0% found this document useful (0 votes)
367 viewsPrinciples of Aperture and Array System Design Part 2
Principles of Aperture and Array System Design Part 2
Uploaded by
JohnCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
367 viewsPrinciples of Aperture and Array System Design Part 2
Principles of Aperture and Array System Design Part 2
Uploaded by
JohnCopyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
/ 238
Aperiodic Arrays
7
7.1 PERIODIC VERSUS APERIODIC ARRAYS
he previous chapter described the rudiments of periodic array design.
A glaring omission in that chapter was the question of whether or not
other design procedures lead to useful arrays. In particular the question of
overdesign was not addressed; that is, do_periodic arrays that avoid
spacin me
performance than some other form of array? Unfortunately there is 20.
single, neat eriodic_array with which to make the
comparison. With the exception of one class of aperiodic arrays, the
random array, such arrays are designed ad hoc and tested.
The complex radiation pattern due to current elements i, located at
positions x, is
flu) = > i, exp (ikxnw) 7.)
Equation (7.1) has two independent design variables, the element loca-
tions x, and their weights i,. Two basic approaches to array design have
developed, one about each of the variables. In a periodic array (Chapters
5 and 6) the elements are equally spaced (interelement distance nominally
123 1s
Aperiodic AFT
ights-
t weid
aa lemen'
-half all the design freedom being in the ol caual-al
Inan aperiog -<—— ‘Stabilized —
n/p
o or 02 03 Of
Figure 85 Radiation pattern of 15-element aperiodic array of length 25A.
Thus the penalty is severe. However in special cases a pattern such as
this one can be useful. For example, if it were necessary to scan no more
than approximately 25°, the poor side radiation characteristics beyond
+12° could be obliterated by a well-designed antenna element having a
radiation pattern that cuts off rapidly at +12° (see Section 5.3, Pattern
Multiplication). Only the first three sidelobes on either side of the main
lobe would remain. Within this restricted scanning sector, the effective
peak sidelobe would not exceed —22 dB. For a restricted problem of this
kind, the three to one savings in the number of elements can be achieved.
Many ad hoc procedures have been designed and are reported in the
literature. Several are listed in the references at the end of this chapter.
7.3. MAIN LOBE VERSUS SIDELOBES
It is evident from the discussion in Chapter 3 that the width and
approximate shape of the main lobe are not severely altered by a spatial
tapering procedure such as described in the previous section. The
width is not affected because the length is fixed. The beam shape is
hardly altered because the current density is an approximation to the
desired current excitation. The dominant effect of the thinning is in the
sidelobes. This effect is due to the nature of the distribution of the phase
shifts across the array, particularly at large scan angles. In a periodic
array the phase at the nth element due to a signal arriving from direction
u, when the beam is steered to broadside, is 6, = kndu. Since phase is
measured modulo 27, the measured quantity is
= 2x (A24_ x) (7.11)
co Main Lobe Versus Sidelobes 131
where K is an integer that reduces the measurement to the interval zero
to 27. Since d, the interelement spacing, is a constant, the set of phases @,,
forms a regular procession for each value of u, resulting in the particular
side radiation pattern associated with this distribution of phases.
In the aperiodic array, the nth phase shift at angle u is
bn (@2-x’) (7.12)
where K' is an integer that constrains the measurement to the 27 interval,
and u isa scale factor on the normalized element distance x,/A. For smal]
values of u, K’ =0 for all values of n, and the phases ¢, are proportional
to the element locations x,. At larger values of u the phases of the outer
elements fall outside the 0 to 27 interval (\K'|>0). The measurement
process, however, reduces the phases by multiples of 27, in effect
“folding” all 27 intervals into the 0 to 2 portion. Because of the
nonuniformity in x,, this folding process causes the distribution of 6, to
approach that of the random array almost independently of the aperiodic
algorithm from which the x, are derived. Only the close sidelobes
therefore, would be expected to approximately follow the radiation
pattern of the underlying distribution. Since the distribution of phases at a
large scan angle tends towards a random distribution, the distant si eo
properties might be expected to be similar to those of a random array
The properties of the random array are discussed in the next chapter.
There it is shown that the average sidelobe level is 1/N relative to the
main lobe. The theoretical average sidelobe level of a 15-element random,
array, therefore, would be ~12dB. Examination of the right portion o¢
Figure 85 shows the average to be in that neighborhood. Furthermore the
theory of the peak sidelobe of the random array described in Chapter g
gives the expected value of the peak sidelobe of a 15-element array of
length 25A as —4.5 dB. Note that the peak sidelobe seen in Figure 5 j.
essentially this value. These facts suggest that the penalty for spatial tape,
is a degradation of the sidelobe properties toward those of a random
array.
Based on these observations the radiation pattern of Figure 85 may b
broken into three regions. The first region is called the controlled region,”
Here the radiation pattern exhibits an approximation to the desi, a.
properties, The next region is a transition region. in 1 which the sidelgn>
pattern degenerates from the design characteristics toward that of qp©
random array. The third portion may be called the st legion we
which the sidelobe
array, 132 Aperiodic Arrays
7.4 COMPARISON WITH THE RANDOM ARRAY
It would be desirable if there existed a theory governing the design of
aperiodic arrays. Unfortunately there is none. With the exception of the
random array, aperiodic arrays are designed ad hoc. The objective of any
aperiodic, algorithmic design is to reduce costs by thinning without
sacrificing the width of the main lobe or suffering unduly large sidelobes
or grating lobes. Based on the discussion in Chapter 3, it is evident that
the width of the main lobe is dominated by the size of the aperture in units
of wavelengths; therefore, beamwidth is easily preserved in a large array
notwithstanding extensive thinning. The two-element interferometer de-
scribed in Section 4.1 is an extreme example. Grating lobes are eliminated
by aperiodic design. Control of the sidelobes, on the other hand, is
seriously jeopardized by the reduction in the number of elements.
The design techniques for the aperiodic arrays referred to earlier[3-10]
all attempted to reduce the large sidelobes. The success has been quite
limited, however. A thinned array, because of its reduced number of
elements, offers the designer fewer degrees of design freedom than a filled
array. This inevitably results in less control of the radiation pattern. The
normalized main lobe, as noted above, remains relatively unaffected; its
first-order property is its width, which is determined by the number of
wavelengths across the array. The loss in design control is reflected in the
side radiation pattern.
Although the object of algorithmic design is the retention of some
control in the sidelobe region, many of the published patterns of the
algorithmically designed arrays found in the papers already cited exhibit
properties similar in character and level to random arrays having similar
design parameters. A comparison between the patterns in Figures 85 and
89 discloses such similarities. Hence it is questionable whether the
sought-for control in the sidelobe region exists to a significant extent.
It is of interest to compare the relative degree of sidelobe control by
algorithmic aperiodic design and random aperiodic design. But how is
such a comparison to be made in the absence of a design theory of the
algorithmic, aperiodic array? An attempt has been made based upon a
statistical comparison between a large group of algorithmic, aperiodic
arrays and a group of random arrays. The measure used for the
+The theory of the random array is developed in Chapter 8, from which the following facts
are taken. In a random array the location of the nth element is a random variable drawn from
a population described by a probability density function (pdf) w(x). Tis far-field radiation
pattern, when the elements are equally excited and isotropic, is f(u) = 5°, exp (jkx,u). The
ensemble average pattern is determined solely by w(x); Ef(u) © fu) = F{wix)} is the
Fourier transform of w(x). The average power pattern, however, has an important additive ‘Comparison With the Random Array iss
comparison was the peak sidelobe[12]. The numerical results of the
study, which is described below, indicate little difference.
‘A data base of 170 random arrays with various parameters was created
by computer, the antenna patterns were calculated, and the peak sidelobe
of each was measured; 70 aperiodic arrays designed by a variety of
algorithms were taken from the literature.t The peak sidelobe of each was
measured and its pertinent parameters tabulated. After suitable normal-
ization of the peak sidelobes, comparison was then made of their two
distributions.
