0
d
2
b
2
d
2
k
2
0
d
2
"
BBk
0
d
b
2
k
2
0
"
BBk
0
d
14
We have the relations:
k
2
0
o
2
m
0
e
0
o
2
LC 15
"
BB
B
Y
c
oC
0
L
C
_
16
(where characteristic impedance Z
c
L=C
_
1=Y
c
). Sub-
stituting k
0
and
"
BB from (15) and (16) in Eq. (14), we have
b
2
o
2
LCo
2
LC
0
d
or
bo
L C
C
0
d
_ _
17
Equation (17) is very important in understanding the
slow-wave effect of periodic structures. We can observe
new value of capacitance CC
0
/d in which capacitance
has been increased by C
0
/d. Therefore, we nd that at low
frequencies where d5l
0
, the loaded line has a shunt
ELECTROMAGNETIC-BANDGAP-ASSISTED BANDPASS FILTERS 1119
capacitance CC
0
/d per unit length. Increased b value
results in reduction of the phase velocity.
2.2. Full-Wave Analysis of One-Dimensional PBG Structure
2.2.1. Fundamental Equation. Maxwells equations are
fundamental equations to form the foundation for electro-
magnetic theory. These equations can be written in differ-
ential form as follows:
rE
@B
@t
18
rHJ
@D
@t
19
r
.
Dr 20
r
.
B0 21
where E is the electric eld intensity, B is the magnetic
ux density, H is the magnetic eld intensity, and D is the
electric ux density. The electric current density J and
electric charge density r are the sources of the electro-
magnetic elds.
The constitutive relations are
De
r
e
0
E 22
Bm
r
m
0
H 23
where e
r
and e
0
are the relative permittivity and free-space
permittivity, respectively and m
r
and m
0
are the relative
permeability and free-space permeability, respectively. The
relation between the electric charge density and electric
current density is dened as the continuity equation
r
.
J
@r
@t
24
Finally, Lorentz force equation determines the total elec-
tromagnetic force on a charge q to be
FqEuB 25
where u is the velocity of the moving charge.
Equations (18)(25) provide the necessary framework
for predicting all macroscopic electromagnetic interactions.
2.2.2. Analytical Techniques. The use of Maxwells
equations in the wave equation and the understanding of
Floquets principle yield the eigenvalue equation for the
periodic structure. The solution of the eigenvalue equation
for the periodic structure can be obtained by the use of
Fourier series. The EM wave may be transverse electric
(TE) or transverse magnetic (TM) depending on the orien-
tation of the structure. We present a mathematical model
for a rectangular patterned UPBGS. The structure of
rectangular patterned uniform PBGSs is shown in Fig. 2.
2.2.2.1. Transverse Electric (TE
x
) Case. The electric eld
can be expanded [22] as a periodic function of plane waves
in x with period d and prescribed propagation constant of
k
x0
as follows
Ex; y ^ zzE
z
x; y ^ zzE
p
xe
jk
x0
x
e
jk
y
y
26
where E
p
(x) is the periodic electric eld that propagates
only in the xy plane, that is, k
z
0. The electric eld must
satisfy the wave equation, so applying the operator
r
2
xy
k
2
to E
z
(x,y) of Eq. (26) and noting that the
dielectric constant is a function of x, we obtain
r
2
xy
E
z
x; y k
2
0
e
r
xE
z
x; y 0 27
Assuming the parallel slabs to be innite in the y and z
directions, Eq. (27) can be simplied to
d
2
dx
2
E
z
x; y k
2
y
E
z
x; y
k
2
0
e
r
xE
z
x; y
28
The periodic electric eld is expanded in a Fourier series
in x with unknown coefcients a
n
, which serves to repre-
sent the dependence on y:
E
p
x
n
a
n
e
j2pn=dx
29
Since the dielectric function is also periodic, it is
appropriate to expand it in another Fourier series with
coefcient b
m
:
e
r
x
m
b
m
e
j2pm=dx
30
Substituting the Fourier expansions for the eld and the
dielectric into Eq. (28) and carrying out the algebraic
y
x z
b
d
Figure 2. One-dimensional lattice of dielectric slabs having
width b and period d.
operations, we obtain
n
2pn
d
k
2
x0
_ _
2
k
2
y
_ _
a
n
e
j2pn=dx
k
2
0
m
a
n
b
m
e
j2pm=dx
e
j2pn=dx
31
In order to determine the unknown coefcients a
n
and b
m
,
Eq. (31) is multiplied by an orthogonal function and
integrated over one unit cell, which produces a Kronecker
delta function for a specic index:
n
2pn
d
k
x0
_ _
2
^
kk
2
y
_ _
a
n
d
2pp
d

2pn
d
_ _
k
2
0
m
a
n
b
m
d
2pp
d

2pm
d

2pn
d
_ _
32
The convolution in Eq. (32) is easily cast into the following
general matrix form
2pn
d
k
x0
_ _
2
k
2
y
_ _
a
n
k
2
0
m
b
nm
a
m
33
where
b
nm
1
d
_
b=2
b=2
e
r
1 exp j
2pn m
d
x
_ _
dx
1
d
_
d=2
d=2
1 exp j
2pn m
d
x
_ _
dx
b
d
e
r
1sinc
pn mb
d
d
nm
34
A generalized linear eigensystem problem is represented
by Ax lBx, where A and B are nn matrices. The value
l is an eigenvalue and xa0 is the corresponding eigen-
vector. The propagating modes in the TE
x
case are solu-
tions of the generalized linear eigensystem in (33).
The solution of Eq. (33) is obtained [22] using the
commercial software Matlab. The resulting eigenvalues
of the matrix are the squares of the frequencies of the
propagating modes in the structure. The solutions of the
frequencies of the propagating modes in the structure are
obtained for the specic values of phase shift. The band
structures for a 1D lattice of dielectric slabs with lling
factor (b/d) of 0.5 is shown in Fig. 3 for k
x0
aA[ p, p]. The
dielectric constant of the dielectric slab is 13.
Figure 3 shows the dispersion diagram of a square
patterned PBGSs. It can be seen that a stopband is
present from normalized frequencies of 0.150.255 GHz.
2.2.3. Transmission-Line Model of PBG Structures. This
section presents the analytical approach of the bandgap
performance of UCPBG. The mathematical formulation of
UCPBGSs is described that has been developed using a
transmission-line model [23].
Basically, a unit UCPBGS consists of a square metallic
patch with four connecting branches. The period of each
cell is approximately a half-wavelength of the center
frequency of the attenuation pole. The slow-wave effect
of the periodic structures causes contraction of the trans-
mission-line wavelength; the period is no longer a half-
wavelength of the same center frequency. Hence, the slow-
wave effect gives rise to a unique compact device struc-
ture. Yang and Itoh [10] designed a periodic UCPBGS
using a complex nite-difference time-domain (FDTD)
method. The method requires numerous numeric codes
to be written for analysis of the specic frequency char-
acteristic of a LPF (lowpass lter) response. The slow-
wave factor (SWF) of the UCPBG structure is approxi-
mately 1.22.4 times higher than that for the unperturbed
line at passband. Rahman and Stuchly [23] proposed a
transmission-line model to calculate the impedance of
each section followed by analysis of the whole structure.
The following section presents the transmission-line
model of a unit UCPBG cell and the dispersion diagrams.
2.2.3.1. UCPBG Theory. The analytic models of two
high-impedance surfaces of UCPBGS are shown in Figs.
4a and 4b. The lattice in Fig. 4a consists of a distribution of
square metal patches; each patch is short-circuited to the
ground plane by a thin pin. The lattice in Fig. 4b consists
of square patches; each is indented with four narrow
connecting branches. The circuits can be considered as
the arrays of reactively loaded resonators coupled by gap
capacitors. If the short-circuiting pins or the connecting
branches are disregarded, the period of each cell would be
a half-wavelength of the resonant frequency. However, the
additional short-circuiting pins or connecting branches
increase the reactive loading of the cell and consequently
cause a lower resonant frequency. This agrees with the
slow-wave effect of the PBG structure. Thus, the 2D
3.00 1.00 0.00 1.00 2.00 3.00
Normalized propagation constant (k
x
d)
2.00
0.00
0.05
0.10
0.15
0.20
0.25
0.30
N
o
r
m
a
l
i
z
e
d

f
r
e
q
u
e
n
c
y

(
f
d
/
c
)
Stopband
Figure 3. Band structure for 1D lattice of dielectric slabs with
Filling Factor of 0.5 having dielectric constant of 13.
UCPBGS arrays shown in Figs. 4a and 4b can be
considered as structures periodically loaded with gap
capacitances and inductances provided by either the
short-circuiting pins or the connecting branches.
The equivalent circuit of the reactive loading part for
Figs. 4a and 4b can be represented by Fig. 4c, where Z
p
is
the reactance of the resonator. The capacitive reactance X
c
represents the coupling gap capacitance between the
neighboring resonators.
With reference to Fig. 4c, the impedance Z
p
of each
resonator section is calculated in two steps using the
following transmission-line formula (where subscript u is
for unloading conditions, l is the length of the transmis-
sion line):
Z
in
Z
0
Z
l
jZ
0
tanb
u
l
Z
0
jZ
l
tanb
u
l
35
Step 1: Taking Z
l
N at l 0 and calculating Z
m
at
l w/2 using Eq. (35), we obtain
Z
m
jZ
0
cot
b
u
w
2
36
Step 2: Taking Z
l
X
L
//Z
m
at l w/2, including the
reactance X
L
represented by a coil, Z
p
can be calcu-
lated using Eq. (35) as follows:
Z
p
Z
0
Z
l
jZ
0
tanb
u
w=2
Z
0
jZ
l
tanb
u
w=2
37
For each cell in Fig. 4a, using the numerical method
given, we can express the inductive loading X
L
of the
central short-circuiting pin, with inductance equal to [24]
L2 10
7
t ln
4t
d
_ _
0:5
d
t
_ _
0:75
_ _
38
where t and d are the length and diameter of the pin,
respectively.
The coupling capacitance between resonators can be
written as [24]
C
we
0
1 e
r
p
cosh
1
a
g
_ _
39
where a,w,g denote the dimensions as shown in Fig. 4. In
Fig. 4 Z
0
is the characteristic impedance of the line and b
u
is the phase constant of a lossless line.
Similarly, considering each cell in Fig. 4b, using the
numerical method given, the total inductance provided by
the four connecting branches, can be calculated as [24]
L0:25 10
7
d
ln
d
s
_ _
0:2235
s
d
_ _
1:193
_ _
k
g
40
where k
g
is a correction factor for the ground plane, which
is given as [24]
k
g
0:57 0:145 ln
s
t
_ _
41
t
d
W
c
w
g
a
s
b
d
z
x
Y
X
Z
p
Z
m
W/2 W/2
Front
view
Front
view
Top view
(a)
Top view
(b)
Z
0,
Z
0,
[
u
X
L
g
1
(c)
[
u
Figure 4. (a) Array of square metal plates
with short-circuiting pins; (b) array of square
metal plates with connecting branches;
(c) equivalent circuit of each resonator section.
and the coupling capacitance between resonators is [24]
C
2be
0
1 e
r
p
cosh
1
w
c
g
_ _
42
Both Z
p
from Eq. (37) in parallel with reactance of the
coupling capacitor X
c
of Fig. 4c contributed to the impe-
dance of unit PBG. Hence, the UCPBG array in Figs. 4a
and 4b can be treated as a transmission line periodically
loaded with lumped impedance Z consisting of Z
p
in
parallel with X
c
with a period of a:
Z
Z
p
X
c
Z
p
X
c
43
The equivalent circuit for the wave propagation in the
x direction shown in Fig. 5 is also an analogous circuit
corresponding to the propagation in the y direction. Ac-
cording to Shumpert [22], the ABCD matrix of a cascade of
unit UCPBG cell, consisting of a transmission-line section
of length a/2, series impedance Z as dened in Eq. (43) and
another transmission-line section of length a/2 is
A B
C D
_
_
_
_
cos
b
u
a
2
jZ
0
sin
b
u
a
2
jY
0
sin
b
u
a
2
cos
b
u
a
2
_
_
_
_
1 Z
0 1
_
_
_
_
cos
b
u
a
2
jZ
0
sin
b
u
a
2
jY
0
sin
b
u
a
2
cos
b
u
a
2
_
_
_
cos b
u
aj
Z
2Z
0
sin b
u
a
Z
0
2
cos b
u
ajZ
0
sin b
u
a
Z
2
Z
2Z
0
cos b
u
a
j
Z
0
sin b
u
a
Z
2Z
2
0
cos b
u
aj
Z
2Z
0
sin b
u
a
_
_
_
_
44
For the conditions of a reciprocal network, ADBC1
and nontrivial solution, the determinant of the matrix
must vanish. We derived the following equation of propa-
gation constant g along with the innite periodic UCPBGS
as
coshga cos b
u
a j
Z
2Z
0
sinb
u
a 45
where Z
0
is the characteristic impedance and b
u
is the
phase constant of the unloaded line; they are the same as
those of the resonators.
With g a jb, Eq. (45) can be rearranged as follows
coshaa cosba j sinhaa sinba
cosb
u
a j
Z
2Z
0
sinb
u
a
46
where a denotes attenuation of the wave propagation.
Since Z is purely reactive for lossless (ideal-case)
resonators, thus rendering the RHS of Eq. (46) a real
value, a purely real value is produced for the left-hand
side (LHS) of Eq. (46) with the condition of either a 0 or
b 0, np/a. This corresponds to passband and stopband
conditions, which are considered below for the wave
propagation in PBG structures:
1. The passband condition of the periodic structure
a 0 and ba0, np/a, in Eq. (46) is reduced to
cosba cosb
u
a j
Z
2Z
0
sinb
u
a 47
2. The stopband condition of the periodic structure
aa0 and b0, np/a, in Eq. (46) is reduced to
coshaa cosb
u
a j
Z
2Z
0
sinb
u
a 48
We have designed the UCPBGS and simulated the struc-
ture using Matlab software. The dimensions of the
UCPBGS presented here are w15.0 mm, a15.13 mm,
g 0.13 mm, g
1
2.615 mm, b 1.885 mm, d3.8mm,
s 6.0 mm, and W
c
9.1mm. The geometry of a modied
UCPBG structure is shown in Fig. 6. This type of UCPBG
structure can nd applications in harmonic suppression of
hairpin lters [2533,37].
The bandgap performance is shown in Fig. 7. As can be
seen in the dispersion diagram in Fig. 7, the rst stopband
is created at 22.8 GHz, followed by a passband.
z z z
a
Z
p
X
c
x
Z
0,
[
u
Z
0,
[
u
Z
0,
[
u
Z
0,
[
u
Figure 5. Equivalent circuit of a periodic structure.
2.3. Uniform PBG Congurations Applied to Microstrip Lines
In the preceding sections the theory and analysis of
periodic structures followed by UCPBG congurations
are presented. In modern microwave devices, microstrip
transmission lines are commonplace. Therefore, in the
following sections, we shall concentrate our investigation
only on PBG-assisted microstrip transmission lines. To
derive the characteristic performance of PBG-assisted
lines, it is more logical to use scattering parameters
(S parameters) instead of showing the Kb diagram.
S parameters are the universally accepted format for
device characterization. Moreover, we shall use the com-
mercially available software tool IE3D (from Zeland Soft-
ware, Inc.) to design and extract S parameters for all
designs due to the exibility of IE3D. IE3D is a method-of-
moment (MoM)-based full-wave analysis tool and hence is
very accurate.
The perturbation in the ground plane of any microstrip
transmission line in the form of PBGS creates stopband.
This stopband is useful for suppression of surface waves,
leakage, and spurious transmission and to improve the
performance of antennas, lters, and other microwave
devices and components. The stopband characteristic is
signicantly inuenced by the shape, size, and period of
the PBGS located on the ground plane. Therefore it is
useful to investigate the performance of the standard 50-O
microstrip transmission line on PBGSs. We will investi-
gate the PBG-assisted line to verify the performance of
three rows of uniform PBGSs as reported by Radisic et al.
[33]. In this section we will investigate uniform circular
patterned PBGSs (UPBGSs). It is well known that the EM
eld is highly concentrated under the microstrip line. We
will investigate one-dimensional (1D) uniform PBGS (one
row) and compare the results with those for two-dimen-
sional (2D) structures (three rows).
2.3.1. Design of Microstrip Transmission Line over
Uniform PBGS (UPBGS). With the inclusion of PBGSs,
the dispersion characteristics of a transmission line
change. First we design a microstrip transmission line
with unperturbed ground plane that does not provide any
stopband characteristics. Then we observe the effect on
the dispersion characteristics in the form of scattering
matrices versus frequency by perturbing the ground plane
with UPBGSs.
Braggs condition is used [34] in the PBGS design to
calculate the central stopband frequency. Under this con-
dition, intercell separation (we call it period) is approxi-
mately equal to half a wavelength of the stopband center
frequency. From the intercell separation, the size of
the PBG element is calculated on the basis of the lling
factor (FF).
The center frequency of the stopband is calculated
approximately with the following expression:
bap 49
where a is the period of the PBG pattern and b is the
wavenumber in the dielectric slab and is dened by the
following expression:
b
2pf
0
c
e
e
p
50
where f
0
is the center frequency of the stopband, e
e
is the
effective relative permittivity of the dielectric slab, and c
denotes the speed of light in free space.
On the basis of Eqs. (49) and (50), we designed three-
row and one-row circular PBGSs in the ground plane of a
50-O microstrip transmission line. The geometries of the
g
w
b

d
W
c
a
s
g
1
Figure 6. Geometry of a modied UCPBG structure to yield
stopband at lower frequency. Dimensions: w15.0mm, a
15.13mm, g 0.13mm, g
1
2.615mm, b 1.885mm, d
3.8 mm, s 6.0 mm, W
c
9.1 mm.
F
r
e
q
u
e
n
c
y

(
G
H
z
)
2
3
1
0
0 40 80 120
Wave number (/m)
Passband
Stopband
Stopband
160 200
4
Figure 7. Plot of propagation wavenumber, passband, and stop-
band frequency range of the 2D UCPBG array. (This gure is
available in full color at http://www.mrw.interscience.wiley.com/
erfme.)
two designs are shown in Fig. 8. The S-parameter perfor-
mances are shown in Fig. 9. It can be seen that both
designs provide identical S-parameter performances. Im-
pressive agreement between the two results justies the
fact that the EM eld is highly concentrated under the
microstrip transmission line. The two side rows in three-
row PBGSs have a minimal effect on the dispersion
characteristics of the line. This nding motivates us to
use 1D PBGSs under only the lines of BPF for harmonic
suppression. PBG-assisted BPF will be discussed in the
next section.
2.4. PBGSs in Harmonic Suppression of BPF
Nowadays microstrip BPFs are extensively used in mod-
ern microwave integrated circuits. The performance of the
overall RF system is affected by the spurious transmission
of a conventional BPF. Yang et al. [1] propose the use of
UCPBGS to suppress the spurious transmission. The
UCPBGS is a complex design on a 2D plane and occupies
more space in a circuit. On the other hand, classical
circular-patterned PBGSs are simple in design and can
be made 1D. 1D PBGSs need less space and create fewer
problems in electronic packaging. We have shown in Fig. 9
that 1D UPBGS provides stopband performance very
similar to that of 2D UPBGS. Therefore 1D UPBGS is
used in a lter to suppress harmonics. The numbers of
PBG elements have signicant inuence on S-parameter
performance [35]. In this section we will investigate the
effect of the number of PBG elements on harmonic sup-
pression of a BPF.
Nonuniform PBGSs realized by the implementation of
binomial and Chebyshev distributions yield better S-para-
meter performance [15]. We will determine whether the
conguration of nonuniform PBGSs can improve harmo-
nic suppression. We will use only binomially distributed
PBG elements.
The harmonics are the multiplication of the fundamen-
tal frequency. 1D and 2D PBGSs generate distinct stop-
bands at a single frequency; therefore it is not possible to
achieve signicant multiple harmonic suppression by 1D
or 2D PBGSs. DGS can yield wider stopbands [19]. We will
use this unique property of DGSs to suppress second and
third harmonics of a lter simultaneously.
2.4.1. Design of UPBGSs for Harmonic Suppression. We
investigate ve UPBGSs assisted BPFs with different
lattices and numbers of UPBGSs: (1) BPF on a 2D array
Substrate
50-Ohm
T-line
(a)
(b)
Circular
PBGSs in the
GND plane
a
a
Figure 8. Uniform circular PBG-engineered
microstrip transmission lines: (a) three rows;
(b) one row.
0 2 4 6 8 10 12 14 16
Frequency (GHz)
90
80
70
60
50
40
30
20
10
0
S
-
p
a
r
a
m
e
t
e
r
s

(
d
B
)
S11_ 3 row
S21_3 row
S11_1 row
S21_ 3 row
Figure 9. S parameters versus frequency of PBG-assisted trans-
mission lines as shown in Fig. 7.
of UPBGSs with square lattice (Fig. 10b), (2) BPFon dense
2D UPBGSs of rectangular lattice (Fig. 10c), (3) BPF on
1D UPBGSs located exactly under the two extreme 50-O
lines and the central coupled line (Fig. 10d), (4) BPF on
UPBGSs that are located under two extreme 50-O lines
only (Fig. 10e), and nally (5) BPF on UPBGSs that are
located under all the lines of a BPF (Fig. 10f). As periodic
structures, both PBGSs and BPFs have their own poles
and zeros. Theoretical formulation of the combined struc-
ture is not a trivial task. Only full-wave analyses of some
structures are reported in the open literature. So far no
thorough investigations into the effect of the position,
number, and lattice structures of UPBGS on BPFs have
been reported. The design investigation of UPBGS-
assisted BPF reveals that PBGS drastically degrades the
passband performance. The design objective of simulta-
neous excellent passband and harmonics suppression is
not fullled. Although many PBG-assisted BPFs have
been reported in the open literature, this problem has
been overlooked. The motivation of thorough full-wave
analysis has been derived to address this issue. In this
section we investigate these effects on BPF with the aid of
IE3D. These thorough investigations are very useful in
understanding the behavior of BPFs in the presence of
PBGSs.
2.4.1.1. Standard BPF. We rst designed a four-section
asymmetric coupled-line BPF (shown in Fig. 10a) on RT/
Duroid with e
r
10.2 and h0.635 mm. The dimensions of
the coupled lines were W
1
0.425 mm, W
2
0.525 mm,
60 ohm
transmission
line
Coupled
line
Coupled
line
Coupled
line
Coupled
line
Coupled
line
PBGs with uniform
distribution
PBGs with uniform
distribution
PBGs with uniform
distribution
PBGs with uniform
distribution
PBGs with uniform
distribution
W
60
W
60 W
60
W
60
Coupled
line
D
0
G
1
G
2
W
2
W
1
L
2
L
1
Ground plane
(a) (b)
(c) (d)
(e) (f)
Ground plane
W50
W50
Ground plane
Ground plane
Ground plane
Ground plane
D
0
D
0
D
0
D
0
Figure 10. Various designs of BPFs assisted by uniform circular PBGSs: (a) standard four-section
asymmetric coupled-line BPF; (b) BPF on 2D square lattice PBGSs; (c) BPF with dense 2D PBGSs
(rectangular lattice); (d) BPF with PBGSs under 50-O and central coupled lines; (e) BPF with
PBGSs under 50-O lines only; (f) BPF with PBGSs under all lines only.
G
1
0.2 mm, G
2
0.7 mm, L
1
L
2
3.625 mm, and W
50
0.6 mm. The passband design frequency was 7.5 GHz.

2.4.1.2. BPF on 2D Array of UPBGSs. In this design the
UPBGSs are etched in the ground plane of the BPF and
their periodicities in the X and Y directions form a square-
patterned lattice structure. The geometry, shown in Fig.
10b, consists of three rows of nine PBG elements. The
UPBGS generates the rst stopband at 15 GHz to sup-
press the second harmonics of the lter.
2.4.1.3. BPF with Dense 2D UPBGSs. Next a 2D array of
UPBGSs is situated beneath all the lines of BPF having
rectangular lattice structures. The Braggs condition is
applied in the x direction to create a stopband at 15 GHz.
The geometry is shown in Fig. 10c.
2.4.1.4. BPF with PBGSs under 50-O and Central Coupled
Lines. In this design PBG elements are etched under 50-O
and central coupled lines only. This design consists of nine
PBG elements total. The geometry of this design is shown
in Fig. 10d. The objective of the design is to eliminate
redundant PBG cells.
2.4.1.5. BPF with PBGSs under 50-O Lines Only. In this
design, PBG elements are located under 50-O lines only.
There are six PBG elements in this design as shown in
Fig. 10e.
2.4.1.6. BPF with UPBGSs under All Lines. UPBGSs are
under all the microstrip lines of BPF, based on the
assumption that the eld is conned below the lines.
Beyond the lines there are no PBG elements. The geome-
try is shown in Fig. 10f.
2.4.2. Results. All the designs are simulated with the
EM Zeland IE3D software. The ground plane has been
assumed to be innite. We used the RT/Duroid substrate
having a dielectric constant of 10.2 and a height of
0.635 mm. We produced theoretical results for different
geometries of UPBGS-assisted BPF. To see the effect of
UPBGS in harmonic suppression, we show the perfor-
mance of the reference BPF. Finally, the results of the
measured insertion loss (IL) performances of a standard
BPFand a UPBGS-assisted BPFare compared in Sections
2.4.2.12.4.2.7.
2.4.2.1. Performance of Reference BPF. An IE3D-simu-
lated S-parameter/frequency plot of a standard asym-
metric coupled-line BPF is shown in Fig. 11. The
presence of spurious transmission around 15 GHz is ob-
vious in the reference BPF. At this frequency the max-
imum value of return loss (RL) is found to be 9dB and IL is
found to be 2.5dB. At 7.5GHz, which is the passband, the
maximum value of return loss (RL) is 22.5 dB and IL is
0.5 dB. At third harmonic, the IL is 0 dB.
2.4.2.2. Performance of BPF on 2D Array of UPBGSs. For
harmonics suppression we applied a 2D array of UPBG in
the ground plane of a reference BPF. The harmonic is
suppressed by application of 2D array of UPBGSs as
shown in Fig. 12. The RL performance at second harmonic
should ideally be zero for proper harmonic suppression.
But the maximum value of RL is here 2 dB and the value
of the transmission coefcient is 9 dB. The IL at third
harmonic is 0dB. We understand that the UPBGSs are not
exact under the lines, so they are not sufciently effective
to suppress the transmission at second-harmonic fre-
quency.
0 5 10 15 20 25
Frequency (GHz)
110
100
90
80
70
60
50
40
30
20
10
0
S
-
p
a
r
a
m
e
t
e
r
s

(
d
B
)
S11
S21
Figure 11. IE3D-simulated S-parameter performance of a stan-
dard coupled-line BPF as shown in Fig. 10a. Substrate is RT/
Duroid having dielectric constant 10.2 and height 0.635mm.
0 5 10 15 20 25
Frequency (GHz)
S
-
p
a
r
a
m
e
t
e
r
s

(
d
B
)
S11
S21
100
90
80
70
60
50
40
30
20
10
0
Figure 12. Theoretical S-parameter performances of a BPF with
square lattice 2D UPBGSs (Fig. 10b). Substrate is RT/Duroid
having dielectric constant 10.2 and height 0.635mm.
2.4.2.3. Performance of BPF on Dense 2D Array of UP-
BGSs. Figure 13 shows the simulation result of a BPF
with 2D PBGSs that are denser than the conventional 2D
square lattice structure. In this case they conform to a
rectangular lattice. We see that at 7.5 GHz the maximum
RL is 25 dB and the maximum RL at 15 GHz is about 0 dB.
The IL at 15 GHz is about 40 dB. Signicant second-
harmonic suppression is achieved here. The third-harmo-
nic value of IL is 0 dB.
2.4.2.4. Performance of BPF with Three-Line UP-
BGSs. We simulated the BPF where the PBGSs are under
two 50-O lines and the central coupled lines. The perfor-
mance is shown in Fig. 14. In this design (Fig. 10d), two
sides of the UPBGSs are situated under two 50-O lines
and three UPBGSs are under the central coupled line.
From the simulation results it can seen that at 15 GHz the
maximum value of RL is about 1 dB; only and the trans-
mission coefcient is found to be approximately 14 dB. The
third-harmonic value of IL is 2 dB. Harmonic suppression
is not satisfactory. In addition to this, the passband
performance at fundamental frequency is distorted sig-
nicantly.
2.4.2.5. Performance of BPF with UPBGSs under Two
Extreme 50-O Lines. Considering the possibility that EM
propagation takes place through the 50-O line, we rst
used UPBGSs under two 50-O lines only to see their effects
on harmonic suppression. The simulation result, shown in
Fig. 15, does not seem promising at all; rather, at funda-
mental, second, and third harmonic frequencies, the per-
formances are poor.
2.4.2.6. Performance of BPF with UPBGSs under All Line-
s. Finally we used PBG under all the lines. The design
provides S-parameter performances as shown in Fig. 16. It
can be seen that at 7.5GHz the maximum RL is more than
30 dB. At 15 GHz the maximum RL is approximately 0 dB
and IL is approximately 30 dB. In this case signicant
improvements in fundamental and second harmonic fre-
quencies are achieved. Only a small ripple in the trans-
mission band is noticed, and this can be controlled by
resizing the UPBGS elements.
0 5 10 15 20 25
Frequency (GHz)
100
90
80
70
60
50
40
30
20
10
0
S
-
p
a
r
a
m
e
t
e
r
s

(
d
B
)
S11
S21
Figure 13. Simulated S-parameter performance of BPF with
dense rectangular lattice 2D UPBGSs (Fig. 10c). Substrate is
RT/Duroid having dielectric constant of 10.2 and height
0.635mm.
0 5 10 15 20 25
Frequency (GHz)
80
70
60
50
40
30
20
10
0
S
-
p
a
r
a
m
e
t
e
r
s

(
d
B
)
S11
S21
Figure 14. Simulated S-parameter performance of a BPF with
UPBGSs on 50-O lines and central coupled line (Fig. 10d).
Substrate is RT/Duroid having dielectric constant 10.2 and height
0.635mm.
0 5 10 15 20 25
Frequency (GHz)
110
100
90
80
70
60
50
40
30
20
10
0
S
-
p
a
r
a
m
e
t
e
r
s

(
d
B
)
S11
S21
Figure 15. Simulated S-parameter performances of a BPF with
PBGSs on two 50-O lines only (Fig. 10e). Substrate is RT/Duroid
having dielectric constant of 10.2 and height of 0.635mm.
2.4.2.7. Comparison of S
21
Performances. Finally we
fabricated the optimized reference BPF and UPBGS-as-
sisted BPF. The measured IL performances up to 20 GHz
are shown in Fig. 17. Compared to the simulation results
in Fig. 11, the measured results in Fig. 17 have shifted in
both fundamental frequencies; this frequency shift may be
due to the fabrication errors. For the reference BPF the
average 3-dB IL bandwidth (IL BW) is 7.18%. Average IL
at second harmonic is found to be 10 dB for the reference
BPF. UPBGS-assisted BPF provides an average IL of
30 dB. The 3 dB IL BW is found to be 16.02%. It can
be concluded that UPBGS-assisted BPF improves per-
formance in terms of IL BW and second-harmonic sup-
pression.
The performance levels of UPBGS-assisted BPF are
shown in Table 1.
2.4.3. Designs of B-PBGs for Harmonic Suppression. We
have designed BPFs with binomially distributed PBGSs
(B-PBGSs). We investigate harmonic suppression due
variation in the number of PBG elements in B-PBG-
assisted BPF (B-BPF). Various designs of B-BPFs are
shown in Fig. 18.
In Fig. 16, the B-PBGSs have a FF of 0.4 (0.4 is the
optimum FF for binomially distributed PBGSs [36]).
Using conducting copper Foil to vary the number of PBG
elements, we compare the IL performances for B-BPF in
several different designs:
Design 1. The ground plane of the reference BPF is
perturbed by PBGSs with binomial distribution. In
this design three lines (consisting of two extreme 50-
O lines and one central coupled line) are perturbed by
B-PBGSs having an FF of 0.4. The rests are loaded
with uniform PBGSs having FF 0.25. The total
number of PBG elements in B-BPFs is 13. The
geometry is shown in Fig. 18a.
Design 2. In this design we use only two PBG elements
from the two extremes and PBG elements of B-BPF
are closed as shown in Fig. 18b.
Design 3. In this case the two 50-O lines of BPF are
loaded by four PBG elements. The remaining PBG
elements are closed. The geometry is shown in Fig.
18c.
Design 4. In this design the BPF is loaded by six PBG
elements from the two extremes. The PBG elements
are under 50-O lines only as shown in Fig. 18d.
Design 5. In this design 10 PBG elements of B-BPFare
used. The design is shown in Fig. 18e. No PBG
elements are situated under the central coupled line.
Design 6. This design consists of 12 PBG elements in
the ground plane of a BPF. Here the central PBG
0 5 10 15 20 25
Frequency (GHz)
80
70
60
50
40
30
20
10
0
S
-
p
a
r
a
m
e
t
e
r
s

(
d
B
)
S11
S21
Figure 16. Theoretical S-parameter performances of a BPF
when UPBGSs are situated under all the lines (Fig. 10f). Sub-
strate is RT/Duroid having dielectric constant 10.2 and height
0.635mm.
0 2 4 6 8 10 12 14 16 18 20
Frequency (GHz)
100
90
80
70
60
50
40
30
20
10
0
S
-
p
a
r
a
m
e
t
e
r
s

(
d
B
)
S21_Ref
S21_UPBG
Figure 17. Measured IL performances of an optimized BPF.
Substrate is Taconic, having dielectric constant of 10 and height
0.635mm.
Table 1. Performance of UPBGS-Assisted BPF
Designs
IL at 2nd
Harmonic (dB)
IL at 3rd
Harmonic (dB)
Reference BPF 2.5 0
BPF on 2D array of UPBGSs 9 0
BPF on dense 2D array
of UPBGSs
40 0
BPF on 3-line UPBGS
(PBGSs on two 50-O
lines central coupled line)
14 2
BPF on two-line UPBGS
(UPBGSs on two 50-O lines)
5 2
BPF on 1D UPBGS
(UPBGSs on all lines)
30 15
element of the central coupled line is omitted. The
design is shown in Fig. 18f.
We present the RL performances for other designs
below. We show only IL performances to indicate harmonic
suppression.
2.4.3.1. Reference Optimized BPF. The experimental
S-parameter performance of a standard BPF is shown in
Fig. 19. We are investigating the harmonics (second and
third harmonics) by varying the number of PBG elements.
For standard BPF the measured second and third harmo-
nics are found to be 13 and 8 dB, respectively.
50 ohm
transmission
line
50 ohm
transmission
line
Coupled
line
Ground plane Ground plane PBGs with Binomial
distribution
PBGs with uniform
distribution
W
50
W
50
W
50
W
50
W
50
L
1
L
2
G
1
G
2 G
1
G
2
(a) (b)
(c) (d)
(e) (f)
Coupled
line
Coupled
line
Coupled
line
Coupled
line
Coupled
line
G
2
G
2
G
1
G
1
L
1
L
2
L
2
L
1
L
2
L
1
L
2
L
1
L
2
D
2
D
2
Ground plane PBGs with Binomial
distribution
PBGs with Uniform
distribution
PBGs with Uniform
distribution
PBGs with Binomial
distribution
PBGs with Binomial
distribution
Ground plane Ground plane
Ground plane
W
50
50 ohm
transmission
line
50 ohm
transmission
line
50 ohm
transmission
line
G
1
G
1
G
2
G
2
D
1
D
0
L
1
D
1
D
0
D
1
D
0
D
1
D
0
D
2
50 ohm
transmission
line
Figure 18. Various designs of B-BPFs with B-PBGSs: (a) elements under all lines; (b) two
elements under 50-O lines only; (c) four elements under 50-O lines; (d) six elements under 50-O
lines; (e) elements under all lines except central coupled lines; (f) elements under all lines except
the central larger PBG elements in the central coupled line.
2.4.4. Performance of B-BPF. We fabricated all the de-
signs and measured performance by vector network ana-
lyzer (VNA). The measured IL and RL performances of
a BPF loaded by binomially distributed PBGS are as
follows:
Design 1. The IL performance of the B-BPF (Fig. 18a)
is shown in Fig. 20. In this design a total of 13 PBG
elements are used to suppress the harmonics. B-BPF
provides second and third harmonics as 32 and 20 dB;
the harmonics are suppressed by 19 and 12 dB,
respectively. However, the IL performance in the
transmission band is seen to be poor. Under these
circumstances we will investigate the inuence of
different numbers of PBG elements on the perfor-
mance of BPF.
Design 2. We investigate the BPF as shown in Fig. 18b
with two PBG elements. During the measurement
the IL performances show second and third harmo-
nics to be 30 and 17.5 dB, respectively (the graph is
not shown here).
Design 3. Four PBG elements etched in the ground
plane of a BPF as shown in Fig. 18d yields the IL
performance, providing second and third harmonics
as 37 and 16 dB, respectively (the graph is not shown
here).
Design 4. During measurement of the IL performance
of a BPF (Fig. 18d) loaded by six PBG elements,
second and third harmonics are found to be 34 and
14 dB, respectively. It is worthwhile to mention that
this conguration hold completely binomially distrib-
uted PBGS under two 50-O lines of a BPF (the graph
is not shown here).
Design 5. The design in Fig. 18e provides 40 and 22 dB
IL as second and third harmonics. The transmission
band at the fundamental frequency is also seen to be
improved. The IL performance for 10 PBG elements is
shown in Fig. 21.
Design 6. As can be seen from Fig. 22, design 5 (Fig. 18f)
yields ILs of 38 and 21dB at second and third harmo-
nics. The transmission band at the fundamental
frequency is similar to that for design 5.
0 5 10 15 20 25
Frequency (GHz)
90
80
70
60
50
40
30
20
10
0
S
-
p
a
r
a
m
e
t
e
r
s

(
d
B
)
S11
S21
Figure 19. Measured S-parameter performance of a standard
BPF. Substrate is Taconic having dielectric constant 10 and
height 0.635mm.
0 5 10 15 20 25
Frequency (GHz)
100
90
80
70
60
50
40
30
20
10
0
I
n
s
e
r
t
i
o
n

L
o
s
s

(
d
B
)
Figure 20. Measured IL performance of a B-BPF shown in Fig.
18a where PBG elements are periodically loading under all the
lines. Substrate is Taconic having dielectric constant 10 and
height 0.635mm.
0 5 10 15 20 25
Frequency (GHz)
90
80
70
60
50
40
30
20
10
0
I
n
s
e
r
t
i
o
n

L
o
s
s

(
d
B
)
Figure 21. Measured IL performance of BPF when 10 PBG
elements are under all lines except the central coupled line.
The IL performances for all the designs are compared
in Table 2, where it can be seen that the 10-PBG element
provides the best performance. 12-PBG elements also
provide almost similar performance. It is also clear from
the investigation that suppressions of second and third
harmonics do not follow the number of PBG elements;
rather, their position and size are the important issues in
harmonic suppression of coupled-line BPF. The important
nding is the role of larger PBG elements of the central
coupled line. Comparing results from Figs. 21 and 22, it
can be concluded that the middle larger element under the
central coupled line degrades the performance.
2.4.4.1. Binomially/Uniformly Assisted BPF Comparison.
In order to investigate the improved performance of B-
BPF over the UBPG-assisted BPF, the combined perfor-
mances are shown in Fig. 23.. It can be seen in Fig. 23
that the average 10 dB RL BW at fundamental frequency
for both B-BPF and U-BPF is 16.2%. The 3 dB IL BW is
found to be 19.56%. The average isolation at second
harmonic is 39 dB for B-BPF, compared to the U-BPF
of 29 dB. The performances of an optimized BPF,
UPBGS-assisted BPF, and B-BPFare compared in Table 3.
2.5. DGS for Harmonic Suppression
2.5.1. Design. In this subsection we report the novel
idea for simultaneous suppression of second- and third-
order harmonics generated in a conventional BPF. We
have already reported that the dumbbell-shaped DGS can
yield a wider stopband [20]. Now we implement this
unique property in an asymmetric four-section coupled-
line BPF to suppress the second- and third-order harmo-
nics. We propose using 1D EBGSs (dumbbell-shaped
DGSs) that are located just under standard 50-O lines.
No DGS will be used under the coupled lines. We also
present a novel idea that the dumbbell-shaped DGSs,
which are designed at two different frequencies, are to
be located under two end 50-O lines only. This design is
sufcient to suppress the second- and third-order harmo-
nics. We investigate a few different designs. Figure 24
shows one of the various designs investigated. It is noted
0 5 10 15 20 25
Frequency (GHz)
110
100
90
80
70
60
50
40
30
20
10
0
I
n
s
e
r
t
i
o
n

L
o
s
s

(
d
B
)
Figure 22. Measured IL performance of a BPF loaded by 12 PBG
elements. Only one larger element under the central coupled line
is closed.
0 2 4 6 8 10 12 14 16 18 20
Frequency (GHz)
100
90
80
70
60
50
40
30
20
10
0
S
-
p
a
r
a
m
e
t
e
r
s

