C H A P T E R 3

Implementation of the Spectral-Domain

Moment-Method Formulation
3.1 INTRODUCTORY REMARKS
This chapter presents a detailed exposition of the theoretical formulation that was implemented
for the rigorous analysis of microstrip patch antennas with capacitive feed probes. The basic
antenna geometries under consideration are shown in Figure 3.1. In its general form, it consists
of one or more rectangular (or square) resonant patch elements that are capacitively coupled to
smaller probe-fed capacitor patches. The capacitor patches can either be circular or rectangular
in shape. The patches do not necessary have to be aligned, but can be orientated arbitrarily. All
of the patches and probes reside within a grounded, planarly multilayered substrate. It is assumed
that the patches, probes and ground plane are all perfect electric conductors (PECs) and that the
multilayered substrate is isotropic, but possibly lossy. The analysis excludes the feed network as
it can usually be analysed separately for probe-fed microstrip antennas.
The present analysis is based on a Greens function/moment method (MM) approach, where the
electric-eld integral-equation (EFIE), representing the boundary condition that the total tangen-
tial electric eld must vanish on the PEC parts of the structure, is discretised by the Galerkin
method. The unknown coefcients are evaluated by solving the resulting matrix equation. Al-
though probe-fed patch antennas have enjoyed extensive theoretical analysis, the present analysis
relies on a unique combination of entire-domain and subdomain basis functions and therefore
introduces unique numerical requirements.
Section 3.2 presents a general overview of the MM formulation that was implemented, while
33
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
Resonant patch
Capacitor patch
Probe
Substrate
Ground plane
Resonant patch
Capacitor patch
Probe
Substrate
Ground plane
Figure 3.1 Basic antenna geometries.
University of PretoriaElectrical, Electronic and Computer Engineering 34
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
Section 3.3 shows how the MM formulation can be cast into its spectral-domain form. The
multilayered-media Greens function and basis functions are addressed in Sections 3.4 and 3.5 re-
spectively. Section 3.6 deals with attachment modes, which are special basis functions required to
model the singular and rapidly varying nature of the surface current density in the vicinity of each
probe-to-patch junction. Section 3.8 deals with issues surrounding the evaluation of interaction-
matrix elements, while Section 3.9 describes different integration strategies that have been used
to evaluate these matrix elements. One of the difculties associated with the spectral-domain
MM (SDMM) is the highly oscillating nature of integrands for basis and testing functions that are
widely separated. In Section 3.9, a recently-published method is extended to handle both situa-
tions where the basis and testing functions are closely spaced or widely separated. When using
the MM, the interactions between all basis and testing functions have to be calculated. However,
depending on the implementation, there are often identical interactions that have to be calculated
repeatedly. On a rectangular grid, these duplicate entries can easily be identied and eliminated,
but it becomes much more difcult with a mixture of lower-order and higher-order basis functions
that are arbitrarily orientated. Special algorithms have been developed to deal with such a mixture
of basis and testing functions, and are addressed in Section 3.10. Section 3.11 deals with the eval-
uation of the excitation-vector elements, while Section 3.12 addresses the solution of the matrix
equation to yield the unknown current-density coefcients. The evaluation of the various network
parameters for an array of antenna elements, is addressed in Section 3.13, while Section 3.14 deals
with the evaluation of the far elds.
3.2 GENERAL FORMULATION
The general problem to be solved here, is shown in Figure 3.2(a). It consists of a PEC structure
that is embedded within a grounded multilayered medium. For purposes of generalisation and il-
lustration, the PEC structure is represented by an arbitrary shape, but in practice it would represent
the patches and probes of the microstrip antenna. The excitation is due to impressed electric and
magnetic current sources, J
imp
and M
imp
, that radiate known incident electric and magnetic elds,
E
inc
and H
inc
, in the presence of the grounded multilayered medium, but in the absence of the
PEC structure.
1
These elds, that impinge upon the surface of the PEC structure, induce on it an
electric current density J, which in turn radiates the scattered elds E
scat
and H
scat
. Therefore, at
every point r, the total electric and magnetic elds, E
tot
and H
tot
, existing in the presence of the
patches, probes and grounded multilayered medium, can be expressed as
E
tot
(r) = E
inc
(r) +E
scat
(r) (3.1)
and
H
tot
(r) = H
inc
(r) +H
scat
(r). (3.2)
1
In the case of probe-fed microstrip antennas, H
inc
can be thought of as the equivalent magnetic current density in the
aperture, where the probe passes through the ground plane.
University of PretoriaElectrical, Electronic and Computer Engineering 35
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
Now, the EFIE is based on the fact that, on the surface of the PEC structure, the total tangential
electric eld must vanish. This can be expressed as
E
tot
tan
(r) = n E
tot
(r)
= n E
inc
(r) + n E
scat
(r) (3.3)
= E
inc
tan
(r) +E
scat
tan
(r) = 0, (3.4)
where n is the normal vector at position r on the surface of the PEC structure. It then follows that
E
scat
tan
(r) = E
inc
tan
(r), (3.5)
where E
inc
tan
represents the incident electric eld, tangential to the structure at r, and E
scat
tan
represents
the scattered electric eld, also tangential to the structure at r.
The surface equivalence theorem [73] can be used to replace the PEC structure with an equivalent
electric current density J, as shown in Figure 3.2(b). The representation in Figure 3.2(b) is also
known as the physical equivalent of the problem in Figure 3.2(a). In practice, the patches are
normally assumed to be innitely thin and therefore J will be the vector sum of the true current
densities on the top and bottom surfaces of the patches. The scattered elds are radiated by J in
the presence of the grounded multilayered medium (but not the embedded PEC structure) and can
be expressed as
E
scat
tan
(r) =
__
S

G(r|r

) J(r

) ds

, (3.6)
where

Gis the dyadic Greens function for a grounded multilayered medium, relating the electric
eld E
scat
tan
at the observation position r to the surface current density J at the source position r

.
In (3.6), S represents the surface area over which J extends and is also known as the support of J.
The EFIE in (3.5) can now be written as
__
S

G(r|r

) J(r

) ds

= E
inc
tan
(r), (3.7)
which is a Fredholm integral equation of the rst kind. In (3.7),

G is the kernel of the equation,
E
inc
tan
is the known excitation and J is the unknown response to be determined.
The rst step in the solution of (3.7), via the MM, is to approximate the unknown current den-
sity J on the patches and probes by a set of N vector basis functions with unknown amplitude
coefcients. It can be expanded as
J(r

) =
N

n=1
I
n
f
n
(r

), (3.8)
where f refers to the vector basis functions (also known as expansion functions) and I refers to
University of PretoriaElectrical, Electronic and Computer Engineering 36
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
n
S
E
tot
(r) = E
inc
(r) +E
scat
(r)
E
tot
tan
(r) = 0
PEC
Medium 0
Medium 1
PEC ground plane
J
imp
M
imp
(a)
n
S
E
tot
(r) = E
inc
(r) +E
scat
(r)
E
tot
tan
(r) = 0
J(r

)
Medium 0
Medium 1
PEC ground plane
J
imp
M
imp
(b)
n
S
E
tot
(r) = E
inc
(r) +E
scat
(r)
E
tot
tan
(r) = 0
J(r

)
Medium 0
Medium 1
PEC ground plane
f
n
(r

)
J
imp
M
imp
(c)
Figure 3.2 Development of the EFIE. (a) Actual problem. (b) Physical equivalent.
(c) Basis functions.
University of PretoriaElectrical, Electronic and Computer Engineering 37
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
the unknown amplitude coefcients (which are generally complex numbers). This is illustrated in
Figure 3.2(c).
Exploiting the linearity of the operator on the left-hand side of (3.7), leads to
N

n=1
I
n
__
S

G(r|r

) f
n
(r

) ds

= E
inc
tan
(r), (3.9)
which is one equation with N unknowns. It alone is not sufcient to determine the N unknown
amplitude coefcients. To resolve the N amplitude coefcients, it is necessary to have N linearly
independent equations. This can be achieved by dening a symmetric product
a, b =
__
S
a
a bds, (3.10)
where a is used to test b over the domain S
a
. In (3.10), a is a vector function, also known
as the testing function (or weighting function) with associated support S
a
. By using Galerkins
method [4, 73, 74], both basis and testing functions are chosen to be the same set of functions.
Therefore, the same set of vector basis functions in (3.8) are also used to test (3.9), yielding a set
of simultaneous equations that can be solved by standard linear algebra techniques. The boundary
condition in (3.9) then becomes
__
S
m
f
m
(r)
N

n=1
I
n
__
S
n

G(r|r

) f
n
(r

) ds

ds =
__
S
m
f
m
(r) E
inc
tan
(r) ds, m = 1, 2, . . . , N
(3.11)
or
N

n=1
I
n

f
m
, E
scat
tan
(f
n
)
_
=

f
m
, E
inc
tan
_
, m = 1, 2, . . . , N. (3.12)
The resulting system of linear equations can also be expressed in matrix notation as
[Z
m,n
] {I
n
} = {V
m
}, (3.13)
in which [Z
m,n
] is the N N-dimensional interaction matrix with elements
Z
m,n
=

f
m
, E
scat
tan
(f
n
)
_
(3.14)
and {V
m
} is the N-dimensional excitation vector with elements
V
m
=

f
m
, E
inc
tan
_
. (3.15)
In (3.13), {I
n
} is the N-dimensional vector that contains the unknown amplitude coefcients of
University of PretoriaElectrical, Electronic and Computer Engineering 38
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
(3.8). The problem can be solved as
{I
n
} = [Z
m,n
]
1
{V
m
} (3.16)
through any of the standard numerical techniques. Once the unknown amplitude coefcients have
been resolved, the electric current density J is essentially known and all other quantities of interest
can then also be calculated.
In the next section, it will now be shown how the interaction-matrix elements can be calculated in
the spectral domain.
3.3 SPECTRAL-DOMAIN FORMULATION
It is often more convenient to calculate the elements of the interaction matrix in the spectral domain
rather than in the spatial domain. This is due to the fact that, in the spectral domain, the dyadic
Greens function for a grounded multilayered medium can be derived in closed form. Usually it
requires that the Fourier transforms of the basis and testing functions should also be available in
closed form, thereby limiting the choice of basis and testing functions somewhat. However, the
spectral-domain formulation can be very efcient if these expressions are available.
In the spatial domain, the elements of the interaction matrix can be expressed as
Z
m,n
=
__
S
m
__
S
n
f
m
(r)

G(r|r

) f
n
(r

) ds

ds
=
_
z
_
y
_
x
f
m
(x, y, z)
_
z

_
y

_
x

G(x, y, z|x

, y

, z

) f
n
(x

, y

, z

) dx

dy

dz

dxdydz,
(3.17)
where all the variables are shown in Figure 3.3, which in turn depicts the general coordinate
system for the theoretical analysis that follows. A two-dimensional Fourier-transform pair can
now be dened as

f(k
x
, k
y
) = F
_
f(x, y)
_
=
_

f(x, y) e
jk
x
x
e
jk
y
y
dxdy (3.18)
f(x, y) = F
1
_

f(k
x
, k
y
)
_
=
1
4
2
_

f(k
x
, k
y
) e
jk
x
x
e
jk
y
y
dk
x
dk
y
(3.19)
in order to transform between a general function f(x, y) and its spectral counterpart

f(k
x
, k
y
). By
using the denition in (3.19), the dyadic Greens function

Gin (3.17) can be related to its spectral
domain counterpart

Gthrough

G(x, y, z|x

, y

, z

) =
1
4
2
_

G(k
x
, k
y
, z|z

) e
jk
x
(xx

)
e
jk
y
(yy

)
dk
x
dk
y
. (3.20)
University of PretoriaElectrical, Electronic and Computer Engineering 39
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
x y
z
r

r
r r

x, y, z at
observation
point
x

, y

, z

at
source point
on S
Figure 3.3 Coordinate system for the theoretical analysis.
This allows (3.17) to be written, after some reorganisation of the terms and operators, as
Z
m,n
=
1
4
2
_
k
y
_
k
x
_
z
_
z

__
y
_
x
f
m
(x, y, z) e
jk
x
x
e
jk
y
y
dxdy
_

G(k
x
, k
y
, z|z

)
__
y

_
x

f
n
(x

, y

, z

) e
jk
x
x

e
jk
y
y

dx

dy

_
dz

dz dk
x
dk
y
. (3.21)
The expressions between the two pairs of square brackets in (3.21) can be recognised as two-
dimensional Fourier transforms, similar to the denition in (3.18). The expression in (3.21) can
therefore nally be simplied to
Z
m,n
=
1
4
2
_
k
y
_
k
x
_
z
_
z

f
m
(k
x
, k
y
, z)

G(k
x
, k
y
, z|z

f
n
(k
x
, k
y
, z

) dz

dz dk
x
dk
y
.
(3.22)
As is evident from this discussion, the spectral-domain forms of both the dyadic Greens function
and the basis functions (per implication, also the testing functions) need to be found. These will
now be addressed in the sections that follow.
3.4 GREENS FUNCTION
Consider the grounded, planarly multilayered medium as shown in Figure 3.4. It is assumed that
there are T + 1 layers and that they are numbered from 0 to T, starting with 0 at the top layer and
ending with T at the bottom layer. The top layer (layer 0), which is normally assumed to be free
space, extends to innity along the z direction, while the bottom layer (layer T) is bounded by a
PEC ground plane right below it. All the layers are also assumed to have innite dimensions along
the x and y directions. The interface between layers t and t + 1 is at z = d
(t)
, while the thickness
of layer t is h
(t)
= d
(t1)
d
(t)
.
University of PretoriaElectrical, Electronic and Computer Engineering 40
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation

(
0
)
,
(
0
)
(
1
)
,
(
1
)
(
2
)
,
(
2
)
(
T

1
)
,
(
T

1
)
(
T
)
,
(
T
)
PEC ground plane
h
(1)
h
(2)
h
(T1)
h
(T)
z = d
(0)
z = d
(1)
z = d
(2)
z = d
(T2)
z = d
(T1)
z = d
(T)
x
y
z
Figure 3.4 Grounded, planarly multilayered medium.
The complex permittivity
(t)
and complex permeability
(t)
of layer t can be expressed as

(t)
=
0

r(t)
_
1 j tan
(t)

(3.23)
and

(t)
=
0

r(t)
_
1 j tan
(t)

, (3.24)
where:

0
= 8.854 10
12
F/m is the permittivity of free space;

0
= 4 10
7
H/m is the permeability of free space;

r(t)
is the relative permittivity of layer t;

r(t)
is the relative permeability of layer t;
tan
(t)
is the electric loss tangent of layer t;
University of PretoriaElectrical, Electronic and Computer Engineering 41
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
tan
(t)
is the magnetic loss tangent of layer t.
The wavenumber in the z direction can be expressed as
k
z(t)
=
_
k
2
(t)
k
2

, m
_
k
z(t)
0
_
, (3.25)
where k

is the wavenumber in the radial direction, given by

k
2

= k
2
x
+k
2
y
. (3.26)
Here, k
x
and k
y
are the wavenumbers in the x and y directions respectively. Furthermore, k
(t)
is
the wavenumber in layer t, given by
k
(t)
=
_

(t)

(t)
, (3.27)
with = 2f the radial frequency and f the frequency.
In order to derive the expressions for the components of the Greens function, which relate the
electric eld in one layer to the electric current density in another layer, the vector wave functions,
which depend on the reection coefcients at the interfaces between layers, as well as the wave
amplitudes within the various layers, have to be found [6, 7, 178]. For an isotropic medium, where
and vary only in one direction (the z direction in this case), the vector wave equations can be
reduced to two scalar equations that are decoupled from each other [178]. They characterise two
types of waves, namely the transverse electric (TE) and transverse magnetic (TM) waves. In the
case of the TE waves, the electric eld is transverse to the z direction, while in the case of the
TM waves, the magnetic eld is transverse to the z direction. The reection coefcients, wave
amplitudes and wave functions associated with both these types of waves, will now be addressed
in the following sections.
3.4.1 Reection Coefcients
First consider a half-space (i.e. there are only two semi-innite planar layers) with layers t and
t + 1, or alternatively, layers t 1 and t. If a wave now travels from layer t to t + 1 or from layer
t to layer t 1, there will be both incident and reected waves in layer t, but only a transmitted
wave in layer t 1. Now, at the interface between the two layers, t and t 1, the Fresnel reection
coefcients are dened as
R
TE
(t),(t1)
=

(t1)
k
z(t)

(t)
k
z(t1)

(t1)
k
z(t)
+
(t)
k
z(t1)
(3.28)
and
R
TM
(t),(t1)
=

(t1)
k
z(t)

(t)
k
z(t1)

(t1)
k
z(t)
+
(t)
k
z(t1)
(3.29)
University of PretoriaElectrical, Electronic and Computer Engineering 42
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
for the TE and TM cases respectively. In (3.28), R
TE
(t),(t1)
can be interpreted as the ratio of the
amplitude of the reected TE wave in layer t1 to the amplitude of the incident TE wave in layer t.
The same goes for R
TM
(t),(t1)
in (3.29), except that it is associated with a TM wave. In multilayered
media, the Fresnel reection coefcients can be calculated at each boundary by considering only
the two layers adjacent to the boundary. There are however two special cases. These are at the top
of layer 0, where there is effectively no boundary and at the bottom of layer T, where it is bounded
by a PEC ground plane. At the top of layer 0, (3.28) and (3.29) reduce to
R
TE
(0),(1)
= 0 (3.30)
and
R
TM
(0),(1)
= 0 (3.31)
respectively, while at the lower boundary of layer T, they become
R
TE
(T),(T+1)
= 1 (3.32)
and
R
TM
(T),(T+1)
= 1 (3.33)
respectively. Layers 1 and T + 1 are merely imaginary layers.
In a multilayered medium, there will of course be multiple reections at each interface between
two adjacent layers. In such a case it is more convenient to work with the generalised reection
coefcients. These are reection coefcients that relate the amplitude of the reected wave to
the amplitude of the incident wave at a specic layer, but also include the effect of subsurface
reections. In the case of TE waves, they are given by

R
TE
(t),(t1)
=
_

_
R
TE
(t),(t1)
, t = 0
R
TE
(t),(t1)
+

R
TE
(t1),(t2)
e
j2k
z(t1)
h
(t1)
1 +R
TE
(t),(t1)

R
TE
(t1),(t2)
e
j2k
z(t1)
h
(t1)
, t = 1, 2, . . . , T
(3.34)
and

R
TE
(t),(t+1)
=
_

_
R
TE
(t),(t+1)
, t = T
R
TE
(t),(t+1)
+

R
TE
(t+1),(t+2)
e
j2k
z(t+1)
h
(t+1)
1 +R
TE
(t),(t+1)

R
TE
(t+1),(t+2)
e
j2k
z(t+1)
h
(t+1)
, t = T 1, T 2, . . . , 0,
(3.35)
where

R
TE
(t),(t1)
is the generalised reection coefcient at the interface between layers t and t 1,
associated with an incident TE wave travelling from layer t to layer t 1. Similarly,

R
TE
(t),(t+1)
is associated with an incident TE wave, travelling from layer t to layer t + 1. By considering the
University of PretoriaElectrical, Electronic and Computer Engineering 43
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
expressions in (3.34) and (3.35) carefully, one can see that they dene recursive relationships. For
example,

