ANTENNA ARRAYS

THE MATHEMATICS OF LINEAR ARRAYS:

For a general linear array of equally spaced elements, the relative amplitude of the
radiated field strength is given by
jn 2 j jn 1
E a0 e j 0
a1e j j 1
a2e j2 j 2
an 2e n 2
e (1)

where
2
dcos ,

In this expression d is the spacing between the elements. The coefficient a0 ,a1,a2, etc.,
are proportional to the current amplitudes in the respective elements. is the
progressive phase shift (lead) from left to right; 0, 1, 2, are the derivations from this
progressive phase shift. Expression (1) may be written as
n 2 n 1
E A0 A1z A 2 z2 An 2z z (2)

where z e j , Am ame j m
.
The coefficients A0 ,A1,A2, are now complex and indicate the amplitude of current in
each element and the phase deviation of that current from the progressive phase shift
of the array. If any of the coefficients are zero, the corresponding element of the array
will be missing, and the actual separation between adjacent elements can be greater
than the “apparent separation” d. The apparent separation is the greatest common
measure of the actual separations.
The following fundamental theorems are due to Schelkunoff, and lay the foundations
for the method:

Theorem I: Every linear array with commensurable separations between the elements
can be represented by a polynomial, and every polynomial can be interpreted as a linear
array.

Since the product of two polynomials is a polynomial, a corollary to Theorem I is:

Theorem II: There exists a linear array with a space factor equal to the product of the
space factors of two linear arrays.

Theorem III: The space factor of a linear array of n apparent elements is the product of
(n-1) virtual couplets with their null points at the zeros of E.

The space factor of an array is defined as the radiation pattern of a similar array of
nondirective or isotropic elements. The degree of the polynomial which represents an
array is always one less than the apparent number of elements. The actual number of
elements is at most equal to the apparent number. The total length of array is the
product of the apparent separation and degree of polynomials.
Consider a simple two-element array in which the currents in the elements are equal
in magnitude. The relative radiation field strength is represented by
E 1 z (3)
j d cos
where z e
Making use of theorem II, a second array can be constructed which will have a
radiation pattern that is the square of that given by above, that is,
2
E 1 z 1 2z z2

It is seen that the array that will produce this pattern is a three element array having
the current ratios 1:2:1. The current in the centre element will lead the left hand
element by , and the current in the right hand element will lead the left hand
element by 2 .
If the polynomial of (3) is raised to mth power, they results the general binomial array.
When the element spacing d is not greater than 2 , such an array produces a pattern
with no secondary lobes. However, the principal lobe is considerably broader than that
produced by a uniform array having the same number of elements. An array having a
narrower principal lobe than that given by the binomial distribution and smaller
secondary lobes than that given by the uniform distribution can be obtained by raising
the polynomial of the uniform array of n elements (where n>2) to any desired power.
For an n-element uniform array
n 2 n 1
E 1 z z2 z z (4)

It has already been shown that when n, the number of elements, is large, the ratio of
the principal maximum to the first secondary maximum is approximately independent
of n and is 13.5 dB for the uniform array. If an array is formed to produce a pattern
that is the square of that given by (4), the ratio of the principal to first secondary
maximum will be 27 dB. This second array is given by
2
n 2 n 1
E 1 z z2 z z

n 1 n 3 2n 2
Or, E 1 2z 3z2 nz n 1 zn 2z z (5)

The current ratios of this array have triangular distribution

1,2,3, , n 1 ,n, n 1 , ,3,2,1 .Raising the uniform array to a still higher power, would,
of course, increase still further the ratio of principal to secondary lobes.
The significance of theorem III, the decomposition theorem, can be understood by
studying the variable z, where z e j and dcos . Since is real, j is a pure
imaginary, and the absolute value of z is always unity. Plotted in the complex plane, z
is always on the circumference of the unit circle. As increases from zero to 180o,
decreases from d to - d and z moves in a clockwise direction. Because of the
symmetry the range of to be considered is zero to 180o . Thus the range of
described by z is 2 d radians. For example, for a separation between elements of
4, varies through radians as goes from zero to 180o, and z describes a
semicircle. For d= 2 the range of is 2 radians and z describes a complete cycle as
varies from zero to 180o. If d> 2 , the range of is greater than 2 radians and z
will overlap itself.
For the simple two element uniform array, given by (3), the field strength is the sum
z 1 , which may be written as z 1 . This value is given geometrically by the
distance between z and the point -1. For the more general case of unequal amplitudes,
where the source intensities are proportional to 1 and –t, the radiated field strength
pattern is given by z t which geometrically is the distance between the points z and
t. Since z is always on the unit circle, the pattern will have a zero only when t is also
on the unit circle, and when t is within the range of z.

