16

I/R*E TRASSACTIONS-ANTENNAS AND PROPAGATION

Design of Line-Source Antennas for Narrow

Beamwidth and Low Side Lobes*
T. T. TAYLORT
Summary-It is well known that the phenomenon of radiation
from he-source antennas isvery similar to thatof the diffraction of
light from narrow apertures. Unlike the optical situation, however,
antenna design techniquepermits the use
of other-than-uniform distributions of field across the antennaaperture. Line sourcesynthesis
is the science of choosing this distribution functionto give a radiation
pattern with prescribedproperties such as, for example, narrow
angular width of the main lobe and low side lobes. In the present
article the mathematicalrelationships involved in the radiation
calculation are studied from the point of view of function theory.
Some conclusions are drawn which outline the major aspects of
synthesis technique very clearly. In particular, the problem of constructing a h e source withan optimum compromise between beamwidth and side-lobe level(analogous tothe Dolph-Tchebycheff
problem in linear array theory) is considered. The ideal pattern is
cos r d u i i , where u=(Za/A) cos e, a is the half-length of the
source, and cosh T A is the side-lobe ratio. Because of theoretical limitations, this pattern cannot be obtained from a physically realizable
antenna; nevertheless its ideal characteristics can be approached
arbitrarily closely. The procedure for doing this is given in detail.

design a space factor with the narrowest beamwidth

and lowest side lobes consistent with both the available
aperture and thenecessity for avoiding a super-gain line
source.
Let the line source, which has a total length 2a, have
a distribution function g ( x ) . Then the space factor
is
given bythefamiliarFourier-transformrelationship
~ ( v=
)

k Jg(x)e*zYdx,

(1)

where v =cos 0 as illustrated in Fig. 1 and k is the free

space wave number, 2n/X. I t willbe advantageous to
introduce new variables as follows:

INTRODUCTION

HE term line source may be applied to any an-

The quantity X/2a will be known as a standard beamtenna which embodiesa long, narrow, and straight width2 and will prove useful later. With these substitugeometry and which depends upon variations in tions, (1) becomes
field or current strength with respect to the longitudinal
co-ordinate as the basis for its directivity. I t is implied
that these currents orfields are continuous functionsof
this longitudinal co-ordinate, hence arrays of discrete The factor 2a/X will be dropped and the integral deelements will not be considered as such.I t is well known, noted F(u):
however, that arrays of a large number of elements are
most conveniently analyzed by regarding them as conF(u) =
g(p)eiPudp.
tinuous distributions, therefore the remarks made here
(5)
will apply-in this sense-to such arrays.
The patternof a line source contains an element factor Eq. (5) expresses the basic relationship which will be
studied here.
and a space factor. The element factor depends upon
thetypeandorientation
of fields orcurrents which
make up a typical segment of the source, and usually
has little directivity, a t least in the direction in which
i t would augment the directivity of the space factor.
m OB~RVER
The spacefactor is the highly directivefactor;itis
dependent for its shape upon the relative variation of
field or current strength along thesource and is the object of the antenna designers art. In shaping the space
factor, the designer may or may not wish to take cognizance of the element factor. In this article it will be
assumed t h a t he does not and that the objective is to

S-,

tz

Fig. 1

* Original manuscript received by the PG-qP, April 26, 1954;

revised manuscript received October 11, 1954. The work reported
in this article was performed at the Hughes Aircraft Company and
has been described in Hughes Aircraft Company Tech. Memo. Yo.
316. The work w a s sponsored by the .4ir Force Cambridge Research
Center, Cambridge, Mass., under Contract AF 19(604)-262.
t California Institute of Technology, Pasadena, Calif.; formerly
Research Physicist, Hughes Aircraft Co., Culver City, Calif.
1 C . L. Dolph, A current distribution for broadside arrays which
optimizes the relationship between beamwidth and side-lobe level,
PROC.I.R.E., vol. 3-1, pp. 335-348; June, 1946.

PROPERTIES
OF

THE

DISTRIBUTION
FUNCTION

I t now becomes necessary t o specify the hypotheses

which the function g ( p ) must satisfy in order that the
results of this analysis will be applicable. 4 n exhaustive
The beamwidth between half-power points of a uniformly illuminated line source is 0.8859 of a standard beamwidth.

generality is not attempted here; nevertheless, it is felt

that a high degree of relevancy to thephysical situation
is achieved.
I t is necessary, in antenna work, that g ( p ) be singlevalued, uniformly bounded, anda t least piece-nrise
. continuous in -a < p < P . However, anyone conversant with
such workwill grant that there
exists a largeclass of practically important cases in which not only g ( # ) , but all its
derivatives as well, arecontinuousoverthe
full interval -T < p <P. This article is specifically concerned
with functions of the latter type and all general statements made henceforth are t o be understood as applying only t o distributions which obey this hypothesis.
The restriction, however,is not as stringentas might be
supposed. All analytically expressible functions which
arenon-singular in the aforementionedinterval,such
as theuniform function cosp / 2 , J & ) (and an infinitude
of others which have no specific names), belong to this
class. In fact, any function which is continuous in the
intervalunder discussion can be approximatedarbitrarily exactly by a member of this class. This article
will also, incidentally, beconcerned only with even distributionfunctions, since theseare sufficient for the
purpose at hand, namely the optimization of the compromise between beamwidth and side-lobe level.
The degree of abruptness with which the source distribution begins and ends a t p = fa will be a matter of
considerable importance. Let the symbol
N be read
is asymptotic to. Then suppose that

g(p)
g(p>

as

K(T -

as

K(p

p
p

+
+

-r

).

WP)(T2

P 2 P 9

Fig. 2

tends to zero of the integral along C of Fig. 3, beginning a t - P + T and following the axis of reals to T - r
in the 6plane. The limiting process is necessary because
of the formalconsideration of negativevalues of a.