R
TE
(t),(t1)
is dependant on

R
TE
(t1),(t2)
and so on, up to the top layer. Therefore in order
to nd

R
TE
(t),(t1)
, one would start at the top layer and work your way down the layers. Similarly,
in order to nd

R
TE
(t),(t+1)
, one would start at the bottom layer and work your way upwards.
In the case of TM waves, the generalised reection coefcients are given by

R
TM
(t),(t1)
=
_

_
R
TM
(t),(t1)
, t = 0
R
TM
(t),(t1)
+

R
TM
(t1),(t2)
e
j2k
z(t1)
h
(t1)
1 +R
TM
(t),(t1)

R
TM
(t1),(t2)
e
j2k
z(t1)
h
(t1)
, t = 1, 2, . . . , T
(3.36)
and

R
TM
(t),(t+1)
=
_

_
R
TM
(t),(t+1)
, t = T
R
TM
(t),(t+1)
+

R
TM
(t+1),(t+2)
e
j2k
z(t+1)
h
(t+1)
1 +R
TM
(t),(t+1)

R
TM
(t+1),(t+2)
e
j2k
z(t+1)
h
(t+1)
, t = T 1, T 2, . . . , 0.
(3.37)
3.4.2 Wave Amplitudes
Let t = denote the layer containing the observation point and t =

denote the layer containing

the source point. The wave amplitude in the source layer

can then be expressed as

A
TE
(

)
=
1
1

R
TE
(

),(

1)

R
TE
(

),(

+1)
e
j2k
z(

)
h
(

)
(3.38)
for TE waves and as
A
TM
(

)
=
1
1

R
TM
(

),(

1)

R
TM
(

),(

+1)
e
j2k
z(

)
h
(

)
(3.39)
for TM waves. The wave amplitude in the observation layer can now also be found through a
recursive process. Let A
TE
()+
or A
TM
()+
denote the wave amplitude in the observation layer for
z z

(i.e. the observation point above or equal to the source point), while A
TE
()
or A
TM
()
denote
the wave amplitude in the observation layer for z < z

(i.e. the observation point below the source

point). A
TE
()+
or A
TM
()+
can be found by starting at the source layer with either (3.38) or (3.39) and
applying either
A
TE
(t)+
= A
TE
(t+1)+
e
j[k
z(t)
k
z(t+1)
]d
(t)
1 R
TE
(t),(t+1)
1 R
TE
(t),(t+1)

R
TE
(t),(t1)
e
j2k
z(t)
h
(t)
,
t =

1,

2, . . . , (3.40)
University of PretoriaElectrical, Electronic and Computer Engineering 44
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
or
A
TM
(t)+
= A
TM
(t+1)+
e
j[k
z(t)
k
z(t+1)
]d
(t)
1 R
TM
(t),(t+1)
1 R
TM
(t),(t+1)

R
TM
(t),(t1)
e
j2k
z(t)
h
(t)
,
t =

1,

2, . . . , (3.41)
for each consecutive layer up to the observation layer. Similarly, A
TE
()
or A
TM
()
can be found by
starting at the source layer with either (3.38) or (3.39) and applying either
A
TE
(t)
= A
TE
(t1)
e
j[k
z(t1)
k
z(t)
]d
(t1)
1 R
TE
(t),(t1)
1 R
TE
(t),(t1)

R
TE
(t),(t+1)
e
j2k
z(t)
h
(t)
,
t =

+ 1,

+ 2, . . . , (3.42)
or
A
TM
(t)
= A
TM
(t1)
e
j[k
z(t1)
k
z(t)
]d
(t1)
1 R
TM
(t),(t1)
1 R
TM
(t),(t1)

R
TM
(t),(t+1)
e
j2k
z(t)
h
(t)
,
t =

+ 1,

+ 2, . . . , (3.43)
for each consecutive layer down to the observation layer.
3.4.3 Wave Functions
The wave function, for an observation point at z in layer due to a source at z

in layer

, can
then be expressed as a superposition of upgoing and downgoing waves. For TE waves, the wave
function can be expressed as
F
TE
()+
(z, z

) = A
TE
()+
_
e
jk
z(

)
z

+

R
TE
(

),(

+1)
e
j2k
z(

)
d
(

)
e
jk
z(

)
z

_
e
jk
z()
z
+

R
TE
(),(1)
e
j2k
z()
d
(1)
e
jk
z()
z
_
(3.44)
if z z

, and as
F
TE
()
(z, z

) = A
TE
()
_
e
jk
z()
z
+

R
TE
(),(+1)
e
j2k
z()
d
()
e
jk
z()
z
_

_
e
jk
z(

)
z

+

R
TE
(

),(

1)
e
j2k
z(

)
d
(

1)
e
jk
z(

)
z

_
(3.45)
if z < z

. Similarly, for TM waves, the wave function can be expressed as

F
TM
()+
(z, z

) = A
TM
()+
_
e
jk
z(

)
z

+

R
TM
(

),(

+1)
e
j2k
z(

)
d
(

)
e
jk
z(

)
z

_
e
jk
z()
z
+

R
TM
(),(1)
e
j2k
z()
d
(1)
e
jk
z()
z
_
(3.46)
University of PretoriaElectrical, Electronic and Computer Engineering 45
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
if z z

, and as
F
TM
()
(z, z

) = A
TM
()
_
e
jk
z()
z
+

R
TM
(),(+1)
e
j2k
z()
d
()
e
jk
z()
z
_

_
e
jk
z(

)
z

+

R
TM
(

),(

1)
e
j2k
z(

)
d
(

1)
e
jk
z(

)
z

_
(3.47)
if z < z

. The derivatives and double derivatives of F

TM
+
and F
TM

with respect to z and z

are also
required in the expressions for the Greens function. After some straightforward manipulation,
they are given by

z
F
TM
()+
(z, z

) = jk
z()
A
TM
()+
_
e
jk
z(

)
z

+

R
TM
(

),(

+1)
e
j2k
z(

)
d
(

)
e
jk
z(

)
z

_
e
jk
z()
z
+

R
TM
(),(1)
e
j2k
z()
d
(1)
e
jk
z()
z
_
, (3.48)

z
F
TM
()
(z, z

) = jk
z()
A
TM
()
_
e
jk
z()
z

R
TM
(),(+1)
e
j2k
z()
d
()
e
jk
z()
z
_

_
e
jk
z(

)
z

+

R
TM
(

),(

1)
e
j2k
z(

)
d
(

1)
e
jk
z(

)
z

_
, (3.49)

F
TM
()+
(z, z

) = jk
z(

)
A
TM
()+
_
e
jk
z(

)
z

R
TM
(

),(

+1)
e
j2k
z(

)
d
(

)
e
jk
z(

)
z

_
e
jk
z()
z
+

R
TM
(),(1)
e
j2k
z()
d
(1)
e
jk
z()
z
_
, (3.50)

F
TM
()
(z, z

) = jk
z(

)
A
TM
()
_
e
jk
z()
z
+

R
TM
(),(+1)
e
j2k
z()
d
()
e
jk
z()
z
_

_
e
jk
z(

)
z

+

R
TM
(

),(

1)
e
j2k
z(

)
d
(

1)
e
jk
z(

)
z

_
, (3.51)

2
z

z
F
TM
()+
(z, z

) = k
z(

)
k
z()
A
TM
()+
_
e
jk
z(

)
z

R
TM
(

),(

+1)
e
j2k
z(

)
d
(

)
e
jk
z(

)
z

_
e
jk
z()
z
+

R
TM
(),(1)
e
j2k
z()
d
(1)
e
jk
z()
z
_
(3.52)
and

2
z

z
F
TM
()
(z, z

) = k
z(

)
k
z()
A
TM
()
_
e
jk
z()
z

R
TM
(),(+1)
e
j2k
z()
d
()
e
jk
z()
z
_

_
e
jk
z(

)
z

+

R
TM
(

),(

1)
e
j2k
z(

)
d
(

1)
e
jk
z(

)
z

_
. (3.53)
The terms between brackets in (3.44) to (3.53) should be multiplied out in order to avoid numerical
overow for large values of k

, which in turn results in large negative imaginary values for k

z
.
University of PretoriaElectrical, Electronic and Computer Engineering 46
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
3.4.4 Components of the Greens Function
Figure 3.5 shows the various current-density components that have to be taken into account for
the purposes of this analysis. In general then, the current density can be expanded in terms of its
individual components as
J(x, y, z) = J
x
(x, y) x +J
y
(x, y) y +J
z
(z) z. (3.54)
Due to the fact that the current density is either horizontal or vertical, but not oblique, this kind
of problem is often referred to as two-and-one-half-dimensional (2.5D) as opposed to a full three-
dimensional problem (3D). The analysis of 2.5D electromagnetic problems, however, still requires
knowledge of all nine components of the dyadic Greens function that relates the electric eld to
the electric current density.
Now, the dyadic Greens function can be expressed in terms of its spectral-domain components as

G = x

G
xx
x+ x

G
xy
y + x

G
xz
z + y

G
yx
x+ y

G
yy
y + y

G
yz
z + z

G
zx
x+ z

G
zy
y + z

G
zz
z,
(3.55)
where, for example, the

G
xy
component provides the x-directed component of the electric eld
due to the y-directed component of the electric current density.
The nine components of (3.55) are given by

G
xx
(k
x
, k
y
, z|z

) =

)
2k
z(

)
k
2

_
k
2
y
F
TE
()
(z, z

) +
k
2
x

()

2
z

z
F
TM
()
(z, z

)
_
, (3.56)

G
yy
(k
x
, k
y
, z|z

) =

)
2k
z(

)
k
2

_
k
2
x
F
TE
()
(z, z

) +
k
2
y

()

2
z

z
F
TM
()
(z, z

)
_
, (3.57)

G
xy
(k
x
, k
y
, z|z

) =

G
yx
(k
x
, k
y
, z|z

)
=

)
2k
z(

)
k
2

k
x
k
y
_
F
TE
()
(z, z

) +
1

()

2
z

z
F
TM
()
(z, z

)
_
, (3.58)

G
xz
(k
x
, k
y
, z|z

) =
jk
x
2k
z(

()

z
F
TM
()
(z, z

), (3.59)

G
yz
(k
x
, k
y
, z|z

) =
jk
y
2k
z(

()

z
F
TM
()
(z, z

), (3.60)

G
zx
(k
x
, k
y
, z|z

) =
jk
x
2k
z(

()

F
TM
()
(z, z

), (3.61)

G
zy
(k
x
, k
y
, z|z

) =
jk
y
2k
z(

()

F
TM
()
(z, z

) (3.62)
University of PretoriaElectrical, Electronic and Computer Engineering 47
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
x
y
z
J
x
J
y
J
z
J
x
J
y
J
z
J
z
Layer 0
Layer 1
Layer T
PEC ground plane
Figure 3.5 Current-density components.
and

G
zz
(k
x
, k
y
, z|z

) =
k
2

2k
z(

()
F
TM
()
(z, z

)
1
j
(

)
(z z

). (3.63)
In Appendix A, it is shown how these Greens-function components can be derived. The Greens-
function components for two [128, 129], three [132, 175] and four layers [28] are available in
the literature, but are often limited to the case where the probe only passes through one of the
substrate layers. For a more general analysis, also including an arbitrary number of substrate
layers, the recursive method becomes inevitable.
The next step in the analysis is to dene appropriate basis functions (testing functions are similar)
over which to expand the electric current density. This will be dealt with in the section that follows.
3.5 BASIS FUNCTIONS
As mentioned earlier on, the antenna structure basically consists of probes, capacitor patches
(which can either be circular or rectangular) and rectangular (or square) resonant pathes. In order
to obtain an efcient computational solution, the current density over each of these parts has
to be approximated by as few a number of basis functions as possible. This can be achieved
by using entire-domain basis functions on the rectangular resonant patches, which account for
most of the antenna size, while resorting to subdomain basis functions only on the probes and
capacitor patches. The latter are usually quite small when compared to a wavelength and therefore
should require relatively few subdomain basis functions. Furthermore, during the analysis of some
antenna congurations, it might happen that the individual basis functions are not always aligned
to one another, but that they have different orientations. In order to simplify the analysis of such an
arrangement, it is necessary to dene a local coordinate system for each basis function as shown
in Figure 3.6. These local coordinate systems are denoted by u and v axes, as opposed to the x
University of PretoriaElectrical, Electronic and Computer Engineering 48
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
and y axes for the global coordinate system.
In mathematical notation, the complete set of basis functions, which is used for the analysis of
microstrip patch antennas with capacitive feed probes, can be expressed as
J =
N

n=1
I
n
f
n
=
N
P

np=1
N
PZ
np

nz=1
I
PZ
np:nz
f
PZ
np:nz
. .
PWS basis functions on probes
+
N
A

na=1
I
A
na
_
f
AZ
na
+f
AP
na
_
. .
attachment modes
+
N
S

ns=1
_
_
N
SU
ns

nu=1
I
SU
ns:nu
f
SU
ns:nu
+
N
SV
ns

nv=1
I
SV
ns:nv
f
SV
ns:nv
_
_
. .
subdomain rooftop basis functions on rectangular capacitor patches
+
N
E

ne=1
_
_
N
EU
ne

nu=1
I
EU
ne:nu
f
EU
ne:nu
+
N
EV
ne

nv=1
I
EV
ne:nv
f
EV
ne:nv
_
_
. .
entire-domain sinusoidal basis functions on resonant patches
. (3.64)
Here, f refers to the different vector basis functions, while I refers to the unknown current-density
coefcients.
The basis functions on the N
P
probes are taken to be overlapping piecewise-sinusoidal (PWS)
functions and also take into account the nite radii of the probes. They are represented by f
PZ
np:nz
,
which is the nz-th of N
PZ
np
basis functions on the np-th probe.
The N
A
attachment modes (being either circular or rectangular) all have two parts: one part, f
AZ
na
,
residing on the probe and one part, f
AP
na
, residing on the patch. The probe part of each attach-
ment mode is modelled by one half of a PWS function right below the probe-to-patch junction.
The patch part (which differs for the circular and rectangular attachment modes) is described in
more detail in Section 3.6. The patch part of the circular attachment mode alone is sufcient for
modelling the electric current density on the circular capacitor patches, whereas the remaining
current density on the rectangular capacitor patches is approximated by overlapping rooftop basis
functions.
Rooftop basis functions yield more exibility in the modelling of the patch current and are usually
more convenient for small patch sizes, while entire-domain sinusoidal basis functions (usually ex-
pressed in terms of sine and cosine functions) yield a smaller number of unknowns in the solution
and are therefore more efcient for resonant pathes. The overlapping rooftop basis functions on
the N
S
rectangular capacitor patches are then represented by f
SU
ns:nu
and f
SV
ns:nv
, where f
SU
ns:nu
is the
nu-th of N
SU
ns
u-directed basis functions on the ns-th rectangular capacitor patch and f
SV
ns:nv
is the
University of PretoriaElectrical, Electronic and Computer Engineering 49
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation

v
u
v
u
v
u
v
u
v
u
Circular
attachment mode
Entire-domain sinusoidal basis functions
Subdomain rooftop basis functions
Entire-domain sinusoidal basis functions
y
x
Figure 3.6 Local coordinate systems for some of the basis functions.
University of PretoriaElectrical, Electronic and Computer Engineering 50
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
nv-th of N
SV
ns
v-directed basis functions on the ns-th rectangular capacitor patch. Similarly, the
entire-domain sinusoidal basis functions on the N
E
rectangular resonant patches are represented
by f
EU
ne:nu
and f
EV
ne:nv
, where f
EU
ne:nu
is the nu-th of N
EU
ne
u-directed basis functions on the ne-th
rectangular resonant patch and f
EV
ne:nv
is the nv-th of N
EV
ne
v-directed basis functions on the ne-th
rectangular resonant patch.
If the complete set of basis functions is used in an analysis, the total number of basis functions on
the structure, N, can be calculated as
N =
N
P

np=1
N
PZ
np
+N
A
+
N
S

ns=1
_
N
SU
ns
+N
SV
ns
_
+
N
E

ne=1
_
N
EU
ne
+N
EV
ne
_
. (3.65)
Note that it is not always necessary to use the complete set of basis functions. For example, if the
structure does not comprise of rectangular capacitor patches, the rooftop basis functions are not
required.
Each of the different basis functions, as have just been discussed, will now be described in some-
what more detail in the sections that follow. Section 3.5.1 addresses the PWS basis functions
on the probes, Section 3.5.2 the subdomain rooftop basis functions on the rectangular capacitor
patches and Section 3.5.3 the entire-domain sinusoidal basis functions on the rectangular reso-
nant patches. The attachment modes are somewhat more specialised and are therefore described
separately in Section 3.6.
3.5.1 Piecewise-Sinusoidal Basis Functions on the Probes
In a thick substrate, the electric current density on the surface of the probes will generally vary
along the length of the probes. Furthermore, if the probes are thin, which is usually the case,
the electric current density should have minimal angular variation around the probe. As shown
in Figure 3.7, the electric current density on the probes can be modelled by a set of PWS basis
functions, residing on the surface of every probe. These basis functions are used to model the
electric current density along the length of each probe. In most approaches [6, 7, 125], piecewise-
linear basis functions are normally used, but these functions result in very complicated expressions
for the interaction-matrix entries where integration over the length of the probe is required. The
PWS functions can approximate piecewise-linear functions very well, but result in much simpler
expressions for the interaction-matrix entries.
In a coordinate system that is local to each probe, the basis functions can be expressed as
f
PZ
np:nz
(u, v, z) =
1
2a
np
f
PZ
nz
(z) z

u
2
+v
2
=a
2
np
, (3.66)
where f
PZ
np:nz
refers to the nz-th basis function on the np-th probe. The basis functions on the np-
University of PretoriaElectrical, Electronic and Computer Engineering 51
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation

u v
z
J(z)
f
PZ
1
(z)
f
PZ
2
(z)
f
PZ
nz
(z)
f
PZ
N
PZ
np
1
(z)
f
PZ
N
PZ
np
(z)
z
PZ
nz
z
PZ+
nz
z
PZ
nz
PEC ground plane
2a
np
(a)
z
J(z)
f
PZ
nz
(z)
Layer 0
Layer 1
Layer T
PEC ground plane
(b)
Figure 3.7 Piecewise-sinusoidal basis functions on probe np. (a) Isolated probe. (b) Probe in a
multilayered substrate.
th probe are numbered from 1 for the bottom-most basis function to N
PZ
np
for the top-most basis
function. The radius of the np-th probe is denoted by a
np
.
The PWS parts of the basis functions are given by
f
PZ
nz
(z) =
_