In Chapter 8 it is shown that the peak sidelobe (PSL) normalized to the
number of elements N varies approximately as the logarithm of an array
parameter n. (For the idealized random array of isotropic radiators, with
monochromatic radiation and neither amplitude nor spatial taper, n =
(L/A)(. +|sin 60) * no. Amplitude or spatial taper decreases n in direct
proportion to the beam broadening. Nonisotropic elements and wide
bandwidth radiation also decrease n.] An experimental curve of the
expected value of PSL/N B, was obtained from the 170 random arrays.
To obtain this curve the random arrays were grouped according to n and
their measured peak sidelobes, normalized to N, were averaged. Eight
such average values were calculated. These are the points shown on the
smooth curve labeled E(B,) in Figure 86. The curve drawn through these
points is taken to be the expected value of the distribution of B, as a
function of n, that is, E[B,(n)].
The sources of the algorithmic designs are listed in Table 8. Descrip-
tions of the procedures by which the patterns were formed are given in
the references. From the published parameters of each algorithmic
aperiodic array, that is, N, L, A, taper, beamwidth, element distribution,
and so on, a value of n for a random array having the same physical
resources was calculated or estimated. The peak sidelobe data for the
algorithmic arrays could then be plotted on the same figure. Figure 86
shows the results. Each point is located by a letter that identifies the
term; Ef(u)f*(u) = (1— IN) fou) f4(u) + LIN is the power pattern associated with w(x)
plus an angle independent constant. The latter term is the variance of a random, noiselike
sidelobe pattern having the statistics, of the envelope of handlimited random noise. Near the
beam-steering angle f, is large and dominates the pattern. However beyond that angle at
which f. drops to the order of 1/N, which normally is only a few beamwidths, the element
distribution has no effect and the random character of the sidelobe pattern dominates.
+There is an implicit assumption that these 70 arrays are somehow representative of a
population of “good” arrays or “good” design techniques. The assumption is completely
unjustified in any theoretical sense. However there is some subjective validity to the
assumption because these array designs, which were taken from the literature, were created
by bright people and it is well known that one rarely publishes one's failures.
1 ble htt)
t+ @ lv) 134
ton *
red
An SM a ft mean for 170
ML
BG random arrays
Ee NG m
LOM
cea tH
GBB mH
mH
H
10 100 1000
Array parameter n
Figure 86 Peak sidelobe normalized to number of elements. From Steinberg, IEEE Trans.
Antennas Propag., AP-21 (May 1973), 366-370.
source (Table 8). The ordinate of each point is the measured value of the
peak sidelobe divided by N. The abscissa is the calculated value of n. The
difference in decibels between an experimental point and the curve is the
random variable examined in the study, that is, 10 log B,/E(B,). It is the
ratio, expressed in decibels, of the normalized peak sidelobe of a sample
array to the expected value for the random array having the same array
parameter n.
Figure 87 shows the cumulative distribution functions (cdf) of the
measured values of 10 log B, /E(B,) for the 170 random arrays and the 70
nonrandom arrays. The data are plotted on a Gaussian probability graph.
The approximately straight-line character of each set of data is suggestive
of a log normal pdf for the ratio. The mean for the random arrays is zero
because E(B,) was derived from its data; there is no further significance
to this fact. The experimental median is exceedingly close to zero, which
is to be expected from a symmetrical distribution such as the log normal.
The mean and median of the algorithmic arrays also are within a few
tenths of a decibel of zero; hence, the mean (and median) algorithmic mertnat
skeuy
(91 1E961 MUIOXS 7F UeULIDYS
lelos61 “sewouys ay ‘pieyoeg “Bury
[slz961 Hoyen.
WAN “LU]OXS
l9lz961 “Ant,
[pLever “oui,
(el s961 “wou 2 naeWIYs]
letle961 ‘1kog
Paystandun “2961 ‘so1uonY pe10UaH,
[st]P961 “unwuoyg 2p “10:
skeue Bun UL W
(sautsos payfonuos ‘sajdnnw jo uoneunua
“uossaidod onowue ‘sawtid “oj) snouw A
Jo soquinn,
aaunog,
Saimwaind onwapenb “Bulseds wsojlunuON — ¢
BupwwesTosd weUdp ‘Ava uy paseds Ajpenboup Fy
Wysiom JO[ABL “paouds K|uostUNUOU “esse AeIND1
sAuue Buu poseds Ajuojiun jo Kwuue syjnou gq
onsiumaaq 3
sade) Aysuap vai jenby
jade, Ausuap Fo] [sg oy
pouray
sdtaay dpoyiady g aqui,
138 aS Aperiodie Arra
Probabi
2g
os
é
of
oe
Random
Aperiodic ~—0.08
Ee (48)
Figure 87 Cumulative distribution functions of peak sidelobes of 170 random
aperiodic arrays. From Steinberg, IEEE Trans. Antennas Propag., AP-21
366-370.
design (as represented by 70 arrays created by 13 different a
than randomization of element lo
The single most successful procedure of the several e)
dynamic programming. It is not a purely algorithmic design
ta
co-workers describe it as a trial-and-error procedure[15}. A
highly constrained approach, better considered as a quasi ‘
method of sidelobe control. The first element is located al
second location is that which gives the best combinat
location is that which gives the best trio based on the fixed |
first two elements, and so on. The mean, median, and star
of 10 log B,/E(B,) for the nine arrays in this group are
1.9 dB, respectively. ys
array
M:
gor
“atic
am
. Sk
tue
tial
rar
oca
\dar
vor 138
5.
10.
16.
Aperiodie Arrays
Maffet, A. L., “Array Factors with Nonuniform Spacing Parameters,” IRE Trans.
Antennas Propag., AP-10 (March 1962), 131-136.
Willey, R. E., “Space Tapering of Linear and Planar Arrays,” IRE Trans. Antennas
Propag., AP-10 (July 1962), 1369-1377.
Ishimaru, A., “Theory of Unequally-Spaced Arrays,” IRE Trans. Antennas Propag.,
AP-10 (November 1962), 691-702.
Skolnik, M. I. and J. W. Sherman, “Planar Arrays with Unequally Spaced Elements,”
Radio Electron. Eng., 28, 3 (September 1964).
Ishimaru, A. and Y. S. Chen, “Thinning and Broadbanding Antenna Arrays by Unequal
Spacing,” IEEE Trans. Antennas Propag., AP-13 (January 1965), 34—42
‘Skolnik, M. I.,“Nonuniform Arrays,” Chapter 6 in Antenna Theory, Part I, Collin and
Zucker (eds.), McGraw-Hill, New York, 1969.
Lo, Y. T. and S. W. Lee, “A Study of Space-Tapered Arrays,” IEEE Trans. Antennas
Propag., AP-14 (January 1966), 22-30.
Steinberg, B. D., “Comparison between the Peak Sidelobe of the Random Array and
Algorithmically Designed Aperiodic Arrays,” IEEE Trans. Antennas Propag., AP-21
(May 1973), 366-370
Doyle, W., “On Approximating Linear Array Factors,” Rand Report RM-3530-PR
(February 1963).
Tighe, R. F., Non-Uniform Two-Dimensional Scanning Arrays, IEEE WESCON
Convention Record, 1963.
Skolnik, M. I., G. Nembhauser, and J. W. Sherman, “Dynamic Programming Applied to
ee Spaced Arrays,” IEEE Trans. Antennas Propag., AP-12 (January 1964),
5-43,
Sherman, J. W. and M. I. Skolnik, Thinning Planar Array Antennas with Ring Arrays,
1963 International Convention Record, Part 1 Theory of the
Random Array
8
8.1 BACKGROUND
he random array is one form of aperiodic array. It_is_a_thinned
array (mean interelement spacing greater than A/2) and therefore is
Jess costly in components than a conventional phased array. Grating lobes
are not present because there are no periodicities in the element locations.
