lEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 42, NO. 7, JULY 1994
&
&
I. INTRODUCTION
--
2dx
--
<U<-
x
2dx
<U<--
2dY
2dY
where U = sin(0) cos(4)
w = sin(0) sin(4).
When sampled with an N x N grid, where N is a power
of 2 for the convenience of the FFT,the actual sample points
will be
N
N
nX
U , = -n = --... - - 1
N dx
2
2
U,
mX
-N dY
m = --
... - - 1.
2
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 42, NO. I, JULY 1994
984
Fig. 1.
Summary of
routes
from
dx
Crosses indicate rectangular array element positions
Circles indicate triangular array element positions
Fig. 2.
the finite array. However there are a few subtle details that
also must be taken into account to make the method give the
best results.
The infinite array immersed element pattern is first divided
by an estimate of the isolated element pattern to deduce the
hypothetical pattern that would be produced by the infinite
array if it were made up of Huygens sources. This stage
is very important, as it enables the transform to the aperture to reveal the true excitations on the array spacing grid
(dz,dy). Next, rather than convolving the aperture excitations with the forward waves, it is more computationally
efficient-and numerically equivalent-to multiply the infinite
array Huygens source pattern with the finite array arrayfactor
before transforming back to the aperture. Fig. 3 also shows
a previous step that transforms the Huygens source pattern
back to the aperture, and then straight back to scan space
on a larger grid, if required, by padding the dataset with
zeros. This enables arrays that are larger than the original
N x N sample grid to be analysed. Back transforming the
product of the multiplication step now reveals the true aperture
excitations. These are truncated to the size and shape of the
ZAcT(m,IC) =
zmne2+(m--n)
n=l
d .
where k = - sin6
985
ri
I
9
'
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
Divideby
Isolated Element Pattern
ConvolutionReplacamt
3
el
Far-Fmld Pattern tor Hygens Sources
Fig. 3.
zmn
= Z(l m - n
I)
~ ( m1 - n l ) e 2 X j s ( m - n )
m, - =
n=l
where PK
-= I C
~ E Z .
ZACT
(g)
N-1
z(l -n
zne2nj+n
1
2 = -FFT(ZACT).
LzT1' I r(e,
l)e'rjG(-n).
4) 12 sin(e)cioci$ = 47r
n= --N
n= --N
zACT
N-1
zAcTp
(%
ZACT
_
N - 1
< k < 2with N points
N
E m ( 0 , d ) = I ( O , d )( a ,
where
+
+
b,)
I ( d , d ) a,(l
ACT,)
a, = Forward wave applied to element m,
1
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 42, NO. I, JULY 1994
986
m=l
m=l
i
where
M
'
i
Active Admittam
e l e m away
elen"
away
elemenm away
m=l
M = Number of elements.
It has been found that for typical finite arrays that have been
analysed using the admittance parameter approach, the value
of c corresponds to a correction of less than f 0 . 5 dB. This
includes the analysis of both widely spaced arrays exhibiting
far field patterns with grating lobes, and of closely spaced
arrays scanned to angles outside real space, which results in
most of the power being reflected back into the array elements.
The directivity of the array is then given by
Fig. 4. Derivation of the infinite array couplings from the infinite array active
impedance.
981
Coupling Y Parameters
up to (tV2,NR)away
I
-.-.
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
Ii
r-J
Matrix Inversiontechnique
iI
lamfn-im
b = Sa.
It is interesting to note that this S parameter approach is
in fact exactly the same assumption as is made in the direct
Fourier windowing technique described in Section 111.
C. Multimode Coupling
The coupling parameters that have been derived above
contain contributions from as many modes as are examined
during the initial infinite array analysis. From this aspect these
methods can be considered to be a multimode analysis.
However, another important input dataset to these methods
is the isolated element pattem. This will vary depending
on which modes are significant within each individual element-information that is impossible to derive from the active
impedance alone. The best estimate of the isolated element
pattem is the field produced from the array element with purely
the fundamental mode present. Hence these methods are much
D.Implementation Algorithms
A block diagram of an algorithm implementing the above
method using the infinite array admittance parameters is shown
in Fig. 5. The methods that use impedance or 5' parameters are
almost identical, but changes can be made to the part of the
algorithm that derives the reverse waves:
1) The S parameter approach requires no matrix inversion,
since the coupling parameters lend themselves immediately to the free excitation analysis used here.
2) As shown on Fig. 5, the S matrix calculated for the finite
array must be calculated only once for a particular array
geometry. This matrix can then be used to examine any
scan angle, amplitude taper, or phase taper. However,
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 42. NO. I , JULY 1994
988
13
12
11
10
9
:I
5
4
3
2
z axis
24
22
20
18
16
14
12
Fig. 6.
