CALIFORNIA STATE UNIVERSITY, NORTHRIDGE PHASE-ONLY ADAPTIVE NULLING WITH DISCRETE VALUES A thesis submitted in partial satisfaction of the requirements for the degree of Master of Science in Engineering by Elise Tina Oestreich August, 1979 ~\ The Thesis of Elise Tina Oestreich is approved: California State University, Northridge ii ' ACKNOWLEDGMENT The author wishes to thank Dr. Ephraim Mendelovicz for suggesting the topic for this thesis and for his continuing encouragement and guidance. iii TABLE OF CONTENTS Page LIST OF FIGURES vi LIST OF TABLES vii ABSTRACT • . viii Chapter 1 - INTRODUCTION 1 1.1 Background and Problem Statement . 1 1.2 Array, Signal and Noise Models . 2 1.3 Derivation of the Signal to Interference Plus Noise Ratio (SIR) for the Fully Adaptive Array 6 Derivation of the Optimum Weight Vector for the Fully Adaptive Array . . . . . . . . . . . . . . 9 1.4 1.5 The Maximum SIR for an ESA with Discrete Phases 11 1.6 Thesis Outline. . . 13 Chapter 2 - BRANCH AND BOUND 14 2 .1 Introduction to Branch and Bound . 14 2.2 SIR Maximization for the ESA with Discrete Phases 15 2.3 Tree Structure 16 2.4 A Fundamental Mathematical Relation 19 2.5 Fathoming 20 2.6 Updating the Current Lower Bound •. 21 Chapter 3 - COMPUTER PROGRAM DESCRIPTION . 22 3.1 Introduction to the Computer Program .• 22 3.2 Fully Adaptive Array Processing 24 3. 2.1 MWQ . . 25 3.2.2 SIRCLB 29 3.3 ESA Processing - Pre-Branch and Bound (SIRCLB) iv 29 Page 3.4 Branch and Bound: ment Priority . . Discrete Weights and Assign- 3.5 Branch and Bound: Loop Processing 3.6 ARFACT and APLOT 47 3.7 Trench-Zohar Complex Toeplitz Matrix Inversion Subroutine 47 Chapter 4 - 35 40 RESULTS 48 4.1 Introduction 48 4.2 Results for Four Element Arrays 48 4.3 Results for Eight Element Arrays 55 4.4 Problems with Algorithm . 59 Chapter 5 - CONCLUSIONS AND FUTURE IVORK 63 67 REFERENCES APPENDICES 68 A. MAXIMIZATION OF THE SIGNAL TO INTERFERENCE RATIO· · B. MAXIMUM SIR FOR A PARTIALLY CONSTRAINED WEIGHT VECTOR 77 c. COMPUTER PROGRAM LISTING v 84 LIST OF FIGURES Page 1.1 Coherent Combiner for K Channels 3 1.2 Signal Arriving in the Direction 8 from Broadside · · · • · · ~ · 4 2.1 Tree Structure for Branch and Bound 17 3.1 Block Diagram for the Computer Program 23 3.2 Interference Covariance Matrix 27 3.3 Flowchart for SIRCLB ESA Processing • 30 3.4 Flowchart for WEIGHT Subroutine, p. 1 32 Flowchart for WEIGHT Subroutine, p. 2 33 3.5 Flowchart for Branch and Bound 36 3.6 Complex Plane Representation of Discrete Set (NBITS = 3) . · · · . . . . . . . . . 41 .... 50 4.1 Adapting Patterns with Jammer in Main Beam, . ESA with 3 Bit Phase Shifters 4.2 Adapting Patterns with Jammer in First Sidelobe, ESA with 3 Bit Phase Shifters 4.3 Antenna Patterns with Jammer in Second Sidelobe, ESA with 3 Bit Phase Shifters 4.4 Comparison of Adapting Patterns with Jammer in Main Beam, ESA with 3 Bit Phase Shifters vs. 4 Bit Phase Shifters · · · · · · 53 4.5 Patterns for Eight Element Linear Array 60 5.1 Approximate Branch and Bound Search . • · 66 . . . vi . . ... ... ...... .. ... 51 52 LIST OF TABLES Page 4.1 Signal to (Noise Pluse Interference) Ratios (dB), Jammer in 3 Different Positions . . . . • . 54 4.2 Signal to (Noise Plus Interference) Ratios, dB 56 4.3 Difference Between Adapted and Quiescent Patterns in Direction of Interference, dB · · · · • . · . 57 Difference Between Adapted and Quiescent Patterns in Direction of Signal, dB . • . . . . . . . . . 58 4.4 vii ABSTRACT PHASE-ONLY ADAPTIVE NULLING WITH DISCRETE VALUES by Elise Tina Oestreich Master of Science in Engineering In this thesis a method, using a branch and bound algorithm, is described for obtaining optimal phase settings for jammer cancellation with an electronically scanning array (ESA) whose only variable parameters are element phases which vary in discrete steps. For the purpose of this study it is assumed that the interference environment is known. Therefore, no estimation of environmental noise input will be required. The signal to interference plus noise ratio (SIR) for a linear array with an adaptive complex element weight vector was given in the literature. In particular the optimum weight vector that maximizes the SIR when the weights are unconstrained has been derived. This solution is not valid when the values allowed for the components of the weight vector are restricted to a discrete set. In particular, in an ESA the only values available are those with constant viii magnitude and discrete phases determined