IEEE TRANSACTIONS ON A N E N N A S AND PROPAGATION, VOL. AP-24, KO.

5, SEPTEMBER

1976

585

Adaptive Arrays
SIDNEY P. APPLEBAUM,
FELLOW, EEE

Abstract-A method for adaptively optimizing signal-to-noise the ratio of an arrayantenna i presented.Optimumelementweightsare s derived for a prescribed environment and a given signal direction. The derivation is extended to the optimization of a “generalized” signal-tonoiseratio wbich permits specification of preferred weights the for normalquiescentenvironment.Therelation of the adaptivearray to sidelobe cancellation is shown, and a real-time adaptive implementation is discussed. For illustration, the performance ofan adaptive linear array is presented for various jammer configurations.

I. INTRODUCTION
RRAY ANTENNAS consisting of many controllable radiating elements are very versatile sensors. The pattern of the array canbe steered by applying linear phase weighting across the array and can be shaped amplitude by and phase weighting theoutputs of thearray elements. Most arrays are built with fixed weights designedto produce a pattern that a compromisebetween resolution, gain,and is low sidelobes. The versatility of the array antenna, however, invites the useof more sophisticated techniques for array weighting. Particularly attractiveare adaptiveschemes that can sense and respond to a time-varying environment. In this report, we showhowadaptivetechniques can be applied to an antenna array to reduce its susceptibility to jamming or interference of any kind. We begin with a derivation of the “control law” for the array weights that will maximize the signal-to-noise ratio (SNR) of the array output in the presence of any spatial configuration of noise sources.

A

V

+
s

n

Fig. 2-1. Functional representation of “optimum” linear coherent combiner.

11. SIGNAL-TO-NOISE OPTIMIZATION It is well known thata uniformly weighted array gives the maximum SNR when the noise contributionsfromthe element channels have equal power and are uncorrelated. These conditions are approximately valid when receiver noise and uniformly distributed sky noise are prethe pk2 = E(nk*nl) (2- 1) dominant noise contributions. (They pertain exactly in linear halfwave space array antennas.)However, when there where the asterisk (*) denotes the complex conjugate. We is directional interference from other in-band transmitters, note that jammers, or natural phenomena,the noise out of the element = E(nl*nk) = C(kl*. (2-2) channels will be correlated, and uniform weighting will not optimize the SNR. When the K channels represent the output of the element of an array antenna, the covariance terms pkl are determined by receiver noise and the spatial distribution of all ManuscriptreceivedApril 20, 1 9 7 6 . This paperis reprintedfrom SyracuseUniversity Research Corporation, Syracuse, NY, Technical noise sources “seen” by the antenna. Here we assume that Report SURC TR 66-001, August 1966 (revisedMarch 1975). This study was performed under the sponsorship of the Advanced Projects thecovariances are known.The desired signal, when it Agency, Ballistic Missile Defense Office (Project Defender), on Con- occurs, is assumed to be present in the Kchannelsin proportract AF 30 (6021-3523, ARPA Order 5 6 1 , Program Code 4720. The author was with the Syracuse University Research Corporation, tion to the knowncomplex numbers s k . The signal in channel Syracuse, NY. He is now with theGeneral Electric Company, Syracuse, k is represented as m k , where a defines the level and time NY 1 3 2 0 1 . Editor’s Note: This paper was first published as a techniczl report. variation of the signal. In a linear arrayantenna with Its circulation was somewhat limited, consequently, this classic report equally spaced elements, the s k are determined by the is being reprintedin this issue with the kind permissionof the Syracuse direction of the desired signal. Thus, if the desired direction University Research Corporation.