(
d
B
)
S11_Ref
S21_Ref
S21_UPBG
S21_BPBG
Figure 23. Measured S-parameter performance of an optimized
BPF along with the IL performances of B-BPF and UPBGS-
assisted BPF.
Table 2. Performance of a Coupled-Line BPF with
Binomially Distributed PBGS along with Different
Numbers of PBG Elements
Insertion loss (IL) (dB)
Number of PBG Elements 2nd Harmonics 3rd Harmonics
0 (reference BPF) 13 8
2 30 17.5
4 37 16
6 34 14
10 42.5 22
12 38 21
13 20 17.5
Table 3. Performance of an Optimized BPF with UPBGSs
and B-PBGS
Type
Average
3dB IL BW
(%)
Average
10dB
RL BW
(%)
Average
2nd Harmonics
(dB)
Standard BPF 7.18 5.96 6
UPBGS-assisted BPF 16.02 15.2 29
B-BPF (with omission of
central PBG elements)
19.56 16.2 39
that the designs differ from each other with respect to the
number of DGSs only. The designs are as follows:
Design 1. Here we use a total of four DGSs under two
50-O lines. Two DGSs will be located under each 50-O
line. The larger square slot of the DGS unit measures
52 52 mm
2
, and the vertical narrow slot is 20
15 mm
2
. The geometry of a BPF with four DGSs in
the ground plane is shown in Fig. 24.
Design 2. In this case we use four DGSs under only the
left-side 50-O line. There are no DGS elements under
the right-side 50-O line. The geometry is identical to
Fig. 24, with the exception that DGS elements are
located only on the left side rather than on both left
and right.
Design 3. Four DGSs are located under the left 50-O
line designed for second-harmonic suppression. An-
other set of four DGSs is located under the right 50-O
line, which is suitable for third-harmonic suppres-
sion. The RHS (right-hand-side) DGSs are smaller in
size than the LHS DGSs. This is obvious, as they are
designed at higher frequency to suppress third har-
monics. The geometry is shown in Fig. 25.
Design 4. Four DGSs are located under the LHS 50-O
line and eight DGSs are placed under the RHS 50-O
line. The eight DGS elements are designed for sig-
nicant third-harmonic suppression. The geometry is
identical to that in Fig. 25.
2.5.2. Results. Theoretical investigations have been
carried out on DGS-assisted BPFs for effective suppres-
sion of second (at 15 GHz) and third (at 22.5 GHz) harmo-
nics. The S-parameter performances for the various
designs are given below (theoretical performances of the
standard optimized BPF used in all designs are shown in
50 ohm
transmission
line
Ground plane
DGSs in
GND plane
Coupled
line
W1
W2
L
2
W
50
G
2
G
1
L
1
Figure 24. Geometry of a DGS-assisted BPF; four DGSs are
located under two 50-O lines.
50 ohm
transmission
line
DGSs in
GND plane
Ground plane
W
50
L
2
W2
W1
L
1
G
2
G
1
Coupled
line
Figure 25. Geometry of a DGS-assisted BPF having four larger
DGSs under LHS 50-O line and four smaller DGSs under RHS
50-O line.
0 5 10 15 20 25
Frequency (GHz)
50
40
30
20
10
0
S
-
p
a
r
a
m
e
t
e
r
s

(
d
B
)
S11
S21
Figure 26. S-parameter performances of an optimized reference
BPF.
0 5 10 15 20 25
Frequency (GHz)
50
40
30
20
10
0
S
-
p
a
r
a
m
e
t
e
r
s

(
d
B
)
S11
S21
Figure 27. S-parameter performances of a DGS-assisted BPF
having a total of four DGSs (2 2) lying under two 50-O lines.
Fig. 26; the second and third harmonics are 6 and 0 dB,
respectively):
Design 1. The S-parameter performances for four DGS
(22) located under two 50-O lines are shown in
Fig. 27. From the IL performance it can be seen that
the ILs at second and third harmonics are 28 and
6 dB, respectively. Input matching is seen to be im-
proved.
Design 2. The S-parameter performances of a BPF
assisted by four DGSs under the LHS 50-O line are
shown in Fig. 28. It can be seen that this design
provides the ILs at second and third harmonics of 38
and 20 dB, respectively.
Design 3. In this design we have placed four DGSs
under the LHS 50-O line for use in second-harmonic
suppression, and four other DGSs are placed under
the RHS 50-O line, designed for use in third-harmonic
suppression. This is a new technique, in which we
have designed DGSs at two frequencies for effective
harmonic suppression. The S-parameter perfor-
mances of this design are shown in Fig. 29. It can
be seen that this design provides second and third
harmonics as 40 and 20 dB, respectively.
Design 4. In this case we attempted to suppress the
third harmonic signicantly. For this purpose eight
DGSs are used under the RHS 50-O lines to generate
deeper and wider stopbands to suppress the third
harmonics. The S-parameter performances of this
design are shown in Fig. 30. It can be seen that the
suppression at third harmonic has improved signi-
cantly to 30 dB.
3. CONCLUSIONS
We have reviewed the theory of periodic structures to
understand the passband and stopband phenomena of
PBGSs. The dispersion diagram of a modied UCPBG
0 5 10 15 20 25
Frequency (GHz)
50
40
30
20
10
0
S
-
p
a
r
a
m
e
t
e
r
s

(
d
B
)
S11
S21
Figure 28. S-parameter performances of a BPF assisted by four
DGSs lying under the LHS 50-O transmission line.
0 5 10 15 20 25
Frequency (GHz)
50
40
30
20
10
0
S11
S21
Figure 29. S-parameter performances of DGS-assisted BPF
where two sets of four (i.e., total eight) DGSs have been placed
under two 50-O lines.
0 5 10 15 20 25
Frequency (GHz)
50
40
30
20
10
0
S
-
p
a
r
a
m
e
t
e
r
s

(
d
B
)
S11
S21
Figure 30. S-parameter performances of DGS-assisted BPF
where four DGSs are used under the LHS 50-O line and eight
DGSs are designed for third-harmonic suppression and placed
under the RHS 50-O line.
structure is presented, based on the transmission-line
model [23] of PBGSs. We have compared the S-parameter
performances of three rows with those of one row of PBG-
assisted transmission lines. It can be seen that they
provide almost identical performances. Therefore we
have placed 1D PBG elements just under the lines of
BPF for the suppression of higher-order harmonics.
We have theoretically investigated BPF for UPBGSs
with different conditions. From all the results it can be
seen that the performance improves if the PBG elements
are located just under the lines. We have also fabricated
optimized reference BPF and a BPF with UPBGS. It can
be seen that, due to inclusion of PBG elements in a
conventional BPF, the IL performance improves signi-
cantly in terms of passband performance and harmonic
suppression.
We designed binomially distributed PBGSs for the
suppression of spurious transmission in a BPF. A compact
microstrip BPF with intrinsic spurious suppression was
reported [1] in which 2D UCPBGSs are used. 2D
UCPBGSs are more complex and take more space to
describe; thus we have proposed BPBGSs that yield
ripple-free transmission with wider stopband in micro-
strip lines. We investigated IL performances of a BPF with
a variable number of PBG elements. It can be seen that
BPBGSs suppress second harmonics signicantly; they
also suppress third harmonics. It is very interesting to
note that in a microstrip transmission line the stopband
scales with the number of PBG elements. But our present
study reveals that it does not hold true for harmonic
suppression of a BPF. This can be explained as the
coupling phenomenon of the poles of BPF and PBG ele-
ments. The PBG elements situated under the central
coupled line may provide signicant coupling of PBGS
and BPFs and may drastically degrade the passband
performances.
Random usage of PBGS under BPF does not always
lead to descent passband and harmonic suppression. In
most cases, passband performance is drastically distorted;
therefore careful design and investigation is required. In
our investigation we observed that the strong coupling
between the central coupled line of BPF and the central
PBG element under the coupled line destroys the band-
pass performance of BPF. After careful design of elimina-
tion of central PBG elements, the designs with both 10 and
12 PBG elements provide better IL performance.
Finally, we have investigated DGS-assisted BPF. Of the
few designs that we have investigated, all suppress both
second and third harmonics signicantly. A new technique
of using two sets of DGSs designed at two different
frequencies works well in second- and third-harmonic
suppression.
BIBLIOGRAPHY
1. F.-R. Yang, K.-P. Ma, Y. Qian, and T. Itoh, A uniplanar
compact photonic-bandgap (UC-PBG) structure and its appli-
cations for microwave circuits, IEEE Trans. Microwave
Theory Tech. 47(8):15091514 (Aug. 1999).
2. Y. Qian, D. Sievenpiper, V. Radisic, Y. Yablonovitch, and T.
Itoh, A novel approach for gain and bandwidth enhancement
of patch antenna, Proc. Radio and Wireless Conf., 1998
(RAWCON98), 1998.
3. Y. Qian, R. Coccioli, D. Sievenpiper, V. Radisic, E. Yablono-
vitch, and T. Itoh, Microstrip patch antenna using novel
photonic band-gap structures, Microwave J. 42(1):6676
(Jan. 1999).
4. R. Coccioli, F.-R. Yang, K.-P. Ma, and T. Itoh, Aper-
ture-coupled patch antenna on UC-PBG substrate,
IEEE Trans. Microwave Theory Tech. 47(11):21232130
(Nov. 1999).
5. R. Gonzalo, P. D. Maagt, and M. Sorolla, Enhanced patch
antenna performance by suppressing surface waves using
photonic-bandgap substrates, IEEE Trans. Microwave Theory
Tech. 47(11):21312138 (Nov. 1999).
6. K. M. Shum, Q. Xue, C. H. Chan, and K. M. Luck, Investiga-
tion of microstrip reectarray using a photonic bandgap
structure, Microwave Opt. Technol. Lett. 28(2):114116 (Jan.
20, 2001).
7. P. S. Hui and A. Alphones, Microstrip patch antenna with
annular ring PBG, Proc. Asia Pacic Microwave Conf. 2000,
Sydney, Australia, Dec. 2000, pp. 13471351.
8. R. Gonzalo, B. Martinez, P. D. Maagt, and M. Sorolla, Patch
antennas on photonic crystal structures (online), http://
www.ecs.umass.edu/ece/allerton/99/34/.
9. Y. Qian, R. Coccioli, F.-R. Yang, and T. Itoh, Passive and active
component design using PBG, Proc. 6th Int. Conf. Terahertz
Electronics, 1998, pp. 4245.
10. F.-R. Yang, Y. Qian, and T. Itoh, A novel uniplanar compact
PBG structure for lter and mixer applications, IEEE MTT-S
(Microwave Theory and Techniques Symp.) Int. Digest, Ana-
heim, CA, June 1319, 1999, Vol. 3, pp. 919922.
11. E. R. Brown, C. D. Parker, and E. Yablonovitch, Radiation
properties of a planar antenna on a photonic-crystal sub-
strate, J. Opt. Soc. Am. B. Opt. Phys. 10(2):404407 (Feb.
1993).
12. A. Harvey, Periodic and guiding structures at microwave
frequencies, IRE Trans. Microwave Theory Tech. 8:3061
(June 1959).
13. D. Nesic and A. Nesic, Bandstop microstrip PBG lter with
sinusoidal variation of the characteristic impedance and
without etching in the ground plane, Microwave Opt. Technol.
Lett. 29(6) (June 20, 2001).
14. S. K. Sharma and L. Shafai, Enhanced performance of an
aperture-coupled rectangular microstrip antenna on a sim-
plied unipolar compact photonic bandgap (UC-PBG) struc-
ture, IEEE-APS (Antennas and Propagation Symp.), Boston,
July 813, 2001, Vol. 2, 498501.
15. N. C. Karmakar and M. N. Mollah, Investigation into non-
uniform photonic bandgap microstripline lowpass lters,
IEEE-Trans. Microwave Theory Tech. 51(2):564572 (Feb.
2003).
16. V. Radisic et al., Novel architectures for high-efciency
ampliers for wireless applications, IEEE Trans. Microwave
Theory Tech. 46(11):19011909 (Nov. 1998).
17. C. Y. Hang, W. R. Deal, Y. Qian, and T. Itoh, High efciency
transmitter front-ends integrated with planar antennas and
PBG, Proc. 2000 Asia-Pacic Microwave Conf., Australia.
18. N. C. Karmakar and M. N. Mollah, Harmonic suppression of a
bandpass lter using binomially distributed photonic band-
gap structures, Proc. IEEE-APS, Colombus, OH, July 2227,
2003, pp. 883886.
19. D. Ahn, J.-S. Park, C.-S. Kim, J. Kim, Y. Qian, and T. Itoh, A
design of the low-pass lter using novel microstrip defected
ground structure, IEEE Trans. Microwave Theory Tech.
49(1):8693 (Jan. 2001).
20. M. N. Mollah and N. C. Karmakar, A novel hybrid defected
ground structure as low pass lter, Proc. IEEE APS-S Int.
2004, Monterey, CA, June 2026, 2004.
21. D. M. Pozar, Microwave Engineering, Addison-Wesley, Read-
ing, MA, 1990.
22. J. D. Shumpert, Modeling of Periodic Dielectric Structures
(Electromagnetic crystal), Ph.D. dissertation, Univ. Michigan,
2001.
23. M. Rahman and M. A. Stuchly, Transmission line-periodic
circuit representation of planar microwave photonic bandgap
structures, Microwave Opt. Technol. Lett. 30(1) (July 5, 2001).
24. I. J. Bahl and P. Bhartia, Microwave Solid State Circuit
Design, New York, Wiley, 1998.
25. G. L. Larkins, R. Socorregut, and Y. A. Vlasov, Superconduct-
ing microstrip hairpin lter with BaTiO
3
patches, IEEE
Trans. Appl. Superconduct. 13(2):724726 (June 2003).
26. C.-M. Tsai, S.-Y. Lee, and C.-C. Tsai, Hairpin lters with
tunable transmission zeros, 2001 IEEE MTT-S Int. Micro-
wave Symp. Digest, May 2001, Vol. 3, pp. 21752178.
27. R. R. Bonetti, and A. E. Williams, Preliminary design steps
for thin-lm superconducting lters, IEEE MTT-S Int. Micro-
wave Symp. Digest, May 810, 1990, Vol. 1, pp. 273276.
28. Y. Di, P. Gardner, P. S. Hall, H. Ghafouri-Shiraz, and J. Zhou,
Multiple-coupled microstrip hairpin-resonator lter, IEEE
Microwave Wireless Compon. Lett. 13(12):532534 (Dec. 2003).
29. A. N. Deleniv, M. S. Gashinova, I. B. Vendik, and A. Eriksson,
Design of an interdigital hairpin bandpass lter utilizing a
model of coupled slots, IEEE Trans. Microwave Theory Tech.
50(9):21532158 (Sept. 2002).
30. G. L. Matthaei, N. O. Fenzi, R. J. Forse, and S. M. Rohng,
Hairpin-comb lters for HTS and other narrow-band applica-
tions, IEEE Trans. Microwave Theory Tech. 45(8):12261231
(Aug. 1997).
31. G. L. Matthaei, N. O. Fenzi, R. Forse, and S. Rohlng,
Narrow-band hairpin-comb lters for HTS and other applica-
tions, IEEE MTT-S Int. Microwave Symp. Digest, 1996, Vol. 2,
pp. 457460.
32. Y. Di, P. Gardner, P. S. Hall, and H. Ghafouri-Shiraz, Multi-
ple-coupled microstrip hairpin-resonator lter, Proc. 3rd Int.
Microwave and Millimeter Wave Technology Conf., 2002, pp.
10511054.
33. V. Radisic, Y. Qian, R. Coccioli, and T. Itoh, Novel 2-D
photonic bandgap structures for microstrip lines, IEEE
Microwave Guided Wave Lett. 8(2):6971 (Feb. 1998).
34. P. J. Schields, Braggs Law and Diffraction, Center for High
Pressure Research, State Univ. New York, Stony Brook, NY;
available online at http://www.eserc.stonybrook.edu/Project-
Java/Bragg/.
35. M. N. Mollah and N. C. Karmakar, Effect of number of
uniform photonic bandgap (U-PBG) elements on the perfor-
mance of microstrip transmission lines, Proc. PACRIM-2003,
Victoria, Canada.
36. M. N. Mollah and N. C. Karmakar, Optimum lling factor and
performance of a transmission line at X-band perturbed by
binomially distributed photonic bandgap (PBG) structure,
Proc. PACRIM-2003, Victoria, Canada.
37. R. L. L. Ling, Improved Performance of Planar Filters on
Photonic Bandgaps, M.Sc. thesis, Nanyang Technological
Univ., Singapore, May 2004.
ELECTROMAGNETIC COMPATIBILITY
CLAYTON R. PAUL
Mercer University
Macon, Georgia
1. INTRODUCTION
Electromagnetic compatibility (EMC) concerns the reduc-
tion of interference in electronic systems. For a more com-
plete discussion of EMC, see Ref. 1. All gures in this
article are taken from that reference. Interference has oc-
curred when an electromagnetic emission causes the un-
intended operation of a device. A typical interference
scenario consists of a source (or emitter), a transfer (or
coupling) path, and a receptor (or receiver) as shown in
Fig. 1. Hence there are three ways to reduce the interfer-
ence in the receptor: (1) reduce the level of the source, (2)
increase the attenuation of the transfer path, and (3) re-
duce the susceptibility of the receptor. Generally the op-
timal procedure is to work from left to right, that is, rst
reduce the level of the source, then increase the attenua-
tion of the transfer path, and nally, if needed, reduce the
susceptibility of the receptor. The source and the receptor
may be intended or unintended. Examples of intended
sources are the emissions from an FM radio transmitter or
an airport surveillance radar. Arcing at the brushes of a
DC motor and the radiated emissions from a digital com-
puter are examples of unintended emissions. Generally,
the transfer path is either through air or via metallic con-
duction as in wires. An unintended receptor for the energy
might be a television set that has received the emissions
from an digital computer. In the case of unintended sourc-
es such as the clock in a digital computer, the levels of
emission can be reduced, as we will see, by increasing the
rise/falltimes of the clock pulses. This serves to reduce the
levels of the high-frequency components without affecting
the functionality of the computer. The levels of intended
sources such as an FM radio transmission can rarely be
reduced. Shielding and ltering are common ways of
increasing the attenuation of the transfer path. Separat-
ing the source and receptor, when possible, also accom-
plishes this. In some cases the susceptibility of the
receptor can be reduced by use of error correcting codes.
Generally interference occurs through radiated emis-
sions or conducted emissions. Susceptibility also occurs
through these two mechanisms. Hence there are four basic
EMC subproblems shown in Fig. 2. These are radiated
emissions, radiated susceptibility, conducted emissions,
and conducted susceptibility. It should be pointed out
that time-varying currents radiate. Hence currents,
whether on cables or on metallic cabinets cause radiated
Figure 1. The basic decomposition of the EMC coupling problem.
1136 ELECTROMAGNETIC COMPATIBILITY
emissions. Typically long cables in the systems that carry
high-frequency noise currents are the primary mecha-
nisms for radiated emissions. These radiated emissions
may cause interference with adjacent electronic systems.
External sources such as the emissions from an FM radio
transmitter or an airport surveillance radar may impinge
on an electronic product. These emissions induce unin-
tended currents on the metallic cabinets or cables of the
systems that may create interference with the system.
This is referred to as a radiated susceptibility problem.
The power cord of an electronic device carries, in addi-
tional to the necessary 60-Hz power signal, other higher-
frequency signals. For example, the harmonics of a digital
clock in a product frequently show up on the power cord.
These noise signals are then conducted onto the power
net of the installation. The power net of an installation is
in essence a large antenna in that the phase wires, and the
neutral wires are hardwired together. When high-fre-
quency noise is conducted out of the product via the power
cord and placed on this large power net, they then radiate
which can cause interference. A frequent example of this
is the interference with a TV set caused by the arcing at
the commutator of a motor as in a vacuum cleaner. These
noise signals, in addition to radiating, also proceed out the
power cord of the vacuum cleaner and hence onto the
power net. These are examples of conducted emissions.
And nally high-frequency signals are conducted along
wires and the metallic cabinets of a device. Hence they
may appear in unintended places within the device caus-
ing interference. A common such case is where noise on
the AC power net is conducted into the product via the
device power cord. Switching on a motor such as in an air
conditioner somewhere in the building will induce noise
on the power net, which can be conducted into a product
via the power cord. Lightning during thunderstorms
induces signals on the power transmission lines. These
also may be conducted into the product, causing interfer-
ence. These scenarios constitute conducted susceptibility
problems.
In addition to these primary concerns in EMC, there
are also other aspects of EMC shown in Fig. 3. Electro-
static discharge or ESD is becoming a frequent mecha-
nism for causing intereference in digital systems. When a
human wearing rubber-soled shoes walks across a nylon
carpet, charge is deposited on the persons body. When the
person approaches a digital device such as a computer
keyboard, this accumulation of charge on the nger cre-
ates an intense electric eld that may be of sufcient mag-
nitude to break down the air, resulting in an intense
discharge much like a miniature lightning discharge.
This discharge can cause improper operation of the device
such as loss of data or permanent destruction of some of
the electronic components in the device. The detonation of
a nuclear device creates an intense radiated electric eld
that can damage or destroy electronic systems. This is re-
ferred to as an electromagnetic pulse (EMP). Even though
a communication center does not receive direct blast dam-
age, this intense electromagnetic eld can damage the
communication electronics and prevent a retaliatory
action. Lightning occurs frequently and can result in large
Figure 2. The four basic EMC subproblems: (a) radiated
emissions; (b) radiated susceptibility; (c) conducted emis-
sions; and (d) conducted susceptibility.
ELECTROMAGNETIC COMPATIBILITY 1137
current channels. Although a direct strike can cause se-
vere damage, indirect strikes can also induce damaging
currents in electronic devices. And nally, it has been
shown that it is possible to indirectly determine informa-
tion being transmitted by monitoring the electromagnetic
emissions. Communication security is obviously of ex-
treme importance in the military context, but it is also
important in the business world in order to protect trade
secrets.
2. REGULATORY REQUIREMENTS FOR
ELECTRONIC SYSTEMS
In the United States, the Federal Communications Com-
mission (FCC) has regulatory authority over wire and ra-
dio communications, and has the responsibility for
limiting interference with communications. Around
1979, the personal computer was being introduced to con-
sumers in the United States. Residents of apartments be-
gan complaining to the FCC about interference being
caused in their television receivers by a personal comput-
er in an adjacent apartment. In order to limit the
interference potential of these personal computers, the
FCC instituted regulations in Title 47 of the Code of Fed-
eral Regulations, Part 15, Subpart J which limited the
radiated and conducted (exiting the power cord) emissions
of all digital devices that are intended to be sold in the
United States. (Look on the back of your calculator, per-
sonal computer, or laser printer to see the FCC compliance
notice.) The FCC denes a digital device rather broadly
as Any unintentional radiator (device or system) that gen-
erates and uses timing pulses at a rate in excess of 9000
pulses (cycles) per second and uses digital techniques.
This definition is rather broad and covers, in addition to
personal computers, virtually anything digital such as
electronic typewriters and laser printers. These regula-
tions prescribe limits on the maximum radiated and con-
ducted emissions of a digital device. The conducted
emissions are dened as those exiting the power cord of
the digital device, and the frequency of the regulation
ranges from 150kHz to 30MHz. The radiated emission
limits extend from 30MHz to well over 1GHz, with this
upper limit depending on the fth harmonic of the system
clock. See Ref. 1 for a discussion of these specific limits. For
example, for a digital device intended for residential use
(as opposed to business use), the limit on noise current ex-
iting the power cord is on the order of 5mA, whereas the
limit on the electric eld of radiated emissions is on the
order of 100500mV/m when measured at a distance of 3m.
It is very important to emphasize that if the conducted
or radiated emissions of a digital device exceed the FCC
limits at any frequency within the regulation frequency
range, it cannot be legally sold in the United States! The
FCC has the statutory authority for imposing nes and/or
jail sentences if anyone intentionally violates the regula-
tion. Hence, even though a company develops, for exam-
ple, a revolutionary device (e.g., a laser printer that has
the dimensions of a laptop computer, prints 50 pages per
minute in color, and can be sold for under $50 U.S.), this is
all for naught if the device fails to meet the FCC limits.
Even if the device fails the FCC limits, additional sup-
pression measures can be taken to bring those emissions
into compliance. However, adding the suppression mea-
sures such as lters and shielding will add to the cost of
the product yet not enhance its functionality and hence
will affect its marketability. Therefore, one of the most
important design goals of electronic products today is to
design the system to comply with the FCC limits.
Virtually all other countries in the world have similar
and sometimes more stringent limits on the emissions of
digital products. Hence EMC is a global concern. For ex-
ample, prior to the institution of the FCC limits in the
United States, most countries in Europe had individual
limits on radiated and conducted emissions of digital de-
vices that were similar to those instituted later in the
United States. Most of these limits were those developed
by CISPR (the French translation meaning International
Special Committee on Radio Interference). CISPR is a
committee of the International Electrotechnical Commis-
sion (IEC), an international body that promulgates
standards in order to facilitate trade between countries.
CISPR published a recommended standard known as
CISPR 22 that concerns information technology equipment
Figure 3. Other aspects of EMC: (a) electrostatic discharge
(ESD); (b) electro-magnetic pulse (EMP); (c) lightning; (d) TEM-
PEST (secure communication and data processing).
(ITE) which also includes digital devices. Many countries
in Europe and throughout the world have adopted the
CISPR 22 standard. The CISPR 22 limits are quite similar
to the FCC limits. In recent years many countries in Eu-
rope have come together to form the European Economic
Community or European Union in order to facilitate trade
between the member nations. They have essentially adopt-
ed the CISPR 22 standard for radiated and conducted
emissions of digital devices that are to be sold in those
countries. This is called a European Norm, and the stan-
dard is EN 55022. Hence, complying with limits on radi-
ated and conducted emissions of digital devices is a
worldwide concern. In addition, the European standards
impose additional regulatory requirements on digital de-
vices that limit the susceptibility of a digital device to elec-
tromagnetic emissions. This requirement has not yet been
imposed by the FCC but has created considerable concern
for manufacturers of digital products that are intended for
sale in Europe.
The measurement of these levels must be conducted by
strict rules that are described in related standards. The
measurement of radiated emissions is generally conducted
in a semianechoic chamber as illustrated in Fig. 4. A
shielded room encloses the test and prevents external
emissions from contaminating the test results. The walls
are lined with materials (carbon-impregnated foam or
more recent ferrite tiles) that absorb electromagnetic
waves in the frequency range of the regulations. This is
done to prevent reections from the shielded room walls
and hence to simulate free space. The oor is a ground
plane; hence the prex semi in the chamber name. The
conducted emissions are noise currents exiting the power
cord of the device. These are routinely measured with a
line impedance stabilization network (LISN) as shown in
Fig. 5. The digital product under test has its power cord
plugged into the LISN, and the LISN is plugged into the
commercial power outlet. The purposes of the LISN are
twofold:
1. The impedance seen between phase and ground and
between neutral and ground of the commercial pow-
er net over the regulatory frequency range of
150kHz30 MHz varies widely; thus the rst pur-
pose of the LISN is to present a constant 50 O im-
pedance between phase and ground and between
neutral and ground over the regulatory frequency
range.
2. The second purpose of the LISN is to prevent noise
on the commercial power net from contaminating
the test. Inductors L1 are present to block this, and
capacitors C2 are present to divert this noise to
ground. Of course, the LISN must also pass the
60 Hz commercial power to the product. At 60 Hz,
L1 is essentially a low impedance and C2 is essen-
tially a high impedance. Hence the LISN is trans-
parent to the 60 Hz power but appears as shown in
Fig. 6 at the regulatory frequencies of 150kHz
30 MHz.
These EMC regulatory requirements cannot be waived.
If a digital device fails to meet these limits at any fre-
quency, it cannot be legally sold in that country! There are
certain other EMC requirements that can be relaxed if
deemed necessary. These are for military products, com-
mercial aircraft and automobiles, and other commercial
Figure 4. Illustration of the use of a semian-
echoic chamber for the measurement of radi-
ated emissions.
vehicles. In the United States, in the early 1960s the
Department of Defense (DoD) instituted a broad range of
EMC requirements concerning, among other things, limits
on the radiated and conducted emissions from virtually all
electronic products that are sold to DoD units (U.S. Army,
Navy, Coast Guard, etc.) These are military standards and
the EMC regulation is MIL-STD-461E. The frequency
range of regulation of the emissions is drastically greater
than the FCC or CISPR 22 regulations, extending in some
cases from 30Hz to 40 GHz. However, the project ofcer
for the contract has the authority to tailor these limits to
t the particular requirements of the system in which the
electronic device is to be installed. In addition to the mili-
tary, commercial aircraft manufacturers in the United
States voluntarily comply with a standard on radiated
and conducted emissions of commercial aircraft known as
DO-160D. These voluntary standards prescribe limits on
radiated and conducted emissions (on cables) as well as
susceptibility in order to prevent or reduce the potential
for interference with critical ight systems on the aircraft.
Manufacturers of commercial automobiles and trucks in
the United States also voluntarily comply with a standard
such as SAE J551-2, which provides, among other things,
limits on the radiated and conducted emissions of the
vehicle as well as its susceptibility to external electromag-
netic emissions. Again these limits are intended to prevent
or reduce the potential for interference with electronic
systems on the vehicle from either internal or external
sources. Once again, these are voluntary but manufactur-
ers generally comply because introduction into the market
of an airplane or vehicle that is susceptible to everyday
emissions from radio towers and other ambient emitters
that the airplane or vehicle routinely encounters would
create a devastating loss of condence with consumers.
3. TIME-DOMAIN VERSUS FREQUENCY-DOMAIN
DESCRIPTION OF DIGITAL SIGNALS
The abovementioned regulatory requirements on digital
devices are given in the frequency domain, that is, at each
frequency within a frequency range. Hence it is important
to investigate the relation between the time-domain wave-
forms of digital signals and their spectral content. The
primary waveform in a digital device is its clock(s) signals.
These are periodic, pulses with period T that are approx-
imately trapezoidal in shape and have risetimes and
Figure 6. Equivalent circuit of the LISN as seen by the
product over its intended frequency range of use.
Figure 5. The line impedance stabilization
network (LISN) for the measurement of con-
ducted emissions.
falltimes of t
r
and t
f
, respectively, as illustrated in Fig. 7.
Typical clock signals have equal risetimes and falltimes or
approximately so. Digital data signals have a similar
waveform but are random in occurrence as opposed to
the deterministic clock signal. Since the clock is periodic,
we may decompose it into its spectral components with the
Fourier series, giving the representation [1]
xt X
0
X
1
cos o
0
t y
1
X
2
cos 2o
0
t y
2
X
3
cos 3o
0
t y
3

1
Hence, the signal consists of the sum of a DC or average
value X
0
, plus an innite number of harmonics having
magnitude X
n
and phase y
n
and frequencies no
0
, where
o
0
(2p/T) and the fundamental (repetition) frequency is
f
0
(1/T). For the trapezoidal waveform in Fig. 7 having
equal risetimes and falltimes, the magnitudes of these
spectral components are [1]
X
n
2AD
sinnpD
npD
sinnpt
r
f
0
npt
r
f
0
2
where the duty cycle is
D
t
T
tf
0
3
and the pulsewidth between 50% points is t.
In order to facilitate an understanding of how the time-
domain properties affect the spectral content, it is possible
to obtain spectral bounds on the magnitude as shown in
Fig. 8 [1]. This Bode plot shows that the magnitudes of the
spectral components fall off with increasing frequency as
0 dB/decade out to the rst breakpoint, which occurs at
1/pt f
0
/pD. Above this breakpoint the levels fall off as
20 dB/decade up to the next breakpoint, which is 1/pt
r
.
Above this the levels fall off as 40 dB/decade. Hence, to re-
duce the high-frequency spectral content, we must increase
the pulse rise/falltimes. This is a primary method of re-
ducing the source. Figure 9 shows a comparison between
the exact levels and the spectral bounds for a 1-V, 1-MHz
clock having a rise/falltime of 20 ns, a 50% duty cycle, and
a 30% duty cycle. Figure 10 shows a spectrum analyzer
display of a 1-V, 10-MHz, 50% duty cycle digital clock sig-
nal having rise/falltimes of 20 and 5 ns illustrating the
considerable reduction of the high-frequency spectral con-
tent achieved by increasing the rise/falltimes of the pulse.
4. METHODS FOR REDUCING EMISSIONS
FOR COMPLIANCE
4.1. Radiated Emissions
We rst describe methods for reducing radiated emissions.
Again, time-varying currents radiate whether they are on
cables or metallic cabinets. However, the predominant ra-
diated emissions for digital products seem to emanate
from wires or printed circuit board (PCB) traces in the
system. The PCB traces are also referred to as lands be-
cause of their rectangular cross sections. There are two
possible types of current. Consider a two-conductor (wire
or PCB land) transmission line as shown in Fig. 11. Dif-
ferential-mode currents are the usual functional currents
on these conductors, and these phasor currents are denot-
ed as
^
II
D
in Fig. 11a. At a line cross section these are equal
in magnitude but opposite in phase. These are the func-
tional or desired currents that are designed for. If we view
each conductor current as a Hertzian dipole (assuming
that the conductor is electrically short and the observation
point is in the far eld at this frequency), then the radi-
ated electric elds,
^
EE
1
and
^
EE
2
, subtract. Since the two cur-
rents are not collocated, they do not cancel. Conversely,
there also exist common-mode currents on the conductors
and these phasor currents are denoted as
^
II
C
. At a cross
section of the line the two currents have the same magni-
tude and are in phase as shown in Fig. 11b. Again viewing
these as Hertzian dipoles, we see that their radiated elec-
tric elds add. Common-mode currents are sometimes re-
ferred to as antenna-mode currents. One way of viewing
Figure 7. The periodic, trapezoidal pulse-
train representing clock and data signals of
digital systems.
Figure 8. Bounds on the one-sided magnitude spectrum of a
trapezoidal pulsetrain.
them is as having a return path to their source consisting
of, for example, parasitic capacitances to a third adjacent
conductor (which is always present). The differential-
mode currents are easily calculated, but the common-
mode currents are not easily calculated (if at all). The
common-mode currents are usually measured with a cur-
rent probe [1]. The important observation here is that,
because the differential-mode current emissions subtract
but the common-mode emissions add, a common-mode
current of magnitude much smaller than that of the dif-
ferential-mode current can cause equal or greater emis-
sion levels. Radiated emissions from common-mode
currents can be devastating in terms of radiated emis-
sions and tend to be one of the major causes (if not the
sole cause) of radiated emission and product noncompli-
ance with the regulations for digital products today. For
example, one can compute that for a 1m length of ribbon
cable (wire separation of 50 mils or 1.27mm) a level of
differential-mode current at 30 MHz of some 20 mA will
cause radiated emissions at 3m that will exceed the FCC
limit, whereas only 8mA of common-mode current will
cause the same level of emission at 30 MHz.
Suppressing the emissions of these two currents
requires different techniques. In the case of differential-
mode currents, capacitors can be placed between the two
wires to divert the current and keep it off the cable. In
general, the longer the cable, the greater the potential for
the current to radiate. The value of the capacitor is chosen
to give a relatively low impedance (relative to the cable
termination impedance) at the frequency of the emission
current. In addition to the technique of current diversion,
as with a parallel capacitor, there is the technique of cur-
rent blocking where inductors or ferrite beads are inserted
in series with each line to block the differential-mode cur-
rents of each line. In this case the value of the series im-
pedance must be chosen to give a large impedance
(relative to the cable termination impedance). It must be
remembered that these differential-mode currents are re-
quired for the product to operate properly. Hence these
suppression measures must be chosen with care so that
these intended currents are not corrupted to the point that
functionality of the product is impaired.
Suppressing the radiated emissions of common-mode
currents requires a different technique. The common tech-
nique is to use common-mode chokes as illustrated in
Fig. 12. A common-mode choke can take many forms. We
have shown a toroid where the two cable wires are wound
the same direction. In the case of differential-mode cur-
rents as shown in Fig. 12b, the magnetic uxes of the dif-
ferential-mode currents tend to cancel, and hence the
choke is transparent to these currents. However, the mag-
netic uxes of the common-mode currents add in the core,
and hence the common-mode currents see a net imped-
ance inserted in each line. These chokes are constructed of
ferrite materials such as manganesezinc or nickelzinc
materials with additional materials added to produce
high-frequency losses. In addition to inserting inductanc-
es selectively in the lines as shown in Fig. 12c, these
ferrite chokes also insert high-frequency resistances (not
shown) and provide selective loss, thus dissipating the en-
ergy in the common-mode currents. There are also chokes
consisting of a ferrite core that is clamped around both
wires such as is frequently seen in personal computer sys-
tems on the keyboard and video cables.
We now will derive a simple model for these emissions
that will illustrate the factors that can be controlled in
order to reduce their emissions. First consider the case of
differential-mode currents on the two wires as shown in
Fig. 13, where the wires are separated by a distance s and
are a total length L. If we assume that the line length is
much less than a wavelength and the observation point is
in their far eld (typically a wavelength or more), then we
may view these as Hertzian dipoles and simply superim-
pose their elds. The resulting maximum electric eld is
then [1]
j
^
EE
D;max
j 1:31610
14
j
^
II
D
jf
2
Ls
d
4
Figure 9. Examples illustrating the spectral bounds for various
duty cycles of a 1-V, 1-MHz, trapezoidal pulsetrain having rise/
falltimes of 20ns: (a) 50% duty cycle; (b) 30% duty cycle.
where d is the distance to the observation point. For res-
idential digital devices in the United States, this is the
regulatory measurement distance of 3 m. At 30 MHz, the
FCC limit is 100mV/m at 3m. We can calculate from this
that a level of about 20 mA on a 1m length of ribbon cable
at this frequency will just exceed this limit. The reason we
derive a simple result such as in (4) is not so much to cal-
culate emission levels but more importantly to determine
the factors that control the emission. For example,
(4) shows that the factors affecting the level of the radiat-
ed eld are (1) the level of the current (the level of the
Fourier component of this current) at this frequency, (2)
the frequency (squared), and (3) the loop area formed by
the conductors, ALs. The level of the current (perhaps
a high-frequency component of a clock signal) can be re-
duced by increasing pulse rise/falltimes. These can be re-
ducing only up to a point where the functionality of the
product becomes impaired. Usually there is considerable
latitude in doing this in the original design. The frequency
here is the frequency of the emission (and the regulation),
which cannot be changed. The nal item, loop area formed
by the two conductors can be easily controlled. Two cases
are illustrated in Fig. 14. Figure 14a shows a digital clock
that serves an application-specific integrated circuit
(ASIC). The loop area of the two connection leads must
be kept to a minimum. This can be easily done by (1) plac-
ing the clock as close as possible to the device being served
and (2) placing the two connection leads as close together
as possible. In the layout of a PCB, this would be the initial
order of business. Figure 14b shows a ribbon cable at-
tached to a stepper motor. The pin assignments in the
connectors are as shown. The prime items, such as A
0
,
are the returns for the currents on the unprimed wires,
such as A. Because of the pin assignments, each return
Figure 10. Experimentally measured spectra of 1-V, 10-MHz, 50% duty cycle trapezoidal pulse-
trains for rise/falltimes of (a) 20ns; (b) 5 ns.
Figure 11. Illustration of the relative radiated emission poten-
tial of (a) differential-mode currents and (b) common-mode cur-
rents.
encompasses three loop areas between wires. If the pin
assignments had been as shown in Fig. 14c, then the in-
dividual loop areas would be reduced by one-third and the
radiated emissions would be, according to the model in
Fig. 14, reduced by about 10 dB. Simple understanding of
the basic mechanisms will often lead to cost-free EMC
suppression.
Next we consider a simple model for the emissions from
common-mode currents shown in Fig. 15. Once again con-
sidering the two conductors as Hertzian dipoles and
superimposing their elds gives [1]
j
^
EE
C;max
j 1:25710
6
j
^
II
D
jf L
d
5
For a 1 m length of ribbon cable at 30 MHz, we can calcu-
late that a level of common-mode current of some 8 mAwill
cause an electric eld at the regulatory measurement dis-
tance of 3 m that exceeds the FCC limit. Hence very small
levels of common-mode currents can be devastating from
the standpoint of complying with the regulatory limits. In
(5) we observe that the two factors that control these
Figure 12. Modeling the effect of a common-mode choke on
(a) the currents of a two-wire line, (b) the differential-mode com-
ponents, and (c) the common-mode components.
Figure 13. A simplied estimate of the maximum radiated emis-
sions due to differential-mode currents with constant distribution.
Figure 14. Common mistakes that lead to unnecessarily large
differential-mode emissions: (a) large clock land areas; (b,c) choic-
es of connector pin assignments in ribbon cables to minimize loop
area.
Figure 15. A simplied estimate of the maximum radiated emis-
sions due to common-mode currents with constant distribution.
emission levels are (1) the level of the current and (2) the
length of the cable. Reducing the levels of the differential-
mode currents tends to also reduce the levels of the com-
mon-mode currents, although the precise relationship is
not clear. Also, we should endeavor to make all connec-
tions (wires and PCB lands) as short as possible.
In order to illustrate the devastating nature of com-
mon-mode currents, we investigate the simple experiment
shown in Fig. 16. Two PCB lands of width 25 mils
(0.635 mm) and length 6 in (15.2 cm) are mounted on a
typical PCB substrate. A 10-MHz dual-inline package
(DIP) oscillator is powered by a 5-V battery (no connec-
tion to the commercial power system) and is connected to
the two lands. The two lands are terminated in a 330-O
resistor. The characteristic impedance of this line is about
330 O, and hence the line is matched. The radiated electric
elds measured in a typical semianechoic chamber at 3 m
are shown in Fig. 17. On these measured emissions are
superimposed the predictions of the common-mode model
in (5) denoted as x. The common-mode currents used in
this model were measured with a current probe. In addi-
tion, the emissions of the differential-mode currents were
measured on each individual wire and also predicted with
a transmission-line model. The emissions computed with
the measured differential-mode currents with (4) are
shown as o and the emissions computed with the com-
puted differential-mode currents with (4) are shown as
. These results show that (1) the emissions of common-
mode currents greatly exceed those from differential-mode
currents and are dominant, and (2) the simple models in
(4) and (5) are sufciently accurate to illustrate the factors
that affect these emissions.
4.2. Conducted Emissions
Now we turn to the suppression of conducted emissions.
As stated earlier, the standard regulatory test for con-
ducted emissions of a digital product is to plug the prod-
ucts power cord into a LISN and the LISN into the
commercial power outlet. Again the noise currents
(150 kHz30 MHz) exiting the phase and neutral conduc-
tors of the power cord consist of differential-mode and
common-mode currents. The differential-mode currents
ow out of the phase conductor and return on the neutral
conductor. The common-mode currents ow out of the
phase and neutral conductors and return on the safety
or green wire of the product. The primary method of
suppressing these emissions is with a power supply
lter shown in Fig. 18. Ordinarily the power supply lter
is at the last point of exit of the power cord from the
product. The important components in the lter are (1) a
Figure 16. An experiment illustrating common-mode currents
on PCBs: (a) device schematic; (b) PCB cross-sectional dimen-
sions.
Figure 17. Measured and predicted emis-
sions of the device of Fig. 16.
common-mode choke denoted by self- and mutual induc-
tances L and M, (2) line-to-line capacitors denoted as C
DR
and C
DL
, and (3) line-to-ground capacitors denoted as C
CR
and C
CL
. The subscripts L and R denote left and right,
respectively. The common-mode choke acts as described
before; it selectively blocks the common-mode noise cur-
rent and is transparent to the differential-mode noise
current. It is necessary to use a common-mode choke rath-
er than individual inductors here since the differential-
mode current contains the important 60 Hz power current.
The 60 Hz power current has a very large magnitude, on
the order of 110A in typical digital products and hence
would saturate the ferrite cores of individual inductors.
The common-mode choke is transparent to the high-level
60 Hz current, and hence the core of the common-mode
choke is not saturated by it. The line-to-line capacitors are
present to divert the differential-mode noise currents, and
the line-to-ground capacitors are present to divert the
common-mode noise currents.
4.3. Shielding
Another suppression measure is obtained with shielding.
Generally, shielding should be used as a last resort for
suppression since (1) it is difcult to implement in its ideal
form (shielding effectiveness values of 4200dB can easily
be calculated but are impossible or too costly to imple-
ment), and (2) shielding is generally quite expensive com-
pared to other simple methods. A shield is a contiguous
metal enclosure. There are two purposes for shields as il-
lustrated in Fig. 19. One is to keep internal elds from
being radiated (possibly causing noncompliance with ra-
diated emission levels) and the other is to keep external
elds from interfering with internal electronics (prevent-
ing susceptibility to, for example, a FM radio transmitter
or an airport surveillance radar).
The simplest model of a shield is a large, conducting
plane of conductivity s and thickness t as shown in Fig. 20.
The most common measure of shielding effectiveness (SE) is
the ratio of the electric eld that is incident on the shield to
the electric eld that exits the shield. In decibels this is [1]
SE
dB
20 log
10
E
incident
E
transmitted
6
This gives the shielding effectiveness as a positive value. For
example, a SE of 100dB means that the electric eld is re-
duced by a factor of 100,000 as it transits the shield wall. If
we assume that (1) the shield is in the far eld of the source
(such as a radar) so that the incident eld is a uniformplane
wave and (2) the incident wave strikes the shield wall nor-
mal to its surface, then a simple result for the shielding ef-
fectiveness can be obtained as [1]
SE
dB
20 log
10
Z
0
4^ ZZ
..
R
dB
20 log
10
e
t=d
..
A
dB
7
Figure 18. A typical power supply l-
ter topology.
Figure 19. Illustration of the use of a
shielded enclosure to (a) contain radiated
emissions and (b) exclude radiated emis-
sions.
where the intrinsic impedance of free space is Z
0
377 O,
the intrinsic impedance of the conducting shield is
^ ZZ
jom=sjoe
_
, and d 1=
pf ms
_
is the skin depth.
The rst term in (7) is said to be the reection loss and ac-
counts for the reection at the left and right interfaces of the
shield walls. The second term is said to be the absorption
loss and accounts for attenuation of the waves as they pass
through the shield. This absorption term can be written as
A
dB
8:69t=d. Hence a shield thickness of one skin depth
gives B9dB of absorption loss, and a shield thickness of ve
skin depths gives B44dB.
From this result one can compute shielding effective-
ness values on the order of 100500 dB. Rarely will these
be realized in actual systems because of the inevitable
penetrations of the shield by openings to allow for cooling
and by cables that must exit the shield. These penetra-
tions must be properly treated (such as placing gratings
with many small, closely spaced holes over openings or
placing lters in every exiting cable), or else the shielding
effectiveness will be drastically reduced.
Other types of shielding act to divert rather than re-
ecting magnetic elds. Ferromagnetic materials have
relative permeabilities greater than unity. For example,
sheet steel has m
r
2000, and mumetal has m
r
30,000.
These materials present a low reluctance path for the
magnetic elds and hence divert them away from elec-
tronics that may be susceptible to them.
4.4. Crosstalk
There is another aspect to EMC that is often overlooked:
the ability of the system to interfere with itself. A common
such situation is crosstalk between adjacent transmission
lines. Consider a three-conductor line shown in Fig. 21.
One conductor is called the generator conductor and
forms, with the reference conductor (a ground plane, a
wire, a PCB land, or an overall shield), the generator cir-
cuit. It is driven by a source that has an open-circuit volt-
age V
S
(t) and source resistance R
S
, and is terminated by a
resistor R
L
. Another conductor, the receptor conductor,
with the reference conductor forms the receptor circuit.
This circuit is terminated at the near end (with reference
to the source on the generator circuit) with a resistor R
NE
and at the far end with a resistor R
FE
. The line has total
length of L. The electric and magnetic elds of the gen-
erator circuit interact with the receptor circuit and cause
induced voltages, V
NE
and V
FE
, at the terminals of the re-
ceptor circuit. These induced voltages may cause interfer-
ence with the devices that are represented by R
NE
and
R
FE
. This is referred to as crosstalk.
Computing this crosstalk is generally a formidable task.
(See COUPLED TRANSMISSION LINES in this Encyclopedia). We
may, however, obtain a simple model for this crosstalk as
long as (1) the line length is much less than a wavelength
at the highest significant frequency of the source, that is, is
electrically short, and (2) the two circuits are weakly cou-
pled [1]. This simple model is shown in Fig. 22. The electric
eld and voltage of the generator circuit induce a current
source in the receptor circuit, via a per unit length (PUL)
mutual capacitance c
m
, whose value is joc
m
L
^
VV
G;DC
where
^
VV
G;DC
is the phasor voltage between the conductors of the
generator circuit. Because the line length is assumed elec-
trically small, this voltage is approximately constant along
the line and can be computed as though it were DC as
^
VV
G;DC