by the number of phase shifter bits. Methods of continuous analysis cannot produce a result for this discrete problem and an exhaustive search is impractical in most circumstances. In contrast to the exhaustive search method, the branch and bound technique needs to examine only a small subset of all possible discrete weight vectors in order to arrive at the best SIR. The method has been tested and the jammer cancellation results in the test cases have been encouraging. As a minimum, the approach described provides upper bounds for adaptation performance for electronically scanned arrays that are controlled with digital phase shifters. Chapter 1 INTRODUCTION 1.1 Background and Problem Statement Radar and communication systems must be designed to cope with directional interference, intentional and unintentional, in order to fill the needs of users. This requirement has led to increasing interest in adaptive antenna techniques which can sense and respond to a time-varying environment. Adaptive arrays have the capacity to null the reception from particular directions and thereby eliminate undesirable signals. An adaptive array can be implemented by asso- ciating a coherent receiver channel with each array element. The complex gain of the receiver is equivalent to a complex weight. This weight has magnitude and phase that vary over a continuous range of values. Adaptive arrays are over-equipped to handle electronic scan- ning since all that is required for that function is variable phase shifters. In many applications electronic scanning is necessary, while adaptive nulling capability is a desirable option, often left out, due to its high cost. Using phase control only reduces this cost and provides the electronic scan function. In addition it should pro- vide an acceptable degree of nulling adaptivity. Arrays having phase control only are typically called electronically scanned arrays (ESA) . This thesis provides a way to find the best adaptive nulling perf omance for an ESA with phases that vary in discrete steps. Herein the term "fully adaptive array" is reserved for arrays with element weights that can vary continuously in magnitude and phase such as those analyzed by Widrow [1] and Applebaum [2] . 1 2 1.2 Array, Signal and Noise Models The signal and noise environment can be represented by a coherent combiner as shown in Figure 1.1. All desirable signals are narrowband and the noise is a narrowband, stationary process, which is additive and Gaussian with zero mean and unit variance. The interferences are also narrowband, zero mean stationary processes which are additive and Gaussian. Signals, noise and interference are represented by complex envelopes. The desired input signal waveform modeled as a planar wave is represented in Figure 1.2. The horizontal line along the bottom of the drawing depicts a linear array with K elements. signal arrives at an angle 8 s from broadside. The incoming In this linear array it is assumed that the elements are equally spaced by a distance A d = (1.1) 2 where A is the wavelength. This value insures that there will be no grating lobes for any scanning angle es. The equation for the signal th in the k-- array element is given by -j 2Tikd ---A- sin 6 8 (1.2) Given that n. and n. are the noise and interference components ~ J from two channels i and j the covariance of n. and n. is given by ~ J (1.3) where * represents complex conjugate. 3 s, • • • Figure 1.1. Coherent Combiner forK Channels 4 f.-~ 0 --..~oo-~-1·-- ~ ~ 2 • • • (K-11 Figure 1.2. Signal Arriving in the Direction e s from Broadside 5 It follows that m .. J1 = E{ n.J* n.} 1 E{n. n~} = m~. = 1 J (1.4) 1J Thus, the covariance matrix 6. M = (m .. ] - is Hermitian. (1. 5) 1J The power in the ith channel is given by m .. = 11 E{n~ n.} = 1 jn.j 1 1 2 (1. 6) Note that the covariance matrix is made up of two components M = M. -1 + --n M (1. 7) where --J_ M. represents the covariance matrix due to interference alone and M denotes the covariance matrix for noise alone, --n where (1.8) M --n with ~ 6. identity matrix of order K and p 6. =thermal noise power in each array element. n The derivation of the covariance matrix M. is shown next. -J.. The voltage due to interference i in element k is shown in Equation 1.9. =v.e 1 2'!Tkd -j - , 1\ . S1n 8. 1 -"kS i = v .e J 1 (1. 9) 6 where (1.10) th th The interference covariance of the k-- and t-- array elements is given by =e j(k-9.)13. = p e 1 J. j(k-9.)13.1 (1.11) The interference covariance matrix may now be given as M. -J_ = j (k-9.)13 J.