In this report, we consider the problem of determining the array weights that will maximize the SNR for any type of noise environment. The problem may be viewed as that of finding an optimum coherent combiner for K channels as shown in Fig. 2-1. We assume in this discussion that all signals have “bandpass” frequency spectra. All signals are represented by their complex envelopes. These are assumed to modulate a common carrier reference that never appears explicitly. Each of the K channels contains noise component whose a complex envelope is denoted by n k . The envelope power in the kth channel is denoted by p k k and the covariance of n k and n, by

2-1). = W. it can be signal and noise outputs of the combiner. W } . In general. Since M is a positive definite Hermitian matrix. M is the covariance matrix of the noise components. ~ * . Since pkZ= P .}W = W.) we start by writing expressions for the than zero whenever W # 0. We will show that the optimum weights are determined by = E{W. however. (2-8) the problem intoonein which all channels have equal l= 1 i power noise components that are uncorrelated.s. 2-2. 24. = M*. s = aS.so that the SNR at the output of the combiner is maximized.After the transformation shown there. Pn = W. The transformation matrix A followed by ..Similarly.EEE ~ S A C I I O N S ON ANTENNAS AEiD PROPAGATION. As noted earlier. the matrix M will be a diagonal matrix. 2-2. the matrix M is Hermitian. u To derive ( . Functional representation of combiner that is equivalent to one shown in Fig. is K This means that a transformation exists that will transform us = u w. It is also positive definite since the output noise power P.13) MW where = .W (210) and where AN. (2. become output may be expressed as = AS (2-1 6 ) V. (see Fig. we have The expected output noise power will be P = E{Ivn12> = E{IW. . This may be conveniently expressedin matrix notation as Assume that the transformation matrixis A and consider us = u w . is 0 rad from mechanical boresight.US* The expectationoperator E willaffect terms.W (2-9) the block diagram shownin Fig. the noise spectively. n] *K (2-1 1) A caret (*) has been used to denote quantities after the transformation.7) 1 m= N = ['. The problem is to choose the weights w. M . rewhere the subscript t means transposed. is an arbitrary constant. that is.*MW (2-14) (2-1 5) M = [&i] = covariance matrix of the noise outputs (2-5) where M = E{N*N. (2-12) onlythenoise (2. The signal output "diagonalized" by a nonsingular coordinate transformation.*E{N*N. M may have nonzero entries in any position. 2-1.} = [pkz]. is greater are column vectors and . S E P T E ~ ~ E 1976 R "n s Fig.NI~I n = E{(WrN)*(N:W)l sk = exp ( j E sin e) d 1 where d is the element spacingand 1 is the wavelength. the signal and noise components. If the noise components are uncorrelated.*N*N.N = N. hence.

The result is (2-43) It is important to emphasize that the solution. of Dividing the magnitude square (2-36) by (2-37). = 1. 19) (2- From (2-27)we have P. * (2-23) (2-24) (2-25) By (2-39.we obtain ? us = Iqfi = JP~AN.. Using (2-26) in (2-25). for the combiner in Fig. (2-27) Thus the optimum weight vector W. If theoutputs of thetransformation matrix A are combined with the weights I?k. Thus. including thenormal "quiescent" environ= pS* (2-31) ment (no jamming). we see that A"MA. may be made equivalent to the combiner of Fig. array designers are willing to compromise on SNR in order to where p is an arbitrary constant. from (2-14) and (2-22). = 11 @/I2. we may write. expressed by (2-42) maximizes the SNR.@. = AfI?oopt= A.. if we substitute t = pS* into (2-18) and (2-27).. M = (A... particularly sideTo show that (2-31) is the optimum.. for equivalent outputs we have W = A. (2-22) Crp3. this is the maximum possible value of the SNR. 2-2. For example.APPLEBAUM : ADAFINE ARRAYS 587 a combiner. = pM-lS*. = pA. 2-1 is the value of W that satisfies the equation MW = If the configurations of Figs. = W" * . is given by 111. This yields used to obtain 30 dB sidelobes with less than 2 dB loss in 1412 5 1~1211~11211~112 (2-32) SNR.} = 1. therefore. sides of (2-32) by P. In most applications. theoutput signal will be us = E@$ (2. obtain we Similarly. W. we need criterion a for the mop. 2-1 by properly relating the weights used in each case.. Thus. the output noise will be u. 2-2 is !Pn x = lCr1211S112. .18) = aqrAS. dividing both mt*I?.. we show how theoptimum weights may be determined adaptively in real time. obtain some control of the pattern shape.we note that combining the channels after the transfonnation matrix A with the weight vector @ is equivalent to using the weight vector A r m without transformation the matrix. In Section 6.. however. It is well known that the optimum choice for the weighting vector @ in Fig. thus. (2-28) (2-29) (2-30) Comparing this with (2-27).. as shown in Fig.*fi*flrW} = wr*E{fi f i r } ) $ ' . we have shown that &* is the optimum value for I?. However. E(fi*fi.A*)-l. (2-42) The SNR corresponding to the optimum weighting can be obtained by substituting (2-16) into (2-38). = @. where the noise components i?k have equal power and are uncorrelated. 2-1 and 2-2 are equivalent.. the covariance matrix the of noise components after A is simply the identity matrix of order K .&" or using (2-16). = pS*.. we obtain (2-26) w. where llSll2 and III?112 = = 3:s (2-33) (2-34) thus. and. apply the Cauchylobe levels. that It then. Comparing (2-19) and (2-21) with (2-9) and (2-IO). = (2-35) (2-20) (2-21) This puts an upper bound on the SNR.. is apparent. or controllaw. (2-39) (2-40) Since the transformation matrix A decorrelates the noise components and equalizes their powers. 2-2. (2-38) P. from (2-30). (2-41) P.W'.*~*~~. wehave The output noise power expressed in terms of the quantities in Fig. fit. P.A*S* and. then the array designer is forced to accept the SNR as the governing criterion for all noise environments.= E { ] P t R} I' = E{@. finally."S = = rp113112 (2-36) (2-37) and P n IPI'IISII~. The optimum value of W may now be obtained from (2-22) : IV. Equation (2-29) expresses the fact that the transformation matrix A diagonalizes the matrix M . GENERALIZED SIGXAL-TO-NOISE OPTIMIZATION If an array is designed so that the weights are adaptively controlled to satisfy (2-42)..*m= 11I?l12. Dolph-Chebyshev weights can be Schwartz inequality to (2-18).