R
L
R
S
R
L
^
VV
S
8a
Similarly, the magnetic eld and current of the generator
circuit induce a voltage source (according to Faradays law)
in the receptor circuit, via a PUL mutual inductance l
m
,
whose value is jol
m
L
^
II
G;DC
where
^
II
G;DC
is the phasor
Figure 20. Illustration of multiple reections within a shield.
Figure 21. The general three-conductor trans-
mission line, illustrating crosstalk.
current along the conductors of the generator circuit. Be-
cause the line length is assumed electrically short, this
current is also approximately constant along the line and
can be computed as though it were DC as
^
II
G;DC

1
R
S
R
L
^
VV
S
8b
Hence the near-end and far-end phasor crosstalk voltages
can be computed from this circuit as
^
VV
NE
R
NE
R
NE
R
FE
jol
m
L
^
II
G;DC
..
inductive coupling
R
NE
R
FE
R
NE
R
FE
joc
m
L
^
VV
G;DC
..
capacitive coupling
jo
R
NE
R
NE
R
FE
1
R
S
R
L
l
m
L
_
R
NE
R
FE
R
NE
R
FE
c
m
L
R
L
R
S
R
L
_
^
VV
S
9a
^
VV
FE

R
FE
R
NE
R
FE
jol
m
L
^
II
G;DC
..
inductive coupling
R
NE
R
FE
R
NE
R
FE
joc
m
L
^
VV
G;DC
..
capacitive coupling
jo
R
FE
R
NE
R
FE
1
R
S
R
L
l
m
L
_
R
NE
R
FE
R
NE
R
FE
c
m
L
R
L
R
S
R
L
_
^
VV
S
9b
These phasor crosstalk voltages are the sum of an induc-
tive coupling term due to the mutual inductance and the
current of the generator circuit, and a capacitive coupling
term due to the mutual capacitance and the voltage of the
generator circuit. Observe that the nal results in (9) vary
directly with the frequency o2pf of the source. Hence
the magnitudes of the crosstalk transfer functions
j
^
VV
NE
=
^
VV
S
j and j
^
VV
FE
=
^
VV
S
j increase with frequency at 20dB/
decade while the phase is 901. A substantial amount of ex-
perimental as well as computed data given in Ref. 1 con-
rm this behavior and the predictions of this simple model.
The model presented above is for single-frequency,
sinusoidal source waveforms. For more general time-
domain waveforms we may substitute
jo ,
d
dt
10
to give the time-domain crosstalk as
V
NE
t
R
NE
R
NE
R
FE
1
R
S
R
L
l
m
L
R
NE
R
FE
R
NE
R
FE
c
m
L
R
L
R
S
R
L
_ _
..
M
NE
dV
S
t
dt
11a
V
FE
t
R
FE
R
NE
R
FE
1
R
S
R
L
l
m
L
R
NE
R
FE
R
NE
R
FE
c
m
L
R
L
R
S
R
L
_ _
..
M
FE
dV
S
t
dt
11b
Hence the near-end and far-end crosstalk voltages depend
on the derivative of the source voltage waveform, also
known as its slew rate. Consider the source waveform to
be that of a digital clock or data signal. The resulting
crosstalk waveforms are shown in Fig. 23. Observe that
(1) the crosstalk pulses occur during the transitions of the
clock or data pulses and (2) their levels depend, in addition
Figure 22. The simple inductivecapacitive coupling model of
the receptor circuit for sinusoidal excitation.
Figure 23. Time-domain crosstalk prediction of the inductive
capacitive coupling model for a trapezoidal pulsetrain excitation.
to the mutual inductance and capacitance, on the rise/fall-
times of the source waveforms. Hence again we see that in
order to reduce the crosstalk magnitudes, we should in-
crease the rise/falltimes of the clock and data pulses. This
rule of thumb also applied to radiated emissions.
We now turn to other methods of reducing crosstalk
useful for wire-type conductors. Clearly, since the cross-
talk depends on the PUL mutual inductance and capaci-
tance between the two circuits, we can reduce the
crosstalk by (1) separating the two circuits further and
(2) orienting the two circuits so that they are orthogonal to
each other. When these methods along with increasing the
rise/falltimes fail to reduce the crosstalk to acceptable lev-
els, we have two other options: (1) enclosing the generator
and/or receptor wires with a shield and (2) twisting the
two wires of the generator circuit together and/or twisting
the two wires of the receptor circuit together to form
twisted pairs. First we address the use of shielded wires.
Suppose that we enclose, for example, the receptor wire
with a concentric shield as shown in Fig. 24. The mutual
capacitance between the generator wire and the receptor
wire will be eliminated, and we obtain the capacitive cou-
pling model shown in Fig. 24. From this we see that the
capacitive coupling can be eliminated if we make the
shield voltage (with respect to the reference conductor,
e.g., a ground plane) zero. Since the line is assumed to be
electrically short, we can ground the shield at only one
end and make the voltage all along it approximately zero.
In order to eliminate inductive coupling with a shield,
the shield must be grounded at both ends as shown in
Fig. 25. The process is as follows. The current of the gen-
erator circuit induces a magnetic eld that threads the
shield-reference conductor loop thereby inducing a cur-
rent owing back along the shield
^
II
S
, which produces its
own magnetic ux that tends to cancel the original ux
penetrating this loop. If the shield were not connected to
the reference conductor (grounded) at both ends, no
shield current could ow and hence no counteracting mag-
netic eld would be produced. Hence, surrounding the
generator wire and/or the receptor wire with a concentric
shield will ideally eliminate (1) capacitive coupling as long
as it is grounded at at least one end and (2) inductive cou-
pling as long as it is grounded at both ends.
The other alternative is to replace the receptor wire with
two wires (a dedicated return) and twist them together to
form a twisted pair. A similar scheme can be applied to the
generator wire producing the same effect. The basic process
whereby a twisted wire eliminates magnetic eld or induc-
tive coupling is illustrated in Fig. 26. The current of the
generator circuit causes magnetic uxes to penetrate the
loops of the twisted pair, inducing in them, by Faradays law,
voltage sources. When we untwist this twisted pair, we see
that the Faraday voltage sources in adjacent twists cancel
out. Suppose, for example, that the reference conductor is a
ground plane and the receptor wires are twisted together as
shown in Fig. 27. Although a twisted pair is truly a bilar
helix, we have shown an approximate model consisting of
Figure 24. A lumped equivalent circuit for capacitive coupling
for the shielded receptor wire.
Figure 25. Illustration of the effect of placing a shield around a
receptor wire on inductive coupling.
Figure 26. Illustration of the effect of a twisted pair of receptor
circuit wires on magnetic eld (inductive) coupling.
adjacent loops abruptly interchanged. The length of each
half-twist is denoted by L
HT
. We can view this as superim-
posing two of our previous generatorreceptor coupling cir-
cuits; that is, we may viewthe twisted receptor pair as being
two receptor wire-ground plane circuits. Hence the induced
sources are as shown in Fig. 28. There are two ways of ter-
minating a two-wire line as shown in Fig. 29. In the un-
balanced case, one end is single-ended while the other is
oating. In the balanced case, either center-tapped trans-
formers or balanced line drivers and receivers are used to
terminate the line. For the unbalanced case, untwisting
the two wires gives the equivalent circuit of Fig. 30. The
Faraday lawinduced sources again tend to cancel. Howev-
er, the induced current sources of the lower wire are short-
circuited out but those attached to the upper wire do not
cancel. Hence the inductive coupling portion of the crosstalk
is canceled out and the capacitive coupling portion is ap-
proximately the same as before the wires were twisted to-
gether. For the balanced termination case, untwisting the
pair of receptor wires gives the equivalent circuit in Fig. 31.
Observe that the Faraday lawinduced voltage sources are
again canceled out by the twist but more importantly, the
balancing of the terminations causes the capacitive-cou-
pling-induced current sources to also cancel out. Hence, (1)
a twisted pair eliminates inductive coupling by virtue of the
twist but (2) the capacitive coupling can be eliminated only
by using balanced terminations. See Ref. 1 for extensive ex-
perimental and computed data that conrm these observa-
tions about shielded wires and twisted pairs in the
reduction of crosstalk.
Figure 27. A simple abrupt loop model of a twisted pair of re-
ceptor circuit wires.
Figure 28. The simple inductivecapacitive coupling model for
the twisted pair of receptor circuit wires.
Figure 29. Terminating a twisted pair: (a) un-
balanced; (b) balanced.
5. SUMMARY
Interference in electronic systems is increasing in impor-
tance because of the necessity to use higher clock speeds in
digital systems and to use the GHz frequency spectrum for
analog systems. For these reasons, the electromagnetic
compatibility (EMC) of electronic systems is rapidly be-
coming an important design criterion. Susceptibility of
electronic systems to electromagnetic emissions as well as
the unintended generation of electromagnetic emissions
by those electronic systems can create loss of consumer
condence if the operation of those devices is impaired.
Perhaps more importantly, virtually all countries through-
out the world now limit the unintended electromagnetic
emissions of digital devices, and some limit the suscepti-
bility of digital devices to electromagnetic emissions.
Hence EMC of digital devices has become a legal issue.
It does little good to produce a novel digital device if it
cannot be legally sold in a country because of its inability
to comply with the legal requirements on its electromag-
netic emissions and/or its susceptibility to them. Hence
EMC will no doubt steadily increase as an important de-
sign criterion for digital products. The majority of electri-
cal engineering curricula do not address this problem
today, although the number of curricula that do is steadi-
ly increasing [2]. Hence the majority of those graduating
engineers have little, if any, knowledge of this important
aspect of design until they enter the job place. It has be-
come imperitive that universities begin to incorporate
EMC into their curricula so that engineers will be better
prepared to design future electronic products.
BIBLIOGRAPHY
1. C. R. Paul, Introduction to Electromagnetic Compatibility,
Wiley-Interscience, New York, 1992.
2. C. R. Paul, Electromagnetics for Engineers: with Applications
to Digital Systems and Electromagnetic Interference, Wiley,
New York, 2004.
ELECTROMAGNETIC FERRITE TILE ABSORBER
CHRISTOPHER L. HOLLOWAY
JAMES R. BAKERJARVIS
ROBERT T. JOHNK
RICHARD G. GEYER
U.S. Department of Commerce
Boulder Laboratories
Boulder, Colorado
Anechoic chambers have been used for over forty years to
simulate different operating environments for a wide
range of frequencies [13]. These chambers provide an ac-
curate and convenient environment for electromagnetic
compatibility (EMC) testing and are important cost-effec-
tive tools for achieving EMC compliance. To perform tests
in anechoic chambers in the 30MHz to 1000MHz frequen-
cy range, good low-frequency electromagnetic absorbers
are needed. Magnetic materials, such as ferrite tile ab-
sorbers, offer the type of low-frequency performance need-
ed for EMC testing inside shielded rooms (or chambers).
The rst anechoic chambers were constructed in the
early 1950s for antenna measurements. These chambers
were equipped with bats of loosely spun animal hair coat-
ed with carbon. This broadband hair absorber was
5.08 cm (2in) thick and provided 20 dB of reectivity
(20 log [reected wave/incident wave]) at normal incidence
from 2.4 GHz to 10 GHz. Later in the decade, the hair ab-
sorber was replaced by a new generation of absorber that
Figure 30. The inductivecapacitive coupling model for the un-
balanced twisted receptor wire pair obtained by untwisting the
wires.
Figure 31. The inductivecapacitive coupling model for a twisted
receptor wire pair and balanced terminations.
ELECTROMAGNETIC FERRITE TILE ABSORBER 1151
offered 40 dB of reectivity at normal incidence. Unlike
the hair absorber, the new absorber was equipped with a
shaped or convoluted front surface. Commercially manu-
factured absorbers emerged in the 1960s and boasted less
than 60 dB of reectivity at normal incidence. This low
reectivity was obtained at high microwave frequencies,
however, where the thickness (tip-to-base dimension) of
the pyramids was at least several wavelengths.
The multiple reections between the electrically thick
pyramids account for the excellent performance. Because
the individual pyramids are relatively large compared with
a wavelength, the sides of the pyramids reect and re-re-
ect the incident wave many times. Upon each reection, a
portion of the wave is absorbed by the pyramid. Hence, an
extremely small portion of the incident wave energy sur-
vives the many reections that occur before it arrives at the
solid absorber base located between the pyramids and the
metal chamber wall. Then the remaining wave is absorbed
further as it travels through the base until it is reected by
the chamber wall. The reected wave is absorbed as it pro-
ceeds through the absorber base and again experiences the
multiple reections between adjacent pyramids before
emerging from the plane dened by the pyramid tips.
Semi-anechoic chambers were rst used for emission
measurements during the 1970s as an alternative to an
open area test site (OATS), a ground-plane facility for
emission measurements performed from 30 MHz to
1000 MHz. These chambers were equipped with a con-
ducting oor to simulate the large ground plane employed
on the OATS. Because it obviously was impractical to con-
struct a chamber with absorbers several wavelengths long
at 30 MHz, the early chambers typically were equipped
with 0.91 to 1.83 m (3 to 6ft) thick absorber. Originally
intended to be operated at higher frequencies, this ab-
sorber restricted the use of these chambers to approxi-
mately the 90 MHz to 1000 MHz frequency range [4].
With the promulgation of emission requirements during
the late 1970s, absorber manufacturers were confronted
with a growing demand for a compact wideband absorber
to operate over the entire 30 to 1000MHz frequency range.
The evolution of compact wideband absorbers was cata-
lyzed in the mid 1980s when techniques to calculate and
measure its reectivity were developed. The rst compu-
tationally efcient, accurate calculation of absorber reec-
tivity became possible when the method of homogenization
was used in a theoretical model [57]. Soon afterward, the
reectivity of compact, wideband absorbers was measured
directly with large test xtures [811].
To achieve small reectivities over the entire 30 to
1000 MHz frequency range, compact wideband absorbers
must use tapered structures (like pyramids or wedges)
that not only operate at frequencies where they are elec-
trically thick, but at the frequencies where they are elec-
trically thin. When an incident wave encounters
electrically thin absorbers, it does not see the ne struc-
ture of the pyramids or wedges. Instead, the incident wave
behaves as though it encountered a solid medium whose
effective conductivity and permittivity vary with the dis-
tance into the medium. These effective material properties
differ from the conductivity and permittivity of the actual
material used to construct the absorber.
An optimized electrically thin absorber provides a tran-
sition from the wave impedance of free space to the wave
impedance of the absorber base. With the correct carbon
loading, most of the incident wave penetrates the tapered
structure (pyramids or wedges) and is absorbed as it trav-
els through their solid base. Moreover, the carbon loading
can be adjusted further to obtain cancellation between the
portion of the incident wave reected by the tapered struc-
ture and the wave that emerges from the absorber after
reection by the metal wall. This cancellation causes an
extremely small reectivity, albeit over a relatively nar-
row frequency range. In general, the carbon loading se-
lected for electrically thin, tapered structures differs from
the loading employed for electrically thick structures [7].
Although it is possible to achieve good low-frequency
performance of carbon-loaded, tapered structures, they
are physically large and require large metal enclosures
to house them. An alternative to these large carbon-loaded,
tapered structures is the ferrite-tile absorber. Electrically
thin ferrite tiles were developed in Japan during the early
1960s [1214] as an alternative to urethane pyramids and
wedges. Because they have a wave impedance close to the
impedance of free space, the tiles allow direct penetration
of the incident wave without significant reection at the
air-to-tile interface. Furthermore, the tiles are magneti-
cally lossy absorbing the penetrating wave almost com-
pletely as it travels through the tile. Subsequently, the
wave is reected by the metal wall behind the ferrite and
travels toward the ferrite/air interface. As with the elec-
trically thin urethane absorber, the thickness of the ferrite
can be selected to achieve narrowband cancellation from
the portion of the incident wave reected at the air-to-tile
interface and from the wave that emerges from the ferrite
tile after reection by the metal wall.
Recently, thin urethane pyramids or wedges that oper-
ate from about 200MHz to 1000 MHz have been combined
with ferrite tiles that are effective from 30 MHz to
600 MHz. The combination of these two absorbers, along
with properly chosen dielectric layers, results in an ultra-
compact wideband hybrid absorber that exhibits excel-
lent performance from 30 MHz to 1000 MHz [15,16]. These
hybrid absorbers are discussed in detail in [3].
In this chapter, we discuss the ferrite tile, ferrite grids,
and hybrid combinations of urethane and ferrite absorbers
typically used in electromagnetic test chambers for the
30 MHz to 1000MHz frequency range. We present formu-
las and calculated reectivities (or reection coefcients).
We also discuss how material properties and reectivities
are measured for the ferrite-tile absorbers.
1. MODELING FERRITE TILES
The performance of absorbers is determined by comparing
the magnitudes of incident and reected plane waves, where
the absorber is assumed to be an innite plane. One such
comparison is the reection coefcient G, dened as
G
E
r
E
i
1
1152 ELECTROMAGNETIC FERRITE TILE ABSORBER
where E
r
is the reected electric eld and E
i
is the incident
electric eld. Alternatively, absorber performance is ex-
pressed as reectivity R in decibels as
R20
.
log
10
jGjdB 2
The smaller the value of R in dB, the better the performance
of the absorber.
It is very straightforward to model a ferrite tile theo-
retically with the aid of classical transmission-line equa-
tions. The reectivity of a solid ferrite tile backed by a
metal wall is given by
G
Z
f
Z
Z
f
Z
3
where Z is dened as
Z
m
e
_
4
and
Z
f
Z
1 e
2gd
1e
2gd
5
where d is the ferrite thickness and g is the propagation
constant dened by
g jo

me
p
6
In these expressions e and m are the complex permittivity
and permeability of the ferrite tile. While Eq. (3) is for a
normal incident plane wave, it can be modied to handle
off-normal incidence by incorporating the angular depen-
dence into the impedance and propagation constant (for
details, see [3,6,17]).
Figure 1 shows the reectivity of a solid ferrite tile and
the material properties are given in Table 1. The gure
shows that the tile performs very well below 600MHz, but
above 600MHz, the tile performance deteriorates.
The higher frequency (4600MHz) performance is im-
proved by using so-called ferrite grids. A ferrite grid (or
wafe) is shown in Fig. 2. The grid is a two-dimensional ar-
ray of square air sections cut into a ferrite-tile matrix. This
structure is modeled with an averaging technique known as
homogenization and the transverse material properties are
approximated by the following expressions [3]:
e
t
e
a
1 g
1
e
0
e
a
g
2e
a
and
m
t
m
a
1 g
1
m
0
m
a
g
2m
a
7
The longitudinal permittivity and permeability are known
exactly [18] as
e
z
1 ge
0
ge
a
and
m
z
1 gm
0
gm
a
8
In these expressions, ga
2
/p
2
is the fraction of space occu-
pied by the absorber, and e
a
and m
a
are complex parameters of
the bulk ferrite. These material properties then are substi-
tuted into Eq. (5) to obtain the reectivity of the ferrite grid.
The reectivity of the ferrite grid strongly depends on
the lling factor g. One of the advantages of this ferrite
Figure 1. Reectivity of a 6.38mm (0.25in) solid ferrite tile at 01
and 451 incident angles.
Table 1. Data on Ferrite Tiles
Ferrite Tiles
Permittivity Permeability
Frequency (MHz) e
0
r
e
00
r
m
0
r
m
00
r
30 10.88 0.16 52.31 236.17
40 10.93 0.37 31.26 181.36
50 11.04 0.50 21.05 147.02
60 11.24 0.51 15.59 123.51
70 11.39 0.27 12.32 106.32
80 11.41 0.09 9.96 93.56
90 11.37 0.07 8.16 83.57
100 11.19 0.04 6.74 75.52
150 11.29 0.53 3.39 51.64
200 10.97 0.07 1.85 38.77
250 11.37 0.26 1.40 31.38
300 10.93 0.18 0.52 26.23
400 11.00 0.17 0.12 19.82
500 11.03 0.02 0.47 15.66
600 10.95 0.19 0.59 12.93
700 10.99 0.28 0.59 11.07
800 11.18 0.23 0.48 9.69
900 11.31 0.14 0.38 8.55
1000 11.36 0.04 0.29 7.69
grid geometry over the standard solid tile is that there is
an additional parameter, the lling factor g, that can be
varied to place the nulls (or minimum) of the reectivity at
a desired frequency. The ferrite grid can be designed for
wider frequency band performance over that of the stan-
dard solid tiles by appropriately choosing the tile thick-
ness, lling factor g, and the material properties of the
ferrite. Figure 3 shows results of the reectivity of a ferrite
grid for various values of the lling factor g. The optimum
reectivity of an 18-mm thick ferrite grid with a lling
factor of g 0.725 was calculated using the material prop-
erties in [19]. The results are shown in Fig. 4 for normally
and obliquely incident waves.
Small, tapered, electric-absorbing structures boast
very low reectivities above about 200MHz [3], whereas
ferrite tiles perform best below 600MHz. Hence, compact
wideband absorbers can be designed by combining tapered
electrical absorbers with ferrite tiles. The reectivity in
the 600MHz to 1000MHz range is also reduced by simply
adding a dielectric layer between the ferrite and the metal
chamber wall. The performance advantages of these so-
called hybrids are presented in [3].
In the following sections, we discuss how the material
properties and the reectivities of the ferrite tiles are
measured.
2. MEASUREMENTS OF MATERIAL PROPERTIES
Accurate measurements of the high-frequency dispersive
magnetic properties of ferrite materials are important for
optimally designing a ferrite tile absorber. Magnetic loss
mechanisms are strongly frequency-dependent and gen-
erally behave nonlinearly. At very low frequencies
hysteretic effects (or braking forces acting on Bloch walls
in motion) dominantly inuence magnetic loss. Eddy cur-
rents also produce an apparent reduction of permeability
and an apparent increase of magnetic loss tangent asso-
ciated with attenuation of the magnetic eld inside a fer-
rite material at low frequencies. However this remains
small in most ferrite materials with low dielectric loss.
At RF and microwave frequencies, domain wall move-
ment and domain rotation contribute to magnetic loss.
The initial permeability spectrum is usually considered to
be that part of the permeability spectrum due to domain
wall motion. As the RF excitation frequency increases, do-
main wall rotation no longer fully responds to the excita-
tion. At this point, magnetization does not move in phase
with the excitation, and losses occur. Then the real part m
0
of the magnetic permeability decreases with increased
frequency. The imaginary part, or magnetic loss index
m
00
, goes through a broad absorptive relaxation.
One characteristic constant of ferrites is the product
f
0
jm
d
jf
c
, where m
d
m
0
d
jm
00
d
is the demagnetized scalar
initial permeability and f
c
is the corner frequency above
which jm
d
j falls off with frequency as 1/f. For some polycrys-
talline ferrites, this product represents the frequency at
which the relative permeability becomes 1 [20]. This rule is
only a rst-order approximation that depends on the spec-
tral overlap of domain wall and domain rotation relaxation
phenomena in the ferrite under test. Another characteristic
a p
d
Figure 2. Illustration of the grid or wafe ferrite tile geometry.
0
10
20
30
40
50
0 250 500 750 1000 1250
Frequency (MHz)
1500 1750
g = 0.10
g = 0.15
g = 0.20
g = 0.25
g = 0.30
g = 0.33
g = 0.40
g = 0.50
2000
R
e
f
l
e
c
t
i
v
i
t
y

R

(
d
B
)
Figure 3. Reectivity of a 16mm ferrite grid for various values of
the volume fraction g for an incident angle of 01.
0
10
20
30
40
50
100 0
R
e
f
l
e
c
t
i
v
i
t
y

R

(
d
B
)
200 1000 300 400 500 600 700 800 900
= 0.0
= 45.0: TE polarization
= 45.0: TM polarization
[
[
[
Figure 4. Reectivity of a ferrite grid at 01 and 451 incident an-
gles for lling factor g 0.725 and a thickness of d18mm
(0.71in).
of ferrite materials is that the peak in the magnetic loss in-
dex usually occurs at the frequency where the real perme-
ability has decreased to one-half its static value [21].
Generally, the absorptive window applications of ferrite
materials are at frequencies below natural gyromagnetic
resonance g
gr
M
s
, where M
s
is saturation magnetization (T)
and g
gr
is the gyromagnetic ratio (28GHz/T or
35.19 MHz m/kA). At frequencies less than g
gr
M
s
, mag-
netic losses in ferrite materials are high, whereas at fre-
quencies greater than natural gyromagnetic resonance,
magnetic losses rapidly decrease. With increasing satura-
tion magnetization, (1) natural gyromagnetic resonance
shifts to higher frequencies, (2) f
c
increases, and (3) the
low-frequency static initial permeability decreases. The
absorptive window, similarly, shifts to higher frequencies.
With an applied static magnetic eld in the z direction,
ferrite materials become uniaxially anisotropic and non-
reciprocal, and the magnetic permeability is described by
the well-known nondiagonalized tensor
m m
0
m
jk
0
jk
0
0 0 m
z
_
_
_
_
9
where m
m
0
jm
00
is the principal direction transverse
component of the magnetic permeability, m
z
m
0
z
jm
00
z
is
the parallel component, and k
k
0
jk
00
is the off-diagonal
transverse component. All components of the magnetic
permeability tensor are complex because of ferrite mag-
netic losses. Particular tensor components depend on fer-
rite composition, porosity, grain size, saturation
magnetization, and RF frequency, temperature, and ap-
plied static magnetic eld strength. Near ferromagnetic
resonance (with an applied static eld) saturated ferrite
losses are expressed by the resonance line width. Without
an applied static eld, the ferrite in the demagnetized
state is isotropic and reciprocal. In this case the perme-
ability tensor reduces to the scalar frequency-dependent
permeability, m
d
m
0
d
jm
00
d
.
Regardless of general spectral characteristics, the mag-
netic properties of ferrite materials must be measured at
microwave frequencies. Experimental procedures for ac-
curate magnetic permeability measurements on demag-
netized ferrites, therefore, depend spectrally on which side
of the natural gyromagnetic resonance the magnetic char-
acterization is desired. At frequencies lower than g
gr
M
s
,
one-port permeameter and two-port coaxial transmission
line measurements provide accurate magnetic permeabil-
ity characterization of ferrite materials. At microwave fre-
quencies above g
gr
M
s
, resonant cavity or dielectric
resonator techniques must be used.
2.1. Transmission Line Measurement
The goal of this section is to present transmission line
measuring methods for permeability [22]. In the measure-
ment, a sample is inserted into a waveguide or a coaxial
line, and the sample is subjected to an incident electro-
magnetic eld. In developing the scattering equations,
usually only the fundamental waveguide mode is assumed
to exist. At low frequencies, the impedance of the sample is
measured. At microwave frequencies, the two-port scat-
tering matrix is measured. Reection coefcient and scat-
tering equations are found from an analysis of the
magnetic eld in the sample holder. Both high- and low-
frequency measurements are considered. The rst tech-
nique, the permeameter, is a low-frequency technique.
When lumped circuit parameters (LCR) are used as
data, this method is limited to frequencies below
50 MHz. The method is extended to a higher frequency
by using reection coefcient data from a network ana-
lyzer. As the operating frequency increases to the point
where there is an appreciable electric eld in the sample,
however, then knowledge of the permittivity is required.
The second technique is the transmission line scattering
parameter technique. This technique is limited to micro-
wave frequencies, because the measured phase change
across the sample becomes inaccurate below 1 MHz. Pre-
vious work using the coaxial line for magnetic measure-
ments has been done, for example, by Von Hippel [23],
Bussey [24], Geyer and BakerJarvis [25], and Hoer [26].
2.1.1. Permeameters. The permeameter consists of a
short-circuited coaxial sample holder (see Fig. 5) with a
sample on the inner conductor of a coaxial line terminated
in a short circuit. The advantages of this approach are the
ease of sample installation and the broad frequency capa-
bility. An air-gap correction is not necessary for perme-
ability measurements because the magnetic eld is
tangential to air gaps. In this technique, a LCR meter is
used to measure admittance Y in the frequency range of
100 Hz to 1 MHz. Then this admittance is converted to a
reection coefcient by G(Y
0
Y)/(Y
0
Y), where Y
0
is
the admittance without the sample. The reection coef-
cient for a sample positioned next to a short-circuited ter-
mination is given by
G exp2g
a
L
a
g
t
L
t
tanhgL
m
0
g
mg
0
tanhgL
m
0
g
mg
0
10
where g is the propagative constant in the material given
by
g j
o
2
m
r
e
r
c
2
vac
2p
l
c
_ _
2
11
L
air
L
L
t
L
a
Sample
Bead Short circuit
Figure 5. Measurement setup for one-port coaxial line permea-
meter. Sample resides adjacent to the short circuit.
Also, c
vac
is the speed of light in a vacuum; L is the sample
length, the subscripts a and t denote air-line section in the
connector and support bead; l
c
is the cutoff wavelength;
and e
0
and m
0
are the permittivity and permeability of
vacuum, respectively, e
r
e
0
r
je
00
r
and m
r
m
0
r
jm
00
r
are the
complex permittivity and permeability relative to a vacu-
um, respectively. This equation is useful for both micro-
wave and low-frequency measurements. At low
frequencies the permittivity of the sample does not con-
tribute to the determination of permeability. We have
shown the frequency limitation for applying the permea-
meter technique is given by
f !
c
4pmL
12
where c is the speed of light in a vacuum.
2.1.2. Two-Port Permeability and Permittivity Determina-
tion. The two-port scattering parameter technique is used
from approximately 50 MHz to microwave frequencies and
yields both permittivity and permeability (see Fig. 6). The
scattering parameters are dened in terms of the reec-
tion coefcient G and the transmission coefcient z by
S
11
R
2
1
G1 z
2
1 G
2
z
2
_ _
13
S
22
R
2
2
G1 z
2
1 G
2
z
2
_ _
14
S
21
R
1
R
2
z1 G
2
1 G
2
z
2
_ _
15
and
g
0
j
o
c
lab
_ _
2
2p
l
nc
_ _
2
16
where g
0
is the propagation constant in a vacuum, c
lab
is
the speed of light, and o is the radian frequency.
R
1
expg
0
L
1
17
and
R
2
expg
0
L
2
18
are the respective reference plane transformations. Equa-
tions (13) through (15) are derived in detail by Hoer and
Rasmussen [26]. The transmission coefcient z is ex-
pressed as
z expgL 19
and the reection coefcient is dened as
G
m
g
m
0
g
0
m
g
m
0
g
0
20
For a coaxial line, the cutoff frequency approaches 0
(o
c
-0).
To obtain both the permittivity and the permeability
from the scattering parameter relationships, at least two
independent measurements are necessary. In the full scat-
tering parameter solution, a solution of the equations is
obtained that is invariant to reference planes for e
r
and m
r
.
A set of equations for single-sample magnetic measure-
ments for an air line of length L
air
is
S
11
S
22
S
21
S
12
exp2g
0
L
air
L
G
2
z
2
1 G
2
z
2
21
and
S
21
S
12
=2 expg
0
L
air
L
z1 G
2
1 G
2
z
2
22
It is possible to obtain an explicit solution to Eqs. (21)
and (22). If x (S
21
S
12
S
11
S
22
) exp[2g
0
(L
air
L)] and y
[(S
21
S
12
)/2] exp[g
0
(L
air
L)], then the physical roots for
the transmission coefcient are given by
Z
x 1
2y

x 1
2y
_ _
2
1
23
and the reection coefcient is expressed as
G
2

x Z
2
xZ
2
1
_
24
The ambiguity in the 7 sign in Eq. (24) is resolved by
considering the reection coefcient calculated from S
11
alone:
G
3
aZ
2
1
a
2
Z
4
2Z
2
2S
11
a
2
a
2
_
2S
11
Z
2
25
where a exp( 2g
0
L
1
). The correct root for G
3
is chosen
by requiring that |G
3
|r1. An estimate of L
1
is needed in
Eq. (25). If G
2
is compared with G
3
, then the 7 sign am-
biguity is resolved and therefore G
2
is determined. Then
the permeability and permittivity are
m
r

1 G
2
1 G
2
1
g
0
L
lnZ2pjn 26
Outer conductor
I
Port 1
III
Port 1
Outer conductor
L
1
air
L
2
air
L
sample
II
Figure 6. A dielectric sample in a transmission line. Port 1 and
Port 2 denote positions of calibration reference planes.
and
e
c
2
o
2
2p
l
c
_ _
2
1
L
2
lnZ2pjn
2
_ _
=m
r
27
where n0,71,72,y. The correct value of n is chosen by
using a group-delay comparison test. At low frequencies,
the correct roots are identied more easily because they
are more widely spaced.
Equations (26) and (27) have an innite number of
roots for magnetic materials, because the logarithm of a
complex number is multivalued. To choose the correct
root, it is necessary to compare the measured group de-
lay with the calculated group delay. The calculated group
delay is related to the change of the wave number k with
respect to the angular frequency by
t
calc:group
L
d
df
r
m
r
f
2
c
2

