You are on page 1of 8

IEEE TRANSACTIONS

ON A N T E N K ~ S

m -

PROPAGATION, VOL.

a 2 0 , NO.

2, MARCH 1972

129

The Peak Sidelobeof the Phased Array Having Randomly Located Elements
BERNARD D. STEINBERG,
FELLOW,

mm

Abstract-A formula is derived for the peak sidelobe level of a phased array in which theelements are randomly located. The parameters of the formula are the number and size of the array elements, size of the array, wavelength, beamsteering angle, and signal bandwidth. The theory is tested by measurement of the peak sidelobe of several hundred computer-simulated random arrays. Unlike thecase for the conventional array the effect of spatial taper (nonuniform density of element location) upon the peak sidelobe level is minor. The peak sidelobe of the twodimensional planar array is approximately 3 dB higher than thet of thelinear array of the same lengthandsamenumber of elements.

the peak sidelobe in terms of a probability or confidence level that thepredicted value will not be exceeded. Specifically, a probabilistic estimator of the peak sidelobe of the linear uniform random array with equally-weighted elements is derived and tested by Monte Carlo computer simulations. The result is a practical design theory which places the pertinent design parameters in evidence and makes their relative weights apparent.Theparameters are the number and size of the array elements, size of the array, wavelength, beamsteering angle, and signal bandwidth.

I. INTRODUCTION HE COST of a large phased array which is designed primarily for high angular resolution rather than for weak signal detection may be reduced manyfold through thinning, i.e., reducing the number of elements inthe aperture below that of the filled array in which the interelement spacing is nominally onehalf-wavelength. Increasing the interelement spacing hasanothersalutory effect; a separation of a few wavelengths reduces mutual coupling to negligible proportions. Thinning, therefore, is attractive from both points of view. But these benefits are not free of penalty. Unless t.he element locations are randomized or made otherwise nonperiodic, grating lobes appear.' Also, the reduction in the number of elements reduces the designer's control of the radiation pattern in 'the sidelobe region, which in turn influences the level of the largest, or peak, sidelobe. In this paper the peak sidelobe of the random array is studied. The paper does' not argue the merits of random array design; its purpose is to offer a tool to the array designer who, whether by necessity or choice, is designing a random array. Since an a priori description of a random array canonly be given statistically, it is logical to seek a solution in a probabilistic sense. Such a solution yields an estimator of

1 1 . LINEAR RANDOM Amy


Consider an array of N unit, isotropic and monochromatic radiators a t locations xn on a line of length L. The x,, are chosen from a set of independent random variables described by some first probability density distribution, initially assumed to be uniform over L. It is assumed that each element, irrespective of its location, is properly phased so that a main lobe of maximum strength is formed a t e,, which is measured from the normal to the array. The reduced angular variable u = sin e - sin 0, contains the beamsteering information. The complex far-fieldradiation pattern f(u) is the Fourier transform of the current density. Since the latter is a set of delta functions, f(u) is proportional to the sum of unit vectors having phase angles kxnu, k = 2u/X. The array factor is
N

f(u> =

C exp ( 5 k u ) .
-1

(1)

The main lobe amplitude is N , occurring a t u = 0. Outside the neighborhood of the main lobe, the unit vectors combine withrandom phases.Hence thermsamplitude is N112 and its square, which is the mean of the power pattern, is N . Thus the power ratio of the average sidelobe to the main lobe is N / N 2 = 1/N. Averages alone, however,may be somewhat misleading. Manuscript received March 17, 1971; revised October 19, 1971. s the calculated power This paper is part of a dissertation submitted t.o the University Fig. 1 illustrates the situation. It i of Pennsylvania, Philadelphia, Pa., inpartial fulfillment of the pattern for a random array of 30 elements in which the requirements for the Ph.D. degree. The author i . ; with the Moore School of Electrical Engineering, probability density function (pdf) of element location is University of Pennsylvania, Philadelphia, Pa. 19104. Extensive theoretical work of the last decade on the aperiodic uniform over L = 70X, the element positions having been generation program. The n [1]-[7](deterministic aperiodic array), 1 8 1 1 1 5 1 chosen byarandomnumber array is found i (nondeterministic randomarray),and [lS]-[19](nondeterministic beamsteering angle O0 = 0. The ordinate scale is in dB, statistical array). The random m a y is characterized by element locat,ions chosen by some random process; in the statistical a m y and the abscissa is linear in degrees. The sidelobe level a conventional array is designed, and a given fraction of the elements averages about -15 dB, which is consistent with the is removed at random. In a l l cases a set of phase shifters is assumed preceding theorem (- 10 log 30 N 15 dB). The theory so as to eophase the elements at the beamsteering angle.