(6)

The quantity . K is a non-zero constant in these equations. The real parameter a , if an integer, is equal to the
multiplicity of the zeros in g ( p ) a t p = +a. Thus if
a = 0, g ( p ) assumes the fixed non-zero value K a t these
two points and the distribution
is said to have a pedestal. If a = 1, the distribution falls t o zero linearly at its
ends as in cosine distributions. If a = 2 , i t falls to zero
quadratically as in cosine-squared distributions. All
values of a greater than - 1, both integral and nonintegral, will be considered. The not-uniformly-bounded
distributions corresponding t o a in the range - 1 < a < O
are included in order to achieve a degree
of completeness
necessary for later arguments, evea though such values
are not regarded as being physically realizable. If g ( p )
is known analytically, a may be found by taking the
limit as p approaches --R of @ + P ) g ( p ) / g ( p ) , or the
limit as p approaches P of @-a) g ( p ) / g ( p ) . Evidently,
it is possible to writeg ( p ) as the product
g(P> =

t = p + i q and z=u+io. Thefunctions g ( p ) and F(u)

are then profiles on the axes of reals of the functions
g ( t ) and F ( z ) . The visible range of the z-plane3 is the
real segment - 2a/X lu 5 2 a / X , and a unit distance in
the z-plane corresponds to onestandardbeamwidth
(see Fig. 2). The integral in (5) becomes the limit at

(7)

iq

[ PLANE

Fig. 3

Eq. (5) becomes

~ ( z=
) lim J c g ( ~ ) e ~ ~ z d ~ .

(8)

S+O

This g ( t ) is, of course, equal to the product

h(4) ( ~ 2 f)*. If we cutthe 4 planefrom to -T andfrom
+a to
x , then ( ~ - 6 ~
is)regular
~
in this cut plane
and should be regarded as being on its principal branch.
Since i t is desired that theasymptoticbehaviourof

where h ( p ) is an even function which doesnot vanish at

p = kn and which fulfills the other requirements applying to g ( P ) .

Let u s now reconsider (5) from the point of view of
function theory. Let the variables p and u be imbedded
in the complex domains t and z respectively, such that

3 A n observer who travels around the line source to examine its

.
radiation pattern will cover all real angles in the range 0 5 0 5 ~ This
coverage is translated into the point set -2a/X5u52a/X,
o=O, in
the z-plane by the relationship zd = (2e cos 0)/X as shown graphically
in Fig. 2. Hence the term visible range is applied to this .point set.
The profile of F(z) on the visible range is the only part of F(z) which
ormally enters into theformation of the radiation patterp-

January

p approaches +a be as indicated in (6), i.e.,

be embodied entirely in the factor ( x 2 - E 2 ) ~i t, is essential that h(l) beregular at least in neighborhoods of
--K and +T and be different from zero a t these points.
Moreover, if g ( p ) and all its derivatives are to be continuous in -a <$<n, h(5) must be regular in a domain
which contains this interval in its interior. Thus
i t is
sufficient if h(() is differentfrom zero a t f = +a and
regular in therace-trackshaped
region R of Fig. 3,
where 6 is the arbitrarily small but non-vanishing radius
of the ends of the region.
The hypotheses of thisanalysismay
thereforebe
summarized as follows:

g($) qs

1. g(p) is the profile on the real interval -a<$<a

of the principal branch of the function h(t)(n2-

function is alwaysuniformlybounded
on a closed
bounded set within its domain of regularity. Thus, for
p in the closed interval a - 7 6 p ST and z in G, I2h(p)
( a + p ) a l is bounded by some constant A , and since
cos p z is entire with respect to z for every value of p in
thestatedinterval,
cos pzI is bounded byanother
constant B. Therefore

Since a > - 1 ,

lim f(z,

p ) a.

2.
3.
4.
5.

7)

I =o

(13)

T-0

h ( ( ) is regular in the region R of Fig. 3.

h ( ( ) does not vanish a t E = &a.
h({) is even.
a is real and must be greater than or equal tozero
forphysicallyrealizable
distributions; however,
values of a between - 1 and zero will be considered
for completeness of analysis.

for every z in G, hence condition 3 is satisfied and F ( z )

is entire.Forfurtherbackground
in connection with
this proof the readershould consult Copson.4

PATTERN
THEOREMS
In this section, some basic theorems concerning
properties of the function F ( z ) will be proved.

the

iq

-2

Theorem I: F(z) is a n even function of z.

This is a very simple theorem. Because of the evenness of h ( ( ) and of the factor ( ~ ~ - pthe
)~
transforma,
tion may be written a s f o l l o ~ s :

s,

r-r

F ( z ) = lim
d

2h(p)(.? - p)a cos pzap,

(9)
Fig. 4

since cos$( -2) =cos p z , F( - 2 ) = F ( z ) , and F ( z ) is even.

Theorem 11: F ( z ) is a n entire functionof z.

Eq. (9) will be adopted as the point of departure in
the proof of this theorem. I t will be necessary and sufficient to show that:
1. The integrand is itself entire for every p in the
interval 0 5 $ 5 a -r.
2. The integrand is continuous with respect to p for

every p in the interval 0 5 p ST - r and for every z

in the entire finiteplane.
3. The limit of (9) is approached uniformly for z in
G, where G is any closed bounded subset of the
entire finite plane.
Conditions 1 and 2 are evidently satisfied. T o demonstrate the fulfillment of condition 3 , it must be shown
that the difference, f(z, r ) , between the integral in (9)
and its limit is uniformly bounded for z in G and that
this bound tends to zero as T tends to zero. Evidently,

PLANE

Theorem 111: as z tends to infinity,F ( z ) has the follozving

asymptotic forms

F(z)

cos

h(?;)(2x)far(l

.(.