Unlike the aperiodic arrays discussed in the last chapter, however, the
tandom array is nondeterministic. Curiously enough, while no_unified
theory of the determi: aperiodic array exists, sucha theory has been
developed for the random array. This is because the mathematics of
: rocesses, that is, statistics, can-be-applie
“Two forms of nondeterministic aperiodic arrays are treated in the
literature, the random array[1-9] and the statistical array [10-13]. The
former is characterized by element locations chosen by some random
Process; in the latter a conventional array is designed and a given fraction
139
fla 140 Theory of the Random Array
s 7 i ce
of the elements is removed at random, Although different design ra
dures are involved, the radiation patterns of the two forms of arrays en
the same statistical Properties; hence the term random is applied to
‘ . f° at
forms in this chapter. 1 of phase shifters is assumed th
cophases the elements at the beam-ste
ering angle, sa ost
The earliest major work Bertinent to the random array was in. 18528
John Ruze, who did not study the random array per se; instead og
examined the effects of random amplitude and phase errors in a aperture
on the radiation pattern of the aperture (Id). Although Ruse was n0'
treating the ray his mathemati; i
ndom array.
mean pattern. Y. T. Lo
Panicali and v Agrawal) followed shortly thereafter,
the random array
» Concentrating on
- This group has be
€n_the dominant contributor to the
subject. The first major work w he same year by i
sequel [2]. These papers OQ
Average Sidelobe Level 14
N
N
t { { { jt) = 3 5 — 8)
a
flO) = Main lobe N Elements
o_o > Cophased at » = 0
N Elements
Sidelobe N Randomly Phased
Figure 88 Random phased array
tandom variables having a first probability density distribution w,(x).t It
is assumed that all elements, irrespective of their locations, are properly
Phased so as to form a main lobe of maximum strength at 6s, (In the
mathematical formulation that follows this is equivalent to redefining the
Teduced angular variable u as u = sin @ — sin 60. Thus the angular variable
includes the beam-pointing direction.+) The current density is the sum of
8 functions at the locations x, on the line segment of length I.
x
i(x) =D 8(x xn) (8.1)
u) is the Fourier transform of
f(u) is proportional to the sum
k =2r/A. The array factor is
ce complex far-field radiation pattern {(
a Because i(x) is a set of 5 functions,
ef unit vectors having phase angles aa
len written
fu) = > exp (ikxntt) (8.2)
probability density function from the
+
The subscript is used here to distinguish the first
used later in this section. Where no
oe, or joint, probability density function >
PRSCTIPL is used, wis implied.
oe beam-ateering information is temporarily included in this formulation of the reduced
aan while some of the statistical properties of the radiation pattern that are
Rae lent of pointing direction are developed. Later, when the pointing direction has an
on the radiation properties, the original definition is reintroduced. 142 Theory of the Random Array
The main-lobe amplitude is N, occurring at u =0 independent of the
random locations x,. Qutside the neighborhood of the main lobe, how-
ever, the phase angle kx,u is a random variable because the element
location x, is a random variable. Hence the unit vectors combine with
random phases. The RMS amplitude gri VN and its square as N,
the latter being the mean of the power pattern (see Section 8.4). Thus the
power ratio of the average sidelobe to the main lobe is N/N? = 1/N. This
is an important theorem.
8.3 AVERAGE ARRAY FACTOR
Another first-order quantity of interest is the average array factor, that is,
the ensemble average of the complex radiation patterns resulting from
many selections of N-element locations. Letting the overbar represent the
ensemble average and temporarily normalizing (8.2) to 1/N yields
Saree
Fu) = D exp (hers) (8.3)
Equation (8.3) also equals
Fw) (8.4)
1 —
ND OP ke)
since the average of a sum equals the sum of its averages. Each term has
the same average value because the random variables come from the
same population. Thus
See y
Fa)- eB =o (85)
Equation (8.5) is recognized as the characteristic function of x. It is the
expected value of exp jkxu) and is related to the pdf ws) tt
a f wrayer ax (8.6)
et the pdf wi(x) be chosen so that the ensemble average of f(t) equals
some desired array factor fu(u), which, in turn, is related to come
continuous current density io(x) by the far-field transform relation
folu) = f.. jo(x)e™ dx 7)
+Equation (8.6) follows from the theorem that the ensemble ‘Average of a function of a
is the integral of that function weighted by the pdf of the random variable
random vari Average Power Pattern 143
1 gives the interesting relation
Saray bl) dh
implies f(u) = fo(u) (8.8)
This equation indicates that the average complex radiation pattern can be
matched to any radiation pattern that is derivable from a real current
density by selecting the pdf of element location according to (8.8). This
has a major implication: it suggests that the array can be drastically
thinned without altering the average pattern.
Normalizing fo(w) by setti
wilk
84 AVERAGE POWER PATTERN
In most radar and communications applications only the amplitude and
the gain properties of the far-field radiation pattern are of interest.
Consequently another first-order quantity is the amplitude of the array
factor, or its square, which is the power pattern. The power pattern of an
array is the product of the radiation pattern f(u) and its complex
conjugate. By representing the conjugate of a complex quantity by ( )*,
the ensemble average or the expected power pattern may be written
Fp > Dew (jk (Xn — Xm JU)
f(u)}
= Ld DS exp Gk — mn)
NI &
= RIN + ENT NI) (8.9)
The second line is based on the theorem that the average of a sum equals
the sum of the averages. In the third line the first quantity is the N terms
for_n=m. The second quantity has N*—N_terms for_nz m. The
coefficient of the second term is f f* [by (8.5)] and also equals fo(u )f8(u),
which is the desired power pattern, by virtue of (8.8). The mean power
pattern, therefore, may be written
Fea = ppt + fawyfBuyn?—NY1
e(;-t)y
= |fou)| (1 w) + (8.10)
Equation (8.10) is an important result. Whereas the mean complex
radiation pattern is the unaltered design pattern fo(u), the mean power
pattern, suffers change. It consists of two terms. The first term is the ‘Theory of the Random A
ired power
Pattern, slightly reduced im strength to account for
ond term. The seco:
nd term is an additive, angle-independent ter
rength 1/N, the i
. es the shape of the raqi
only in the Reighborhood of the beam-pointing direction or
bulk of the radiation Pattern, the random Properties that dev
Statistics of the sum of randomly phased Vectors prevail,
OF Me-FY =
8.8 VARIANCE one SAF eM 3
} SAUCY aay
The variance of the array tactor ig ThE
pace
= Fe
which evaluates to
Ou) = iF FF
=\tr \
“WOE ip
= HO \htuyey
‘The variance goes to zero as u +0; this May have
the stipulation in Section 8.2 that a elements are. p
a beam at 60. For large angles o* = AIN, as was
8.6 SUMMARY OF THE AVERAGE QUANTITHE
‘These averages may be summarized as follow
A. The ensemble average array factor equals
arene the
M of
Meay
S th;
Ke
e le
the
tin
OVE
side)
cha
Pat
pal
se
ion
oug
lop
deer
Ope
dbs
s
he 144 ‘Theory of the Random Array
desired power pattern, slightly reduced in strength to account for the
second term. The second term is an additive, angle-independent term of
strength 1/N, the quantity found earlier to be the power ratio of the mean
sidelobe level to the main lobe. The presence of this pedestal implies that
the designer only has limited control of the sidelobe region. Unlike the
radiation patterns of most conventional arrays, the mean sidelobe level
does not decrease monotonically with angle; on the contrary, the ex-
pected value of the sidelobe level is the same everywhere except in the
neighborhood of the main lobe; there |fo|’~ 1 and rises well above the
pedestal 1/N. The shape of the main lobe and of those nearby sidelobes
whose strengths are much greater than 1/N are essentially unchanged.