for all noise environments. Using (3-10) and (3-11) in (3-6) and (3-9). The GSN is a measure of how close the weights are to the generalized signalvector T inthe“dot”on“inner”product sense. we can obtain the following expressions : (3-11) .*MW) (3-7) The output noise power of an adaptive array controlled 1PI2 accordingto (3-2)will not remainconstant as the noise 1 environment varies. If these are statistically independent of the noise sources in the quiescent environment. the most desirable array weights are given by the weight vector W.W. we have. .):M.M . (3-2) From the results of the previous section. that the weight (3-16) vector is held fixed at W. represents an optimumcompromise between gain. then from the results of the previous section. Then the GSN will be and (3-4) Using (3-l).. we see that the effectof the jammer on the GSN with adaptive control isreducedby For an adaptive array controlled by (3-2).. and define the column vector T by the equation M.thatappears in analyses sidelobe of cancellation. Designating this ratio by (JIN).. M W = pT* = M. we also know that controlling the array so as to always satisfy (3-2) is equivalent to maximizing the ratio (3-3) (GSN). 1 . (3-14) the numerical valueof the GSN by itself has nosignificance. we may write M = M. Let the covariance matrix in the normal quiescent environment be M.W.W. This is a more general criterion maximizing the SNR. this mayb e written as When the weights are fixed. (3-8) ing the array combiner to maintain a constant noise level.). decided to optimize to anequivalent signal vector T instead of the actual signal vector S.UT* (3-1) M In the normal quiescent environment where W = M.):M. etc.. I (W.). and (GSN). It is only significant in comparisons in which the vector T is held k e d . SEPTEMBER 1976 control of array weights that will give more flexibility in beam shaping. (GW. (3-12) The ratio in the denominatorof (3-1 1) is the ratio of the jammer noise output to the quiescent noise output when the weights are fixed at W.*M..1M. we have thefactor r.. Comparison of the GSN of a fixed array with that of an adaptive array is of interest. we note from (3-13) that the GSN varies almost as the inverse of the jammer-to-noise ratio. Suppose. sidelobes. the weights should be adjusted so that + Mi (3-10) where M j is the covariance matrix of the noise components due to the additional noise sources. in the normal quiescent environment. than which it includes as a special case when T = US.1M w. Suppose that. The point of view we wish to present is that in choosing W.M . (3-5) and (3-8) both reduce to = W. (3-3).We refer we may write (3-11) and (3-12) as to T as a generalized signal vector and the ratio. A tractable approach can be developed as follows.. in effect.%. however.M . becauseof the presenceof additional noise sources (jamming). lP12 .W. as being optimum.. to continueoptimizing on T.588 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION.a it is figure-of-merit that describes the performance of an adaptive array in a specific noise environment. where p is a normalizing constant.= 1 *. We assume that W . we notethatthecovariancematrix differs from M.):M. Since the “level” of T i s arbitrary.= 1 1’4 (W. If the environment changesso that the covariance matrix becomes M . as (3-13) the generalized signal-to-noise ratio (GSN).(W. This may require using an AGC follow= -(( W. 1 (W. and relative to output noise. = . then.W. From (3-14). (3-9) In order toput (3-6) and (3-8) intomoremeaningful forms. Thisfactor r is analogous to the jammer cancellation ratio.