.IEEE ?RANSAClIONS

130

O N ANTENNAS AND PROPAGATION, MARCH

1972

which is limited in bandwidth to W and in time duration to T is 2WT [20]. The complex radiation pattern of a random array is such a band-limited function, the limit being due to the finite length of the array. The comparison is evident from the Fourier transforms re1a.ting signal spectrum F ( f ) and timefunction f(t), and currentdensity i(s/X) and far-field complex ra,diat,ion pattern f(u) :
Fig. 1. Pattern of 70-wavelength random array of 30 isotropic elements.

is inadequate, however, in not alerting the designer to the peak value, which is about -7 dB, or 8 dB higher than
the average level.

1 1 1 . PEAKSIDELOBE ESTIMATOR
Estimation of the peak sidelobe requires knowledge of the pdf of pattern magnitude. By the central limit theorem (large N ) the quadrature components of the array factor in the sidelobe region are asymptotically normal [ S I and, therefore, its amplitudedistribution is Ilayleigh. Hence the asymptotic pdf is

where t t)u, f w x/X, W t)L/X, and T w nonredundant port,ion of the visible domain of u. The latter interval is usually less than two because the power pat.tern is symmetrical about t,he beamsteering angle u = 0.This follows from (1) and the defining expression for the power pattern of a.n array of unit radiators

f(u>f*(u) =

c c exP Mz72 - Gn)u


n m

The integral

=Lo
m

p ( A ) dA

exp

is the probability thatthe magnitude of an arbitrary sample of the radiation pattern, away from the region of the main lobe, exceeds somethreshold A,. Its complement, 1 - a, is the probability that such a sample is less than Ao. I f n independent samples are taken

whichalso equals f( -u)F ( -u) since the index n is indistinguishable from m. The visible domain is -1 - sin eo 5 u 5 1 - sin eo. The length of the nonredundant portion i s 1 I sin 00 I. Consequently, the number of independent samples needed to specify the complex radiation pattern is 2(L/X)(1 I sin eo I). Ha.lf this number may be associated with the amplitude of the array fa,ctor and half with its phase. Therefore, the power pattern is uniquely specified by

is the probability that none exceeds Ao. From (3), Ao2= - N In (1 - Oln). It is convenient to normalize this expression to N , the average sidelobe level, and to give the dimensionless power ratio Ao2/N a new symbol, B. Thus

-In (1 - P

n ) v

In n - InIn

,9-l.

(4)

B may be interpreted as a statistical estimator of the power ratio of the peak-to-average sidelobe of a set of n independent samples. B is a confidence level; it is the probability that none of n independent samples of the sidelobe power pattern exceeds the mean value by the factor B. n is an array parameter, which is a function of all the relevant array properties other than N . It is proportional to the number of sidelobes in the visible region. It may be calculated in several ways. An interesting method utilizes the Nyquist sampling theorem. That theorem, usually applied to band-limited random processes, asserts that the number of samples required to specify a wave

independent samples, the average angularinterval between samples being AIL. n is dominated by t.he length of the arrayin units of wavelength and secondarily influenced by the beamsteering angle. Equations (4) and (5), however, are insufficient to provide an unbiased estimate of the peak sidelobe. The probability is zero that any finite set of samples of a power pattern falls exactly upon the crest of the largest sidelobe. Hence such estimation is downward biased. A correction t.0 (4) may be obtained by calculating the difference between the largest of a set of samples and the height of the lobe from which t,he sample is taken. Consider a. random sample point a t u1close to a lobe crest. In t.he neighborhood of ul,the power pattern can be represented by the Ta.ylor expansion

STEMBERG: PEAH SIDELOBE OF PHASED ARRAY

131

where Pt(ul) is the first, derivative with respect to u. of P ( u = ul),etc. For sufficiently small int,ervals the first t.hree terms of t,he series suffice to describe t.he lobe crest.. Its peak (at. u = up)is located by sett,ing the derivat.ive of t,he t.runca.t,edseries to zero. Thus dP/du = 0 = P' P"(uP - ul), and up - u1 = -Pt/P". Insert,ing up into (6) (truncated)andsubtract,ing P ( u I ) gives the difference AP between the peak value P(u,) and t.he sample value P ( u l ) in t.erms of the first t x o derivatives at the sample point,:

mated by

where it is understood t.hat P = P(u1). The sum of (9) and P(u1) is the approximate height of the lobe crest,. If P(u1) is t<helargest. of a complete set of independent samples of the power pat.tern, the probabilit,y is high t*hat it falls on the largest lobe. Hence P ( u I ) a.lsois an estimat,e of the largest sidelobe. Dividing t.his sum by N normalizes it t,o. the average value, as in (4).The first term is exactly B. The second term is