Away from the main lobe where the sidelobe level of the design pattern is
less than or nearly as great as 1/N, the character of the design pattern is
submerged below the random component and disappears. Consequently
the pdf of element location influences the shape of the radiation pattern
only in the neighborhood of the beam-pointing direction. Throughout the
bulk of the radiation pattern, the random properties that develop from the
statistics of the sum of randomly phased vectors prevail. avyobea
F)e-F ke mee
Pety = nothicg
= A fer F .
8.5. VARIANCE fare - f a . fi f an
Fe EP et FF
The variance of the array factor is mE spe
=F Ao} or of the equivalent form Pr{A < Ac}, where Ao is
some arbitrary sidelobe level, is desired. This is easily obtained from
(8.19) in the following manner.
8.8 PEAK SIDELOBE ESTIMATOR B
Equation (8.19) means that the probability that a sample of the sidelobe
pattern, taken at an arbitrary angle u = sin @—sin # away from the
neighborhood of the main lobe, falls between Ao and Ay +A is
Pr{Ao< A =Ao+ dA} = J wi(A) dA (8.20)
If the upper limit in (8.20) is made infinite, the integral*
en J w(A) dA =e“ (8.21)
is the probability that an arbitrary sample of the radiation pattern away
from the region of the main lobe exceeds some value Ao. Let the value Ao
be identified with the sidelobe tolerance; for example, let Ao represent the
sidelobe specification of an array design problem. The complement of
(8.21), 1—a, is the probability that such a sample is less than Ao. The
Statistics of a single sample of the side radiation pattern is not very
interesting. One is interested in a probability statement of this type
tegarding the entire sidelobe region. Three steps are required to obtain
such a statement. First, the side radiation is conceptually sampled n times
at the minimum interval in u that assures that the samples are indepen-
+The remainder of this chapter is drawn largely from [9] Peak Sidelobe Estimator B 151
dent. The probability is then calculated that none of the n independent
samples exceeds Ao. Second, the number of independent samples and/or
the minimum sampling interval for independence are calculated. Third, a
correction is calculated to account for the high probability that the largest
sample falls somewhat below the crest of the largest sidelobe.
The first calculation involves n independent samples of the sidelobe I A
pattern. If n such samples are taken, =[z]-exp(-B.
’ Py 3 [+]-exp'
B=(1-a)" =|1-exp mane J (8.22)
[ ) BP = |_expl-B)
is the probability that none exceeds Ao. Here Ao’ is the power level of the , fA
tolerance, or threshold, level and N is the theoretical average sidelobe (/? ‘)
level. Therefore the quantity Ao’/N is the power ratio of the threshold = = «el
level to the theoretical average sidelobe level. Since this ratio reappears
often in the theory, it is convenient to give it a new symbol, B. Solving for /— "=
Ac/N = B yields LA
B=-In(i-B"") 2a OF
In (8.23) B is the probability that none of the n independent sidelobe
samples exceeds B. Since is a confidence level, 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. Because n must introduce all the
relevant array properties other than N into the equation for B, n is called
the array parameter.
The second of the three tasks is to calculate nm, the number of
independent samples of the sidelobe power pattern required to specify the
pattern completely in the visible region. The number may be calculated in
several ways. One method is by the application of the Nyquist sampling
theorem. That theorem, usually applied in the theory of bandlimited
random processes, asserts that the number of samples required to specify
a wave that is limited in bandwidth to W (Hz) and in time duration to T
(seconds) in 2WT [18], which alternately may be expressed as a minimum
sampling rate, 2W. The complex radiation pattern of a random array is
such a “bandlimited” function, the “limit” being due to the finite length of
the array. The statistical similarity is evident in Figure 91, which shows a
pattern of an aperiodic array and a photograph of bandlimited noise
approximately to the same scales. The mathematical comparison is
evident from the Fourier transforms relating signal spectrum F(f) and
time function f(t), and the transforms relating current density i(x/A) and
far-field complex radiation pattern f(u): “Eh-SE (961 ArenUUT) TI-AV “Sodorg souuaUY “suD4y ZAI “ueUUDYS pur “sosNEYWEN, “yLUjO¥S Wor
enboun juawe[9-sz Jo waned (P) “asiou WopUL paywH|pULG puL Wayed Aeue sIpoLade Jo uosuedWO 16 aanBLy
oy ee of ve ce of st 92 vz gi vt zt or 80 90 vo zoo
a ee ToT 90
: 7 {voneziundo Jo uorbey er
L | | \ \ im
| y A
{ | \{ zo
fro =
epee
+ so
f Anawus apee
F joaun so
i= boii 7 [tas [foe ea jeetesl nl ot
152 (S61) 4vpvy OF suoHDayddy YIM S094], UOHDUUOJUT PUP KIGDqoLg “pAEMpOON WOIY “AP
Partuypurg Joy LAAT = ¥ “astou paruUTps
1 (q) ‘as1ou Wopu
Paywipurg pur waned Ave o1po19%
on
153 154 Theory of the Random Array
uae fote( Go p)t vax
ao O55 i. fiy= f Fexp(2nftyay c
(8G gL 6)02)) + arene e
Y alead-t(6, 0) ceag
adel “ fy= f icryexp [i2*(2) «Ja(2) (8.24)
La
The quantities 1, f, and W may be identified with u, x/A, and L/A. Since
eet /) the minimum sampling rate in the time domain is 2W, the minimum
eu ) angular sampling rate in the domain of the reduced angular variable is
U-u /2L/A. The minimum number of samples required is 2L/A times an
‘interval U in the visible region. For an end-fire array, the interval U is the
_V==-_ entire visible region, which is of length two. When the beam is steered
wie pi elsewhere, however, it takes fewer samples to describe the power pattern
| because the power pattern is symmetrical about the beam-steering
Tbradsdel position. Thus for most situations U <2.
—“—~_|_ To show that the power pattern is symmetrical about the pointing
Veu-© > direction 6, it is useful to change the reduced angular variable once more
rvs 8 so that the steering angle again becomes explicit. As it has been used
/ throughout the text except in the last few sections, the variable is
iU[=‘L ) «=sin@ and the beam-steering position uo = sin 6. The fact of sym-
metry follows from the defining expression for the array factor of an
array steered to wo, which is
moe wh
fu) = = i, exp Likxi(u ~ uo)] (8.25)
Its power pattern is
NW
P(u)= f(u)f*(u) = = > ik exp Lik(x: — Xm )(U = Uo)] (8.26)
Now let the weights be real, which is the common design procedure, or if
‘ic “' complex let all their phases be the same. In either case ii* equals the
wr
~ product of their amplitudes which, for simplicity, is written below as inim.
jy. Next consider two directions u, and us equally spaced on opposite sides
vet" of the beam-steering direction wo. Let the spacing be Aw. Thus Au =
- Uz Wo = Wo~ Us. At Ua
yy -AV)
P(uz)= SD iin exp [K(x — Xp )AU]
uv). while at uy
“LY P(u,)= DD itin exp [ik (a —Xn)Au]
—_—
=x2
it exp [ik (Xm — x1) Au] Peak Sidelobe Estimator B 155
But the index | is indistinguishable from m. Hene
the power pattern of an array has even symmetry
The function P(u) is symmetrical about uo independent of the element
distribution w,(x) and independent of whether i(x) is even provided only
that it is real (or cophased) for all x. The length U is that portion of the
visible region needed to describe the entire visible region. The visible
region extends from —1~uo to 1— uo. Due to the symmetry discussed
above the size of U is 1+|us|. Consequently the number of independent
samples needed to specify the complex radiation pattern in this interval is
2LJA)(1 +|uol). However the power pattern requires only half this
number to specify it, for the phase of the radiation pattern contains as
much information as the amplitude. Hence the number of independent
samples in the power pattern is
n= £ (1+ |uo)) (8.27)
It varies between L/A and 2L/A. The dominant factor is the length of the
nits of wavelength. The array parameter is only secondarily
influenced by the beam-steering angle.