M. e . = sin e. hence. is IV. metric based on the quiescent covariance matrix M.W. and W is then equal to W.*M.).*MW . (4-4) /. we recognize that the weight vector W.W.*M-'(MW . the change in the noise level will be AP.W.1)Ps The first term in (3-22) represents the output noise power due to the jammingsources.and.P 3 (4-5) The expression for 4. Now. we consider here a linear. (3-23) This can be expressed in matrix notation as It is easy to verify that (3-23) does reduce to (3-22). the resultant pat11 W . uniformly spaced array that is weighted in accordance with (3-2).APPLEBAUM:ADAFTWEARRAYS 589 Without the AGC.M. we get .n 8. G. The covariance matrix in the quiescent environment would then be a diagonal matrix of the form + ( W . In so doing.*M. - = . We assume first that in the quiescent environment the noise output of the element channels have equal powers and are uncorrelated. exp j ( r . This implies that both terms in (3-23) are nonnegative. represents an where optimumcompromise between gain.W. where g = W. If the amplitude weights are all equal. when the jamming power is large compared to any weight vector W. can be manipulated into the I= 1 form where 4 = (Wq)r*(MqM-"j)Wq /3 = 2nd e. q will have its minimumvalue when the second term is minimized.. When the environment K number of array elements. identity matrix of order K. 1. + ( M W . may no longer be optimum.M lMjWq. = (3-19) (3-20) (3-21) (W. The second term is a measure In (4-3).and M.M . exp -M . in the quiescent environment. they will not be equal. The pattern obtained with conversely.).M . however. (3-22). A suitable Q1 criterion is to choose a weight vector W that minimizes the a2 exp -jBs quantity q.(M\).Il.M. sidelobes.'Mi also positive are definite Hermitian matrices. (3-22) a. = W. noise power output of each element. Since the first term is fixed. Substituting W = (3-18) as it should be. The minimumvalue of the second term is 0.).W. the control law for the weights W in .*MqWq.. factors in the quiescent environment. M j .).Wq from (3-3). M . changes due to the introduction of jamming noise sources. Since Mq.(Wq). = PqlK Assume. the importance minimizing of the deviation. The deviation is measured with respect to a P. will be the quiescent noise.M - ". It is interesting to observe that the control law given by (3-2) can also bedeveloped from another point ofview.W.*MM-'MqW. 1.W.*Mj W To illustratetheconcepts and results of the previous section. MW = M. LINEAR ADAPTIVE ARRAY M-'M. and this occurs if. How should the weights be changed so that the Now let the desired weight vector the in quiescent environeffects of the jamming are reduced? We want a criterion mechanical ment for a signal in the direction 0.W.W. Notethatinthe quiescent environment M reduces to M . the output noise level will decrease when additional noise sources are added to the environment.' M i are positive definite Hermitian matrices.(W. It is also interestingto observe that the value of q is the negative of APn (see (3-21)) when (3-24) satisfied.. again.). (3-24) Since the weights onthearrayarecontrolledinacThis is thesameas (3-2). and M are positive definite Hermitian matrices.*M. the u1 are amplitude weights (real numbers) and of the deviation of the weight vector W from the quiescent + optimum W. APE = (M-'M.(P) = a.1)(B .. as has already been remarked. and only if. In general.thatthe weightvector W.*M. cordance with (3-2).. because of the minus sign in (3-21).(W . M. minimizing thejammingpower will K receive greater weight.). This implies that. from which reduces the jamming effects and still recognizes the boresight be optimality of W. thejamming is small compared tothe quiescent noise. and other p . is emphasizedwhen the noise introduced by terns will be of the form (sin Kx)/(sin x).).

