Synthesis of Far Field Patterns: Electromagnetic Sources and Electromagnetic Fields
Synthesis of Far Field Patterns: Electromagnetic Sources and Electromagnetic Fields
Synthesis of Far Field Patterns: Electromagnetic Sources and Electromagnetic Fields
Abstract In this chapter, we will focus on the synthesis of antenna arrays and
treat the problem from a slightly different perspective. We at first to synthesize
the required radiation pattern with a continuous current distribution in free space,
and then spatially sample the continuous current source and realize it with discrete
radiation elements. We show that the effective number of degrees of freedom (NDF)
can be adopted as a useful information to obtain the direct synthesis pattern, which
can be used as a very good initial value for further optimization. The sidelobe levels
and the ripples in the main beams can be effectively controlled using an efficient
hybrid optimization algorithm, in which the extrema of the objective radiation pattern
are assigned based on the properties of the entire function while their positions are
flexibly adjusted. Meantime, we investigate the relationship between the radiation
pattern of the continuous current and that of the spatially sampled current and discuss
the aliasing effect on the radiation pattern due to spatial sampling. Eleven examples
are provided to demonstrate that the sidelobe levels and the ripples in the main beams
can be controlled much more effectively with the hybrid method.
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 233
G. Xiao, Electromagnetic Sources and Electromagnetic Fields, Modern Antenna,
https://doi.org/10.1007/978-981-99-9449-6_7
234 7 Synthesis of Far Field Patterns
Taking the direct synthesis result as the initial value, the sidelobe levels and the ripples
in the main beams can be effectively controlled using an efficient hybrid optimization
algorithm. As for the second step, we will investigate the relationship between the
radiation pattern of the continuous current and that of the spatially sampled current.
We will compare the continuous array factor and the discrete array factor, and discuss
the aliasing effect on the radiation pattern due to spatial sampling.
For a time harmonic source J(r1 ) in domain Vs , its electric field in free space is
expressed by [1–4].
{ ( )
∇∇
E(r) = − j ωμ0 I + 2 G(r, r1 ) · J(r1 )dr1 (7.1)
k0
Vs
where I is the identity operator, and G(r, r1 ) is the scalar Green’s function. In three-
dimensional space, G(r, r1 ) = exp(− jk0 R)/4π R, and R = |r − r1 |.
( )2
At far field region, r ≫ |r1 | and R ≈ r − r̂ · r1 − r̂ · r1 /2r . For large R, the
variation in phase is much more significant than the variation in amplitude,
| − jk0 ΔR | | |
|e | | |
| | ≫ |e− jk0 R Δ 1 |.
| R | | R|
We often take 1/R ≈ 1/r ,exp(− jk0 R) ≈ exp(− jk0 r ) exp( jk · r1 ), and replace
( · r1 ), where g0 = exp(−
the scalar Green’s function by g0 exp( jk ) jk0 r )/4πr . Mean-
time, we replace the gradient ∇G by − jk0 g0 exp( jk · r1 )r̂ . The far field can be
approximately expressed by
{
( )Δ Δ
In the expression, k = k x âx + k y â y + k z âz is the wave vector. Because the far
field is a spherical TEM wave, the corresponding magnetic field can be obtained by
1
H(r) ≈ r̂ × E(r). (7.3)
η0
F(θ, ϕ) = lim r e jk0 r E(r) ≈ − I−rr · e jk0 r̂·r1 J(r1 )dr1 (7.4)
r →∞ 4π
Vs
7.1 Electromagnetic Far Field in Free Space 235
and the normalized magnetic far field can be written in terms of F(θ, ϕ) as
( ) 1 1
lim r e jk0 r H(r) = lim r e jk0 r r̂ × E(r) ≈ r̂ × F(θ, ϕ). (7.5)
r →∞ η0 r →∞ η0
The total radiation power can be evaluated by integrating the square of the far
field over the solid-angle domain Ω, namely,
{
1
Prad = |F(θ, ϕ)|2 dΩ. (7.6)
2η0
Ω
{ { 2π { π
Note that Ω [·]dΩ = 0 0 [·] sin θ dθ dϕ.
(A) Line Current Source
Consider a line current source on the z-axis as shown in Fig. 7.1. Its current density
is expressed by
{
Dz /2
j ωμ0
F(θ ) = sin θ e jkz z I (z)dz θ̂ (7.8)
4π
−Dz /2
where k z = k0 cos θ . Particularly, if we choose I (z) = δ(z), the line current becomes
an infinitesimal dipole. Its far field turns out to be
j ωμ0
Fdi p (θ ) = sin θ θ̂ f di p,z (θ )θ̂ (7.9)
4π
j ωμ0
Fdi p,z (θ ) = sin θ. (7.10)
4π
Obviously, the factor Fdi p,z (θ ) can be regarded as the element factor of the
infinitesimal dipole. We sometimes call it polarization factor. It is dependent on the
polarization of the current source, as(indicated
) by the second subscript in Fdi p,z (θ ).
In later sections, we may use Fdi p, p θ p to represent the p̂-polarized infinitesimal
dipole factor, where θ p is the angle of the radial vector r and the polarization direction
p̂.
In this book, we define the rest part of Eq. 7.8 as the continuous array factor,
{
Dz /2
where Ωz = 2π/Dz = k0 /Nz . Substituting Eq. 7.12 into Eq. 7.8 yields
∞
∑
F(θ ) = Fdi p,z (θ )Dz In f n (θ ) = Fdi p,z (θ )Fa,z (θ ) (7.13)
n=−∞
where Fdi p,z (θ ) is defined in Eq. 7.9. The continuous array factor Eq. 7.11 in this
situation is identified as
∞
∑
Fa,z (θ ) = Dz In f n (θ ). (7.14)
n=−∞
It is important to note from the properties of the sinc function that, under the
condition of Dz = Nz λ, we have f n (θn ) = 1 and f n (θm ) = 0 for m /= n.
Each mode function f n (θ ) represents a beam that is axially symmetrical with
respect to the line source. The beam of the n-th mode is regarded to locate in the
visible region if its peak direction satisfies −π ≤ θn ≤ π , corresponding to the mode
number range of −Nz ≤ n ≤ Nz .
The transversal wavenumber kρ is expressed by
/ /
kρ = k02 − k z2 = k0 1 − (n/Nz )2 .
Apparently, when the peak direction of the main lobe of a mode falls in the visible
region, the mode belongs to the propagation group and is a propagation mode. All
other modes that fall out of the propagation group are evanescent modes.
If a current source only consists of the propagation modes, its far field can be
rigorously expressed by
∑
Nz
F(θ ) = Fdi p,z (θ )Dz In f n (θ ) = Fdi p,z (θ )Fa,z (θ ) (7.18)
n=−Nz
∑
Nz
Fa,z (θ ) = Dz In f n (θ ). (7.19)
n=−Nz
We can interpret Eq. 7.19 in two ways. The first one is to treat Fa,z (θ ) as the
continuous array factor for the continuous current source, which can be considered
as an array consisting of infinitesimal dipoles. Then Fdi p,z (θ ) is naturally the element
factor and Eq. 7.18 is simply the result based on the pattern multiplication principle.
The second one is to interpret Fa,z (θ ) as F(θ )/Fdi p,z (θ ), that is, we just merge the
two terms together and treat Fa,z (θ ) as the radiation function synthesized using the
238 7 Synthesis of Far Field Patterns
mode functions. In this book, we treat Fa,z (θ ) as the continuous array factor. We
will usually focus on synthesizing Fa,z (θ ) instead of F(θ ). With this strategy, the
synthesis can be fulfilled with a kind of standard algorithm because Fa,z (θ ) is simply
the superposition of the sinc mode functions.
For a line current with length of 4λ, there are nine propagation modes. Their
curves in the visible region are plotted in Fig. 7.2a. The first eight evanescent modes
that have the largest amplitudes in the visible region are shown in Fig. 7.2b. The
beam patterns corresponding to f 0 (θ ), f −1 (θ ), and f 4 (θ ) are shown in Fig. 7.3.
where the two continuous array factors are the Fourier Transforms of the two current
components,
Fig. 7.2 Mode functions for the linear current with length of 4λ. a Nine propagation modes. b Eight
evanescent modes
{y /2
D {x /2
D
{y /2
D {x /2
D
( ) j ωμ0
Fdi p, p θ p = sin θ p , p = x, y.
4π
In the spherical coordinate system, k x = k0 sin θ cos ϕ, k y = k0 sin θ sin ϕ, and
k z = k0 cos θ . θx is the angle between the position vector r and the x-axis, θ y is
angle between the position vector r and the y-axis. θ̂x and θ̂ y are respectively the
corresponding unit vectors, as shown in Fig. 7.4. Explicitly we have
( )
Δ Δ
∞
∑ ∞
∑
I y (x, y) = I ymn e j (mΩx x+nΩ y y ) (7.24)
m=−∞ n=−∞
where Ωx = 2π/Dx and Ω y = 2π/D y . Substituting Eqs. 7.23 and 7.24 into Eq. 7.20
and carrying out the integrations yield
∞
∑ ∞ [
∑ ]
( )
F(θ, ϕ) = Fdi p,x (θx )Ixmn θ̂x + Fdi p,y θ y I ymn θ̂ y f mn (θ, ϕ). (7.25)
m=−∞ n=−∞
The two-dimensional mode function f mn (θ, ϕ) is the product of two sinc functions
with respect to k x and k y , respectively,
[ ] [ ]
Dx Dy ( )
f mn (θ, ϕ) = sinc (k x + mΩx ) sinc k y + nΩ y . (7.27)
2 2
( )
( Each mode) function reaches its peaks at the two directions of θmn, ϕmn and
π − θmn, ϕmn . The angles of the peaks can be determined with equations
{
k0 sin θmn cos ϕmn + mΩx = 0
. (7.28)
k0 sin θmn sin ϕmn + nΩ y = 0
(The wave vector )of the (m, n) mode at the peak direction (θmn , ϕmn ) is denoted
as k xmn , k ymn , k zmn and can be calculated with
⎧
⎨ k xmn = k0 sin θmn cos ϕmn
⎪
k ymn = k0 sin θmn sin ϕmn . (7.29)
⎪
⎩
k zmn = k0 cos θmn
7.1 Electromagnetic Far Field in Free Space 241
( )
At the other peak point π − θmn, ϕmn , k xmn and k ymn are the same as those
described in Eq. 7.29. However, k zmn changes its sign, i.e., k zmn = −k0 cos θmn .
In order to analyze the properties of these modes in a simple way, we choose
Dx × D y = N x λ × N y λ, where N x and N y are two integers. In the visible region,
0 ≤ θmn ≤ π and 0 ≤ ϕmn ≤ 2π . It can be derived from Eq. 7.28 that
( )2
(mΩx )2 + nΩ y = k02 sin2 θmn ≤ k02
which leads to
( )2
(m/N x )2 + n/N y ≤ 1. (7.30)
Each two-dimensional mode function in the visible region describes a beam that
has two main lobes symmetrically located in the two sides of the source plane. For
a 2λ × 2λ current sheet, the two beams associated with f 00 (θ, ϕ) and f 02 (θ, ϕ) are
shown in Fig. 7.5. Note that when θmn = π/2, the two main lobes overlap and form
one single end-fire beam, like the beam of f 02 (θ, ϕ) in this case.