Another way to estimate n is from the expected number of lobes of the
radiation pattern in the interval 1+ |wo|. Since the nominal lobe width is
A/L, the number of lobes in this interval is (1+ |uo|)(A/L). From sampling
th ory an average of two samples_per lobe is required to specify the
radiation pattern; this number is 2(L/A)(1+|ud). Again, as above, only
half this number is needed to specify the amplitude of the array factor.
Thus the array parameter n is given by (8.27).
A more rigorous method* is to calculate the autocorrelation function of
the array factor and to determine from it the spacing between indepen-
dent samples. The quantity 1 + |uo| divided by this quantity is n. Consider
Ef(us)f*(us) = EY > exp [jk (xmtts — Xatt2)]
= Nob [k(t — u3)] + (N? = No. (kts) b.(— kus)
(8.28)
where E is the expected value, k = 27/A, and
$.(Z) = Ee" = 1 f e” dx
ee
sin (ZL 2)
= “Z~ER Z=ku (8.29)
+Due to Professor R. S. Berkowitz, Moore School of Electrical Engineering, University of
Pennsylvania. 156 Theory of the Random Array
when the array elements are distributed uniformly and independently
from —L/2 to L/2. Note that .(Z) is real. Hence Ef(u;)f*(uz) is real.
Similarly
Ef(us)f (ua) = ry 5) eXp [Jk (mts + Xntts)]
= Nox [k (us + u3)] + (N? = Nydyfhund. (hus) (8.30)
and is real. Represent
F(u) = a(u) + jb(u) (8.13)
Then
12+ bibs + j(—aib2 + azbi)
142— bib2+ j(aibz + azb,)
Flu)f*(us)
Fu) f (us) =
Now, since the autocorrelation functions Ef,f$ and Ef.f: are purely real,
the cross correlations Ea;b; and Eazb, are zero, and the expected values
are
Ef. f§ = Ea,a,+ Ebibz
Efif2 = Ea,a:— Eb,bz
Adding and subtracting these quantities give
Ea(us)a(u2) = s[Ef(us)f*(us) + Ef(us)f(u2)] (8.31)
Eb (ui)b (uz) = 3LEf (us)f*(ua) — Ef(us)f(u2))
Further, assuming L >A (the large-array case) and considering only
the sidelobe region (away from the main lobe) such that \u|>A/L for
both u, and u: yields
(ku), be (Hku2), be [k(ui + u2)] = 0
Hence
Ea,a,= Eb, b= N b.thk(us— us) (8.32)
Therefore a(u) and b(u) can be represented as independent, wide-sense,
stationary random processes with autocorrelation functions
NsinGaLula)
3
Lula (8.33)
Ra(u) = Rou) => N o.cku De
where u = uz. The spacing for independence between samples is that
value of u =u. for which the correlation is zero. Thus u, = A/L and
n=Za + |uol) (8.27)
exel jKamy thaued] ar
) sol Sve] an ewE[r\”) gr)
tu , areal Peak Sidelobe Estimator B 157
It is well to pause to consider the peak sidelobe theory given thus far.
What has been calculated is the sidelobe level Ao that will not be
exceeded with some given probability 8 for any uniformly distributed
random array specified in terms of the number of elements N, the size of
the array in units of wavelength L/A, and the beam-steering angle 60.
Equations (8.23) and (8.27) express the relationships between the parame-
ters. This result may be generalized to an arbitrary spatial distribution of
array elements by recognizing that the only effect of changing the pdf of
element location is in the parameter n. The fundamental expression (8.23)
does not change. On the other hand, (8.27) does change because the lobe
width alters and, therefore, the number of samples required to specify the
pattern changes. Since the array parameter n is proportional to the
reciprocal of the nominal lobe width, as discussed earlier in this section,
the array parameter for any arbitrary spatial taper is (8.27) divided by the
ratio of the beamwidth for that taper to the beamwidth of a uniformly
distributed random array of the same length. Generally speaking, if m: and
n; are array parameters for two arrays of the same length, same
wavelength, and same beam-steering angle but having different beam-
widths Au, and Aus, the relation between these quantities is
1. Aus (8.34)
nz Au
In Figure 92, B = Ao’/N is plotted as a linear ordinate and n as a
logarithmic abscissa. The confidence level 8 is the parameter of the
family. The curves are nearly straight lines in the logarithmic variable
log n. A good approximation (<10% error) for 8 =0.5 from which it is
easy to make calculations is
B=Inn~-In(1—B) (8.35)
where In is the natural logarithm. A better approximation is
B~Inn-In Ing (8.36)
taper. This is a ver
The peak-to-average sidelobe ratio calculated in this section does not
include the effects of several other design factors such 38 nonuniform
weights in the antenna elements, limitations TO the theory when the
number of elements is small, nonisotropic antenna elements, and broad-
Sand waveforms passing through the system. These factors are consi-
dered in Chapter 9. 158 ‘Theory of the Random Array
10 10? 10° 10°
L
£4 tuoi)
Figure 92 Plot of B, the power ratio of the largest of n independent samples of the array
factor for a random array to the theoretical average value, versus the array parameter n.
8.9 AVERAGE AND VARIANCE OF B
The confidence level B = Pr (no sidelobe exceeds B) is the probability
distribution function of B. Its derivative with respect to B is the pdf of B.
Thus
W(B) Bane"
aB x
(8.37)
where B, B”,..., B* are the moments of B. The first two moments are the
ones of greatest interest, for these give the average value and a measure
of the variability about the average. The average is the first moment B.
The variance is B’— B*.
Moments of BR may-be obtained from the logarithm of the characteris
tic, or moment-generating, function by the method of
Known as semiinvariants)(20). The characteristic function is g
a(S) = [ e*W(B) aB
°
=nferrane
°
yas A
Abe ke
Average and Variance of B _ 159
ao
7 ry sb
Let x =e" and dx =—e"" dB. Then —b) . @
1 @
ba(S)= J nx S(1= x)! dx
=nB(—S,n)
where
Bop. a)= fx?" 2)" de, p>0, q>0
°
Pq)
T(p +4)
is the beta function of p and q and I(-) is the gamma function.+ From the
properties of the gamma function
Tin-S+1)=(n—-S)P(n-S)
=(n—S)(n—S-1)---(1-S)TA-S ;
(n n 1d) ) ra)
T(n) =(n-1! rad
Therefore nse)
a“
oS) G=R=S—T OT. pus)b-V!
. In do(S)=Inn!— ¥ In(k-S) | EMS U5) 5)
In(k (1-3 5 wnn-i) 1
=Ialk) +) (2) =Inn!=inn!~¥ m(1-2) KS (nash (is)
--3im(1-3)
Now vo C+ (-3))
ze [()- sy +(= sy- (i
AS”
rt
Ms
In ba(S) =
where A, are the semiinvariants, A, is the expected or average value of B,
and A; is its variance. The logarithm may be expanded into its power
“series representation
2
ind'+wy=w-% 4
+See text on advanced calculus, such as (21),
in(odi-g eat) tn +(-g
mint k! C3) = lala! Ja ( E) Theory of the Random Array
Ba
Inda(S)= ($+ St
M=(r=DIS kT
a
where n is the array parameter. Note that, for large n, B ~ fi" (1/k) dk =
Inn. Comparison of this expression with (8.25) shows that B ~
B-In(1—8), which means that B differs from \its mean by a small
constant that is determined solely by the confidence level p<,
The variance of B is
le
-Sp-2 eas ne (8.39)
o
and therefore is approximately independent of the size of the array.
8.10 THE ESTIMATOR B,
Early in Section 8.8 three tasks were indicated as necessary to obtain the
desired peak sidelobe statistics. First, an estimator B of the power ratio
of the peak sidelobe to the mean sidelobe level was calculated in terms of
the number of elements N, a confidence level 8, and an array parameter n.