1
l
2
c
28

1
c
2
f e
r
m
r
f
2 1
2
de
r
m
df
r
m
r
f
2
c
2

1
l
2
c
L 29
The measured group delay is given by
t
meas:group
1
2p
df
df
30
where f is the phase of Z. To determine the correct root,
the calculated group delays are found from Eq. (29) for
various values of n in the logarithm term, where ln Z
ln|Z|j(y 2pn). The calculated and measured group
delays are compared to yield the correct value of n.
When there is no loss in the sample under test, the solu-
tion is divergent at integral multiples of one-half wave-
length in the sample. This occurs because the phase of S
11
cannot be measured accurately for small |S
11
|. In the low
loss limit, both of the scattering equations reduce to the
relationship Z
2
-1, which is a relationship only for the
phase velocity, and therefore solutions for e
r
and m
r
are not
separable. This singular behavior is minimized in cases
where permeability is known a priori, as shown in previ-
ous work performed by Baker-Jarvis [22].
A measurement on a ferrite material is given in Fig. 7
using both the permeameter and two-port techniques.
This gure indicates that the permeameter method breaks
down around 50 MHz.
2.2. Dielectric-Resonator Measurements
At frequencies above g
gr
M
s
, resonant-cavity [2731] or di-
electric-resonator techniques [3236] must be used to ac-
curately characterize ferrite magnetic properties.
Commonly used xtures for determining the demagne-
tized scalar permeability m
d
are either a H
011
mode cavity
for measuring one cylindrical sample or a rectangular
cavity operating in the TE
102
mode with a small spherical
sample placed at the center [37]. Both of these techniques
generally employ perturbation theory, which restricts
sample size and, accordingly, allowable magnetic energy
partial lling factors and sensitivity to measure small
changes in the Q-factor. More recently, Latrach, Le Roux,
and Jecko [31] have used a TM
mn0
cavity for permeability
measurements of ferrite disk samples from 2 GHz to
8 GHz. By dielectrically loading the TM
mn0
cavity, the
measured resonant frequency was changed. The method
presented here, discussed in more detail in [3436], uses
low-loss H
011
dielectric resonators, containing the ferrite
sample under test, whose complex permittivity is given by
e
f
e
0
f
je
00
f
. The aspect ratios and permittivities of the di-
electric resonators are chosen to spectrally characterize a
single sample over a broad frequency range. The H
011
res-
onant system used in these measurements is illustrated in
Fig. 8. The resonators are coupled to the external micro-
wave source through two loop-terminated coaxial cables,
1000
100
10
0.1
Frequency (GHz)
Permeameter
Two-port
Figure 7. The measured real part of the permeability using both
the permeameter and scattering techniques.
Teflon sleeve
Spacer
Spacer
Dielectric
ring resonator
Ferrite
sample
Air gap
Metal
Metal
Adjustable coaxial
cable terminated by loop
Figure 8. Parallel-plate H
011
resonant cong-
uration used to measure magnetic permeability.
and these are adjusted so that the measured loaded
Q-factor is equal to the unloaded Q-factor within any
prescribed accuracy.
Formulations for the electromagnetic elds in a ferrite
rod and in coaxial cylindrical dielectric layers surrounding
the rod are found in several papers and textbooks [3852].
At the boundaries between the cylindrical interfaces, the
well-known continuity conditions between tangential elec-
tric and magnetic eld components must be satised. At
the conductive ground planes, tangential electric elds
must vanish. These well-known boundary conditions cre-
ate a system of linear equations with respect to the con-
stant coefcients in the eld expressions, which have
nontrivial solutions only if the corresponding determinant
vanishes. For xed values of dimensions and material
properties of the resonant system, it is possible to nd
eigenvalues as roots of the determinant equation. Mate-
rial properties are determined by measuring resonant fre-
quencies and unloaded Q factors of the dielectric ring
resonators with and without the ferrite sample.
The H
011
resonant frequencies of a ring resonator with
a ferrite sample are determined by e
0
rr
, e
0
f
, and m
0
d
, where e
0
rr
is the relative permittivity of the dielectric ring resonator.
For ferrite sample diameters less than one-quarter of the
external ring resonator diameter, these resonant frequen-
cies are primarily determined by e
0
rr
and m
0
d
.
The rst step in the measurement process is to verify
the complex permittivity of the ferrite sample under test
with a TM
0n0
cavity. Second, the complex permittivities of
each dielectric ring resonator are found from measure-
ments of the resonant frequencies and unloaded Q factors
of the empty ring resonators operating in the H
011
mode,
given the geometrical dimensions of the resonators and
taking into account conductive microwave losses of the
upper and lower ground planes. Then values of the imag-
inary parts of e
f
and e
r
are calculated, assuming a linear
increase with frequency. Next, the scalar permeability m
0
d
,
computed from measurements of the resonant frequency
of these resonators, which contain a completely demagne-
tized sample and operate in the H
011
mode, is evaluated
from the H
011
eigenvalue equation in determinant form:
Fe
0
f
; m
0
d
; f
0
0 31
where f
0
is the measured resonant frequency of the H
011
mode for a resonator containing the demagnetized ferrite
sample.
After m
0
d
is determined, the imaginary part of the per-
meability, m
00
f
m
0
f
tan d
m,f
, is found as a solution to the
equation
Q
1
Q
1
c
p
e
0
r
tand
e;r
p
e
0
f
tand
e;f
p
m
0
d
tand
m;f
32
where Q is the unloaded Q factor for the H
011
mode; Q
c
is
the Q factor representing conductor losses in the metal
plates for the H
011
mode; p
e
0
r
is the electric-energy-lling
factor for the dielectric ring resonator; p
e
0
f
; p
m
0
d
are the fer-
rite sample electric- and magnetic-energy-lling factors;
tan d
e,r
and tan d
e,f
are the dielectric loss tangents of the
ring resonator and the ferrite sample under test; and tan
d
m,f
is the magnetic loss tangent of the ferrite.
Typical measurement data for bulk ceramic ferrites by
this technique are given in Figs. 9 and 10. Uncertainties in
m
0
d
for typical dimensional uncertainties in the geometric
parameters of the dielectric ring resonators and the ferrite
sample under test are estimated to be 70.8%, that is, m
00
d
is
71 10
5
. Combined complex permeability data of two
ferrites with differing saturation magnetizations over
greater than four frequency decades, where both
20 10
Frequency (GHz)
2
1.0
0.1
d
M
s
= 95 kA/m: Al Ferrite
M
s
= 11 kA/m: Ca-V Garnet
M
s
= 142 kA/m: Y Garnet
M
s
Figure 9. Measured relative permeability m
0
d
of several alumi-
num-doped, calcium-vanadium, and yttrium garnet ferrites with
9 kA/mrM
s
r147kA/m as a function of frequency.
20 10
Frequency (GHz)
2
10
1
10
2
10
3
10
4
10
5
d
M
s
= 95 kA/m: Al Ferrite
M
s
M
s
M
s
Figure 10. Measured relative magnetic loss index m
00
d
of several
aluminum-doped, calcium-vanadium, and yttrium garnets with
95kA/mrM
s
r147kA/m as a function of frequency.
two-port coaxial transmission line and dielectric resonator
data have been combined, are shown in Figs. 11 and 12.
The rapid increase in magnetic loss, as natural gyromag-
netic resonance is approached, is clearly seen. In addition,
the large differences in nonlinear magnetic loss behavior
over a broad frequency range are apparent.
The total attenuative losses of ferrite-tile absorbers
should include both the dielectric and magnetic proper-
ties of the ferrite materials under examination. For a
plane-wave TEM eld, the attenuative loss a is given by
a
2o
p
2c
e
0
f
m
0
f
1 tan
2
d
e; f
1=2
1 tan
2
d
m; f
1=2
tand
e; f
tan
m; f
_
33
where c is the speed of light.
3. REFLECTIVITY MEASUREMENTS
In designing a ferrite absorber that performs well over a
selected frequency range, it is important to measure the
reectivity of the absorber for normal and oblique inci-
dence. However, the large test xtures currently used to
characterize an absorber from 30 MHz to 1000 MHz allow
measuring reectivities only at normal incidence [811].
Although an arch is used to measure reectivities at
oblique angles, it is useful only at frequencies above ap-
proximately 600 MHz [53,54]. An alternative to these tech-
niques is a method under development at the National
Institute of Standards and Technology (NIST) in Boulder,
CO. This alternative technique uses time-frequency meth-
ods to measure reectivity at arbitrary angles and polar-
ization [5557].
A block diagram of the bistatic free-space reectivity
measurement system at NIST is depicted in Fig. 13. The
measurement system consists of a time- or frequency-
domain network analyzer, two TEM horn antennas, and
interconnecting RF/microwave cables with a 50O charac-
teristic impedance. More detailed descriptions of time- and
frequency-domain versions of this measurement system
are given in [5559]. A rectangular ferrite-tile sample (or
metal plate reference) is placed in the plane of the two
TEM horn antennas at a distance of 1 to 4m. The level of
the sample center is adjusted to match the antenna aper-
ture centers, and the normal to the sample surface bisects
the angle between the two horn antenna apertures. The
measurements are performed in an ordinary room (labo-
ratory, warehouse, etc.) or in an absorber-lined chamber
(semi- or fully anechoic). To minimize the effects of oor,
100 10 1 0.1 0.01 0.0001
Frequency (GHz)
10
3
10
2
10
1
10
0
10
1
d
M
s
M
s
Dielectric
resonator data
Coaxial transmission
line data
Figure 11. Combined two-port coaxial transmission line and di-
electric resonator relative permeability measurements for yttri-
um and calcium-vanadium garnet ferrites from 1 MHz to 10GHz.
100 10
Dielectric
resonator
data
Coaxial transmission
line data
1 0.1 0.01 0.0001
Frequency (GHz)
10
2
10
1
10
0
10
1
10
2
10
3
10
4
10
5
d
M
s
M
s
Figure 12. Combined two-port coaxial transmission line and di-
electric resonator measurements of relative magnetic loss index
for yttrium and calcium-vanadium garnet ferrites from 1 MHz to
10GHz.
Time or
frequency-domain
network analyzer
Transmitting
TEM horn
Receiving
TEM horn
Ferrite tile
panel or metal
reference plate
Figure 13. Bistatic free-space RF absorber reectivity measure-
ment system.
ceiling, and wall reections, the sample under test should
be situated centrally within the room volume.
Obtaining the scattering characteristics of the ferrite
tile sample under test requires the following steps:
1. A time-domain waveform is obtained for the ferrite-
tile sample under test by placing it close to the TEM
horn antennas (typically from 1 to 4m from the an-
tenna apertures).
2. The sample is removed and a background time-do-
main waveform is obtained.
3. The waveform of step (2) is subtracted from that of
step (1). The resulting waveform consisted of the
ferrite-tile sample response plus a component that
emanates from the shadow region behind the sam-
ple. This process removes significant systematic
effects due to antenna-to-antenna coupling and spu-
rious environmental reections.
4. The shadow-region waveform component is readily
eliminated by time-gating, which deletes the unde-
sired waveform component(s). Time-gating works
only if there is sufcient separation between the
sample and the wall directly behind it, which allows
the absorber response to die out before the shadow-
region component arrives.
5. Steps (1)(4) are repeated for a rectangular (or
square) metal reference plate. The NISTsystem em-
ploys a 3m3 m metal plate.
6. The time-gated absorber and metal reference plate
waveforms obtained in steps (5) and (6) are each
Fourier transformed to obtain scattering amplitude
spectra.
7. Then the amplitude spectra are divided to obtain the
backscatter coefcient which is given by
BCf ; y
jFTgatedabsorber waveformj
jFTgatedreference waveformj
34
where f is the frequency, y is the angle of incidence,
FT denotes the Fourier transform, and BC is the
backscatter coefcient. The backscatter coefcient is
a real, frequency-domain quantity that directly com-
pares the reection characteristics of the ferrite-tile
absorber system under test with that of the metal
plate reference signal. The smaller the backscatter
coefcient, the better the performance of the ferrite-
tile system.
Examples of gated time-domain reference and ferrite-
tile-panel waveforms are shown in Figs. 14 and 15,
respectively. These waveforms are obtained from a bistat-
ic measurement at a 301 angle of incidence with both the
receiving and transmitting antennas horizontally polar-
ized. A synthetic time-domain method is used [59]
in which the acquired frequency-domain data are Fouri-
er transformed into time-domain waveforms. The refer-
ence waveform of Fig. 14 is obtained from a 3 m3 m
0.0E+0 4.0E9 8.0E9 1.2E8 1.6E8 2.0E8
0.040
0.020
0.000
0.020
0.040
Time (s)
R
e
c
e
i
v
e
d

s
i
g
n
a
l

(
V
)
Figure 14. Gated time-domain metal reference plate waveform.
0.0E+0 4.0E9 8.0E9 1.2E8 1.6E8 2.0E8
0.010
0.005
0.000
0.005
0.040
Time (s)
R
e
c
e
i
v
e
d

s
i
g
n
a
l

(
V
)
Figure 15. Gated time-domain ferrite-tile waveform.
0
5
10
15
20
25
B
a
c
k
s
c
a
t
t
e
r

c
o
e
f
f
i
c
i
e
n
t

(
d
B
)
1E
9
1E
8
Frequency (Hz)
1E
7
1E
6
Figure 16. Backscatter coefcient results obtained from the
waveforms of Figs. 14 and 15 in conjunction with Eq. (34).
metal reference plate, located 3.0 m from the antenna
aperture centers. It is interesting to note that the refer-
ence waveform is nearly a perfect doublet, a result of the
differentiating properties of TEM horns in the transmit-
ting mode [60]. The corresponding waveform obtained for
a 2.4 m2.4 m ferrite-tile panel backed by a plywood
layer and metallic ground plate is depicted in Fig. 15.
Reections from the front and back surfaces of the ferrite
tiles produce the initial triplet (5 ns to 7 ns). Wavefront
curvature and nite sample size are responsible for the
portion of the waveform that occurs after 7 ns. The fre-
quency-domain back-scatter coefcient characteristics
from these gated waveforms are depicted in Fig. 16 for
a frequency range of 3 MHz to 500 MHz. The character-
istic notch that occurs slightly above 300 MHz is a phe-
nomenon typical for ferrite tile systems backed by a
ground plane.
4. CONCLUSION
In this article, ferrite tile electromagnetic absorbing ma-
terials were discussed. Methods to model performance,
measure material properties, and measure the reectivity
of this type of absorbing material were presented. The ad-
vantages of the ferrite-tile absorber are that they are
small and offer very low reectivity in the frequency range
of 30 MHz to 600MHz. However, above 600MHz the per-
formance of these ferrite tiles begins to degrade. Dielectric
layers behind the ferrite tiles and tapered electric absorb-
ing structures in front of the tiles are used to improve the
high-frequency performance of the ferrites. A discussion
on how well these hybrid absorbers work is found in [3].
The next generation of ferrite-tile absorbers, consisting of
thin magnetic layers [6164] and a combination of ferrite
layers with chiral materials [3], is currently being devel-
oped.
BIBLIOGRAPHY
1. A. J. Simmons and W. H. Emerson, An anechoic chamber
making use of a new broadband material. Convention Record
of the IRE 1953 National Convention: Part 2-Antennas and
Communications, pp. 3141.
2. W. H. Emerson, Electromagnetic wave absorbers and anecho-
ic chambers through the years. IEEE Trans. Antennas Prop-
agat., 21: 484490, 1973.
3. C. L. Holloway et al., Comparison of electromagnetic absorber
used in anechoic and semi-anechoic chambers for emissions
and immunity testing of digital devices. IEEE Trans. Elect-
romag. Compat., 39: 3347, 1997.
4. R. F. German, Comparison of semi-anechoic chambers and
open-eld site attenuation measurements. Proc. 1982 IEEE
International Symposium on Electromagnetic Compatibility,
Santa Clara, CA, September 810, 1982, pp. 260265.
5. E. F. Kuester and C. L. Holloway, Improved low-frequency
performance of pyramid-cone absorbers for application in
semi-anechoic chambers, Proc. 1989 IEEE National Sympo-
sium on Electromagnetic Compatibility, Denver, CO, May 23
25, 1989, pp. 394399.
6. E. F. Kuester and C. L. Holloway, A low-frequency model for
wedge or pyramid absorber arrays-I: theory, IEEE Trans.
Electromagn. Compat., 36: 300306, 1994.
7. C. L. Holloway and E. F. Kuester, A low-frequency model for
wedge or pyramid absorber arrays-II: computed and mea-
sured results, IEEE Trans. Electromagn. Compat., 36: 307
313, 1994.
8. N. Ari, D. Hansen, and H. Garbe, Analysis and measurements
of electromagnetic scattering by pyramidal absorbers, 8th Int.
Zurich Symp. Technical Exhibition Electromagnetic Compat-
ibility, Zurich, March 79, 1989, pp. 301304.
9. D. Hansen, N. Ari, and H. Garbe, An investigation into
the scattering and radiation characteristic of RF-absorbers,
Proc. 1988 IEEE International Symposium on Electromag-
netic Compatibility, Seattle, WA, August 24, 1988, pp.
99105.
10. S. Takeya and K. Shimada, New measurement method of RF
absorber characteristics by large square coaxial line. Proc.
1988 IEEE International Symposium on Electromagnetic
Compatibility, Dallas, TX, August 913, 1993, pp. 397402.
11. P. Pues, Electromagnetic wave absorber measurement in a
large coax, 9th Int. Zurich Symp. Technical Exhibition Elec-
tromagnetic Compatibility, Zurich, March 1991, pp. 541546.
12. Y. Naito et al., Characteristics of the ferrite absorbing wall,
Trans. I.E.C.E., Japan, 52-B (1): 26, 1969 (available in En-
glish in Electronics and Communication in Japan, p. 76,
1969).
13. Y. Natio and E. Fujiwara, Thickness of electromagnetic wave
absorber utilizing ferrite, Trans. I.E.C.E., Japan, 53-B (9):
537, 1970 (available in English in Electronics and Communi-
cation in Japan, p. 94, 1970).
14. Y. Natio, On the permeability dispersion of a spinel ferrite,
Electronics and Communication in Japan, 56-C (2): 118123,
1973.
15. T. Ellam, An update on the design and synthesis of compact
absorber for EMC chamber applications, Proc. 1994 IEEE In-
ternational Symposium on Electromagnetic Compatibility,
Chicago, August 2226, 1994, pp. 408412.
16. K. Ishino et al., Realization of compact semi- and fully an-
echoic chambers using a new developed composite absorber,
Proc. 1994 IEEE International Symposium on Electromag-
netic Compatibility, Chicago, August 2226, 1994, pp. 413
418.
17. L. M. Brekhovskikh, Waves in Layered Media, New York: Ac-
ademic Press, 1960, ch. 1.
18. E. F. Kuester and C. L. Holloway, Comparison of approxima-
tions for effective parameters of articial dielectrics, IEEE
Trans. Microwave Theory Techn., 38 (11): 17521755, 1990.
19. M. Takahashi, A new structure of electromagnetic wave
absorber, Toyo Corporation, Technical Report, October 1991.
20. R. G. Geyer et al., Spectral characterization of ferrites for use
as magnetic reference materials, Conf. Precision Electromag.
Meas. Dig., 1992, pp. 107108.
21. J. L. Snoek, Dispersion and absorption in magnetic ferrites at
frequencies above one Mc/s. Physica XIV, 4: 207217, 1948.
22. J. BakerJarvis et al., Transmission/reection and short-cir-
cuit line methods for measuring permittivity and permeabil-
ity, Natl. Inst. Stand. Technol., Tech. Note 1355, May, 1992.
23. A. R. V. Hippel, Dielectric Materials and Applications, Cam-
bridge, MA: M.I.T. Press, 1954.
24. R. B. Goldfarb and H. E. Bussey, Method for measuring com-
plex permeability at radio frequencies, Rev. Sci. Instrum., 58
(4): 624627, 1987.
25. R. G. Geyer and J. BakerJarvis, Spectral characterization of
ferrites for use as magnetic reference materials. CPEM92
Digest, Conference on Precision Electromagnetic Measure-
ments, 1992, pp. 107108.
26. C. A. Hoer and A. L. Rasmussen, Equations for the radiofre-
quency magnetic permeameter, NBS Journal of Research,
67C:6976, 1963.
27. R. C. LeCraw and E. G. Spencer, Tensor permeabilities of fer-
rites belowmagnetic saturation, IRE Conv. Rec. NewYork, pt.
5, 1956, pp. 6674.
28. J. J. Green and T. Kohane, Testing of ferrite materials for mi-
crowave applications, Semicond. Prod. Solid State Technol.,
7:4654, 1964.
29. H. E. Bussey and L. A. Steinert, Exact solution for a gyro-
magnetic sample and measurements on a ferrite, IRE Trans.
Microwave Theory Tech., 6: 7276, 1958.
30. W. MullerGronau and I. Wolff, A microwave method for
the determination of the real parts of the magnetic and di-
electric material parameters of premagnetized microwave
ferrites. IEEE Trans. Microwave Theory Tech., 32:377382,
1983.
31. M. Latrach, P. LeRoux, and F. Jecko, Accurate method for
experimental determination of real part of initial permeabil-
ity of demagnetized ferrite samples. In Proc. 10eme Colloque
Optique Hertzienne et Dielectriques, Rennes, September 68,
1989, pp. 339342.
32. N. Ogasawara et al., Highly sensitive procedures for mea-
suring permeabilities (m
7
) for circularly polarized elds in
microwave ferrites. IEEE Trans. Magnetics, 12:256269,
1976.
33. J. Krupka, Resonant modes in shielded cylindrical ferrite and
single crystal resonators. IEEE Trans. Microwave Theory
Tech., 37:691697, 1989.
34. R. G. Geyer and J. Krupka, Microwave behavior of ferrites:
theory and experiment. Conf. Digest on Precision Electromag-
netic Measurements, June 1720, 1996, pp. 206207.
35. R. G. Geyer and J. Krupka, Complex permeability measure-
ments of microwave ferrites. In Microwave Processing of Ma-
terials V, Materials Research Society Symposium Proceedings,
430:257262, 1996.
36. J. Krupka and R. G. Geyer, Complex permeability of de-
magnetized microwave ferrites near and above gyromag-
netic resonance, IEEE Trans. Magnetics, 32 (3):19241933,
1996.
37. J. J. Green and F. Sandy, Microwave characterization of par-
tially magnetized ferrites. IEEE Trans. Microwave Theory
Tech., 22:641645, 1974.
38. M. L. Kales, Modes in waveguides containing ferrites, J. Appl.
Phys., 24:604608, 1953.
39. H. Suhl and L. R. Walker, Topics in guided wave propa-
gation through gyromagnetic media. Pt. I-The com-
pletely lled cylindrical guide, Bell Syst. Tech. J., 33:579
659, 1954.
40. E. Snitzer, Cylindrical dielectric waveguide modes. J. Opt.
Soc. Amer., 51:491498, 1961.
41. R. A. Waldron, Electromagnetic wave propagation in cylin-
drical waveguides containing gyromagnetic media, J. Brit.
IRE, 18: 597612, 677690, and 733746, 1958.
42. R. A. Waldron, Theory of the mode spectra of cylindrical
waveguides containing gyromagnetic media. J. Brit. IRE,
19: 347356, 1959.
43. R. A. Waldron, Features of cylindrical waveguides containing
gyromagnetic media. J. Brit IRE, 695706, 1960.
44. R. A. Waldron, Loss properties of ferrite-loaded cylindrical
waveguides containing gyromagnetic media. J. Brit. IRE,
321334, 1963.
45. R. A. Waldron, Properties of ferrite-loaded cylindrical wave-
guides in the neighborhood of cutoff. Proc. IEEE, 109B (suppl.
no. 21): 9094, 1962.
46. R. A. Waldron, Properties of inhomogeneous cylindrical wave-
guides in the neighborhood of cutoff, J. Brit. IRE, 25: 547555,
1963.
47. J. E. Tompkins, Energy distribution in partially ferrite-lled
waveguides, J. Appl. Phys., 29:399400, 1958.
48. R. A. Waldron, Ferrites, New York: Van Nostrand, 1961.
49. P. J. B. Clarricoats, Microwave Ferrites, London: Chapman
and Hall, 1961.
50. A. J. BadenFuller, Ferrites at Microwave Frequencies, Lon-
don: Peter Peregrinus, 1987, Chap. 3.
51. H. D. Godtmann and W. Hass, Magnetodynamic modes in ax-
ially magnetized rods between two parallel conducting sheets.
IEEE Trans. Microwave Theory Tech., 15:478480, 1967.
52. A. M. Duputz and A. C. Priou, Computer analysis of micro-
wave propagation in a ferrite circular waveguide-optimiza-
tion of phase-shifter longitudinal eld section, IEEE Trans.
Microwave Theory Tech., 22:601613, 1974.
53. E. F. Knott, J. F. Shaeffer, and M. T. Tuley, Radar Cross Sec-
tion: Its Prediction, Measurement and Reduction, Dedham,
MA: Artech House, Inc., 1985, Chap. 10.
54. R. E. Hiatt, E. F. Knott, and T. B. A. Senior, A study of VHF
absorber and anechoic rooms, College of Engineering, Depart-
ment of Electrical and Computer Engineering, Radiation Lab-
oratory, University of Michigan, Ann Arbor, Michigan, Tech.
Report No. 5391-1-F, February, 1963.
55. S. Tofani, A. Ondrejka, and M. Kanda, Time-domain method
for characterizing the reectivity of absorbing materials from
30 to 1000MHz, IEEE Trans. Electromagn. Compat., 33:234
240, 1991.
56. S. Tofani et al., Bistatic scattering of absorbing materials
from 30 to 1000MHz, IEEE Trans. Electromagn. Compat., 34:
304307, 1992.
57. R. Johnk et al., Time-domain measurements of the electro-
magnetic backscatter of pyramidal absorbers and metallic
plates, IEEE Trans. Electromagn. Compat., 35:429433, 1993.
58. R. T. Johnk and A. R. Ondrejka, Electrical material properties
from a freespace time-domain RF absorber reectivity mea-
surement system, Proc. 1997 IEEE International Symposium
on Electromagnetic Compatibility, August 1122, 1997, Aus-
tin, TX.
59. IEEE standard PAR-1128, Recommended practice for RF ab-
sorber evaluation, to be published.
60. R. A. Lawton and A. R. Ondrejka, Antennas and the associ-
ated time-domain range for the measurement of impulsive
elds, Nat. Bur. Stand. (U.S.) Tech. Note 1008, 1978.
61. H. W. Helberg and V. Kose, Die breitbandige absorption elect-
ro-magnetischer wellen durch dunne ferritschichten, Zeitsch-
rift fur Angewandte Physik, 19:509514, 1965.
62. F. Mayer and J. P. Chaumat, Dielectromagnetic materials for
absorber-lined chambers (ALC). Proc. 9th International Zu-
rich Symposium on Electromagnetic Compatibility, Zurich,
March 1214, 1991, pp. 569572.
63. H. Komori and Y. Konishi, Wideband electromagnetic wave
absorber with thin magnetic layers, IEEE Trans. Broadcast-
ing, 40:219222, 1994.
64. Z. Cohn, J. Daly, and C. Parker, Advanced ferrite materials for
anechoic chambers, Ceramics Trans., 47:269284, 1995.
ELECTROMAGNETIC FIELD COMPUTATION
IN PLANAR MULTILAYERS
KRZYSZTOF ARKADIUSZ
MICHALSKI
Texas A&M University
College Station, Texas
1. INTRODUCTION
In a variety of applications, such as geophysical prospect-
ing, remote sensing, detection of landmines and unexplod-
ed ordinance, noninvasive testing, planning of
hyperthermia, modeling of wave propagation in wireless
communication, design of microstrip circuits and anten-
nas, modeling of integrated circuit interconnects, charac-
terization of silicon wafer defects, interpretation of near-
eld scanning optical microscopy, modeling of nano- and
extreme-ultraviolet (EUV) lithography, and other emerg-
ing technologies, it is necessary to compute electromag-
netic elds in planar, layered media. The purpose of this
article is to present a succinct yet reasonably complete
introduction to this topic. The emphasis is on the devel-
opment of dyadic Green functions (DGFs), which make it
possible to compute the elds due to any conguration of
sources. In particular, a complete set of electric and mag-
netic DGFs is derived for planar, multilayered media, in
an efcient and convenient-to-use form, based on a trans-
mission-line equivalent circuit along the axis normal to
the stratication. Attention is limited to media with at
most uniaxial anisotropy, which, although important in
practice, still allow the simple transmission-line represen-
tation of the electromagnetic elds. Also, the analysis is
restricted to multilayers of innite lateral extent, al-
though a large part of the presented material can also be
adapted for laterally shielded geometries.
2. STATEMENT OF THE PROBLEM
Consider a plane-stratied medium excited by arbitrary
electric and magnetic currents (J, M)
1
occupying a volume
V, as illustrated in Fig. 1a. It is desired to determine the
resulting electric and magnetic elds (E, H) at an arbi-
trary location r. Of interest are also the far-zone radiation
elds and the elds excited in the multilayer by an
obliquely incident plane wave.
We assume that the stratication is perpendicular to
the z axis and of innite lateral extent along x and y. The
layers are numbered from 1 to N in the direction of in-
creasing z, where the number of layers N is arbitrary. The
nth layer, with interfaces at z z
n
and z z
n1
, is lled
with a homogeneous, linear, uniaxial medium [1, p. 5],
characterized by permittivity and permeability dyadics,
2
e
n
and l
n
, respectively. The optical axis of the medium
is assumed to be normal to the interfaces. The multilayer
may be shielded from below and/or above by plates that
present three possible boundary conditions: (1)
^ nnEZ
s
^ nn ^ nnH, where ^ nn is the interface normal
vector and Z
s
is a specied surface impedance; (2)
^ nnE0, which denes a perfect electric conductor
(PEC); and (3) ^ nnH0, which denes a perfect magnet-
ic conductor (PMC).
We approach this problem in the frequency domain,
which means that all eld quantities are phasors evalu-
ated at an implicitly understood radian frequency o. Once
a phasor [e.g., E(r)] is found, the corresponding physical,
time-domain eld can be obtained by the inverse Fourier
transform [3, p. 9]
3
Er; t
1
2p
_
1
1
Ere
jot
dt
Re
1
p
_
1
0
Ere
jot
dt
1
Also, we will initially assume that the medium properties
may vary along the z direction, which will obviate the use
of layer subscripts. Only at a later stage will we restrict
the medium to have piecewise-constant parameters along
the z axis. Under these assumptions, the elds are gov-
erned by Maxwells equations [4, p. 745]
=E jom
0
l
.
H M;
=Hjoe
0
e
.
EJ
2
where e
0
and m
0
are the free-space permittivity and per-
meability, respectively, and where
e I
t
e ^ zz^ zze
z
; l I
t
m ^ zz^ zzm
z
3
In the above, I
t
^ xx^ xx ^ yy^ yy is the transverse unit dyadic and
e and m denote the transverse, and e
z
and m
z
the longitu-
dinal permittivities and permeabilities, respectively, all
relative to free space. For lossy media, the dyadics e and l
become complex-valued.
To solve the problem before us, it will be helpful to in-
troduce the concept of a dyadic Green function (DGF),
4
as
follows. Consider a Hertzian dipole with moment I placed
at r
0
in a layered medium, as illustrated in Fig. 2. The
current density of this point source is given as
Jr I dr r
0
4
where d denotes the Dirac delta function [3, p. 568].
The electric eld due to this dipole at any point r can be
1
Throughout this article, vectors are denoted by boldface letters,
unit vectors are distinguished by carets, and dyadics (i.e., second-
rank Cartesian tensors) are denoted by doubly underlined bold-
face letters.
2
A symbol ab is called a dyad, and a sum of dyads is called a
dyadic [2, App. 3]. A dot product of a dyad and a vector is dened
as follows: ab
.
c ab
.
c; c
.
abc
.
ab.
3
The e
jot
time convention adopted here is prevalent in engineer-
ing disciplines. In physics, the alternative e
iot
convention is al-
most exclusively employed. The formulas presented here can be
converted to that convention by the substitution j-i.
4
Also referred to as Greens tensor.
ELECTROMAGNETIC FIELD COMPUTATION IN PLANAR MULTILAYERS 1163
expressed as
5
Er Grjr
0
.
I hGrjr
0
; Jr
0
i 5
where Grjr
0
is the (electric) DGF of the layered medium.
When expressed in Cartesian coordinates, Grjr
0
can be
thought of as a 3 3 matrix, whose columns render the x,
y, and z components of the electric eld at r, due to an x-,
y-, and z-oriented unit-strength dipole at r
0
[2, p. 17]. For
example, the x component of E(r), due to a unit-strength
y-oriented dipole at r
0
, is given as G
xy
rjr
0
^ xx
.
Grjr
0
.
^ yy.
The DGF fully characterizes the inuence of the layered
medium on the eld of the dipole. Hence, it rigorously in-
cludes all physical effects, such as multiple reections and
transmissions at the interfaces [5].
The DGF concept can be extended in an obvious way to
magnetic elds and also to magnetic currents. The mag-
netic current concept is useful in the modeling of small
coils [6], coaxial antenna feeds [7,8], as well as via holes
and slots in printed-circuit boards (PCBs) [9,10]. Since we
consider linear media, superposition applies,
6
and the
electric and magnetic elds due to arbitrary current
z
N
N1
N
(E, H )
m
n
2
1
z
N1
z
m+1
z
m
z
n+1
z
n
z
3
z
2
z
1
optional PEC, PMC, or surface
impedance plane
(a) (b)
= =
(c
N,
j
N
)
= =
(c
N1,
j
N1
)
= =
(c
m ,
j
m
)
= =
(c
n,
j
n
)
= =
(c
2,
j
2
)
= =
(c
1,
j
1
)
r
r'
(z)
(y)
(x)
O
V
J
M
Z
s
Z
s
:
V
m
(z)
:
I
m
(z)
Z
m
: :
Z
N1
:
Z
N
:
k
z1
:
k
z2
:
k
zn
:
k
z,N1
k
zm
:
k
zN
:
Z
1
:
Z
2
:
Z
n
:
i
:
(z)
V
:
(z)
+
+
Figure 1. Electric and magnetic currents in a
layered medium: (a) physical conguration;
(b) transmission-line analog.
E(r) = G(r | r
0
) Il
(z)
(x)
(y)
O
r
0
r
Il
=
Figure 2. Hertzian dipole with moment I located at r
0
in a lay-
ered medium. The eld at r is determined by the dyadic Green
function Grjr
0
and the orientation and amplitude of the dipole.
5
The notation /,S is used for integrals of products of two func-
tions separated by the comma over their common spatial support,
with a dot over the comma indicating a dot product. Source coor-
dinates are distinguished by primes.
6
The superposition principle states that, in a linear system, the
effect due to an aggregate of causes is the sum of the effects due to
all causes taken individually [11, p. 481].
1164 ELECTROMAGNETIC FIELD COMPUTATION IN PLANAR MULTILAYERS
distributions (J, M) may be expressed as
Er hG
EJ
rjr
0
; Jr
0
i hG
EM
rjr
0
; Mr
0
i 6
Hr hG
HJ
rjr
0
; Jr
0
i hG
HM
rjr
0
; Mr
0
i 7
where G
PQ
rjr
0
is the DGF relating P-type elds at r and
Q-type currents at r
0
. In view of the translational symme-
try of the medium with respect to the transverse coordi-
nates, these DGFs can also be expressed as
G
PQ
rjr
0
G
PQ
q q
0
; zjz
0
8
where q is the projection of r on the xy plane. Note that
this property does not hold in laterally shielded environ-
ments.
Let (J
a
, M
a
) and (J
b
, M
b
) be two separate source dis-
tributions, which give rise to the elds (E
a
, H
a
) and (E
b
,
H
b
), respectively. From the Maxwells equations (2), the
following reciprocity theorem can be established
7
[12, p.
50; 13, p. 22]:
M
a
; H
b
h i J
a
; E
b
h i M
b
; H
a
h i J
b
; E
a
h i 9
By substituting various combinations of dipole sources
and their corresponding responses (6) and (7) into the
above, the reciprocity properties of the DGFs can be de-
rived. For example, if we postulate M
a
J
b
0,
M
b
^ yy dr r
1
, and J
a
^ xxdr r
2
, we obtain
G
EM
xy
r
2
jr
1
G
HJ
yx
r
1
jr
2
. Proceeding in a similar fash-
ion, we can show that
G
EJ
rjr
0
G
EJ
r
0
jr
T
10
G
HM
rjr
0
G
HM
r
0
jr
T
11
G
EM
rjr
0
G
HJ
r
0
jr
T
12
One important use of reciprocity in electromagnetics is as
a check of equations and their computer implementation.
For example, from (10) we obtain G
EJ
yz
r
2
jr
1
G
EJ
zy
r
1
jr
2
,
which means that the y component of the electric eld
measured at r
2
, due to a unit-strength z-oriented dipole
at r
1
, is equal to the z component of the electric eld
measured at r
1
, due to a y-oriented unit-strength dipole
at r
2
(see Fig. 3). Reciprocity can also be used to obviate
some derivations and to reduce the programming effort.
For example, we note that in view of (12), G
EM
can be
computed from G
HJ
.
It should now be clear that the problem stated at the
beginning of this section, and illustrated in Fig. 1a, can be
reduced to that of nding the four DGFs of the layered
medium. In what follows, we will show that these DGFs
can be conveniently expressed in terms of the scalar Green
functions of the transmission line analogue of the medi-
um, illustrated in Fig. 1b, whose parameters will be de-
ned in due course. This transmission-line equivalent
circuit will also prove useful in the derivation of the far-
zone radiated elds, as well as the elds excited in the
multilayer by an incident plane wave.
3. TRANSMISSION-LINE ANALOG
Let the currents (J, M) radiate in a region lled with a
medium characterized by z-dependent parameters e ez
and llz, as illustrated in Fig. 4a. For this problem, we
will derive a transmission-line equivalent circuit depicted
in Fig. 4b. Since the medium is homogeneous and of in-
nite extent in any transverse (to z) plane, the analysis is
facilitated by the Fourier transformation of all elds with
respect to the transverse coordinates, dened as follows
Ff r
~
ff k
r
; z
_
1
1
_
1
1
f re
jkr
.
q
dx dy 13
F
1
~
ff k
r
; z f r
1
2p
2
_
1
1
_
1
1
~
ff k
r
; ze
jkr
.
q
dk
x
dk
y
14
where
q ^ xxx ^ yyy; k
r
^ xxk
x
^ yyk
y
15
Hence, we apply (13) to Maxwells equations (2), noting
that the operator nabla transforms as = !jk
r
^ zzd=dz.
Furthermore, we separate the transverse and longitudinal
r
2
r
1
O O
Measurement
dipole
(z) (z)
(y) (y)
(x) (x)
Source
dipole
(a)
r
2
r
1
Measurement
dipole
Source
dipole
(b)
Figure 3. Illustration of reciprocity. The eld
measurements in situations (a) and (b) are iden-
tical.
7
This holds provided the permittivity and permeability dyadics
are symmetric, e e
T
and l l
T
, where the superscript T indi-
cates a transposed dyadic. This condition is clearly satised for
the uniaxial media considered here.
parts of the resulting equations, to obtain [14]
d
dz
~
E E
t
1
joe
0
e
k
2
1
n
e
k
r
k
r
.
_ _
~
HH
t
^ zz
k
r
~
JJ
z
oe
0
e
z
~
MM
t
^ zz
16
d
dz
~
HH
t
1
jom
0
m
k
2
1
n
h
k
r
k
r
.
_ _
~ zz
~
EE
t
k
r
~
MM
z
om
0
m
z
z
~
JJ
t
17
joe
0
e
z
~
EE
z
jk
r
.
~
HH
t
^ zz
~
JJ
z
18
jom
0
m
z
~
HH
z
jk
r
.
~ zz
~
EE
t

~
MM
z
19
where
kk
0
em
p
;
n
e
e
z
e
;
n
h
m
z
m
20
and where k
0
o

m
0
e
0
p
is the free-space wavenumber, k is
the medium wavenumber, and n
e
and n
h
are the electric
and magnetic anisotropy ratios, respectively. It is under-
stood that all media parameters may be z-dependent.
The subsequent analysis is greatly simplied if one de-
nes a spectral-domain coordinate system based on
k
r
; ^ zz k
r
, as illustrated in Fig. 5, with the unit vectors
^ uu; ^ vv given by [15]
^ uu
k
r
k
r
^ xxcos x ^ yy sinx;
^ vv
^ zz k
r
k
r
^ xxsinx ^ yy cos x
21
where x is the angle the vector k
r
makes with the positive
x axis and k
r
k
2
x
k
2
y
_
. In this system, the transverse
electric and magnetic elds can be expressed as [16,17]
~
EE
t
^ uuV
e
^ vvV
h
; ^ zz
~
EE
t
^ uuV
h
^ vvV
e
22
~
HH
t
^ zz ^ uuI
e
^ vvI
h
;
~
HH
t
^ uuI
h
^ vvI
e
23
On substituting this into (16), (17) and projecting the re-
sulting equations on ^ uu and ^ vv, we obtain two sets of trans-
mission-line (TL) equations of the form
dV
a
dz
jk
a
z
Z
a
I
a
v
a
;
dI
a
dz
jk
a
z
Y
a
V
a
i
a
; a e; h
24
where
k
a
z
k
2
k
2
r
=n
a
_
25
Z
e
1
Y
e

k
e
z
oe
0
e
; Z
h
1
Y
h

om
0
m
k
h
z
26
and where the source functions v
a
and i
a
are as given in
Table 1. Hence, as anticipated by the notation introduced
in (22), (23), the components of
~
EE
t
and
~
HH
t
in the spectral
uv plane may be interpreted as voltages and currents
8
on
a transmission-line analog of the medium along the z axis,
as illustrated in Fig. 4b. This analog comprises two trans-
mission lines, with z-dependent propagation constants
k
a
z
and characteristic impedances Z
a
, where a (e,h). Since
(k
y
)
k
y
k
x
kj
(k
x
)
z kj
v
(z)
x
y
u
Figure 5. Spectral-domain coordinate system.