By the fundamental theorem of algebra, a polynomial of the (n-1)th degree have (n-1)
zeros (some of which may be multiple zeros) and can be factored into (n-1) binomials.
Thus
E z t1 z t2 z tn 1 (6)
from which theorem III follows directly.
It is evident that relative radiation field in any direction is given by the products of the
distance from z (corresponding to the chosen direction) to the null points of the array.

Example -1: Uniform array

Consider the uniform array that is represented by
n 2 n 1
E 1 z z2 z z

1 zn zn 1
Or, E (7)
1 z z 1
The null points of such an array, given by the roots of (7), are in this case the nth
roots of unity (excluding z = 1, which is the principal maximum). In the complex plane
the roots of unity all lie on the unit circle, and divide the circle into n equal parts. The
roots are
2 2 2
j j2 jm
n n n
z e ,e , e
 2
It is seen that the null pints of the array are given by m m where m = 1, 2, 3,
n
...(n-1). Since dcos , the null points of the radiation pattern are given in
terms of the angle by
2 m
cos m
d n d
When z = 1, E has a principal maximum. Other maxima occurs approximately
midway between the nulls. As z moves around the circle the radiation pattern is given
by the product of the lines connecting the null points to z. A plot of E as a function of
is shown below.

Using cos 1
, E can be drawn as a function of .
d

Example -2: Four element broadside

For this case d , =0 and = cos .

The range of z is 2 d 2 .
The relative field strength pattern is given by
 z4 1
E 1 z z2 z3
z 1
2 3
j j j
Or E z e 2 z e 2 z e 2 (8)

The nulls are spaced equally on the unit circle as shown below. As increases from
zero to 180o, decreases from through 0 to - and the curve of figure (b) results.

This is plotted in polar coordinates as a function of below.

Example – 3: Four element end fire

Consider a uniform four-element end fire array having an element spacing of one-
quarter wavelength and a progressive phase shift of 2
radians. For this array

d , and cos 1
2 2 2
The range of is radians.
As before the expression of E is given by (8) and the three nulls occur at
2 3
, , . However in this case the range of is only from 0 to , so the null
2 2 2
3
at obviously have little effect on the pattern.
2

An improved pattern can be obtained with the same number of elements by spacing
the nulls equally in the range of . This gives rise to the array that has the circle
diagram of figure below and the pattern given by
2 3
j j j
E z e 3 z e 3 z e 3

2 2
j j j j
Or, E z3 1 e 3 e 3 z2 1 e 3 e 3 z 1

2 3
j j j
Or, E 1 2e 3z 2e 3 z2 e 3 z3 (9)

The current amplitudes of the array are 1: 2: 2: 1 and the progressive phase shift
5
between the elements will be radians. The resultant pattern as a
2 3 6
function of is shown as curve “B” in the following figure.
If the overall length of the array is maintained constant, but the number of elements is
increased, it is possible to improve the directivity still further if the nulls are properly
spaced in the range of operation. Curve “C” of the following figure shows the pattern
that results when the number of elements is increased to seven with spacing reduced
3
to one-eighth wavelength so that the overall length is still . To obtain this result the
4
nulls were equi-spaced in the range 2 d 2 . Curve “D” shows the pattern
obtained for 13 elements at 16 spacing, with the nulls again equi-spaced in the range
of .

ANTENNA SYNTHESIS:
It is a simple and straightforward job to compute the radiation pattern of an array
having specified configuration and antenna currents. A somewhat more difficult
problem is the design of an array to produce a prescribed radiation pattern. Making
use of the Fourier analysis, the methods of the preceding sections may be extended to
accomplish this result.
It is convenient to consider an array having an odd number of elements with certain
symmetry of current distribution about the centre element. The polynomial of an array
with n = 2m+1 elements is
m 1
E A0 A1z A 2 z2 A m zm A m 1z A 2m z 2m (10)

Now the absolute value of z is always unity, so equation (10) can be divided by z m
without changing the value of E . That is

E A0 z m
A1z m 1
A2z m 2
Am 1z 1
Am Am 1z A2mzm (11)

It is now specified that the currents in corresponding elements on either side of the
centre element be equal in magnitude, but that the phase of the left-side element shall
lag that of the centre element by the same amount that the corresponding right-side
element leads the centre element (or vice versa). That is the coefficients of the
corresponding elements are made complex conjugate with
Am a0 Am k ak jbk Am k ak jbk

Then the sum of terms of two corresponding elements may be written

k
Am k Z Am k Zk ak Zk Z k
jbk Zk Z k

= 2ak cos k 2bk sin k

Since Zk e jk
The expression for E is now

1
E 2 a0 a1 cos a2 cos 2 am cos m b1 sin b 2 sin 2 bm sin m
2

m
a0
Or, E 2 ak cos k bk sin k (12)
2 k 1

There are the first 2m+1 terms of a Fourier series in which the coefficients of the
cosine terms are the ak ’s, and the coefficient of the sine terms are - bk ’s. Now any

radiation pattern specified as a function f may be expanded as a Fourier series

with an infinite number of terms. Such a pattern may be approximated to any desired
accuracy by means of the finite series (12). When this is done the required current
distribution of the array can be written down directly. From the theory of Fourier
series, this approximation is in the least-mean-square sense; i.e., the mean square
difference between the desired and the approximate pattern for from 0 to 2 is
minimized.