l ) P j . willbe J ( t ) exp j(I . therefore. = ' [ .(#Ij)Bj*. Using the property H* = H . - (4-18) (4-19) Since C(0) = K.Bj* = C@ Bj) (4-28) where and U is a K x K matrix of ones. is just thesum of the two covariance matrices thus.pj).Hence. BtWq = Gq(P) (4-27) rl 0 1 and it is easy to show that B. (4.adaptive the control reduces the gain in .OBjI). Thisis shown in Fig. where pi is the envelope jamming power in each channel. (416) ~q + KPj To obtain the weight vector W from (4-IO).Bj*. jammer will be + + ( ) W = = (Pq + KPj) H'UH) Wq Wq .l)Bj. andthe second. it is easy to linear array in the presence of a jammer consists of two verify that parts. Then if the jamming signal in the first element channel is J ( t ) .( + KPj) H*UHWq.3) 1 From (4-7). whichis subtracted from the first. (4-26) M. it can be conveniently expressed as The pattern obtained with theweight vector W may now be expressed as G@) = B. - ( Pq + KPj ' j ) G. The covariance matrix due to the jamming signal.' . where (4-1 1) Bj = 2nd sin ej.590 TRANSACTIONS ON ANTENNAS AND PROPAGATION. therefore. 4-1. quiescent plus jammer. (4-30) = pqlK pjH*UH. The gain of the array in the direction of the Using (4-17) in (4-lo). thejamming signal intheIth channel. assuming narrow bandwidth. ' The noise environment we wish to studyis that caused by a single jammer added to the quiescent environment. The covariance matrix for the total noise environment. (4-9) UHW. however. MW = MqWq = pqWq or W = pqM Wq. will be pj exp -j(k . The covariance of the jammingsignals in the kth and Ith channels. substituting (4-27) and (4-28) into (4-26) gives M = Mq M j (4-1 ) 5 G@) = Gq(P) ' j Gq(Pj)C@. will be lil (4-22) H"UHWq = Gq(Pj)Bj* where r 1 1 Substituting (4-22) in (4-19).W = B.)P If the array weights remained fixed at Wq in the presence of the jammer. (4-24) Mj = pj(exp C-Ak . we require the This says that the pattern of the adaptively controlled inverse of M . = Gq(Bj) 1 10) (4and finally. the gain the direction of the jammerwould in be Gq(Pj). Let thejammerbelocated at the angle Bj frommechanical boresight.The first is the quiescent pattern Gq@).Wq (4-25) - ( Pq + KPj ) Gq(Bj)B. is a (sin Kx)/ (sin x) shaped beam centered on the jammer. (431) reduces to G@j) = Pq From the definitions of Wq and H we have HW. ( Pq + KPj ) GqGj>* (4-32) a 1 a 2 exPj@j - P3 P3 (4-20) UK e x ~ j ( K 1(j . Because of its simple structure. = pjH*UH where H i s the diagonalmatrix (4. ( 1 . get we W = W.12) This is a Hermitian matrix of order K in which all terms onthe samediagonalareequal. SEPTEMBER 1976 any noise environment will be therefore. 3.