At the peak of a mode function, all other mode functions are zeroes. We can write
that
{
( ) 1, if p = m and q = n
f mn θ pq , ϕ pq = . (7.31)
0, otherwise
Similar to the situation of the line source, the propagation modes belong to the
propagation set P of the current sheet. They can significantly contribute to the far
field. On the contrary, the modes that do not belong to the set P are evanescent modes.
The peak direction of each propagation mode corresponds to a point in the unit
spherical surface in the k-space. All dots from the propagation modes form a constel-
lation on the k-surface. An example of the constellation of the propagation group
in the k-space is shown in Fig. 7.6. Each circle corresponds to the wave vector
( )
k xmn , k ymn , k zmn of a propagation mode. As all propagation modes are mirror
symmetrical with respect to the source plane, only the upper half of the constel-
lation is illustrated in Fig. 7.6. Like in the one-dimensional case, we can verify that if
(m, n) ∈ P, k zmn is real. The mode is a propagation mode in ±z-direction. Otherwise,
the mode is an evanescent one. In the peak direction, the evanescent mode decays
exponentially when it leaves the source plane. The main lobes of a propagation mode
fall in the visible region of the source. However, for an evanescent mode, only some
of its sidelobes enter the visible region. Consequently, the far field is mainly deter-
mined by the propagation modes. Although part of the sidelobes of the evanescent
modes may fall into the visible region and contribute to the far field, the effect is
much smaller than that of the propagation modes.
(C) Two Parallel Planar Sources
The far field of a single layer of current sheet is mirror symmetrical with respect
to the source plane. The symmetry may be destroyed if there are multiple layers of
currents.
Consider the two current sheets depicted in Fig. 7.7. They are placed in parallel
to the xoy plane. Both current sheets are rectangularly shaped with the same size of
Dx × D y . We choose Dx = N x λ and D y = N y λ again. The center of the upper current
sheet J1 locates at (0, 0, d), while the center of the lower current sheet J2 locates
at (0, 0, −d). Each current has a x-component and a y-component. The radiation
far field of the two current sheets can also be decomposed into two polarization
components,
( )
F(θ, ϕ) = Fdi p,x (θx )Fa,x (θ, ϕ)θ̂x + Fdi p,y θ y Fa,y (θ, ϕ)θ̂ y (7.33)
which is of the same form as Eq. 7.20. However, the continuous array factors for the
two parallel current sheets are changed to
/
{
Dy 2
{x / 2
D
[ ]
Fa,x (θx ) = e jkx x+ jk y y e jkz d I1x (x, y) + e− jkz d I2x (x, y) d xd y (7.34)
/
−D y 2 −Dx / 2
7.1 Electromagnetic Far Field in Free Space 243
/
{
Dy 2
{x / 2
D
( ) [ ]
Fa,y θ y = e jkx x+ jk y y e jkz d I1y (x, y) + e− jkz d I2y (x, y) d xd y (7.35)
/
−D y 2 −Dx / 2
where θx , θ y , θ̂x , and θ̂ y are defined exactly in the same way as in the single current
sheet.
Expanding the two layers of currents separately with two-dimensional Fourier
series,
⎧ ∞ ∞
⎪
⎪ ∑ ∑
⎪
⎪
⎪ I1x (x, y) = I1xmn e j (mΩx x+nΩ y y )
⎪
⎪
⎪
⎪ m=−∞ n=−∞
⎪
⎪ ∞ ∞
⎪
⎪ ∑ ∑
⎪
⎪
⎪ I 2x (x, y) = I2xmn e j (mΩx x+nΩ y y )
⎨
m=−∞ n=−∞
(7.36)
⎪
⎪
∞
∑ ∑∞
⎪ I (x, y) =
⎪ I1ymn e j (mΩx x+nΩ y y )
⎪
⎪ 1y
⎪
⎪
⎪
⎪
m=−∞ n=−∞
⎪
⎪ ∞
∑ ∑∞
⎪
⎪
⎪
⎪
⎩ I 2y (x, y) = I2ymn e j (mΩx x+nΩ y y )
m=−∞ n=−∞
(7.37)
The continuous array factors are expressed with the expansion coefficients of the
current as
244 7 Synthesis of Far Field Patterns
⎧ ∞ ∞
⎪ ∑ ∑ [( )]
⎪
⎪ F (θ ) = D D I1xmn e jkz d + I2xmn e− jkz d f mn (θ, ϕ)
⎪
⎨ a,x x x y
m=−∞ n=−∞
. (7.38)
⎪
⎪ ( ) ∑∞ ∑∞
[( )]
⎪
⎪ − jk z d
⎩ Fa,y θ y = Dx D y + I2ymn e f mn (θ, ϕ)
jk z d
I1ymn e
m=−∞ n=−∞
The mode function f mn (θ, ϕ) and the constellation for the two-layer currents are
respectively the same as those for the single layer sheet with its center at the origin, as
defined by Eqs. 7.27 and 7.32, respectively. The displacement in z-axis only causes
a phase-shift in all the propagation modes, which is exp( jk z d) for the upper sheet
and exp(− jk z d) for the lower one. For evanescent modes, the displacement in z-axis
causes an additional fading factor of exp(−|k z |d). The far fields of each current sheet
are mainly determined by the propagation modes associated with that current sheet.
Weighting the far fields from the two current sheets by the phase shift corresponding
to the center position in the z-axis and summing them up, we can obtain the total far
field, as expressed by Eq. 7.37
(D) Source Arrays
Consider a source array with N elements. Assume that the n-th element has a current
source Jn (r1 ) distributing over region Vsn , and its center locates at rcn . The far field
of the array is expressed by
∑
N
F(θ, ϕ) = e jk0 r̂·rcn Fn (θ, ϕ) (7.39)
n=1
Fn (θ, ϕ) = − I−rr ·
4π Vsn
If all elements have the same radiation far field Fele (θ, ϕ) but have different
excitations In , namely, Fn (θ, ϕ) = In Fele (θ, ϕ), then the far field of the source array
can be expressed in a simplified form as
∑
N
f a (θ, ϕ) = In e jk0 r̂·rcn . (7.41)
n=1
Equation 7.40 is the pattern multiplication principle. It is popular for predicting the
radiation patterns of antenna arrays. In the design of an antenna array, we may choose
an antenna element with required stand-alone radiation pattern and concentrate on
7.1 Electromagnetic Far Field in Free Space 245
the synthesis of the array factor. However, when all elements are put together to
form an array, there inevitably exist mutual electromagnetic couplings among them.
The active radiation patterns of the elements may change and become different to
their stand-alone patterns. The total radiation pattern of the antenna array can be
predicted using the pattern multiplication principle only approximately even if all
elements have identical structures and have identical stand-alone radiation patterns.
Therefore, the realized radiation pattern has to be analyzed with Eq. 7.39 if higher
accuracy is required.
(E) Spherical Harmonic Modes
Since the transverse components dominate the far fields, the radial component in the
spherical harmonic expansion of the electric far field can be discarded. As a result,
the spherical mode expansion for the electric far field is generally expressed by
∞ ∑
∑ n
[ ]
E f ar (r) ≈ n (θ, ϕ) + eπ nm (r )πn (θ, ϕ) .
eψnm (r )ψm m
(7.42)
n=1 m=−n
The expansion coefficients are given by Eqs. 4.96 and 4.100. The excited spherical
modes of a given source can be determined with the two coefficients anm and bnm
defined in Eq. 4.129,
{r0 [ √ ]
d
anm = −η0 [r1 jn (k0 r1 )]r1 Jπ nm (r1 ) + Q n r1 jn (k0 r1 )Jr nm (r1 ) dr1 (7.43)
dr1
0
{r0
bnm = −η0 k0 jn (k0 r1 )r12 Jψnm (r1 )dr1 . (7.44)
0
Based on the asymptotic behavior of the spherical Hankel function for large
argument, the field part in the Green’s function becomes [5],
1 n+1 − jk0 r
gn (r ) = r h n (k0 r ) ≈ j e
k0
| |
Fig. 7.8 Radiation patterns for the even modes of |ψm |
n (r) (n = 15)
In this section, we are to determine the source distribution in free space for
synthesizing a prerequisite radiation pattern.
Categorized in a brief manner, there are mainly two kinds of methods for synthe-
sizing radiation patterns. The first kind of approaches are based on optimization
techniques. The second group includes direct synthesis methods. We at first summa-
rize the fundamental principles of the two methods separately, and then discuss a
hybrid synthesis method that takes the direct synthesis result as an initial state and
perform further optimization.
7.2 Methods for Synthesis of Far Field Patterns 247
A general optimization procedure for the synthesis of a radiation far field is to find
the source distribution J(r1 ) in domain Vs that can minimize the error function
{
| |
Err = |Fobj (θ, ϕ) − F J (θ, ϕ)|2 dΩ (7.47)
Ω
where Fobj (θ, ϕ) is the required radiation pattern function and F J (θ, ϕ) is the
radiation function of the current J(r1 ) calculated with Eq. 7.4.
To implement the optimization scheme, we have to discretize Eq. 7.47 and replace
it by some kind of approximate expressions with finite degrees of freedom. This trans-
formation usually includes the discretization of the current source and discretization
of the radiation function.
(1) Discretization of the Current Source
At first, we need to discretize the current distribution in the source domain. Two
methods are commonly used: (1) to express the current with discrete dipoles; (2) to
expand the current with vector basis functions.
In the first method, we directly sample the current distribution at Ns properly
chosen discrete points. The current source is emulated with the infinitesimal dipoles
at the sampling positions. The amplitudes and directions of the dipoles can be adjusted
for minimizing the error function. Obviously, this is basically a spatial impulse
sampling.
Denote In and p̂n as the amplitude and the unit vector of the current density of the n-
th dipole at the sampling point rsn , respectively. The current source is approximately
expressed by
∑
Ns
J(r1 ) ≈ In δ(r − rsn )p̂n . (7.48)
n=1
j ωμ0 ∑
Ns [ ( )
Δ Δ
] ∑ Ns
F J (θ, ϕ) = − In e jk0 r̂·rsn I − r r · p̂n = In e jk0 r̂·rsn Fn (θ, ϕ).
4π n=1 n=1
(7.49)
In this case, Fn (θ, ϕ) is actually the normalized radiation pattern of the n-th
infinitesimal dipole element.
If we adopt the second method, some general vector basis functions can be used
for expanding the current source. According to the support of the basis function, we
may roughly classify the vector basis functions as low order basis functions and entire
domain basis functions. If the source region is partitioned with a mesh structure, a
low order basis function is defined over a single mesh or a pair of meshes, while an
248 7 Synthesis of Far Field Patterns
entire domain basis function is usually defined on the mesh structure of a subdomain
or on the entire domain of the source region. In this case, an entire domain basis
function is generally an aggregate of the low order basis functions.