+ a)

,>

- l+Cr

T(Z)l+Q

for Re z

F(Z)

cosr(-

-)

(14)

-l+ff

/2(R)(2y)1+ar(1 a )

~ ( z ) lz(T)(2R)1-r(1

>0

.(-

z)l+a

for Re z

<0

(15)

eklrl

+ cy)2RI .I

for Re z

0. (16)

l+-

Proof of (14) depends upon deforming the contour C

into the contour C1 C2 Ca and subsequently allowing the
radius of the small arcs to shrink to zero (see Fig. 4).

Taylor: Design of Line-Source Antennas for h'arrozc Beamwidth and Low Side Lobes

1955

I t is easily shown that evenfor -1 <a<O the contributionsalongthe

small arcsvanish.We
begin by
saying :
z = rei@
n

--

x-

< cp < - ;

C$

is constant

r+*

1I

(17)

19

But the magnitude of one or the other of the two integrals 11 or 1 3 (depending upon the sign of 4 ) tends to
KY-('+~)
err sin IQI for large Y . Therefore 1121 will, for
large r , always be negligible withrespect to 111 or
131 or, if 4 =0, to both 1111 and 1181. The total integral is thus the sum of (23) and (24). Since h( -T) =
h ( ~ this
)
sumis simply (14). Exp.(15)may be derived by
deforming the contour as in Fig. 5, or by appealing t o

The angle 4 determines the direction taken by the two

parallel legs of the contour. The length of these legs is
b, 7 < b < 6. Let 11 be the integral alongC1. Then

PLANE

Fig. 5

The only significant contribution to this integral takes

place at some finite t , specifically a t t = 0 if 01 6 0 or at
1 =cy if a>O. Hence two of the factors of the integrand
can beremoved as r tends to infinity; these
become
simply h( -T) (2~)". Thus:
e-ix(z-(l+u)/z)j&( -r ) ( 2 ? f ) a
br
I1 =
zl+a
tae-'df.
(22)

2~~
1,

Since r goes to zero independently of r , we may substitute zero for the lower limit. Then as r tends to infinity,
the upper limit tends to infinity and the integral tends
to I?(1+CY).
Hence,
e-iH(Z-(l+e)/2)
= h(-x-)(2n)1+ar(1
CY)
. (23)

Theorem I. Exp. (16)is simply the limiting form of both

(14) and (lj),and can also be derived by the methods
used in obtaining (14) and working on the undeformed
contour C. It is interesting t o notice that when a is a
non-negative integer there are no singularities
at ( = AT.
TABLE I
asymptotic form of P(z)

01

sin 7rz
h(7r)(27r) n-z

h(7r) (2742

cos 7rz

2r(2)1+a

Similarly,

13

eir(z-(l+a)/Z)

h(r)(2~)l+~r(l CY)~-

On C2 we have

2x-(z)I+a

(24)

The contour of Fig. 4 can be swung around until it

conforms to that of Fig. 5 without making a separate analysis. The net result, that one asymptotic
expression
suffices for all Z, zI sufficiently large, is shown in Table
I above. As a further consequence of Theorem 111, it is
seen t h a t the asymptotic behaviour of F(z) depends only
upon the manner in which g(#) acts ut the ends of the line
source. If there is a pedestal, the asymptotic form must
be sin m / m . If g(#) falls t o zero linearly at the ends of
the line source the asymptotic form must be cos T Z / T Z ~ ;
if quadratically, it must be sin T Z / T Z ~ , etc. TlhI'l-decay
rate of_,the..~-~o_s_e~.:_side.~obe-~inF(z>,i?.th,y?..i:revocabIy
.._-z--.L -.. .
fixed bysLnditions at theends.of the.,lin.e_spurce. .

20

I4R.E T~A~SACTIOnS-ANTEl~l\rAS
AND PROPAGATION

Theorem I V : Let the zeros of F ( z ) be grouped in pairs:

fzl,-t 22, etc., and numbered in such a way that themoduli
of the members of successive paivs d o not decrease. Then
members
the
of nth
the
zero pair
tend
to positions
the
&
( n+a/,?)
injinity.
a ston fends

enough. The approximation, then, is

32
k=1

(T

+ ya + k ) 2

This canonical product is expressible in I functions, for

A i

[r(T

++:2

1)12

1.~1

Hencefrom ( 2 8 ) theasymptoticform of F ( z ) , as
becomes much greater than the radiusof the circle H , is

\-I

r T+-+I+z
(

?LANE

r T+-+I-~

>(

where K is a constant including both the coefficient of

in the polynomial and [r(T+a/2+1)]*.Now

z2~f

Fig. 6

In Fig. 6 , circles of very small fixed radius E have been

drawnaroundthepoints
f [ T + ( a / 2 ) + k ] where T
and K are integers and K = 1. 2 , 3 , .
. The large circle
H has a radius of T + ( a / 2 ) + ( 1 / 2 ) . K o w if T is made
large enough but finite, this system of circles will cover
all the zeros of F ( z ) , for if the circle H is sufficiently
large, F ( z ) will approachitsasymptotic
form arbitrarily exactly in the region outside of H and the zeros
of theasymptoticform of F ( z ) do lie at thepoints
f [T+ ( 4 2 ) + k ] . The question which must be settled
is that of the number of zeros within H. These zeros
constitute a bounded set and their number cannot be
infinite, for if it were they would have a limit point and
F ( z ) would have an essential singularity within H , a
condition which contradictstheentirety of F ( z ) . Let
the number of these zeros be 2 M , M an integer. Then
F ( z ) can be approximatedarbitrarilyexactly
b5- the
product of a polynomial of degree 211Z and a canonical
product on zeros outside of H located at the points
rt: [ T + ( a / 2 ) + k ] where k runs from one to infinitye5
The only defect in this approximation is embodied
in the small distances between the actual zeros of F ( z )
and the pointsf [ T + ( a / 2 ) + k ] but these distances can
bemade smaller than any fixed E bytaking T large

--

The technique of synthesizing the space factor of a line source

by setting up the product of a polynomial and an entire transcendental function where the latter has the same zero arrangement as
that just discussed here has been mentioned earlier by the author in
less general termsin an unpublished paper, An antenna pattern
synthesis method with discussion of energy storage considerations,
presented at theU.R.S.I. Washington Meeting; April 1931. Here, the
quantity T+(01/2) is equal to the ?V/2 in that paper: only cases corresponding to non-negative integer values of 01 mere treated.