The first step is to select a T vector. we see that y = 1 + . In that case. 4-1. we note that against the jammer. foraJammer in the sidelobe region. In this case. To apply the concepts of the adaptive array we assume that the outputsof the channels are weighted and summed. Kpj/p. Since G(B) is a voltage gain pattern. purpose The of the auxiliaries is to provide independent replicas of jamming signals in the sidelobes of the main pattern for cancellation. we obtain r = 1. The problem is to find an appropriate control law for the weights. It is interesting to note that this result is the same as the reduction in voltagegain of the pattern in the direction of the jammer and that it is almost independent of the choice of Wqand. = ( Pq + KPj 1 pq ) Mj. < 1 and.is Kpj/pq. It consists of amain. it is tempting to conclude that the performance of the array against the jammer power had been improved by the square of the inverse of Equation (4-33). Forthe sake of completeness andasa second illustration of the applicationof the concepts.KPj Pj The concepts presented in this report actually are a generalization of the coherent sidelobe cancellation techniques. The amount of desired target signal received by the auxiliaries is negligible compared tothetarget signal in themain channel. hence.pi). we will < have ( J / N ) . we shall show that sidelcbe cancellation may be viewed as a special case of an adaptive array. is the jammer-to-noise ratio in the “cancellation” beam. we choose the K + 1 column vector Now the maximumpossible value of ( J / N ) . In that case.. of the shapeof the quiescent beam. The auxiliary antenna gains are designed to approximate the average sidelobe level of the main antenna gain pattern.and this will occur only if the jammeris at the peak of the main beam of a (sin Kx)/(sin x ) quiescent beam. which is to be expected. (Resultant pattern is difference between quiescent and cancellation pattern. since the signal collectedby the auxiliaries isnegligible and the main antenna has a carefully designed pattern. indicating no improvement of performance T = .) the direction of the jammer by the factor n yq Theratio in thedenominator of (4-33). hence. 5-1.APPLEBAUM : ADAFTWE ARRAYS 591 “ D e s i r e d Signal” Jamming Signal 1 1 C a n c e l l a t i o n Pattern / B Fig. The proper measure of the improvement against the jammer is the cancellation ratio r defined by (3-16) and (3-17). and MjMj = KpjMj (4-35) therefore. high gain antenna whose output is designated as channel “0” and K auxiliary antennas. . V. r z y. A sidelobe cancellation system is shown in Fig. C(g . SIDELOBE CANCELLATION . Thus. The adaptive control will be most effective against jammers in the sidelobe region of the quiescent pattern. however. would not be correct. the adaptive control will “cancel” the jammer power by approximately the jammer-to-noise ratio in the cancellation beam. Sketch of pattern of adaptively controlled array. << Kpj/pq: so that y(J/N). (4-37) Using (4-37) in (3-17). which are repeated for convenience: (3-16) and (3-17) To evaluate (3-17) in the present context. This.

is a ratio in which the level of W' does not matter. fixed at 8. it must clearly be minimizing and the output power. because of the unique form of the - + The control loops for an adaptive array can be implemented using thesamecircuitry as isused forcoherent sidelobe cancellation. = 9 .wl.)%. at a nonzero value if we are sure that no solution to (5-2) will require wto be 0. 5-1. it is possible to achieve an optimum combiner with only K control loops. The multiplication and summation occur on carrier frequencies at IF. since the inner product . (5-8) The optimum controllaw would then be + Since theinnerproduct W i T = f i 0 is fixed. (5-8) is the sidelobe cancellation control law for minimizing the output noise power. we can auxiliary channel weights.the GSN will never be optimized with w. T vector. Equa(5-5) tion (5-8) has been obtained using this criterion in previous studies of sidelobe cancellation. The basic idea is shownthe in functional block diagram in Fig. then any multiple of W' will also be optimum the channels and W' is the K 1 column vector of all the since the figure of merit. . Now let M be the K by K covariance matrix of the let the parameter p in (5-2) vary freely. ... x. andthe p o = poo' of M' = noise power output of the main channel control law. However.* -.To do this. In (5-2) we have set M W = . The signal in each channel u. GSN. Sidelobe cancellation system. With w. IMPLEMENTATION Note that pLl: is the cross-correlation of the output of the Ith auxiliary with the output of the main channel. with the sum signal substracting the correlation from the desired vector component t. we observe that if the M'W' = pT (5-2) weight vector W' is optimum for a given noise environment where M' is the K + 1 by K + 1 covariance matrix of all (a given M'). andthecontrol law is given by (5-7) with w. Equation (5-2) now may be fix w. Equation (5-3) may be written as two separate equations: a scalar equation. and then the weighted signals are summed. optimizes the ratio of the inner product (5-4) to the output noise power.. However. (5-6) and a matrix equation.. = 0. Thus. and then using a high gain amplifier. SEPTEMBER 1976 W 0 Y 4 Fig. to satisfy (5-6). VI. ~ -I-A. This means. 6-1.w. (5-7) The control law expressed by (5-2) can be implemented using K + 1 control loops to control the K 1 weights wo. p . Thenotation used there gives the complex envelope of signals on phase coherent carriers. Thus. in effect. (5-8). The weights w k are derived by correlating u. that we may weights.592 IEEE TRANSACTIONSON ANTENNASAND PROPAGATION. partitioned as follows : W. MW = -wOA. instead of auxiliary channelsonly and W be the K column vector of the fixing p and trying to control w. is multiplied by the weight w.'T = w.* W = p . only need control loops for the we (5-3) auxiliaries.*.A.