_
sin[k
F
(z
PZ
nz
+ z
PZ+
nz
z)]
sin(k
F
z
PZ+
nz
)
nz 1, z
PZ
nz
z z
PZ
nz
+ z
PZ+
nz
sin[k
F
(z z
PZ
nz
+ z
PZ
nz
)]
sin(k
F
z
PZ
nz
)
nz 2, z
PZ
nz
z
PZ
nz
z z
PZ
nz
.
(3.67)
Here, z
PZ
nz
is the z position of the nz-th PWS function, while z
PZ+
nz
and z
PZ
nz
are the segment
sizes just above and below z
PZ
nz
, as shown in Figure 3.7(a).
In order to closely approximate piecewise-linear basis functions, the value of k
F
is chosen as
k
F
=
1
100 max
_
z
PZ+
1
, z
PZ+
2
, . . . , z
PZ+
N
PZ
max
_, (3.68)
University of PretoriaElectrical, Electronic and Computer Engineering 52
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
where
N
PZ
max
= max
_
N
PZ
1
, N
PZ
2
, . . . , N
PZ
N
P
_
. (3.69)
By using the value of k
F
in (3.68), the PWS functions deviate only very slightly from piecewise-
linear functions over the same support.
In the spectral domain, the probe basis functions can be expressed as

f
PZ
np:nz
(k
u
, k
v
, z) = f
PZ
nz
(z) J
0
(k

a
np
) z (3.70)
where J
0
() is the Bessel function of the rst kind of order 0 and
k
2

= k
2
u
+k
2
v
= k
2
x
+k
2
y
. (3.71)
In Appendix B, it is shown how the spectral-domain form of these basis functions can be derived.
3.5.2 Rooftop Basis Functions on the Rectangular Capacitor Patches
Due to the probe-to-patch junction on capacitor patches, the electric current density on the rectan-
gular capacitor patches has to be partially modelled by an attachment mode. However, the patch
part of the attachment mode alone is not sufcient to model the electric current density on the
rectangular capacitor patches and therefore it has to be augmented with other basis functions. Due
to the geometry of the rectangular capacitor patch, and its relative small size, subdomain rooftop
basis functions [137139] are a good choice. Figure 3.8 illustrates the use of subdomain rooftop
basis functions on a rectangular capacitor patch.
In a coordinate system that is local to each rooftop basis function, the rooftop basis functions are
expressed as
f
SU
ns:nu
(u, v) =
_
1
|u|
l
ns
_
rect
_
v
w
ns
_
u, |u| l
ns
, z = z
S
ns
(3.72)
for the u-directed basis functions and as
f
SV
ns:nv
(u, v) =
_
1
|v|
w
ns
_
rect
_
u
l
ns
_
v, |v| w
ns
, z = z
S
ns
(3.73)
for the v-directed basis functions. Here, f
SU
ns:nu
refers to the nu-th u-directed basis function on the
ns-th rectangular capacitor patch, while f
SV
ns:nv
refers to the nv-th v-directed basis function on the
ns-th rectangular capacitor patch. Also, z
S
ns
is the z position of the ns-th rectangular capacitor
patch.
As shown in Figure 3.8, u and v point along the local u and v directions for the ns-th capacitor
patch. The ns-th rectangular capacitor patch has dimensions of l
ns
and w
ns
along the u and v
University of PretoriaElectrical, Electronic and Computer Engineering 53
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation

u
v
z
u
v
z

l
n
s
l
n
s

w
n
s
w
n
s
f
SV
ns:nv
(u, v)
f
SU
ns:nu
(u, v)
Figure 3.8 Subdomain rooftop basis functions on rectangular
capacitor patch ns.
directions respectively. Each capacitor patch is also divided into a grid of cells, where the size of
a single cell is l
ns
by w
ns
. As shown in Figure 3.8, each rooftop basis function then spans two
such cells.
In the spectral domain, the subdomain rooftop basis functions can be expressed as

f
SU
ns:nu
(k
u
, k
v
) =
8
l
ns
k
2
u
k
v
sin
2
_
k
u
l
ns
2
_
sin
_
k
v
w
ns
2
_
u, z = z
S
ns
(3.74)
and

f
SV
ns:nv
(k
u
, k
v
) =
8
w
ns
k
2
v
k
u
sin
2
_
k
v
w
ns
2
_
sin
_
k
u
l
ns
2
_
v, z = z
S
ns
(3.75)
respectively. In Appendix B, it is shown how the spectral-domain form of these basis functions
can be derived.
3.5.3 Sinusoidal Basis Functions on the Resonant Patches
As the rectangular resonant patches are not directly coupled to any other part of the structure,
entire-domain sinusoidal basis functions [31, 123125], which normally originate from a cavity-
model analysis of rectangular patch antennas, should be sufcient to model the electric current
density on these patches. The entire-domain sinusoidal basis functions are illustrated in Figure 3.9.
In a coordinate systemthat is local to each rectangular resonant patch, the entire-domain sinusoidal
University of PretoriaElectrical, Electronic and Computer Engineering 54
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation

W
n
e
Ln
e
u
v
z
f
EU
ne:nu
(u, v)
f
EV
ne:nv
(u, v)
Figure 3.9 Entire-domain sinusoidal basis functions on rectangular resonant
patch ne. Here, p
EU
ne:nu
= 2 and q
EU
ne:nu
= 1, while p
EV
ne:nv
= 2 and q
EV
ne:nv
= 1.
basis functions are expressed as
f
EU
ne:nu
(u, v) = sin
_
p
EU
ne:nu

L
ne
_
u +
L
ne
2
__
cos
_
q
EU
ne:nu

W
ne
_
v +
W
ne
2
__
u,
|u|
L
ne
2
, |v|
W
ne
2
, z = z
E
ne
(3.76)
for the u-directed basis functions and as
f
EV
ne:nv
(u, v) = sin
_
p
EV
ne:nv

W
ne
_
v +
W
ne
2
__
cos
_
q
EV
ne:nv

L
ne
_
u +
L
ne
2
__
v,
|u|
L
ne
2
, |v|
W
ne
2
, z = z
E
ne
(3.77)
for the v-directed basis functions. Here, f
EU
ne:nu
refers to the nu-th u-directed basis function on
the ne-th rectangular resonant patch, while f
EV
ne:nv
refers to the nv-th v-directed basis function on
University of PretoriaElectrical, Electronic and Computer Engineering 55
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
the ne-th rectangular resonant patch. Also, z
E
ne
is the z position of the ne-th rectangular resonant
patch.
As shown in Figure 3.9, u and v point along the local u and v directions for the ne-th resonant
patch. The patch has dimensions of L
ne
along the u direction and W
ne
along the v direction.
The (p
EU
ne:nu
, q
EU
ne:nu
) mode is associated with the nu-th u-directed basis function on the ne-th
rectangular resonant patch, while the (p
EV
ne:nv
, q
EV
ne:nv
) mode is associated with the nv-th v-directed
basis function. Here, p
EU
ne:nu
and p
EV
ne:nv
can take on values of 1, 2, . . . , while q
EU
ne:nu
and q
EV
ne:nv
can
take on values of 0, 1, 2, . . . .
In the spectral domain, the entire-domain sinusoidal basis functions can be expressed as

f
EU
ne:nu
(k
u
, k
v
) =
p
EU
ne:nu
/L
ne
k
2
u
(p
EU
ne:nu
/L
ne
)
2
_
(1)
p
EU
ne:nu
e
jk
u
L
ne
/2
e
jk
u
L
ne
/2
_

jk
v
k
2
v
(q
EU
ne:nu
/W
ne
)
2
_
(1)
q
EU
ne:nu
e
jk
v
W
ne
/2
e
jk
v
W
ne
/2
_
u, z = z
E
ne
(3.78)
and

f
EV
ne:nv
(k
u
, k
v
) =
p
EV
ne:nv
/W
ne
k
2
v
(p
EV
ne:nv
/W
ne
)
2
_
(1)
p
EV
ne:nv
e
jk
v
W
ne
/2
e
jk
v
W
ne
/2
_

jk
u
k
2
u
(q
EV
ne:nv
/L
ne
)
2
_
(1)
q
EV
ne:nv
e
jk
u
L
ne
/2
e
jk
u
L
ne
/2
_
v, z = z
E
ne
(3.79)
respectively. As with the other basis functions, the derivation of these expressions can be found in
Appendix B.
3.6 ATTACHMENT MODES
As mentioned before, the attachment modes are special basis functions that are used to accurately
model probe-to-patch transitions. They ensure the continuity of the electric current at the junction
and also model the singular behaviour of the electric current density at the junction. In this section,
two types of attachment modes will be addressed. These will be referred to as the rectangular and
circular attachment modes respectively.
2
Both approaches have been used by different researchers,
mainly for the analysis of probes connected to resonant patches on a thin substrate. From the
literature, it is not clear whether they can be used for probes in thick substrates and for patches
that are well below resonant size. Therefore, both of these approaches will be investigated to
determine which one is the most appropriate. As they differ substantially, they will be considered
separately.
2
The two attachment modes are not formally known as rectangular and circular attachment modes, but are termed so
here in order to distinguish between them.
University of PretoriaElectrical, Electronic and Computer Engineering 56
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
3.6.1 Rectangular Attachment Mode
The rectangular attachment mode can only be used to model probe-to-patch junctions for rectan-
gular patches. This attachment mode was rst proposed by Aberle and Pozar [129, 132], and has
since been extended by Chen and Lee [6, 7]. As shown in Figure 3.10(a), the probe part of the
attachment mode can be modelled by one half of a PWS function right below every probe-to-patch
junction. In a coordinate system that is local to each patch on which an attachment mode resides,
the probe part of each attachment mode can therefore be expressed as
f
AZ
na
(u, v, z) =
1
2a
na
f
AZ
na
(z) z

(uu
AZ
na
)
2
+(vv
AZ
na
)
2
=a
2
na
, (3.80)
where f
AZ
na
refers to the probe part of the na-th attachment mode. If the current density on the
probe, associated with the attachment mode, is approximated by N
PZ
np
PWS basis functions, the
probe part of the attachment mode can also be viewed as the (N
PZ
np
+ 1)-th basis function on the
probe. The radius of the probe associated with the attachment mode, is denoted by a
na
, while its
position relative to the centre of the patch, is denoted by (u
AZ
na
, v
AZ
na
).
The PWS part of the basis function is given by
f
AZ
na
(z) =
sin[k
F
(z z
A
na
+ z
A
na
)]
sin(k
F
z
A
na
)
, z
A
na
z
A
na
z z
A
na
, (3.81)
where z
A
na
is the z position of the na-th attachment mode (i.e. the top of the associated probe) and
z
A
na
is the segment size just below z
A
na
, as shown in Figure 3.10(a). The value of k
F
is given by
(3.68).
In the spectral domain, the probe part of each attachment mode can be expressed as

f
AZ
na
(k
u
, k
v
, z) = f
AZ
na
(z) J
0
(k

a
na
) e
j(k
u
u
AZ
na
+k
v
v
AZ
na
)
z. (3.82)
It is derived in a similar way as the spectral-domain form of the other PWS basis functions on the
probe.
The patch part of the entire-domain attachment mode is based on the the eigenmodes of the
magnetic-wall cavity model for a rectangular patch with an uniform current lament source. In a
coordinate system that is local to each patch on which an attachment mode resides, the patch part
of the attachment mode can be written as [6, 7]
f
AP
na
(u, v) =
1
2
_
2
0
J
AP
na
(u, v, u
AZ
na
, v
AZ
na
) d, (3.83)
where
u
AZ
na
= u
AZ
na
+a
na
cos() (3.84)
University of PretoriaElectrical, Electronic and Computer Engineering 57
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
and
v
AZ
na
= v
AZ
na
+a
na
sin() (3.85)
are the coordinates for any position on the surface of the probe. In (3.83), J
AP
na
(u, v, u
AZ
na
, v
AZ
na
)
is the electric current density that would exist at (u, v) on the patch due to a current lament at
( u
AZ
na
, v
AZ
na
). Therefore, in order to nd the electric current density that would exist at (u, v) due
to the total current on the probe, J
AP
na
(u, v, u
AZ
na
, v
AZ
na
) has to be integrated over the circumference
of the probe.
After some manipulation, which is shown in Appendix B, J
AP
na
(u, v, u
AZ
na
, v
AZ
na
) can be expressed
as [6, 7, 129, 132]
J
AP
na
(u, v, u
AZ
na
, v
AZ
na
) =
1
2W
na

i=0

i
cos
_
i
W
na
_
v
AZ
na
+
W
na
2
__

_
g
s
(, L
na
, u
AZ
na
, u)f
c
(i, W
na
, v) u +
_
i
W
na
_
g
c
(, L
na
, u
AZ
na
, u)f
s
(i, W
na
, v) v
_
,
|u|
L
na
2
, |v|
W
na
2
, z = z
A
na
, (3.86)
where
g
s
(, L
na
, u
AZ
na
, u) =
sin[(u + u
AZ
na
)] sgn(u u
AZ
na
) sin[(L
na
|u u
AZ
na
|)]
sin(L
na
)
, (3.87)
g
c
(, L
na
, u
AZ
na
, u) =
cos[(u + u
AZ
na
)] + cos[(L
na
|u u
AZ
na
|)]
sin(L
na
)
, (3.88)
f
s
(i, W
na
, v) = sin
_
i
W
na
_
v +
W
na
2
__
(3.89)
and
f
c
(i, W
na
, v) = cos
_
i
W
na
_
v +
W
na
2
__
, (3.90)
with
sgn(u) =
_

_
1, u > 0
0, u = 0
1, u < 0
, (3.91)
=

r(eff)
k
2
0

_
i
W
na
_
2
(3.92)
and the Neumann numbers

i
=
_
_
_
1, i = 0
2, i > 0.
(3.93)
In (3.92), k
0
= 2/
0
, where
0
is the free-space wavelength. Also,
r(eff)
is the effective relative
University of PretoriaElectrical, Electronic and Computer Engineering 58
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
permittivity of the layers below the patch. For two layers between the probe-fed patch and the
ground plane, it can be calculated as [65, 66, 179]

r(eff)
=

r(1)

r(2)
_
h
(1)
+h
(2)

r(1)
h
(2)
+
r(2)
h
(1)
_
1 tan
(1)
tan
(2)

. (3.94)
Figure 3.10(b) shows the patch part of the attachment mode where the patch size is small as
compared to a wavelength, while Figure 3.10(c) shows the same for a resonant patch.
3
From these
two gures it can clearly be seen that, when the patch is small, the attachment mode basically only
models the singular nature of the electric current density at the junction position, but when the
patch is of a resonant size, the attachment mode also includes the resonant modes.
In the spectral domain, the patch part of the attachment mode can be expressed as

f
AP
na
(k
u
, k
v
) =
j(k
u
u +k
v
v)
k
2
z(eff)
J
0
(k

a
na
) e
j(k
u
u
AZ
na
+k
v
v
AZ
na
)

(k
u
k
v
u +k
2

v)
k
2
z(eff)
A(k

)J
0
(
_

r(eff)
k
0
a
na
) e
jk
u
u
AZ
na

1
W
na

i=0

i
cos
_
i
W
na
_
v
AZ
na
+
W
na
2
__
(1)
i
e
jk
v
W
na
/2
e
jk
v
W
na
/2
[k
2
v
(i/W
na
)
2
][k
2
u

2
]

_
k
v
u +k
u
(i/W
na
)
2

2
v
__
cos(u
AZ
na
) sin(k
u
L
na
/2)
sin(L
na
/2)
j
sin(x
AZ
na
) cos(k
u
L
na
/2)
cos(L
na
/2)
_
J
0
(
_

r(eff)
k
0
a
na
), z = z
A
na
, (3.95)
where
A(k

) =
j
k

_
cos(k

v
AZ
na
) sin(k
v
W
na
/2)
sin(k

W
na
/2)
j
sin(k

v
AZ
na
) cos(k
v
W
na
/2)
cos(k

W
na
/2)
_
, (3.96)
k
2

=
r(eff)
k
2
0
k
2
u
(3.97)
and
k
2
z(eff)
=
_

r(eff)
k
2
0
k
2
u
k
2
v

=
_
k
2
v
k
2

_
. (3.98)
The derivation of the spectral-domain form of the patch part of the rectangular attachment mode
can be found in [6,7], but is also included in Appendix B for completeness sake. Also, in [6,7], the
resonant mode is subtracted from the electric current density on the patch, while angular variation
of the electric current density on the probe is also taken into account. This is not done here and
therefore the expressions are not exactly the same as those in [6, 7].
3
Although the new antenna elements do not include resonant patches that are driven directly with probes, such cong-
urations will be studied in Chapter 4 in order to establish the abilities of the attachment mode.
University of PretoriaElectrical, Electronic and Computer Engineering 59
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation

u
v
z
2a
na
z
A
na
z
A
na
f
AZ
na
(z)
u A
Z
n
a
v
A
Z
n
a
(a)

W
n
a
L
n
a
u
v
z
e
_
_
f
u
(u, v)
_
AP
na
_
(b)

W
n
a
Ln
a
u
v
z
e
_
_
f
u
(u, v)
_
AP
na
_
(c)
Figure 3.10 Rectangular attachment mode na. (a) Probe part of the attachment mode. (b) Patch part of
the attachment mode for a small patch. (c) Patch part of the attachment mode for a resonant patch.
University of PretoriaElectrical, Electronic and Computer Engineering 60
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
3.6.2 Circular Attachment Mode
In order to model the probe-to-patch transitions for circular patches, it is necessary to use circular
attachment modes. Such an attachment mode was developed by Pinhas and Shtrikman [127], and
has since also been used for the analysis of rectangular probe-fed patches [124, 125]. As shown in
Figure 3.11(a), the probe part of the circular attachment mode can also be modelled by one half of
a PWS function right below every probe-to-patch junction. In a coordinate system that is local to
the top of the probe associated with the attachment mode, the probe part of each attachment mode
can therefore be expressed as
f
AZ
na
(u, v, z) =
1
2a
na
f
AZ
na
(z) z

u
2
+v
2
=a
na
, (3.99)
where f
AZ
na
refers to the probe part of the na-th attachment mode. If the current density on the
probe associated with the attachment mode, is approximated by N
PZ
np
PWS basis functions, the
probe part of the attachment mode can also be viewed as the (N
PZ
np
+ 1)-th basis function on the
probe. The radius of the probe associated with the attachment mode is denoted by a
na
.
The PWS part of the basis function is given by
f
AZ
na
(z) =
sin[k
F
(z z
A
na
+ z
A
na
)]
sin(k
F
z
A
na
)
, z
A
na
z
A
na
z z
A
na
, (3.100)
where z
A
na
is the z position of the na-th attachment mode (i.e. the top of the associated probe) and
z
A
na
is the segment size just below z
A
na
, as shown in Figure 3.11(a). The value of k
F
is given by
(3.68).
In the spectral domain, the probe part of the attachment mode can be expressed as

f
AZ
na
(k
u
, k
v
, z) = f
AZ
na
(z) J
0
(k

a
na
) z. (3.101)
It is derived in a similar way as the spectral-domain form of the other PWS basis functions on the
probe.
The patch part of the circular attachment mode is represented by a radial current density, spreading
over a small ctitious disk with radius b
na
, concentric with the probe. It can be written as [127]
f
AP
na
(u, v) =
_