• Low order vector basis function
The most popular low order vector basis function for surface current sources is the
RWG basis function defined on triangular mesh structures. A RWG basis function
f RW G (r) is defined on a pair of triangles Tn± with a common edge ln and two free
nodes rn± [6], as shown in Fig. 7.9a,
⎧ ln ( )
⎨ 2 A+ r(− rn+ , ) r ∈ Tn+
f RW G (r) = − 2lAn− r − rn− , r ∈ Tn−
⎩
0, elsewhere
Multi-branch RWG (MB-RWG) is a natural extension for the RWG basis function.
A MB-RWG basis function f M B−RW G (r) is defined on a positive triangle Tn+ and
−
several negative triangles Tn,i , i = 1, · · · , Nn . The positive triangle and the negative
triangles share the common edge ln [7], as shown in Fig. 7.9b. The expression for
f M B−RW G (r) is similar to that of f RW G (r), only with the negative part replaced by
the multi-branch triangles.
For a volume current density, we can adopt tetrahedral mesh structure and expand
the current with the Schaubert–Wilton–Glisson (SWG) basis function, which is a
popular low order basis function for volume currents. A SWG basis function is
defined on a pair of tetrahedrons tn± with the common face Sn and two free vertices
rn± [8], as shown in Fig. 7.9c,
⎧ Sn ( )
⎨ 3V + r(− rn+ , ) r ∈ tn+
f SW G (r) = − 3VSn− r − rn− , r ∈ tn−
⎩
0, elsewhere
RWG basis functions and SWG basis functions all have guaranteed the continuity
of the current in the normal directions on the interfaces. They are widely used in
numerical algorithms for analyzing electromagnetic radiation and mutual coupling
problems. If we adopt them to expand the current sources, many algorithms developed
in the computational electromagnetics society may be conveniently implemented for
far field pattern synthesis.
Expanding the current with the low order basis function bn (r1 ) gives
∑
Ns
J(r1 ) = In bn (r1 ). (7.50)
n=1
We may treat the current expanded with Eq. 7.50 as a kind of antenna array, in
which the current of one basis function is an array element. The normalized radiation
pattern of an element is the far electric field of the basis current,
{
j ωμ0 ( )Δ Δ
When the mesh structure of the source domain is fixed, Fn (θ, ϕ) is also determined.
For the basis function with regular shapes, Eq. 7.51 can be evaluated analytically. Note
that the elements in this kind of array are overlapped, which is the main difference
from the conventional arrays with separately placed discrete elements.
• Entire domain basis function
In analyzing the electromagnetic radiation and mutual coupling problems, the number
of the low order basis function is usually very large in order to obtain result with
satisfactory accuracy. If we use them directly as the basis function for synthesis, the
resultant optimization problem may become too time-consuming to solve. A better
way is to use higher order basis functions to catch the radiation property more effi-
ciently. The entire domain basis functions are defined over the whole source domain.
They can be generated using singular-value decomposition (SVD) or by solving
some kind of associated generalized eigen equations. Entire domain basis functions
can be expressed in terms of the low order basis functions weighting by the corre-
sponding singular vectors or eigen vectors [9]. Synthetic basis functions (SBFs) [10],
characteristic basis functions (CBFs) [11], and characteristic mode basis functions
(CMBFs) [12] are widely-used entire basis functions in computational electromag-
netics. They can be readily applied for problems of synthesizing radiation patterns.
The entire domain basis functions can grasp the electromagnetic radiation charac-
teristics of the sources more accurately. To achieve similar accuracy, the required
number of the entire domain basis functions is usually much smaller than that of the
low order basis functions. Consequently, if we adopt the entire basis functions for
synthesis of radiation patterns, the optimization burden can be reduced significantly.
For current sources in regular shaped region, we can use harmonic basis functions
to expand the current density. Especially, as we have discussed previously, we can
250 7 Synthesis of Far Field Patterns
sphere. These functions can all be classified as entire domain basis functions because
they are defined over the whole source region. Their radiation fields are available
using the mode functions expressed by Eqs. 7.15, 7.27, and 7.46, respectively.
The current source expanded with entire domain functions can also be treated as
a kind of antenna array, the current of each entire domain function composes one
array element. Obviously, all elements are overlapped over the whole source domain.
Although the array structure has become complicated, the number of the elements is
largely reduced.
In practical applications, it is important to take into account the characteristics
of the radiating structures and the excitation ports. For antenna arrays, the current
distribution on an element can be well characterized by the entire basis function
associated with the so-called antenna mode, i.e., the current on the element when
it is excited alone. If we ignore the mutual coupling, then all elements have the
same entire basis functions, hence, have the same radiation pattern of Fele (θ, ϕ).
The differences in the position and the excitations are taken into account in the array
factor. If we want to include the structural parameters simultaneously in the synthesis,
the problem becomes much more complex because the antenna modes may change
in the optimization process [13].
(2) Discretization of the Pattern Function
Secondly, we have to discretize the radiation pattern functions Fobj (θ, ϕ) and
F J (θ, ϕ). Two methods are commonly used for this purpose: the point matching
method and the mode matching method.
• Point matching method
We sample the far fields at M discrete directions (θm , ϕm ), m = 1, · · · , M, and
minimize the error function,
∑
M
| |
Err = |Fobj (θm , ϕm ) − F J (θm , ϕm )|2 . (7.52)
m=1
Discretizing the source with one of the methods discussed in the previous section
yields the matrix form of the error function
[ ]H [ ]
Err = ZI − Bobj · ZI − Bobj (7.53)
[ ]t
where I = I1 , I2 , · · · , I Ns is the column vector containing the expansion coef-
[ ]t
ficients of the current source, B = Fobj,1 , Fobj,2 , · · · , Fobj,M is column vector
containing the sampling data for the objective radiation pattern function, and Z is
the transfer matrix with size of M × Ns . The upper script “t” means transpose and
“H” represents conjugate transpose.
7.2 Methods for Synthesis of Far Field Patterns 251
Making use of the orthogonality of the vector basis functions and truncating the
terms in the summation to a properly chosen mode degree Ntr , we can write the error
function as
n (
∑
Ntr ∑
| | | | )
Err = | f obj,ψnm − f J,ψnm |2 + | f obj,π nm − f J,π nm |2 . (7.56)
n=1 m=−n
As pointed out previously, the synthesis problem becomes much more complicated
if we take the structural parameters of the antenna into the optimization process.
We are not going to dig into this complicated topic in this book. Many powerful
optimization methods have been successfully applied in antenna synthesis, such as
the genetic algorithms (GAs), particle swarm optimization (PSO) method, simulated
annealing (SA) algorithm, sequential convex optimization, ant colony optimization
method, and some other algorithms [14–19]. However, to get optimal solutions,
optimization algorithms are usually time consuming. If no special techniques are
taken to accelerate the algorithm, the synthesis time may increase exponentially with
the number of unknowns. Therefore, it may become quite difficult to use optimization
methods for the synthesis of very large antenna arrays.
252 7 Synthesis of Far Field Patterns
Direct synthesis methods aim to realize the radiation pattern of a discrete source or a
continuous current source without resorting to optimization algorithms. As early as in
1946, Woodward and Lawson proposed to realize a desired pattern by sampling it at
various discrete locations and interpolating it using a composing function associated
with a harmonic current of uniform amplitude distribution and uniform progressive
phase. The excitations of the source are then obtained by summing up the harmonic
currents required for the interpolation. Prony’s method can be used to realize a
specified pattern by determining the strengths of the N point sources at complex
positions z α , α = 1, . . . , N , each element has a directivity that can be slightly
controlled by the real part of the position [20, 21].
Consider the case that we sample both the source current and the radiation func-
tions at N discrete points. As the transfer matrix Z is now a square matrix, it seems
that we may directly synthesize the current source by solving the matrix equation
ZI = Bobj . (7.57)
However, for a source in a bounded region, the information carried in the far field
must be limited. The NDF of the far field is finite and is determined by its source.
If the far field is oversampled, then Eq. 7.57 is certainly ill-posed. If the far field is
under sampled, we may have not made full use of the information carried by the far
field and may be not able to obtain satisfactory synthesis result by solving Eq. 7.57.
Obviously, it is of great significance to correctly evaluate the NDF of the far field
associated with a source in a bounded region.
Theoretically, the normalized far electric field of a source in a bounded region in
free space can be expanded with spherical harmonic modes using Eq. 7.46. Therefore,
the NDF of the far field is infinite since it may consist of infinite number of orthogonal
spherical modes. However, for realizing a required radiation pattern Fobj (θ, ϕ), it is
not necessary to use all spherical harmonic modes. A commonly adopted strategy is
to approximate the radiation pattern with the spherical harmonic modes satisfying
the required accuracy. Or equivalently, we approximate the required radiation pattern
with the truncated sum of the spherical harmonic basis functions as
∑
Ntr ∑
n
[ ]
Fobj (θ, ϕ) ≈ Fsph (θ, ϕ) = f obj,ψnm ψm
n (r) + f obj,π nm πn (r) .
m
(7.58)
n=1 m=−n
The truncation degree Ntr of the spherical modes is determined in the approxi-
mation process. Since the order of the spherical harmonic modes is −n ≤ m ≤ n,
there are 2n + 1 spherical modes with the same degree n. The total number of modes
is counted to be 2Ntr (Ntr + 1) + 2Ntr . It is reasonable to define the total number of
the spherical harmonic modes as the required effective NDF of the far field in this
synthesis problem, namely,
Note that the factor 2 is added because there are TE modes and TM modes in the
field. In this way, we have converted the task of realizing Fobj (θ, ϕ) to the task of
realizing Fsph (θ, ϕ), the NDF of which is a finite number of N D F f ar .
Assume to realize Fsph (θ, ϕ) with a source in region Vs . Obviously, the effective
NDF of the current source should not be much smaller than N D F f ar in order to get
a satisfactory synthesis result. Therefore, we have to find a proper estimate for the
effective NDF of the source.
Denote the smallest spherical surface containing the source region by S0 . Its
radius is r0 . The far electric field of the source can be expanded with spherical
harmonic modes using Eq. 7.46. Since the vector spherical harmonic basis functions
are normalized, the amplitudes of| the far | fields of the spherical harmonic modes can
be evaluated by the coefficients | f ψnm | = |bnm | and | f πnm | = |anm |. In free space,
the two coefficients are given by Eqs. 7.43 and 7.44. Obviously, they are determined
by the current source and the distance to the source.