Let A

=[z-T-(a/2)].

Then

By appealing to the Stirling approximation

for r(p),
valid where I pI is large and where a wedge-shaped
region covering the negative axis of reals is excluded,
it can be shown that

in a region which is even more extensive than that defined by R e z>O. Then certainly for R e z > O

But by Theorem 111. F ( z ) is asymptotic to a constant

x cos ~ [ z (-1 +a)/2]/.rrz1fcr in the region defined by

04 Line-Source
Antennas

Taylor:
Design

1966

for Sarrow Beamuridih and L02c Side

Lobes

R e z> 0. Hence we must conclude that

M= T an.d that
there are 2T zeros with.in the circle H. Therefore the zero
within the E neighborhood of T+(a/2) +1 must have
been a member of the T+lst pair of zeros, that within
an E neighborhood of T+(01/2) + 2 must have been a
member of the T+2nd pair, etc. Therefore z, tends t o
n+(a,/2) as ?a tends t o infinity, and the theoremis true.

BASICDESIGNC O N S I D E R 4 T I O N S
\$e are now in a position to state some definite conclusions regarding the design of line sources for highly
directive patterns. IlTe have seen that the pattern of a
finite line source, theeven distribution functionof which
decays with the exponent a at the ends of the source,
must have thefollowing essential characteristics:
1. F ( z ) must be aneven entire function of z , z =u+iv.
2. Theremoteside
lobes of F ( z ) mustdecayas

1uI

--(l+a).

3. The remote zeros of F ( z ) must lie on the axis of

reals and the members of the nth zero pair must
approach the positions k [n+ (a/2)] as n tends to
infinity. The remote zeros are separated by unit
intervals on the real axis.

Since, from the function-theoretic point

of view, two entire functions with the same asymptotic behavior are
but trivially different, we may expect any entire function possessing the characteristics noted to be the space
factor of some hypothetical line source. Space factors
will differ from one another, then, onlyin two aspects:
1. The value of a.
2. The placement of the central zeros; i.e., the zeros
near the origin in the z-plane.
As with the space factors of arrays of discrete elements, zeros are precious items. In highly directive patterns they are used to hold down the average level of
F ( z ) in the side-lobe region and their absence over even
a relatively short range permits a large lobe t o appear.
As an aid in visualizing this, the harmonic function log
F ( z ) = U ( u ,v) may be regardedas describing an elastic
surfacewhich is supported at infinityand weighted
down by small heavy weights a t each of the zeros.
Since we are primarily concerned with the profile of
F ( z ) along the axisof reals, it is important to realize that
the average value of this profile in the side-lobe region
can best be minimized if all the zeros are on the axis of
reals. A zero just off this axis produces a dip but not a
null in the profile. Furthermore, it is a matter of experience6 that if, in a given portion of the side-lobe region,

4 s an example, consider the entire functionfl(z) =cos srz, which

hasequally spaced zeros and uniform lobes. Move the two zeros
which straddle the origin together to form a double zero at the
origin. The function becomes
22

f2(z) =

a fixed number of zeros are available, the average sidelobe level will belower if these zeros are spread out
approximately uniformly rather than grouped
in twos
or threes or some other multiplicity. Therefore the first
design conclusion is that for highly directive patterns,
the central zeros of F ( z ) should be simple zeros on the
axis of reals. The remote zeros must follow this law
anyway, hence, f o p highly directizle patterns all the zeros
of F ( z ) should be simple zeros o n the axis of reals.
Let us next consider the selection of a value of a.
First of all, to insure that
g ( p ) will be uniformly bounded
a must be greater than or equal to zero. But as a is increased the zeros of F ( z ) abdicate from thecentral
region of the z-plane, since the nth zero pair tends to the
position k [n+(a/2)]. Increasing a by one unit is
tantamount to throwing away one zero. I t is true that
high values of a mean rapid decay rates of the remote
side lobes, but since any side-lobe structure within the
central region can be obtained anyway, and since the
central region can be extendedto cover thevisible range,
- (2a/X) 5 u5 (2a/X), a rapid decay rate of the remote
side lobes will not be a normaldesideratum in linesource design. It follows, then,thut 01 should have the
smallest possible value, zero, and that the best distribution
function f o r generating a highly directize pattern will have
a pedestal.
If a is to be zero, the remote
zeros must be asymptotic to the integers; the whole design technique becomes
a question of positioning the central zeros alongthe
axis of reals in such a way that the best beamwidth is
obtained compatible with the level of the central side
lobes and in such a way that the members of the nth
zero pair smoothly approach the positions& n.
AN

ENTIRE
FUKCTION
WITH UNIFORM
SIDELOBES

Beforeproceedingwith
thetechnique of practical
line-source design, it will be essential to formulate the
ideal space factor, i.e., an entire function with uniform
side lobes. Later itwill become apparent thata practical
line source can have a space factorwhose characteristics
approach the characteristics of this ideal space factor
arbitrarily closely and that theideal cannot be imitated
exactly without resorting t o super-gain.
G. J. van der Maas has shown that as the number of
elements of a Dolph-Tchebycheff array is indefinitely
increased, the space factor approachesa function which
(expressed in the present notation)is
Fo(u, A ) = cos

TdU2

- A?.