Next the array parameter was evaluated in terms of A/L and 60 (or
Uo = sin 60). The statistical estimator B, however, is not an unbiased
estimator. The third task is to remove this bias.
Figure 93 illustrates the problem. A few lobes of a power pattern are
drawn. Independent sampling is shown at intervals of A/L. The largest
Figure 93 A power pattern where B is a lower bound estimator of the peak sidelobe. The Estimator B, 161
lobe has maximum value B,, and the largest sample on that lobe is B, a
distance Au from the location of the peak. The probability is zero that Au
exactly equals zero. Hence B < B, with a probability of unity; that is, it is
a downward biased estimator.
A better estimator is obtained by adding to B the expected value of AB.
The problem is to determine the increment from a threshold value to the
local maximum of the squared magnitude of f(u). For this purpose it is
convenient to represent the array factor f(u) once again as a sum of
quadrature components a(u) and b(u):
f(u) = a(u) + jb(u) (8.13)
The power pattern is the sum of the squares of the components.
P(u)= f(w)f*(u) = a*(u) + b*(u) (8.40)
Consider a random sample point u =u, close to a lobe crest. In the
neighborhood of u:, P(u) can be represented by the Taylor expansion
P(u) = PC) + Pusu = ws) +3P"(U\u— wy +--+ (B41)
where P’(u,) is the first derivative with respect to u of P(u =u), and so
forth. For sufficiently small distances between the sample point and the
crest the first three terms of the series suffice, and the lobe becomes
described by the parabola
P(u) = P(us) + P'(u)(u — us) +4P "(uu =u) (8.42)
This representation is satisfactory because the sample spacing is A/L,
which equals the nominal lobe width; hence, the mean distance of the
largest sample of a lobe from the crest is about A/4L. The location of the
crest u =u, is found by setting the derivative of (8.42) equal to zero.
aP
du
P'+P"(u, — us)
Pp’
Pp (8.43)
Uy — th =
where P’ = P'(u,), and so forth. Inserting (8.43) into (8.42) and subtract-
ing P(u,) gives the difference AP between the peak value P(u,) and the
sample value P(u,). 07
sre -
e
2
¢
(8.44)
To evaluate (8.44) it is helpful to express the derivatives of P in terms
of the amplitudes and first and second derivatives of the quadrature 162 ‘Theory of the Random Array
components. From (8.40)
P' =2(aa' + bb') (8.45)
aldara rhb”)
co P" = 2(aa"+a"*+ bb" +b”) (8.46)
where a = a(u,), and so on. Inserting (8.45) and (8.46) into (8.44) gives
a?q’? +2aa’ bb’ + b*b”
AP ga" a+ bb +b aed
where a and b are random Gaussian variables, as noted in Section 8.7,
and are independent and stationary, as shown in Section 8.8. The
probability density functions of each of the six variables in (8.47) are
known (e.g., Chapter 3 of Ref. [22]). The quadrature components a and b
are uncorrelated with their first derivatives, that is, aa’ = bb’ = 0, but are
highly correlated with their second derivatives. Equation (3.16) of [22]
shows that for large thresholds the pdf of a" (or b”) is a narrowly peaked
Gaussian distribution. Consequently a” (or b") can be regarded as
nonrandom and equal to its conditional mean —R,a (or —R2b). The
coefficient is given by
_ 1 @ Rw)
RRO du? lane ay
where R(u) is the autocorrelation function (8.34) of the quadrature
components calculated in Section 8.8. Thus aa" = — Ra? and aa" + bb" =
~R2(a* + b*)=—RzP. The other terms in the denominator are a’? and b”,
Their pdf's are exponential (calculated from (3.14) of [22])
aye _ a?
wa = ner ( Am) (8.49)
and therefore are not narrowly peaked. However for P > N the contribu-
tions of these terms are not large, so that each may be represented in the
denominator by its average value. From (8.49) a = b” = NR>. Hence the
denominator of (8.49) may be approximated by the nonrandom quantity
R:P CF) (8.50)
The average increment, then, is approximately the average of the
numerator of (8.47) divided by (8.50). The average of the middle term in
the numerator of (8.47) is zero, as observed above. The remaining
numerator terms average to
NR.(a*+b*)= NR;P (8.51)
—R2P +2NR.= The Estimator B,
and the average of (8.47) is approximately
NR2P N
AP=~—Rpd INP) 1-2NIP
where it is understood that P = P(u,).
‘As an example of the theory as developed so far, consider an array
design in which L = 100A, 6 =0, and N = 50. Let the desired confidence
level be 6 = 0.9. From Figure 92, B is found to equal 7, and P = BN =
350. The average increment AP is, from (8.52), 50/1— (100/350) = 70.
Hence, the peak sidelobe does not exceed 420 with probability 0.9. The
main lobe strength is N* = 2500. Therefore the relative peak sidelobe is
420/2500 = —7.5 dB. That is, the probability is 0.9 that the largest sidelobe
does not exceed —7.5 dB with respect to the main lobe.
Another interesting example is the median peak sidelobe. The median
peak sidelobe is the peak sidelobe of the Sth percentile of a large group
of random arrays having the same design parameters L/A, n, and 60. It is
represented by the B = 0.5 curve of Figure 92. For this value of 6 and the
design parameters used in the example above, B is found to equal 5.
Therefore, P = BN = 250, AP = 50/1 — (100/250) = 83, and the median
peak sidelobe to main lobe ratio is 333/2500 = —8.7 dB.
The difference between the peak sidelobe at the 50% confidence level
and the 90% confidence level is only 1.2 dB. The small difference between
the 50th and the 90th percentiles implies a highly compact t distribution,
which could have been anticipated either from the experimental variance
data of Section 7.4 or the variance calculation in Section 8.9. For example,
in the latter the variance of B was shown to be 7/6. The standard
deviation is 7/6 = 1.3. The difference between B for 8 = 0.9 and B for
B = 0.5 is 7—5 = 2. This is 1.5 standard deviations, a distance from center
that typically includes the bulk of most distributions.
The use of the theory can be somewhat simplified by redrawing Figure
92 to include the average increment after suitable normalization. Equation
(8.52) is approximately the expected value of the difference between the
height of the lobe crest and the sample value P(u:). Normalizing (8.52) to
N gives
1 2
7-778! a
When B is 5 or larger, the asymptotic assumptions prove [by comparison
with experimental data (see Figure 95)] to be valid. At small values of B,
however, (8.53) grows large. It is one-half B at B =4, it equals B at
B =3, and it blows up at B =2. In this region it is not a satisfactory
increment to B. The difficulty can be avoided without affecting the quality
AB ) tees (8.53)
lL = M Vv pa?
)
a oe Vv
SRT ju1w~ BM=2ZM ~ BN-2N
BW 164 ‘Theory of the Random Array
of the estimator at large B by truncating the series expansion of (8.53) at
two terms. This removes the singularity at B = 2. The increment then
reduces to
2
1 +e (8.54)
and the estimator of the peak becomes
2
Be=Btl+E (8.55)
when B is large. This equation is plotted in Figure 94 for B >3. The
independent variable is the array parameter n =(L/A)(1+|sin 6). The
parameter of the family is 8, the confidence level.
b= 0.999
099
12
"
10 ne
07
7 05
G 03
e7
6
5
4
10 100 7000
Array parameter = (L/N)(1 + | sin Uo |)
Figure94__ Probabilistic estimator of peak sidelobe of a random array. From Steinberg, IEEE
Trans. Antennas Propag., AP-20 (March 1972),
Equation (8.55) has been tested by computer simulation of random
arrays[9]. Figure 95 shows the results for 4) = 0 and eight different values
of n ranging from 7 to 700. Ten random arrays were simulated for the
largest value (n = 700). For the remaining seven groups, 20 arrays each
were formed. Of the 150 arrays, 70 had 30 elements and 80 had 100 The Estimator B, 165
Jo =0
20 Arrays
10 Arrays
10 100 1000
Array parameter n = (L/A}(1 + | sin Dol)
Figure 95 Each sample point is the indicated percentile of the experimental distribution of
the peak sidelobe of a group of linear random arrays. The solid curves are from Figure 94.