The upper bound of the amplitudes of the mode coefficients can be estimated to
be
⎛ ⎞
{r0 | |
√ {r0
| |
|anm |max = η0 ⎝ Jπnm
max | d [r1 jn (k0 r1 )]r1 |dr1 + Q n J max |r1 jn (k0 r1 )|dr1 ⎠
| dr | r nm
1
0 0
(7.61)
{r0
| |
|bnm |max = η0 k0 Jψnm
max | jn (k0 r1 )r 2 |dr1 (7.62)
1
0
| |
= |Jπ nm (r1 )|max , Jrmax | |
nm = |Jr nm (r 1 )|max , and Jψnm = Jψnm (r 1 ) max . For
max max
where Jπnm
the sake of convenience, we introduce three mode generation functions for evaluating
the excitation strength of the source. They are defined as follows
254 7 Synthesis of Far Field Patterns
⎧
⎪ {r0 | |
⎪
⎪ | d |
⎪
⎪ Tππ,n (k0 , r0 ) = η0 | | [r1 jn (k0 r1 )]||r1 dr1
⎪
⎪ dr
⎪
⎪
1
⎪
⎪
0
⎪
⎪ {r0
⎨ √
Tπr,n (k0 , r0 ) = η0 n(n + 1) | jn (k0 r1 )|r1 dr1 .
⎪
⎪
⎪
⎪
⎪
⎪
0
⎪
⎪ { r0
⎪
⎪
⎪
⎪ Tψψ,n (k0 , r0 ) = η0 k0 | jn (k0 r1 )|r12 dr1
⎪
⎩
0
With these quantities, Eqs. 7.61 and 7.62 can be concisely rewritten as
Equations 7.63 and 7.64 give the estimates for the upper bounds of the amplitudes
of the far fields of the spherical modes that can be generated by a current source with
max
maximum coefficients of Jπnm , Jπmax max
nm , and Jπ nm in the source region. All spherical
modes of the same degree have the same generation functions. Typical plots for the
generation functions are shown in Fig. 7.10. It can be observed that when n > n c ,
the amplitudes of the generation functions tend to decrease exponentially. n c is the
cutoff mode degree defined in Chap. 5 [22]. Their positions are indicated by the
small squares in the plots. Therefore, the far fields of the spherical modes with
degrees larger than n c tend to be very small even if the mode currents to excite them
are of the same level. It may be considered that the excitation efficiency for these
spherical modes is very low. For a current source bounded in a sphere with radius
of r0 , we may approximately think that the spherical harmonic modes with degrees
larger than n c have neglectable contributions to the effective NDF of the source.
Reasonably, we define the total number of the modes with degrees not larger than n c
as the effective NDF of the source within S0 , which is
[ (/ )]
N D Fc ≈ 2n c (n c + 1) + 2n c = 2 (k0 r0 )2 + 0.5 1 + 4(k0 r0 )2 − 1 . (7.65)
The operation symbol [x] means to take the integer nearest to x. Note that the
factor 2 is also added. For large k0 r0 , it can be simplified as
[ ]
[ ] A0
N D Fc ≈ 2 (k0 r0 ) = 2π 2
2
(7.66)
λ
Fig. 7.10 The three generation functions for different source region. Squares: cutoff mode degree.
a r0 = 1λ, 5λ, and 10λ. b r0 = 15λ, 25λ, and 35λ
spherical modes can effectively contribute to the far field of the source. The far fields
of the modes that do not satisfy the criterion n ≤ n c exponentially decay with the
increase of the degree n, hence, their contributions to the far field can be ignored.
Although the criterion may be chosen to be slightly larger than n c , N D Fc estimated
with Eq. 7.66 is generally recommended as the effective NDF for the source.
To fulfill the synthesis of the required radiation pattern with a volume source, we
usually take Ntr ≈ n c , and determine the radius of the source region with
√
r0 ≥ Ntr (Ntr + 1)/k0 . (7.67)
Under this condition, we may solve Eq. 7.57 directly or using some kind of
optimization algorithms.
However, what we have determined is the effective NDF that the source should
possess in order to realize a satisfactory synthesis. Obviously, the source is not
necessary to be unique. This provides us redundant space to take into account the
structure limitations on the radiators in the optimization.
An alternative way is to carry out the synthesis in other mode function spaces
instead of the spherical harmonic mode space, such as the mode function space for
the line sources and that for the current sheets. In these mode spaces, the line sources
and the rectangular current sources can be determined almost uniquely, as will be
discussed in later sections.
In the synthesis of radiation pattern with line sources and planar sources, we obtain
a continuous line current I (z) in the one-dimensional case and a continuous current
sheet I (x, y) in the two-dimensional case. For the sake of convenience, we may
256 7 Synthesis of Far Field Patterns
( )
generally merge the polarization factor Fdi p, p θ p into the radiation pattern function
F(θ, ϕ) to get the continuous array factor.
( ) The realized far field of the continuous
current source always includes Fdi p, p θ p .
Now we are to discuss the relationship between the radiation field of a continuous
current source and that of an antenna array with discrete radiation elements. Based on
the sampling theory, we are to reveal the relationship between the continuous array
factor of a continuous current and the discrete array factor of an array with discrete
elements. We take the one-dimensional case as example. The results and conclusions
can be extended to the two-dimensional situations.
Consider a line current I (z) distributed over the whole z-axis. The Fourier
Transform of the current is expressed by
{∞
FI (k z ) = e jkz z I (z)dz. (7.68)
−∞
Here we deliberately use exp( jk z z) in order to be consistent with Eq. 7.8, where k z
is the z-component of the wave vector k. It can be considered as the spatial angular
frequency in z-direction. In the visible region in free space, we have k z = k0 cos θ .
Therefore, FI (k z ) is the spatial spectrum of I (z). If there exists a constant kmax such
that FI (k z ) = 0 for |k z | > kmax , then I (z) is spatially band-limited.
The line current can be obtained from its spectrum with inverse Fourier Transform,
{∞
1
I (z) = e− jkz z FI (k z )dk z . (7.69)
2π
−∞
Fa,z (k z ) = FI (k z ) ∗ Fw (k z ).
7.2 Methods for Synthesis of Far Field Patterns 257
It is understood that Fa,z (k z ) = Fa,z (k0 cos θ ) is a function different from Fa,z (θ ),
but we use the same function symbol for the sake of brevity.
Assume that we only use the propagation modes to synthesize the far field pattern.
The resultant line current I (z) is strictly spatially band-limited if it is periodically
distributed over the whole line, i.e., −∞ < z < ∞. As shown in Fig. 7.11a, in this
case, its spatial spectrum contains at most 2Nz + 1 discrete spectral lines in the range
of −k0 ≤ k z ≤ k0 with spacing of Ωz = 2π/Dz = k0 /Nz ,
∑
Nz
Fc (k z ) = In δ(k z − nΩz ). (7.72)
n=−Nz
When the current is confined within the domain −Dz /2 ≤ z ≤ Dz /2, its spectrum
is obtained to be
∑
Nz
Fa,z (k z ) = Fc (k z ) ∗ Fw (k z ) = In Fw (k z − nΩz ) (7.73)
n=−Nz
which is exactly the same as Eq. 7.19. Obviously, the line current source with finite
length is not spatially band-limited even if it consists of only the propagation modes.
However, the main beam falls in the visible region if there are no evanescent modes
included in the current, as illustrated in Fig. 7.11b.
In practical applications, it is generally difficult to realize a continuous current
distribution. We may need to spatially sample the continuous current and realize the
far field pattern using discrete elements. Assume that the sampling spacing is ds ,
which corresponds to a spatial frequency of ks = 2π/ds . Denote the sampling points
as z p = pds , p = −Ns , · · · , 0, · · · , Ns . We can express the sampled current with
the sum of a series of impulses,
Fig. 7.11 Typical spatial spectrum structure. a Periodic current over the whole line. b Line current
with finite length. Visible region: −k0 ≤ k z ≤ k0
258 7 Synthesis of Far Field Patterns
∑
Ns
( ) ( )
Id (z) = I zp δ z − zp .
p=−Ns
Based on Eq. 7.11, we obtain the discrete array factor Fd (k z ) for the sampled
current,
{
Dz /2
∑
Ns
( )
Fd (k z ) = e jkz z Id (z)dz = I z p e jkz z p . (7.74)
−Dz /2 p=−Ns
which is exactly the conventionally defined array factor for an array antenna with
discrete elements. Moreover, making use of the property of Fourier Transform, we
can verify that Fd (k z ) is a periodic function of k z consisting of a superposition of the
shifted replicas of Fa,z (k z ),
∑ ∞
1 ∑
Ns
Fd (k z ) = I (z n )e jkz zn = Fa,z (k z − pks ). (7.75)
n=−Ns
ds p=−∞
Since the continuous array factor Fa,z (k z ) associated with a line current with
finite length is not spatially band-limited, as depicted in Fig. 7.11b, there are always
overlaps between the shifted replicas of Fa,z (k z ), which will inevitably cause aliasing.
For a broadside antenna array, we may synthesize it with a continuous line current
consisting of only propagation modes. Its main beam falls in the visible region. The
levels of the sidelobes outside the visible region are usually much smaller than that
in the visible region. If we sample the current with a spacing of half wavelength,
the related sampling spatial frequency is 2π/(0.5λ) = 2k0 in the k-space. It is 2
times higher than the largest spatial frequency k0 that bounds the visible region, so
the aliasing effect is expected to be small. However, if we sample with spacings
larger than 0.5λ, then severe overlaps of the replicas may cause large distortion to
the radiation pattern and grating beams may appear in the visible region.
The array factors obtained by sampling with a spacing of λ/2, λ/3, and λ/4 are
illustrated in Fig. 7.12. The corresponding sampling spatial frequencies are respec-
tively 2k0 , 3k0 , and 4k0 . The influence of the sampling spacing on the radiation pattern
in the visible region is zoomed-in in Fig. 7.13. Obviously, the aliasing not only affects
the sidelobe levels but also the ripples in the main beam. A smaller sampling spacing
corresponds to a larger sampling frequency and causes smaller aliasing effect.
We usually choose the spacing between two adjacent elements of the array to
be 0.5λ in our practical designs. The aliasing effect is basically small for broadside
antenna arrays. Although using higher sampling frequency may cause less aliasing
effect, but requires a larger number of elements.
For the end-fire antenna array, the aliasing effect may cause a back lobe with very
high level, as illustrated in Fig. 7.14. As its level is almost the same as that of the
main beam, it is actually a grating lobe.
7.2 Methods for Synthesis of Far Field Patterns 259
Fig. 7.12 The periodic array factors in the range of −3k0 ≤ k z ≤ 3k0 . Sampling spacing: a λ/2,
b λ/3, c λ/4. Visible region: −k0 ≤ k z ≤ k0
Fig. 7.13 The influence of the sampling spacing on the radiation pattern in the visible region
Fig. 7.14 Aliasing effect on an end fire array. a Continuous current with a single end-fire main
beam. b Array factor by sampling with λ/2 spacing
260 7 Synthesis of Far Field Patterns
Fig. 7.15 Aliasing effect on evanescent modes. a Continuous current with only evanescent modes.
b Array factor by sampling with λ/2 spacing
7.2 Methods for Synthesis of Far Field Patterns 261
Theoretically, if we use the continuous line current as the radiating source, the real-
ized far field of the continuous line current contains the polarization factor Fdi p,z (θ ).