. COMPUTER SIMULATION mined by the “integrators” shown in Fig. Thus. the weights will be uncorrelated with ul. For each Wk. has only positive eigenThis differs from the optimum control law. (7. + where T is the timethe of “integrator” constant circuits. Since M is a positive definite Hermitian matrix. 7-1 through 7-9 show the results of the computer these are high ( . The K equations obtained from (6-5) may be represented in matrix form as z . the irnpleintroduces an error analogous to the “servo error” of a type 0 servo. The preceding discussion demonstrates that the steady.neglected in the preceding analysis. single-pole circuits. one. K (6-2) where (A.1) .. may make the loops state solution of the control loops is essentially the desired unstable if the loop gains are allowed to become excessive. 6-1. The dynamic behavior of the control loops is deterVII. is also a positive definite Hermitian matrix. and it was assumed that the arrayweights were determined by the control law (M k] W = T*. if we apply the expectation operator to both sides of (6-4) we get (uk*ul + 5) = fk* G = for k = 1. This equation determines the dynamic behavior of the expected value of w k . 6-1.The differential 2 simulation of an adaptive array. the K equatjons represented by(6-2) may be written as dW + + GT*. I I I Fig. In practice Figs. hasonly positive eigenvalues. In practice.. 1.= -(GM dt Recalling that uk*u. + when k = 1 all other k. It. The servo error discussed previously was neglected. by the values. second-order effects. we have Wk = G [ tk* - uk* I= 1 C K WlU1) (6-1) or K w I 1=1 loops The are designed sothe that weights w k will vary slowly compared tobandwidth the of the signals ul. 6-1 is stable under all conditions for the assumptions made. Hence. MW = T*. The array used in these equations describing thedynamic behavior of theloops studies was a 21-element linear array with half-wavelength can be shown to be spacing. Thus. (3-2). * .This term GM therefore.APPLEBAUM: ADAPTIVE ARRAYS 593 kth Array Elements 1 K f I Same f o r each e l r m e n t . high gain. (6-6) It can be shown easily that matrix equations of this form are stable if the matrix GM + 1. is an element of thecovariance matrix M . addition of a term inversely proportional to gain. Its effect can be made negligible with sufficiently mentation shown in Fig. Functional block diagram of implementation of element control loop for adaptive array.

0 SIN 9 0.0 -1.- 3 I o ~ - ~ . Patterns with eight “randomly” located narrowband jammers.5 0.0 -0. 7-2.5 I 10 .5 I + & + I z -0.5 I I 10 .5 10 .594 IEEE TRANSACXIONS ON ANTENNAS AND PROPAGATION. Quiescent patterns(no jammers present). Nulls are produced at locations of jammers. 1 SIN e 2 0 (4 20 -1 10 100 0-10 -10h -20 -20 c9 v m v W a d -40 -30 c.5 0 0. (a) -20 dB sidelobes (quiescent). (a) -20 dB sidelobes. SIK e (b) Fig. U ~ -40. SEPTEMBER 1976 *O 1 -50 -60 -Lo 1‘ I I I 0 I -0. (3) Fig. (b) -40 dB sidelobes.:I ~~ ~ ~~ -50 -50 -60 -60 -1. Nulls below -60 dB are plotted as 60 dB. . 7-1. (b) -40 dB sidelobes (quiescent).

7-4.0 I -Lo -0. -0. Each cluster is about half a sidelobe width wide. 7-3.5 I 1 0 : e SIN e 20 - (a) (a) 10- 20 0- 10.5 * SIN O I -60 0. (b) -40 dB sidelobes (quiescent).0 -0. Same as Fig. Jammer-to-noise ratio in each element is 0 dB.5 1.5 1. Jammer configurations consist of five clusters of three each.APPLEBAUM: ADAprrvE ARRAYS 595 10 2o 1 20 l0 [ o A 4 R -60 I -1I. 0 -10- h -20 -30 -40 -10 m a v s m a v -20 -30 -50 -60 -50 -1. (a) -20 dB sidelobes (quiescent). .5 0 SIN 0. 7-3 with 15 dB jammer-to-noise ratio.0 e SIN (b) Fig.5 I 0 1 0.O * . e (b) Fig.