(35)

Here A is an adjustable real parameter having the property that cosh T A is the side-lobe ratio. This function
( 3 5 ) is evidently the profile on the axis of reals of the
entire function
Fo(2,

cos 7rz

___
z21 2

21

A)

cos T . \ / Z % - A2,

(36)

(3)

The remote behavior is unchanged but the two central lobes of f2(z)
have risen to 1.38 compared with 1.00 for f~(z).

I G. J. vander
Maas, -4 simplified calculation for DolphTchebycheff arrays, Jour. Appl. Phys., vol. 25, pp. 121-124; January, 1954.

TRANSACTIONS-ANTENNAS
I*R*E

88

AATD PROPAGATIOh'

Januarg

which can be shown to have the zeros

Let the second product be denoted Q*+-l)(z) t o agree
This entire function is clearly the ideal spacefactor.
with the notation of the unpublished paper referred to
Its side-lobe region ( u 2 > A 2 )is populated by an infinite earlier. This product has
someconvenientequivalent
number of equal lobes of unit amplitude; its main lobe expressions:
appears in the region u2<A 2 and has adjustable ampliC44) ;
tude, as was pointed out. Nevertheless this space factor
is unrealizable beecause the remote side lobes do not
decay and the membersof the nth zero pair are asymptotic to k [n- (1/2)], both of which conditions correspond to an 01 value of -1, a value which contradicts
[ ( i Z - 1)!p
the uniform boundedness .of g ( p ) . However, any good
(46)
(fi - 1 z)!(fi - 1- z ) !
practical space factor must have characteristics resembling those of Fo(z, A ) in the central region of the zplane, but in the remote regions its side lobes must decay as l/z and its zeros must tend to the integers, not
the integers minus one-half. The problem is becoming
clearer: in the central region the zero positions must be
The entire function with thezeros of (42) is
related to those of Fo(z, A ) but in the remote regions,
the nth zero must tend to 5 % A
. solution to this problem is proposed below.

SYNTHESIS OF A PRACTICAL

Here C is an arbitrary constant; if it is taken equal to

cosh T A then Fo(o,A ) = F ( o , A , %).Let us suppose that
this hasbeen done, then

LIKE S O U R C E

APPROACHING
ARBITFL~RILY
CLOSELY

IDEAL

TO THE

Let the practical space factor resemble the following

function in the central region :

lim F ( z , A ,
7i-

.ri)

Fo(z, A ) ;

(49)

for as fi increases, more and more of the zeros of F ( z , A ,

f i ) follow the pattern of (39) and simultaneously u-1,
hence (49) is as stated. Thus F ( z , A , f i ) defines a twoparameter family of entire functions whose members apthe zeros of which are
proach the ideal space factor arbitrarily closely a s fi i s inzn = f a d A , ( n - 3)'.
(39) creased. The value of ri, which is finite in any practical
case, is of considerable significance. In relationship t o
The constant u is a numbersomewhatgreaterthan
the pattern, the points k fi divide the region of uniform
unity. One can visualize (38) as an ideal space factor
side lobes
< -rt) from the region of decaying side
slightly dilated horizontally. Kow let us choose u such lobes
> e) where here, as before, z =u+iv. -4s fi is
that eventually one of the zeros of Fl(z, A ) becomes increased, the region of uniform side lobes extends farequal to the corresponding integer, designated ri. Then ther out from the main beam. In the region of uniform
side lobes the lobes actually decrease slightly as
2% = 5 f i = k u d A 3
(-rt - p ,
(40:)
increases and their initial
level is veryslightlylower
(that
is,
better)
than
the
design level, 71. (Note that
whereupon u is defined b~7
nA =arc cosh 7.)The chief disparity between the pracfi
(r=
(41:) tical pattern and theideal is in the beamwidth, which is
.\/A?
(e - +)Z
greater than the ideal by a factor almost exactly equal
to cr. However, fi does not have to
be very large to make
We now propose to construct an entire function with u only a few per cent greater thanunity.8
the following zeros:
T o summarize,themembers of the pattern family
F
(
z , A , r i ) have two independent characteristics:
z , = k u.\/A2
( n - +)?
for 1 I
IZ < n
Fl(2,

A ) = cos Tr

(38)

(I U I

(I U I

IU I

z*= f n

The canonical product is

for ri

5n<

x .

(42)

1. The design side-lobe ratio, 71.

2. The boundary of the region of uniform side lobes,
e, an integer.
6 For example,if the design side-lobe level is to be 25 db, = 17.78;
-4 = 1.137. Then if n = 5, u = 1.077 and thebeamwidth of the practical
pattern will be only 7.7 per cent greater than that of Fo(z,1.137),
the ideal pattern.

Taylor: Deign of Line-Source Antennas for Narrow

Beamwidth

1955

Low
and

Side Lobes

23

TABLE I1
(1)

(3)

Design side7
lobe ratio
(Side-lobe
voltage ratio)
(in db)

0
1.00000
5
1.77828
34.49
10
3 116228
15
5.62341
20
10.0000
25
17.7828
30
31 .6228
64.78
56.2341
35 2.25976
100.0000
40

(2)

(4)

18080
-

(in deg.)
28.65
~.-

45.93
51.17
56.04

Values of the parameter u

A2

(5)

68.76

0.00000
0.14067
0.33504
0.58950
0.90777
1.29177
1.74229

2.84428

1.33333
1.29351
40.33
1.12549
60.55

Once these two parameters have been chosen, the pattern, the distribution function, and
all other relevant
data may be calcultited. In selecting FZ, it is essential to
avoid valuesthat are too small; the change
from Fl(z, A )
to F ( z , A , F Z ) ,in which the trans-ri zeros migrate to the
integers, must be such as never to increase the spacing
between any of these zeros. In order that this condition
be fulfilled, ti must be chosen such that a unit increase
in li does not increase u. Practically, this means thatfor
a space factor with a design side-lobe ratio of 25 db, FZ
must be a t least 3, and that for a design side-lobe ratio of 40 db, i t must be at least 6. Apart from this limitationonehasconsiderableliberty
in the choice of
ti;large values makeCT more nearly equalto unity, thereby sharpening the beam;however, the additionalbenefit
obtained in this way soon becomesnegligible. If r~ is
increased sufficiently beyond the endpoint of the visible
range, 2a/X, all the visible side lobestend to become uniform. Such large values of +
have
i
the effect of supergaining the linesource,aphenomenonwhich
will be
considered in detail in the next section. The choice of
FZ may alsobeinfluencedbyconsiderations
of gain.
The beamwidth of the ideal space factor (in standard
beamwidths) is a function
of the side-lobe ratio only.
I t may be derivedg from (35) and is given rigorously as
follows:

Bo = A,(

1.20000
1.14712
1.12133
1.09241

1.07692 1.09091
1.06667
1.07514 1.08838
1.06534 1.10727
1.08492 1.10203 1.12754
1.063501.16908
1.072681.24393
1.11631
1.09528
1.18689
1.06112 1.06949 1.08043
1.05816 1.06554 1.07490
1.08698 1.07728 1.06834 1.06083
1 .OS463
1.05538 1.06079
1.06619 1.06934
1.05052
1 .OS231
1 .OS386
1.04587 1.04923
1.04068 1.04241 1.04298
1.14286
1.13635

tains the design side-lobe ratio in decibels-a quantity

whichisusuallychosenfirst
in anysynthesis procedure. Column (2) is self-explanatory. Column (3) gives
PO expressed in degrees. Thismust be multiplied by
X/2a to get the ideal beamwidth for a given line source.
Column (4)gives the parameter A 2 . Column (5) gives
the parameter u for several possible choices
of FZ.As

50

4 Oo

3'0

2 0'

(arc cosh q ) 2 -

%-

A curveshowing the relationship (51) is presented in

Fig. 7. The beamwidth of an F ( z , A , FZ)space factor is
given by the following extremely good approximation:

Table I1 (above) shows relationships among various

quantities discussed here. Column (1) of the table con9 R. J. Stegen,,,
"Excitation coefficients and beamwidths of
Tchebycheff arrays, PROC.I.R.E., vol. 41, pp. 1671-74; November,
1953.
Stegen has derived essentially the same formula by considering
Dolph-Tchebycheff arrays of large numbers of elements.

1.11111
1.18672

IO

20

30

40

50

SIDELOBERATIO,DECIBELS

Fig. 7-Beam

width vs. side-lobe ratio for ideal space factor.

an example,suppose a linesourcewitha30-dbsidelobelevelisrequired
and that the aperture-to-wavelength ratio is 50, that is (2a/X) =SO. Then the ideal
beamwidth is 60.55/50 or 1.211 deg. If it is decided t o
use a n of eight, u = 1.05052, and the practical beamwidth, thatis, the beamwidth of the function F ( z , A , ti),
is 1.272 deg. The zeros of the pattern and the pattern
itself may be readilycalculatedusing
u and A2. The
places in the table where there is no entry for u correspond to choices of rt which are too small, as has been
explained earlier.
~

I.RdE TRAhTSACTIONS-ANTENNAS AhTD PROPAGATION

84

January

GAINAND SUPER-GAIN
IN LINESOURCE
ANTENNAS
In this article the term
gain of a line source means the

specific gain changes rapidly with respect to this quantity; when, however, 2a/X passes into the region where
ratio of power flux radiated in the direction of maximum the side lobes of F ( u ) have become low and are getting
intensity to the value of this flux averaged over all di- lower by virtue of their asymptotic decay properties,
rections. If 8, g5 are spherical co-ordinates in terms of the specific gain becomes practically constant and unwhich the radiation field is described and if P(4, 8) is dergoes but little further changeif 2a/X increases all the
the element factor (with dimensions of power-per-unit way to infinity. As an example, the specific gain of a
solid angle),thetotal
power flux pattern is simply uniform distribution is plotted with respect to 2a/X in
P(8,g5) S (cos 8) 2. With el, 41 representing the direction Fig. 8. Evidently, the limiting value of the specific gain
is an important line-source parameter, especiaIly since
of maximum intensity, thegain is givenby
it depends only upon the functional form of g ( p ) . This
p(el,+1) COS el)
is shown by the following expression, in which the ParG=
. (52) seval formulalofor Fourier integrals hasbeen employed:
JozJop(o, $1
e) 12 sin eded4
4a

l2

-L

I cos

I t has been mentioned that P(8, 4) usually has little

directivity, especially in the 8 co-ordinate. If S(cos 8) is
strongly directive and not endfire (8, not near 0 or r),
then a very good approximation for the gain is obtained
by assuming that S has the properties of an
impulse
function and thatP(81,4 ) can therefore be factored out
of the 6 integral
G &

P(81,41)

T h e second factor in the above will be called Go and is

the gain which the entire antenna would have if the
element factor were isotropic; from the point of view of
line-source design, Go is the quantity which one would
naturally discuss when comparing various line sources
on the basis of gain. In the present case the beam is
presumed to be broadside and cos 81=0. Replacing cos
8 by v, Gobecomes

The quantity XGo/4a, which will be called the specific

gain, is of considerable importance in what is t o follow.
I t is, of course, given by

Now consider a line source with a distribution g ( p ) of

fixed functional form and let the ratioof physical length
to wavelength be increased. This produces no change in
F(u), which is related to g ( p ) through (5), however, it
obviously changes the specific gain via the limits of the
integralin (55). Forvery small values of 2a,Q, the

-0

-2 d X

Fig. 8

If g ( p ) is now varied, it is easy to prove that (XGo/4u),

takes on its greatestpossible value, namely unity,when
g ( p ) is uniform. I t is in this sense, and only in thissense,
that a uniform distribution has a specific gain greater
than that of any other type of line-source distribution.
When 2 a D isfinite,distributionswitharbitrarily
highspecific
gains can be synthesized-at
leaston
paper. Starting with the space factor
of the uniform
distribution, sin m / m c , one can form the space factors
for such super-gain distributions merely by rearranging
the existing zeros to give an increased concentration of
zeros in the visible range. This larger supply of zeros
can be used to reduce the width of the main lobe while
holding the relative level of the visible side lobes constant, thusenhancing the specific gain. At the same time
however, (XGo/$a), must decreasebecause i t had its
l o E. C. Titschmarsh, Theory of Fourier Integrals, Oxford University Press, pp. 50-51; 1948.