To evaluate t,his expression the array fact.or is represented as a sum of quadrature comp0nent.s a ( u ) and b ( u ) , the power pattern P ( u ) = a2(u) b 2 ( u ) ,and the deriva- TogeOher they form a better estimator than (4),except tives P' = 2(aa' bb')and Prt = 2(m" aJ2 bb" b'2), in the neighborhood of t.hesingu1arit.y at B = 2. The where a = a(u,), etc. The increment difficulty in this region can be avoided without affecting the quality of t.he estimator for large B by truncating the a2a'z 2aa'bb' b2bf2 AP = series in (10) a t two t.erms, thereby removing the singuaa" at2 bb" bI2 hrity. The approximate mean increment reduces to the average value of which can be est,imat,ed.In the.side- 1 2/B, and the estimat,or of the normalized peak belobe region, a and b are asymptot.ically Gaussian random comes

+ + +

-+ + + + +

variables [8] 6hich ca.n be shown to be independent. and 2 Bp=B+l+stationary [21]. The pdfof the six variables in (7) are B' known (e.g., [22, ch. 31). The quadrat.ure components a and b are uncorrelated with their fist derivatives, i.e., The power ra.tio of the pea.k sidelobeto the main lobe is aa' = = 0, but are highly correlat.edwith their second derivatives. In [22, eq. (3.16)] it. is shown that for large peak sidelobe = peak avg = B p . ( i ) = B 1 2/B N main lobe avg main thresholds t,he pdf of arr (or b") is a narrowly peaked Gaussian distribution. Consequently, a" (or b") can be (12) regarded as nonrandom and equal to it,s condit.iona1 mea.n -&a (or - a b ) . The coeEcient Experimental data indicate that the estimator closely matches the data when B 2 3. The fa.& that the match is satisfactory for B as small as 3 implies t,hat (11) is useful even for small a.rrays. Using B = 3 in (4)gives the smallest array for which the est.imator is satisfactory. For where R ( u ) is t.he aut.ocorrelation function of t.he quadra- the broadside case (eo = 0), the value is 2 X for #? = 0.9, ture components. Thus aa" = -&a2, and aa" bb" = and 14X for fl = 0.5. Equation (11) is plotted in Fig. 2 -R2(a2 b2) = -R2P. The ot.her t.erms in the denomi- for B > 3. The independent variable is the a.rray paramnator are af2and bt2.Their pdf are exponent-ial(calculated eter n = (L/X)(1 1 sin 00 ; the paramet.er of the famfrom [22, eq. (3.15)]) ily is 8, the confidence level. Fig. 3 shows a test of (11) for 00 = 0 and eight different values of n ranging from 7 to 700. Ten random arrays were formed by computer simulation for the largest value and, therefore, are not narrowly peaked. However, for a ( n = 700). For the remaining seven groups, 20 arrays threshold large compared t.o N , the contributions of these each were formed. Of the 150 arrays, 70 had 30 elements terms are not large so that each may be represented in and 80 had 100 elements. The array fact.or was calculated the denominaaor by it.s average value. From (8) 3 = for ea.ch army and itspeak sidelobe found and measured. = NR2. Hence the denominator of (7) may be approxi- For each of the eight groups the experimental distribution mated by the nonrandom quantit.y -R2P 2NR2 = of the peak sidelobe was determined, from which the per-R!P( 1 - 2N/P). centile data of Fig. 3 were obt.ained. These data are plotted The average value of (7) is approximately tjhe average along with the theoretical curves of Fig. 2 for correspondof its numerator divided by this quantity. The a.verageof ing values of @. The agreement is highly satisfactory. the middle numerator t.ermis zero, as observed previously. In Fig. 4 the predict>edrise in the peak sidelobe due to The remaining numera.tor terms average to N%(u2 b2) = beamsteering angle is tested. Ten random arrays of length N&P, and t,he average of (7) becomes well approxi- lOOX were formed. Each array was steered successively

-.-

+ +

I)

. - .

132

IEEE TPLNSACTIONS ON ANTEWAS AND PROPAGATION,MARCH

1972

= ~.QQQ

to eo = 0", 15", 30", 45", and 60". The median of the peaks of each set is plott,ed against the arra.y pa.rameter and shown in relation to the theoretical curve, for p = 0.5. The data points for this particular set of random arra.ys fall above the curve; nevertheless, it is evident t.hat t.he growth in the median peak is consistent with the theory.