(z)
(b)
0 z
Z
:
(z)
I
:
(z) v
:
(z)
i
:
(z)
V
:
(z)
k
z
(z)
:
+
+
j = (x, y)
V
J
M
r' r
z' z (z)
(E, H )
(a)
(c (z),j(z))
= =
o
Figure 4. Currents in a uniaxial medium with z-dependent pa-
rameters: (a) physical conguration; (b) transmission-line analog.
8
Note, however, that these voltages and currents have the
units of V/m and A/m, respectively.
the longitudinal eld components may be found from the
transverse components and the eld sources by means of
(18), (19), the original vector problem of Fig. 4a has in ef-
fect been reduced to the scalar TL problem of Fig. 4b. In
fact, combining (22), (23) and (18), (19), we obtain the
complete spectral domain elds as
~
EEk
r
; z ^ uuV
e
z ^ vvV
h
z ^ zz
1
joe
0
e
z
z
jk
r
I
e
z
~
JJ
z
k
r
; z
27
~
HHk
r
; z ^ uuI
h
z ^ vvI
e
z ^ zz
1
jom
0
m
z
z
jk
r
V
h
z
~
MM
z
k
r
; z
28
where V
a
(z) and I
a
(z) implicitly depend on k
r
(but not on x).
We observe that, outside the source region, (V
e
, I
e
) and (V
h
,
I
h
) represent elds that are transverse magnetic (TM) and
transverse electric (TE) to z, respectively. We also note
from Table 1 that transverse x- or y-directed electric and
magnetic currents, always excite both the TM and TE
transmission lines, whereas longitudinal z-directed elec-
tric (magnetic) currents excite only the TM (TE) trans-
mission line.
The preceding three-dimensional (3D) analysis can be
specialized to the two-dimensional (2D) case, where there
is a translational symmetry along one of the transverse
axessay, the y axis, and it may be assumed that the sca-
lar components of all elds and sources have the form
9
f r f
2D
x; b; ze
jby
29
where b is a known or sought-after propagation constant
[1822 (Chap. 6 in Ref. 20]). When (29) is substituted into
(13), (14), one obtains
Ff r 2pd k
y
b
~
ff
2D
k
r
; z 30
where
~
ff
2D
k
r
; z
_
1
1
f
2D
x; k
y
; z e
jkxx
dx 31
and
f
2D
x; k
y
; z
1
2p
_
1
1
~
ff
2D
k
r
; z e
jk
x
x
dk
x
32
Hence, the transmission-line analog derived here is still
applicable in 2D, with the understanding that the two-di-
mensional Fourier transform pair should be replaced by
the one-dimensional transforms (31), (32), with k
y
b.
4. TRANSMISSION-LINE GREEN FUNCTIONS
To nd the transmission-line voltages and currents ap-
pearing in (27), (28), it will be helpful to introduce trans-
mission-line Green functions (TLGFs) as the voltages and
currents excited by unit-strength impulsive sources.
Hence, let V
i
(z|z
0
) and I
i
(z|z
0
) denote, respectively, the
voltage and current at z, due to a 1-A shunt current source
at z
0
, and let V
v
(z|z
0
) and I
v
(z|z
0
) denote, respectively, the
voltage and current at z, due to a 1-V series voltage source
at z
0
, as illustrated in Fig. 6.
10
It then follows from (24)
that the so-dened TLGFs satisfy the equations
dV
i
dz
jk
z
ZI
i
;
dI
i
dz
jk
z
YV
i
dz z
0
33
dV
v
dz
jk
z
ZI
v
dz z
0
;
dI
v
dz
jk
z
YV
v
34
In view of the linearity of the TL equations (24), we can
use superposition
11
to express V and I at any point z as
Vz V
i
zjz
0
; iz
0
_
V
v
zjz
0
; vz
0
_
35
Iz I
i
zjz
0
; iz
0
_
I
v
zjz
0
; vz
0
_
36
The complete analogy with (6), (7) should be noted.
Let (v
a
,i
a
) and (v
b
,i
b
) be two separate TL source distri-
butions, which give rise to (V
a
,I
a
) and (V
b
,I
b
), respectively.
Since both sets satisfy the TL equations (24) subject to the
same boundary conditions, it can be shown that [4; p. 194;
24, p. 147]
v
a
; I
b
h i i
a
; V
b
h i v
b
; I
a
h i i
b
; V
a
h i 37
which is the transmission-line counterpart of the reciproc-
ity theorem (9). By substituting various combinations of
point sources and their corresponding responses (35), (36)
into the above, it is found that the TLGFs possess the re-
Table 1. Transmission-Line Sources in Terms of Field
Sources
a e (TM) a h (TE)
v
a
k
r
oe
0
e
z
~
JJ
z
~
MM
v
~
MM
u
i
a
~
JJ
u

k
r
om
0
m
z
~
MM
z
~
JJ
v
9
The term 2.5D is occasionally used in this context, with the 2D
designation reserved for the b 0 case.
10
Here and throughout this article, we omit the superscript a,
whenever the equations apply to both the TM and TE transmis-
sion lines.
11
The principle of superposition for linear networks can be stated
as follows. The response due to several independent voltage and
current sources is equal to the sum of the responses due to each
independent source acting alone, that is, with all other indepen-
dent sources made inoperative. In the case of distributed sources,
the sums become integrals [23, p. 63].
ciprocity properties
V
i
zjz
0
V
i
z
0
jz; I
v
zjz
0
I
v
z
0
jz;
V
v
zjz
0
I
i
z
0
jz
38
which are the transmission-line counterparts of (10)(12).
The TL equations (33), (34) give rise to second-order
differential equations of the SturmLiouville type
[4, p. 278; 25, p. 53; 26, p. 291]. For example, equations
(33) imply
d
dz
Pz
d
dz
Qz
_ _
V
i
zjz
0
dz z
0
39
with
I
i
zjz
0
P
d
dz
V
i
zjz
0
40
where we have introduced the notation
P
1
jk
z
Z
; Q
jk
z
Z
41
Assuming that the solution domain is restricted to
z
n
rzrz
n1
, to correspond to the nth layer in Fig. 1, we
impose the boundary conditions
V
i
I
i
_ _
z z
n
Z
n
;
V
i
I
i
_ _
z z
n1
Z
!
n
42
where Z
n
and Z
!
n
denote the left-looking and right-
looking impedances at z
n
and z
n1
, respectively. Let
V
z and V
!
z be two homogeneous solutions of (39) that
satisfy the boundary conditions at the left and at the right
ends of the solution domain, respectively. Then, the solu-
tion of (39), subject to (42), can be found as
V
i
zjz
0

V
z
o
V
!
z
>
PW V
; V
!
43
where z
o
min(z, z
0
), z
4
max(z, z
0
), and where
W V
; V
!
V
dV
!
dz
V
!dV
dz
44
is the Wronskian determinant. It can be shown that the
denominator in (43), which is referred to as the conjunct, is
independent of z. The solution of (34) follows by a dual pro-
cedure, in which voltages are replaced by currents, imped-
ances by admittances, and vice versa.
12
The simple and elegant result (43) clearly depends
on the availability of the homogeneous solutions V
z
and V
!
z in an analytical closed form. In Section 7,
we derive such solutions for the most practically impor-
tant case of piecewise homogeneous media.
5. SPECTRAL-DOMAIN DYADIC GREEN FUNCTIONS
On substituting (35), (36) into (27), (28) and referring to
Table 1, one obtains, after some simple transformations,
the spectral domain counterparts of (6), (7), specifically
~
EEk
r
; z h
~
GG
EJ
k
r
; zjz
0
;
~
JJk
r
; z
0
i
h
~
GG
EM
k
r
; zjz
0
;
~
MMk
r
; z
0
i
45
~
HHk
r
; z h
~
GG
HJ
k
r
; zjz
0
;
~
JJk
r
; z
0
i
h
~
GG
HM
k
r
; zjz
0
;
~
MMk
r
; z
0
i
46
with the spectral DGFs given as
13
~
GG
EJ
k
r
; zjz
0
^ uu^ uuV
e
i
^ vv^ vvV
h
i
^ zz ^ uu
k
r
oe
0
e
z
I
e
i
^ uu^ zz
k
r
oe
0
e
0
z
V
e
v
^ zz^ zz
1
joe
0
e
0
z
k
2
r
joe
0
e
z
I
e
v
dz z
0
_ _
47
~
GG
EM
k
r
; zjz
0
^ uu^ vvV
e
v
^ vv ^ uuV
h
v
^ zz^ vv
k
r
oe
0
e
z
I
e
v
^ vv^ zz
k
r
om
0
m
0
z
V
h
i
48
(a)
k
z
(z)

Z(z)

1A
z' z (z)
V
i
(z | z' )
I
i
(z |z' )
+
(b)
k
z
(z)

Z(z)

1V
z' z (z)
V
v
(z | z' )
I
v
(z | z' )
+
+
Figure 6. Network problems for determination
of transmission-line Green functions: (a) current
source excitation; (b) voltage source excitation.
12
Two equations of the same mathematical form are called dual
equations, and quantities occupying the same position in dual
equations are called dual quantities. When two equations are
duals of each other, a systematic interchange of symbols changes
the rst equation into the second, and vice versa [27, p. 98].
13
It is understood that the TLGFs in these equations depend on z
and z
0
, such as V
e
i
V
e
i
zjz
0
, and also (implicitly) on k
r
.
~
GG
HJ
k
r
; zjz
0
^ uu^ vvI
h
i
^ vv ^ uuI
e
i
^ zz^ vv
k
r
om
0
m
z
V
h
i
^ vv^ zz
k
r
oe
0
e
0
z
I
e
v
49
~
GG
HM
k
r
; zjz
0
^ uu^ uuI
h
v
^ vv^ vvI
e
v
^ zz ^ uu
k
r
om
0
m
z
V
h
v
^ uu^ zz
k
r
om
0
m
0
z
I
h
i
^ zz^ zz
1
jom
0
m
0
z
k
2
r
jom
0
m
z
V
h
i
dz z
0
_ _
50
where the primed and unprimed media parameters are
evaluated at z
0
and z, respectively. It can readily be con-
rmed, using (38), that these DGFs possess the reciprocity
properties
~
GG
EJ
k
r
; zjz
0

~
GG
EJ
k
r
; z
0
jz
T
51
~
GG
HM
k
r
; zjz
0

~
GG
HM
k
r
; z
0
jz
T
52
~
GG
EM
k
r
; zjz
0

~
GG
HJ
k
r
; z
0
jz
T
53
which are the counterparts of the spatial-domain relations
(10)(12).
The spectral-domain electric DGF (47) has been widely
used in the analysis of planar microstrip antennas,
resonators, and transmission lines by the spectral-domain
approach (SDA) [16,2833]. For planar circuits with
innitesimal metallization thickness, only the transverse
part of
~
GG
EJ
is needed. This transverse dyadic, which has a
particularly simple form, can also be expressed as [34]
~
GG
EJ
t
k
r
; zjz
0
I
t
V
h
i
^ uu^ uuV
e
i
V
h
i

1
2
I
t
V
e
i
V
h
i

1
2
2 ^ uu^ uu I
t
V
e
i
V
h
i

54
where we have used the fact that I
t
^ uu^ uu ^ vv^ vv.
To demonstrate the application of the SDA, consider a
half-space medium excited by a coaxial transmission line
opening into a PEC ange at z z
1
, as illustrated in Fig. 7.
The medium may be uniaxial, with z-dependent permit-
tivity and permeability dyadics. It is desired to nd the
electric and magnetic elds in the half-space, assuming
that the coax (coaxial TL) propagates the dominant TEM
mode. This problem, which is relevant to coaxial feeds for
microstrip antennas, PCB vias, and noninvasive testing,
will also serve to introduce some results and notation
needed in the subsequent development. We assume, with-
out loss of generality, that the coax is concentric with the z
axis. The eld for z4z
1
can be found from the equivalent
problem depicted in Fig. 7a, where the annular coax ap-
erture has been short-circuited and covered by an equiv-
alent magnetic surface current M
S
E
a
^ zz, where E
a
is
the aperture electric eld in the original problem. Al-
though E
a
is not known, its distribution may be approx-
imated by that existing in the cross section of an innite
coax, which results in the azimuthally oriented frill cur-
rent [27, p. 112]
M
S
q ^ uu
K
r
; aorob; z z
1
55
with K V
c
/log(b/a), where V
c
is the voltage at the
mouth of the coax.
14
Noting that the corresponding vol-
ume current density is M(r) M
S
(q)d(z z
1
), we may use
(E,H)
r
r
(y)
(z)
(x)
a
b
PEC
Magnetic
frill current
[ c(z),j(z)]
z
1
z
1
z
1
K
z
(z)
e
Z
e
(z)
V
e
(z)
I
e
(z)
+
+
+
v
e
Short
O
Figure 7. Coax-excited half-space medium:
(a) computational model showing an equivalent
magnetic surface current over a shorted-cir-
cuited annular aperture: (b) spectral-domain
transmission-line equivalent circuit.
14
We will assume that V
c
is known, although this seldom is the
case. In general, one must resort to numerical techniques to de-
termine the aperture eld [35].
(6) and (8) to express the electric eld in the half-space as
Er hG
EM
q q
0
; zjz
1
; M
S
q
0
i
F
1
f
~
GG
EM
k
r
; zjz
1
.
~
MM
S
k
r
g
56
where the second equality on the right side is the result of
the well-known fact that a spatial-domain convolution of
two functions corresponds to a product of their transforms
in the spectral domain [36, p. 23]. Similarly, the magnetic
eld may be expressed as
Hr hG
HM
q q
0
; zjz
1
; M
S
q
0
i
F
1
f
~
GG
HM
k
q
; zjz
1
.
~
MM
S
k
r
g
57
To nd the spectral-domain magnetic current density
~
MM
S
,
we apply the Fourier transformation (13) to (55). The in-
tegration is facilitated by changing to polar coordinates in
both domains, namely, (x, y)-(r, j) and (k
x
, k
y
)-(k
r
, x),
and invoking an integral representation of the Bessel
function of order n
J
n
z
j
n
2p
_
p
p
e
jz cosjW njW
dj 58
where W is an arbitrary angle [37, p. 106].
15
From (58), we
derive the formula
1
2p
_
p
p
cos
sin
nje
jk
r
rcosjx
djj
n
J
n
k
r
r
cos
sin
nx 59
and apply it to evaluate the j integral in a closed form.
The r integration is also evaluated in a closed form,
16
with
the nal result
~
MM
S
k
r
^ vv2pj
K
k
p
J
0
k
r
a J
0
k
r
b
..
$k
r
^ vv
~
MM
S
k
r
60
where ^ vv is the unit vector dened in (21). Note that, as
evident from Table 1, this source excites only the TM
transmission line, as illustrated in Fig. 7b, where
v
e

~
MM
S
. Finally, we substitute (60) together with (48)
and (50) into (56), (57) and apply the inverse Fourier
transformation (14) to obtain the desired electric and mag-
netic elds. The x integrals are evaluated in closed forms
by means of the formula
1
2p
_
p
p
cos
sin
nxe
jk
r
r cosxj
dx j
n
J
n
k
r
r
cos
sin
nj 61
which can also be derived from (58). As a result, the non-
zero eld components are found as [8]
E
r
r 2pKS
1
$k
r
k
r
V
e
v
zjz
1
_ _
62
E
z
r 2pK
jZ
1
k
1
n
e
1
S
0
f$k
r
I
e
v
zjz
1
g 63
H
j
r 2pKS
1
$k
r
k
r
I
e
v
zjz
1
_ _
64
with the notation
S
n
f
~
ff k
r
g
1
2p
_
1
0
~
ff k
r
J
n
k
r
rk
r
dk
r
1
4p
_
1
1e
jp
~
ff k
r
H
2
n
k
r
rk
r
dk
r
65
where H
2
n
is the Hankel function of the second kind
and order n.
17
The integrals in (65) are recognized as
Hankel (or Fourier-Bessel) transforms, but in the electro-
magnetics community they are generally known as
Sommerfeld integrals [40,41]. The real-axis integration
paths in (65) must be properly indented around the singu-
larities of
~
ff k
r
, as discussed in Section 8. Note that (62)
(64) can be used, in particular, when the upper half-space
in Fig. 7 is a layered medium with piecewise constant pa-
rameters. The TLGFs for this case are derived in Section 7.
6. SPATIAL-DOMAIN DYADIC GREEN FUNCTIONS
The spatial-domain counterparts of (47)(50) are obtained
by the Fourier inversion
G
PQ
q; zjz
0
F
1
~
GG
PQ
k
r
; zjz
0
66
which is performed by rst projecting the unit vectors
^ uu; ^ vv on the xy coordinate system via (21) and then ap-
plying the transformation (14). Since all spectral inte-
grands are of the form sinnx
~
ff k
r
or cosnx
~
ff k
r
, with
n0, 1, or 2, the integration is facilitated by changing to
polar coordinates in both domains and using the formula
F
1
cos
sin
nx
~
ff k
r
_ _
j
n
cos
sin
njS
n
~
ff k
r
_ _
67
which follows from (61) and (65). We note that the second-
order Hankel transforms in (67) can be expressed as
S
2
f
~
ff k
r
g
2
r
S
1
~
ff k
r
k
r
_ _
S
0
f
~
ff k
r
g 68
15
This formula is also valid with n in the exponent of the inte-
grand changed to n, which can be conrmed by using the re-
lationship J
n
(z) ( 1)
n
J
n
(z) [38, p. 358]. Also, in view of the
periodicity of the integrand, the limits of integration can be
changed to any 2p range.
16
This is done by invoking the formula J
1
z J
0
0
z [38, p. 361].
17
The second integral form in [65] can be derived from the rst by
using the formulas J
n
z
1
2
H
1
n
z H
2
n
z and H
1
n
ze
jp

e
jnp
H
2
n
z [3, p. 453; 39, p. 231], under the assumption that
the integrand functions
~
ff k
r
are even (odd) in k
r
for even (odd) n.
Hence only the Hankel transforms of orders 0 and 1 are
needed.
18
Using (67), (68) in (66), we obtain the non-zero
Cartesian components of the space-domain DGFs as
G
EJ
xx
q; zjz
0
cos
2
jS
0
fV
e
i
g sin
2
jS
0
fV
h
i
g
cos 2j
r
S
1
V
e
i
V
h
i
k
r
_ _
69
G
EJ
xx
q; zjz
0
G
EJ
yx
q; zjz
0

sin2j
2
S
0
fV
e
i
V
h
i
g
sin2j
r
S
1
V
e
i
V
h
i
k
r
_ _
70
G
EJ
xz
q; zjz
0

Z
0
jk
0
e
0
z
cos jS
1
fk
r
V
e
v
g 71
G
EJ
yy
q; zjz
0
sin
2
jS
0
fV
e
i
g cos
2
jS
0
fV
h
i
g
cos 2j
r
S
1
V
e
i
V
h
i
k
r
_ _
72
G
EJ
yz
q; zjz
0

Z
0
jk
0
e
0
z
sinjS
1
fk
r
V
e
v
g 73
G
EJ
zx
q; zjz
0

Z
0
jk
0
e
z
cos jS
1
fk
r
I
e
i
g 74
G
EJ
zy
q; zjz
0

Z
0
jk
0
e
z
sinjS
1
fk
r
I
e
i
g 75
G
EJ
zz
q; zjz
0

Z
2
0
k
2
0
e
z
e
0
z
S
0
fk
2
r
I
e
v
g
Z
0
jk
0
e
z
dqdz z
0
76
G
EM
xx
q; zjz
0
G
EM
yy
q; zjz
0

sin2j
2
S
0
fV
e
v
V
h
v
g
sin2j
r
S
1
V
e
v
V
h
v
k
r
_ _
77
G
EM
xy
q; zjz
0
cos
2
jS
0
fV
e
v
g sin
2
jS
0
fV
h
v
g
cos 2j
r
S
1
V
e
v
V
h
v
k
r
_ _
78
G
EM
xz
q; zjz
0

1
jk
0
Z
0
m
0
z
sinjS
1
fk
r
V
h
i
g 79
G
EM
yx
q; zjz
0
sin
2
jS
0
fV
e
v
g cos
2
jS
0
fV
h
v
g
cos 2j
r
S
1
V
e
v
V
h
v
k
r
_ _
80
G
EM
yz
q; zjz
0

1
jk
0
Z
0
m
0
z
cos jS
1
fk
r
V
h
i
g 81
G
EM
zx
q; zjz
0

Z
0
jk
0
e
z
sinjS
1
fk
r
I
e
v
g 82
G
EM
zy
q; zjz
0

Z
0
jk
0
e
z
cos jS
1
fk
r
I
e
v
g 83
G
HJ
xx
q; zjz
0
G
HJ
yy
q; zjz
0

sin2j
2
S
0
fI
h
i
I
e
i
g
sin2j
r
S
1
I
h
i
I
e
i
k
r
_ _
84
G
HJ
xy
q; zjz
0
cos
2
jS
0
fI
h
i
g sin
2
jS
0
fI
e
i
g
cos 2j
r
S
1
I
h
i
I
e
i
k
r
_ _
85
G
HJ
xz
q; zjz
0

Z
0
jk
0
e
0
z
sinjS
1
fk
r
I
e
v
g 86
G
HJ
yx
q; zjz
0
sin
2
jS
0
fI
h
i
g cos
2
jS
0
fI
e
i
g
cos 2j
r
S
1
I
h
i
I
e
i
k
r
_ _
87
G
HJ
yz
q; zjz
0

Z
0
jk
0
e
0
z
cos jS
1
fk
r
I
e
v
g 88
G
HJ
zx
q; zjz
0

1
jk
0
Z
0
m
z
sinjS
1
fk
r
V
h
i
g 89
G
HJ
zy
q; zjz
0

1
jk
0
Z
0
m
z
cos jS
1
fk
r
V
h
i
g 90
G
HM
xx
q; zjz
0
cos
2
jS
0
fI
h
v
g sin
2
jS
0
fI
e
v
g
cos 2j
r
S
1
I
h
v
I
e
v
k
r
_ _
91
G
HM
xy
q; zjz
0
G
HM
yx
q; zjz
0

sin2j
2
S
0
fI
h
v
I
e
v
g
sin2j
r
S
1
I
h
v
I
e
v
k
r
_ _
92
G
HM
xz
q; zjz
0

1
jk
0
Z
0
m
0
z
cos jS
1
fk
r
I
h
i
g 93
G
HM
yy
q; zjz
0
sin
2
jS
0
fI
h
v
g cos
2
jS
0
fI
e
v
g
cos 2j
r
S
1
I
h
v
I
e
v
k
r
_ _
94
G
HM
yz
q; zjz
0

1
jk
0
Z
0
m
0
z
sinjS
1
fk
r
I
h
i
g 95
G
HM
zx
q; zjz
0

1
jk
0
Z
0
m
z
cos jS
1
fk
r
V
h
v
g 96
18
This follows from the recurrence relation J
n1
z
2n=zJ
n
z J
n1
z [38, p. 361].
G
HM
zy
q; zjz
0

1
jk
0
Z
0
m
z
sinjS
1
fk
r
V
h
v
g 97
G
HM
zz
q; zjz
0

1
k
2
0
Z
2
0
m
z
m
0
z
S
0
fk
2
r
V
h
i
g
1
jk
0
Z
0
m
z
dqdz z
0
98
where Z
0
m
0
=e
0
_
denotes the intrinsic impedance of free
space. Although these expressions are given for source
points on the z axis, they are readily generalized by the
substitutions x-(x x
0
), y-(y y
0
), and
r !R
x x
0
2
y y
0
2
_
;
j !f arctan
y y
0
x x
0
99
where the quadrant of the argument must be noted when
the inverse tangent is evaluated.
19
The new cylindrical
coordinates (R, f) are illustrated in Fig. 8. It may be di-
rectly veried, using (38), that the resulting DGFs satisfy
the reciprocity relations (10)(12).
20
It will be instructive to briefly revisit the problem of
Fig. 7 using the space-domain approach. In view of (55),
the electric and magnetic elds in the half-space can be
expressed as
Er K
_
b
a
dr
0
_
p
p
dj
0
G
EM
q q
0
; zjz
1
.
^ uu
0
100
Hr K
_
b
a
dr
0
_
p
p
dj
0
G
HM
q q
0
; zjz
1
.
^ uu
0
101
where G
EM
and G
HM
are as given by (77)(83) and (91)
(98), respectively. Note that the integrals above must be
evaluated by numerical quadratures. These equations can
provide an excellent check for the DGFs and the numer-
ical procedures used to evaluate the Sommerfeld integrals,
since the results must agree with those computed from the
spectral-domain formulas (62)(64) [43]. It is also possible
to demonstrate analytically the equivalence between
(100), (101) and (62)(64); consider H
x
, for example.
From (101), we obtain
H
x
r K
_
b
a
dr
0
_
p
p
dj
0
sinj
0
G
HM
xx
q q
0
; zjz
1
cos j
0
G
HM
xy
q q
0
; zjz
1
2pK
_
b
a
dr
0
_
p
p
dj
0
1
4p
sinj
0
S
0
fI
e
v
I
h
v
g
sin2f j
0
S
2
fI
e
v
I
h
v
g
102
where the Hankel transforms S
0
and S
2
are as given
by (65) with r replaced by R, and where we have omitted
the arguments (z|z
1
) of the TLGFs for notational simplic-
ity. Note that we have used (68) to simplify the second
integral expression above. If we now change the order of
the spatial and spectral integrals in the last equation, we
obtain
H
x
r 2pK
1
2p
_
1
0
dk
r
k
r
_
b
a
dr
0
_
p
p
dj
0
1
4p
sinj
0
J
0
k
r
RfI
e
v
I
h
v
g
sin2f j
0
J
2
k
r
RfI
e
v
I
h
v
g
103
Next, we expand the Bessel functions in (103) using Graf s
addition theorem [38, p. 363]
J
n
k
r
R
cos
sin
nw
1
k 1
J
k
k
r
rJ
nk
k
r
r
0
cos
sin
ka 104
where the angles w p(fj
0
) and a (jj
0
) are indi-
cated in Fig. 8, which makes it possible to evaluate the
spatial integrals in a closed form. As a result of the or-
thogonality of the trigonometric functions over the 2p
range, many cancellations occur, leading to the remark-
ably simple nal result
H
x
r 2pK sinjS
1
$k
r
k
r
I
e
v
zjz
1
_ _
105
where $k
r
has been dened in (60). The other eld com-
ponents in (100), (101) can be transformed in a similar
fashion, to arrive at (62)(64).
The discussion above pertains to the general, 3D case.
In 2D, where all eld components have the form of (29),
the space-domain DGFs can be obtained by applying the
one-dimensional inverse transform (32) to (47)(50), with
k
y
set to b and x replaced by (x x
0
). Hence, the spectral
integrals that arise in the 2D DGFs are ordinary Fourier
transforms, rather than Sommerfeld integrals, with the
(x)
(z)
(y)
y
x
y
x
[
jj
j
j
j
:
p
Figure 8. Cylindrical coordinates for off-axis source points.

19
The intrinsic FORTRAN function ATAN2 does this automati-
cally and returns a result in the range ( p, p] [42, p. 178]. Note
that under this transformation dqdz z
0
!dr r
0
.
20
Note that f-p f when the locations of the source and eld
points in Fig. 8 are swapped, whereas R is unaffected.
integrands expressed in terms of the TLGFs, exactly as in
the 3D case [44].
We note that our Fourier representation of the DGFs is
tantamount to a two-dimensional eigenfunction expansion
in the transverse to z plane, with a closed-form solution in
the z direction [45]. The appearance of the delta function
terms in (76) and (98) is consistent with such z-propagat-
ing eigenfunction representations [4648].
21
However,
these terms do not represent the total source region sin-
gularity of the DGFs, since the remaining terms are still
highly singular when R|r r
0
|-0 [50]. The level of
these singularities is the same in layered media as in a
homogeneous space (see the Appendix at the end of this
article). In layered media, however, the DGFs are given in
terms of the Sommerfeld integrals, and there is a direct
relationship between the large-k
r
divergence of the spec-
tral integrands and the space-domain singularity, as dis-
cussed in the next section. It is found that in the source
region G
EM
and G
HJ
behave as R
2
, while G
EJ
and G
HM
exhibit more severe R
3
singularities. Since such singu-
larities are not generally integrable over a volume, special
care has to be taken when the integrals (6), (7) are eval-
uated with rAV [51]. The classical approach is to exclude
from V an innitesimal volume V
d
surrounding the eld
point [5254], [12, p. 101]. The integral over VV
d
, which
is referred to as the principal value (PV) integral, is then
convergent, and the integral over V
d
gives rise to the so-
called depolarization dyadic. Although both the PV inte-
gral and the depolarization dyadic depend on the shape of
the exclusion volume, the sum of the two contributions is
unique.
22
The depolarization dyadic for several simple
shapes, such as spherical, cubical, and cylindrical (with
a needle shape and a disk as special cases), have been
evaluated for both isotropic [54] and uniaxial [57] media.
23
An alternative approach is regularization, which uses the
static form of the kernel to compensate for the singularity
[5962].
Still another way to regularize the integral represen-
tations with hypersingular DGF kernels is to convert
them into the so-called mixed-potential forms, which ex-
press the elds in terms of the vector and scalar potentials
[14,43,6371]. Such potential representations, although
not unique, have the advantage of being less singular (in
space domain) and faster convergent (in spectral domain)
than the eld forms. The mixed-potential integral equa-
tions (MPIEs) have been frequently used in the analysis of
antennas, printed-circuit boards, arbitrarily shaped scat-
terers, and other structures embedded in layered media.
Space limitations do not allow us to develop this important
topic further here.
6.1. Comparison with Other Approaches
Before leaving this section, we wish to point out some sa-
lient features of the formulation presented here and to
comment on other approaches. First, we note that the
DGFs have been scalarized; that is, they are expressed
in terms of the scalar Green functions of the transmission-
line analog of the layered medium along the axis normal to
the stratication [16,7274]. The TLGFs are associated
with the transverse eld components, whose continuity at
the media interfaces naturally translates into the conti-
nuity of the voltages and currents on the transmission-
line network representation of the layered medium. As
indicated in Fig. 1b, two transmission-line networks arise
that represent the TM (a e) and TE (a h) partial elds.
However, since the TM and TE networks have identical
congurations and differ only in the propagation constants
and characteristic impedances, it sufces to work out the
solution for just one of them. The advantage of the TL
representation is that it enhances the engineering insight
and is amenable to the familiar network analysis tech-
niques. By using these techniques, the TLGFs for any
number of layers can easily be found, as discussed in the
next section. Hence, the DGF expressions (69)(98) are
general, applicable to arbitrary source congurations and
any number of layers. They are also in a convenient ready
to use form, which cannot be further simpliedexcept,
possibly, by invoking (68).
24
The transmission-line analogue employed here is
based on the transverse eld components and it treats
the longitudinal components as dependent variables [4, p.
745]. However, it is also possible to scalarize the Maxwell
equations in layered media by taking the longitudinal
eld components, (E
z
, H
z
), as the independent variables
(or wave potentials), from which the transverse compo-
nents are derived [7577]. A closely related approach is
based on the longitudinal components of either the mag-
netic and electric vector potentials (A
z
, F
z
), or the Hertz
vectors (P
z
, P
z
) [29,78, p. 242; 7982]. Although such rep-
resentations achieve the TM-TE decomposition of the
elds, they do not naturally lead to transmission-line
equivalent circuits, because the chosen wave potentials
are not continuous across the boundaries between dis-
similar media. Perhaps to overcome this drawback, a for-
mulation based on the longitudinal components of the
electric and magnetic displacement vectors (D
z
, B
z
) has
also been introduced [83].
Some authors have adopted Sommerfelds original ap-
proach [40,84,85], based on the horizontal and vertical
components of the Hertz vector (P
x
, P
z
)
25
[8891]. How-
ever, the extension of this method, or its variant based on
21
The occurrence of the Dirac deltas in G
EJ
and G
HM
indicates
that these DGFs belong to the class of distributions (or general-
ized functions) [49, Chap. 1]. In the present context, a distribution
may be dened as a function, which has meaning only when in-
tegrated against another, well-behaved function.
22
It should be mentioned here that in the special case of a uniform
current and a cylindrical volume with the axis parallel to the di-
rection of current ow, it is possible to evaluate the electric eld
inside the source region without invoking the concepts of exclu-
sion volume or principal value [55,56].
23
It is interesting to note that the delta function terms in (76) and
(98) give rise to depolarization dyadics corresponding to a disk-
shaped exclusion volume perpendicular to the z axis [58].
24
However, using S
2
to simplify equations does not necessarily
lead to a more efcient formulation, because J
2
is usually com-
puted from J
0
and J
1
.
25
It has been realized that Sommerfelds choice of Hertz poten-
tials is nonunique, and that one could also use, for example, (P
x
,
P
y
) [86,87](86), (87).
(A
x
, A
z
), to multilayered media is cumbersome, because
the chosen wave potentials are coupled in the boundary
conditions at the interfaces [55,9296].
Another approach is based on the eigenfunction expan-
sion of the DGFs in terms of vector wavefunctions [39,
Chap. 11]. This method, although elegant and historically
important, leads to representations that require much
processing to render them tractable [89,97,98].
Still another formalism begins with a 3D Fourier trans-
form representation of the DGF for a homogeneous medi-
um [99, p. 38; 100105]. The integration over the spectral
variable associated with the z direction is then performed
in closed form by the calculus of residues [12, p. 823]. This
results in a 2D Fourier representation, which makes ev-
ident the TM and TE dyadic components and is suitable
for application of the scattering superposition method, in
which the spectral representation of the primary, whole-
space DGF is augmented by homogeneous solutions, to
account for the presence of the stratied medium. The
augmented terms constitute upward- and downward-
propagating waves, with coefcients that obey certain
recurrence relations, which are determined by imposing
the continuity of the transverse DGF components at the
media interfaces. Finally, in the 3D case, the resulting
two-dimensional spectral integrals are converted to So-
mmerfeld integrals.
In conclusion, the alternative techniques do not appear
to offer any substantial advantages over the method es-
poused here.
7. TRANSMISSION-LINE ANALYSIS FOR PIECEWISE
HOMOGENEOUS MEDIA
Although the theory presented thus far is applicable to
media with parameters that have arbitrarily z depen-
dence, only for a very few special media proles can the
transmission-line solution be found in an analytical closed
form [106]. Therefore, we next focus attention on a mul-
tilayered medium with piecewise constant parameters, as
illustrated in Fig. 1a, which is the most important and
most frequently encountered case in practice. The trans-
mission-line equivalent circuit now consists of a tandem
connection of uniform transmission-line sections, where
section n, with terminals at z
n
and z
n1
, has propagation
constant k
a
zn
and characteristic impedance Z
a
n
, as illus-
trated in Fig. 1b.
In this section, we rst analyze the situation where
there are no sources on the TL network, except possibly at
innity. This case is important in its own right, and it
prepares the ground for the derivation of the TLGFs,
which is presented next. We close with a brief discussion
of the asymptotic behavior of the TLGFs.
7.1. Source-Free Case
The voltage and current on the nth TL section, which is
assumed to be source-free, satisfy the homogeneous forms
of the transmission-line equations (24), which lead to
d
2
dz
2
k
2
zn
_ _V
n
z
I
n
z
0; I
n
z
1
jk
zn
Z
n
d
dz
V
n
z 106
Hence, choosing e
jkznz
as the fundamental solutions, we
may write
26
V
n
z
Z
n
I
n
z
_ _
V
n
e
jkznzzn
..
V
n
z
V
n
e
jkznzzn
..
V
n
z
107
where V
n
z and V
n
z represent the forward- and back-
ward-propagating voltage waves, respectively, and where
V
n
and V
n
are as yet undetermined complex coefcients.
The plus/minus signs between the forward and backward
waves correspond, respectively, to the upper/lower terms
on the left side of the equation. Note that, for later con-
venience, we use (z z
n
), rather than just z, in the expo-
nents. Thus, the phase reference point for the traveling
waves has been arbitrarily selected at the left terminals of
the TL section. An exception must be made if section 1 in
Fig. 1b is of innite extent; in this case, we move the phase
reference to the right terminalsspecifically we use z
2
in
place of z
1
in (107).
We can also express (107) as
V
n
z
Z
n
I
n
z
_ _
V
n
z1 G
!
n
z 108
where
G
!
n
z
V
n
z
V
n
z
109
is the right-looking voltage reection coefcient at z on the
nth TL section. If we also dene the right-looking imped-
ance at z as
Z
!
n
z
V
n
z
I
n
z
Z
n
1 G
!
n
z
1 G
!
n
z
110
where the second equality follows from (108), we obtain
G
!
n
z
Z
!
n
z Z
n
Z
!
n
z Z
n
111
Hence, the reection coefcient at any location may be
computed from the impedance, and vice versa. From (109)
26
The present choice of fundamental solutions leads to a travel-
ing-wave representation of the voltages and currents. Another
possible choice, sin(k
zn
z) and cos(k
zn
z), leads to a standing-wave
representation [4, p. 203].
and (107), it follows that
G
!
n
z
n
G
!
n
e
j2y
n
112
where
y
n
k
zn
d
n
; d
n
z
n1
z
n
113
and where G
!
n
G
!
n
z
n
1
. It should be noted that the
terminal reection coefcients shown above are evaluat-
ed on the inner sides of the TL section. This distinction is
important, because unlike the impedances, the reection
coefcients are discontinuous across the TL terminals.
27
Using (110)(112) and applying fact that Z
!
n
z
n1

Z
!
n
Z
!
n1
z
n1
, it is readily found that the right-
looking reection coefcients satisfy the recurrence
relation
G
!
n
G
n1;n
G
!
n1
e
j2y
n1
1G
n1;n
G
!
n1
e
j2y
n1
114
where
G
i; j
Z
i
Z
j
Z
i
Z
j
115
Note that G
i,j
is analogous to the Fresnel reection coef-
cient across an interface between two half-spaces lled
with media i and j, looking from medium j [107, p. 62] The
recursion (114) is applied beginning with the rightmost TL
section and it proceeds backward, toward the left end. If
the Nth layer is of innite extent, as in Fig. 1, then it is
reectionless, and the starting value is G
!
N
0.
Using the translation formula [112] in (108), we can
express the latter as
V
n
z
Z
n
I
n
z
_ _
V
n
e
jk
zn
zz
n
1 G
!
n
e
j2k
zn
z
n1
z
116
For simplicity, the terminal voltages will be denoted as
V
n
V
n
(z
n
) and V
n1
V
n
(z
n1
). Using the upper equation
(116), we can eliminate the coefcient V
n
in favor of V
n
,
which results in
V
n
z
Z
n
I
n
z
_ _
V
n
e
jk
zn
zz
n
1 G
!
n
e
j2yn
1 G
!
n
e
j2k
zn
z
n1
z
117
Hence, the voltage and current at any location within a
source-free TL section can be computed from the voltage
at the left terminals. Evaluating the upper equation (117)
at z z
n1
, we nd the right-looking voltage transmission
coefcient across the nth TL section as
t
!
n

V
n1
V
n
1 G
!
n
e
jy
n
1 G
!
n
e
j2y
n
118
This formalism is useful if the excitation is to the left of
the nth TL section. If the sources are to the right of the nth
section, (107) is still applicable, except that V
n
now plays
the role of the incident wave. Hence, we dene the left-
looking reection coefcient
G
n
z
V
n
z
V
n
z