2b
2
na
, 0 < a
na
, z = z
A
na
_

2b
2
na
+
1
2
_
, a
na
b
na
, z = z
A
na
0, > b
na
, z = z
A
na
,
(3.102)
University of PretoriaElectrical, Electronic and Computer Engineering 61
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation

u v
z
2a
na
2b
na
z
A
na
f
AZ
na
(z)
z
A
na
(a)

u
v
z
2a
na
2b
na
z
A
na
f
AP
na
(u, v)
(b)
Figure 3.11 Circular attachment mode na. (a) Probe part of the attachment mode. (b) Patch part of the
attachment mode.
where
=
_
u
2
+v
2
(3.103)
and
= u u +v v. (3.104)
As is shown in Figure 3.11(b), it has a maximum value at the probe-to-patch junction and then
tapers off towards zero at the edge of the attachment mode.
In the spectral domain, the patch part of the attachment mode can be expressed as

f
AP
na
(k
u
, k
v
) =
_
jJ
2
(k

b
na
)
k

jJ
0
(k

a
na
)
k

+
jJ
0
(k

b
na
)
k

__
k
u
k

u +
k
v
k

v
_
=
_
j2J
1
(k

b
na
)
k
2

b
na

jJ
0
(k

a
na
)
k

__
k
u
k

u +
k
v
k

v
_
, z = z
A
na
, (3.105)
where J
1
() is the Bessel function of the rst kind of order 1 and J
2
() is the Bessel function of
the rst kind of order 2. In order to avoid numerical overow problems for very small values of
k

, it is advisable to use the rst form in (3.105) together with the small-argument form of J
2
().
University of PretoriaElectrical, Electronic and Computer Engineering 62
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
The term associated with J
2
() can then be approximated as
jJ
2
(k

b
na
)
k

jk

b
2
na
8
. (3.106)
The derivation of (3.105) can be found in Appendix B.
The circular attachment mode can be used to model the complete electric current density on the
circular capacitor patches. However, in the next section a higher-order circular attachment mode
will be presented that appears to be somewhat more accurate, especially when only the attachment
mode is used to model the electric current density on a circular capacitor patch.
3.6.3 Higher-Order Circular Attachment Mode
The probe part of the higher-order circular attachment mode is exactly the same as the one de-
scribed in the previous section. The patch part of the higher-order circular attachment mode is
also represented by a radial current density, spreading over a small ctitious disk with radius b
na
and which is concentric with the probe. However, on the surface of the patch, the current that ows
towards the centre of the patch, is smaller than that for the normal circular attachment mode, which
was presented in the previous section. The patch part of the higher-order circular attachment mode
can be written as
f
AP
na
(u, v) =
_

3
2b
4
na
, 0 < a
na
, z = z
A
na
_

3
2b
4
na
+
1
2
_
, a
na
b
na
, z = z
A
na
0, > b
na
, z = z
A
na
.
(3.107)
Here, the higher-order terms force the current density that ows inwards, towards the centre of the
attachment mode, to be lower relative to the current density that ows outwards, towards the edge
of the attachment mode. It is important to note that not just any functions can be chosen to achieve
the desired effect. The current must be continuous across the junction, while the Fourier transform
of the attachment mode must also be available in closed form.
In the spectral domain, the patch part of the higher-order circular attachment mode can be ex-
pressed as

f
AP
na
(k
u
, k
v
)
=
_
jJ
4
(k

b
na
)
k

+
j4J
3
(k

b
na
)
k
2

b
na

jJ
0
(k

a
na
)
k

+
jJ
0
(k

b
na
)
k

__
k
u
k

u +
k
v
k

v
_
=
_
j16J
1
(k

b
na
)
k
4

b
3
na
+
j4J
1
(k

b
na
)
k
2

b
na
+
j8J
0
(k

b
na
)
k
2

b
2
na

jJ
0
(k

a
na
)
k

__
k
u
k

u +
k
v
k

v
_
, z = z
A
na
,
(3.108)
University of PretoriaElectrical, Electronic and Computer Engineering 63
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
where J
3
() is the Bessel function of the rst kind of order 3 and J
4
() is the Bessel function of
the rst kind of order 4. In order to avoid numerical overow problems for very small values of
k

, it is advisable to use the rst form in (3.108) together with the small-argument forms of J
3
()
and J
4
(). The terms associated with J
3
() and J
4
() can then be approximated as
jJ
4
(k

b
na
)
k

+
j4J
3
(k

b
na
)
k
2

b
na
k

jk
3

b
4
na
384
+
jk

b
2
na
12
. (3.109)
As with the other basis functions, the derivation of (3.108) can be found in Appendix B.
The complete set of basis functions has now been dened. In the following section, it will be
shown how the system of linear independent equations can be constructed by making use of the
dyadic Greens function and various vector basis and testing functions.
3.7 SYSTEM OF LINEAR EQUATIONS
In Section 3.2, it has been shown that the system of linear equations can be represented in matrix
notation as
[Z
m,n
] {I
n
} = {V
m
}, (3.110)
where the subscript m is associated with the testing functions and the subscript n is associated
with the basis functions. This system can be expanded as
_

_
_
Z
PZ,PZ
mp:mz,np:nz
_
Z
PZ,A
mp:mz,na
_
Z
PZ,SU
mp:mz,ns:nu
_
Z
PZ,SV
mp:mz,ns:nv

_
Z
A,PZ
ma,np:nz
_
Z
A,A
ma,na
_
Z
A,SU
ma,ns:nu
_
Z
A,SV
ma,ns:nv

_
Z
SU,PZ
ms:mu,np:nz
_
Z
SU,A
ms:mu,na
_
Z
SU,SU
ms:mu,ns:nu
_
Z
SU,SV
ms:mu,ns:nv

_
Z
SV,PZ
ms:mv,np:nz
_
Z
SV,A
ms:mv,na
_
Z
SV,SU
ms:mv,ns:nu
_
Z
SV,SV
ms:mv,ns:nv

_
Z
EU,PZ
me:mu,np:nz
_
Z
EU,A
me:mu,na
_
Z
EU,SU
me:mu,ns:nu
_
Z
EU,SV
me:mu,ns:nv

_
Z
EV,PZ
me:mv,np:nz
_
Z
EV,A
me:mv,na
_
Z
EV,SU
me:mv,ns:nu
_
Z
EV,SV
me:mv,ns:nv

_
Z
PZ,EU
mp:mz,ne:nu
_
Z
PZ,EV
mp:mz,ne:nv

_
Z
A,EU
ma,ne:nu
_
Z
A,EV
ma,ne:nv

_
Z
SU,EU
ms:mu,ne:nu
_
Z
SU,EV
ms:mu,ne:nv

_
Z
SV,EU
ms:mv,ne:nu
_
Z
SV,EV
ms:mv,ne:nv

_
Z
EU,EU
me:mu,ne:nu
_
Z
EU,EV
me:mu,ne:nv

_
Z
EV,EU
me:mv,ne:nu
_
Z
EV,EV
me:mv,ne:nv

_
_
I
PZ
np:nz
_
_
I
A
na
_
_
I
SU
ns:nu
_
_
I
SV
ns:nv
_
_
I
EU
ne:nu
_
_
I
EV
ne:nv
_
_

_
=
_

_
_
V
PZ
mp:mz
_
_
0
_
_
0
_
_
0
_
_
0
_
_
0
_
_

_
(3.111)
University of PretoriaElectrical, Electronic and Computer Engineering 64
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
by inserting the six different types of basis functions, as dened in (3.64), into (3.110). As can
be seen, the interaction matrix consists of 36 submatrices, where the entries in each submatrix de-
scribe the interaction between a specic type of basis function and testing function. For example,
in (3.111), Z
SU,EV
ms:mu,ne:nv
refers to the interaction between the mu-th u-directed subdomain rooftop
testing function on the ms-th rectangular capacitor patch and the nv-th v-directed entire-domain
sinusoidal basis function on the ne-th rectangular resonant patch. Due to the fact that the Galerkin
method is used (i.e. the testing functions and basis functions are chosen to be the same set of
functions), the interaction matrix in (3.111) is symmetric and therefore only slightly more than
half of the total number of entries in the matrix have to be calculated. More specically, if N is
the total number of basis functions, only (N
2
+ N)/2 of the matrix entries have to be calculated,
as opposed to N
2
if the matrix were not symmetric.
The submatrix elements in (3.110) that have to be calculated, are given by
Z
PZ,PZ
mp:mz,np:nz
=

f
PZ
mp:mz
, E
scat
tan
(f
PZ
np:nz
)
_
,
mp = 1, 2, . . . , N
P
, mz = 1, 2, . . . , N
PZ
mp
, np = 1, 2, . . . , N
P
, nz = 1, 2, . . . , N
PZ
np
, (3.112)
Z
PZ,A
mp:mz,na
=

f
PZ
mp:mz
, E
scat
tan
(f
AZ
na
+f
AP
na
)
_
,
mp = 1, 2, . . . , N
P
, mz = 1, 2, . . . , N
PZ
mp
, na = 1, 2, . . . , N
A
, (3.113)
Z
A,A
ma,na
=

f
AZ
ma
+f
AP
ma
, E
scat
tan
(f
AZ
na
+f
AP
na
)
_
,
ma = 1, 2, . . . , N
A
, na = 1, 2, . . . , N
A
, (3.114)
Z
PZ,SU
mp:mz,ns:nu
=

f
PZ
mp:mz
, E
scat
tan
(f
SU
ns:nu
)
_
,
mp = 1, 2, . . . , N
P
, mz = 1, 2, . . . , N
PZ
mp
, ns = 1, 2, . . . , N
S
, nu = 1, 2, . . . , N
SU
ns
, (3.115)
Z
A,SU
ma,ns:nu
=

f
AZ
ma
+f
AP
ma
, E
scat
tan
(f
SU
ns:nu
)
_
,
ma = 1, 2, . . . , N
A
, ns = 1, 2, . . . , N
S
, nu = 1, 2, . . . , N
SU
ns
, (3.116)
Z
SU,SU
ms:mu,ns:nu
=

f
SU
ms:mu
, E
scat
tan
(f
SU
ns:nu
)
_
,
ms = 1, 2, . . . , N
S
, mu = 1, 2, . . . , N
SU
ms
, ns = 1, 2, . . . , N
S
, nu = 1, 2, . . . , N
SU
ns
, (3.117)
Z
PZ,SV
mp:mz,ns:nv
=

f
PZ
mp:mz
, E
scat
tan
(f
SV
ns:nv
)
_
,
mp = 1, 2, . . . , N
P
, mz = 1, 2, . . . , N
PZ
mp
, ns = 1, 2, . . . , N
S
, nv = 1, 2, . . . , N
SV
ns
, (3.118)
University of PretoriaElectrical, Electronic and Computer Engineering 65
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
Z
A,SV
ma,ns:nv
=

f
AZ
ma
+f
AP
ma
, E
scat
tan
(f
SV
ns:nv
)
_
,
ma = 1, 2, . . . , N
A
, ns = 1, 2, . . . , N
S
, nv = 1, 2, . . . , N
SV
ns
, (3.119)
Z
SU,SV
ms:mu,ns:nv
=

f
SU
ms:mu
, E
scat
tan
(f
SV
ns:nv
)
_
,
ms = 1, 2, . . . , N
S
, mu = 1, 2, . . . , N
SU
ms
, ns = 1, 2, . . . , N
S
, nv = 1, 2, . . . , N
SV
ns
, (3.120)
Z
SV,SV
ms:mv,ns:nv
=

f
SV
ms:mv
, E
scat
tan
(f
SV
ns:nv
)
_
,
ms = 1, 2, . . . , N
S
, mv = 1, 2, . . . , N
SV
ms
, ns = 1, 2, . . . , N
S
, nv = 1, 2, . . . , N
SV
ns
, (3.121)
Z
PZ,EU
mp:mz,ne:nu
=

f
PZ
mp:mz
, E
scat
tan
(f
EU
ne:nu
)
_
,
mp = 1, 2, . . . , N
P
, mz = 1, 2, . . . , N
PZ
mp
, ne = 1, 2, . . . , N
E
, nu = 1, 2, . . . , N
EU
ne
, (3.122)
Z
A,EU
na,me:mu
=

f
AZ
ma
+f
AP
ma
, E
scat
tan
(f
EU
ne:nu
)
_
,
ma = 1, 2, . . . , N
A
, ne = 1, 2, . . . , N
E
, nu = 1, 2, . . . , N
EU
ne
, (3.123)
Z
SU,EU
ms:mu,ne:nu
=

f
SU
ms:mu
, E
scat
tan
(f
EU
ne:nu
)
_
,
ms = 1, 2, . . . , N
S
, mu = 1, 2, . . . , N
SU
ms
, ne = 1, 2, . . . , N
E
, nu = 1, 2, . . . , N
EU
ne
, (3.124)
Z
SV,EU
ms:mv,ne:nu
=

f
SV
ms:mv
, E
scat
tan
(f
EU
ne:nu
)
_
,
ms = 1, 2, . . . , N
S
, mv = 1, 2, . . . , N
SV
ms
, ne = 1, 2, . . . , N
E
, nu = 1, 2, . . . , N
EU
ne
, (3.125)
Z
EU,EU
me:mu,ne:nu
=

f
EU
me:mu
, E
scat
tan
(f
EU
ne:nu
)
_
,
me = 1, 2, . . . , N
E
, mu = 1, 2, . . . , N
EU
me
, ne = 1, 2, . . . , N
E
, nu = 1, 2, . . . , N
EU
ne
, (3.126)
Z
PZ,EV
mp:mz,ne:nv
=

f
PZ
mp:mz
, E
scat
tan
(f
EV
ne:nv
)
_
,
mp = 1, 2, . . . , N
P
, mz = 1, 2, . . . , N
PZ
mp
, ne = 1, 2, . . . , N
E
, nv = 1, 2, . . . , N
EV
ne
, (3.127)
Z
A,EV
ma,ne:nv
=

f
AZ
ma
+f
AP
ma
, E
scat
tan
(f
EV
ne:nv
)
_
,
ma = 1, 2, . . . , N
A
, ne = 1, 2, . . . , N
E
, nv = 1, 2, . . . , N
EV
ne
, (3.128)
University of PretoriaElectrical, Electronic and Computer Engineering 66
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
Z
SU,EV
ms:mu,ne:nv
=

f
SU
ms:mu
, E
scat
tan
(f
EV
ne:nv
)
_
,
ms = 1, 2, . . . , N
S
, mu = 1, 2, . . . , N
SU
ms
, ne = 1, 2, . . . , N
E
, nv = 1, 2, . . . , N
EV
ne
, (3.129)
Z
SV,EV
ms:mv,ne:nv
=

f
SV
ms:mv
, E
scat
tan
(f
EV
ne:nv
)
_
,
ms = 1, 2, . . . , N
S
, mv = 1, 2, . . . , N
SV
ms
, ne = 1, 2, . . . , N
E
, nv = 1, 2, . . . , N
EV
ne
, (3.130)
Z
EU,EV
me:mu,ne:nv
=

f
EU
me:mu
, E
scat
tan
(f
EV
ne:nv
)
_
,
me = 1, 2, . . . , N
E
, mu = 1, 2, . . . , N
EU
me
, ne = 1, 2, . . . , N
E
, nv = 1, 2, . . . , N
EV
ne
(3.131)
and
Z
EV,EV
me:mv,ne:nv
=

f
EV
me:mv
, E
scat
tan
(f
EV
ne:nv
)
_
,
me = 1, 2, . . . , N
E
, mv = 1, 2, . . . , N
EV
me
, ne = 1, 2, . . . , N
E
, nv = 1, 2, . . . , N
EV
ne
. (3.132)
The remaining submatrix elements can be found by making use of symmetry. Also, note that the
diagonal submatrices of the interaction matrix are themselves symmetric and therefore some of
their entries can also be found by making use of symmetry within the submatrix.
The excitation vector only has one subvector to calculate. Its non-zero elements can be expressed
as
V
PZ
mp:mz
=

f
PZ
mp:mz
, E
inc
tan
_
, mp = 1, 2, . . . , N
P
, mz = 1. (3.133)
The actual evaluation of the interaction-matrix entries accounts for most of the computational
burden and will now be addressed in the section that follows.
3.8 EVALUATION OF THE INTERACTION-MATRIX ENTRIES
As is evident from (3.22), the evaluation of each entry in the interaction matrix requires integration
over the k
x
and k
y
wavenumbers, as well as possible integration over the z and z

variables. The
integrations over z and z

can be done analytically, but the integrations over k

x
and k
y
require
numerical integration in the complex plane, where special care has to be taken with respect to
poles, branch points and branch cuts.
As will be explained later, it is more convenient to follow the approach in [164168] and to dene
a new coordinate system, as shown in Figure 3.12.
4
This new coordinate system is represented by
the and axes, which is chosen in such a way that the axis starts from the centre of the testing
4
The new coordinate system was previously used to calculate the interaction between basis and testing functions that
are widely separated [164168]. Here, it is used for all interactions.
University of PretoriaElectrical, Electronic and Computer Engineering 67
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation

x
y

u
v
u
v

n
Testing function f
m
Basis function f
n

m
,
n
Figure 3.12 Coordinate system used for the evaluation of the interaction-matrix entries.
function and runs through the centre of the basis function. Between the two functions, there is
therefore a displacement of
m,n
along the axis, while there is no displacement along the
axis. In the rest of this section, it will now be shown how the expression in (3.22) can be cast into
a new form so that the integration is performed over the k

and k

wavenumbers, as opposed to
the k
x
and k
y
wavenumbers. The benet of this change in variables is explained in Section 3.9,
which deals with integration strategies.
In the new coordinate system of Figure 3.12, the interaction-matrix entries can be expressed as
Z
m,n
=
1
4
2
_
k

_
k

_
z
_
z

f
m
(k

, k

, z)

G(k

, k

, z|z

f
n
(k

, k

, z

) dz

dz dk

dk

,
(3.134)
where the expressions for

G(k

, k

, z|z

),

f
n
(k

, k

, z

) and

f
m
(k

, k

, z) still have to be deter-

mined.
The expression for

G(k

, k

, z|z

) is very straightforward to derive. It can simply be done by

rotating the x and y axes until the x axis coincides with the axis. Now, in terms of its individual
components, the dyadic Greens function can be expressed as

G =

G
z
z +

G
z
z + z

G
z

+ z

G
z

+ z

G
zz
z,
(3.135)
where the individual components are given by

(k

, k

, z|z

) =

G
xx
(k
x
= k

, k
y
= k

, z|z

), (3.136)
University of PretoriaElectrical, Electronic and Computer Engineering 68
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation

(k

, k

, z|z

) =

G
xy
(k
x
= k

, k
y
= k

, z|z

), (3.137)

G
z
(k

, k

, z|z

) =

G
xz
(k
x
= k

, k
y
= k

, z|z

), (3.138)

(k

, k

, z|z

) =

G
yx
(k
x
= k

, k
y
= k

, z|z

), (3.139)

(k

, k

, z|z

) =

G
yy
(k
x
= k

, k
y
= k

, z|z

), (3.140)

G
z
(k

, k

, z|z

) =

G
yz
(k
x
= k

, k
y
= k

, z|z

), (3.141)

G
z
(k

, k

, z|z

) =

G
zx
(k
x
= k

, k
y
= k

, z|z

), (3.142)