In the visible region, the far field is Fa,z (k0 cos θ )Fdi p,z (θ ). It is always zero in the
direction of the axis of the line current. We cannot use a continuous line current
to realize an end-fire radiation even if the synthesized continuous array factor is an
end-fire type.
(2) Antenna arrays with discrete elements can be realized by spatially sampling the
continuous current.
Three array factors are introduced in this book: the conventional array factor, the
continuous array factor, and the discrete array factor. The continuous array factor
Fa,z (k z ) is used for continuous current sources, such as the line currents and the
current sheets. We can treat a continuous current source as a continuous array, and
consider the polarization factor Fdi p,z (k z ) as its element factor. In practical appli-
cations, except some special distributions that can be created with typical radiation
structures, it is not convenient to realize a general continuous current. We may need to
spatially sample the continuous current and approximately realize the current using
discrete elements.
The discrete array factor Fd (k z ) is obtained by uniformly sampling the continuous
current source with spacing ds . Generally, we choose ds = λ/2. If we put the sampling
positions at the centers of the elements of an antenna array, then Fd (k z ) has exactly
the same expression as the conventional array factor Fa (k z ). Most importantly, the
discrete array factor Fd (k z ) is the superposition of the shifted replicas of Fa,z (k z ), as
shown in Eq. 7.75. It is efficient to synthesize the continuous array factor Fa,z (k z )
using the direct synthesis method, but it is usually not quite efficient to directly
synthesize the discrete array factor Fd (k z ) with optimization algorithms. Therefore,
to get Fd (k z ) from Fa,z (k z ) with Eq. 7.75 is a very good strategy despite the fact that
the final performance may be slightly deteriorated by the aliasing effect.
(3) Only propagation modes should be used for the synthesis.
If a continuous current is used for realizing the radiation pattern, the evanescent
modes in the current may affect the reactive electromagnetic energy in the radiating
system and total radiation efficiency, but they have little contribution to the continuous
array factor, as shown in Fig. 7.15a. However, the evanescent modes may have
significant effect on the discrete array factor Fd (k z ). As illustrated in Fig. 7.15b,
when the current is spatially sampled with spacing of λ/2, because of the aliasing
effect, the evanescent modes may cause grating beams in the visible region that
are not easy to predict or to control. To make matters worse, the presence of the
evanescent modes may deteriorate the behavior of the synthesis problem or even
make it ill-posed. Therefore, it is critical to exclude the evanescent modes from the
synthesis algorithm.
262 7 Synthesis of Far Field Patterns
The relationship Eq. 7.75 provides a solid basis for establishing an efficient hybrid
method for synthesizing the discrete array factors. The first step is to synthesize
a required continuous array factor with a continuous current source. The second
step is to sample the continuous current with uniform spacings in each direction.
We can realize it with discrete radiating elements whose current distribution can be
effectively controlled with some kind of excitations. The spatial sampling strategy
and the feeding techniques all have significant effect on the radiation performance
of the realized device.
Although the direct method is very efficient for synthesizing the continuous array
factor, the realized radiation pattern may usually have high sidelobe levels and large
ripples in the main beams, thus may probably not meet the practical requirements.
However, we will show that the direct synthesis result can provide a very good initial
basis for further optimization of the performance of the antenna array.
Consider a linear array at first. We always choose the length of the array as
Dz = Nz λ. Because the two end-fire modes can be used for synthesizing the array
factor, the NDF of the far field is 2N + 1. In practical designs, the phase information
of the far field is not easy to explicitly and accurately specify. A popular strategy
is to assume that the far field is of equi-phase and synthesize the radiation function
with real values using Eq. 7.19.
We are to synthesize the continuous array factor using a hybrid optimization
method at first, and then sample the continuous array factor to get the discrete array
factor. The hybrid optimization algorithm is composed as follows [23].
Step 1. Realize the continuous array factor with a continuous line current I (z) with
length of Dz = Nz λ, where the current only consists of the propagation
modes,
∑
Nz
I (z) = In e jnΩz z , Ωz = 2π/Dz . (7.76)
n=−Nz
∑
Nz
∑
Nz [ ]
Dz
Fa,z (k z ) = In f n (k z ) = Dz In sinc (k z + nΩz ) . (7.77)
n=−Nz n=−Nz
2
The standard direct synthesis procedure in Sect. 7.2.3 can be used to obtain In ,
which will be used as the initial values for further optimization.
Step 2. Optimize the sidelobe levels and the ripples in the main beams by
minimizing the error function
7.2 Methods for Synthesis of Far Field Patterns 263
∑
NS L
| ( SL ) ( SL )|2 ∑
N Ri p
| ( ) ( )|
Err = | |
Fa,z k z,m − FSL k z,m + | Fa,z k ri p − FRi p k ri p |2
z,m z,m
m=1 m=1
(7.78)
SL
where k z,m is the location of the m-th peak of the sidelobe in the range of −k0 ≤
ri p
k z ≤ k0 and k z,m is the( location
) of the m-th ripple of the main lobe in the same range.
( SL ) ri p
FSL k z,m and FRi p k z,m are respectively the objective values for the far field at
( SL ) ( )
SL ri p ri p
k z,m and k z,m . Fa,z k z,m and Fa,z k z,m are respectively the values of the radiation
pattern at these positions calculated with Eq. 7.77. If the error is less than the criterion
for convergence, the optimization algorithm is stopped. Otherwise, renew In with
Eq. 7.79 and repeat the previous steps,
( )(q)
∂ Err
In(q+1) = In(q) −β (7.79)
∂ In
where β is a constant for adjusting the converging speed. (∂ Err/∂ In )(q) is calculated
by taking the derivatives of Eq. 7.78 with respect to In .
Step 3. Calculate the optimized continuous current I (z) with Eq. 7.76. Sample I (z)
at the 2Ns + 1 discrete points rsp = pds ẑ to obtain the current coefficients
for the array factor,
( )
Is ( p) = I rsp , p = −Ns , · · · , 0, · · · Ns (7.80)
where ds is the spatial sampling spacing. The common choice is to set ds = λ/2 and
Ns = Nz . The discrete array factor is determined by
∑
Ns
f d (k z ) = Is ( p)e jk0 r̂·rsp . (7.81)
p=−Ns
The continuous array factor Fa,z (k z ) is a sum of the sinc type mode functions,
which are all entire functions. As a result, the local maxima and the local minima
of Fa,z (k z ) are interlaced. The sidelobe peaks appear either at a maximum or at a
minimum. For sidelobes, the maxima of Fa,z (k z ) have to be assigned positive values
while the minima of Fa,z (k z ) have to be assigned negative values. For the ripples in
the main beams, the values assigned to the maxima have to be larger than the average
level of the main beam, while the values assigned to the minima have to be smaller
than the average level. Otherwise, the optimization algorithm may not converge or
may converge to a local optimal state.
The NDF of the far field gives a limitation to the unknowns in the optimization
algorithm. Basically, the total number of the sidelobe levels and the main lobe ripples
to control should be approximately equal to that of the NDF of the far field. If the
264 7 Synthesis of Far Field Patterns
unknowns to control is too large, the optimization algorithm tends to become ill-
posed. On the contrary, if it is too small, the results may be not satisfactory because
we have not made full use of the freedoms of the source.
We can combine the ripples and sidelobe peaks together and rewrite Eq. 7.78 in
matrix form as
( )H ( )
Err = ZI − Fobj · ZI − Fobj (7.82)
[ ]t
I = I−N , · · · , I0 , · · · , I N is the (2N + 1) × 1 column vector containing the
expansion coefficients of the current I (z) expressed by Eq. 7.76. Fobj is the N Peak ×1
column vector containing the prerequisite values of the sidelobe levels and the ripples
in the main beams, i.e.,
[ ( ) ( peak ) ( )]t
peak peak
Fobj = Fobj k z,1 , · · · , Fobj k z,m , · · · , Fobj k z,N Peak .
( )
peak
N Peak = N SL + N Ri p . Fobj k z,m is the preset value of the continuous array
factor at the peaks of the sidelobes or the ripples. Z is the transfer matrix with size
of N Peak × (2N + 1). Its entries are expressed by
[ ]
( peak ) Dz
Z (m, n) = Dz sinc k z,m + nΩz . (7.83)
2
Take care that the entries change their values at each iteration, so we cannot solve
the optimization problem directly with least square method.
The line current locates on the z-axis with length of Dz = Nz λ. In free space,
each harmonic component of the current source generates an electromagnetic field
described with a sinc type mode function, as expressed by Eq. 7.15, which has the
same form as the composing function in Woodward and Lawson method. The radi-
ation pattern required to be synthesized is given by Fobj (θ ). It can be approximated
with the superposition of the mode functions as
∑
Nz
Fobj (θ ) = Fdi p,z (θ )Dz In f n (θ ). (7.84)
n=−Nz
Naturally, we can use the mode function f m (θ ) as the test function to discretize
Eq. 7.84 and get its matrix form
7.3 Pattern Synthesis for Line Source 265
TI = Fobj . (7.85)
[ ]t
I = I−Nz , · · · I0 , · · · I Nz is the column vector containing the coefficients of the
[ ]t
linear current. Fobj = Fobj,−Nz , · · · Fobj,0 , · · · Fobj,Nz is the column vector for the
required radiation pattern with
{π
Fobj,m = Fobj (θ ) f m (θ )dθ .
0
Fobj (θn ) 1
In = = Fa,z (θn ). (7.86)
Dz Fdi p,z (θn ) Dz
θn is the peak direction of the n-th mode beam. Obviously, there is no necessary to
solve any equation or utilizing any optimization algorithm. The current distribution
can then be calculated directly or using the Fast Fourier Transform (FFT).
We have merged the polarization factor Fdi p,z (θ ) into the radiation pattern in
Eq. 7.86. As a matter of fact, what we have directly synthesized with Eq. 7.86 is
the continuous array factor Fa,z (θ ). A qualified prerequisite radiation field Fobj (θ )
for a continuous line current must be of the form of Fa,z (θ )Fdi p,z (θ ). Specifically,
it must be satisfied that Fobj (0) = 0 and Fobj (π ) = 0. Note that Fa,z (θ ) may take
nonzero but finite values in the whole visible region and can be of the array factor of
a broad side antenna or an end-fire antenna. In practical designs, it is reasonable to
focus on synthesizing the array factor. If we plan to realize the radiation pattern with
continuous line current, the realized far field pattern is then obtained by multiplying
the array factor with Fdi p,z (θ ); If we plan to realize the radiation pattern with an
antenna array consisting of discrete radiation elements, then we can sample the line
266 7 Synthesis of Far Field Patterns
current and get the discrete array factor. The realized far field pattern is obtained by
multiplying the element factor Fele (θ ) instead of Fdi p,z (θ ).