Example – 4: Synthesized bidirectional array

Design an array that will produce approximately a pattern shown below.
This pattern is defined by
f 1 0< < 3
f 0 3< < 2 3
f 1 2 3< <

It will, of course, by symmetrical about the line of the array, 0 . If the spacing is

chosen to be 2 , then cos . The corresponding function is

F 1 +
2
F 0 +
2 2
F 1 - +
2
Choosing for an end fire array results in the function shown below. The Fourier
series expansion for this function is
k
1 2 1 k
F sin cos k
2 k 1
k 2

Comparison with (12) determines the coefficients

a0 12
1 k
ak sin
k 2
bk 0 k 0

The pattern obtained using the value of m = 4, is given from equation (11) as
1 1 1 3
E z 3
z 1
z z (13)
3 2 3

This is a five element array having the current ratios indicated and an overall length of
three wavelengths (the apparent spacing between elements is one half wavelength, but
four of the elements are missing). The pattern produces by this array is shown in the
above figure and marked “C”. “A” is the desired pattern.
In the above example the apparent element spacing was arbitrarily chosen to one-half
wavelength, which made the range of equal to 2 radians. If the element spacing is
less than 2 , the range of will be less than 2 radians. This means that although

the radiation pattern as a function of , that is f , is completely specified for the

whole range of , the corresponding F is specified only over its range, which is less

than the interval of 2 radians required for the Fourier expansion. It is possible then
to complete the interval with any function that satisfies Dirichlet’s conditions.
Naturally, the function chosen would be one which would simplify the series as much
as possible or make it converge rapidly. It is evident that when the apparent spacing is
less than 2 there is an unlimited number of solutions that will satisfy the conditions

of the problem. If the apparent spacing of the elements is greater than 2 , the range

of is more than 2 radians. Except for some special cases it is then not possible to
obtain the prescribed directional pattern by this method.
When an apparent spacing less than 2 is used, F is specified over only one

portion of the required 2 radians, and the function used to fill in the remainder of the
interval can be chosen at will by the designer. A judicious choice of a “fill-in” function
will produce a desirable pattern with a minimum number of elements, and conversely
a poor choice of function may result in a poor pattern or high Q0-factor for the array.
This is illustrated below -

Example -2: Synthesized unidirectional array

Let it be required to design an end-fire array that will have an approximately
semicircular pattern given by
f 1 0< < 2
f 0 2< <

The apparent spacing is to be 4.

Then for this problem cos and the range of is radians. By choosing
2
different values of , the range of which is used can be shifted anywhere in the
interval 2 radians, which is required for the Fourier expansion. This is shown in the
above figure for three values of . In this example there is a finite discontinuity
within the range of , so the coefficients of the series will decrease at a rate that is of
the order of 1 n . If the functions were continuous in the range, the series would

converge at a rate that would be at least of the order of 1 n2 . Because no choice of a

fill-in function can remove this discontinuity within the range of , it is anticipated
that, in this case, the fill-in function may not have much effect on the number of
terms required.
Case – 1 A possible choice for and for the fill-in function is illustrated in figure
below. This choice would apper to be good, because it results in the following
conditions:
1. f f . Therefore only odd harmonics will be present.

2. f is an even function, so coefficients of the sine terms will be zero.

3. f has an average value of zero, so dc term (or centre element) is eliminated.

The antenna array resulting from the choice and the corresponding pattern is shown
below for m = 7 (eight) element. This pattern has two serious defects. It approaches a
relative value of 0.5 at 0 where it should be unity, and it also approaches 0.5 at