From Steinberg, [EEE Trans. Antennas Propag., AP-20 (March 1972).
elements. The array factor was calculated for each array, and its peak
sidelobe found and measured. For each of the eight groups the experi-
mental distribution of the peak sidelobe was determined, from which the
percentile data of Figure 95 were obtained. Also plotted are theoretical
curves from Figure 94 for corresponding values of B.
It is evident that the estimator closely matches the data for B, as small
as 5 (or B ~3), which implies that the estimator is a useful solution even
for small arrays. The smallest broadside array (@ =) for which itisa
satisfactory estimator is 2A for 6 = 0.9 and 14A for B =0.5. Figure 96
shows the relationship for other values of B.
The normalized quantity B, is the ratio of the unnormalized peak
sidelobe to the average sidelobe level. For design purposes the ratio of the
peak sidelobe to the main lobe is a more useful quantity. The ratio B, is
converted to this quantity through the use of the theorem of Section 8.2;
that is, the power ratio of the average sidelobe to the main lobe is 1/N.
Combining these factors gives the desired power ratio:
Peak sidelobe
Main lobe
N (8.56) 166 ‘Theory of the Random Array
24 T_T
16
0 05 1.0
Figure 96 Minimum array length for which B, is a satisfactory estimator of the peak
sidelobe,
8.11 SUMMARY OF ASYMPTOTIC, ISOTROPIC THEORY
By defining
B,
peak sidelobe of power pattern of a random array of isotropic radiators
theoretical average sidelobe power level
the results of this chapter may be summarized:
2
B,=B+1+5, Be3 (8.55)
where |
-In(1-B"") (8.23)
=Inn-In(I-B), p=0s (8.35)
N is the number of elements in the array, equal to the theoretical average
sidelobe power level; B is the confidence level, that is, the probability that Problems 167
no sidelobe exceeds the theoretical average level N by the factor B, ; and
n is the number of independent samples in the sidelobe pattern of a
random array
n= E+ hud) (6.27)
for a linear array with uniform pdf of element location.
The power ratio of the peak sidelobe to the main lobe is B,/N. This
ratio may be written
PSL_B+1+2/B
ML ONT (8.56)
PROBLEMS
8.1. A linear random array has 30 elements that are distributed accord-
ing to a uniform pdf over the aperture.
(a) What is the average sidelobe level outside the immediate
neighborhood of the main lobe?
(b) What is the average sidelobe level if the pdf is changed to a
raised cosine distribution?
8.2. A 100ft aperture is available for the construction of an L-band
(1 ft wavelength) array. The desired beamwidth is 12.5 mrad, and
the average sidelobe level must not exceed —17 dB with respect to
the main lobe.
(a) Specify the minimum number of elements required to
achieve the sidelobe specification.
(b) Recommend a pdf of element location that satisfies the
beamwidth requirement.
(c) How many elements would be required to fill the array (A/2
element spacing)?
(d) Calculate the thinning factor (ratio of number of elements in
the aperiodic array to number in the filled array).
8.3. It is desired to build a random linear array having a parabolic pdf,
20 dB average sidelobes, and a second-moment-of-the-aperture
beamwidth (Au.) of 10 mrad (107 rad). The wavelength is 3.2 cm
(X-band).
(a) Find the length.
(b) Find the number of elements.
(c) Also find the number of elements required to construct a
filled array (spacing = A/2) of the same size. 8.4.
8.5.
8.6.
8.7.
8.8.
8.9.
‘Theory of the Random Array
(d) What thinning factor results from the random design?
The side radiation pattern (u > A/L) of arandom, linear array of N
isotropic, equally excited elements is randomly sampled. What is
the probability that a random sample
(a) equals or exceeds the average sidelobe level?
(b) is 34B or more larger than the ASL?
(c)_ is 6dB or more larger than the ASL?
(d) is 34B or more smaller than the ASL?
Find B, for a random, linear array of 100 wavelengths at a level of
confidence B =0.5, 0.7, and 0.9. The steering angle is 0°.
What is the ratio of PSL to ASL for a 40-element linear random
array phased to point to —45°? The length is 200 ft, the wavelength
is 18in., and the desired level of confidence is 80%.
Given a 30-element linear random array 70 wavelengths in length,
what is its peak sidelobe at the 70% confidence level? The steering
angle is 30°.
What is the ratio of peak to average sidelobe of a 50-element linear
random array of 400 wavelengths? The beam-steering angle is 0°.
The answer is to be given at a confidence level of 90%. Also find
the PSL.
A random linear array of 300 wavelengths length has 100 elements.
Its radiation pattern, which is steered to broadside, is sampled
randomly.
(a) What are the probabilities that none of n independent
samples (n = array parameter) exceeds —13 dB, —11.5 dB,
and —10.5 dB?
(b) What are the probabili
exceed these values?
ies that the peak sidelobe does not
8.10. A uniform distribution of 1000 randomly (but collinearly) placed
X-band (3.2 cm) elements results in an expected value of PSL =
—20 dB relative to the main lobe when the array is steered to 30°.
Find the length of the array.
REFERENCES
|. Lo, Y. T., “A Mathematical Theory of Antenna Arrays with Randomly Spaced
Elements,”
IRE Trans. Antennas Propag., AP-12 (May 1964), 257-268.
. Lo, Y. T., “A Probabilistic Approach to the Problem of Large Antenna Arrays,” Radio
Sci. 68D (September 1964), 1011-1019.
Lo, Y. T. and S. W. Lee, “Sidelobe Level of Nonuniformly Spaced Antenna Arrays,”
IEEE Trans. Antennas Propag., AP-13 (September 1965), 817-818. References 169
10.
u
12.
1B.
14,
15.
16.
Te
18.
19.
20.
21
22,
Lo, Y. T. and S. W. Lee, “A Study of Space-Tapered Arrays,” IEEE Trans. Antennas
Propag., AP-14 (January 1966), 22-30.
Lo, ¥. T. and R. J. Simcoe, “An Experiment on Antenna Arrays with Randomly
Spaced Elements,” IEEE Trans. Antennas Propag., AP-15 (March 1967), 231-235.
Lo, Y. T. and V. D. Agrawal, “A Method for Removing Blindness in Phased Arrays,”
Proc. IEEE, 56 (September 1958), 1586-1588.
Panicali, A. R. and Y. T. Lo, “A Probabilistic Approach to Large Circular and
Spherical Arrays,” IEEE Trans. Antennas Propag., AP-17 (July 1969), 514-522.
Agrawal, V. D. and Y. T. Lo, “Distribution of Sidelobe Level in Random Arrays,”
Proc. IEEE, $7 (October 1969), 1764-1765.
Steinberg, B. D., “The Peak Sidelobe of the Phased Array Having Randomly Located
Elements,” [EEE Trans. Antennas Propag., AP-20 (March 1972).
Allen, John L., “Some Extensions of the Theory of Random Error Effects on Array
Patterns,” Chapter III, Part 3 in Phased Array Radar Studies, Lincoln Lab Report 236
(November 1961).
Maher, T. M. and D. K. Cheng, “Random Removal of Radiators from Large Linear
Arrays," IEEE Trans. Antennas Propag., AP-11 (March 1963), 106-112.
Lo, Y. T., “Random Periodic Arrays,” Radio Sci., 3 (May 1968), 425-436.
Skolnik, M. I., “‘Nonuniform Arrays,” Chapter 6 in Antenna Theory, Part I, Collin and
Zucker (eds.), McGraw-Hill, New York, 1969.
Ruze, J., “The Effect of Aperture Errors on the Antenna Radiation Pattern,” Nouvo
Cimento, Suppl. 3, 9, (1952), 364-380.