IV. NONISOTROPIC ELEMENTS


The foregoing theory assumes isotropic array elements. When the elements a,re not isotropic, the element pattern suppresses sidelobes of the array factor which are outside the central portion of the main lobe of the element pattern.The effect is to reduce t.he number of sidelobes which can compete to be t.he largest, thereby decreasing the a.rray pa.rameter n a.nd the peak sidelobe. Consider a sidelobe of the arra.y factor at some angle el # 0. Let. eo = 0. If the elements radiated isotropically the probability thatthat sidelobe would exceed some threshold B is a = exp ( - B ) , as before. However, the lobe st.rength a t 81 is reduced by Gl < 1, the relative gain of the element. power pattern. This is equivalent to increasing t,he t-hreshold by its reciprocal. Hence the probability that the sidelobe at el exceeds the threshold B is a 1 = exp ( -B/G1) < CY. The proba.bility that itdoes not exceed the threshold is its complement, and the probability t,hat none exceeds B is

I I l l l l '

I I Ill1

I I IIII

10

100

1000

ARRAY PARWETER

n =

(LI~)c~+IsINE~J)

Fig. 2. Probabilistic est.imator of peak sidelobe of random array. N is number o f array elements, PSL/ML is power ratio of peak sidelobe to main lobe, 6 is probability or confidence level that. no sidelobeexceeds ordinate, L is array length, X is wavelength, en is beamsteering angle.

'.8

B:O.

N=100,30

where L/X is the number of independent samples of the magnitude of the array factor when eo = 0 (see ( 5 ) ) . By equatingthisproduct to [l - exp ( - B ) ] " a reduced number n is found whichis the equivalent number of independent sa.mples of an a.rray of isot.ropic radiators having the same peak sidelobe level. Taking logarithms of both sides
1000

10
ARRAY PARAMETER

100 (L/X)(l+[SlNe,l)

Fig. 3. Each sample point is indicated percentile of experimental distribution of peak sidelobe of group of linear random arrays. Solid curves are from Fig. 2.

1 - exp -Gi

-">

= 12

In [I - exp ( - B ) ] .

87B

Since B is on the order of 5 or more,exp ( - B ) and exp -B/Gi are smallenought.o permit, replacement of t.he logarithm by the first term of t,heseriesexpansion In (1 x ) N x leading to

65I
'

N=100

G is the normalized square magnit.ude of the element fa.ctor


g ( u ) which, near the origin, may be approximated by the quadratic g ( u ) 'v g(0) ( 1/2)g"(0)uz. The curvature

10
ARRAY PARAMETER

I I lllll I 100 n = CL/~)(l+151N8~1)

L=lOoX I I I IIII 1000

Fig. 4 Effect of beamsteering angle 80 upon peak sidelobe. Each sample point is median of experimental distribution of peak sidelobe o f ten linear random arrays.

is relat.ed to t.he current density i ( x ) across the element. It is calculat.ed by twice differentiating tlhe Fourier t.ransform relation g(u) = Ji( x) exp ( jksu) dl: and evaluating the second derivative g"(u) = J - kzs2i(x ) exp ( jksu) dx at the origin. Thus g"(0) = -k2Jx2i( x) d x p -kzIZg(O), where P is the second moment of the current excitation

STEINBERG: PEAK SIDELOBE OF PH&SED A R R A Y

133

across the element., and

G(u) = I g ( u ) / g ( O ) l2 = 1 - k2Pu2, for small angles. Similarly, 1/G N 1 k212u2 and 1/Gi - 1 2r: k 2 h i 2 = 4d12u:/X2. Since ui = Xi/L, X/L being the sampling interval, (13) becomes

Equation (15) may be written

I I I IIII

10
AURAY PARW,ETER

I I I IIII 100

I I I IILj

l0OC

L/~>.sB~

where 2 = L2/8P12B. The first term integrates to a(?r/2) lI2. The second term is less t.han 5 percent of the first term for L/Xa 2 2 (evaluated from [23, eq. (26.2.12)]. Since the dependence of B upon n is logarithmic (see (4) ) , a 5-percent error isexceedingly small. Thus

Fig. 5 . Effect, of elementsize 1 = S A uponpeaksidelobe. Each sample point, is median of experimental dktribution of peak sidelobe of ten linear random arrays.