where it should be zero. Using more terms of the series will not remedy these
defects, which are inherent in the particular chosen function. This function is
discontinuous at the value of corresponding to 0 and , and the series
converges to the average of the value taken by the function on the two sides of the
discontinuity. Therefore the function chosen above is an unsuitable choice.
Case 2  The discontinuities at values of corresponding to 0 and can be
eliminated by a different choice of , and a suitable fill-in function. A possible
function is that shown in figure (a) below. The corresponding antenna array and
resulting pattern are also shown, and it is seen that this function is a suitable one.
THE TCHEBYSCHEFF DISTRIBUTION:
A particular, but very important, problem in antenna synthesis is the following: for a
given linear antenna array, determine the current ratios that will result in the
narrowest main lobe, for a specified side-lobe level; or, in other words, determine the
current ratios that will result in the smallest side-lobe level for a given beam-width of
the principal lobe. The current distribution that produces such a pattern will be
considered as being the optimum in the above sense. The directive gain of this
“optimum” distribution is less than that of a uniform array. It is evident that the
optimum pattern is obtained when all the side lobes have the same level. The problem
is simply that of finding the spacing of nulls which makes this true. The answer is
given in terms of Tchebyscheff polynomials.
The Tchebyscheff polynomials are defined by
1
Tm x cos mcos x -1 < x < +1
1
Tm x cosh mcosh x x 1

The general shape of Tm x is shown below.

By inspection, T0 x 1 , T1 x x

The higher order polynomials can be derived as follows

1
T2 x cos 2cos x =cos 2

where cos 1
x Or x = cos

Now since
cos 2 2cos2 1
T2 x 2x2 1
Similarly, it can be shown that
Tm 1 x 2Tm x T1 x Tm 1 x

So that
T3 x 4x3 3x
T4 x 8x 4 8x2 1

and so on.
Consider the Tchebyscheff polynomial of mth degree
1
Tm x cos mcos x cos m

The nulls of the pattern is given by the roots

cos m 0

that is, by

0 2k 1
k k = 1, 2, m
2m
Next consider the function . For a broadside array for which 0,

dcos

As varies from 0 to 2 to , goes from d to 0 to - d and the range of id s

2 d.

Now let x x0 cos 2 . Then, as varies from 0 to 2 to , goes from d to 0 to -

d , and x will vary from x x0 cos d to x0 to x0 cos d x 0 cos d .

The nulls in the Tchebyscheff pattern occur at values of x given by

xk0 cos 0
k

so that the corresponding position for the nulls on the unit circle will be given by
xk0 x0 cos 0
k 2

Or, 0
k 2cos 1
xk0 x0 2cos 1
cos 0
k x0 (14)

Equation (14) gives the required spacing of the nulls on the unit circle for a pattern
whose side lobes are all equal. The degree m of the polynomial used will be equal to
the number of nulls on the unit circle, and this will be one less than n, the apparent
number of elements. The value of x0 is determined by the desired ratio b of principal
to side-lobe amplitudes. The value of x0 is given in terms of b by

Tm x0 b
It can be calculated by noting that if b cosh , then

x0 cosh m

The graphical-analytical method for obtaining the pattern of the array from the
location of the nulls yields a detailed and accurate plot of relative field strength versus
the defined angle . For many purposes a rough sketch of the pattern may be
adequate, and this can be obtained directly from the known properties of the
Tchebyscheff polynomial. Thus knowing the location of the nulls in the pattern and
the amplitude of all the side lobes relative to the principal lobe, the pattern as a
function of may be sketched with good accuracy. The pattern as a function of

azimuth angle is then determined using the transform cos 1

d . The binomial

expansion method of calculating the required current distribution from the location of
the nulls provides satisfactory for small arrays, but tends to become unwieldy foe
larger arrays.

Example
Design a four element broad side array having spacing 2 between elements. The

pattern is to be optimum with a side lobe level which is 19.1 dB down (b = 9.0).

For d 2 , the range of operation is 2 d 2 . Since there will be 3 nulls,

use T3 x 4x3 3x . Then

3
T3 x0 4x0 3x0 9

Solving for x0 ,
1
cosh b 2.887

x0 cosh 2.887 3 1.5

The nulls given by cos m cos 3 0 . Therefore

0 2k 1 2k 1
k k = 1, 2, 3
2m 6

Then 0
1 6, 0
2 3 6, 0
3 5 6

k 0
k xk0 cos 0
k
xk0 x0 0
k 2cos 1
xk0 x0
0
k (Radians)

1 6 0.866 0.577 109.5o 1.910

 2 3 6 0.000 0.000 180.0o

3 5 6 -0.866 -0.577 250.5o 4.37

The polynomial representing the array is

E z e j1.91 z ej z e j4.37

= z3 1.667z2 1.667z 1

The required relative currents in the elements are 1:1.667:1.667:1.

In concluding this section it is pertinent to note that the Tchebyscheff “optimum”
design is strictly applicable only for a line array of isotropic sources. Actual antennas
are not isotropic and the effect of element directivity should not be forgotten.
Moreover, instead of the typical fan-beam pattern of an equi-phase line array, the
more usual optimization problem calls for a maximum in only one direction in space,
with the peak field strength in all other directions minimized.