~ [ p.J. e (1.12) 6. where p. =interference power in each array element. J. For the purpose of this study it is assumed that the interference covariance is given and unchanging so that no estimation of the interference environment such as interference power(s) and direction(s) are required. When there are multiple jammers, the interference covariance matrix elements are of the form of a summation of elements ~ described in Equation 1.12. 1.3 Derivation of Signal to Interference Plus Noise Ratio (SIR) for the Fully Adaptive Array In this section, it will be shown that the signal to interference 7 plus noise ratio (SIR) is given by lwt w*lz ~ ~q SIR=---- wtM w (1.13) where t represents transpose, t is the conjugate transpose, and w= = adaptive complex array weight vector (1.14) w = = unadapted (quiescent) weight vector representing the optimum compromise between such performance factors as resolution, gain, and low sidelobes (1.15) ~q In order to derive Equation 1.13 it should be noted that the signal output from the combiner (Figure 1.1) is a real scalar value. K v s = L: (1.16) In vector notation it is given as v ~ s (1.17) was modeled as a planar waveform and is completely defined when the direction 8 s is given. 8 In order to maximize v s in Equation 1.17 it is sufficient to set w = s* (1.18) ~q so that all the voltages in the array elements add up coherently. When Equation 1.18 is satisfied, the array illumination is uniform with the attendant high sidelobes. The signal power can now be obtained from Equations 1.17 and 1.18 as p s = (1.19) The noise output may be expressed as the scalar value v (1. 20) n where N = = noise and interference vector The expected noise power· can be represented by = = 9 Since the expectation operator only operates on the random noise terms, p n or p (1. 21) n where (1. 22) Therefore, the ratio of the signal power to the noise (plus interference) power is given by p SIR= s p ]wt w*]2 = n 1.4 - wt -q M w (1.13) Derivation of the Optimum Weight Vector for the Fully Adaptive Array The corresponding optimum weight vector which maximizes the SIR is given in this section but its derivation is deferred to Appendix A. This result is due to Applebaum [2]. However, the proof given in the appendix follows other authors. The optimum weight vector that maximizes Equation 1.13 is given by (1. 23) The numerator in Equation 1.13 is shown in the appendL~ to be a quadratic form, (1. 24) 10 with Similarly, the denominator of Equation 1.13 is a quadratic form. Since M is the covariance matrix for a stationary random process, the denominator is positive definite. The equation for the signal to interference ratio is now given by the ratio of quadratic forms SIR = - - - ' - wt M w (1. 25) Therefore the following theorem from Gantmacher [8] can be used. Theorem: Given Hermitian matrices C and D of order n x n with D positive definite, the equation 1£- >..nl = o has in general n distinct values for A (eigenvalues). Then: The largest characteristic value A n is the maximum of the ratio of two quadratic forms and is given by wt c w A n = max ~---------- and this maximum is assumed only for the eigenvector ~ associated with the characteristic value A which satisfies the following homogeneous n equation 11 c w= A. n D W ~ Applying this theorem to Equation 1.25, it is shown in Appendix A that a) The largest eigenvalue ~ of the linear combination ! -.AM is its only nonzero value and is given by = b) Thus ~ W M-l W -q -q The eigenvector corresponding to .AM is provides the maximum obtainable SIR for the array and ~M is the excitation required to achieve it. 1.5 The Maximum SIR for an ESA with Discrete Phases The solution given in the previous section is not valid when the values allowed for the components of Ware restricted to a discrete set. In particular, in an ESA the only available values are l ej21f(k-l)/L l _ _ b , k-1, .•. , L-2 ~ where b is fue number of phase shifter bits. Methods of continuous analysis cannot produce a result for this discrete problem. stances. An exhaustive search is impractical in most circum- To demonstrate this, the total computational count will now be derived. Note that there will be LN points to be considered where N is the number of array elements. The computational count for the 12 SIR is obtained as follows. From Equation 1.13 jwt w*l2 SIR = -q - (1.13) wtM w N + 1 multiplications are required to calculate the numerator, assuming that Wand W are N-dimensional. -q 2 N + N multiplications are performed in calculating the denominator, assuming that M is N X N. This results in the following number of complex multiplications required each time the SIR is calculated 2 N + 2N + 1 ~ 2 N (1. 