Gilbert, E. N. and S. P. Morgan, “Optimum Design of Directive Antenna Arrays
Subject to Random Variations,” Bell Syst. Tech. J. 34 (May 1955), 637-663.
Cramer, H., Mathematical Methods of Statistics, Princeton University Press, 1946.
Rice, S. 0., “Mathematical Analysis of Random Noise,” Bell Syst. Tech. J., 23, (July
1944), 282-332; 24 January 1945), 46-156.
Woodward, P. M., Probability and Information Theory, with Applications to Radar,
McGraw-Hill, New York, 1953,
Skolnik, M. I, G. Nemhauser, and J. W. Sherman, III, “Dynamic Programming
Applied to Unequally Spaced Arrays,” IEEE Trans. Antennas Propag., AP-12
(January 1964), 35-43.
Kendell, M. G. and A. Stuart, The Advanced Theory of Statistics, Vol. 1, Hafner, New
York, 1963,
Franklin, P., Methods of Advanced Calculus, McGraw-Hill, New York, 1944.
Stratonovitch, R. L., Topics in the Theory of Random Noise, Vol. II, Chapter 3, Gordon
and Breach, New York, London, and Paris, 1967. Extensions to Peak
Sidelobe Theory for
the Random Array
9
T he preceding chapter developed the basic theory of the peak side-
lobe of the random array. This theory is highly idealized in several
significant ways. The purpose of this chapter is to extend the theory so as
to make it more realistic for practical array design. Eorexample, 20.2784
has_isotropic element s. Similar} arra
chromatic r: imes the element weights are not all equal.
These and other deviations from the uniform, isotropic, asymptotic
theory are discussed in this chapter.
9.1 UNEQUALLY WEIGHTED ELEMENTS
The normalized* average sidelobe level outside the neighborhood of the
main beam was calculated to be 1/N in Sections 8.2 and 8.4. This result
Relative to the main lobe.
170 Unequally Weighted Elements im
pertains to a random array of equally excited elements [see (8.1)]. Equal
weight is the most practical excitation. In a transmitting array it leads to
designs having common elements each driven at or near peak efficiency.
In a receiving array it leads to a maximization of SNR. A further and less
obvious advantage is a lower sidelobe level.
In contrast to (8.1), let the random array be generalized to
x
iw=*% >, 4,5 (x — Xn) (9.1)
where the a, either are random variables described statistically or
deterministic values derived from some algorithm. In either case the
complex radiation pattern is
Hu) =D a exp (ik) (9.2)
Let the a, be uncorrelated, real random variables. The main-lobe amp-
litude f(0) also is a random variable, the expected value of which is
Ff) = 45 a, =a (9.3)
lobe. Outside the the neighborhood of the main lobe the contributions from
each element add with random phase angles (as in the discussion in
Section 8.2). If a, and x, are assumed to be independent, the ensemble
average, or expected value, is
Ef(u) =fRd a,
= GF {w(x)}
2 fou) (9.4)
as in (8.8). The power pattern is
fu) f*(w) =< DD anak exp [jk (Xe — XU]
=> aat +155 ana exp ike — Xm )u] (9.5) m Extensions to Peak Sidelobe Theory
and its expected value [following (8.9) and (8.10)] is
a ws
Efaofeu)= wD a)
(9.6)
where a” is the mean square or the second moment of the pdf of a ; that is,
a” = f% aw(a) da. Equation (9.6) is the mean power pattern. As in (8.10),
it consists of an angle-independent term and a term proportional to the
design power pattern. At angles away from the main lobe the latter term
may be ignored. The angle-independent pedestal is the theoretical asymp-
totic average sidelobe level. The ratio of this quantity to the square of
(9.3) is the average sidelobe power level normalized to the average power
level:
Average sidelobe _ a”
‘Average main lobe ~ Na? on
Note that (9.7) is greater by the factor a”/a” than 1/N, the value when all
elements are equally weighted.
The second moment of a variable is the sum of its squared first moment
and its variance (or second central moment). Thus 2.55.45 p37
= are! a -2
@=a@+o0 f A-s 9.8)
where o,° is the variance of a; that is, f(a a)*w(a) da. The
average normalized sidelobe power level is, therefore,
a
Average sidelobe _ 1+ 0.°/a?
‘Average mainlobe ~ -N’ 9)
In the special case a,=1, for which the sidelobe statistics were
calculated in Chapter 8, o.’ = 0, and (9.9) reduces to 1/N, as in Section 8.4.
In the general case a.’ >0, and (9.9) is the correct expression.
Variation in element weight is only one factor that causes the variai
array for a variety of reasons, any one of which will cause th r
of COT or 5) to be great 979) to be greater than unity. Such an increase represents a
growth in the average sidelobe level relative to constant signal strength
across the array.
Although no single number represents this increase, a few examples Unequally Weighted Elements 173
readily yield a range of such values. Antenna elements having nearby
conductors, such as ground planes, reflectors, or inadvertent scatterers in
the local environment, may find the local field canceled, at one extreme,
or approximately doubled at the other extreme. This variation in field
strength across the array may be modeled by a uniform distribution from
a =0 to a =24. For this distribution
=
aia
aa [ aida
which implies that the average sidelobe power level would be 33% larger
than it would be if the signal strength were constant across the array.
Another important distribution is the Rayleigh pdf
ve?
4a°
3
(9.10)
w(ay=Ser", a =0 9.11)
o
This distribution results when the number of signal components of
comparable strength at each element is large and when the size of the
array is sufficiently great so that the phase relations between the signal
components are independent from one part to another of the array.
Signals arriving via scatter circuits and multipath circuits exhibit Rayleigh
fading statistics. An example of the latter is HF propagation through the
ionosphere. In an array more than a few thousand feet in length the
amplitude variations in signal strength, due to the time varying refractive
properties of the ionosphere, are decorrelated across the array.
The first and second moments of (9.11) are
a= [awiayda= 03 (9.12)
a= { a*w(a) da = 20° 9.13)
from which the increase in average sidelobe power level is
(9.14)
This increase is within a few percent of (9.10), each representing a growth
of approximately 1 dB in the average sidelobe relative to constant signal
strength across the array. 174 Extensions to Peak Sidelobe Theory
The effect upon the peak sidelobe also is readily calculated. The pdf of
sidelobe amplitude remains Rayleigh; therefore the derivations in Sec-
tions 8.7 and 8.8 are unchanged (except for the increase in mean square
value by a7/a*). The peak sidelobe estimator of Chapter 8 is a ratio of
peak to average value. Since the average level increases nominally by
1 dB, so does the peak sidelobe.
9.2 NONISOTROPIC ELEMENTS+
Practical antenna elements are far from isotropic. The simplest radiators
approximate dipoles, which exhibit peak gain about 2 dB greater than the
isotropic radiator. Since the array gain is the product of element gain and
the gain due to the number of elements in the array, element gain is an
inexpensive way to build up array gain. The limitation that the designer
faces is that the element beamwidth shrinks as its gain increased,
restricti the nominal angulat_scanning_range “of the array to the
beamwidth of the antenna elements. However a gain of 5 dB is common,
and one of 10 dB is not rare. Even a gaii 0 dB permits angle scan of
the array over approximately 20°. Thus it is pertinent to examine the
effect of element gain upon peak sidelobe level.
Since the complex radiation pattern of the radiating element and of the
array structure are multiplicative (Section 5.3), only within the main lobe
of the element pattern are the sidelobes of the array factor weighted
strongly. The dominant effect of element gain is reduction of the number
of sidelobes that can compete to be the largest. In effect, the parameter n
is made smaller. Roughly speaking, for the broadside linear array, n_is
reduced from the value given in (8.27) by the factor sin (Aur/2), where
Aug is the beamwidth of the element.
Consider a linear array steered to uo. Let each element have a
rectangular beam of width Aus, as sketched in Figure 97. Only the
sidelobes of the array factor within the interval |u|