Z
n
z Z
n
Z
n
z Z
n
119
and the left-looking impedance
28
Z
n
z
V
n
z
I
n
z
Z
n
1 G
n
z
1 G
n
z
120
and from (119) and (107), we obtain
G
n
z
n1
G
n
e
j2y
n
121
where G
n
G
n
z
n
. Applying (119)(121) and the fact that
Z
n
z
n
Z
n
Z
n1
z
n
, we readily derive the recurrence
relation
G
G
n1;n
G
n1
e
j2y
n1
1G
n1;n
G
n1
e
j2y
n1
122
From this recursion, the left-looking reection coefcients
can be computed by the same procedure as that used for
the right-looking coefcients, except now we begin in sec-
tion 1 and progress forward, toward the right end of the
TL network. The starting value, G
1
, is easily found. For
example, if the rst layer is backed by a plate with surface
impedance Z
s
, as illustrated in Fig. 1, then
G
Z
s
Z
1
Z
s
Z
1
123
which becomes 1 if the plate is PEC, and 1 if it is
PMC. For a good, but not perfect, electrical conductor of
conductivity s, Z
s
can be found as
Z
s
1j
sd
; d
2
om
0
s
124
where d is the skin depth (not to be confused with the
Dirac delta) [12, p. 201; [108]].
27
The continuity of the impedances is a consequence of the con-
tinuity of the voltages and currents when the sources are absent.
28
The minus sign arises because we maintain the positive-z
direction as the reference direction for the current.
Using the translation formula [121] in (107), the latter
can be expressed as
V
n
z
Z
n
I
n
z
_ _
V
n
e
jk
zn
zz
n
1 G
n
e
j2k
zn
zz
n
125
where the coefcient V
n
can be eliminated in favor of
V
n1
, resulting in
V
n
z
Z
n
I
n
z
_ _
V
n1
e
jk
zn
z
n1
z
1 G
n
e
j2yn
1 G
n
e
j2k
zn
zz
n
126
Hence, the voltage and current at any point z within a
source-free TL section can also be computed from the volt-
age at the right terminals. Finally, evaluating (126) at z
z
n
, we nd the left-looking voltage transmission coefcient
for section n as
t
n

V
n
V
n1
1 G
n
e
jy
n
1 G
n
e
j2yn
127
To illustrate the application of this formalism, consider a
TL network corresponding to an unshielded multilayer
excited in the upper half-space (layer N). Let the incident
eld be represented by a voltage wave with the amplitude
V
N
. The resulting total voltage and current in the top lay-
er are given by (125) with nN, where G
N
represents the
overall reection coefcient looking into the stack from the
upper half-space, which can be computed from (122). Note
that the voltage at the left terminals of section N is
V
N
1 G
N
V
N
128
In layer 1 there is no reection; hence
V
1
z V
1
e
jk
z1
zz
2
129
and V
2
V
1
. The overall transmission coefcient of the
stack can now be found as
T
1
V
N
1 G
V
2
V
N
;
V
2
V
N
V
2
V
3
V
3
V
4

V
N1
V
N
N1
n2
t
n
130
where it is understood that the product is equal to one if
the lower limit exceeds the upper.
As another interesting application, consider an arbi-
trary distribution of electric and magnetic currents radi-
ating near a planar interface between a uniaxial medium
(layer 1) and an isotropic medium (layer 2). The issue is
whether it is possible to adjust the uniaxial medium pa-
rameters so that there is no reection at the interface for
any frequency and any conguration of the sources. Since
the elds of arbitrary sources in layered media are
expressible as Sommerfeld integrals whose integrands
comprise both TM and TE TLGFs, these requirements
will be satised if the TM and TE Fresnel reection coef-
cients given in (115) vanish for all values of k
r
, that is, if
G
e
1;2
0 and G
h
1;2
0, and thus Z
e
1
Z
e
2
and Z
h
1
Z
h
2
. In view
of (25), (26), these conditions imply
1
e
1
k
2
0
e
1
m
1
e
1
e
z1
k
2
r
_

1
e
2
k
2
0
e
2
m
2
k
2
r
_
131
and
1
m
1
k
2
0
e
1
m
1
m
1
m
z1
k
2
r
_

1
m
2
k
2
0
e
2
m
2
k
2
r
_
132
which will be simultaneously satised for all k
r
if
e
z1
e
2
2
e
1
; m
1
e
1
m
2
e
2
; m
z1
e
2
m
2
e
1
133
where e
1
may be selected arbitrarily. Hence, we may pos-
tulate a lossy medium with
e
1
e
r1
j
s
1
oe
0
134
where the dielectric constant e
r1
and conductivity s
1
are
arbitrary. It is interesting to note that, since e
z1
and m
z1
are
inversely proportional to e
1
, the resulting effective electric
and magnetic conductivities along the optic axis are neg-
ative. Even though such media may be not realizable in
practice, they have played an important role in the design
of the so-called perfectly matched layers (PMLs) for re-
ectionless truncation of nite-difference and nite-ele-
ment meshes [109111]. Since such PMLs must be
surrounded by impenetrable enclosures to be practical,
let us consider the effect of terminating the perfectly ab-
sorbing medium characterized by (133), (134) with a PEC
ground plane at a distance d
1
below the interface. The re-
ection coefcient looking into the uniaxial layer can be
found using (122) with n2. Since in the present case
G
a
1;2
0 and G
1
1, we obtain
G
2
a
e
j2k
a
z1
d
1
; k
e
z1
k
h
z1
e
1
e
2
k
2
2
k
2
r
_
135
Using (134) here and making the substitutions k
r
k
2
sinW,
the reectivity of this PML is found as
29
R
a
y e
4Z
2
s
1
d
1
cos W
136
where it is assumed that k
2
, Z
2
(the intrinsic impedance of
medium 2) and W are real-valued. Here, W can be inter-
preted as the angle of incidence, measured from the in-
terface normal, of the continuous spectrum of plane waves
representing the eld radiated by the nite sources. We
note that the reectivity, which is the same for the TM and
TE waves, can be made arbitrarily small by properly
29
Reectivity is dened as the magnitude-squared reection coef-
cient.
choosing the product s
1
d
1
. However, this PML works best
for normal incidence and loses its effectiveness for waves
incident at grazing angles (W 901) and for evanescent
waves (in which case W is complex-valued [112, p. 231]).
For completeness, we also mention two alternative
methods that have been extensively used in multilayer
analysis [113116]. The rst is based on the relationship
[3, p. 44; 12, p. 192]
V
n
I
n
_ _
cos y
n
jZ
n
siny
n
j
Z
n
siny
n
cos y
n
_
_
_
_
V
n1
I
n1
_ _
137
which can be derived from (107) by eliminating the trav-
eling-wave coefcients V
n
in favor of the terminal voltages
and currents. The matrix in (137) is unimodular, that is, it
has a unity determinant.
30
This property can be exploited
in the analysis of periodic multilayers [117, p. 60], which
have many important applications, such as gain-enhanc-
ing superstrates for microstrip antennas [118], or multi-
layer reective coatings for EUV lithography [119]. The
second method is based on the formula [12, p. 189; 120]
V
n
V
n
_
_
_
_
1
T
n1;n
1 G
n1;n
G
n1;n
1
_
_
_
_
e
jy
n1
0
0 e
jy
n1
_
_
_
_
n1
V
n1
_
_
_
_
Q
n1
V
n1
V
n1
_
_
_
_
138
where T
I, j
1G
i, j
is the Fresnel transmission coefcient
from medium j to medium i. This equation can also be de-
rived from (107), this time by eliminating the terminal
voltages and currents in favor of the traveling-wave coef-
cients in adjacent layers. If layer 1 is of innite extent,
(138) is used with y
1
0.
Both matrix methods outlined above are useful in com-
putation of the guided-wave poles of a multilayer, which
correspond to the source-free solutions of the Maxwell
equations, or waveguide modes. We will illustrate this for
the second method, assuming an N-layer stack, as illus-
trated in Fig. 1. In this case, a repeated application of
(138) leads to the relationship
V
N
V
N
_
_
_
_
Q
N1
Q
N2
Q
1
V
1
V
1
_
_
_
_
q
11
q
12
q
21
q
22
_
_
_
_
V
1
V
1
_
_
_
_
139
which links the amplitudes of the forward- and backward-
propagating voltage waves in the top and bottom layers. If
layer 1 is shielded from below, the positive- and negative-
propagating wave amplitudes can be expressed as
V
1

1
2
V
1
Z
1
I
1
, with the total voltage and current at
z
1
constrained by the impedance boundary condition V
1
Z
s
I
1
. In layer N, assuming a stack open above, there
should be no backward-propagating wave, thus V
N
0.
Using these constraints in (139), we obtain
1 q
11
G
1
q
12
0 q
21
G
1
q
22
_
_
_
_
V
N
I
1
_ _
0 140
with G
1
given by (123). Nontrivial solutions to this homo-
geneous system only exist for those values of k
r
, which
make the determinant vanish. This leads to the modal
dispersion relation
q
21
G
1
q
22
0 141
which also includes the cases of PEC backing G
1
1,
PMC backing G
1
1, and no backing G
1
0. The zeros
of (141) in the complex k
r
plane determine the propagation
constants of the guided-wave modes of the multilayer and
are also the poles of the spectral integrands appearing in
(69)(98) [121,122]. A similar dispersion relation can be
derived for a multilayer shielded from above.
The advantage of the matrix formulations is that they
lead to analytic modal dispersion functions, which are
amenable to the root nding methods based on Cauchys
theorem [123,124]. However, they are not well suited for
use in the DGFs, because of the occurrence of growing ex-
ponentials when k
r
-N, which may cause numerical over-
ows in a computer implementation [6,125].
Finally, we mention that it is also possible to develop
explicit recursive relationships between the coefcients of
the forward- and backward-propagating waves in different
layers, without using the matrix technique [13, p. 49; 126].
7.2. Source-Excited Case
Consider next a source-excited TL section n, as illustrated
in Fig. 9, where G
n
and G
!
n
are the voltage reection coef-
cients and Z
n
and Z
!
n
are the total impedances, looking
out of the left and right terminals, respectively. It will
sufce to treat in detail only the current excitation case;
the solution for the voltage source will follow by duality.
First, we note that the SturmLiouville theory of Sec-
tion 4 can be directly applied to this problem. The homo-
geneous solutions satisfying the boundary conditions at
the left and right terminals of TL section n have already
been constructed in (125) and (116). Hence, we can choose
V
n
z e
jkznzzn
1 G
n
e
j2kznzzn
142
V
!
n
z e
jk
zn
zz
n
1 G
!
n
e
j2k
zn
z
n1
z
143
The corresponding conjunct is readily found as
PWV
n
; G
!
n

2
Z
n
D
n
144
30
This matrix is sometimes referred to as the ABCD matrix.
D
n
1 G
n
G
!
n
e
j2yn
145
Using these equations in (43), we nally obtain
V
i
zjz
0

Z
n
2D
n
V
n
z
o
V
!
n
z
>
146
and I
i
(z|z
0
) then follows from (40). However, this product
form of solution can be cumbersome when used in the
DGFs, where it may be necessary to combine terms arising
from different TLGFs. Therefore, we derive below a more
convenient TLGF form, which also offers more physical
insight.
Assume at rst that the TL section in Fig. 9 is either of
innite extent, or is matched at both ends, so that there
are no reections. Let this TL be driven by a unit-strength
current generator at z
0
. Note that the voltage V
i
must be
continuous at the source location z
0
and must have the
form of outgoing waves. Furthermore, the current source
is in effect loaded by two impedances Z
n
connected in par-
allel. Hence, the solution must have the form [4, p. 206]
V
i
zjz
0

Z
n
2
e
jk
zn
jzz
0
j
; I
i
zjz
0

1
2
e
jk
zn
jzz
0
j
147
where the upper and lower signs correspond to z4z
0
and
zoz
0
, respectively. Next, to account for the presence of the
terminations, as in Fig. 9, we augment the particular so-
lution given above by a homogeneous solution, as follows
V
i
zjz
0

Z
n
2
e
jk
zn
jzz
0
j
A
n
e
jk
zn
z
B
n
e
jk
zn
z
148
I
i
zjz
0

1
2
e
jk
zn
jzz
0
j
A
n
e
jk
zn
z
B
n
e
jk
zn
z
149
where the coefcients A
n
and B
n
can be found by enforcing
the boundary conditions (42). This last step is straightfor-
ward, and we omit the details.
31
Substituting the so-
obtained coefcients A
n
and B
n
back into (148), (149) com-
pletes the solution procedure. The nal result can be
expressed compactly as
V
i
zjz
0

Z
n
2
e
jk
zn
jzz
0
j
1
D
n
4
s 1
R
s
n
e
jk
zn
z
s
n
_ _
150
I
i
zjz
0

1
2
e
jk
zn
jzz
0
j
1
D
n
4
s 1
1
s
R
s
n
e
jk
zn
z
s
n
_ _
151
R
1
n
G
n
; z
1
n
z z
0
2z
n
; 152
R
2
n
G
!
n
; z
2
n
2z
n1
z z
0
; 153
R
3
n
G
n
G
!
n
; z
3
n
2d
n
z z
0
; 154
R
4
n
G
n
G
!
n
; z
4
n
2d
n
z z
0
; 155
Of course, Eqs. (146) and (150) are equivalent, and the
former can be directly manipulated into the form of the
latter.
The TLGF representation (150), (151) can be given an
interesting ray (geometric optical) interpretation [127], as
illustrated in Fig. 10. Hence, the rst terms in the brack-
ets represent the direct ray between the source and the
eld point (ray 0), whereas the second terms represent
four categories of rays that undergo partial reections at
the upper and lower slab boundaries before reaching the
observation point (rays 1 through 4). This ray interpreta-
tion is more evident if the inverse of the resonant de-
nominator D
n
is expanded as [127]
1
D
n
K
k 0
O
k
O
K 1
1 O
; O G
n
G
!
n
e
j2yn
156
With nite K, this formula leads to a hybrid ray modal
representation, whereas for K-N, it gives a pure ray
representation. However, the simple ray picture shown in
Fig. 10 obtains for only a single layer, and it becomes
unwieldy in a multilayer geometry [127].
The remaining TLGFs, I
v
and V
v
, are governed by equa-
tions (34), which are dual to equations (33) satised by V
i
and I
i
. Hence, the expressions for I
v
and V
v
can be ob-
tained from those for V
i
and I
i
, respectively, by simply re-
placing the characteristic impedances by admittances, and
k
z,n1
k
z,n+1
k
zn
Z
n1
Z
n+1
z
n+1
Z
n
z'
d
n
z
n
1V
1A
(z)
(
n
, Z
n
)

(
n
, Z
n
)

+
Figure 9. Voltage and current sources in a transmission-line
section.
Source
point
Field
point
1 4
3 2
0
z
n+1
z
z'
z
n
Figure 10. Ray interpretation of the TLGFs.
31
Note that we are in effect using the scattering superposition
method at the transmission-line level.
vice versa, which has also the effect of changing the signs
of the reection coefcients. We should mention here the
important fact that all TLGFs can be derived from either
V
i
or I
v
[4, p. 194]. For, suppose we have found V
i
, as given
by (150) above. Then, I
i
follows from the rst equation in
(33), which in turn implies knowledge of V
v
, in view of the
last reciprocity relation in (38). Finally, V
v
determines I
v
via the rst equation in (34). Thus, all the required infor-
mation is contained in V
i
; an alternative statement applies
to I
v
. Because, as is evident from (18), (19), the E
z
and H
z
eld components are specied by, respectively, the current
on the TM transmission line and the voltage on the TE
transmission line, it is convenient to use I
e
v
and V
h
i
as the
basic TLGFs.
32
As we show later in this section, I
e
v
and V
h
i
are also the least singular of all the TLGFs on conversion
to the space domain.
Although the formulas given above assume observation
points within the source TL section, they can easily be
extended to arbitrary observation points outside section n,
by using the results derived earlier for the source-free
case. Hence, if z is within section mon, we use the source
section formulas to compute the voltage V
n
at the left ter-
minals of section n, and then invoke (126) with nm and
V
m1
V
n
n1
k m1
t
k
157
Similarly, if z is within section m4n, we compute the volt-
age V
n1
at the right terminals of the source section, and
then use (117) with nm and
V
m
V
n1
m1
kn1
t
!
n
158
We note that it would sufce to implement only one of
these cases and to handle the other case by using the re-
ciprocity relations (38). However, if both cases are imple-
mented, the reciprocity relations can provide a useful
check of the formulation and the computer program.
The formulation adopted here easily specializes to the
case where the source or observation layer is a half-space,
and thus either the left- or right-looking reection coef-
cient becomes zero; one simply omits the terms involving
the vanishing reection coefcients. Also, an important
feature of this formulation is that all exponentials that
occur are nongrowing as k
r
-N, which greatly reduces
the danger of overows in the computation of Sommerfeld
integrals (see Section 8).
Finally, an important property of the TLGFs should be
mentioned, which also has important consequences for the
evaluation of the Sommerfeld integrals. Namely, the
TLGFs are even functions of k
zn
, if n corresponds to a -
nite-thickness layer. This can be directly conrmed by
noting that (150), (151) are unaffected by the replacement
k
zn
-k
zn
, which also causes the replacements Z
n
-Z
n
,
G
n
!1= G
n
, and G
!
n
!1= G
n
. Hence, the choice of the
square-root branch in k
zn
is arbitrary, which means that
there are no branchpoints in the k
r
plane associated with
any of the nite layers [13, p. 112]. However, this obser-
vation applies only to the full TLGFs, including the direct-
ray terms.
7.3. Asymptotic Behavior
It is important to understand the asymptotic, large-k
r
be-
havior of the TLGFs, because it affects the convergence of
the Sommerfeld integrals and determines the source-re-
gion singularity of the DGFs in the space domain.
33
This
knowledge is needed, for example, in extrapolation meth-
ods [128], or to extract the quasistatic parts from the
spectral kernels, both as a measure to accelerate the So-
mmerfeld integrals and to isolate the singular parts of the
kernels [80, p. 36; 129]. As is evident from (150), (151), the
TLGFs decay exponentially as k
r
-N, unless z z
0
, which
represents the worst-case situation. We therefore focus
attention on the case where z lies within the source layer.
The large-k
r
, quasistatic forms of V
a
i
and I
a
i
are readily
found from (150) and (151) by expanding the inverse res-
onant denominator in the geometric series (156) and re-
placing the reection coefcients by their asymptotic
values. Each term of the series then represents a quasi-
static image of the source in the lower or upper interface.
Hence, keeping only the direct components and two dom-
inant image terms, we nd that, as k
r
-N
V
a
i
$
Z
a
n
2
e
jk
a
zn
jzz
0
j
w
a
n1;n
e
jk
a
zn
z
1
n
w
a
n1;n
e
jk
a
zn
z
2
n
_ _
159
I
a
i
$
1
2
e
jk
a
zn
jzz
0
j
w
a
n1;n
e
jk
a
zn
z
1
n
w
a
n1;n
e
jk
a
zn
z
2
n
_ _
160
where we have used the asymptotic forms
G
e
i; j
$ w
e
i; j

k
e
i
k
e
j
k
e
i
k
e
j
; k
e
n
e
n
e
zn
p
161
G
h
i; j
$ w
h
i; j
k
h
i
k
h
j
k
h
i
k
h
j
; k
h
n

m
n
m
zn
p
162
which follow from (115) on noting that the characteristic
impedances (26) behave as
Z
e
n
$
k
r
joe
0
k
e
n
; Z
h
n
$
jom
0
k
h
n
k
r
163
The asymptotic behavior of I
a
v
and V
a
v
can be deduced
from (159) and (160) by invoking duality, as explained
earlier. We summarize these ndings in Table 2, where
we also indicate the small-R behavior of the corresponding
32
Some authors [45] prefer to express I
e
v
in terms of the current
reection coefcients, which are just the negatives of the voltage
reection coefcients used here.
33
This follows from the nal value theorem for Fourier transforms
[12, p. 826]. It can be seen from [25] that the case k
r
-N corre-
sponds to o-0, that is, the static case. If the k
r
-Nlimit is ap-
plied only to the reection coefcients, the resulting expressions
are referred to as the quasistatic forms.
Sommered integrals.
34
By referring to this table, one can
easily ascertain the asymptotic behavior of the spectral
integrands, as well as the source region singularities of
the DGF components (69)(98). It should be noted that not
only can singularities can arise as a result of the direct ray
[the rst terms in brackets in (159), (160)]; one of the
dominant images (the second and third terms) can also
give rise to a singularity when z and z
0
are both on the
same interface between adjacent layers.
8. SPECTRAL INTEGRALS AND THEIR COMPUTATION
Consider the formula [131, p. 455]
S
0
e
jk
a
z
jzj
2jk
a
z
_ _
n
a
e
jkr
a
4pr
a
; r
a
n
a
r
2
z
2
_
164
which is the Sommerfeld identity [40] extended to uniaxial
media. From (164), another important identity can be de-
rived [132, p. 164; 133, p. 428; 134, p. 264]:
S
1
e
jk
a
z
jzj
2jk
a
z
k
r
_ _
e
jkjzj
e
jkr
a
4pjkr
165
In the case of innite, homogeneous media, these identi-
ties and their derivatives can be used to obtain closed-form
expressions for the Sommerfeld integrals that arise in
(69)(98) [133, p. 376; 135; 136, p. 136]. We include the
resulting whole-space DGFs in the Appendix. In layered
media, however, such closed forms do not exist and one
must resort to numerical quadrature.
The spectral integrands appearing in (69)(98) are in
general multivalued, as a result of the branchpoints in-
troduced by the square-root definition of k
a
zn
[137, p. 238].
However, as pointed out earlier, these integrands are even
functions of the k
a
zn
corresponding to nite layers,
35
and
thus the only branchpoints that appear are those associ-
ated with the half-space media. If the multilayer is sand-
wiched between two isotropic half-spaces (which represent
the substrate and cladding in optical applications), the
branchpoints occur at 7k
1
and 7k
N
, and the complex k
r
plane consists of four overlying Riemann sheets (or two
sheets, if the multilayer is shielded on one side, or if the
media of the substrate and cladding are the same, i.e.,
k
1
k
N
), joined at the branchcuts emanating from the
branchpoints [13, p. 66; 85, p. 56; 137, p. 241]. Although
the exact shape of these cuts is arbitrary, it is often con-
venient to specify them by the condition Imk
a
zn
0,
which ensures that Imk
a
zn
is of the same sign every-
where on a given sheet.
36
It is then customary to dene
the proper (also referred to as the top or upper) sheet
as the one for which poargfk
a
zn
g 0. The spectral inte-
grands also in general exhibit pole singularities in the
complex k
r
plane, contributed by the zeros of the resonant
denominator and/or the reection coefcients appearing in
the TLGFs. For the layered medium geometry of Fig. 1,
the poles can be found as the roots of the dispersion rela-
tion (141), as explained in the previous section. If the con-
dition poargfk
a
zN
g 0 is imposed in the cladding, the
roots of (141) are the proper poles, which are located on the
top Riemann sheet and correspond to the bound modes of
the structure. Otherwise, they are the leaky-wave poles
located on the lower sheet [113].
The integration paths in the integrals (65) must be
properly indented around the poles and branchpoints and
must approach innity on the proper Riemann sheet, to
ensure that the elds are bounded as z-7N [105,138].
An example path satisfying these requirements, referred
to as the Sommerfeld integration path (SIP), is illustrated
in Fig. 11, where only the right half of the path applies to
the rst integral form in (65). Note that the poles and
branchpoints occur in pairs in the second and fourth quad-
rants of the k
r
plane [4, p. 465].
37
In the limit of vanishing
losses, all singular points move to the real axis. The Hank-
el function in the second integral in (65) introduces a log-
arithmic branchpoint [139] at the origin (not shown), and
the path of integration must be indented around it into the
Branch
cuts
Integration
path
Rekj
lmkj
k
N
k
1
Poles
Figure 11. Integration path in the complex k
r
plane.
Table 2. Worst-Case Asymptotic Behavior of
Transmission-Line Green Functions and Corresponding
Spatial-Domain Singularities
TL Green Function
Asymptotic
Behavior
Space-Domain
Singularity
V
e
i
; I
h
v
Ok
1
r
OR
3
V
e
v
; I
h
i
; V
h
v
; I
e
I
Ok
0
r
OR
2
V
h
i
; I
e
v
Ok
1
r
OR
1
34
The big oh symbol means order of magnitude [130, p. 9].
35
Provided that the whole-space terms are not extracted.
36
The so-dened branchcuts are referred to as the Sommerfeld, or
fundamental, branchcuts.
37
Figure 11 corresponds to the situation where the media lling
the lower and upper half-spaces are lossy and lossless, respec-
tively. In the e
iot
convention, the singular points are complex
conjugates of those shown here. Also, the SIP and the branchcuts
are reected in the real axis and H
1
n
replaces H
2
n
in [65]
[4, p. 466].
lower half of the k
r
plane, which is emphasized in the
notation by writing the lower integration limit as Ne
jp
.
If the SIP is used to compute the Sommerfeld integrals,
the pole and branchpoint singularities that occur nearby
must be avoided or, preferably, extracted and handled an-
alytically [41,83, p. 161; 140143]. Since the Bessel and
Hankel functions are oscillatory on the real axis and the
integrands are divergent, the integration is inefcient,
unless extrapolation techniques (sequence accelerators)
are employed. One proven and effective procedure is the
integration then summation method [128,144146]. In
this approach, the GaussKronrod or Patterson quad-
rature [147] is applied between consecutive zeros of the
Bessel function and the resulting sequence of partial sums
is accelerated by one of the extrapolation methods
[148,149].
The integration may be more efcient if the SIP is de-
formed to another path, on which the integrand is rapidly
convergent [41,88,150,151]. The contour deformation,
which is based on Cauchys integral theorem [1, p. 294],
is performed in a way, such that the conditions of Jordans
lemma [12, p. 823; 139] are satised and no contribution
results from any part of the path of integration at innity.
The residues of the poles captured by the deformation,
as well as contributions from any branchcuts swept in
this process, must be properly accounted for [152155]. If
the integration contour is restricted to the proper Riem-
ann sheet, the pole contributions represent the guided-
wave modes (or discrete spectrum) of the multilayer, and
the branchcut integrals represent the radiation modes
(continuous spectrum) [13, p. 333]. Near the surface of
a layered medium, the latter assume the characteristics of
lateral waves that propagate along the top and bottom
interfaces [3, p. 484]. If the multilayer is shielded from
below and above, the branchcuts are absent, but an in-
nite number of poles appear on the top (and, in this case,
only) sheet. Hence, the spectrum is purely discrete, and
the Sommerfeld integrals can be converted into series of
pole residues [156].
One particularly attractive integration contour is the
steepest-descent path (SDP), on which the integrand is
most rapidly convergent [3, p. 462; 13, p. 107; 152,157].
However, this contour veers into a lower Riemann sheet,
and thus there is a possibility of capturing some improper
poles during the path deformation. The contribution from
these leaky-wave poles, as well as any captured proper
poles and branchcuts, must not be overlooked. In fact, the
difculty of tracking the location of these singularities vis-
a` -vis the SDP is a major obstacle in the practical imple-
mentation of the SDP integration. The SDP, which passes
through the saddle point, also directly leads to the asymp-
totic techniques for the approximate evaluation of the in-
tegrals [4, Chap. 4; 13, p. 106]. These approximations are
often performed in the complex angular spectrum plane,
which is related to the k
r
plane by k
r
k
1
sinw or k
r
k
N
sinw [3, Sect. 157; 4, p. 541; 130, Sect. 17; 137, p. 241].
Either of these transformations unfolds one pair of
branchcuts in the k
r
plane and maps the associated two
Riemann sheets onto a single sheet in the w plane. An-
other interesting transformation has been proposed,
which unfolds all four sheets of the Riemann surface [158].
Other techniques that should be mentioned are the fast
Hankel transform (FHT) method
38
[159,160] and the dis-
crete complex image method (DCIM) [161165]. In the
FHT method the Sommerfeld integral is converted to a
discrete convolution by a logarithmic transformation of
the variables and the approximation of the transformed
integrand [i.e., the
~
ff k
r
function in (65)] as a sum of sinc
functions. This convolution has a form of a linear digital
lter, which processes the input function [i.e.,
~
ff k
r
] to
produce the output function, namely, the Sommerfeld
integral. The coefcients of digital lters of various
lengths, corresponding to Hankel transforms of orders 0
and 1, have been computed and are available in the liter-
ature [166,167]. The FHT method is effective only when
there are no singularities near the real axis, as in the case
of highly lossy media, or these singularities can be ex-
tracted and handled analytically. Also, the integrand en-
velope must be strictly decreasing as k
r
-N.
The basic idea of the DCIM is to approximate the spec-
tral kernel on a suitable path in the complex k
r
plane by a
sum of complex exponentials, using an established system
identication procedure [168], and then to use an identity,
such as (164), to evaluate the Sommerfeld integrals in
closed form.
39
The lateral range of applicability of DCIM is
limited, unless the guided-wave poles are extracted from
the integrands prior to the exponential t. More recently,
a related technique has been proposed, based on rational,
rather that exponential, function tting, which does not
suffer from this limitation [169].
A remarkable property of both the FHT method and the
DCIM is that no numerical integration is required and no
evaluation of Bessel (or Hankel) functions is involved. As a
result, these techniques are much faster than the most
efcient numerical quadrature methods. However, they
also share a significant drawback that they have no built-
in convergence measures, and thus the accuracy of the
result is usually difcult to ascertain.
Before any of the techniques discussed here is applied,
a recommended and sometimes necessary practice is to
improve the convergence of the Sommerfeld integrals by
the subtraction of the large-k
r
, quasi-static terms, which
have closed-form space-domain counterparts [34,170]. The
added benet of this procedure is that it helps identify and
isolate the singular parts of the space-domain Green func-
tions. In the important special case where the source and
observation points are on the same interface in a multi-
layer medium, the Sommerfeld integrals can be acceler-
ated by the extraction of the half-space portion of the
integrand, which makes the remainder exponentially con-
vergent [171,172]. The extracted half-space part can be
efciently evaluated by SDP integration [152].
It should also be pointed out that the techniques de-
scribed here are in practice usually applied to Sommerfeld
integrals associated with electromagnetic potentials, rath-
er than elds [41,129,173]. Since elds are related to po-
tentials through differential operators, the potential forms
38
Also known as the linear digital lter method.
39
The layered media Green functions computed this way are often
referred to, somewhat inappropriately, as closed-form Greens
functions.
are less singular (in spatial domain) and more rapidly
convergent (in spectral domain), and thus more tractable,
than the eld forms. If repeated evaluation of Sommerfeld
integrals is required, as is the case when integral equation
techniques are used for objects in layered media, a viable
alternative to the on demand computation is interpola-
tion from a grid of precomputed and tabulated values
[88,138,150,174]. The interpolation is in general over the
three-dimensional (R, z, z
0
) space. However, if the source
and observation points are within the same layer, only
two-dimensional interpolation is required, because, as is
evident from (150), (151), the DGFs can always be split
into parts that depend on either R; z z
0
or R; z z
0
.
In the 2D case, the spectral integrals have the form of
(32), where the integration can be limited to the positive
real k
x
axis, with the exponential function replaced by
2j sin(k
x
x) or 2cos(k
x
x), depending on whether the inte-
grand
~
ff
2D
is even or odd in k
x
, respectively. Most of the
techniques available for the Sommerfeld integrals can be
adapted for this case. The singularities in the k
x
plane are
related to their k
r
-plane counterparts by the mapping
k
x

k
2
r
b
2
_
, where the propagation constant b may
be complex-valued. Hence, the location of the poles and
branchpoints changes with b, which may require a corre-
sponding deformation of the integration path [18,175].
Finally, we note that the 2D counterpart of the So-
mmerfeld identity (164) is
1
2p
_
1
1
e
jk
a
z
jzj
2jk
a
z
e
jk
x
x
dk
x
n
a
p
4j
H
2
0
g
a
n
a
x
2
z
2
_
166
where g
a
k
2
b
2
=n
a
_
, with poarg{g
a
}r0 [176; 177,
p. 48].
9. FAR-ZONE RADIATION FIELDS
In some important applications, such as antenna engi-
neering, the far-zone (radiation) elds due to arbitrary
source distributions in layered media are of interest. In
this section we demonstrate that these elds can be easily
computed using the formalism developed in the previous
sections.
Hence, let (J, M) be known currents occupying a vol-
ume Vanywhere in a layered medium, which is not shield-
ed from above, as illustrated in Fig. 1a. We assume, for
simplicity, that the medium of the upper half-space is iso-
tropic. Let z
0
be a point on the z axis, such that the layered
medium and the volume Vare conned to the region zrz
0
,
as illustrated in Fig. 12. Our goal is to derive leading-order
asymptotic expressions for the elds radiated by (J, M),
valid when the eld point is far from V in the half-space
region z4z
0
.
It should be clear from (47)(50) and the TLGFs derived
in Section 7 that, in the region z4z
0
, the spectral DGFs
may be expressed as
~
GG
PQ
k
r
; zjz
0

~
GG
PQ
k
r
; z
0
jz
0
e
jk
zN
zz
0
;
k
zN
k
2
N
k
2
r
_
167
In view of (8)(66), the space-domain counterpart of (167)
may be written as [178, Sect. 13]
G
PQ
rjr
0

1
2p
2
_
1
1
_
1
1
2jk
zN
~
GG
PQ
k
r
; z
0
jz
0
e
jkr
.
r
0
0
e
jk
.
r
2jk
zN
_ _
dk
x
dk
y
168
kk
r
^ zzk
zN
; r
0
0
q
0
^ zzz
0
169
The integrand in (168) has been factorized into a slowly
varying part and a part, enclosed in curly braces, that var-
ies rapidly as r |r|-N.
40
The integrals of the form(168)
are amenable to the stationary phase method, or its equiv-
alent, the steepest descent method [3, p. 594: 4, p. 386].
The stationary phase point is given by the conditions
@
@k
x
k
.
r 0;
@
@k
y
k
.
r 0 170
which yield the solution
k
s
x
k
N
x
r
; k
s
y
k
N
y
r
; k
s
zN
k
N
z
r
171
where the superscript s has been introduced to distin-
guish the stationary phase point values. This result has a
clear physical interpretation; the stationary phase point
corresponds to a plane wave with its propagation vector k
pointing to the far-zone eld point [179]. Hence, we can
write
k
s
r
k
N
sinW ^ xxcos f ^ yy sinf; k
s
zN
k
N
cos W 172
where the spherical angles (W, j) specify the observation
direction. If we now replace the slowly varying part of the
integrand in (168) by its value at the stationary phase
point and perform the spectral integration in cylindrical
j = (x, y)
V
J
M
r'
z
N z
0
(z)
O
r
0
'
r
r r
0
'
T
o
a
d
is
ta
n
t
fie
ld
p
o
in
t
Figure 12. Geometry for the evaluation of the far-zone eld.
40
The reason for arbitrarily including the denominator 2jk
zN
in
the rapidly varying part will become evident in due course.
coordinates, we arrive at
G
PQ
rjr
0
$ 2jk
s
zN
~
GG
PQ
k
s
r
; z
0
jz
0
e
jk
s
r
.
r
0
0
S
0
e
jk
zN
z
2jk
zN
_ _
;
r !1
173
where S
0
is the zero-order Hankel transform dened in
(65). Finally, we invoke the Sommerfeld identity (164), to
obtain
G
PQ
rjr
0
$
e
jk
N
r
2pjr
e
jk
s
zN
z
0
k
s
zN
~
GG
PQ
k
s
r
; z
0
jz
0
e
jk
s
r
.
q
0
;
r !1
174
The far-zone electric eld can now be computed from (6),
with the DGFs transformed according to (174). Note that
at the stationary phase point ~ uu ^ qq and ^ vv ^ uu, where the
directions of the unit vectors ^ qq and ^ uu ^ zz ^ qq are xed by
the observation angle j. Furthermore, since there are no
reected waves for z4z
0
, the spectral-domain DGFs can be
simplied by using the transmission line relationships
(33),(34). As a result, we obtain the far-zone electric eld
components as [180,181]
E
W;j
$
e
jk
N
r
2pjr
e
jk
s
zN
z
0
k
N
f
EJ
W;j
e
jk
s
r
.
q
0
; J
_ _
Z
N
f
EM
W;j
e
jk
s
r
.
q
0
; M
_ _ _ _
;
r !1
175
where
f
EJ
W
V
e
i
z
0
jz
0
^ qq
Z
2
0
m
N
e
0
z
sinW cos WI
e
v
z
0
jz
0
^ zz 176
f
EJ
j
cos WV
h
i
z
0
jz
0
^ uu 177
f
EM
W
cos WI
e
v
z
0
jz
0
^ uu 178
f
EM
j
I
h
v
z
0
jz
0
^ qq
e
N
Z
2
0
m
0
z
sinW cos WV
h
i
z
0
jz
0
^ zz 179
It is understood here that the TLGFs, which can be com-
puted as explained in Section 7, are evaluated with
k
r
k
s
r
k
N
sinW. Finally, the far-zone magnetic eld com-
ponents corresponding to (175) can be found as [27, p. 133]
H
j
E
W
Z
N
; H
W