The second integral is (15) which eva,luat,es to (16). The first integral represents the increase in n. due to steering off t,he array normal. For small steering angles

Io (5)
exp
dy

As an example, consider an element, of length I uniformly illuminated over it.s length. Thus

s X

-7n

?n3 m -6a2

+-

- *

and the condition on B becomes B 2 6/.K2.9, which is and easily satisfied for s 2 1/2. For smaller e1ement.sthe effect, upon t.hesideIobe pattern disappears and the isot.ropic case pert.ains. The condition is similarly satisfied for most practicalcurrent distributions. The restrict.ion on (16), V. WIDE-BAND SIGNAL t.herefore, may be dropped. For the uniformly illuminated element Spreading the signal energy over a band of frequencies further reduces n. Let there be two frequency components, w1 and w 2 , in the signal passing t.hrough the array. The composit,e complex pat.tern is the weighted sum f(u) = Fig. 5 shows two test.s of (17), one for L = lOOX and f ( u , w ~ ) Q f(u,~) where , q is t.he relat,ive strengths of N = 100, and t.he other for L = 4001 and N = 30. I n the two component,s. More generally, f(u) = fs(w)f(u,w) dw, each test ten uniform random arrays were generat.ed. The where s ( w ) , w = 2 r v , is hhe Fourier spectrumof the signal. element,wasassumed t.0 be illuminated uniformly. The va,lues of s were 1/2, 1, 2, 4, 8, and 16 for t.he 4oox array This process leaves the centra1 port,ion of the pataterninand 1, 2, 4 , 8 for bhe lOOX array. The st.eering angle was tact while smoothing the portions away from the main eo = 0. The peak sidelobe of eacharray wa,sfound for each lobe. Consequently, the sidelobecrest,s shrink and t.he value of s, and t.he median of each group is plotked in the nulls fill in, t.hereby diminishing t,he probability tha,t. t,he figure along with the t,heoretical curve for /3 = 0.5. The distant sidelobes contribute the pea.k of the random sideisotropic case (s = 0) is also included. The variation of lobe pattern. For this calculation it is useful to make (1) an explicit peak sidelobe with element size appears consistent wit.h funct,ion of w ; replacing k by w / c , c = speed of light,, theory. To account, for eo # 0 t.he lower limit in (15) is shift.ed by m, the number of independent samples between the origin and sin eo, i.e., 1 sin 00 I = m(X/L). Equation (15) becomes Consider a crest in the sidelobe pat,tern occurring at UO,WO$ LIX -y + nNJ exp-dy = where wo is the nomina,l cent,erof the signal spect.rum. I n --m 22 0 the neighborhood o f wo and at u = uo, f(u,w) may be

/-: ... 1 ....

-c

L -

134

IEEE TRANSACTIONS ON A K ~ N N A S AND PROPAGATIOK, MIARCH 1972

approximated by

8-0.5

np:

Lo j

where the coefficient of (w - wo)2 is proportional to the curvature at. the crest with respect to w. This quantit.y, being a function of a random variable, is it.self a random I I I I I I IIII I I I 1 1 1 1 1 l I I I I IIll variable, the dist.ribution of which is needed for the calcu10 100 1000 lation of the effect of bandwidth upon the peak sidelobe. ARRAY PARAMETER n CLta~cl~lsrtao~) However,because of the weak (approximately 1ogarit.h- Fig. 6. Effect of number of elements h T uponpeaksidelobe. Each mic) dependence of B upon n, an estimate of t.he effect of sample point is median of experimental distribution of peak sidelobe of ten linear random arrays. bandwidth can be obtained by replacing the ra.ndom variable with a typical value, a.s, for example, that which is obtained from the ensemble average array factor. This signal spectrum. Hence function is the Fourier t.ransform of t.he first probability density distribution of element location [SI. For the uniform random array of length L it is

fo = 1J L l 2
2
-LIZ

exp

f?)
= wo

- sin d u / 2 c
OLu/2C

(19)

Near a lobe crest (u


f(uo,wo

uo, w

+ So)
(20)

under fairly general conditions. The crossover occurs near (J,,/AWB~~ = L/X or a t abandwidth which is approximately the reciprocal of the travel time of light across the array.