Five examples are provided to validate the synthesis method. In the examples, the
far field patterns of the prototypes to be synthesized are all continuous array factors.
The realized far field pattern is calculated with Eq. 7.18. The normalized radiation
pattern is shown in Fig. 7.17. The sidelobe levels are as high as -16 dB.
Fig. 7.16 Prototype of the continuous array factor. a Single pencil beam. b With transition bands
Fig. 7.17 Realized pattern for the single beam antenna. a Polar pattern. b Main beam
7.3 Pattern Synthesis for Line Source 267
If we assign a transition band of 5° at both sides of the main beam, and set the
sidelobe level to be 0.002 (−54 dB), the realized sidelobe levels are greatly reduced,
as shown in Fig. 7.18.
The line currents used to realize these patterns are calculated with Eq. 7.12, only
including the 201 propagation modes. The normalized amplitudes for the resultant
line currents are plotted in Fig. 7.19. The current for the antenna with transition bands
is more concentrative to the center.
Example 7.2 Single beam line antenna with equi-ripples and equi-sidelobe levels.
peak
Step 1. Find the locations of all extrema k z,m of the base pattern in the visible
( 0 ≤ )k z ≤ k0 and calculate the values of the continuous array
region of −k
peak
factor Fa,z k z,m at the extrema.
Step 2. Calculate the transfer matrix with Eq. 7.83 and renew In with Eq. 7.79.
Step 3. Evaluate the error function. If the convergence criterion is satisfied, then
stop; otherwise, repeat the previous steps.
We set Err = 0.001 in this example. The convergence behavior of the algorithm
is shown in Fig. 7.22. The algorithm converged after 21 iterations. The optimized
7.3 Pattern Synthesis for Line Source 269
result is shown in Fig. 7.20a and b. The ripples and sidelobe levels all meet the
requirements.
The line current obtained by using the direct synthesis method and that after
optimization are plotted in Fig. 7.23. In order to realize the discrete array factor, we
have sampled the optimized current with spacing of 0.5λ, as depicted with the blue
circles in Fig. 7.23. In spectral domain, it corresponds to a sampling frequency of 2k0 .
The corresponding normalized discrete array factor is shown in Fig. 7.24. Because
of aliasing effect, the ripples and the sidelobe levels all have deteriorated slightly.
We have also sampled the current at a closer spacing of 0.25λ, corresponding to a
higher sampling frequency of 4k0 . The discrete array factor is shown in Fig. 7.24. It
can be clearly seen that the aliasing effect is smaller than that of the 0.5λ spacing.
Example 7.3 Antenna with equi-ripples and step-wise equi sidelobe levels.
range of 120◦ < θ or θ ≤ 60◦ . We are to synthesize it using a line source with length
of L = 100λ.
The continuous array factor obtained using the direct synthesis method and opti-
mization algorithm are shown in Fig. 7.25a. After optimization, the ripples and the
sidelobe levels satisfactorily meet the requirements in the whole visible region.
The discrete array factors obtained by sampling the continuous current with
0.5λ spacing and 0.25λ spacing are shown in Fig. 7.25b. When sampling with
0.5λ spacing, the sidelobe levels at the two edges of the discrete array factor have
slightly increased because of the aliasing effect, while they are close to the optimized
continuous array factor when sampling with 0.25λ spacing.
The synthesized continuous currents are plotted in Fig. 7.26, together with the
sampling points with 0.5λ spacing for the discrete array. For such a broadside
radiation, the current distributions vary relatively slowly in space.
Example 7.4 Antenna with equi-ripples and very low sidelobe levels.
The sidelobe levels can be made lower if we set a transition region at both edges of
the main beam. We make two modifications of the optimization algorithm. Firstly, we
exclude the nearest sidelobe at each edge of the main beam in the iteration and leave
them varying freely. Secondly, we do not require all sidelobes have equal levels, but
only require that the levels of all sidelobes, except the two first sidelobes, are lower
than −60 dB. The optimization procedure converged after 75 iterations with error of
0.0001. The optimized continuous array factor is shown in Fig. 7.27a. Although the
two first sidelobe levels are quite high, the edges of the main beam are very sharp.
The other sidelobe levels and the ripples in the main beam are controlled very well.
Since the sidelobe levels are very low, the discrete array factors obtained by
sampling the continuous current with 0.5λ spacing and 0.25λ spacing are almost
unaffected by the aliasing effect, as shown in Fig. 7.27b.
Fig. 7.28 Synthesized current. a Real part. b Real part of the current in 20λ ≤ z ≤ 50λ. c Imaginary
part. d Imaginary part of the current in 20λ ≤ z ≤ 50λ
The radiation pattern of a single planar current is always mirror symmetrical with
respect to the source plane. If the current has only x-component on the sheet with
size of Dx × D y = N x λ × N y λ, then its far field is linearly polarized and can be
expressed by the corresponding part in Eq. 7.25, which is
∞
∑ ∞
∑
F(θ, ϕ) = Fdi p,x (θx )Dx D y Ixmn f mn (θ, ϕ)θ̂x = Fdi p,x (θx )Fa,x (θ, ϕ)θ̂x .
m=−∞ n=−∞
(7.87)
The simplest synthesis strategy is to assume that the current consists of only those
propagation modes, i.e., (m, n) ∈ P. In this case, k zmn is real.
274 7 Synthesis of Far Field Patterns
Fa,x (θ, ϕ) is the required two-dimensional continuous array factor for the x-
polarized current sheet. Accordingly, the x-component of the current distribution
for realizing the prescribed x-polarized far field pattern can be explicitly expressed
as
1 ∑Nx ∑Ny
Ix (x, y) = Fa,x (θmn , ϕmn )e j (mΩx x+nΩ y y ) , (m, n) ∈ P (7.89)
Dx D y m=−N n=−N
x y
As an example, we are to realize a far field pattern with the shape of two letters
“6G”. The source area is chosen to be 60λ × 60λ. We directly synthesize the current
distribution with the following steps,
Step 1. Define the area for “6G” in the k x − k y plane where k z is real, which is the
blue disk in Fig. 7.30. The pattern values are assigned to 1.0 within the main
beam; to 0.5 at the edges of the main beam; and 0.0 at other places.
Step 2. Use Eq. 7.29 to select the mode whose peak direction (θmn , ϕmn ) falls in the
“6G” area.
Step 3. Use these selected modes to synthesize the current distribution with Eq. 7.89.
In order to plot the realized far field pattern, Eq. 7.87 can be used to calculate
the data for F(θ, ϕ) · θ̂x at the selected directions.
In this example, there are totally 11,289 propagation modes. Among them, 1504
modes are selected with their main beams pointing to the “6G” pattern.
The realized far field pattern is shown in Fig. 7.31. Since it is symmetrical with
respect to the source plane, only the upper half part of the radiation pattern is plotted.
The amplitude of the current distribution is shown in Fig. 7.32. Apparently, the
current is concentrated in the center area and may have very small amplitude in most
of the other areas.
In general situations, the current usually have two orthogonal components. The
synthesis procedure is still applicable. Assume to synthesize the radiation pattern
Fig. 7.31 Realized far field pattern. a Front side view. b Top view
276 7 Synthesis of Far Field Patterns
with only propagation modes. Making use of the interpolation formula Eq. 7.25, we
obtain the sampling equation at (θmn , ϕmn ),
1 ( )
F(θmn , ϕmn ) = Fdi p,x (θxmn )Ixmn θ̂xmn + Fdi p,y θ ymn I ymn θ̂ ymn . (7.90)
Dx D y
Recall that (θmn , ϕmn ) is the peak direction of the mn-th mode where the mode
function f xmn (θmn , ϕmn ) = 1. θ̂xmn and θ̂ ymn are the transverse unit vectors at that
direction. Separating the two components leads to
[ ( ) ][ ] [ ]
Fdi p,x (θxmn ) Fdi p,y θ ymn
( cos ) γ Ixmn 1 F(θmn , ϕmn ) · θ̂xmn
=
Fdi p,x (θxmn ) cos γ Fdi p,y θ ymn I ymn Dx D y F(θmn , ϕmn ) · θ̂ ymn
(7.91)
( )
with cos γ = θ̂xmn · θ̂ ymn . We can determine the sampling values for Ixmn and I ymn
by solving Eq. 7.91 and then obtain the two components of the current on the plate
for realizing the required radiation pattern.
For circularly polarized radiation, the far field F(θ, ϕ) consists of two orthogonal
components with roughly equal amplitude and a phase difference π/2. The current
can also be obtained with Eq. 7.91, so the synthesis procedure is applicable. Note that
at the directions around the z-axis,(θ is small. Hence,
) θ̂xmn and θ̂ ymn are approximately
orthogonal. We may simply take θ̂xmn · θ̂ ymn ≈ 0 in the synthesis procedure.
levels of the optimized continuous array factor are all below −40 dB, as shown in
Fig. 7.34b. However, when it is realized with discrete array consisting of elements
with 0.5λ spacing, several sidelobe levels at the edges may become higher than −
35 dB due to the aliasing effect, as shown in Fig. 7.34c.
The convergence property is shown in Fig. 7.35. In this example, the error
decreases fast at the beginning, but no longer decreases when Err ≈ 0.1. The
optimized current is plotted in Fig. 7.36.
Fig. 7.33 The C-shaped footprints. a Prototype. b Direct synthesis method. c Optimized. d Discrete
array by sampling with 0.5λ spacing
Fig. 7.34 The array factors. a Direct synthesized continuous array factor. b Optimized continuous
array factor. c Discrete array factor by sampling with 0.5λ spacing
278 7 Synthesis of Far Field Patterns
In order to show the optimization results more clearly, the continuous array factors
along the middle lines on the k x − k y plane are plotted in Fig. 7.37. The sidelobe
levels of the optimized pattern are clearly below the required level of –40 dB.
The antenna array obtained by sampling the continuous current with 0.5λ spacing
has 3721 elements in this example. For readers’ reference, we have tried to thin the
antenna array by simply dropping the element when the amplitude of the sampled
current is smaller than 1/64 of the maximum of the current. The number of the
element is reduced from 3721 to 685. The discrete array factor is shown in Fig. 7.38.
Both the sidelobe levels and the ripples in the main beam have significantly increased
due to the thinning [24, 25].
We want to emphasize several points about the hybrid algorithm here:
1. The key technique is that while controlling the values of the extremum points,
we limit the varying range of their positions but not fix their positions.
2. The minima and maxima of the patterns must be interlaced according to the
properties of the entire functions. It is important to properly assign the target
data for the sidelobe levels and the ripples.