1955

Taylor: Design of Line-Source dn.tennasfor Sarrou, Bea-mwidth

and

Lou1 Side Lobes

25

The answer to the question of how large a value of

will depend upon many factors, most of which cannot be considered here. I t seems reasonable, however,to look upon
source distributions for which y exceeds yo by a factorof
ten
or
more
with
extreme
caution.
Thequantity
n m
( A G o / h ) , is especially sensitive to any shifting of the
zeros into or toward the
visible range and it can undergo
a ten-fold change with very little provocation.
Several a r t i ~ l e s ~ Jdealing
~ - ~ ~directlyorindirectly
with the subject of super-gain have appeared. Woodward and LawsonI5 basetheir article upon a two-dimenWhen this ratio is evaluated for a uniform distribution,
sional formulation of the aperture problem but it conit will be designated yo; the expression for this is
tains concepts closely related to those developed here.
r-/ ,2I t remains t o discuss F ( z , A , rZ) line sources from the
(.i81
V =
point of view of super-gain. If 12 is less than 2a/X, the
change from the space factor of a uniform distribution
to an F ( z , A , a ) space factor involves no migration of
Since (XGo/4a),= 1 for auniform distribution, y o is zeros into or toward the visible range, but only a rearnumerically equal to the specific gain in that case and rangement of the zeros already there, and the increase
is therefore also illustrated by Fig. 8.
in y / y o will be inconsequential if it occurs at all. Even
I t is seen from (57) that y is the ratio of the total area if 12 is somewhatlargerthan
2a/h, therelativearea
to thevisible area under ] F ( u ) 2 ; evidently, then, super- under the invisible lobes may still be small, especially if
gaining must increase the relative level of the invisible the design side-lobe level is low. Of course, if ti is inlobes. Large values of y also imply large ratio between creasedindefinitely,apoint
willbe
reachedwhere
mean square valueof g ( p ) and visible area under F(u) 2 , further increase of fi will havelittle effect upon the
i
n r
visible pattern but will continue to raise the level of the
invisible lobes, making them approach the design level ;
the source can thus acquire an arbitrarily
high degreeof
super-gaining. There is, of course, no need to resort t o
large values of fi in order to realize most of the beamsharpeningpotentialities of F(z, A , a) sources,and
This means that, given two line sources with radiating hence there will rarelybe any occasion to usesuch
elements which are identical and with respect to which sources under conditions which would make y / y o large,
the approximationof (53) applies, then themean square even though this is theoretically possible.
values of g ( p ) in the two cases will be t o each other as
By letting 12 tend to infinity, one approaches theideal
the two valuesof y, when both sources are radiating the space factor, Fo(z, A ) ; the latter, therefore, has an insametotal power. This is but one of manyways of finite super-gain ratio associated with it. For this reason,
reaching the conclusion that line sources with an apit is inconceivable that a non-super-gain source could
preciable amount of super-gaining are impractical.
out-perform the ideal space factor with regard to the
Up to this point, the term super-gain has
been used
to refer to line-source distributions whose specific gains
l1 tV. W . Hansen, Notes on Microwaves, published in limited
are higher than that of a uniform distribution with the
edition by engineers of Sperry Gyroscope Co., Inc., vol. 3, pp. 88same ratioof physical length t o wavelength. It has been 100: 1941-1944.
S. A. Schelkunoff, A mathematical theory of linear arrays,
proved that suchsourcesmusthaveassociatedwith
Bell Sys. Tech. Joz~r.,vol. 22, pp. 80-107; January, 1943.
them values of y which are greater than yo. The conC. J. Bouwkamp and S.,G. de Bruijn, The problem of optimum
current distribution, Phillips Res. Reports, vol. 1, pp. 135verse is not necessarily true, for high values of y may antenna
158; 1945-1946.
result from the movement of zeros into the visible range
l4 A. Laemmel, Source distribution functions for small high gain
for the purpose of giving more character to the pattern, antennas, Microwave Res. Inst., Polytech. Inst. of Brooklyn,
ReDort R-137-47. PIB-88: h r i l 4. 1947.
as in beam shaping, in ways which would not increase
I5 H. J. Riblet, Note on <he maximum directivity of an antenna,
I.R.E., vol. 36, pp. 620-623; May, 1948.
the specific gain.Thedebilities of super-gain sources PROC.
l6 R. $1. \$ilmotte, Note on practical limitations in the directioresult, not from the increased gain, but from the in- ity of antennas, PROC.I.R.E., col. 36, pp. 878; July, 1948.
l7 P. M
. Woodward and J. D. Lawson, The theoretical precision
creased value of y. For this reason, it would be preferwith which an arbitrary radiation pattern may be obtained from a
able to extend the definition of a super-gain source to source of finite size, Jozrr. IEE, vol. 95, pp. 363-370; 1948.
I8 T. T. Taylor, A discussion of the maximum directivity of an
include all sources for which y is greaterthan y o ,
antenna, PRoC.I.R.E.,
36, p. 1135; September, 1948.
whether the specific gain is greater than that of the corL. J. Chu, Physical limitations of omni-directional antennas,
respondinguniformdistributionornot.Super-gain
is J . Appl. Plzys.,vol. 19, pp. 1163-11 75; December, 1948.
20 N. Yaru, A note on super-gain antenna arrays, PROC.
I.R.E.,
used in this wider sense hereafter.
vol. 39, pp. 1081-1085; September, 1951.