E
j
Z
N
180
The simple far-zone approximations derived here are not
applicable when the observation point approaches the sur-
face of the multilayer. In that case, there are singularities
near the stationary phase point and a modied saddle-
point technique must be used [3, p. 461; 12, p. 830].
10. PLANE-WAVE EXCITED FIELDS
In this section we show how the transmission-line formal-
ism can be used to compute the electromagnetic eld in
any layer of a multilayer stack with piecewise constant
parameters, due to a plane wave incident in the upper
half-space (layer N). We assume, for simplicity, that the
medium of the top layer is isotropic. The direction of ar-
rival of the plane wave is specied by the angles (W
i
, j
i
),
and its polarization by the angle c
i
, as illustrated in Fig.
13. Hence, the propagation vector of this wave can be
expressed as k
i
k
i
r
^ zzk
zN
, where
k
i
r
k
N
sinW
i
^ xxcos f
i
^ yy sinf
i
; k
zN
k
N
cos W
i
181
and the electric eld as
E
i
E
0
^ qq
i
cos W
i
^ zz sinW
i
cos c
i
^ uu
i
sinc
i
e
jk
zN
zz
N
e
jk
i
r
.
q
182
where E
0
is the specied eld amplitude and z
N
is the
phase reference point, selected to coincide with the upper
interface of the multilayer. Note that the directions of the
unit vectors ^ qq
i
and ^ uu
i
are xed by the angle j
i
, which
species the plane of incidence. Our goal is to nd the total
eld (E
n
, H
n
) established at any point of any layer n, as a
result of the plane wave (182) incident in the upper half-
space.
For the elds to match at the interfaces, the depen-
dence of (E
n
, H
n
) on the transverse coordinates must be
the same in all layers and must match that of the incident
plane wave (182). Furthermore, both the incident eld
and the total eld must satisfy the source-free forms of
(16)(19), with k
r
k
i
r
. Consequently, the transmission-
line analog of the layered medium developed in Section 7
may be employed to nd (E
n
, H
n
). With (22) in mind,
(y)
(z)
(x)
E
0
i
k
i
kj
i
i
Figure 13. Geometry for the evaluation of the plane-wave inci-
dent eld.
where now ~ uu; ^ vv ^ qq
i
; ^ uu
i
, we express the transverse part of
(182) as [180,181]
E
i
t
r ^ qq
i
V
e
N
z ^ uu
i
V
h
N
ze
jk
i
r
.
q
183
where
V
e
N
z E
0
cos c
i
cos W
i
..
V
e
N
e
jk
zN
zz
N
;
V
h
N
z E
0
sinc
i
..
V
h
N
e
jk
zN
zz
N
184
We may interpret V
e
N
z and V
h
N
z are the leftward-prop-
agating incident voltage waves exciting the TM and TE
transmission line networks, respectively, in the TL section
corresponding to the upper-half-space (layer N). In view of
(27),(28), the total electric and magnetic elds in any layer
n can be found as
E
n
r ^ qq
i
V
e
n
z ^ uu
i
V
h
n
z ^ zz
Z
N
e
N
e
zn
sinW
i
I
e
n
z
_ _
e
jk
i
r
.
q
185
H
n
r ^ qq
i
I
h
n
z ^ uu
i
I
e
n
z ^ zz
m
N
Z
N
m
zn
sinW
i
V
h
n
z
_ _
e
jk
i
r
.
q
186
where V
a
n
z and I
a
n
z, with a (e, h), are the total voltage
and current at a point z in layer n of the TL network.
These voltages and currents can be found as explained in
Section 7. Hence, in the upper layer they are given by
(125), with nN. In any other layer n, they can be found
from (126), where
V
a
n1
V
a
N
1 G
N
a
N1
k n1
t
k
a
187
with V
a
N
given in (184). Note that in deriving this equa-
tion, use has been made of (127) and (128). It is understood
in the above that all transmission-line voltages and cur-
rents are evaluated with k
r
k
i
r
k
N
siny
i
.
With reference to Fig. 13, observe that the polarization
of the incident plane wave is parallel for c
i
0, and per-
pendicular for c
i
p/2. From (184) we see that the paral-
lel- and perpendicularly polarized waves only excite the
TM and TE transmission lines, respectively, and that both
lines are excited by an arbitrarily polarized wave. In the
case of normal incidence, the eld is TEM and the plane
of incidence, and thus also j
i
and c
i
, are undened. How-
ever, this situation may be treated as a limiting case of
parallel or perpendicular polarization, as W
i
-0.
11. SUMMARY AND CONCLUSION
A complete set of electric- and magnetic-type dyadic Green
functions (DGFs) is derived for plane-stratied, multilay-
ered, uniaxial media, based on the transmission-line net-
work analog along the axis normal to the stratication.
The DGFs are expressed in terms of transmission-line
Green functions (TLGFs), which are the voltages and cur-
rents on the transmission-line network excited by unit-
strength voltage or current sources. A practical algorithm
is given for the efcient computation of the TLGFs. Also,
various state-of-the-art techniques for computation of the
Sommerfeld integrals are reviewed. Finally, the transmis-
sion-line analog is applied to derive the far-zone elds and
the plane-wave-excited elds in the layered medium. The
unied formulation presented here is compact, convenient
to implement, and computationally efcient. It also affords
much insight into the layered medium Green functions,
because the behavior of the transmission-line voltages and
currents is well understood.
12. APPENDIX: WHOLE-SPACE DGFS
In this appendix we give closed forms of the whole-space,
three-dimensional DGFs for a uniaxial medium with the
parameters of layer n. These expressions have been ob-
tained by retaining only the direct ray terms of the TLGFs
(corresponding to ray 0 in Fig. 10) in (69)(98) and using
the identities (164),(165) and their derivatives to evaluate
the resulting Sommerfeld integrals.
41
Equivalent forms
of these DGFs were derived previously [135] using a
different approach.
First, to simplify the equations, we dene
C
a
n
n
a
n
e
jknR
a
n
4pR
a
n
; D
a
n
e
jknR
a
n
4pjk
n
R
2
;
R
a
n
n
a
n
R
2
z z
0
2
_
A:1
S
a
n
R
R
a
n
; C
a
n

z z
0
R
a
n
A:2
L
a
n
j
k
n
R
a
n
1
k
n
R
a
n
2
; U
a
n
1
j
k
n
R
a
n
A:3
Using this notation, the whole-space terms of (69)(76)
can be written as
G
EJ
xx
rjr
0
jk
n
Z
n
f1 L
e
n
cos
2
fn
e
n
S
e
n
2
3L
e
n
1gC
e
n
sin
2
fC
e
n
C
h
n
cos2fD
e
n
D
h
n
g
A:4
41
The delta function terms in [76] and [98] are canceled in this
process; however, the z derivatives must be taken in the sense of
distributions [182].
G
EJ
xy
rjr
0
G
EJ
yx
rjr
0
jk
n
Z
n
fsin fcos fn
e
n
S
e
n
2
3L
e
n
1C
e
n
sin fcos fC
e
n
C
h
n
sin2fD
e
n
D
h
n
g
A:5
G
EJ
yy
rjr
0
jk
n
Z
n
f1 L
e
n
sin
2
fn
e
n
S
e
n
2
3L
e
n
1C
e
n
cos
2
fC
e
n
C
h
n
cos2fD
e
n
D
h
n
g
A:6
G
EJ
xz
rjr
0
G
EJ
zx
rjr
0
jk
n
Z
n
cos fS
e
n
C
e
n
3L
e
n
1C
e
n
A:7
G
EJ
yz
rjr
0
G
EJ
zy
rjr
0
jk
n
Z
n
sin fS
e
n
C
e
n
3L
e
n
1C
e
n
A:8
G
EJ
zz
rjr
0
jk
n
Z
n
1 L
e
n
C
e
n
2
3L
e
n
1
C
e
n
n
e
n
A:9
Similarly, the whole-space parts of (84)(90) are found
as
G
HJ
xx
rjr
0
G
HJ
yy
rjr
0
jk
n
sin fcos fC
e
n
U
e
n
C
e
n
C
h
n
U
h
n
C
h
n
sin2fC
e
n
D
e
n
C
h
n
D
h
n
A:10
G
HJ
xy
rjr
0
jk
n
sin
2
fC
e
n
U
e
n
C
e
n
cos
2
fC
h
n
U
h
n
C
h
n
cos2fC
e
n
D
e
n
C
h
n
D
h
n
A:11
G
HJ
yx
rjr
0
jk
n
cos
2
fC
e
n
U
e
n
C
e
n
sin
2
fC
h
n
U
h
n
C
h
n
cos2fC
e
n
D
e
n
C
h
n
D
h
n
A:12
G
HJ
xz
rjr
0
jk
n
sin fS
e
n
U
e
n
C
e
n
A:13
G
HJ
zx
rjr
0
jk
n
sin fS
h
n
U
h
n
C
h
n
A:14
G
HJ
yz
rjr
0
jk
n
cos fS
e
n
U
e
n
C
e
n
A:15
G
HJ
zy
rjr
0
jk
n
cos fS
h
n
U
h
n
C
h
n
A:16
It is noted that the whole-space G
EJ
is symmetric. One
also observes that considerable simplications in both G
EJ
and G
HJ
occur for isotropic media, in which case
R
e
n
R
h
n
R, C
e
n
C
h
n
, D
e
n
D
h
n
, S
a
n
R=R sinY and
C
a
n
z z
0
=R cos Y. In particular, G
EJ
properly reduc-
es to its isotropic form [2, p. 221] and G
HJ
becomes skew-
symmetric (i.e., it is equal to the negative of its transpose,
with zero diagonal).
In view of the duality between G
EJ
and G
HM
, the
whole-space components of the latter can be derived
from by making the substitution of symbols e-m and
m-e, which also entails the superscript replacement e-h
and h-e. Finally, the components of the whole-space G
EM
can be obtained from (A.10)(A.16) by using the recipro-
city relation (12), which in the present case (innite, ho-
mogeneous medium) simplies to G
EM
ij
rjr
0
G
HJ
ji
rjr
0
,
i.e., G
EM
is simply the transpose of G
HJ
.
The above expressions make evident the source region
singularities of the DGFs. Namely, G
EJ
and G
HM
behave
as R
3
, and G
HJ
and G
EM
behave as R
2
, where R rep-
resents either R
e
or R
h
.
BIBLIOGRAPHY
1. J. A. Kong, Electromagnetic Wave Theory, 2nd ed., Wiley,
New York, 1990.
2. J. Van Bladel, Electromagnetic Fields, Hemisphere, New
York, 1985.
3. A. Ishimaru, Electromagnetic Wave Propagation, Radiation,
and Scattering, Prentice-Hall, Englewood Cliffs, NJ, 1991.
4. L. B. Felsen and N. Marcuvitz, Radiation and Scattering of
Waves, Prentice-Hall, Englewood Cliffs, NJ, 1973.
5. M. Paulus and O. J. F. Martin, Light propagation and scat-
tering in stratied media: A Greens tensor approach, J. Opt.
Soc. Am. A 18:854861 (April 2001).
6. M. Huang and L. C. Shen, Computation of induction logs in
multiple-layer dipping formations, IEEE Trans. Geosci. Re-
mote Sens. 27:259267 (May 1989).
7. C. H. Chan, Spectral Greens functions for a thin, circular,
magnetic-current loop for microstrip via hole analysis, Mi-
crowave Opt. Technol. Lett. 2:157159 (May 1989).
8. D. Zheng and K. A. Michalski, Analysis of coaxially fed mi-
crostrip antennas of arbitrary shape with thick substrates,
J. Electromagn. Waves Appl. 5(12):13031327 (1991).
9. M. R. Abdul-Gaffoor, H. K. Smith, A. A. Kishk, and A. W.
Glisson, Simple and efcient full-wave modeling of electro-
magnetic coupling in realistic RF multilayer PCB layouts,
IEEE Trans. Microwave Theory Tech. 50:14451457 (June
2002).
10. N. Kinayman, G. Dural, and M. I. Aksun, A numerically ef-
cient technique for the analysis of slots in multilayer media,
IEEETrans. Microwave Theory Tech. 46:430432 (April 1998).
11. M. Javid and P. M. Brown, Field Analysis and Electromag-
netics, McGraw-Hill, New York, 1963.
12. R. E. Collin, Field Theory of Guided Waves, 2nd ed., IEEE
Press, New York, 1991.
13. W. C. Chew, Waves and Fields in Inhomogeneous Media,
IEEE Press, New York, 1995.
14. K. A. Michalski and J. R. Mosig, Multilayered media Greens
functions in integral equation formulations, IEEE Trans. An-
ten. Propag. 45:508519 (March 1997) (invited review paper).
15. T. Itoh, Spectral-domain immittance approach for dispersion
characteristics of generalized printed transmission lines,
IEEE Trans. Microwave Theory Tech. MTT-28:733736
(July 1980).
16. R. Kastner, E. Heyman, and A. Sabban, Spectral domain it-
erative analysis of single- and double-layered microstrip an-
tennas using the conjugate gradient algorithm, IEEE Trans.
Anten. Propag. 36:12041212 (Sept. 1988).
17. Y. T. Lo, S. M. Wright, and M. Davidovitz, Microstrip anten-
nas, in K. Chang, ed., Handbook of Microwave and Optical
Components, Wiley, New York, 1989, Vol. 1, Chap. 13.
18. K. A. Michalski and D. Zheng, Rigorous analysis of open
microstrip lines of arbitrary cross section in bound and leaky
regimes, IEEETrans. Microwave Theory Tech. 37:20052010
(Dec. 1989).
19. J. F. Kiang, S. M. Ali, and J. A. Kong, Integral equation so-
lution to the guidance and leakage properties of coupled di-
electric strip waveguides, IEEE Trans. Microwave Theory
Tech. 38:193203 (Feb. 1990).
20. N. Fache, F. Olyslager, and D. De Zutter, Electromagnetic
and Circuit Modelling of Multiconductor Transmission
Lines, Clarendon Press, Oxford, 1993.
21. M. Paulus and O. J. F. Martin, Greens tensor technique for
scattering in two-dimensional stratied media, Phys. Rev. E
63:066615 (2001).
22. J. Bernal, F. Medina, and R. R. Boix, Full-wave analysis of
nonplanar transmission lines on layered medium by means
of MPIE and complex image theory, IEEE Trans. Microwave
Theory Tech. 49:177185 (Jan. 2001).
23. P. E. Mayes, Electromagnetics for Engineers, Dept. Electrical
and Computer Eng., Univ. Illinois, Urbana, IL, 1965.
24. W. L. Weeks, Electromagnetic Theory for Engineering Appli-
cations, Wiley, New York, 1964.
25. D. G. Dudley, Mathematical Foundations for Electromagnet-
ic Theory, IEEE Press, New York, 1994.
26. G. W. Hanson and A. B. Yakovlev, Operator Theory for Elec-
tromagnetics. An Introduction, Springer-Verlag, New York,
2002.
27. R. F. Harrington, Time-Harmonic Electromagnetic Fields,
McGraw-Hill, New York, 1961.
28. D. M. Pozar, Input impedance and mutual coupling of rect-
angular microstrip antennas, IEEE Trans. Anten. Propag.
AP-30:11911196 (Nov. 1982).
29. N. K. Das and D. M. Pozar, A generalized spectral-domain
Greens function for multilayer dielectric substrates with
application to multilayer transmission lines, IEEE Trans.
Microwave Theory Tech. MTT-35:326335 (March 1987).
30. W. C. Chew and Q. Liu, Resonance frequency of a rectangu-
lar microstrip patch, IEEE Trans. Anten. Propag. 36:1045
1056 (Aug. 1988).
31. K. A. Michalski and D. Zheng, A spectral domain method for
the analysis of the fundamental mode leakage effect in mi-
crostrip lines on uniaxial substrates, Microwave Opt. Tech-
nol. Lett. 4:158161 (March 1991).
32. T.-S. Horng, W. E. McKinzie, and N. G. Alexopoulos, Full-
wave spectral-domain analysis of compensation of microstrip
discontinuities using triangular subdomain functions, IEEE
Trans. Microwave Theory Tech. 40:21372147 (Dec. 1992).
33. T. Becks and I. Wolff, Analysis of 3-D metallization struc-
tures by a full-wave spectral domain technique, IEEE Trans.
Microwave Theory Tech. 40:22192227 (Dec. 1992).
34. G. Plaza, F. Mesa, and F. Medina, Treatment of singularities
and quasi-static terms in the EFIE analysis of planar struc-
tures, IEEE Trans. Anten. Propag. 50:485491 (April 2002).
35. C. M. Butler, A. Q. Martin, and K. A. Michalski, Analysis of a
cylindrical antenna in a circular aperture in a screen,
J. Electromagn. Waves Appl. 8(2):149173 (1994).
36. I. N. Sneddon, Fourier Transforms, McGraw-Hill, New York,
1951.
37. G. Tyras, Radiation and Propagation of Electromagnetic
Waves, Academic Press, New York, 1969.
38. M. Abramowitz and I. A. Stegun, eds., Handbook of Math-
ematical Functions, Dover, New York, 1965.
39. C. T. Tai, Dyadic Green Functions in Electromagnetic Theory,
2nd ed., IEEE Press, New York, 1994.
40. A. Sommerfeld, U
ber die Ausbreitung der Wellen in der

drahtlosen Telegraphie, Ann. Phys. 28:665736 (1909);
81:11351153 (1926).
41. M. Tsai, C. Chen, and N. G. Alexopoulos, Sommerfeld inte-
grals in modeling interconnects and microstrip elements in
multi-layered media, Electromagnetics 18:267288 (1998).
42. M. Metcalf and J. Reid, FORTRAN 90/95 explained, Oxford
Univ. Press, Oxford, UK, 1996.
43. P. Yla -Oijala, M. Taskinen, and J. Sarvas, Multilayered me-
dia Greens functions for MPIE with general electric and
magnetic sources by the Hertz potential approach, in J. A.
Kong, ed., Progress in Electromagnetics Research, EMW
Publ., Cambridge, MA; 2001, Vol. PIER 33, pp. 141165.
44. C.-I. G. Hsu, R. F. Harrington, K. A. Michalski, and D.
Zheng, Analysis of a multiconductor transmission lines of
arbitrary cross-section in multilayered uniaxial media,
IEEE Trans. Microwave Theory Tech. 41:7078 (Jan. 1993).
45. Y. Yamaguchi, H. Miyashita, and I. Chiba, A general method of
construction for dyadic Green functions of plane-stratied me-
dia in uniform open or closed waveguides with separable cross
sections, Electron. Commun. Japan Pt. 1 82(1):3142 (1999).
46. C. Tai, Equivalent layers of surface charge, current sheet,
and polarization in the eigenfunction expansion of Greens
functions in electromagnetic theory, IEEE Trans. Anten.
Propag. AP-29:733739 (Sept. 1981).
47. P. H. Pathak, On the eigenfunction expansion of electromag-
netic dyadic Greens functions, IEEE Trans. Anten. Propag.
AP-31:837846 (Nov. 1983).
48. K. A. Michalski, Missing boundary conditions of electromag-
netics, Electron. Lett. 22(17):921922 (1986).
49. J. Van Bladel, Singular Electromagnetic Fields and Sources,
Clarendon Press, Oxford, 1991.
50. W. A. Johnson, A. Q. Howard, and D. G. Dudley, On the ir-
rotational component of the electric Greens dyadic, Radio
Sci. 14:961967 (Nov.Dec. 1979).
51. W. S. Weiglhofer, Electromagnetic eld in the source region:
A review, Electromagnetics 19:563578 (1999).
52. J. Van Bladel, Some remarks on Greens dyadic for innite
space, IEEETrans. Anten. Propag. AP-9:563566 (Nov. 1961).
53. K. Chen, A simple physical picture of tensor Greens function
in source region, IEEE Proc. 65:12021204 (Aug. 1977).
54. A. D. Yaghjian, Electric dyadic Greens functions in the
source region, IEEE Proc. 68:248263 (Feb. 1980).
55. P. Weidelt, Electromagnetic induction in three-dimensional
structures, J. Geophys. 41:85109 (1975).
56. R. C. Robertson, Computation of the electric eld within a
region of uniform current density, Int. J. Electron. 69:707
716 (1990).
57. W. S. Weiglhofer and A. Lakhtakia, New expressions for de-
polarization dyadics in uniaxial dielectric-magnetic media,
Int. J. Infrared Millim. Waves 17(8):13651376 (1996).
58. W. C. Chew, Some observations on the spatial and eigen-
function representations of dyadic Greens functions, IEEE
Trans. Anten. Propag. 37:13221327 (Oct. 1989).
59. J. G. Fikioris, Electromagnetic eld inside a current-carry-
ing region, J. Math. Phys. 6:16171620 (Nov. 1965).
60. S. W. Lee, J. Boersma, C. L. Law, and G. A. Deschamps, Sin-
gularity in Greens function and its numerical evaluation,
IEEE Trans. Anten. Propag. AP-28:311317 (May 1980).
61. J. J. H. Wang, A unied and consistent view on the singu-
larities of the dyadic Greens function in the source region,
IEEE Trans. Anten. Propag. AP-30:463468 (May 1982).
62. A. Lakhtakia and W. S. Weiglhofer, Time-harmonic electromag-
netic eld in a source region in a uniaxial dielectric-magnetic
medium, Int. J. Appl. Electromagn. Mech. 8:167177 (1997).
63. K. A. Michalski, The mixed-potential electric eld integral
equation for objects in layered media, Arch. Elek. U
bertra-
gung. 39:317322 (Sept.Oct. 1985).
64. K. A. Michalski and D. Zheng, Electromagnetic scattering
and radiation by surfaces of arbitrary shape in layered me-
dia, Part I: Theory, IEEE Trans. Anten. Propag. 38:335344
(March 1990).
65. D. R. Wilton, Review of current status and trends in the use
of integral equations in computational electromagnetics,
Electromagnetics 12:287341 (JulyDec. 1992).
66. K. A. Michalski, Mixed-potential integral equation (MPIE)
formulation for nonplanar microstrip structures of arbitrary
shape in multilayered uniaxial media, Int. J. Microwave Mill-
im. Wave Comput. Aided Eng. 3(4):420431 (1993) (invited
article).
67. K. A. Michalski, Formulation of mixed-potential integral equa-
tions for arbitrarily shaped microstrip structures with uniaxial
substrates, J. Electromagn. Waves Appl. 7(7):899917 (1993).
68. R. D. Cloux, G. P. J. F. M. Maas, and A. J. H. Wachters,
Quasi-static boundary element method for electromagnetic
simulation of PCBS, Philips J. Res. 48(12):117144 (1994).
69. J. Sercu, N. Fache, F. Libbrecht, and P. Lagasse, Mixed
potential integral equation technique for hybrid micro-
strip-slotline multilayered circuits using a mixed rectangu-
lar-triangular mesh, IEEE Trans. Microwave Theory Tech.
43:11621172 (May 1995).
70. R. Bunger and F. Arndt, Efcient MPIE approach for the
analysis of three-dimensional microstrip structures in lay-
ered media, IEEE Trans. Microwave Theory Tech. 45:1141
1153 (Aug. 1997).
71. M. Vrancken and G. A. E. Vandenbosch, Semantics of dyadic
and mixed potential eld representation for 3-D current dis-
tributions in planar stratied media, IEEE Trans. Anten.
Propag. 51:27782787 (Oct. 2003).
72. L. Vegni, R. Cicchetti, and P. Capece, Spectral dyadic Greens
function formulation for planar integrated structures, IEEE
Trans. Anten. Propag. 36:10571065 (Aug. 1988).
73. P. Bernardi and R. Cicchetti, Dyadic Greens functions for
conductor-backed layered structures excited by arbitrary
tridimensional sources, IEEE Trans. Microwave Theory
Tech. 42:14741483 (Aug. 1994).
74. S.-G. Pan and I. Wolff, Scalarization of dyadic spectral
Greens functions and network formalism for three-dimen-
sional full-wave analysis of planar lines and antennas, IEEE
Trans. Microwave Theory Tech. 42:21182127 (Nov. 1994).
75. J. A. Kong, Antenna radiation in stratied media, in J. A.
Kong, ed., Research Topics in Electromagnetic Wave Theory,
Wiley, New York, 1981, pp. 211234.
76. Y. S. Kwon and J. J. H. Wang, Computation of Hertzian di-
pole radiation in stratied uniaxial anisotropic media, Radio
Sci. 21:891902 (Nov.Dec. 1986).
77. W. C. Chew, J. S. Zhao, and T. J. Cui, The layered medium
Greens functiona new look, Microwave Opt. Technol. Lett.
31(4):252255 (2001).
78. J. R. Wait, Electromagnetic Wave Theory, Harper & Row,
New York, 1985.
79. P. E. Wannamaker, G. W. Hohmann, and W. A. SanFilipo,
Electromagnetic modeling of three-dimensional bodies in
layered earths using integral equations, Geophysics 49:60
74 (Jan. 1984).
80. V. W. Hansen, Numerical Solution of Antennas in Layered
Media, Research Studies Press, Taunton, UK, 1989.
81. S. Barkeshli and P. H. Pathak, On the dyadic Greens func-
tion for a planar multilayered dielectric/magnetic media,
IEEE Trans. Microwave Theory Tech. 40:128142 (Jan.
1992).
82. S. Barkeshli, On the electromagnetic dyadic Greens func-
tions for planar multi-layered anisotropic uniaxial material
media, Int. J. Infrared Millim. Waves 13(4):507527 (1992).
83. A. K. Bhattacharyya, Electromagnetic Fields in Multilay-
ered Structures, Artech House, Boston, 1994.
84. A. Sommerfeld, Partial Differential Equations in Physics,
Academic Press, New York, 1949.
85. A. Ban os, Dipole Radiation in the Presence of a Conducting
Half-Space, Pergamon Press, New York, 1966.
86. A. Erteza and B. K. Park, Nonuniqueness of resolution of
Hertz vector in presence of a boundary, and the horizontal
dipole problem, IEEE Trans. Anten. Propag. AP-17:376378
(May 1969).
87. K. A. Michalski, On the scalar potential of a point charge
associated with a time-harmonic dipole in a layered medi-
um, IEEE Trans. Anten. Propag. AP-35:12991301 (Nov.
1987).
88. G. J. Burke, E. K. Miller, J. N. Brittingham, D. L. Lager,
R. J. Lytle, and J. T. Okada, Computer modeling of antennas
near the ground, Electromagnetics 1:2949 (Jan.March 1981).
89. R. R. Chance, A. Prock, and R. Silbey, Molecular uorescence
and energy transfer near interfaces, Adv. Chem. Phys. 37:
165 (1987).
90. M. A. Taubenblatt and T. K. Tran, Calculation of light scat-
tering from particles and structures on a surface by the cou-
pled-dipole method, J. Opt. Soc. Am. A10:912919 (May 1993).
91. R. Schmehl, B. M. Nebeker, and E. D. Hirleman, Discrete-
dipole approximation for scattering by features on surfaces
by means of a two-dimensional fast Fourier transform tech-
nique, J. Opt. Soc. Am. A 14:30263036 (Nov. 1997).
92. J. R. Wait, Fields of a horizontal dipole over a stratied uni-
form half-space, IEEE Trans. Anten. Propag. AP-14:790792
(Nov. 1966).
93. A. P. Raiche, An integral equation approach to three-dimen-
sional modelling, Geophys. J. Roy. Astrophys. Soc. 36:363
376 (1974).
94. C. H. Stoyer, Electromagnetic elds of dipoles in stratied me-
dia, IEEE Trans. Anten. Propag. AP-25:547552 (July 1977).
95. J. S. Bagby and D. P. Nyquist, Dyadic Greens functions for
integrated electronic and optical circuits waveguides, IEEE
Trans. Microwave Theory Tech. MTT-35:206210 (Feb. 1987).
96. Z. Xiong, Electromagnetic elds of electric dipoles embedded
in a stratied anisotropic earth, Geophysics 54:16431646
(Dec. 1989).
97. G. P. S. Cavalcante, D. A. Rogers, and A. J. Giarola, Analysis
of electromagnetic wave propagation in multilayered media
using dyadic Greens functions, Radio Sci. 17:503508
(MayJune 1982).
98. D. H. S. Cheng, On the formulation of the dyadic Greens
function in a layered medium, Electromagnetics 6(2):171
182 (1986).
99. L. Tsang, J. A. Kong, and R. T. Shin, Theory of Microwave
Remote Sensing, Wiley, New York, 1985.
100. L. Tsang, E. Njoku, and J. A. Kong, Microwave ther-
mal emission from a stratied medium with nonuniform
temperature distribution, J. Appl. Phys. 46:51275133 (Dec.
1975).
101. S. M. Ali and S. F. Mahmoud, Electromagnetic elds of bur-
ied sources in stratied anisotropic media, IEEE Trans. An-
ten. Propag. AP-27:671678 (Sept. 1979).
102. T. Sphicopoulos, V. Teodoridis, and F. E. Gardiol, Dyadic
Green function for the electromagnetic eld in multilayered
isotropic media: An operator approach, IEE Proc. H 132:
329334 (Aug. 1985).
103. J. E. Sipe, New Green-function formalism for surface optics,
J. Opt. Soc. Am. B 4:481489 (April 1987).
104. E. W. Kolk, N. H. G. Baken, and H. Blok, Domain integral
equation analysis of integrated optical channel and ridge
waveguides in stratied media, IEEE Trans. Microwave
Theory Tech. 38:7885 (Jan. 1990).
105. M. Paulus, P. Gay-Balmaz, and O. J. F. Martin, Accurate and
efcient computation of the Greens tensor for stratied
media, Phys. Rev. E 62:57975807 (Oct. 2000).
106. L. Vegni, F. Bilotti, and A. Toscano, Microstrip disk antennas
with inhomogeneous articial dielectrics, J. Electromagn.
Waves Appl. 14:12031227 (2000).
107. P. Yeh, Optical Waves in Layered Media, Wiley, New York,
1988.
108. I. P. Theron and J. H. Cloete, On the surface impedance used
to model the conductor losses of microstrip structures, IEE
Proc. Microwave Anten. Propag. 142:3540 (Feb. 1995).
109. J. Berenger, Perfectly matched layer for the absorption of
electromagnetic waves, J. Comput. Phys. 114:185200 (Oct.
1994).
110. Z. S. Sacks, D. M. Kingsland, R. Lee, and J. Lee, A perfectly
matched anisotropic absorber for use as an absorbing bound-
ary condition, IEEE Trans. Anten. Propag. 43:14601463
(Dec. 1995).
111. S. D. Gedney and A. Taflove, Perfectly matched layer ab-
sorbing boundary conditions, in: A. Taflove and S. C. Hag-
ness, eds., Computational Electrodynamics: The Finite-
Difference Time-Domain Method, Artech House, Boston,
2000, pp. 285348.
112. L. M. Brekhovskikh, Waves in Layered Media, 2nd ed.,
Academic Press, New York, 1980.
113. J. Chilwell and I. Hodgkinson, Thin-lms eld-transfer ma-
trix theory of planar multilayer waveguides and reection
from prism-loaded waveguides, J. Opt. Soc. Am. A 1:742753
(July 1984).
114. L. M. Walpita, Solution for planar optical waveguide equa-
tions by selecting zero elements in a characteristic matrix,
J. Opt. Soc. Am. A. 2:595602 (April 1985).
115. Y. L. Chow, N. Hojjat, S. Safavi-Naeini, and R. Faraji-Dana,
Spectral Greens functions for multilayer media in a conve-
nient computational form, IEE Proc. Microwave Anten.
Propag. 145:8591 (Feb. 1998).
116. L. Polerecky, J. Hamrle, and B. D. MacCraith, Theory of the
radiation of dipoles placed within a multilayer system, Appl.
Opt. 39:39683977 (Aug. 2000).
117. M. Born and E. Wolf, Principles of Optics. Electromagnetic
Theory of Propagation, Interference and Diffraction of Light,
7th ed., Cambridge Univ. Press, Cambridge, UK, 1999.
118. H. Y. Yang and N. G. Alexopoulos, Gain enhancement meth-
ods for printed circuit antennas through multiple superst-
rates, IEEE Trans. Anten. Propag. AP-35:860863 (July
1987).
119. C. Montcalm, B. T. Sullivan, H. Pepin, J. A. Dobrowolski, and
M. Sutton, Extreme-ultraviolet Mo/Si multilayer mirrors
deposited by radio-frequency-magnetron sputtering, Appl.
Opt. 33:20572068 (1994).
120. J. M. Vigoureux, Polynomial formulation of reection and
transmission by stratied planar structures, J. Opt. Soc.
Am. A 8:16971701 (Nov. 1991).
121. R. E. Smith, S. N. Houde-Walter, and G. W. Forbes, Mode
determination for planar waveguides using the four-sheeted
dispersion relation, IEEE J. Quantum Electron. 28:1520
1526 (June 1992).
122. R. Rodriguez-Berral, F. Mesa, and F. Medina, Systematic
and efcient root nder for computing the modal spectrumof
planar layered waveguides, Int. J. RF Microwave Comput.
Aided Eng. 14:7383 (Jan. 2004).
123. L. M. Delves and J. N. Lyness, A numerical method for lo-
cating the zeros of an analytic function, Math. Comput.
21:543560 (Oct. 1967).
124. T. Li, On locating all zeros of an analytic function within a
bounded domain by a revised Delves/Lyness method, SIAM
J. Num. Anal. 20:865871 (Aug. 1983).
125. N. Fache, J. Van Hesse, and D. De Zutter, Generalised space
domain Greens dyadic for multilayered media with special
application to microwave interconnections, J. Electromagn.
Waves Appl. 3(7):651669 (1989).
126. W. C. Chew and S. Chen, Response of a point source embed-
ded in a layered medium, IEEEAnten. Wireless Propag. Lett.
2:254258 (2003).
127. I. T. Lu, L. B. Felsen, and A. H. Kamel, Eigenreverberations,
eigenmodes and hybrid combinations: A new approach to
propagation in layered multiwave media, Wave Motion
6:435457 (May 1984).
128. K. A. Michalski, Extrapolation methods for Sommerfeld in-
tegral tails, IEEE Trans. Anten. Propag. 46:14051418 (Oct.
1998) (invited review paper).
129. N. J. Champagne, J. T. Williams, and D. R. Wilton, Analysis
of resistively loaded, printed spiral antennas, Electromag-
netics 14:363395 (JulyDec. 1994).
130. R. Mittra and S. W. Lee, Analytical Techniques in the Theory
of Guided Waves, Macmillan, New York, 1971.
131. E. Arbel and L. B. Felsen, Theory of radiation from sources
in anisotropic media, Part 1: General sources in strat-
ied media, in E. C. Jordan, ed., Electromagnetic Theory
and Antennas, Part 1. Macmillan, New York, 1963, pp.
391420.
132. P. C. Clemmow, The Plane Wave Spectrum Representation of
Electromagnetic Fields, Pergamon Press, New York, 1966.
133. H. C. Chen, Theory of Electromagnetic Fields. A Coordinate-
Free Approach, McGraw-Hill, New York, 1983.
134. J. R. Wait, Geo-Electromagnetism, Academic Press, New
York, 1982.
135. W. S. Weiglhofer, Dyadic Greens functions for general uni-
axial media, IEE Proc. H 137:510 (Feb. 1990).
136. I. V. Lindell, Methods for Electromagnetic Field Analysis,
Clarendon Press, Oxford, 1992.
137. A. Hessel, General characteristics of traveling-wave anten-
nas, in R. E. Collin and F. J. Zucker, eds., Antenna Theory
Part 2, McGraw-Hill, New York, 1969, pp. 151258.
138. P. Gay-Balmaz and J. R. Mosig, Three dimensional pla-
nar radiating structures in stratied media, Int. J. Micro-
wave Millim. Wave Comput. Aided Eng. 7:330343 (Sept.
1997).
139. W. R. LePage, Complex Variables and the Laplace Transform
for Engineers, Dover, New York, 1980.
140. N. K. Uzunoglu, N. G. Alexopoulos, and J. G. Fikioris, Radi-
ation properties of microstrip dipoles, IEEE Trans. Anten.
Propag. AP-27:853858 (Nov. 1979).
141. J. R. Mosig and F. E. Gardiol, A dynamical radiation model
for microstrip structures, in P. W. Hawkes, ed., Advances in
Electronics and Electron Physics, Academic Press, New
York, 1982, Vol. 59, pp. 139237.
142. W. A. Johnson and D. G. Dudley, Real axis integration of
Sommerfeld integrals: Source and observation points in air,
Radio Sci. 18:175186 (MarchApril 1983).
143. V. Teodoridis, T. Sphicopoulos, and F. Gardiol, The reection
from an open-ended rectangular waveguide terminated by a
layered dielectric medium, IEEE Trans. Microwave Theory
Tech. MTT-33:359366 (May 1985).
144. A. D. Chave, Numerical integration of related Hankel trans-
forms by quadrature and continued fraction expansion, Geo-
physics 48:16711686 (Dec. 1983).
145. J. R. Mosig, Integral equation technique, in T. Itoh, ed.,
Numerical Techniques for Microwave and Millimeter-
Wave Passive Structures, Wiley, New York, 1989, pp.
133213.
146. S. K. Lucas and H. A. Stone, Evaluating innite integrals
involving Bessel functions of arbitrary order, J. Comput.
Appl. Math. 64:217231 (1995).
147. G. Evans, Practical Numerical Integration, Wiley, New York,
1993.
148. H. H. H. Homeier, Scalar Levin-type sequence transforma-
tions, J. Comput. Appl. Math. 122:81147 (2000).
149. A. Sidi, Practical Extrapolation Methods. Theory and Appli-
cations, Cambridge Univ. Press, Cambridge, UK, 2003.
150. J. Chen, A. A. Kishk, and A. W. Glisson, A 3D interpolation
method for the calculation of the Sommerfeld integrals to
analyze dielectric resonators in a multilayered medium,
Electromagnetics 20(1):115 (2000).
151. J. R. Mosig and A. A. Melecon, Greens functions in lossy
layered media: Integration along the imaginary axis and
asymptotic behavior, IEEE Trans. Anten. Propag. 51:3200
3208 (Dec. 2003).
152. K. A. Michalski, On the efcient evaluation of integrals aris-
ing in the Sommerfeld halfspace problem, in R. C. Hansen,
ed., Moment Methods in Antennas and Scatterers, Artech
House, Boston, 1990, pp. 325331.
153. S. Barkeshli, P. H. Pathak, and M. Marin, An asymptotic
closed-form microstrip surface Greens function for the ef-
cient moment method analysis of mutual coupling in micro-
strip antennas, IEEE Trans. Anten. Propag. 38:13741383
(Sept. 1990).
154. B. Popovski, A. Toscano, and L. Vegni, Radial and asymptotic
closed form representation of the spatial microstrip dyadic
Greens functions, J. Electromagn. Waves Appl. 9:97126
(1995).
155. T. J. Cui and W. C. Chew, Fast evaluation of Sommerfeld in-
tegrals for EM scattering and radiation by three-dimension-
al buried objects, IEEE Trans. Geosci. Remote Sens. 37:887
900 (March 1999).
156. F. Olyslager and H. Derudder, Series representation of
Green dyadics for layered media using PMLs, IEEE Trans.
Anten. Propag. 51:23192326 (Sept. 2003).
157. P. Cornille, Numerical saddle point method, J. Math. Anal.
Appl. 38:633639 (1972).
158. R. E. Smith and S. N. Houde-Walter, The migration of bound
and leaky solutions to the waveguide dispersion relation,
J. Lightwave Technol. 11:17601768 (Nov. 1993).
159. W. L. Anderson, Computation of Greens tensor integrals for
three-dimensional electromagnetic problems using fast
Hankel transforms, Geophysics 49:17541759 (Oct. 1984).
160. R. Hsieh and J. Kuo, Fast full-wave analysis of planar
microstrip circuit elements in stratied media, IEEE Trans.
Microwave Theory Tech. 46:12911297 (Sept. 1998).
161. D. G. Fang, J. J. Yang, and G. Y. Delisle, Discrete image the-
ory for horizontal electric dipoles in a multilayered medium,
IEE Proc. H 135:297303 (Oct. 1988).
162. K. A. Michalski and J. R. Mosig, Discrete complex image
mixed-potential integral equation analysis of microstrip
patch antennas with vertical probe feeds, Electromagnetics
15:377392 (JulyAug. 1995) (invited paper).
163. M. I. Aksun, A robust approach for the derivation of closed-
form Greens functions, IEEE Trans. Microwave Theory
Tech. 44:651658 (May 1996).
164. F. Ling and J. Jin, Discrete complex image method for
Greens functions of general multilayer media, IEEE Micro-
wave Guided Wave Lett. 10:400402 (Oct. 2000).
165. P. Yla -Oijala and M. Taskinen, Efcient formulation of
closed-form Greens functions for general electric and mag-
netic sources in multilayered media, IEEE Trans. Anten.
Propag. 51:21062115 (Aug. 2003).
166. N. B. Christensen, Optimized fast Hankel transform lters,
Geophys. Prosp. 38:545558 (1990).
167. D. Guptasarma and B. Singh, New digital linear lters for
Hankel J
0
and J
1
transforms, Geophys. Prosp. 45:745762
(1997).
168. T. K. Sarkar and O. Pereira, Using the matrix pencil method
to estimate the parameters of a sum of complex exponen-
tials, IEEE Anten. Propag. Mag. 37:4855 (Feb. 1995).
169. V. I. Okhmatovski and A. C. Cangellaris, Evaluation of layered
media Greens functions via rational function tting, IEEE
Microwave Wireless Compon. Lett. 14:2224 (Jan. 2004).
170. L. Beyne and D. De Zutter, Greens function for layered lossy
media with special application to microstrip antennas, IEEE
Trans. Microwave Theory Tech. 36:875881 (May 1988).
171. J. L. Tsalamengas, TE-scattering by conducting strips right
on the planar interface of a three-layered medium, IEEE
Trans. Anten. Propag. 41:16501658 (Dec. 1993).
172. L. Tsang, C. Huang, and C. H. Chan, Surface electric elds
and impedance matrix elements of stratied media, IEEE
Trans. Anten. Propag. 48:15331543 (Oct. 2000).
173. K. A. Michalski and D. Zheng, Electromagnetic scattering and
radiation by surfaces of arbitrary shape in layered media, Part
II: Implementation and results for contiguous half-spaces,
IEEE Trans. Anten. Propag. 38:345352 (March 1990).
174. F. Ling and J. Jin, Full-wave analysis of multilayer micro-
strip problems, in W. C. Chew, J. Jin, E. Michielssen, and J.
Song, eds., Fast and Efcient Algorithms in Computational
Electromagnetics, Artech House, Boston; 2001, pp. 729780.
175. F. Mesa, C. Di Nallo, and D. R. Jackson, The theory of sur-
face-wave and space-wave leaky-mode excitation on micro-
strip lines, IEEE Trans. Microwave Theory Tech. 47:207215
(Feb. 1999).
176. O. J. F. Martin and N. B. Piller, Electromagnetic scattering
in polarizable backgrounds, Phys. Rev. E 58:39093915
(Sept. 1998).
177. F. Olyslager, Electromagnetic Waveguides and Transmission
Lines, Clarendon Press, Oxford, 1999.
178. S. W. Lee, Basics, in Y. T. Lo and S. W. Lee, eds., Antenna
Handbook: Theory, Applications, and Design, Van Nostrand-
Reinhold, New York, 1988, Chap. 1.
179. W. C. Chew, A quick way to approximate a Sommerfeld-
Weyl-type integral, IEEE Trans. Anten. Propag. 36:1654
1657 (Nov. 1988).
180. K. A. Michalski and C.-I. G. Hsu, RCS computation of coax-
loaded microstrip patch antennas of arbitrary shape, Elec-
tromagnetics 14(1):3363 (1994).
181. M.-H. Ho, K. A. Michalski, and K. Chang, Waveguide excited
microstrip patch antennatheory and experiment, IEEE
Trans. Anten. Propag. 42:11141125 (Aug. 1994).
182. Y. Rahmat-Samii, On the question of computation of the
dyadic Greens functions at source region in waveguides
and cavities, IEEE Trans. Microwave Theory Tech. MTT-
23:762765 (Sept. 1975).
ELECTROMAGNETIC FIELD MEASUREMENT
MOTOHISA KANDA
National Institute of Standards
and Technology
To establish standards for conducting electromagnetic
(EM) eld measurements, measurements must be made
(1) in anechoic chambers, (2) at open-area test sites, and
(3) within guided-wave structures; and a means to trans-
fer these measurements from one situation to another
must be developed. The underlying principles of these
measurement and transfer standards are (1) measure-
ments and (2) theoretical modeling. Thus, a parameter or
a set of parameters is measured, or a parameter is calcu-
lated by established physical and mathematical princi-
ples.
Various electromagnetic eld sensors for measuring ra-
diofrequency (RF) electric and magnetic elds are dis-
cussed below. For electric eld measurements, electrically
short dipole antennas with a high-input-impedance load,
such as a eld-effect transistor (FET) and a high-frequen-
cy diode detector, are discussed. Since the input imped-
ance of an electrically short dipole antenna is
predominantly a capacitive reactance, very broadband fre-
quency responses can be achieved with a high-impedance
capacitive load. However, because conventional dipole an-
tennas support a standing-wave current distribution, the
useful frequency range of these dipole antennas is usually
limited by their natural resonant frequencies. In order to
suppress these resonances, a resistively loaded dipole an-
tenna has been developed. To obtain a standard antenna
with increased sensitivity at a specific frequency, a half-
wave tuned dipole antenna with a diode is used, to mea-
sure the induced open-circuit voltage. Also used was a
tuned receiver with a half-wave tuned dipole antenna, to
further improve antenna sensitivity.
For magnetic eld measurements, this article discusses
an electrically small, resistively loaded loop antenna to
achieve a broadband response. Resistive loading is
achieved either with the loading resistance at the loop
terminal or by uniform resistive loading along the loop
antenna. This short-circuit current loop conguration
gives a very at frequency response over a wide fre-
quency range.
In the region near a transmitting antenna or a scat-
terer, the electric and magnetic eld vectors are not nec-
essarily (spatially) orthogonal or in phase. For time-
harmonic elds, the endpoints of the eld vectors trace
out polarization ellipses, and the Poynting vectors lie on
the surface of a cone with its endpoint on an ellipse. In
these cases, the electric and magnetic elds may be mea-
sured separately, or, using the single-loop antenna ele-
ment described in this article, they may be measured
simultaneously.
Photonic sensors are also discussed; they provide the
wide bandwidth and low dispersion necessary to maintain
the delity of time-domain signals. Since they consist of
electro-optic modulators and optical bers, they are free
from electromagnetic interference, and there is minimal
perturbation of the eld being measured.
Throughout the discussion, the interplay between mea-
sured quantities and predicted (modeled) quantities is
emphasized. The ability of measurements and the restric-
tions imposed by rigorous theoretical analysis of given
models are discussed for the frequencies from 10 kHz to
40 GHz and upward.
1. ELECTRIC FIELD SENSORS
1.1. An Electrically Short Dipole Antenna with a
Capacitive Load
Most electric eld sensors consist of dipole antennas. The
induced open-circuit voltage V
oc
at the dipole antenna ter-
minal is given by
V
oc
E
inc
L
eff
1
where E
inc
is the normal incident electric eld strength
and L
eff
is the effective length of the dipole antenna. For
an electrically short dipole antenna whose physical
length is much shorter than the wavelength, the effective
length L
eff
and driving point capacitance C
a
are approxi-
mately [1]
L
eff
LO 1
4O 2 ln4
2
and
C
a
4pe
0
L
2O 2 ln 4
3
where L is the physical length of the dipole antenna, e
0
is
the free-space permittivity, O is the antenna thickness
factor O2ln(L/a), and a is the antenna radius.
For an electrically short dipole antenna with a capac-
itive load C, the transfer function is given by [1]
Sf
V
o
f
E
inc
f

hk=2
1 C=C
a
4
1190 ELECTROMAGNETIC FIELD MEASUREMENT
where
C
a
4ph
cz
0
O 2 ln 4
5
k
O 1
O 2 ln 4
6
and where
a antenna radius
C capacitance of load
C
a
capacitance of antenna
E
inc
incident electric eld
h half of the physical length of the dipole antenna
V
o
output voltage of the antenna
z
0
free-space impedance
c speed of light in free space
O antenna thickness factor, O2 ln(2h/a)
Since the input impedance of an electrically short di-
pole antenna is almost purely capacitive, it is possible to
achieve a frequency-independent transfer function with a
capacitive load, as indicated in Eq. (4). In practice, the
load impedance is seldom purely capacitive, but also may
have a resistive component. This resistive component will
cause a 6-dB/octave rolloff at the low end of the frequency
range.
1.2. An Electrically Short Dipole Antenna with a Diode
A common standard sensor used for EM eld measure-
ments is an electrically short dipole antenna with a diode
load connected to a DC metering unit by an RF lter
transmission line. The main advantage of including a di-
ode is to make the frequency response of the sensor very
at, even at higher frequencies where an FET amplier is
not readily available. When used with a high-input-im-
pedance detector, the sensor can serve as a portable and
compact transfer standard.
Figure 1 represents a dipole antenna, which can be
used to determine the signal from the transmission line
for a given incident electric eld [2]. This circuit is valid
for the frequency range to be considered (DC and
f 41 MHz). In Fig. 1, the small antenna resistance has
been neglected. The stray gap capacitance of the lter line
from the antenna terminal connections has been included
in the effective shunt capacitance C
d
. The antenna capac-
itance is given by C
a
.
When a rst-order, nonlinear differential equation as-
sociated with the Thevenins equivalent nonlinear circuit
(shown in Fig. 1) is solved for the detected DC voltage V
o
,
for a small induced RF voltage V
i
[2] we obtain
V
o

a
4
V
i
1 C
d
=C
a
_ _
2
7
where a (C38 V
1
) is derived from the diode characteris-
tics, while for large V
i
V
o