G
z
(k

, k

, z|z

) =

G
zy
(k
x
= k

, k
y
= k

, z|z

) (3.143)
and

G
zz
(k

, k

, z|z

) =

G
zz
(k
x
= k

, k
y
= k

, z|z

). (3.144)
The vector basis function

f
n
(k

, k

, z

) can be expressed in terms of its three components as

f
n
(k

, k

, z

) =
_

(k

, k

, z

)
_
n

+
_

(k

, k

, z

)
_
n

+
_

f
z
(k

, k

, z

)
_
n
z, (3.145)
where
_

(k

, k

, z

)
_
n
=
_

f
u
_
k
u
= k

cos(
n
) +k

sin(
n
), k
v
= k

sin(
n
) +k

cos(
n
), z

__
n
cos(
n
) e
jk

m,n
+
_

f
v
_
k
u
= k

cos(
n
) +k

sin(
n
), k
v
= k

sin(
n
) +k

cos(
n
), z

__
n
sin(
n
) e
jk

m,n
, (3.146)
_

(k

, k

, z

)
_
n
=
_

f
u
_
k
u
= k

cos(
n
) +k

sin(
n
), k
v
= k

sin(
n
) +k

cos(
n
), z

__
n
sin(
n
) e
jk

m,n
+
_

f
v
_
k
u
= k

cos(
n
) +k

sin(
n
), k
v
= k

sin(
n
) +k

cos(
n
), z

__
n
cos(
n
) e
jk

m,n
(3.147)
and
_

f
z
(k

, k

, z

)
_
n
=
_

f
z
(k
u
= k

, k
v
= k

, z

)
_
n
e
jk

m,n
. (3.148)
Here,
n
is the angle from the u axis (associated with the basis function) to the axis. It can be
calculated as

n
=
n
, (3.149)
where
n
is the rotation angle, of the local coordinate system of the testing function, with respect
University of PretoriaElectrical, Electronic and Computer Engineering 69
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
to the (x, y) coordinate system. Furthermore, is the rotation angle of the (, ) coordinate
system with respect to the (x, y) coordinate system. All these angles are illustrated in Figure 3.12.
Through (3.146) and (3.147), the two-dimensional Fourier transform of the testing function is
basically rotated through an angle of (
n
).
5
Furthermore, the expressions in (3.151) to (3.153)
all contain a e
jk

m,n
term due to the
m,n
displacement of the testing function along the
axis.
In a similarly way, the vector testing function

f
m
(k

, k

, z) can be expressed in terms of its three

components as

f
m
(k

, k

, z) =
_

(k

, k

, z)
_
m

+
_

(k

, k

, z)
_
m

+
_

f
z
(k

, k

, z)
_
m
z, (3.150)
where
_

(k

, k

, z)
_
m
=
_

f
u
_
k
u
= k

cos(
m
) +k

sin(
m
), k
v
= k

sin(
m
) +k

cos(
m
), z
__
m
cos(
m
)
+
_

f
v
_
k
u
= k

cos(
m
) +k

sin(
m
), k
v
= k

sin(
m
) +k

cos(
m
), z
__
m
sin(
m
), (3.151)
_

(k

, k

, z)
_
m
=
_

f
u
_
k
u
= k

cos(
m
) +k

sin(
m
), k
v
= k

sin(
m
) +k

cos(
m
), z
__
m
sin(
m
)
+
_

f
v
_
k
u
= k

cos(
m
) +k

sin(
m
), k
v
= k

sin(
m
) +k

cos(
m
), z
__
m
cos(
m
) (3.152)
and
_

f
z
(k

, k

, z)
_
m
=
_

f
z
(k
u
= k

, k
v
= k

, z)
_
m
. (3.153)
In this case,
m
is the angle from the u axis (associated with the testing function) to the axis. It
can be calculated as

m
=
m
, (3.154)
where
m
is the rotation angle, of the local coordinate system of the testing function, with respect
to the (x, y) coordinate system.
In general then, for basis function n and testing function m, both with , and z components, the
5
If a two-dimensional function is rotated through an angle (
n
), its two-dimensional Fourier transform is also rotated
through (
n
) [180].
University of PretoriaElectrical, Electronic and Computer Engineering 70
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
interaction between them can be expressed as
Z
m,n
=
1
4
2
_
k

_
k

_
_

(k

, k

, z)
_
m

(k

, k

, z|z

)
_

(k

, k

, z

)
_
n
+
_

(k

, k

, z)
_
m

(k

, k

, z|z

)
_

(k

, k

, z

)
_
n
+
_

(k

, k

, z)
_
m

(k

, k

, z|z

)
_

(k

, k

, z

)
_
n
+
_

(k

, k

, z)
_
m

(k

, k

, z|z

)
_

(k

, k

, z

)
_
n
+
_

f
z
(k

, k

)
_
m

G
I
z
(k

, k

, z

)
_

(k

, k

, z

)
_
n
+
_

f
z
(k

, k

)
_
m

G
I
z
(k

, k

, z

)
_

(k

, k

, z

)
_
n
+
_

f
z
(k

, k

)
_
m

G
II
zz
(k

, k

)
_

f
z
(k

, k

)
_
n
_
dk

dk

, (3.155)
where

G
I
z
(k

, k

, z

) =
_
z
_
f
z
(z)
_
m

G
z
(k

, k

, z|z

) dz

, (3.156)

G
I
z
(k

, k

, z

) =
_
z
_
f
z
(z)
_
m

G
z
(k

, k

, z|z

) dz

(3.157)
and

G
II
zz
(k

, k

) =
_
z
_
z

_
f
z
(z)
_
m

G
z
(k

, k

, z|z

)
_
f
z
(z

)
_
n
dz

dz. (3.158)
The expressions in (3.156) to (3.158) will be referred to as expanded Greens-function components
for the purposes of this analysis. This is due to the fact that they contain integrations that have to
be performed over z and/or z

. In Appendix A, it is shown how they can be evaluated.

The numerical integration over k

and k

in (3.155) is a rather formidable task and will now be

addressed in the section that follows.
3.9 INTEGRATION STRATEGIES
Fast numerical integration methods, which are required for the efcient evaluation of the entries
in the interaction matrix, have been and still seem to be the topic of intensive research. This is
understandable when realising that most of the computational time goes into the evaluation of
these integrals, especially in the spectral domain.
For the purposes of this study, a few different integration strategies have been implemented. Their
usage depends on the separation distance between basis and testing functions, as well as on types
of basis and testing functions that are used. There is a common integration strategy that is used
for all cases where there is overlap of the basis and testing functions, while two different integra-
University of PretoriaElectrical, Electronic and Computer Engineering 71
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
tion strategies were implemented for cases where the basis and testing functions are completely
separated. The usage of the latter two depends on whether the basis and testing functions are
axisymmetric (i.e. no angular variation) or not. Each of the integration strategies will now be
considered in more detail.
3.9.1 Integration Strategy for Overlapping Basis and Testing Functions
A general entry of the interaction matrix can be expressed in a compact form as
Z
m,n
=
1
4
2
_

m,n
(k

, k

) e
jk

m,n
dk

dk

, (3.159)
where
m,n
(k

, k

) is a smooth function and e

jk

m,n
is the shift exponential due to the separa-
tion between the basis and testing functions. The integrals in (3.159) can be simplied by changing
to polar coordinates. This is facilitated by setting
k

= k

cos() (3.160)
and
k

= k

sin(). (3.161)
Here, k

is the wavenumber in a radial direction and is the angular variable. By making this
change of variables, the expression in (3.159) can now be written as
Z
m,n
=
1
4
2
_
2
0
_

0

m,n
(k

, ) e
jk

m,n
cos()
k

dk

d. (3.162)
As can be seen, the double innite integrals in (3.159) have now been replaced by one nite inte-
gral and one semi-innite integral. For numerical integration purposes, this formof the interaction-
matrix entries, is usually preferred over the form in (3.159) [123, 144].
The nite integration over in (3.162) can usually be performed without any difculty, provided
that the integrand is not very oscillatory. However, the integration over k

is much more involved

and requires a very careful treatment. This is due to poles, branch points and branch cuts in the
complex k

plane. They are associated with the multilayered Greens function and appear in the
fourth quadrant of the complex k

plane (they also appear in the second quadrant, but these are
not of importance here), as shown in Figure 3.13. In the lossless case, the singularities move onto
the real k

axis. They are however bounded by two values on the real axis, k
,1
and k
,2
. These
are given by [95, 178]
k
,1
= k
0
_
min
_

r(0)

r(0)
,
r(1)

r(1)
, . . . ,
r(T)

r(T)

(3.163)
and
k
,2
= k
0
_
max
_

r(0)

r(0)
,
r(1)

r(1)
, . . . ,
r(T)

r(T)

. (3.164)
University of PretoriaElectrical, Electronic and Computer Engineering 72
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
e
_
k

_
m
_
k

_
k
,1
2k
,2
k
,max
k
,1

,1

,2

,3
Branch cut
Branch point
Pole
Figure 3.13 Integration path in the complex k

plane for overlapping basis

and testing functions.
In order to avoid the singularities on the real k

axis, the integration contour is deformed into

the rst quadrant of the complex k

plane [61, 151], as shown in Figure 3.13. The integration

contour is divided into three parts:
,1
,
,2
and
,3
. In theory,
,3
runs along the real k

axis
up to innity, but for the purposes of numerical integration, it has to be terminated at some value
k
,max
. The value of k
,max
depends on the basis and testing functions that are used. Its value
for the different combinations of basis and testing functions has been found through extensive
numerical experimentation and is presented in Table 3.1. The deformation of the integration path
should be large enough to avoid the singularities, but also not too large, since the integrand grows
exponentially away from the real axis. Also, the evaluation of Bessel functions with large complex
arguments, becomes very difcult [181, 182]. Another possibility is not to deform the integration
path at all, but to use residue extraction at the singularities [146,183]. However, a thick substrate is
usually associated with many surface-waves poles and, as such, the pole-tracking can be become
extremely tedious. Therefore, this method will not be considered here.
The numerical integration along each part of the k

integration contour is performed through

Gaussian quadrature. The three parts of the integration contour,
,1
,
,2
and
,3
, are divided
into I

,1
, I

,2
and I

,3
intervals respectively. Subsequently, N

,1
, N

,2
and N

,3
point Gauss-
Legendre quadrature formulas are then applied to each interval on the three parts of the integration
contour. Similarly, the integration contour is divided into I

intervals with N

point Gaussian-
Legendre quadrature applied to each interval. Once again, the number of integration intervals
and points required for the Gaussian quadrature, has also been determined through numerical
experimentation, and is presented in Table 3.2.
The present integration method only works well for basis and testing functions that overlap or
University of PretoriaElectrical, Electronic and Computer Engineering 73
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
Table 3.1
Value of k
,max
for (k

, ) integration.
Interaction k
,max
Z
PZ,PZ
mp:mz,np:nz
max
_
min
_
80/
_
max(a
mp
, a
np
) max(z
mz
, z
nz
), 320/|z
mp
z
np
|

, 4k
,2
_
Z
PZ,A
mp:mz,na
max
_
min
_
80/
_
max(a
ma
, a
np
) max(z
ma
, z
nz
), 80/|z
mp
z
na
|

, 4k
,2
_
Z
A,A
ma,na
max
_
500/
_
max(a
ma
, a
np
), 4k
,2
_
Z
PZ,SU
mp:mz,ns:nu
max
_
min
_
min
_
80/
_
a
mp
z
mp
, 20/l
ns
_
, 20/|z
mp
z
ns
|

, 4k
,2
_
Z
A,SU
ma,ns:nu
max
_
min
_
min
_
500/

a
ma
, 20/l
ns
_
, 20/|z
ma
z
ns
|

, 4k
,2
_
Z
SU,SU
ms:mu,ns:nu
max
_
min
_
20/ min(l
ms
, l
ns
), 20/|z
ms
z
ns
|

, 4k
,2
_
Z
PZ,SV
mp:mz,ns:nv
max
_
min
_
min
_
80/
_
a
mp
z
mp
, 20/w
ns
_
, 20/|z
mp
z
ns
|

, 4k
,2
_
Z
A,SV
ma,ns:nv
max
_
min
_
min
_
500/

a
ma
, 20/w
ns
_
, 20/|z
ma
z
ns
|

, 4k
,2
_
Z
SU,SV
ms:mu,ns:nv
max
_
min
_
20/ min(l
ms
, l
ns
, w
ms
, w
ns
), 20/|z
ms
z
ns
|

, 4k
,2
_
Z
SV,SV
ms:mv,ns:nv
max
_
min
_
20/ min(w
ms
, w
ns
), 20/|z
ms
z
ns
|

, 4k
,2
_
Z
PZ,EU
mp:mz,ne:nu
max
_
min
_
min
_
40/
_
a
mp
z
mp
,
_
(p
EU
ne:nu
/L
ne
)
2
+ (q
EU
ne:nu
/W
ne
)
2
+10
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
mp
z
ne
|

, 4k
,2
_
Z
A,EU
na,me:mu
max
_
min
_
min
_
500/

a
ma
,
_
(p
EU
ne:nu
/L
ne
)
2
+ (q
EU
ne:nu
/W
ne
)
2
+10
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
ma
z
ne
|

, 4k
,2
_
Z
SU,EU
ms:mu,ne:nu
max
_
min
_
min
_
10/l
ms
,
_
(p
EU
ne:nu
/L
ne
)
2
+ (q
EU
ne:nu
/W
ne
)
2
+10
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
ms
z
ne
|

, 4k
,2
_
Z
SV,EU
ms:mv,ne:nu
max
_
min
_
min
_
10/w
ms
,
_
(p
EU
ne:nu
/L
ne
)
2
+ (q
EU
ne:nu
/W
ne
)
2
+10
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
ms
z
ne
|

, 4k
,2
_
Z
EU,EU
me:mu,ne:nu
max
_
min
_
max
__
(p
EU
me:mu
/L
ne
)
2
+ (q
EU
me:mu
/W
ne
)
2
,
_
(p
EU
ne:nu
/L
ne
)
2
+ (q
EU
ne:nu
/W
ne
)
2
_
+10
_
(1/L
ne
)
2
+ (1/W
ne
)
2
, 20/|z
me
z
ne
|

, 4k
,2
_
Z
PZ,EV
mp:mz,ne:nv
max
_
min
_
min
_
40/
_
a
mp
z
mp
,
_
(p
EV
ne:nv
/W
ne
)
2
+ (q
EV
ne:nv
/L
ne
)
2
+10
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
mp
z
ne
|

, 4k
,2
_
Z
A,EV
na,me:mv
max
_
min
_
min
_
500/

a
ma
,
_
(p
EV
ne:nv
/W
ne
)
2
+ (q
EV
ne:nv
/L
ne
)
2
+10
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
ma
z
ne
|

, 4k
,2
_
Z
SU,EV
ms:mu,ne:nv
max
_
min
_
min
_
10/l
ms
,
_
(p
EV
ne:nv
/W
ne
)
2
+ (q
EV
ne:nv
/L
ne
)
2
+10
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
ms
z
ne
|

, 4k
,2
_
Z
SV,EV
ms:mv,ne:nv
max
_
min
_
min
_
10/w
ms
,
_
(p
EV
ne:nv
/W
ne
)
2
+ (q
EV
ne:nv
/L
ne
)
2
+10
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
ms
z
ne
|

, 4k
,2
_
Z
EU,EV
me:mu,ne:nv
max
_
min
_
max
__
(p
EU
me:mu
/L
ne
)
2
+ (q
EU
me:mu
/W
ne
)
2
,
_
(p
EV
ne:nv
/W
ne
)
2
+ (q
EV
ne:nv
/L
ne
)
2
_
+10
_
(1/L
ne
)
2
+ (1/W
ne
)
2
, 20/|z
me
z
ne
|

, 4k
,2
_
Z
EV,EV
me:mv,ne:nv
max
_
min
_
max
__
(p
EV
me:mv
/W
ne
)
2
+ (q
EV
me:mv
/L
ne
)
2
,
_
(p
EV
ne:nv
/W
ne
)
2
+ (q
EV
ne:nv
/L
ne
)
2
_
+10
_
(1/L
ne
)
2
+ (1/W
ne
)
2
, 20/|z
me
z
ne
|

, 4k
,2
_
Note: the terms containing |z
m
z
n
| should only be used if |z
m
z
n
| = 0.
University of PretoriaElectrical, Electronic and Computer Engineering 74
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
Table 3.2
Integration intervals and points for (k

, ) integration.
Interaction I

,1
I

,2
I

,3
N

,1
N

,2
N

,3
Z
PZ,PZ
mp:mz,np:nz
n/a

n/a

1 4 4
_
k
,max
a
max

+
1
4
_
4 4 12
Z
PZ,A
mp:mz,na
n/a

n/a

1 4 4
_
k
,max
a
max

+
1
4
_
4 4 12
Z
A,A
ma,na
n/a

n/a

1 4 4
_
k
,max
a
max

+
1
4
_
4 4 12
Z
PZ,SU
mp:mz,ns:nu
8 4 1 4
_
k
,max
l
ns
2
_
4 4 12
Z
A,SU
ma,ns:nu
8 8 1 4
_
k
,max
l
ns
2
_
4 4 20
Z
SU,SU
ms:mu,ns:nu
8 1 1 4
_
k
,max
l
max
2
_
4 4 12
Z
PZ,SV
mp:mz,ns:nv
8 4 1 4
_
k
,max
w
ns
2
_
4 4 12
Z
A,SV
ma,ns:nv
8 8 1 4
_
k
,max
w
ns
2
_
4 4 20
Z
SU,SV
ms:mu,ns:nv
8 8 1 4
_
k
,max
lw
max
2
_
4 4 12
Z
SV,SV
ms:mv,ns:nv
8 8 1 4
_
k
,max
w
max
2
_
4 4 12
Z
PZ,EU
mp:mz,ne:nu
max(pq
EU
n
, 8) 4 1 4
_
k
,max
L
ne
2
_
4 4 12
Z
A,EU
na,me:mu
max(pq
EU
n
, 8) 4 1 4
_
k
,max
L
ne
2
_
4 4 12
Z
SU,EU
ms:mu,ne:nu
max(pq
EU
n
, 8) 4 1 4
_
k
,max
L
ne
2
_
4 4 12
Z
SV,EU
ms:mv,ne:nu
max(pq
EU
n
, 8) 4 1 4
_
k
,max
L
ne
2
_
4 4 12
Z
EU,EU
me:mu,ne:nu
max(pq
EU
m
, pq
EU
n
, 8) 4 1 4
_
k
,max
L
max
2
_
4 4 12
Z
PZ,EV
mp:mz,ne:nv
max(pq
EV
n
, 8) 4 1 4
_
k
,max
W
ne
2
_
4 4 12
Z
A,EV
na,me:mv
max(pq
EV
n
, 8) 4 1 4
_
k
,max
W
ne
2
_
4 4 12
Z
SU,EV
ms:mu,ne:nv
max(pq
EV
n
, 8) 4 1 4
_
k
,max
W
ne
2
_
4 4 12
Z
SV,EV
ms:mv,ne:nv
max(pq
EV
n
, 8) 4 1 4
_
k
,max
W
ne
2
_
4 4 12
Z
EU,EV
me:mu,ne:nv
max(pq
EU
m
, pq
EV
n
, 8) 4 1 4
_
k
,max
LW
max
2
_
4 4 12
Z
EV,EV
me:mv,ne:nv
max(pq
EV
m
, pq
EV
n
, 8) 4 1 4
_
k
,max
W
max
2
_
4 4 12
Note:
pq
EU
m
= 2(p
EU
me:mu
+q
EU
me:mu
) and pq
EU
n
= 2(p
EU
ne:nu
+q
EU
ne:nu
).
pq
EV
m
= 2(p
EV
me:mv
+q
EV
me:mv
) and pq
EV
n
= 2(p
EV
ne:nv
+q
EV
ne:nv
).
a
max
= max(a
mp
, a
np
, a
ma
, a
na
) where applicable.
l
max
= max(l
ms
, l
ns
) and w
max
= max(w
ms
, w
ns
).
lw
max
= max(l
ms
, l
ns
, w
ms
, w
ns
).
L
max
= max(L
me
, L
ne
) and W
max
= max(W
me
, W
ne
).
LW
max
= max(L
me
, L
ne
, W
me
, W
ne
).