5 0 ' SIN 0. SIN (b) e SIN (b) e Fig. .5 (a) (a) 1 h 10.0 -0. Nine jammers are clustered andspan almost two sidelobes.5 0 0. SEPTEMBFX 1976 20 3 m -20 -30 d -40 -50 -60 1 0 : 0..5 z! rn 1 1. Same as Fig. IEEE IXANSACTIONS ON ANTENNAS AND PROPAGATION.+ -0. 0 -10 -20 I 7 0 0.0 -60 -1.5 10 . 7-6. 7-3 with 30 dB jammer-to-noise ratio.5 I -! lO I ' I -0. 7-5.0 SIN e 20 e -10 .i -1. . Fig..5 1 1 1.0 ..

-60 I I * I - 1 1.5 10 . u 5 d -30 -40 -50 -1.5 SIN cb) e SIN e (b) .0 -0.5 SIN 0 ' 0.APPLEBAUM: A D A m ARRAYS 597 2o 1 20 10 0 1 -10 h -20 m v a 5 d -30 -40 -50 .!i 1.0 1 ' .5 I * 1.0 1 e (a) 0 -10 :i h 20 0 -10 -20 v m . 5 0 ' d O.0 I -60 -Lo I + -d.0 0 -0.5 0 0.5 0 0. -.

’ 7-9. Thus. Except for format changes in some figures. Deep notch is produced in center of main beam. andnarrowbandwidth was a s s d e d . and 40 dB sidelobes: The matrix M was computed for various spaiial configurations of jammers. 30.. Various vectors T were used to give different quiescent beams. Company. There is much less effect on pattern level than in first case. Wide bandwidth be can simulated by using narrdwband calculations and spreading the jamniers spatiSlly. The desired target direction corresponds to the center of the plots.5 0 0. many of the figures show configurations of clustiired jammers. 7-9.5 10 . two plots are given in each figure with the same jammer environment assumedin both cases. only 20 and 40 dB Dolph-Chebyshev patterns are shown. sin 8. The antenna gain patterns in Section 7 were replotted on a Calcomp plotter (originally they were plotted using a line printer). where 0 I 8 I n/2 from .-A(sin 8) (7-2) f. Stauffer. Main effectis drastic lowering of level of pattern. Chapman for handling these format changes. ACKNOWLEDGMENT This report has been reprinted because of the recurring interest in this study. . SEPTEMBER1976 - io ‘“I 0 0- -10 3 m -20 - -60 -1 -1. The author also wishes A(sin 0) spread of the cldster in sin 8 space. This report has also in presence the of jammers. the text corresponds to that in previous editions. to thank J.. for useful discussions. Shape of main beam is unaltered by presence of jammer since cancellation beam has same shape. -z - - . J. (b) -30 dB Dolph-Chebyshev quiescent pattern.1 I sin 8 I 1. Each cluster may be regardedas asingle wideband jammer. (a) One jammer is a t boresight knd cluster of three is in sidelobe.0 t I I I -0. center of the cluster. The equivalent bandwidth A of a f cluster of jammers expressed as a fraction of the center frequency is Af . Except for Fig. SIN e SIN (3) e (a) Fig.. Quiescent patter is (sin Kx)/(sin x ) beam. The abscissa is sin 8 for been identified as SPL TR 69-76 and SPL TR 66-1. presently of the General Electric sin 8. 598 20 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION. and help The results are shown as plotsof the pattern of the array with the artwork of the original report. suggestions. For ease of comparison. The author extends his thanks to D. This occurs because main beam of quiescent pattern is broader than main beam of (sin KY)/ (sin x ) “cancellation” beam. W. The quiescent beamsstudied were a (sin Kx)/(sin x ) beam and Dolph-Chebyshev beams with 20.

Sign up to vote on this title
UsefulNot useful