V
i
1 C
d
=C
a
8
Equation (7) indicates that for a small induced RF voltage
V
i
, the output DC voltage V
o
is a square-law function of the
induced voltage. On the other hand, Eq. (8) indicates that,
for a large induced voltage V
i
, the output DC voltage V
o
is
directly proportional to the induced voltage.
1.3. A Resistively Loaded Dipole Antenna with a Diode
A conventional dipole antenna essentially supports a
standing-wave current distribution and is, therefore,
highly frequency-sensitive. For the antenna to have a
at frequency response beyond any natural resonant fre-
quency, a traveling-wave dipole antenna was realized by
use of continuously tapered resistive loading [3,4]. If the
internal impedance per unit length Z
i
(z) as a function of
the axial coordinate z is expressed as
Z
i
z
60C
h jzj
9
then the current distribution I
z
(z) along the linear anten-
na is that of a traveling wave
I
z
z
V
i
60C1 j=kh
1
jzj
h
_ _
e
jkjzj
10
where 2h is the dipole antennas total physical length, k is
the wavenumber, V
i
is the driving voltage, and C is given
by
C2 sinh
1
h
a
C2ka; 2kh jS2ka; 2kh
_ _
j
kh
1 e
j2kh
11
where a is the radius of the dipole and C(x,y) and S(x,y) are
the generalized cosine and sine integrals. The main ad-
vantage of a resistively loaded dipole antenna with a diode
is the very at frequency response of the sensor system. A
shortcoming of this sensor system is the relatively low
sensitivity. To overcome this problem, a standard half-
wave tuned dipole antenna should be used.
1.4. A Tuned Half-Wave Dipole Antenna
The magnitude of the electric eld component at a given
point in an electromagnetic eld is determined from the
open-circuit voltage V
oc
induced in a standard half-wave
i
(t ) v
o
(t ) v
d
(t ) v
R
d
C
a
C
Filter
line
Figure 1. Thevenins equivalent circuit of an electrically short
dipole antenna with a shunt diode detector.
ELECTROMAGNETIC FIELD MEASUREMENT 1191
receiving dipole antenna as obtained from Eq. (1). The in-
duced voltage is measured across the center gap of the di-
pole antenna, which is oriented parallel to the electric eld
vector of the incident eld.
The RF voltage induced in the half-wave standard di-
pole antenna is rectied by a high-impedance Schottky
barrier diode connected in shunt across the center gap of
the antenna. The diode output is ltered by a balanced RC
network, and this DC voltage is measured with a high-
impedance DC voltmeter.
The effective length h
e
of a thin dipole antenna near
resonance and the required total length L for resonance
are given by [1]
h
e
l
p
tan
ph
l
_ _
12
and
L
l
2
1
0:2257
lnl=D 1
_ _
13
where D is the diameter of the standard dipole antenna.
To further increase the sensitivity of a standard anten-
na, a half-wave tuned dipole antenna with a narrowband
receiver should be used. In this case, the transfer function
S(f) is given by
Sf
V
L
f
E
inc
f

h
e
Z
a
Z
a
Z
r
14
where V
L
is the voltage across the receiver load, E
inc
is the
incident electric eld, h
e
is the effective length of the tuned
dipole, Z
a
is the antenna impedance, and Z
r
is the receiver
input impedance (B50 O). The input impedance of the an-
tenna Z
a
is complicated and is given in Ref. 1. When the
cable losses are significant, they should be included in Z
a
.
2. MAGNETIC FIELD SENSORS
2.1. An Electrically Small Loop Antenna with a
Loading Resistance
A magnetic eld sensor consists of an electrically small,
balanced loop antenna. The voltage V
i
induced in an elec-
trically small loop antenna by an electromagnetic wave
incident on the loop antenna is determined from Maxwells
equations and Stokes theorem, and is given by [1]
V
i
_
E
i
d jomH
i
NS 15
where E
i
is the tangential electric eld induced around the
loop antenna, l is the circumference of the loop antenna, o
is the angular frequency of E
i
, m is the permeability of the
loop antenna core, H
i
is the component of the magnetic
eld normal to the plane of the loop antenna, N is the
number of loop antenna turns, and S is the area of the loop
antenna. The induced voltage V
i
of an electrically small
loop antenna is proportional to frequency, the number of
loop turns, and the area of the loop antenna.
To make the response of a loop antenna at over the
frequency range of interest, the Q of the antenna has to
be reduced through a loading resistance. The resonance
of a loop antenna is the result of the combined effect of
the distributed capacitance of the loop antenna, the gap
capacitance, and the capacitance of the amplier along
with the inductance of the loop antenna. The equivalent
circuit for an electrically small loop antenna is shown in
Fig. 2. Here V
i
is the induced voltage, L is the loop in-
ductance, C is the capacitance, R is the loading resis-
tance, and V
o
is the voltage across the loading resistance.
Then the response of an electrically small loop antenna is
given by [1]
V
o
V
i
j
1
d
1
Q
j d
1
d
_ _ 16
where
Q
R
X
0
; X
0
o
0
L
1
o
0
C
; d
o
o
0
; o
0
LC
p 17
The inductance L and the capacitance C of a loop antenna
can be given by
Lmb ln
b
a
_ _
18
and
C
2eb
ln
b
a
_ _ 19
where m is the permeability of the medium, b is the loop
antenna radius, and a is the radius of the loop wire.
The transfer function of S(f) of an electrically small
loop antenna can be obtained by combining Eqs. (15)
and (16):
Sf
V
o
H
i
o
o
mNS
1
1
Q
j d
1
d
_ _ 20
V
i
R C
L
V
o
Figure 2. Thevenins equivalent circuit of an electrically small
loop antenna with a loading resistance.
The normalized transfer function S
n
(f) of a loop antenna
with a loading resistor is given by
S
n
f
1
1
Q
j d
1
d
_ _ 21
as a function of the normalized frequency d o/o
0
and is
given in Fig. 3 for various Qo1. Figure 3 shows that the
upper frequency end o
h
of the 3-dB rolloff point is given by
d
h
Q1 and, similarly, the corresponding low frequency o
l
of the 3-dB rolloff point is given by d
l
/Q1. Thus, from
these conditions, we obtain
d
h
d
l
o
h
o
l
o
2
0
1 22
or
o
0
o
h
o
l
p
23
The self-resonant frequency of a loop antenna is therefore
the geometric mean of the highest and lowest cutoff fre-
quencies.
3. AN ELECTROMAGNETIC FIELD SENSOR FOR
SIMULTANEOUS ELECTRIC AND MAGNETIC FIELD
MEASUREMENTS
The electric and magnetic eld sensors discussed above
measure either the electric or magnetic eld only and,
therefore, cannot measure complicated EM elds such as
those with reactive near-eld components and multipath
reections. For this reason, a single sensor capable of per-
forming simultaneous electric and magnetic eld mea-
surements was developed [5,6]. In this case, a loop
antenna is loaded at diametrically opposite points with
equal impedances. Across one load, the magnetic loop re-
sponse adds to the electric dipole response, whereas across
the other load, the magnetic loop response subtracts from
the electric dipole response. Thus, by taking the sum and
difference of currents across loads at diametrically oppo-
site points, the magnetic loop response and electric dipole
response can be separated. Thus, the sum current gives a
measure of the magnetic eld, whereas the difference cur-
rent gives a measure of the electric eld.
To explain the basic characteristics of a doubly loaded
loop antenna, the currents I
1
and I
2
at each load are given
Normalized frequency (c/c
0
)
Normalized frequency (c/c
0
)
Q = 0.001
Q = 0.01
Q = 0.05
Q = 0.1
Q = 0.5
Q = 1
Q = 0.005
10
4
80
10
4
10
3
10
2
10
1
10
0
10
1
10
2
10
3
60
40
20
0
M
a
g
n
i
t
u
d
e

o
f

t
r
a
n
s
f
e
r

f
u
n
c
t
i
o
n

(
d
B
)
90
45
45
90
0
10
4
10
3
10
2
10
1
10
0
Q = 0.01 Q = 0.1
Q = 0.5
Q = 1
10
1
10
2
10
3
10
4
P
h
a
s
e

o
f

t
r
a
n
s
f
e
r

f
u
n
c
t
i
o
n

(
d
e
g
r
e
e
s
)
Q = 0.05 Q = 0.005
Q = 0.001
Figure 3. The normalized transfer function of a
loop antenna as a function of normalized frequency
for different values of Q.
by [5]
I
1
2pbE
inc
f
0
Y
0
1 2Y
0
Z
L
f
1
Y
1
12Y
1
Z
L
_ _
24
and
I
2
2pbE
inc
f
0
Y
0
1 2Y
0
Z
L
f
1
Y
1
1 2Y
1
Z
L
_ _
25
where b is the radius of the loop, E
inc
is the incident elec-
tric eld, Z
L
is the load impedance, Y
0
is the admittance
for the magnetic loop response, and Y
1
is the admittance
for the electric dipole antenna response of a loop. In gen-
eral, Y
0
is much larger than Y
1
. f
0
and f
1
are Fourier co-
efcients of the incident wave. For a loop antenna
orientation of maximum electric and magnetic eld re-
sponse, f
0
jbb/2 and f
1
1
2
: Taking the sum and difference
of these currents yields [5]
I
S
1
2
I
1
I
2
2pbE
inc
f
0
Y
0
12Y
0
Z
L
26
and
I
D
1
2
I
1
I
2
2pbE
inc
f
1
Y
1
1 2Y
1
Z
L
27
This indicates that the sum current can be used to mea-
sure the magnetic eld and the difference current can be
used to measure the electric eld. In general, 2Y
0
Z
L
41 for
the magnetic eld loop antenna current. Therefore, when
the antenna is oriented for maximum response, I
S
can be
approximated as
I
S
j
E
inc
2Z
L
pb
2
b 28
This indicates that the magnetic loop current is approxi-
mately proportional to the product of frequency and the
area of the loop antenna, and is inversely proportional to
the load impedance. Similarly, for the electric eld dipole
current, assuming that 2Y
1
Z
L
51; we have
I
D
pbE
inc
Y
1
29
which is approximately proportional to the product of the
circumference of the loop antenna and frequency, since Y
1
has a capacitive susceptance (positive) and increases with
frequency. This device is intended to measure not only the
polarization ellipses of the electric and magnetic eld vec-
tors in the near eld region but also the time-dependent
Poynting vector and thus describe the energy ow.
4. PHOTONIC ELECTROMAGNETIC FIELD SENSORS
Properly designed photonic EM eld sensors provide the
wide bandwidth and low dispersion necessary to maintain
the delity of time-domain signals so that both amplitude
and phase information can be retrieved [7]. They are free
from electromagnetic interference, and there is minimal
perturbation of the eld being measured.
A number of photonic passive sensors for EM eld mea-
surements have been reported in the literature [7]. These
are systems in which the sensor head contains no active
electronics or power supplies. Optical measurement sys-
tems of the typical photonic EM eld sensors is shown
schematically in Fig. 4. Light from a laser is launched into
an optical ber link and serves as an optical signal carrier.
At the sensor head, the EM eld induces a voltage across
the modulator crystal and changes its index of refraction.
The crystal index changes occur at the frequency of the
impressed EM eld and result in a modulation in the am-
plitude of the optical carrier. At the receiver end of the
ber the light is converted to an electrical signal by a pho-
todiode and is suitably amplied for analysis with a spec-
trum analyzer, oscilloscope, or other signal processor. The
electro-optic interaction is weak, and, except for very high
elds, the gain of a small antenna is usually required to
obtain adequate modulation.
For the measurement of a pulsed electric eld, an an-
tenna with a at broadband response is most desirable. A
resistively loaded dipole antenna gives a nonresonant fre-
quency response and is, therefore, ideal for use in the
time-domain measurement of electromagnetic elds [3,4].
Electro-optical modulators that are driven by antenna
feeds are characterized as a function of the voltage ap-
plied to their electrodes. A convenient parameter for char-
acterizing a modulators performance is the voltage V
p
that is needed across the electrode to drive a modulator
between maximum and minimum optical power levels. In
general, a modulator transfer function is nonlinear. How-
ever, in most cases of interest, the eld-induced RF volt-
ages on the modulators electrodes are small compared to
V
p
, and a small-signal linear transfer function evaluated
at the operating bias point is appropriate. The most de-
sirable operating bias point is where the modulators sen-
sitivity and linear range are at maximum. The physical
characteristics of four modulators used in photonic EM
eld sensors are given in Table 1.
4.1. Pockels Cell
Pockels cell modulators often are used in EM eld sensors.
The small signal transfer function takes the form given in
Table 1, when the constant phase shift F
0
is set equal to
p/2 by introducing a quarter-wave retardation plate into
the cell. This biases the modulator at its point of maxi-
mum sensitivity and linearity. The characteristic voltage
of a bulk crystal modulator is limited by (1) the magnitude
of the electro-optic coefcients for available materials, (2)
the minimum dimensions of crystals that can be handled
as discrete optical elements, and (3) the maximum crystals
lengths acceptable for the highest desired operating fre-
quency. For operation to 5 GHz, the lowest obtainable val-
ues for V
p
are on the order of 100V for LiNbO
3
. It is
relatively easy to increase V
p
and reduce the modulators
sensitivity in order to measure high elds. In one appli-
cation of this technology for measuring elds from elec-
tromagnetic pulses, a crystal of Bi
4
Ge
3
O
12
(BGO), which
has cubic symmetry and reduced temperature sensitivity,
is used in a Pockels cell modulator and has a V
p
of
approximately 2100V.
4.2. MachZehnder Interferometer
Modulators based on MachZehnder (MZ) interferometers
and fabricated using optical guided wave (OGW) technol-
ogies have found widespread use in the beroptic commu-
nications industry. Their use as EM eld sensors also has
been investigated [7]. Photolithography is used to fabri-
cate OGW modulators by dening the waveguide channels
and then diffusing titanium into the surface of a LiNbO
3
crystal along these channels. The transfer function for an
MZ interferometer is identical to that of the Pockels cell.
The static phase shift F
0
is due to differences in the optical
path length in the interferometer arms. Significantly, V
p
for this device is typically two orders of magnitude smaller
than that for a Pockels cell modulator of the same material
and frequency response. The higher sensitivity is typical
of most OGW modulators and arises from the higher elds
created in the crystal with the closely spaced electrodes
that are achievable using photo lithography. The principal
problems to date with the MZ interferometer have been
the difculty in obtaining the correct value for F
0
during
fabrication, and its temperature and wavelength depen-
dence.
4.3. Directional Coupler
Directional couplers are also OGW devices that have been
investigated for use as EM eld sensors [7]. In the inter-
action region, the two waveguides lie close enough togeth-
er that the evanescent eld of the lightwave in one guide
couples into the other guide. In such a coupled-mode sys-
tem, the energy in one lightguide can be switched to the
other guide by changing the relative propagation con-
stants in the channels with the electro-optic effect. If the
fabrication parameters are chosen correctly, the transfer
function given in the literature [7] for the directional cou-
pler can be written in the form given in Table 1.
5. ELECTROMAGNETIC FIELD STANDARDS
5.1. Transverse Electromagnetic Cell
Transverse electromagnetic (TEM) transmission-line cells
are devices used for establishing standard EM elds in a
Table 1. Transfer Functions for Electro Optic Modulators
Modulator General Small Signal
Pockels cell cos
2
p
2
V
n
f
0
2
_ _
p
2V
p
MachZehnder cos
2
p
2
V
n
f
0
2
_ _
p
2V
p
4-port coupler
1
1 3V
2
n
sin
2
p
2
1 3V
2
n
_
_ _
%
1:6
V
p
3-port coupler
1
2
2V
n
14V
2
n
sin
p
2
2
p
14V
2
n
_
_ _
2
V
p
sin
p
2
2
p
1:79
V
p
Light beam
Polarizer
Pockels cell
Electrode
MachZehnder interferometer
4Port coupler 3Port coupler
Lithium niobate
substrate
P
2
P
in
P
1
0.6
0.6
P
2
P
in
P
1
Quarter-wave plate
Electro-optic crystal
P
in
P
out
Single mode
channel
waveguide
Analyzer
40 m
24 m gap
Resistively loaded dipole antenna
(a) (b)
(c) (d)
Figure 4. Measurement system of a photonic
EM eld sensor.
shielded environment [8]. Their application is becoming
increasingly widespread because of their versatility, mea-
surement accuracy, and ease of operation.
A TEM cell is essentially a 50-O triplate transmission
line with the sides closed in, to prevent radiation of RF
energy into the environment and to provide electrical iso-
lation. A properly designed cell, terminated in its charac-
teristic impedance, is capable of producing a calculable
electric and magnetic eld for calibrating an electrically
small antenna or RF sensor. The cell consists of a section
of rectangular coaxial transmission lines tapered at each
end to adapt to standard coaxial connectors. The line and
tapered transitions are designed to have a nominal char-
acteristic impedance of 50 O along their length, to ensure
minimum-voltage standing-wave ratio. A fairly uniform
EM eld is established between the plates inside the cell
when RF energy is conducted in the line from a transmit-
ter connected to the cells input port. A 50-O termination is
connected to the cells output port. The expression for
determining the electric eld E in the cell is given by
E
V
b

PZ
0
p
b
30
where V is the RMS voltage on the septum (center con-
ductor), b is the separation distance between the septum
and lower or upper walls, P is the net power ow to the
cell, and Z
0
is the real part of the cells characteristic
impedance D50 O.
Awave traveling through a cell has essentially the free-
space impedance (D120p O), thus providing a close ap-
proximation to a far-eld plane wave propagating in free
space. The design of TEM cells can be based on an ap-
proximate equation for the characteristic impedance of a
rectangular transmission line [8]
Z
0

377
4
p
q
2
p
ln sinh
pg
2q
_ _ _ _
Dc
e
0
_ _
1
31
where p, q, and g are as shown in Fig. 5 and Dc/e
0
is related
to the fringing capacitance between the edges of the sep-
tum and the sidewalls. For large gaps (g/p40.2), this
fringing term approaches zero [8].
The upper useful frequency for a cell is limited by dis-
tortion in the test eld caused by multimoding and reso-
nances that occur within the cell at frequencies above the
cells multimode cutoff. Resonant frequencies associated
with these modes can be found from the expression [8]
F
res
f
2
mn
cl
2L
_ _
2
32
where f
mn
are the frequencies of the higher-order mode(s)
excited inside the cell, c is the wave propagation velocity
(D3.010
8
m/s), L is the resonant length of the cell in
meters, and l, m, and n are integers corresponding to mul-
tiples of the resonant length and the particular waveguide
mode. The inuence of the rst-order TE modes does not
become significant until approaching their resonances.
Since most cells are designed with the center plate (sep-
tum) centered symmetrically, the odd-order TE modes are
not excited in an empty cell. The presence of a device
placed in the cell will, however, excite these modes in vary-
ing degrees, depending on its size, shape, and placement.
5.2. Waveguide Chamber
For the frequency range of 3001000 MHz, one can use a
waveguide section with a rectangular cross section with a
width-to-height (aspect) ratio of 2:1. The length of a guide
cell must exceed two wavelengths over the specied fre-
quency band, in order to create a fairly uniform eld with-
in the guide. Electromagnetic power is transmitted
through the guide to a matched resistive load, and the
maximum frequency is limited by the requirement that
power propagates in the guide in the dominant TE
10
mode.
In this well-known case, the direction of the electric eld
vector is across the narrow face of the guide.
Assuming good conductivity of the waveguide walls, an
air dielectric, and sinusoidal excitation, the lowest cutoff
frequency f
co
is
f
co
c
2a
33
where c is the wave propagation velocity (D3.0 10
8
m/s)
and a is the guide width.
The longest or cutoff wavelength is given by l
co
2a.
The wavelength l
wg
inside the guide for these operating
conditions is
1
l
wg
1
l
0
_ _
2
1
l
co
_ _
2
34
where l
0
is the free-space wavelength.
The transverse impedance Z
w
of the wave traveling in
the guide is
Z
w
E
y
H
z
z
0
l
wg
l
0
_ _
z
0
1 l
0
=2a
2
_ 35
where z
0
is the intrinsic impedance of free space (D120p).
The direction of the electric eld vector is across the
narrow face of the guide, and its RMS magnitude at the
2p
2w
g g
2q
Figure 5. Cross-sectional view of a TEM cell.
center of a rectangular waveguide is given by
E
2Z
w
P
z
ab
_
36
and, similarly, the RMS magnitude of the magnetic eld in
the guide center is given by
H
2P
z
Z
w
ab
37
where P
z
is the total power ow in the guide and ab is the
cross-sectional area of the waveguide ( 0.5 a
2
).
5.3. Microwave Anechoic Chambers
Microwave anechoic chambers currently are used for a
variety of indoor antenna measurements, electromagnetic
eld measurements, electromagnetic interference (EMI)
measurements, and electromagnetic compatibility (EMC)
measurements. The primary requirement is that a trans-
mitting antenna at one location within a chamber or at a
chamber wall generates a known eld throughout a vol-
ume of the chamber, which has dimensions sufcient to
perform EM eld measurements. This volume is frequent-
ly called a quiet zone, and the level of reected waves
within it will determine the performance of the anechoic
chamber.
Electromagnetic eld measurements in an anechoic
chamber usually are performed in the near-eld region
of a transmitting standard antenna. To establish the stan-
dard eld, the radiated eld intensity in the near-eld
region of the transmitting antenna is calculated. The
antennas typically used for the anechoic chamber mea-
surements consist of a series of open-ended waveguides at
frequencies below 450 MHz, and a series of rectangular
pyramidal horn antennas at frequencies above 450 MHz.
The electric eld strength at a specific distance from
the radiating antenna is calculated from measurements of
the power delivered to the transmitting antenna and a
knowledge of the gain of the antenna as a function of fre-
quency and distance to the eld point. The equation used
to calculate the electric eld E on the boresight axis of the
transmitting antenna is
E
1
d
z
0
P
net
G
4p
_

30P
net
G
p
d
38
where P
net
is the net power delivered to the transmitting
antenna, z
0
is the free-space impedance (D120p O), G is
the near-eld gain of the transmitting antenna at the giv-
en frequency and distance, and d is the distance from the
center of the aperture of the transmitting antenna (horn
or open-ended waveguide) to the on-axis eld point.
The net power P
net
delivered to the transmitting an-
tenna is the difference between the incident P
inc
and re-
ected P
re
powers as measured with a calibrated
directional coupler (four ports) with calibrated power me-
ters. In order for measurements to be accurate, an an-
echoic chamber must provide a truly free-space test
environment. The performance of a rectangular RF an-
echoic chamber can be determined by measuring the rel-
ative insertion loss versus separation distance between a
source antenna and a receiving antenna.
Antenna insertion loss is the ratio of power received by
a receiving antenna or probe to the power accepted by the
transmitting antenna. If the anechoic chamber is a perfect
free-space simulator, the relative insertion loss between
two polarization-matched antennas will vary with dis-
tance according to the Friis transmission formula [9]
P
r
P
t
G
r
G
t
l
4pd
_ _
2
39
where P
t
is the net power delivered to the transmitting
antenna, P
r
is the power received by the receiving anten-
na, G
t
is the near-eld gain of the transmitting antenna,
G
r
is the near-eld gain of the receiving antenna, d is the
separation distance between the two antennas, and l is
the wavelength. Experimental data can be compared with
the calculated free-space transmission loss, using appro-
priate near-eld transmitting antenna gains. The differ-
ence between the measured and calculated transmission
loss is a measure of reections from chamber surfaces.
5.4. Open-Area Test Sites
An open-area test site typically is used for the antenna
calibration in the frequency range of 10 kHz1000 MHz. A
calibration consists of determining the antenna factor that
permits a receiver (RF voltmeter) to be used with the cal-
ibrated antenna to conduct measurements of eld
strength.
At frequencies below about 50 MHz, loop antennas are
calibrated in a quasi-static, near-zone, magnetic eld pro-
duced by a balanced single-turn transmitting loop with a
10 cm radius. Above 25 MHz, dipole antennas are calibrat-
ed in a far-zone electric eld, which is evaluated in terms
of the open-circuit voltage induced in a self-resonant re-
ceiving dipole antenna. Between 30 kHz and 300MHz,
vertical monopole antennas and small probes are evalu-
ated in an elliptically polarized electromagnetic eld pro-
duced by a transmitting monopole antenna above a
conducting ground screen.
Field strength can be evaluated using two independent
techniques: (1) the standard eld method and (2) the
standard antenna method. For the standard eld meth-
od, a transmitted eld is calculated in terms of the type
and dimensions of a transmitting antenna, its current
distribution or net delivered power, the frequency of the
transmitted signal, the distance from the transmitting
antenna to the eld point, and the effect of ground reec-
tions (if present). For the standard antenna method, an
unknown eld is measured with a calculable receiving
antenna. The voltage or current induced in a standard
antenna by the component of eld being evaluated is
measured. The eld strength is then calculated in terms
of this induced voltage, the dimensions and form of the
receiving antenna, and its orientation with respect to the
eld vector.
All the techniques described above for eld strength
standards are applicable only to steady-state RF elds
with sinusoidal time variation. They are not intended for
use with pulsed elds or other broadband applications.
5.5. Magnetic Field Strength Standards for Loop Antennas
at 10 kHz50 MHz (Standard Field Method)
The response of an electrically small receiving loop anten-
na is proportional to the average normal component of
magnetic eld strength incident on the antenna. A calcu-
lable quasi-static magnetic eld can be produced to cali-
brate these antennas using a circular single-turn balanced
transmitting loop. Up to 30MHz, the current in a loop
with a 10 cm radius is approximately constant in ampli-
tude and phase around the loop. The receiving loop an-
tenna being calibrated is positioned on the same axis as
the transmitting loop at a distance of 1.53 m. The normal
component of the magnetic eld, averaged over the area of
the receiving loop, is given by [10]
H
bIr
1
r
2
1
m0
1
2m1!
1 3 2m1
2 4 2m2
br
1
r
2
R
0
_ _
m1
h
2
2m1
bR
0
40
where
H RMS value of magnetic eld
I RMS current in transmitting loop
r
1
radius of transmitting loop
r
2
radius of receiving loop
R
0

d
2
r
2
1
r
2
2
_
d axial distance between the two loops
b 2p/l
0
l
0
free-space wavelength
h
2
n
nth-order spherical Hankel function of second kind
The current in the transmitting loop antenna is measured
with a vacuum thermocouple calibrated with direct cur-
rent. The thermocouple is at the top of the loop winding.
While coaxial loop antennas normally are used for cal-
ibration purposes, the two-loop antennas also can be po-
sitioned in the same plane. Coplanar loop antennas are
advantageous under certain conditions (e.g., with some
ferrite core antennas in which the core length is large). In
the coplanar loop antenna setup, the calibrating value of
H would be half of that given by Eq. (41).
The calibration and subsequent measurement of mag-
netic eld strength H often are expressed in terms of the
electric eld E that would exist if the measurement were
made in free space, in which case E/HD120p O. When
such a eld strength meter is used to make measurements
near the ground, the indicated value of the electric eld is
not necessarily valid. The same is true for measurements
made in the near zone of a transmitting antenna. Howev-
er, the value of the magnetic component H can still be
measured correctly.
For calibrating loop antennas or magnetic eld sensors
at a higher eld, it is possible to use the calculable mag-
netic eld generated in a TEM cell, or a waveguide cham-
ber, or at the center of a at multiturn coil, or at the
midpoint of a Helmholtz coil pair.
5.6. Electric Field Strength Standards for Dipole Antennas
from 25 to 1000 MHz (Standard Antenna Method)
The magnitude of the electric eld component at a given
point in a locally generated eld is determined from the
open-circuit voltage V
oc
induced in a standard (calculable)
half-wave receiving dipole antenna. The induced voltage
is measured across the center gap of the dipole antenna,
which is oriented parallel to the electric eld vector of the
incident eld. In using the standard antenna method, a
plane-wave eld can be generated by a suitable transmit-
ting antenna, such as a log-periodic or half-wave dipole
antenna. The magnitude of this incident eld is measured
with the standard dipole antenna by the relation
E
inc
V
oc
L
eff
41
where E
inc
is the eld strength of the locally generated
eld, V
oc
is the open-circuit voltage induced in the stan-
dard dipole antenna, and L
eff
is the effective length of the
standard dipole antenna.
The RF voltage V
oc
picked up by the l/2 standard dipole
is detected by a high-impedance Schottky barrier diode
connected in shunt across the center gap of the antenna.
The diode output is ltered by a balanced RC network, and
this DC voltage is measured with a high-impedance DC
voltmeter. The RF-to-DC-characteristic of the dipole an-
tenna and its lter circuit is obtained experimentally. As-
suming a cosinusoidal current distribution on an
innitesimally thin dipole, the effective length of a half-
wave dipole antenna in free space is given by Eq. (12).
5.7. Electric Field Strength Standards for Vertical Monopole
Antennas from 30 kHz to 300MHz (Standard Field Method)
Several approaches were considered for generating a stan-
dard (calculable) eld to calibrate vertically polarized an-
tennas. The system chosen for this measurement consists
of a thin cylindrical transmitting monopole antenna over a
metallic ground plane. The eld strength is calculated in
terms of the magnitude and distribution of the monopole
antenna current, and other factors such as (1) monopole
height, (2) horizontal distance from the transmitting an-
tenna to the eld point, (3) vertical height of this point
above the ground plane, and (4) electrical conductivity of
the ground plane.
The height of the transmitting monopole antenna is
adjustable, with a maximum height of about 3m. The
electrical height of this antenna is l/4 (resonant) at
25 MHz, but only 0.0003 l at 30 kHz. At frequencies above
25 MHz, the antenna height is reduced to a l/4 value. The
base diameter of the monopole antenna is about 1 cm. The
monopole antenna is excited through a coaxial cable from
a transmitting room located beneath a concrete ground
slab that is covered by a conducting metal screen to form
the electrical ground plane.
Equations (42), (43), and (44) give the magnitudes of
the three eld components E
z
, E
r
, and H
f
, respectively, of
a transmitting l/4 monopole antenna above a perfect
ground plane of innite extent:
E
z
30I
0
e
jbr
1
r
1
e
jbr
2
r
2
_ _
42
E
r
30I
0
r
0
e
jbr
1
r
1
_ _
z
l
4
_ _
e
jbr
2
r
2
_ _
z
l
4
_ _ _ _
43
H
f
I
0
4pr
0
e
jbr
1
e
jbr
2
44
where
E
z
vertical electric eld component
E
r
horizontal electric eld component
H
f
magnetic eld, encircling the monopole antenna
I
0
RMS base current of the monopole antenna
b 2p/l the wavelength constant
r
1
[d
2
(z l)
2
]
1/2
r
2
[d
2
(z l)
2
]
1/2
r
0
[d
2
z
2
]
1/2
l monopole antenna length
d horizontal distance between the monopole antenna
and the eld point
z vertical distance from the ground plane to the eld
point
For frequencies near self-resonance, the monopole an-
tenna base current is measured with an RF ammeter con-
sisting of a thermoconverter that has been calibrated with
known values of DC current. At lower frequencies, where
the monopole antenna input impedance Z
in
is a high-ca-
pacitance reactance, the base current is calculated from
Ohms law in terms of the base voltage measured with a
high-input impedance voltmeter and the theoretical input
impedance. At very low frequencies, Z
in
may be calculated
from the antenna capacitive reactance (11).
Z
in
1
joC
a
45
where
C
a
5:56 10
11
h
lnh=a 1
46
where C
a
is the monopole antenna input capacitance (F), h
is the monopole antenna height (m), and a is the monopole
antenna radius (m).
The standard eld equations are relatively simple for a
ground plane with innite extent and innite conductiv-
ity. In addition, the current on a vertical monopole anten-
na with nite diameter departs from the sinusoidal
current distribution of a lamentary monopole antenna.
This does not seriously affect the calculated values of
current-related eld components, such as the magnetic
eld or the far-zone electric eld. However, the low-fre-
quency near-zone quasi-static electric eld components
are more nearly charge-related and are given by the spa-
tial derivative of the current distribution. Hence, there is
greater uncertainty in calculating the electric eld com-
ponents at frequencies well below that of a l/4-resonant
monopole antenna.
If a transmitting monopole antenna is electrically
shortthat is, if the height is less than l/4 and the fre-
quency is below resonancethe current distribution is
triangular. The eld equations are a little more compli-
cated; only the vertical electric eld is given below [11]:
E
z
j30I
0
sinbl
e
jbr
1
r
1
e
jbr
2
r
2
2 cosble
jbr
0
_ _
47
The EM eld values in the half-space above a perfect
ground are the same as those in each half-volume of a
center-fed l/2 dipole antenna in free space. The input im-
pedance of a monopole antenna above perfect ground is
half that of a dipole antenna in free space. The power re-
quired to generate a given eld strength is half that re-
quired for a dipole antenna, but the radiated power goes
into half the volume, so the eld is the same. Measure-
ments of Z
in
with a commercial impedance meter are per-
formed to check the theoretical values from 0.5 to 50 MHz.
Measurements of the monopole antenna capacitance can
be made at lower frequencies with a commercial Q meter.
6. FUTURE DIRECTIONS
Established techniques for EM eld measurements will
probably be extended to higher frequencies; however, such
work would not involve fundamental changes in the in-
strumentation or measurement strategy. The measure-
ment methods described earlier are suitable only for (1)
measuring plane-wave sinusoidal elds of a given fre-
quency and (2) calibrating the devices that measure
such elds.
The challenge of the future resides in the development
of standards for measurement of elds that are nonsinu-
soidal and/or nonplanar. The fundamental requirement
for EM eld sensors is that their outputs provide ampli-
tude and phase information simultaneously over a broad
spectrum for nonsinusoidal elds. For elds containing
more than one frequency component, it is clear that such
simultaneous measurement is necessary. Similarly, if
phase information is preserved, then measurements of
single-frequency nonplanar elds can be made in terms of
true energy density or in terms of the Poynting vector.
The standard techniques of today rely on very high-re-
sistance transmission lines to convey DC voltage and cur-
rent to regions external to the eld under measurement.
Future standards of measurement employ optically
sensed EM eld sensors whose optical sensing signal
paths will not perturb the EM eld under measurement,
but will also convey eld amplitude and phase information
to a region external to the eld for measurement.
Optically sensed sensors are being built today. However,
additional work is needed to transform these experimen-
tal devices into stable elements for standards applications.
Also, standard elds need to be developed, in order to cal-
ibrate these probes of the future.
BIBLIOGRAPHY
1. M. Kanda, Standard probes for electromagnetic eld mea-
surement, IEEE Trans. Anten. Propag. AP-41:13491363
(1993).
2. M. Kanda, Analytical and numerical techniques for analyzing
an electrically short dipole with a nonlinear load, IEEETrans.
Anten. Propag. AP-28:7178 (1980).
3. M. Kanda, A relatively short cylindrical broadband antenna
with tapered resistive loading for picosecond pulse measure-
ments, IEEE Trans. Anten. Propag. AP-26:439447 (1978).
4. M. Kanda and L. D. Driver, An isotropic electric-eld probe
with tapered resistive dipoles for broadband use, 100kHz to
18GHz, IEEE Trans. Microwave Theory Tech. MTT-35:124
130 (1987).
5. M. Kanda, An electromagnetic near-eld sensor for simulta-
neous electric and magnetic eld measurements, IEEE Trans.
Electromagn. Compat. EMC-26:102110 (1984).
6. L. D. Driver and M. Kanda, An optically linked electric and
magnetic eld sensor for Poynting vector measurements in
the near elds of a radiating source, IEEE Trans. Elect-
romagn. Compat. 30:495503 (1988).
7. M. Kanda and K. D. Masterson, Optically sensed EM-eld
probes for pulsed elds, Proc. IEEE 80:209215 (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 near-zone magnetic eld of a small circu-
lar-loop 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, Prentice-Hall, 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 elds 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 eld is articially applied to illuminate
the inaccessible unknown scatterers to give rise to scat-
tered elds, which is measured at some accessible area.
There are some cases, however, such as passive remote
sensing, in which the scattered electromagnetic elds
from the unknown scatterers are not due to such arti-
cially applied incident electromagnetic elds. Usually, the
scattered electromagnetic elds 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 three-dimensional map of
the world around us. This is a highly automated process,
and most of us do not stop to reect on how difcult 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 lters 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 elds is provided, and
the scattered eld 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 eld are
identical, at least in principle, to those used in other, com-
pletely different elds. These interdisciplinary applica-
tions of the inversion techniques are drawing increasing
attention. Hence, although we focus on electromagnetic
inverse problems here, other elds 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
Next Page

05a

Transféré par

Informations du document

Description originale:

Titre original

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

05a

Transféré par

Droits d'auteur :

Formats disponibles

ELECTROMAGNETIC-BANDGAP-ASSISTED

cos y=2 j sin y=2

0.6 mm. The passband design frequency was 7.5 GHz.

Figure 5. Spectral-domain coordinate system.

Figure 8. Cylindrical coordinates for off-axis source points.

ber die Ausbreitung der Wellen in der

Vous aimerez peut-être aussi