LuoSw/2c 1 - (1/2)(Ww/2c)2 + Sw) =cos w&uo/2c wolu0/2c

VI. FINJTE NTT~WER OF ELE~~ENTS


The maximum possible sidelobe amplitude is N , which is equa.1to the main lobe. On the other hand, theasymp totic pdf (2) on which the theory is based is unbounded. Hence discrepancies may be expected for small N . It is pertinent to determine the smallest value of N for which the asymptotic theory is an adequate representation. A measure of the difference between the finite sample distribution and the asymptotic distribution is the integral of (2) from N to a, which is exp (- N ) This quantity shrinks rapidly with N , e.g., exp (-5) < 1W2 while exp ( - 10) < 10-1 and exp ( - 15) N 3 X lo-. The latter values roughly bound the transitionregion between a high qualityand low quality approximation. Thus N 2 15 would be satisfactory for most problems. An additiona.1constraint follows from the fact that the maximum value of A cannot exceed N : the maximum value which B can attain also is numerically equal to N ; i.e., E,, = Amm2/N= N 2 / N = N . However, the maximum theoretical value of B from the asymptotic theory is unbounded. Hence the asymptotic theory is satisfactory only when N is sufficiently greater than the value of B calculated from the theory so that t.heeffect of finite sample size is not felt. Fig. 6 shows a test to determine how small N may be with respect to this second problem. The rnedian peak sidelobe of ten arrays each is plotted as a function of n for three values of N (10, 30, and 100). Also plotted is the probabilistic estimator, for B = 0.5. The curve should approximate the data except for those tests where the normalized peak approaches N , in which cases the data

The magnitude of the crest,, compared to its magnitude for zero bandwidth, ist.he integral of (20)weighted by the signal spect,rum and normalized to the same weighted integral wit.h Sw = 0. This quantit,yis the gain g(%) by o is reduced due to hhe nonzero which the lobe crest at. u aidt,h of the signal spectrum. As an example, for the uniform spectrum of width Aw

Equation (21) has the same form as (14).By equating the coeEcients of t,he angular variables, solving for I and yuthing it into (lS), the array parameter is found to be .?z = oo/AwB112 (L/X) I sin eo 1, subject to the condition that it not exceed the number of independentsample points given by ( 5 ) . Thus n becomes the smaller of the preceding value or (5). Inclusion of the curvature distributioninthe derivation wouldmodify the bandwidth term by a coefficient but would not alter its functional form. A similar comment pertains to the shape of the

STEINBERG: P E d g SIDELOBE OF PHASED ARRAY

135

should drop below the curve. The data for N = 100 and N = 30 do continue to increase with n over the entire range shown. In both cases the limiting va.lue N is not approached. The data for N = 10, on the otherhand, drop and remain below the theoretical curve when they reach within a factor of two of the ma.ximumpossible value, suggesting that thenumber of elements rat.her than the number of sidelobes has begun to dominate. These observations lead to the conclusion that the minimum number of elements for which the theory is satisfactory is the larger of 15 or 2B(n,B),or

N,,

(15,2Bj.

(23)

VII. SPATIAL TAPER OR NONUNIFORM ELEMENT


DISTRIBUTION Tapering the current density across an array is a commonly used procedure for reducing the sidelobe level. It might be expected, therefore, also to be useful in random arrays. This appears not to be the case. The analog of durrent density for the random array is the pdf of element location [SI. Its effect upon the peak sidelobe is through the array parameter n, which is proportional to the -number of sidelobes. With a spatially tapered aperture, i.e., higher density of elements near the cent,er than near the edges, the beam cross section is increased, and n decreases proportionally. However, the slope of the peak sidelobe level with n is small (see (4)). tapering can usually be neglected. For example, changing from a uniform aperture to a triangular excitation in a conventional array increases the lobe cross section by 44 percent and reduces the peak sidelobe by approximately 13 dB. The lobe width of a random array with triangula.r pdf is similarly increased by 44 percent, and ita array parameter n is reduced by the same amount. However, this reduction in n has a very modest effect upon the peak sidelobe. Consider a lOOX arraysteered to eo = 2 6 O . Thearrayparameter n = (L/X) (1 I sin eo I) = 144. The array parameter for a triangular array of the same length is 100. At the 90-percent level of confidence (from Fig. 2) B, = 8.51 and 8.18, respectively, for the two arrays, which amounts to 0.2 dB difference, demonstrating that the effect of spatial taper upon peak sidelobe level is slight.