due to symmetry, no integration is required for axisymmetric basis and testing functions.
University of PretoriaElectrical, Electronic and Computer Engineering 75
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
where the separation between them is very small. The method becomes more and more inefcient
as the basis and testing functions are moved further apart. To illustrate this, consider the behaviour
of the two integrands in Figures 3.14(a) and (b). The integrand in Figure 3.14(a) is for the inter-
action between basis and testing functions with no separation between them, while the integrand
in Figure 3.14(b) is for the interaction between basis and testing functions with a nite separation
distance between them. It can be seen that the integrand in Figure 3.14(a) is fairly smooth, but
that the shift exponential causes the integrand in Figure 3.14(b) to be oscillatory. These oscilla-
tions become more severe as the separation distance is increased and therefore the present method
becomes less and less efcient. In order to efciently integrate the integrand associated with
widely-separated basis and testing functions, one has to resort to other more appropriate methods.
These will now be discussed in more detail.
3.9.2 Integration Strategy for Separated Basis and Testing Functions
A new technique has recently been developed by Sereno-Garino et al. [164168] for a more ef-
cient analysis of the interaction integrals when the basis and testing functions are widely separated.
It makes use of integration-contour deformation in order to avoid the oscillations that are caused
by the shift exponential.
A general entry of the interaction matrix can once again be expressed in its compact form as
Z
m,n
=
1
4
2
_

m,n
(k

, k

) e
jk

m,n
dk

dk

, (3.165)
where
m,n
(k

, k

) is a smooth function and e

jk

m,n
is the shift exponential due to the sepa-
ration between the basis and testing functions. In this case, however, the integration variables are
not substituted with their polar counterparts, as in the previous section, but are left in rectangular
coordinates. The new technique can best be described by expressing (3.165) as
Z
m,n
=
1
4
2
_

m,n
(k

) e
jk

m,n
dk

, (3.166)
where

m,n
(k

) =
_

m,n
(k

, k

) dk

. (3.167)
Here, it can be seen what benet the new coordinate system in Figure 3.12 adds to the evaluation
of the integrals. The inner integration over k

can be evaluated without much trouble as there is

no separation between the basis and testing functions along the direction. This then in turn also
implies that
m,n
(k

) is a smooth function without any oscillations due to the shift exponential.

The integration over k

can be facilitated by choosing an integration contour over which the shift

exponential decays.
Sereno-Garino et al. suggested that the k

integration contour should stay close to the real k

axis,
University of PretoriaElectrical, Electronic and Computer Engineering 76
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation

e
_
k

m
_

m
,
n
(
k

)
e

j
k

m
,
n
c
o
s
(

)
k

_
0
3750
7500 0

2
0
0.9e-5
1.8e-5
(a)

e
_
k

m
_

m
,
n
(
k

)
e

j
k

m
,
n
c
o
s
(

)
k

_
0
3750
7500 0

2
-2.0e-5
-1.0e-5
0
1.0e-5
2.0e-5
(b)
Figure 3.14 Integrand of Z
m,n
, as a function of e
_
k

_
(but evaluated over the deformed k

integration contour) and , for rooftop basis and testing functions with: l
ns
= w
ns
= l
ms
=
w
ms
= 2 mm,
n
=
m
= = 0, h
(1)
= 1.6 mm, h
(2)
= 15 mm,
r(0)
=
r(2)
= 1,
r(1)
= 4.25,
tan
(0)
= tan
(2)
= 0, tan
(1)
= 0.02 and f = 1.8 GHz. (a) Completely overlapping basis and
testing functions (
m,n
= 0). (b) Separated basis and testing functions with
m,n
= 8 mm.
University of PretoriaElectrical, Electronic and Computer Engineering 77
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
while the k

integration contour should run up along the left-hand side of the negative imaginary k

axis, go through the origin, go around the branch points and poles, and then come down the right-
hand side of the negative imaginary k

axis. This has the wanted effect that the shift exponential
e
jk

m,n
decays to zero along the negative imaginary k

axis.
For the purposes of this study, however, the integration contours as shown in Figure 3.15 were
used instead. Now, before discussing the benet of this choice, it has to be realised that the
k

and k

integration contours cannot just be chosen at will and neither can they be considered
independently. The k

integration contour affects the location of the singularities in the complex

k

plane and therefore these two contours have to be considered simultaneously. The k

integration
contour, as shown in Figure 3.15(b), is basically similar to the one suggested by Sereno-Garino et
al. and stays on the real k

axis, except for small deformations to avoid singularities that would

appear at the origin of the complex k

plane when k
,1
k

k
,2
. The k

integration contour,
as shown in Figure 3.15(a), which consists of four parts,
,1
,
,2
,
,3
and
,4
, mainly differs
from the one suggested by Sereno-Garino et al. in that the
,1
and
,4
parts of the contour do
not approach the negative imaginary k

axis at the start and end points of the contour, but instead
they move away from the negative imaginary k

axis at angle of 45

. The reason for this choice

will soon be explained.
First consider the behaviour of the integrand of Z
m,n
for two separated rooftop basis and testing
functions over parts
,1
,
,2
,
,3
and
,4
, as shown in Figures 3.16(a) to (d). From these gures
it can be seen that the integrand is fairly smooth over k

and that the corresponding integral can

therefore be evaluated without much trouble. In order to motivate the usage of the k

integration
contour in Figure 3.15(a), consider the behaviour of the integrand in Figures 3.17(a) and (b).
The integrand in Figure 3.17(a) is for the interaction between basis and testing functions with a
small separation distance between them, while the integrand in Figure 3.17(b) is for the interaction
between basis and testing functions with a larger separation distance between them. Now, as can be
seen from these two gures, the integrand along the the negative imaginary k

axis is only smooth

for larger separation distances between basis and testing functions, but becomes very ragged for
small separation distances (this behaviour also becomes more apparent for thicker substrates). This
implies that the integration contour, as suggested by Sereno-Garino et al., would be troublesome
for small separation distances. However, in both Figures 3.17(a) and (b), it can be seen that the
k

integration contour in Figure 3.15(a) always result in a smooth integrand, regardless of the
separation distance (except for overlap) between the basis and testing functions.
For numerical integration purposes, the k

integration contour starts at (k

,max
, k
,max
) and
ends at (2k
,2
+ k
,max
, k
,max
), while the k

integration contour starts at (k

,max
, 0) and ends
at (k
,max
, 0). A value of 15/
m,n
for k
,max
has been found to work sufciently well for all com-
binations of basis and testing functions, while the values for k
,max
are summarised in Table 3.3.
These have also been found through extensive numerical experimentation.
University of PretoriaElectrical, Electronic and Computer Engineering 78
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
e
_
k

_
m
_
k

_
k
,1
2k
,2
2k
,2
+k
,max
k
,max
k
,max

,1

,2

,3

,4
0.1k
,1
Branch cut
Branch point
Pole
(a)
e
_
k

_
m
_
k

_
k
,1
2k
,2
k
,max
0.1k
,1
(b)
Figure 3.15 Integration path in the complex k

and k

planes for non-overlapping basis and

testing functions. (a) The k

plane. (b) The k

plane.
University of PretoriaElectrical, Electronic and Computer Engineering 79
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
e
_
k

_
e
_
k

m
_

m
,
n
(
k

,
k

)
e

j
k

m
,
n
_
-1e4
0
1e4 -100
-50
0
-2e-9
0
2e-9
(a)
e
_
k

_
e
_
k

m
_

m
,
n
(
k

,
k

)
e

j
k

m
,
n
_
-1e4
0
1e4 0
20
40
-2e-9
0
2e-9
(b)
e
_
k

_
e
_
k

m
_

m
,
n
(
k

,
k

)
e

j
k

m
,
n
_
40
100
160 -1e4
0
1e4
-2e-9
0
2e-9
(c)
e
_
k

_
e
_
k

m
_

m
,
n
(
k

,
k

)
e

j
k

m
,
n
_
160
210
260 -1e4
0
1e4
-2e-9
0
2e-9
(d)
Figure 3.16 Integrand of Z
m,n
, as a function of e
_
k

_
and e
_
k

_
(but evaluated over the deformed k

integration contour), for rooftop basis and testing functions with: l

ns
= w
ns
= l
ms
= w
ms
= 1 mm,

n
=
m
= 0, = 45

,
m,n
= 113.14 mm, h
(1)
= 1.6 mm, h
(2)
= 15 mm,
r(0)
=
r(2)
= 1,
r(1)
= 4,
tan
(0)
= tan
(1)
= tan
(2)
= 0 and f = 1.91 GHz. (a)
,1
part of k

integration contour. (b)

,2
part of
k

integration contour. (c)

,3
part of k

integration contour. (d)

,4
part of k

integration contour.
University of PretoriaElectrical, Electronic and Computer Engineering 80
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
e
_
k

_
m
_
k

m
_

m
,
n
(
k

,
0
)
e

j
k

m
,
n
_
-1000
0
1000 -1000
-450
100
-1e-9
0
1e-9
(a)
e
_
k

_
m
_
k

m
_

m
,
n
(
k

,
0
)
e

j
k

m
,
n
_
-1000
0
1000 -1000
-450
100
-6e-9
0
8e-9
(b)
Figure 3.17 Integrand of Z
m,n
over the complex k

plane (at k

= 0) for rooftop basis and

testing functions with: l
ns
= w
ns
= l
ms
= w
ms
= 1 mm,
n
=
m
= 0, = 45

,
h
(1)
= 1.6 mm, h
(2)
= 15 mm,
r(0)
=
r(2)
= 1,
r(1)
= 4, tan
(0)
= tan
(1)
= tan
(2)
= 0
and f = 1.91 GHz. (a) Closely-spaced basis and testing functions with
m,n
= 2.83 mm.
(b) Widely-separated basis and testing functions with
m,n
= 21.21 mm.
University of PretoriaElectrical, Electronic and Computer Engineering 81
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
Table 3.3
Value of k
,max
for (k

, k

) integration.
Interaction k
,max
Z
PZ,PZ
mp:mz,np:nz
not applicable
Z
PZ,A
mp:mz,na
not applicable
Z
A,A
ma,na
not applicable
Z
PZ,SU
mp:mz,ns:nu
max
_
min
_
min
_
200/

a
mp
, 4/ min(l
ns
, w
ns
)
_
, 20/|z
mp
z
ns
|

, 4k
,2
_
Z
A,SU
ma,ns:nu
max
_
min
_
min
_
200/

a
mp
, 4/ min(l
ns
, w
ns
)
_
, 20/|z
mp
z
ns
|

, 4k
,2
_
Z
SU,SU
ms:mu,ns:nu
max
_
min
_
4/ min(l
ms
, w
ms
, l
ns
, w
ns
), 20/|z
mp
z
ns
|

, 4k
,2
_
Z
PZ,SV
mp:mz,ns:nv
max
_
min
_
min
_
200/

a
mp
, 4/ min(l
ns
, w
ns
)
_
, 20/|z
mp
z
ns
|

, 4k
,2
_
Z
A,SV
ma,ns:nv
max
_
min
_
min
_
200/

a
mp
, 4/ min(l
ns
, w
ns
)
_
, 20/|z
mp
z
ns
|

, 4k
,2
_
Z
SU,SV
ms:mu,ns:nv
max
_
min
_
4/ min(l
ms
, w
ms
, l
ns
, w
ns
), 20/|z
mp
z
ns
|

, 4k
,2
_
Z
SV,SV
ms:mv,ns:nv
max
_
min
_
4/ min(l
ms
, w
ms
, l
ns
, w
ns
), 20/|z
mp
z
ns
|

, 4k
,2
_
Z
PZ,EU
mp:mz,ne:nu
max
_
min
_
min
_
200/

a
mp
, 2
_
(p
EU
ne:nu
/L
ne
)
2
+ (q
EU
ne:nu
/W
ne
)
2
+20
2
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
mp
z
ne
|

, 4k
,2
_
Z
A,EU
na,me:mu
max
_
min
_
min
_
200/

a
mp
, 2
_
(p
EU
ne:nu
/L
ne
)
2
+ (q
EU
ne:nu
/W
ne
)
2
+20
2
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
ma
z
ne
|

, 4k
,2
_
Z
SU,EU
ms:mu,ne:nu
max
_
min
_
min
_
4/l
ms
, 2
_
(p
EU
ne:nu
/L
ne
)
2
+ (q
EU
ne:nu
/W
ne
)
2
+20
2
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
ms
z
ne
|

, 4k
,2
_
Z
SV,EU
ms:mv,ne:nu
max
_
min
_
min
_
4/w
ms
, 2
_
(p
EU
ne:nu
/L
ne
)
2
+ (q
EU
ne:nu
/W
ne
)
2
+20
2
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
ms
z
ne
|

, 4k
,2
_
Z
EU,EU
me:mu,ne:nu
max
_
min
_
max
_
2
_
(p
EU
me:mu
/L
ne
)
2
+ (q
EU
me:mu
/W
ne
)
2
,
2
_
(p
EU
ne:nu
/L
ne
)
2
+ (q
EU
ne:nu
/W
ne
)
2
_
+20
2
_
(1/L
ne
)
2
+ (1/W
ne
)
2
, 20/|z
me
z
ne
|

, 4k
,2
_
Z
PZ,EV
mp:mz,ne:nv
max
_
min
_
min
_
200/

a
mp
, 2
_
(p
EV
ne:nv
/W
ne
)
2
+ (q
EV
ne:nv
/L
ne
)
2
+20
2
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
mp
z
ne
|

, 4k
,2
_
Z
A,EV
na,me:mv
max
_
min
_
min
_
200/

a
mp
, 2
_
(p
EV
ne:nv
/W
ne
)
2
+ (q
EV
ne:nv
/L
ne
)
2
+20
2
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
ma
z
ne
|

, 4k
,2
_
Z
SU,EV
ms:mu,ne:nv
max
_
min
_
min
_
4/l
ms
, 2
_
(p
EV
ne:nv
/W
ne
)
2
+ (q
EV
ne:nv
/L
ne
)
2
+20
2
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
ms
z
ne
|

, 4k
,2
_
Z
SV,EV
ms:mv,ne:nv
max
_
min
_
min
_
4/w
ms
, 2
_
(p
EV
ne:nv
/W
ne
)
2
+ (q
EV
ne:nv
/L
ne
)
2
+20
2
_
(1/L
ne
)
2
+ (1/W
ne
)
2
_
, 20/|z
ms
z
ne
|

, 4k
,2
_
Z
EU,EV
me:mu,ne:nv
max
_
min
_
max
_
2
_
(p
EU
me:mu
/L
ne
)
2
+ (q
EU
me:mu
/W
ne
)
2
,
2
_
(p
EV
ne:nv
/W
ne
)
2
+ (q
EV
ne:nv
/L
ne
)
2
_
+20
2
_
(1/L
ne
)
2
+ (1/W
ne
)
2
, 20/|z
me
z
ne
|

, 4k
,2
_
Z
EV,EV
me:mv,ne:nv
max
_
min
_
max
_
2
_
(p
EV
me:mv
/W
ne
)
2
+ (q
EV
me:mv
/L
ne
)
2
,
2
_
(p
EV
ne:nv
/W
ne
)
2
+ (q
EV
ne:nv
/L
ne
)
2
_
+20
2
_
(1/L
ne
)
2
+ (1/W
ne
)
2
, 20/|z
me
z
ne
|

, 4k
,2
_
Note: the terms containing |z
m
z
n
| should only be used if |z
m
z
n
| = 0.
University of PretoriaElectrical, Electronic and Computer Engineering 82
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
For the actual implementation of this method, adaptive Romberg integration was used for the eval-
uation of the inner integral, which is dened in (3.167). The evaluation of the outer integral, as
dened in (3.166), can also be accelerated, especially over parts
,2
and
,3
of the k

integration
contour where
m,n
(k

)e
jk

m,n
can be very oscillatory for large separation distances. Due
to the fact that
m,n
(k

) has a smooth behaviour over the k

integration contour, it can be ap-

proximated with cubic-spline interpolation over each part of the k

integration contour and the

integration can therefore be carried out analytically. If
m,n
(k

) is evaluated in a reduced set of

points, then between the points k

i
and k

i+1
,
m,n
(k

) can be approximated as

m,n
(k

) A
i
_
k

i
_
3
+B
i
_
k

i
_
2
+C
i
_
k

i
_
+D
i
, (3.168)
where A
i
, B
i
, C
i
and D
i
are the cubic-spline coefcients. The integral of
m,n
(k

)e
jk

m,n
,
over the interval between k

i
and k

i+1
, can then be evaluated as
_
k

i+1
k

m,n
(k

) e
jk

m,n
dk

_
k

i+1
k

i
_
A
i
_
k

i
_
3
+B
i
_
k

i
_
2
+C
i
_
k

i
_
+D
i
_
e
jk

m,n
dk

=
_
A
i
_
(k

i+1
k

i
)
3
j
m,n
+
3(k

i+1
k

i
)
2
(
m,n
)
2
+
6(k

i+1
k

i
)
j(
m,n
)
3

6
(
m,n
)
4
_
+B
i
_
(k

i+1
k

i
)
2
j
m,n
+
2(k

i+1
k

i
)
(
m,n
)
2
+
2
j(
m,n
)
3
_
+C
i
_
(k

i+1
k

i
)
j
m,n
+
1
(
m,n
)
2
_
+D
i
_
1
j
m,n
__
e
jk

m,n
e
j(k

i+1
k

i
)
m,n

_
A
i
_
6
(
m,n
)
4
_
+B
i
_
2
j(
m,n
)
3
_
+C
i
_
1
(
m,n
)
2
_
+D
i
_
1
j
m,n
__
e
jk

m,n
.
(3.169)
After some numerical experimentation, it was found that the number of interpolation points that are
required for sufcient accuracy, amounts to 16 on parts
,1
and
,4
of the k

integration contour,
while 6 points are needed on parts
,2
and
,3
. Another observation regarding
m,n
(k

), is that
it is not dependent on the separation distance
m,n
, and therefore it only has to be computed
once. Thereafter it can be used with any value of
m,n
, provided that the relative alignment
of the basis and testing functions stays the same. This is often the case when analysing antenna
arrays.
An important point, that has to be kept in mind when using this integration technique, is that it
cannot be used for the case where there is any overlap between the basis and testing functions.
The reason therefore is that the shift exponential does not decay to zero in the negative imaginary
k

plane, but instead grows exponentially. Sereno-Garino et al. pointed out that there should be no
physical overlap between the basis and testing functions when using this method. However, during
University of PretoriaElectrical, Electronic and Computer Engineering 83
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation

No overlap
m,n
decays along negative m
_
k

_
axis
Overlap
m,n
grows along negative m
_
k

_
axis
f
m
f
n
f
m
f
n
Figure 3.18 Condition for integration along the negative imaginary k

axis.
this study, it has been found that this condition alone is not sufcient. It has been found, through
numerical experimentation, that for this integration technique to work, there should be no overlap
between the basis and testing functions in the direction. This is illustrated in Figure 3.18, where
it can be seen that, although there might not be any physical overlap between the two functions, it
is still possible to have overlap in the direction. Note that this situation does not occur for basis
and testing functions with circular support, but only occurs when one of the functions does not
have circular support. In such a situation, one would be forced to use the less efcient integration
method in Section 3.9.1.
Even though this technique is much more efcient than the one in the previous section, for basis
and testing functions that are widely separated, there is still a more efcient method for axisym-
metric functions. This will now be addressed in the section that follows.
University of PretoriaElectrical, Electronic and Computer Engineering 84
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
3.9.3 Integration Strategy for Separated Axisymmetric Basis and Testing Functions
The interaction integrals for axisymmetric basis and testing functions that are separated, can be
evaluated very efciently. These functions include the PWS basis functions on the probes, as well
as the circular attachment modes. They are both characterised by the fact that, in a polar coordinate
system, they have no angular variation (i.e. they are only functions of k

). The advantage of this

is that the double integrals in (3.162) can be reduced to a single integral.
For basis and testing functions with no angular variation (note that the Greens function has no
angular variation either), the expression in (3.162) can be written as
Z
m,n
=
1
4
2
_
2
0
_

0

m,n
(k

) e
jk

m,n
cos()
k

dk

d. (3.170)
Then, by realising that
J
0
(k

m,n
) =
1
2
_
2
0
e
jk

m,n
cos()
d, (3.171)
(3.170) can be reduced to
Z
m,n
=
1
2
_

0

m,n
(k

)J
0
(k

m,n
)k

dk

. (3.172)
The integrand in (3.172) can be broken up into of a smooth part,
m,n
(k

)k

, as well as an oscil-
lating part, J
0
(k

m,n
). Such integrals can be evaluated very efciently by using the method of
averages [146,177]. With this method, the integral is approximated by a summation of integrations
over half cycles of the integrand. An acceleration device, such as Shanks nonlinear transforma-
tion, is then used to speed up the summation. The deformed integration contour of Figure 3.13 is
also used in this case.
An additional method that can be used to speed up the overall analysis, is to identify all entries in
the interaction matrix that are duplicates and to avoid calculating them more than once. This will
now be addressed in the section that follows.
3.10 IDENTIFICATION OF DUPLICATE ENTRIES IN THE INTERACTION MATRIX
Usually, there are many entries in the interaction matrix that are duplicates. This is especially so
when an antenna array is analysed. A tremendous amount of computational time can be saved
if these entries are identied beforehand and not calculated more than once. In Section 3.7, it
has already been pointed out that the interaction matrix is symmetric and that close to half of
the entries need not be calculated right from the outset. Depending on the antenna geometry and
choice of basis functions, the remaining number of entries that has to be calculated, can further be
University of PretoriaElectrical, Electronic and Computer Engineering 85
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
reduced by an order of magnitude or so.
As have been shown in Section 3.7, the interaction matrix consists of a number of submatrices
where each submatrix contains the entries associated with a specic combination of basis and
testing functions. The process of nding the duplicate entries is implemented by rst initialising
all the entries in the interaction matrix. A set of certain geometric parameters is then calculated
for the rst entry in a specic submatrix. This entry then becomes the reference entry. The same
set of parameters is now calculated in turn for every other entry in the submatrix. Wherever these
two sets correspond, the address of the reference entry is stored in the other entrys location. After
the geometric parameters of all the entries in the submatrix have been compared to that of the
reference entry, the reference entry is changed to the second entry (or to the next entry where an
address of a reference entry has not been stored yet) in the submatrix and the process of comparing
the geometric parameters of the other entries to that of the reference entry, is repeated again for
each remaining entry. Of course, if an address of a reference entry has already been stored for a
specic entry, that entry does not have to be checked again. Also, the number of entries that have to
be checked against the reference entry, becomes smaller as the process is repeated, due to the fact
that the previous reference entries do not have to be checked again. This process, as has just been
described, is followed for each submatrix in the interaction matrix. Finally, only the entries with
no address information are actually calculated, while the other values are simply fetched from
the corresponding address that is stored in that entrys location. The process of nding all the
duplicate entries only have to be performed once as it is independent of frequency. It is therefore
advisable to store the matrix with the address information off line so that it can be reused again.
It is possible that some entries in one submatrix can correspond to entries in another submatrix.
Therefore, some of the submatrices are grouped together during the process of nding the duplicate
entries. The submatrices that can be grouped together are:

_
Z
PZ,SU
mp:mz,ns:nu

and
_
Z
PZ,SV
mp:mz,ns:nv

_
Z
A,SU
ma,ns:nu

and
_
Z
A,SV
ma,ns:nv

_
Z
PZ,EU
mp:mz,ne:nu

and
_
Z
PZ,EV
mp:mz,ne:nv

_
Z
A,EU
ma,ne:nu

and
_
Z
A,EV
ma,ne:nv

_
Z
SU,SU
ms:mu,ns:nu

,
_
Z
SU,SV
ms:mu,ns:nv

and
_
Z
SV,SV
ms:mv,ns:nv

_
Z
SU,EU
ms:mu,ne:nu

,
_
Z
SU,EV
ms:mu,ne:nv

,
_
Z
SV,EU
ms:mv,ne:nu

and
_
Z
SV,EV
ms:mv,ne:nv

_
Z
EU,EU
me:mu,ne:nu

,
_
Z
EU,EV
me:mu,ne:nv

and
_
Z
EV,EV
me:mv,ne:nv

.
All the interaction-matrix entries can basically be divided into three groups for the purpose of nd-
ing the duplicate entries. Firstly, there are entries for which both the basis and testing functions
are axisymmetric. These include the Z
PZ,PS
mp:mz,np:nz
, Z
PZ,A
mp:mz,na
and Z
A,A
ma,na
entries. Secondly, there
are entries for which the basis functions have rectangular support, and for which the testing func-
University of PretoriaElectrical, Electronic and Computer Engineering 86
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
tions are axisymmetric. These include the Z
PZ,SU
mp:mz,ns:nu
, Z
PZ,SV
mp:mz,ns:nv
, Z
A,SU
ma,ns:nu
and Z
A,SV
ma,ns:nv
entries. Thirdly and nally, there are entries for which both the basis and testing functions have
rectangular support. These include the Z
SU,SU
ms:mu,ns:nu
, Z
SU,SV
ms:mu,ns:nv
and Z
SV,SV
ms:mv,ns:nv
entries, the
Z
SU,EU
ms:mu,ne:nu
, Z
SU,EV
ms:mu,ne:nv
, Z
SV,EU
ms:mv,ne:nu
and Z
SV,EV
ms:mv,ne:nv
entries, as well as the Z
EU,EU
me:mu,ne:nu
,
Z
EU,EV
me:mu,ne:nv
and Z
EV,EV
me:mv,ne:nv
entries. The algorithms that were developed to determine whether
two entries are duplicates, are not trivial and therefore each of them, together with the geometric
parameters that need to be compared for entries in each of the three groups, will be discussed
separately.
Before the different algorithms are discussed, it is necessary to dene two new functions that will
be used. The rst one is ang

(). It takes the angle as argument and returns the same angle,
but limited between and . The function can be dened as
ang

() =
_

_
2 int
_

2
_
2, 2 int
_

2
_
>
2 int
_

2
_
+ 2, 2 int
_

2
_

, otherwise.
(3.173)
The second function, and
1
(a, b), is a special and function that takes two arguments, a and b.
The function is dened as
and
1
(a, b) =
_

_
1, |a| = 0, |b| = 0
1, |a| > 0, |b| > 0
1, otherwise.
(3.174)
3.10.1 Interaction between Axisymmetric Functions
The identication of duplicate interaction-matrix entries, where both basis and testing functions
are axisymmetric, is the most simple case of all. Let Z
m,n
be the reference entry and Z
m

,n
the
entry being compared to the reference entry. Furthermore, let a
m
and a
n
be the radii of the testing
and basis functions associated with Z
m,n
, while a
m
and a
n
are associated with Z
m

,n
. The test-
ing and basis functions associated with Z
m,n
, are located at z
m
and z
n
along the z direction, while
those associated with Z
m,n
, are located at z
m
and z
n
. For Z
m,n
, the separation distance between
basis and testing functions is
m,n
, while for Z
m

,n
, the separation distance is
m

,n
. The
interactions Z
m

,n
and Z
m,n
, can then be checked for equality through the following algorithm:
IF
m

,n
=
m,n
THEN
IF (z
m
= z
m
AND z
n
= z
n
) OR (z
m
= z
n
AND z
n
= z
m
) THEN
IF (a
m
= a
m
AND a
n
= a
n
) OR (a
m
= a
n
AND a
n
= a
m
) THEN
Z
m

,n
Z
m,n
University of PretoriaElectrical, Electronic and Computer Engineering 87
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
END IF
END IF
END IF
3.10.2 Interaction between Axisymmetric Functions and Functions with
Rectangular Support
Figure 3.19 shows the reference interaction Z
m,n
and the interaction Z
m

,n
being compared to the
reference interaction for the case where the testing function is axisymmetric and where the basis
function has rectangular support. Once again, let a
m
be the radius of the testing function associated
with Z
m,n
, and a
m
the radius of the testing function associated with Z
m

,n
. The dimensions of
the rectangular patch associated with Z
m,n
, are L
n
and W
n
, while those associated with Z
m

,n
,
are L
n
and W
n
. The angles
n
and
n
, correspond to those in Figure 3.12. The testing and
basis functions associated with Z
m,n
, are located at z
m
and z
n
along the z direction, while those
associated with Z
m,n
, are located at z
m
and z
n
. For Z
m,n
, the separation distance between basis
and testing functions is
m,n
, while for Z
m

,n
, the separation distance is
m

,n
. Furthermore,
(p
n
, q
n
) is the set of modes on the basis function associated with Z
m,n
, while (p
n
, q
n
) is the set
of modes on the basis function associated with Z
m

,n
. The interactions Z
m

,n
and Z
m,n
, can then
be checked for equality through the following algorithm:
IF
m

,n
=
m,n
THEN
IF (z
m
= z
m
AND z
n
= z
n
) THEN
IF a
m
= a
m
AND L
n
= L
n
AND W
n
= W
n
AND p
n
= p
n
AND q
n
= q
n
THEN
IF |ang

(
n
) ang

(
n
)| = (0 OR ) THEN
IF (p
n
+q
n
) mod 2 = 0 THEN
SIGN 1
ELSE
SIGN and
1
(|ang

(
n
) ang

(
n
)|, 0)
END IF
Z
m

,n
SIGN Z
m,n
ELSE IF |ang

(
n
) + ang

(
n
)| = (0 OR ) THEN
IF (p
n
+q
n
) mod 2 = 0 THEN
SIGN 1
ELSE
SIGN and
1
(|ang

(
n
) + ang

(
n
)|, 0)
END IF
SIGN SIGN and
1
(q
n
mod 2, 0)
University of PretoriaElectrical, Electronic and Computer Engineering 88
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation

a
m
a
m

u
v
u
v

L
n
W
n
L
n

W
n

p
n
= 2
q
n
= 1
p
n
= 2
q
n
= 1
Z
m

,n
= Z
m,n

m,n

,n

Figure 3.19 Checking Z

m

,n
and Z
m,n
for equality where axisym-
metric basis functions and basis functions with rectangular support are
involved.
Z
m

,n
SIGN Z
m,n
END IF
END IF
END IF
END IF
To visually illustrate what this algorithm does, consider the situation depicted in Figure 3.19. It
can be seen that the circled current component (which is the nearest current component to the
axisymmetric function) associated with the Z
m,n
interaction, points towards a ctitious line that
is orthogonal to itself and that goes through the centre of the testing function. On the other hand,
the circled current component associated with the Z
m

,n
interaction, points away from a similar
virtual line. Given that the other dimensions, modes and angles correspond, Z
m,n
is therefore
equal to Z
m

,n
. Furthermore, if
m,n
= 0 and
m

,n
= 0, and given that all the relevant
University of PretoriaElectrical, Electronic and Computer Engineering 89
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
dimension, mode and angle requirements are satised in the algorithm, Z
m,n
is always equal to
Z
m

,n
, and the SIGN does not need to be calculated.
Another point that has to be noted here is that, for the purposes of this algorithm, a subdomain
rooftop basis function can be considered to be a special case of the entire-domain sinusoidal basis
function with the (p = 1, q = 0) set of modes. Also, as have been noted earlier on, there are
submatrices in the interaction matrix with entries that are equal. In this case, for example, the
Z
m,n
interaction can be that between a circular attachment mode and a u-directed entire-domain
sinusoidal basis function, while the Z
m

,n
interaction can be that between a circular attachment
mode and a v-directed entire-domain sinusoidal basis function. Now, in the previous algorithm,
if the basis function associated with Z
m,n
, is u-directed, while the basis function associated with
Z
m

,n
, is v-directed, an angle of (
n
+/2) has to be used in the place of
n
. On the other hand,
if the basis function associated with Z
m,n
, is v-directed, while the basis function associated with
Z
m

,n
, is u-directed, an angle of (
n
/2) has to be used in the place of
n
.
3.10.3 Interaction between Functions with Rectangular Support
Figure 3.20 shows the reference interaction Z
m,n
and the interaction Z
m

,n
being compared to
the reference interaction for the case where both the testing function and the basis function have
rectangular support. The dimensions and angles are the same as in the previous discussion, except
that both functions have rectangular support in this case. The interactions Z
m

,n
and Z
m,n
, can
then be checked for equality through the following algorithm:
IF
m

,n
=
m,n
THEN
IF (z
m
= z
m
AND z
n
= z
n
) OR (z
m
= z
n
AND z
n
= z
m
) THEN
IF L
m
= L
m
AND W
m
= W
m
AND p
m
= p
m
AND q
m
= q
m
AND L
n
= L
n
AND W
n
= W
n
AND p
n
= p
n
AND q
n
= q
n
THEN
IF |ang

(
m
) ang

(
m
)| = (0 OR )
AND |ang

(
n
) ang

(
n
)| = (0 OR ) THEN
IF (p
m
+q
m
) mod 2 = 0 THEN
SIGN 1
ELSE
SIGN and
1
(|ang

(
m
) ang

(
m
)|, 0)
END IF
IF (p
n
+q
n
) mod 2 = 0 THEN
SIGN SIGN 1
ELSE
SIGN SIGN and
1
(|ang

(
n
) ang

(
n
)|, 0)
University of PretoriaElectrical, Electronic and Computer Engineering 90
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation

u
v
u
v
u
v
u
v

Fictitious axisymmetric function

Fictitious axisymmetric function
L
m
W
m
L
n
W
n
L
m
W
m
L
n

W
n

p
m
= 3
q
m
= 2
p
n
= 2
q
n
= 1
p
m
= 3
q
m
= 2
p
n
= 2
q
n
= 1
Z
m

,n
= Z
m,n

m,n

,n

Figure 3.20 Checking Z

m

,n
and Z
m,n
for equality where two basis functions with rectangular
support are involved.
END IF
Z
m

,n
SIGN Z
m,n
ELSE IF |ang

(
m
) + ang

(
m
)| = (0 OR )
AND |ang

(
n
) + ang

(
n
)| = (0 OR ) THEN
IF (p
m
+q
m
) mod 2 = 0 THEN
SIGN 1
ELSE
SIGN and
1
(|ang

(
m
) + ang

(
m
)|, 0)
END IF
University of PretoriaElectrical, Electronic and Computer Engineering 91
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
IF (p
n
+q
n
) mod 2 = 0 THEN
SIGN SIGN 1
ELSE
SIGN SIGN and
1
(|ang

(
n
) + ang

(
n
)|, 0)
END IF
SIGN SIGN and
1
(q
m
mod 2, 0)
SIGN SIGN and
1
(q
n
mod 2, 0)
Z
m

,n
SIGN Z
m,n
END IF
END IF
ELSE IF L
m
= L
n
AND W
m
= W
n
AND p
m
= p
n
AND q
m
= q
n
AND L
n
= L
m
AND W
n
= W
m
AND p
n
= p
m
AND q
n
= q
m
THEN
IF |ang

(
m
) ang

(
n
)| = (0 OR )
AND |ang

(
n
) ang

(
m
)| = (0 OR ) THEN
IF (p
m
+q
m
) mod 2 = 0 THEN
SIGN 1
ELSE
SIGN and
1
(|ang

(
m
) ang

(
n
)|, )
END IF
IF (p
n
+q
n
) mod 2 = 0 THEN
SIGN SIGN 1
ELSE
SIGN SIGN and
1
(|ang

(
n
) ang

(
m
)|, )
END IF
Z
m

,n
SIGN Z
m,n
ELSE IF |ang

(
m
) + ang

(
n
)| = (0 OR )
AND |ang

(
n
) + ang

(
m
)| = (0 OR ) THEN
IF (p
m
+q
m
) mod 2 = 0 THEN
SIGN 1
ELSE
SIGN and
1
(|ang

(
m
) + ang

(
n
)|, )
END IF
University of PretoriaElectrical, Electronic and Computer Engineering 92
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
IF (p
n
+q
n
) mod 2 = 0 THEN
SIGN SIGN 1
ELSE
SIGN SIGN and
1
(|ang

(
n
) + ang

(
m
)|, )
END IF
SIGN SIGN and
1
(q
m
mod 2, 0)
SIGN SIGN and
1
(q
n
mod 2, 0)
Z
m

,n
SIGN Z
m,n
END IF
END IF
END IF
END IF
To visually illustrate what this algorithm does, consider the situation depicted in Figure 3.20. The
interaction can be viewed as two separate interactions, namely the interaction between the testing
function and a ctitious axisymmetric function, as well as the interaction between the basis func-
tion and the same ctitious axisymmetric function. Here, the ctitious axisymmetric function is
located between the testing and basis functions. Now, consider the situation in Figure 3.20 for the
Z
m,n
interaction. It can be seen that the circled current component (which is the nearest current
component to the ctitious axisymmetric function) on the testing function points towards a cti-
tious line that is orthogonal to itself and that goes through the centre of the ctitious axisymmetric
function. Similarly, the circled current component on the basis function, also points towards a sim-
ilar virtual line. On the other hand, for the Z
m

,n
interaction, it can be seen that the circled current
component on the testing function points towards the virtual line, but that the circled current com-
ponent on the basis function, points away from the virtual line. Given that the other dimensions,
modes and angles correspond, Z
m,n
is therefore equal to Z
m

,n
. Furthermore, if
m,n
= 0
and
m

,n
= 0, the same algorithm, as has just been described, can be used. However, the basis
functions associated with the Z
m

,n
interaction are rst rotated so that the basis function associ-
ated with Z
m

,n
points in the same direction as the basis function associated with Z
m,n
. This is
done to avoid problems when checking the angle requirements in the algorithm.
As in the previous discussion, it is also possible that, for example, the Z
m,n
interaction can have
a u-directed entire-domain sinusoidal basis function, while the Z
m