(:T ;::::;)
b(n,m) = u(n,m )
n,m = 0 . . . size of finite array
where
I(n1,ml) =
(nZ,mZ)=(O,O)
x (a(n2, m2) ~ O L D
m2)),
(G
br\rEw(nl,m l ) = (u(n1, m l ) - I(n1,m1))K
+ ( b ~ ~ ~ ( n l , m l-) )K( )l .
The new values b(n1, m l ) are obtained from the old values
for all positions ( n 1 , m l ) within the finite array. This stage
is repeated as many times as necessary to obtain convergence
(typically 10 to 20 iterations).
The gain factor K is necessary to prevent the system
from becoming unstable. Suitable values of K have been
determined experimentally and are approximately 0.1 and 0.2
for the admittance and impedance parameter transformations,
respectively.
VI. VALIDATION
WITH MEASURED
RESULTS
989
Fig. 7. Triangular basis infinite array active reflection coefficient over the
periodic rectangle at 8.7 Ghz.
Fig. 8. Rectangular basis infinite array active reflection coefficient over the
periodic rectangle at 8.7 Ghz.
i.e., the comer points of the rectangle have exactly the same
active impedance and active reflection coefficient as the center
point. The infinite array arrayfactor repeats in the same way,
but the infinite array immersed element pattern is restricted to
real space due to the real properties of the isolated element
pattern.
Predicted
Coupling
-13.3
-74'
-31.1
-85"
-47.6
101"
-41.1
65"
-35.0
144'
-42.0
91"
-33.5
-123"
-39.4
-108"
-47.1
-78"
-44.4
34"
Measured
Coupling
-13.4
-75"
-33.2
-115"
-47.2
109"
-50.0
129"
-35.9
124'
-43.1
102"
-37.6
-131"
-41.9
-114"
-41.9
-79"
-42.6
-10"
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 42, NO. I, JULY 1994
q 4
.-.->
1
+
U
.
I2
n
0
-90
---
-2
__
I -60
Isolated element
-30
0
30
Elevation a n g l e /degrees
Directivities a t 8.70 GHz
60
90
-30
0
30
A z i m u t h a n g l e /degrees
Directivities a t 8.70 GHz
60
90
q 4
.-1
>
.-+c
.E
c3
-90
- 60
991
4
x
.>
.c
.-E
-2
-60
-90
Fig. 11.
lsalaied dement
-30
0
30
Elevation a n g l e /degrees
Directivities a t 8.70 GHz
60
90
60
90
Measured
--
Y Parameter analysis
q 4
x
.-
:.0
-2
-90
- 60
-30
0
30
A z i m u t h a n g l e /degrees
Directivities a t 8.70 GHz
H plane cuts.
VII. CONCLUSION
The methods presented here produce excellent results, and
can be summarized as follows:
1) Direct Fourier windowing technique: This is the fastest
method to execute. It can be implemented for any type
1 -
IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 42, NO. 7, JULY 1994
992
correct. However, the patterns produced are good firstorder estimates. This method also allows a first-order
prediction of the individual element active reflection
coefficients to be made for any amplitude and phase
distribution applied to the finite array.
3) Impedance/admittance parameter free excitation analysis: These routes take the longest of all the routes
discussed, due to the additional transformation to free
excitation. However, when applied to simple singlemode elements such as dipoles (minimum scatterers with
respect to impedance parameters) or waveguide apertures (minimum scatterers with respect to admittance
parameters), the predictions of finite array patterns and
active reflection coefficients are very accurate indeed.
Work done during the course of this study supports the
work of Wasylkiwskyj and Kahn [12], which suggests that
an element can only be considered to behave as an element
within an infinite array, when it is embedded in the center of
at least a 25 x 25 array. However, even in an array of this size
the center element pattern will exhibit a fast ripple due to the
finite edges of the array.
The method of analysis via the coupling parameters accounts for all of the problems identified by Hannan [3]. This
includes effects such as the difference between the realised
element gain and the theoretical value
cos 6. The differences are accounted for by the mutual and element-to-element
couplings, and by the isolated element patterns. The software
produced during the course of this work contains pattern power
integration algorithms that have been shown to consistently
correlate well with the value of total applied forward power
minus total reflected power in the finite array [8]. This has been
shown for single excited elements, normally spaced arrays,
widely spaced arrays with grating lobes and scan blindness,
and for closely spaced arrays scanned to imaginary space with
no main beam or grating lobes propagating.
REFERENCES
[ l ] T. S. Bird, Mutual coupling in finite coplanar rectangular waveguide
arrays, Electron. Lett. vol. 23, pp. 1199-1201, Oct. 1987.
[2] P. Geren, An
approach to large array analysis, IEEE APS, vol.
4, no. 9.
[3] P. W. Hannan, The element gain paradox for a phased array antenna,
IEEE Trans. Antennas Propagat., vol. AP-12, pp. 423-433, 1964.