greatest possible value before the rearrangement of the

zeros took place. It follows t h t super-gaining must increase the value of the ratio of hG0/4a to ( X G o / h ) , with
respect to the value it would have if the distribution were
uniform. This ratio, which has been called the supergain ratio,h is given by

y / y o can be tolerated in a particular situation

1701.

compromisebetweenbeamwidth
and side-lobe level, be useful. This check is based upon the fact that there
especially if the visible range is extensive enough t o in- are two independent methods
for calculating g ( n ) , When
clude a t leastonecompletepair
of side lobes. The a=O, h ( ~ =g(7r)
)
and, from Theorem 111, there is a unibeamwidth-side-loberelationshipgiven
in(51) and il- versal asymptotic expression for F ( z ) ;
lustrated in Fig. 7 maythereforeberegarded
as the
practical optimum compromise for the space factor of a
line source. The words, " . . . for the space factor of
. . . ' are used advisedly, because it is possible to exploit In other words
theelementfactor
(when thereare collinear current
elements, for example) and use a space factor with an
increasing side-lobe characteristicto give a flat sidelobe characteristic in the total pattern. A small reduction of beamwidth can be achieved thus, as Sinclair and Substituting the valueof 2n g ( n ) obtained from (63), we
Cairns21 haveshown for arrays of discrete elements, but find that the F(m) must satisfy the following condition:
such techniques are beyond this article's scope.

THEINYERSETRANSFORM
AND THE
LIMITINGSPECIFIC
GAIN
In calculating the distribution function g(p) to give
the spacefactor F ( z , A , r t ) , thesynthesismethod
of
WoodwardB will be used. Let g(p) be analyzed into its
Fourier components:

By applying the Fourier transform

(5) to this expression, term by term, the following result is obtained:

TZ

The limiting value of the specific gain is easily exhibited in terms of the Dm. Appeal is made to the Parseval formula for Fourier series,1 and the integralin the
denominator of (56) becomes

The integral in the numerator is simply 2nD0, and the

gain expression becomes

ID"I2

m=-m

The terms sin a(z-m)/n(z-m) are displaced from one

another by integral steps along the axis of reals in the
z-plane, hence the principal maximum of any coincides
with a zero of all others. I t follows that F(nz) = 27rD,, or

1.0

0.9

Thus the Fouriercoefficients of g(p) are simply the patternvalues

at the integers.Sincethespacefactors
which have beenconsidered areevenfunctions,
D-,
= D m , and g ( p ) in the range p 25 4 is given by
0.7

If the space factor belongs to the F ( z , A , rt) family, (63)

is veryeasyto
usebecause F ( m , A , rt) is zero for
m 2 e, and the summation terminates.
I t is of the utmost importance that the F(m) be calculated correctly, hence the following simple check may

0.6
0

-5

Fig. 9

G. SiEclair and F. V. Cairns, "Optimum patterns for arrays of

non-isotroolc sources," T a n s . I.R.E., vol. PG-IP-1, pp. S(r59;
~~

February,'1952.
22 p. pvI. \Voodmard, "A method of calculating the field over a
plane aperture required to produce a given polar diagram,'! Jour
I.E.E., 1701. 93, pp. 1554-1558; 1946.

Using this expression, the quantity (XGo/$U),, has been

plotted as a function of A for four F ( z , A , fi) line sources
with design side-lobe ratios of 10, 15, 20, 25 db (Fig. 9).

Taylor: Design of Line-Source Antennas for iyarrozc Beamwidth

1955

and

Lou! Side Lobes

,lO DEDESIGN

SIDE LOBE R A T I O

n=5

I FI
I .o

0.5

--I

-I

IFI
1.0

1.c

0.5 ..
0.5

-7r

-u
8

1.a
1.0

0.5

Fig. 10

I*R*E TRANSACTION8-AMTEII';YAS

AND PROPAGATIOA'

Janua.ry

P(P)

1.0

SIDE

LOBE RATIO
?i= I O

0.5

IO
Fig. 11

According to (57), when y A 1 the limiting value of the

specific gain is an excellent approximationforthe
specific gain itself. The quantity
y will be approximately
equal to unity
if the following conditionsaresimultaneously satisfied :

1. The ratio 2a/X is of the order of or greater than

five, for 1<?OS1.02 when 2a/h25.
2. The source in question has little or no super-gaining andtherefore y&;yo. This will betrue for
F ( z , A , rt) sources when rt is of the order of or less
than 2alX.
These conditions roughly define the domain of utility
of the curvesof Fig. 9 from the point of view of estimating the specific gain. T o be more quantitative, onewould
haveto find the indefinite integral of the function
I F(u, A , rt) z. This hardly seems worthwhile, since the
majority of practical cases,especially in the field of highgain antennas, will fall well within the domain outlined
by the two conditions stated above.

As practical examples of the methodsproposed in this

article, complete calculations for 10, 15, 20, and 25 d b
line sources with 1z =5 have been performed. The results
are illustrated in Fig. 10 (preceding page) which shows
the source distribution and the space factor for each of
these cases. Fig. 11 shows the corresponding data for a
20 d b side-lobe source with rt= 10.
L~CRNOWLEDGMENT

The writer is indebted t o IT.G. Stems for initially

posing the problemconsidered in this article and
for
calculating the 25 d b source andpattern of Fig. 10.
Most of the other numerical data appearing here were
calculated or checked under the directionof Essor Maso
and Dr. IT.L. Doyle; this work is gratefully acknowledged. The writer is further indebted to Dr. L. C. Van
Atta, Dr. R. S. IVehner, and especially to Dr. R. S.
Elliott for their careful reading of the manuscript and
helpful suggestions.