3. For the line sources and planar current sheets, we can simply assign real values
to Fo , or complex values with identical phase, like in synthesizing circularly
7.4 Pattern Synthesis for Rectangular Planar Source 279
polarized beams. As can be deduced from Eq. 7.26, if they are assigned real
values, we can find a set of real Fourier coefficients Ixmn and I ymn for the real
target pattern, and the corresponding current components satisfy
{
Ix (x, y) = Ix∗ (−x, −y)
. (7.92)
I y (x, y) = I y∗ (−x, −y)
280 7 Synthesis of Far Field Patterns
Note that there are no other limitations on Fo except that it is real. If the current does
not satisfy Eq. 7.92, the realized patterns are generally not of equi-phase. However,
we cannot expect to realize with them a pattern that has much better performance
in controlling the sidelobe levels and the ripples in the main beams than what we
can synthesize using the currents that satisfy Eq. 7.92, as we can see that whether
the currents are subject to Eq. 7.92 or not, the effective NDFs of the target far field
patterns are the same, the numbers of the extrema to be optimized are also the same.
4. Another technique is also very important for improving the efficiency of the
algorithm. When we assign real values to the required radiation pattern, the
mode coefficients of the currents are all real, so the radiation patterns keep to be
real in the whole iterations.
5. We can borrow the techniques in computational electromagnetics and partition
the visible region in the k x − k y plane using a rectangular mesh structure with a
uniform cell size of about 0.1 λ × 0.1 λ. The mesh structure keeps unchanged in
the iterations. All extrema of the patterns are assumed to appear on the grids and
can be easily located by comparing it with the pattern values at the surrounding
grids.
In some situations, we need to control the radiation pattern in both sides of the source
plane. It may be required to form different multiple beams with different polarizations
to cover different areas at the two sides, as shown in Fig. 7.39. A natural solution
is to use two antenna arrays that are placed back-to-back for this purpose, with one
array responsible for one side. In order to avoid interference between them, the two
arrays have to be placed far enough away. We also have to carefully design the array
structures to reduce the level of their back lobes, or add metal backplanes between
them to reduce mutual couplings.
linearly
circularly linearly circularly
polarized
polarized polarized polarized
beams
beams beams beams
back
lobes
(a) (b)
Fig. 7.39 Concept for possible applications of radiators consisting of two current sheets. a Two
separate arrays. b Single antenna with two current sheets
7.4 Pattern Synthesis for Rectangular Planar Source 281
If two parallel layers of current sheets are used to replace the single current sheet,
the effective NDF of the far fields can be doubled. However, the effective NDF will
no longer significantly increase if more than two layers of current sheets are added.
We can use two layers of current sheets to remove the symmetry of the far fields and
control the radiation pattern in the two sides separately. The designs with two layers of
sources can be more compact and efficient than using two separate antenna arrays [3].
When two current sheets are placed in parallel, their radiation fields have a phase
difference due to the displacement of their locations. The overall far field pattern may
be not mirror symmetrical anymore. Although in general situations, the two layers of
currents may have different shapes and their propagation modes may have different
constellation structures, we here only discuss the case that the two parallel currents
have the same shape and the same size. The propagation group of each current sheet is
the same as that of the single current sheet discussed in the previous section. We will
show that the direct synthesis method for the single current sheet can be extended to
handle the case of two parallel current sheets. If the far field pattern in the upper half
space and the lower half space are specified, the propagation modes of the currents
on the two sheets can be directly synthesized.
We use the coordinate system in Fig. 7.7. Assume that the current sheets both
have only x-component. The θ̂x -polarized radiation far field of the two current sheets
can be expressed as
where the continuous array factor for the x-polarized two current sheets is
∑∑( )
Fa,x (θ, ϕ) = Dx D y I1xmn e jkz d + I2xmn e− jkz d f mn (θ, ϕ), (m, n) ∈ P.
m n
(7.94)
Similar to handling the single layer current, the property of the sinc functions
enables us to solve the coefficients of the currents directly with the following equation
1
I1xmn e jkzmn d + I2xmn e− jkzmn d = Fa,x (θmn , ϕmn ). (7.95)
Dx D y
The far fields at the peak directions of the propagation modes in both the upper
half space and the lower half space have to be used to solve the current coefficients.
Making use of Eq. 7.29 we have
{
k xmn = k0 sin θmn cos ϕmn = −mΩx = −k0 m/N x
. (7.96)
k ymn = k0 sin θmn sin ϕmn = −nΩ y = −k0 n/N y
From which k xmn , k ymn , θmn , and ϕmn can be determined. Hence, the z-component
of the wave vector is obtained by
282 7 Synthesis of Far Field Patterns
⎧ /
⎨ k 1 − (m/N )2 − (n/N )2 , θ ≤ π/2
k zmn =
0
/ x
(
y
)
mn
. (7.97)
⎩ −k 1 − (m/N )2 − n/N , θ > π/2
2
0 x y mn
Note that the sign is different at the upper and lower half space. Substituting it
into Eq. 7.95, we derive that
{ [ ]
I1xmn = −Amn e j2|kzmn |d Fa,x (θmn , ϕmn ) − Fa,x (π − θmn , ϕmn )
[ ] (7.98)
I2xmn = Amn Fa,x (θmn , ϕmn ) − e j2|kzmn |d Fa,x (π − θmn , ϕmn )
je− j|kzmn |d
Amn = . (7.99)
2Dx D y sin(2|k zmn |d)
The realized far field pattern can be evaluated with Eq. 7.93. The two current
distributions can be calculated with
⎧ ∑∑
⎪
⎪ I1x (x, y) = I1xmn e j (mΩx x+nΩ y y )
⎨
m n
∑∑ , (m, n) ∈ P. (7.100)
⎪
⎪ (x, = j (mΩx x+nΩ y y )
⎩ I 2x y) I 2xmn e
m n
In the following examples, we are to synthesize the array factors Fa,x (θ, ϕ). The
current sources are assumed to be composed of the propagation modes, including
the end-fire modes.
The prototype of the radiation pattern in the k-space is shown in Fig. 7.40. We
want to realize a non-mirror-symmetrical far field pattern with a cross-shaped foot-
print in the upper half space and a disk-shaped footprint in the lower half space. In
these examples, instead of defining the footprint from the specified radiation pattern,
we directly assign the footprint in the normalized k x − k y plane for the sake of
convenience. The size of the bar of the cross-shaped beam is assumed to be
⎧ ⎧
⎪ kx ⎪ ky
⎪
⎨ −0.1 ≤ ≤ 0.1 ⎪
⎨ −0.1 ≤ ≤ 0.1
k0 k0
, or
⎪
⎪ k ⎪
⎪ kx
⎩ −0.6 ≤ y ≤ 0.6 ⎩ −0.6 ≤ ≤ 0.6
k0 k0
Fig. 7.40 The prototype of the radiation pattern. a Upper half space; b Lower half space
The structure of the two current sheets is shown in Fig. 7.7. At the first try of
synthesis, we use a smaller source area and choose Dx = D y = 10λ. The distance
between the two sheets is set to be λ/4. In the normalized k x − k y plane shown
in Fig. 7.41, the yellow area is the footprint to be realized, the cyan area is the
propagation mode area with (m, n) ∈ P.
In this example, we can count that there are totally 317 propagation modes. With
the beam structure shown in Fig. 7.41, we can further check that 21 of the propagation
modes are used to form the cross-shaped footprint, and 29 of them are used to form
the disk-shaped footprint.
For the sake of brevity, we assume that the sampling value of the far field is 1.0
in the yellow area and zero in the cyan area, i.e.,
{
1.0, in yellow region
Fa,x (θmn , ϕmn ) = .
0, in cyan region
The normal perspective view and the upside-down view of the realized radiation
pattern are shown in Fig. 7.42. Because the far field pattern of the prototype is
assumed to change sharply in the edges of the footprint, and the size of the current
sheets is relatively small, the sidelobe levels in the realized pattern are relatively
Fig. 7.41 The radiation footprint in the normalized k x − k y plane. a Upper half space. b Lower
half space
284 7 Synthesis of Far Field Patterns
high. However, the cross-shaped footprint and the disk-shaped footprint are clearly
demonstrated.
∗
Since the far field are real-valued, we can check from Eq. 7.98 that I1xmn = I2xmn ,
the resultant currents on the two sheets are conjugate to each other. Hence in the
figures, we only plot the currents of one sheet. The normalized current distribution
on the upper sheet is shown in Fig. 7.43. The current distribution on the middle line
in the x-direction is shown in Fig. 7.44.
Fig. 7.43 The current on the upper sheet. a Real part. b Imaginary part
In order to reduce the level of the sidelobes and realize a radiation pattern that
resemble the prototype more closely, we re-synthesize it with a much larger source
area of Dx = D y = 100λ. In addition, we have added a transition point at the edges
of the main beam. The distance between the two sheets is still λ/4. Larger current
sheets result in more propagation modes and can provide higher capability to shape
the details of the radiation pattern. With the larger source area, there are totally 31417
propagation modes, 2821 of them are used for forming the cross-shaped footprint,
and 4161 of them are used to form the disk-shaped footprint.
The realized radiation pattern is shown in Fig. 7.45. As expected, the footprints
are much clearer and the sidelobe levels are much lower when larger current sheets
are used.
The normalized current distribution in the upper sheet is plotted in Fig. 7.46, and
the current distribution on the middle line of the sheet in x-direction is shown in
Fig. 7.47. They are calculated with Eq. 7.100. It can be seen that in most area of the
sheet the current is very small, which makes it possible for further thinning.
This example is used to demonstrate that the polarization and the radiation pattern
can both be controlled with a two-layer current source.
The specified radiation footprint is defined in the normalized k x − k y plane, as
shown in Fig. 7.48. It is a circularly polarized (CP) square beam in the upper half
Fig. 7.46 Current distribution on one sheet. a Real part. b Imaginary part
286 7 Synthesis of Far Field Patterns
space, the edge length of which is 0.125. The pattern in the lower half space consists
of two linearly polarized (LP) beams forming the shapes of the two letters “S” and
“J”, as shown in Fig. 7.48b. In the CP beam, the polar angle is very small. The unit
vector θ̂x and θ̂ y are almost perpendicular. We simply assume that the θ̂x component
of the far field is 1.0 and the θ̂ y component of the far field is j1.0. For the LP beams,
their far fields only have the θ̂x component with value of 1.0. Furthermore, we have
scaled by 0.5 all the amplitudes of those values at the edges of the beams so as that
we can avoid sharp transitions in the radiation pattern, which is depicted in blue in
Fig. 7.48.
We synthesize it using a source area of Dx = D y = 40λ. The distance between the
two sheets is λ/4. There are totally 5025 propagation modes in this case, 169 of them
are used for forming the square-shaped CP footprint, and 1346 of the propagation
modes are used to form the SJ-shaped LP footprint.
The realized radiation pattern is shown in Fig. 7.49. Note that the square beam is
circularly polarized and the SJ-beams are linearly polarized. The axial ratios (ARs) at
the cutting planes with different azimuthal angle ϕ are plotted in Fig. 7.50. It clearly
reveals that the ARs are less than 1 dB in the square CP beam where θ ≈ 0.