independent samples over the hemisphere is approximately rLLz/X2. The same result pertains to a three-dimensional array in which LlLz is the projected area upon a plane perpendicular to the axis of the main lobe of the element factor. Symmetry in the pattern reduces the number of independent samples. With the array steered to the zenith (0, = 0) each lobe in every polar cut has an image lobe in the same p h e . Thus the range of variation of n with eo is a factor of two. The logarit,hmicrelation (4) between peak sidelobe and the array para.meter minimizes the importance of the detailed variation. The dominant feature is the approximate squaring of n when a. fixed number of elements N is spread from a linear array toa planar array of the same length and width. The result is (approximately) a doubling, or % d B increase, in the peak sidelobe. IX. CONCLUSIONS
1) N (number of elements) is the dominant quantity, influencing the peak sidelobe linearly. 2) X (wavelength), L (array size), I (element size), and Au (signal bandwidth) have logarithmic effect8 upon the peak and, therefore, are less influential in determining the peak value than N . 3) 0, (beamteering angle) has a minor influence. 4) The effect of spatial taper of the element density upon the peak sidelobe is slight. 5 ) The peak sidelobe of the random planar array is approximately 3 dB larger than that of the ra.ndom linear array of the same length and same number of elements.
~

ACKNOWLEDGMENT Theauthor wishes to a.cknowledge with appreciation the many valuable suggestions made by Prof. R. S. Berkowitz of the the Moore Schoolof Electrical Engineering. In addition, discussions with A. E. Zeger of General Atronics Corporation, and the fruitful ideas which developed from them, are much apprecia,t,ed,as is the machine and program aasist,ance of other former colleagues at General Atronics, particularly J. T. Beardwood, 1 1 1 , M. L. Cohen, and E. D. Banta. REFERENCES
[I] D. D. King, R. F. Packard,and R. K. Thomas,UnequallyIRE Trans. Antennas spaced, broad-band antenna arrays, Propagat., vol. AP-8, pp. 380-384, July 1960. [2] S. S. Sandler, Some equivalence between equally and unequallyspacedarrays, IRE Trans. Antennas Propagat., vol. Ap-8, pp. 496500,Sept. 1960. [3] A. L. Maffett, (Array factors with nonuniform spacing parameters, IRE Trans. Antennas Propagat., vol. AP-10, pp. 131-136, Mar. 1962. [4] R E. Willey,SpacetaDeringoflinearandplanararrays, IRE Trans. Antennas Propagat., vol. AP-IO, pp. 369-357, July 1962. [5] A. Ishimaru, Theory of unequauy-spaced arrays, IRE Trans. Antennas Propagat., vol. AP-10, pp. 691-702, Nov. 1962: 1 6 1 M. I. Skolnikand J. N. Sherman,Planararrays mth unequally spaced elements, Rudw Electron. Eng., vol. 2s, no. 3, Sept. 1964. [7] A. Ishimam and Y. S. Chen, Thinning and broadbanding antennaarrays by unequal spacings, IEEE Trans. Antenrm Propugat., vol. AP-13, pp. 34-42, Jan. 1965. [SI Y. T . Lo, A mathematicaltheory of antennaarrayswith

ARRAY VIII. PLANAR Extension of the peak sidelobe theory to two- and three-dimensional arrays requires only a reevaluation of the arrayparameter n. Consider aa an example a rectangular planar array having sides L and Lz and uniform pdf of element location. The average patterns in planes through the center of the array and parallel to either of the two sides is given by (19). The angular interval for independent sampling of the patternamplitude in these orthogonal planes is A/L and X/& (Section 111). The area in the U I - w plane associated with each sample point is on the order of X2/LL. The visible area of the plane, which is a circle of unit radius, is T . Hence the maximum number of

136

IEEE TRANSACTIONS ON ANTl3NNAS AND PROPAGATION, VOL.

AP-20,

NO. 2, MARCH

1972

randomly spaced elements, IRE Tram. Antennas Propagat., vol. AP-12, pp. W7,268, May 1964. [9] , A probabilatlc approachto the problem of large antenna arrays, Rudw Sci., vol. 68D, pp. 1011-1019, Sept. 1964. [lo] Y. T. Lo and S. W. f ; e e , Sidelobe level of nonuniformly IEEE Trans. Antennas Propagd. spaced antenna arrays, (Cmmun.), vol. AP-13, pp. 817-818, Sept. 1965. [Ill , A study of space-tapered arrays, IEEE Trans.Antennas Propagat., vol. AP-14, pp. 22-30, Jan. 1966. [12] Y. T.Lo and R. J. Simcoe, An experiment on antenna arrays with randomly spaced elements, IEEE Trans. Antennas . Propagat., vol. AP-15, pp. 231-235, Mar. 1967. [13] Y. T. Lo and V. D. Agrawal, A method for removing blindness in phased arrays, Proc IEEE (Lett.), vol. 56, pp. 1586-1588, Sept. 1968. [14] A. R. Panicaliand Y. Lo,A erobabilisiticapproach to large circular and sphencal arrays, IEEETrans. Antennas Propagat., vol. AP-17, pp. 514-522, July 1969. [15] V. D. Agrawal and Y. T. Lo, Distribution of sidelobe level in random arrays, Proc IEEE (Lett..), vol. 57, pp. 1764-1765, Oct. 1969.