,n
interaction can have a v-
directed entire-domain sinusoidal basis function. In such a case, the same arguments (with respect
to the angles) as in the previous discussion, can also be applied here. Another situation that has to
be catered for in this case, is where the basis function in Z
m,n
is compared to the testing function
in Z
m

,n
and vice versa. It is treated in a similar way as has just been described.
University of PretoriaElectrical, Electronic and Computer Engineering 93
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation

u v
z
2a
mp
V
P
mp
E
inc
tan
g
z
PZ
1
f
PZ
1
(z)
PEC ground plane
Figure 3.21 The source region on probe mp.
The only part of the system of linear equations in (3.13) that still needs to be addressed, is the
evaluation of the excitation vector elements. This will now follow.
3.11 EVALUATION OF THE EXCITATION-VECTOR ELEMENTS
Assume that each probe is excited by a voltage source as shown in Figure 3.21. The source region
can then be described by the so-called delta-gap model [4, 73]. Using the delta-gap model, it is
assumed that a constant voltage of V
P
mp
is applied over the narrow feed gap g of probe mp. The
incident electric eld E
inc
tan
is then also constant over the feed gap and zero elsewhere. It can be
related to the feed-gap voltage and feed-gap width through
E
inc
tan
(u, v, z) =
V
P
mp
g
z

u
2
+v
2
=a
2
mp
, z
PZ
1
z z
PZ
1
+ g. (3.175)
Another way to visualise the delta-gap model, is to place a narrow band of impressed magnetic
current density in the feed gap. This magnetic current density would ow in a direction around
the probe and would set up the incident electric eld across the feed gap. It would be related to
the incident electric eld through
M
imp
(u, v, z) = n E
inc
tan
(u, v, z)

u
2
+v
2
=a
2
mp
, (3.176)
where n is the normal vector on the surface of the probe. This representation of the delta-gap
model should not be confused with the magnetic-frill model [4, 73].
University of PretoriaElectrical, Electronic and Computer Engineering 94
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
Each nonzero entry of the excitation vector can now be expressed as
V
PZ
mp:mz
=
__
S
mp:mz
f
PZ
mp:mz
(r) E
inc
tan
(r) ds, mp = 1, 2, . . . , N
P
, mz = 1, (3.177)
where
f
PZ
mp:1
(u, v, z) =
1
2a
mp
f
PZ
1
(z) z

u
2
+v
2
=a
2
mp
(3.178)
is the basis function at the bottom of probe mp. Here, f
PZ
1
(z) is one half of a PWS function as
given by (3.67). The expression in (3.177) for the nonzero entries of the excitation vector can then
be simplied to
V
PZ
mp
=
1
2a
mp
_
2
0
f
PZ
1
(z = z
PZ
1
)
V
P
mp
g
g a
mp
d
= V
P
mp
, mp = 1, 2, . . . , N
P
. (3.179)
All the elements that are required for the solution of the systemin (3.13), have nowbeen addressed.
Some issues regarding the actual solution of the system of linear equations will now be discussed
in the next section.
3.12 SOLUTION OF THE CURRENT-DENSITY COEFFICIENTS
Due to the fact that this formulation does not generate linear equations with an excessive number
of unknowns, Gauss-Jordan elimination, LU decomposition or any similar method [184], can be
used very effectively to solve for the unknown current-density coefcients I
n
. If the antenna that
is analysed, has more than one port, and if the network parameters are required, the system of
linear equations will have to be solved once for every port of the antenna. This is due to separate
excitation vectors that have to be set up for each port. In such a case, it would be advisable
to also store a copy of the interaction matrix so that it can be reused. If this is not done, the
interaction matrix would have to be recalculated for each excitation vector, thereby increasing the
computation time excessively. Furthermore, for a multiport antenna, each solution of the linear
system of equations would only provide the current-density coefcients that correspond to one
excitation vector. However, when calculating the far elds of a multiport antenna, one would
normally be interested in the elds that are generated due to a combination of port excitations.
This can be done by setting up one excitation vector for all of the ports simultaneously or by
summing the individual current-density coefcients in an appropriate way when the system of
linear equations is solved for multiple excitation vectors. These are all factors that have to be
kept in mind when implementing the formulation. The remaining sections will now deal with
observables that can be calculated once the system of linear equations has been solved. These
typically include the network parameters and far elds.
University of PretoriaElectrical, Electronic and Computer Engineering 95
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
3.13 EVALUATION OF THE NETWORK PARAMETERS
In the analysis of antennas or antenna arrays, there is almost always a requirement to calculate the
network parameters of the structure. These are usually expressed in terms of the impedance matrix
_
Z
P

, the admittance matrix

_
Y
P

or the scattering matrix

_
S
P

. With the current formulation,

it is most convenient to calculate the admittance matrix rst and then to calculate the impedance
matrix directly from the admittance matrix. Finally, the scattering matrix can then be calculated
by making use of both the admittance and impedance matrices. The scattering matrix is normally
used by software packages that are capable of microwave network analysis. Such packages can be
used, for example, to analyse an antenna array together with its feed network. This is achieved by
analysing the two structures separately, after which the scattering parameters of the two structures
are connected to yield an overall scattering matrix for the entire structure. However, for such an
analysis to be accurate, there should be no electromagnetic coupling between the two structures.
This is usually the case for probe-fed microstrip antennas.
For the purposes of this analysis, it can be assumed that port mp is associated with probe mp, port
np with probe np and so on. Therefore, the total number of ports will also be equal to the total
number of probes N
P
. Now, entry Y
P
mp,np
(which relates the current at port mp to the voltage at
port np) of the admittance matrix can be calculated as
Y
P
mp,np
=
I
PZ
mp:1
V
P
np
, (3.180)
where I
PZ
mp:1
is the current at port mp (also the current-density coefcient of the bottommost basis
function on probe mp) and V
P
np
is the applied voltage at port np. After all entries of the admittance
matrix have been calculated, the impedance matrix is simply calculated through
_
Z
P

=
_
Y
P

1
. (3.181)
With both the admittance matrix and the impedance matrix known, the scattering matrix can be
calculated as [185]
_
S
P

=
_
Z
P
0

1/2
__
Y
P
0

_
Y
P
___
Y
P
0

+
_
Y
P
_
1
_
Z
P
0

1/2
, (3.182)
where
_
Z
P
0

=
_

_
_
Z
P
0
_
1
0 0
0
_
Z
P
0
_
2
0
.
.
.
.
.
.
.
.
.
.
.
.
0 0
_
Z
P
0
_
N
P
_

_
(3.183)
University of PretoriaElectrical, Electronic and Computer Engineering 96
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
is a diagonal matrix with the characteristic impedance
_
Z
P
0
_
np
of each port on the diagonal of the
matrix. Furthermore,
_
Y
P
0

=
_

_
_
Y
P
0
_
1
0 0
0
_
Y
P
0
_
2
0
.
.
.
.
.
.
.
.
.
.
.
.
0 0
_
Y
P
0
_
N
P
_

_
(3.184)
is a diagonal matrix with the characteristic admittance
_
Y
P
0
_
np
of each port on the diagonal of the
matrix. Here,
_
Y
P
0
_
np
= 1/
_
Z
P
0
_
np
. Also in (3.182),
_
Z
P
0

1/2
=
_

_
_
_
Z
P
0
_
1
0 0
0
_
_
Z
P
0
_
2
0
.
.
.
.
.
.
.
.
.
.
.
.
0 0
_
_
Z
P
0
_
N
P
_

_
(3.185)
is a diagonal matrix with the square root of the characteristic impedance of each port on the
diagonal of the matrix, while
_
Y
P
0

1/2
=
_

_
_
_
Y
P
0
_
1
0 0
0
_
_
Y
P
0
_
2
0
.
.
.
.
.
.
.
.
.
.
.
.
0 0
_
_
Y
P
0
_
N
P
_

_
(3.186)
is a diagonal matrix with the square root of the characteristic admittance of each port on the
diagonal of the matrix.
3.14 EVALUATION OF THE FAR FIELDS
In the spectral domain, the evaluation of the far elds is relatively straightforward. The far elds
can be calculated by using the method of stationary phase [4, 178]. This method is commonly
used for the evaluation of integrals where the integrand is highly oscillatory, but where the con-
tributions to the integral mainly comes from the so-called stationary point and its surrounding
University of PretoriaElectrical, Electronic and Computer Engineering 97
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
neighbourhood. The radiation integrals that are used to obtain the far elds of antennas, are typ-
ical of such integrals. The method of stationary phase essentially nds an asymptotic solution to
such integrals.
In general, the far eld E
scat
at point r can be related to its spectral-domain counterpart, at position
z
r
along the z direction, through [6, 141]
E
scat
(r) =
1
j2

E
scat
(k
xs
, k
ys
, z
r
) k
zs
e
jk
zs
z
r
e
jk
(0)
r
r
. (3.187)
The stationary point is dened by
k
xs
= k
(0)
sin() cos(), (3.188)
k
ys
= k
(0)
sin() sin() (3.189)
and
k
zs
= k
(0)
cos(), (3.190)
where (, ) denotes the direction of the far eld, and where k
(0)
is the wavenumber in the top
layer that extends to innity along the positive z direction. The far eld can also be expressed as
E
scat
(r) = E
scat
(r, , ) = E
scat
(, )
e
jk
(0)
r
r
, (3.191)
where one part is a function of angular position only and the other part is a function of distance
only. The e
jk
(0)
r
/r part is often dropped from far-eld expressions, leaving only the angular
dependence of the far eld. By making use of (3.187), this part can be expressed as
E
scat
(, ) =
1
j2

E
scat
(k
xs
, k
ys
, z
r
) k
zs
e
jk
zs
z
r
. (3.192)
The spectral-domain form of the scattered eld

E
scat
(k
xs
, k
ys
, z
r
) is now rst expressed in terms
of its spatial-domain counterpart by invoking the two-dimensional Fourier transform of (3.18).
This then allows

E
scat
(k
xs
, k
ys
, z
r
) to be expressed in terms of the spatial-domain current density
on the structure, by making use of the spatial-domain Greens function. Mathematically, this is
expressed as

E
scat
(k
xs
, k
ys
, z
r
) =
_

E
scat
(x, y, z
r
) e
jk
xs
x
e
jk
ys
y
dxdy
=
_
y
_
x
__
z

_
y

_
x

G(x, y, z
r
|x

, y

, z

) J(x

, y

, z

) dx

dy

dz

_
e
jk
xs
x
e
jk
ys
y
dxdy. (3.193)
University of PretoriaElectrical, Electronic and Computer Engineering 98
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
Now, the Greens function can be written in terms of its spectral-domain counterpart, resulting in

E
scat
(k
xs
, k
ys
, z
r
) =
_
k
ys
_
k
xs
__
y
_
x
_
z

G(k
xs
, k
ys
, z
r
|z

__
y

_
x

J(x

, y

, z

) e
jk
xs
x

e
jk
ys
y

dx

dy

_
dz

e
jk
xs
x
e
jk
ys
y
dxdy
_
e
jk
xs
x
e
jk
ys
y
dk
xs
dk
ys
. (3.194)
The term is square brackets can be identied as a two-dimensional Fourier transform, enabling
one to express

E
scat
(k
xs
, k
ys
, z
r
) as

E
scat
(k
xs
, k
ys
, z
r
) =
_
k
ys
_
k
xs
__
y
_
x
_
z

G(k
xs
, k
ys
, z
r
|z

J(k
xs
, k
ys
, z

) dz

e
jk
xs
x
e
jk
ys
y
dxdy
_
e
jk
xs
x
e
jk
ys
y
dk
xs
dk
ys
=
_
z

G(k
xs
, k
ys
, z
r
|z

J(k
xs
, k
ys
, z

) dz

=
N

n=1
I
n
_
z

G(k
xs
, k
ys
, z
r
|z

f
n
(k
xs
, k
ys
, z

) dz

. (3.195)
At this point, it is now convenient to express the far eld E
scat
(, ) and the basis function

f
n
in
terms of their components in the global (x, y, z) coordinate system. These are given by
E
scat
(, ) = E
scat
x
(, ) x +E
scat
y
(, ) y +E
scat
z
(, ) z (3.196)
and

f
n
(k
xs
, k
ys
, z

) =
_

f
x
(k
xs
, k
ys
, z

)
_
n
x +
_

f
y
(k
xs
, k
ys
, z

)
_
n
y +
_

f
z
(k
xs
, k
ys
, z

)
_
n
z (3.197)
respectively.
The basis-function components in the global (x, y, z) coordinate system can be determined from
the components in each basis functions local (u, v, z) coordinate system. This is done through
_

f
x
(k
xs
, k
ys
, z

)
_
n
=
_

f
u
_
k
u
= k
xs
cos(
n
) +k
ys
sin(
n
), k
v
= k
xs
sin(
n
) +k
ys
cos(
n
), z

__
n
cos(
n
)
e
jk
xs
x
n
e
jk
ys
y
n

f
v
_
k
u
= k
xs
cos(
n
) +k
ys
sin(
n
), k
v
= k
xs
sin(
n
) +k
ys
cos(
n
), z

__
n
sin(
n
)
e
jk
xs
x
n
e
jk
ys
y
n
, (3.198)
University of PretoriaElectrical, Electronic and Computer Engineering 99
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
_

f
y
(k
xs
, k
ys
, z

)
_
n
=
_

f
u
_
k
u
= k
xs
cos(
n
) +k
ys
sin(
n
), k
v
= k
xs
sin(
n
) +k
ys
cos(
n
), z

__
n
sin(
n
)
e
jk
xs
x
n
e
jk
ys
y
n
+
_

f
v
_
k
u
= k
xs
cos(
n
) +k
ys
sin(
n
), k
v
= k
xs
sin(
n
) +k
ys
cos(
n
), z

__
n
cos(
n
)
e
jk
xs
x
n
e
jk
ys
y
n
(3.199)
and
_

f
z
(k
xs
, k
ys
, z

)
_
n
=
_

f
z
(k
u
= k
xs
, k
v
= k
ys
, z

)
_
n
e
jk
xs
x
n
e
jk
ys
y
n
, (3.200)
where x
n
and y
n
dene the (x, y) position of the n-th basis function in the global (x, y, z) coordi-
nate system and where the angle
n
has already been dened in Figure 3.12.
By using (3.195) to (3.197), the three components of the far eld can be expressed as
E
scat
x
(, ) =
N

n=1
I
n
_

G
xx
(k
xs
, k
ys
, z
r
|z

)
_

f
x
(k
xs
, k
ys
, z

)
_
n
+

G
xy
(k
xs
, k
ys
, z
r
|z

)
_

f
y
(k
xs
, k
ys
, z

)
_
n
+

G
I
xz
(k
xs
, k
ys
, z
r
)
_

f
z
(k
xs
, k
ys
)
_
n
_
, (3.201)
E
scat
y
(, ) =
N

n=1
I
n
_

G
yx
(k
xs
, k
ys
, z
r
|z

)
_

f
x
(k
xs
, k
ys
, z

)
_
n
+

G
yy
(k
xs
, k
ys
, z
r
|z

)
_

f
y
(k
xs
, k
ys
, z

)
_
n
+

G
I
yz
(k
xs
, k
ys
, z
r
)
_

f
z
(k
xs
, k
ys
)
_
n
_
(3.202)
and
E
scat
z
(, ) =
N

n=1
I
n
_

G
zx
(k
xs
, k
ys
, z
r
|z

)
_

f
x
(k
xs
, k
ys
, z

)
_
n
+

G
zy
(k
xs
, k
ys
, z
r
|z

)
_

f
y
(k
xs
, k
ys
, z

)
_
n
+

G
I
zz
(k
xs
, k
ys
, z
r
)
_

f
z
(k
xs
, k
ys
)
_
n
_
. (3.203)
In these, the expanded Greens-function components are given by

G
I
xz
(k
xs
, k
ys
, z
r
) =
_
z

G
I
xz
(k
xs
, k
ys
, z
r
|z

)
_

f
z
(z

)
_
n
dz

, (3.204)
University of PretoriaElectrical, Electronic and Computer Engineering 100
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation

G
I
yz
(k
xs
, k
ys
, z
r
) =
_
z

G
I
yz
(k
xs
, k
ys
, z
r
|z

)
_

f
z
(z

)
_
n
dz

(3.205)
and

G
I
zz
(k
xs
, k
ys
, z
r
) =
_
z

G
I
zz
(k
xs
, k
ys
, z
r
|z

)
_

f
z
(z

)
_
n
dz

. (3.206)
In Appendix B, it is shown how these expanded Greens functions can be evaluated.
Far elds are most often expressed in spherical coordinates through
E
scat
(, ) = E
scat

(, )

+E
scat

(, )

, (3.207)
where the two components, E
scat

and E
scat

, can be determined from their counterparts in a rectan-

gular coordinate system. This in done through
E
scat

(, ) = E
scat
x
(, ) cos() cos() +E
scat
y
(, ) cos() sin() E
scat
z
(, ) sin() (3.208)
and
E
scat

(, ) = E
scat
x
(, ) sin() +E
scat
y
(, ) cos(). (3.209)
Finally, the antenna gain associated with the two far-eld components, can be calculated as
G

(, ) =
4|E
scat

(, )|
2
2
(0)
P
in
(3.210)
and
G

(, ) =
4|E
scat

(, )|
2
2
(0)
P
in
, (3.211)
where

(0)
=

(0)

(0)
(3.212)
is the intrinsic impedance of the top layer, and
P
in
=
1
2
e
_
N
P

np=1
V
P
np
_
I
PZ
np:1
_

_
(3.213)
is the total input power to the antenna.
3.15 CONCLUDING REMARKS
This chapter presented a detailed exposition of the theoretical formulation that was implemented
for the analysis of microstrip patch antennas with capacitive feed probes. It started off with an
overview of the general SDMM formulation for a structure that is embedded within a grounded
multilayered medium. The various basis function were discussed at length, while a new higher-
University of PretoriaElectrical, Electronic and Computer Engineering 101
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )
Chapter 3 Implementation of the Spectral-Domain Moment-Method Formulation
order circular attachment mode was presented for the modelling of circular capacitor patches (the
suitability of the various attachment modes will be investigated in the next chapter). It was shown
how different integration strategies can be used to speed up the calculation of interaction integrals.
Here, a new method for the calculation of the interaction between basis and testing functions
that are widely separated, was modied to also handle small separation distances. A number of
algorithms were developed to identify duplicate entries within the interaction matrix. This has a
signicant effect on the overall computational time that is required for the analysis. In the next
chapter it will be shown how effective this method is. Finally, it was shown how observables such
as the network parameters and far elds, can be calculated. To summarise, although parts of this
analysis have been implemented and reported on before, this analysis as a whole is unique in the
combination of basis functions and numerical methods that have been implemented. As such, no
results for this analysis are available in the literature. In Chapter 4, this analysis will be applied to
a variety of antenna congurations and the results will be compared to measurements, as well as
the results of other commercial codes.
University of PretoriaElectrical, Electronic and Computer Engineering 102
U Un ni iv ve er rs si it ty y o of f P Pr re et to or ri ia a e et td d M Ma ay yh he ew w- -R Ri id dg ge er rs s, , G G ( (2 20 00 04 4) )