Fig. 7.48 The radiation footprint in the normalized k x − k y plane. a Upper half space. b Lower
half space
7.4 Pattern Synthesis for Rectangular Planar Source 287
The synthesized current distribution on one sheet is plotted in Fig. 7.51. The
x-component of the current distribution is much more complicated than the y-
component because x-component contributes to all beams in the two sides, while the
y-component mainly contributes to the CP beam.
An important issue is the effect of the distance between the two current sheets.
It can be seen that, although the interpolation of the pattern is mainly determined
using the sinc mode functions, the phase shift term exp( jk z d) is also a function
with respect to θ because k z = k0 cos θ . This will affect the interpolation behavior.
A simple strategy is to choose small d so that exp( jk z d) is a slow varying function
compared to the sinc functions. To further exploit this issue, we consider a broad-side
antenna array in which k zmn ≈ k0 in the main beam. From Eqs. 7.98 and 7.99 we
have
{ [ ]
I1xmn ≈ −Amn Fa,x (θmn , ϕmn ) − Fa,x (π − θmn , ϕmn )
[ ] . (7.101)
I2xmn ≈ Amn Fa,x (θmn , ϕmn ) − Fa,x (π − θmn , ϕmn )
1
|Amn | ≈ . (7.102)
2Dx D y sin(2k0 d)
Obviously, when d = λ/8, sin 2k0 d = 1, |Amn | takes its minimal value. According
to Eq. 7.101, for a specified radiation pattern Fa,x (θ, ϕ), the two current coefficients
|I1xmn | and |I2xmn | also reach their minimal values. This means that, for a broadside
antenna, if the two current sheets are λ/4 away, the currents required to realize the
prescribed pattern are nearly minimum. Because the mode currents are orthogonal,
we can consider that the total power is approximately minimum. In practical designs,
we may use some kind of optimization algorithm to optimize the total power to get
better efficiency.
It can be further checked that the phase shift caused by the term exp( jk0 d cos θ )
is smaller than π/4 for all propagation modes if we choose d = λ/8. The effect to
the interpolation is negligible as has been demonstrated with the example.
Assume to synthesize a far field pattern with a layer of surface current Js (θ, ϕ) on
a spherical surface with radius r0 . We put the center of the spherical surface at the
origin of the spherical coordinate system and carry out the synthesis with spherical
harmonic expansion method.
The far field of the surface current Js (θ, ϕ) can be expressed with Eq. 7.46. The
harmonic coefficients of the far field are separately related to the corresponding
harmonic components of the surface current,
⎧
⎨ f π nm = −η0 j n d [r0 jn (k0 r0 )]r0 Jπ nm
dr0 . (7.103)
⎩
f ψnm = −η0 k0 j n+1 jn (k0 r0 )r02 Jψnm
7.5 Pattern Synthesis for Current on a Spherical Surface 289
Jψnm and Jπ nm are the harmonic coefficients of the surface current expressed by
Eqs. 4.56 and 4.57. This is the forward problem. Note that there is no Jr nm component
in the surface current.
In a synthesis problem, we are to determine the surface current Js (θ, ϕ) from a
given far field pattern Fobj (θ, ϕ). The first step is to approximate the required pattern
Fobj (θ, ϕ) with Fsph (θ, ϕ), as shown in Eq. 7.58. The harmonic coefficients f obj,ψnm
and f obj,π nm for Fsph (θ, ϕ) can be calculated with Eq. 7.59. We need to determine
the truncation number of the spherical mode degree Ntr according to the required
approximation accuracy and then determine the size of the source region. However,
in practical designs, the size and shape of the source may be subjected to limitations
from the structure of the electronic system, the installation requirements, the cost,
and so on. Consequently, we must take into account the effect of these limitations
and choose a proper Ntr for truncating the spherical modes.
In the next step, we calculate the expansion coefficients of the surface current
from Eqs. 4.56 and 4.57. The results are
⎧ ωε0
⎪
⎨ Jπnm = − j n [u j (u )]' u f obj,π nm
0 n 0 0
ωε0 . (7.104)
⎪
⎩ Jψnm = − f obj,ψnm
j n+1 u 20 jn (u 0 )
∑
Ntr ∑
n
[ ]
Js (θ1 , ϕ1 ) ≈ Jψnm ψm
n (θ1 , ϕ1 ) + Jπnm πn (θ1 , ϕ1 ) .
m
(7.105)
n=1 m=−n
of choosing a larger Ntr is quite marginal. More importantly, the synthesis problem
may become ill-posed if Ntr is too large.
We use two examples to demonstrate the synthesis procedure.
We are to synthesize the pattern using spherical surface current with radius of
r0 = 5λ. We choose the truncation number of the mode degree as Ntr = n c = 31.
The amplitudes of f obj,ψnm and f obj,π nm are determined with Eq. 7.59 and the mode
coefficients of the surface current Jψnm and Jπnm can be obtained with 7.104. In
this example, they have nonzero values only at m = ±1. With the pattern given by
Eq. 7.106, we can further derive that Jψn,−1 = Jψn1 and Jπ n,−1 = −Jπn1 . Therefore,
the surface current can be expressed as
∑
Ntr
[ ( ) ( 1 )]
Js (θ1 , ϕ1 ) = Jψn1 ψ1n (θ1 , ϕ1 ) + ψ−1 −1
n (θ1 , ϕ1 ) + Jπn1 πn (θ1 , ϕ1 ) − πn (θ1 , ϕ1 )
n=1
7.5 Pattern Synthesis for Current on a Spherical Surface 291
from which we can calculate the current component Jsθ and Jsϕ on the spherical
surface. The results are plotted in Fig. 7.53. The amplitude of the total current density
along the meridian line with ϕ = 0 is plotted in Fig. 7.54. It is nearly symmetrical with
respect to the equator of the sphere, but the phase is approximately anti-symmetric.
The realized radiation pattern can be predicted by summing up the far fields of the
selected spherical harmonic modes. The radiation pattern in the xoz plane is shown
in Fig. 7.55a, and the three-dimensional pattern is sown in Fig. 7.55b. We want to
emphasize that the result is the realized radiation pattern of the continuous current
on the spherical surface. No polarization factor is needed in this situation.
Example 7.11 Antenna with two axially symmetrical beams.
The radiation pattern of the prototype to be synthesized is described as
{
cos ϕ θ̂ − sin ϕ ϕ̂, 0◦ ≤ θ ≤ 10◦ , 80◦ ≤ θ ≤ 100◦
Fobj (θ, ϕ) = . (7.107)
0, elsewhere
It has two main beams at 0◦ ≤ θ ≤ 10◦ and 80◦ ≤ θ ≤ 100◦ . Note that the polar-
ization of the far field changes with direction. We are to synthesize it using a surface
current on a larger spherical surface with radius of r0 = 10λ. The corresponding
cutoff mode number is n c = 63. We choose a slightly larger truncation number of
the mode degree as Ntr = 69. The surface current is plotted in Fig. 7.56, while the
realized pattern is plotted in Fig. 7.57.
| |
Fig. 7.53 Current distribution on the spherical surface. r0 = 5λ. a | Jsϕ |. b |Jsθ |. c |Js |
Fig. 7.55 Radiation patterns. r0 = 5λ. a On the xoz plane. b Three-dimensional pattern
| |
Fig. 7.56 Current distribution on the spherical surface. r0 = 10λ. a | Jsϕ |. b |Jsθ |. c |Js |
Obviously, as a larger current surface is used, the ripples in the main beam and
the sidelobe levels are all significantly reduced.
7.6 Summary 293
7.6 Summary
When we use a line current or current sheets to synthesize a radiation pattern, the
realized far field always contains the polarization factor caused by the infinitesimal
dipole. However, because of the diversity in the current, the polarization factor is no
longer inherently included in the far field if we synthesize it with a spherical surface
current.
In practical designs, it may be difficult to provide a qualified far field pattern with
exact values in the whole visible region, especially the phase information. We can
assume that all far fields have equal phases and simply assign them with real values.
A feasible strategy is to preset the vector values of the main beams of the pattern to
be synthesized at properly chosen sampling directions, and assume that the pattern
is zero except at the main beams. Then we can determine the expansion coefficients
of f ψnm and f π nm in two ways. The first one is to solve Eq. 7.54 with point-matching
method,
( ) ∑Ntr ∑
n
[ ( ) ( )]
Fobj θ p , ϕ p = n θ p , ϕ p + f obj,πnm πn θ p , ϕ p .
f obj,ψnm ψm m
(7.108)
n=1 m=−n
When the radius of the current surface is determined, the effective NDF of the
system is determined. We can choose the same number of sampling positions so that
we can solve Eq. 7.108 directly. The second method is to project the far field on
the unit sphere that is partitioned with a kind of mesh structure, then express the far
field with a kind of vector basis functions on the mesh, and finally use Eq. 7.59 to
calculate f obj,ψnm and f obj,πnm .
The procedure only demonstrates that a radiation pattern can be theoretically
synthesized with a current on the spherical surface. The algorithm may be extended
for synthesizing surface currents on other curvatures under certain conditions. Some
related issues will be discussed in Chap. 8. However, we have not provided discus-
sions on how to practically realize the obtained surface currents, which may be quite
difficult. Although it is possible to spatially sample the continuous surface current
to get a discrete antenna array on the spherical surface, the relationship between
the continuous array pattern and the discrete array pattern needs further investi-
gating. Meanwhile, we have to design proper feeding structures and evaluate their
impact on the performance carefully. It is possible to use a metal spherical shell as a
conducting ground to the spherical current, and put the other part of the device in the
interior region of the metal shell, including the main part of the feeding structures,
the amplifiers and the base-band circuits.
Electromagnetic mutual couplings exist between the elements in a practical
antenna array. To account for the mutual coupling effect, the active element pattern
(AEP) is often adopted to replace the original radiation pattern of the element [26].
In most situations, we may simply ignore the mutual couplings, or approximately
consider that the AEP of all elements are the same, hence, we can synthesize the
radiation pattern based on the pattern multiplication principle, and meantime, use
the hybrid optimization algorithm to synthesize the discrete array factor. However,
294 7 Synthesis of Far Field Patterns
the mutual couplings are generally different for elements at different positions. To
synthesize an antenna array with better performance, we have to address the elements
with different AEPs and cannot simply apply the multiplication principle. The hybrid
optimization algorithm needs to be modified in this situation.
References
22. Xiao GB, Hu M (2023) Nonuniform transmission line model for electromagnetic radiation in
free space. Electronics 12(6):1355
23. Xiao GB, Wang X, Zang T (2023) Optimization of far-field patterns based on direct synthesis
results. IEEE Trans Antennas Propag. https://doi.org/10.1109/TAP.2023.3321429 (In press)
24. Haupt RL (2005) Interleaved thinned linear arrays. IEEE Trans Antennas Propag 53(9):2858–
2864
25. Leeper D (1999) Isophoric arrays-massively thinned phased arrays with well controlled
sidelobes. IEEE Trans Antennas Propag 47(12):1825–1835
26. Pozar DM (1994) The active element pattern. IEEE Trans Antennas Propag 42(8):1176–1178