T.

[16] J. L. Allen,Some extensions of the theory of random error effects onarraypatterns, in Phased Array. Radar Studies. Lexington, Mass.:Lincoln Lab.Rep. 236, ch. 3, pt. 3, Nov. 1961. [17] T. M. Maher and D. K. Cheng, Random removal of radiators from large linear arrays, IEEE Trans. Antennas Propagat., vol. AP-11, pp. 106-112, Mar. 1963. [18] Y. T. Lo, Random periodic arrays, Rudw Sci., vol. 3, pp. 425-436. Mav 1968. [19] M. I. SkOl&, Nonu.hiform arrays, in Antenna Theury, pt. 1, R. Collin and F. J. Zucker, Eds. New York: McGraw-Kill, 1969, ch. 6. [20] P. M. Woodward, Probabiility and Infmmation Themy, with Applications to R c r d a p . New York: McGraw-Hill, 1953. [21] R.S. Berkowit.z, Moore School Elec. Eng., Univ. Pennsylvania, Philadelphia, 1971, unpublished memo. 1221 R. L. Stratonovitch, Topics in the Theory of Random Noise, vol. 2. New York: Gordon and Breach, 1967, ch. 3. 1231 M. Abramowitz and I. A. Stegun, Eds., Handbook of M a t h matieal Fundions (Applied Mathematics Series 55). Washington, D. C . : U.S. Govt. Print. Off., 1964.

Gain Measurements of Standard Electromagnetic Horns in the K and K, Bands

[4]. Beatty [5] has given a general discussion of the errors that arise in horn gain measurements, and Bowman [6] has critically reviewed work on this problemprior to 1968, giving estimates of theaccuracy that one might expect to at,t,ainin such determinations. I n order that t,he horns could be calibrated by the twoant,enna technique, two nearly identical pairs of horns were constructed for each band. The two K-band horns, which differ by a small amount in length were electroformed. The K,-band horns, which have identical dimensions, consist, of smooth brass sheets soldered together on I. INTRODUCTION a mandrel. All t.he horns have their inner surfaces silverHE ACCURATE calibration of a radio telescope an- plated. The dimensions of the horns are summarized in tenna by the gain comparison method relies on the Fig. 1. ppecise gain det,ermination of the standard antenna used The calibrat.ion is based on accurate measurements of for the comparison. This art.icle summarizes the calibra- the traiumission loss between the pairs of horns with diftion of two standard horns in K band and K , band which ferent separations. The gains are then worked out with were used in t,he calibrat.ion of a 20-ft diameter millimet,er- aid of the proximity corrections of Chu and Sempiak [l]. %raveantenna. Previous accurate horn gain determinations Finally, the measured gains, a t 4 frequencies in K,, band at longer Wavelengths have been described by Chu and and 3 frequencies in K band, are compared with calcuSemplak [l], Slayton [2], Jull and Deloli [3], and Jakes lated values from the formula of Schelkunoff and Friis [ 7 ] . Before the comparison is made, the estimated ohmic losses in eachhornaresubtractedfrom the calculated Manuscript received March 18, 1971; revised September 20, 1971. gain. These loss estimates are based on measurements of G. T. Wrixonwas with the Radio Astronomy Laboratory, the Space Sciences Laboratory, and the Department of Electrical the Q of each horn in a resonant cavity configuration. Engineering,University of California, Berkeley. He is now with The generally good agreement between theory and measthe Crawford Hill Laboratory, Bell Telephone Laboratories, Inc., urement suggests that Schelkunoffs formula is accurate Holmdel, N. J. 07733. W. J. Welch is with the Radio Astronomy Laboratory, the Space to within about 0.1 dB for horns with flare angles and Sciences Laboratory, and the Department of Electrical Engineering, gains comparable to the ones used in thepresent study. University of California, Berkeley, Calif. 94720.

Abstract-Measurements are described in which the gains of standard-hornsi n the K and K , bands were determined at number a of wavelengths. The two-antenna method was used. It is believed, afteradetailed error analysis,that the gain i s known within a limiting errorof about 0.06 dB ateachwavelengthmeasured. A method is described for measuring the ohmic losses in the horn, so that the measured gains may be compared with theoretical estimates. Gain values calculated from the formula of Schelkunoff and Friis and corrected for ohmic losses agree within about 0.1 dB on the average with the measured gains.

You might also like