26) This can be approximated by 4N 2 real multiplications. Searching for the optimum weight vector by exhaustion amounts to calculating the SIR for LN cases with a total real operational count of (1. 27) For example, consider a small array with 16 elements and 4-bit phase shifters. Then there are 4(16) 16 (16) real multiplications. 2 = 1.89 X 10 22 Assuming optimistically that one nanosecond is required for each multiplication, using the exhaustive search method would take 1.89 X 10 3600 X 24 22 X X 10 365 -9 = 600,000 years 13 to complete. An efficient search algorithm suitable for this problem is known as branch and bound. This method reduces the number of points in the search space and still leads to a proven solution. This procedure has been used to obtain the results in this thesis. 1.6 Thesis Outline The branch and bound algorithm is described in Chapter 2. Chapter 3 provides a description of the computer program used to implement the branch and bound algorithm and generate numerical results. A functional block diagram is presented as well as flowcharts for the main processing and subroutines. Included in the descriptions are the branching rule and the Trench-Zohar Toeplitz matrix inversion method. In addition, a program listing is provided in Appendix C. Chapter 4 contains the numerical results generated by the computer program. The branch and bound program has been used to evaluate four and eight element linear arrays with element spacing A/2. Results have been obtained for arrays with 2 to 4 bit phase shifters. An untapered array illumination is assumed for the unadapted pattern since sidelobe control in the unadapted pattern is not essential to demonstrate the adaptation capacity of the ESA. One source of interference was stationed in different angular positions and the nulling capacities of the arrays were determined. The thesis concludes with Chapter 5 in which the conclusions derived from this study are summarized and suggestions for future work are made. Chapter 2 BRANCH AND BOUND 2.1 Introduction to Branch and Bound The need for an efficient search algorithm suitable for obtaining the maximum signal to interference ratio (SIR) for a linear array with complex element weight vector ~ whose components are restricted to a discrete set was demonstrated in the previous chapter. It was shown that for a small 16 element array with 4 bit phase shifters, 600,000 years would be required to exhaustively search all cases on a digital computer. The branch and bound algorithm is an efficient search technique which reduces the number of points in the search space and still leads to a proven solution in a realistic length of time. The version used to obtain the results of this thesis is described by Cohen [3]. Cohen considers clutter suppression in relation to the determination of transmitter coherent pulse train amplitudes. That problem is cast in identical mathematics as the problem at hand. In the array problem N dimensional complex formulation was chosen over 2N dimensional real formulation. Complex formulation preserves the Toeplitz form of the covariance matrices allowing the use of faster inversion algorithms (Trench-Zohar [4], [5]). chosen. For this reason complex formulation was This differs from Cohen's implementation and brought about a different branching rule as well as the need to use mathematical results and extensions thereof from Spafford [6]. Spafford gives the optimum subvector of weights over the continuum when the rest of the weight vector has been constrained to the allowed discrete values. This 14 15 result is essential in the use of the branch and bound algorithm. 2.2 SIR Maximization for fue ESA with Discrete Phases In Chapter 1 it was shown that the equation for the signal to interference ratio for a fully adaptive array is given by SIR(W) == Jwt w*J ~ ~q 2 (2.1) wt M w For an ESA the elements of the weight vector W are constrained to the discrete set (2. 2) where b is the number of phase shifter bits. He wish to find the maximum signal to interference ratio for an ESA using the branch and bound algorithm. Mathematically this can be formulated as follows. Find: Max SIR(W), (2.3) w EA m with m == 1,2, ... , N For the fully adaptive array the weight vector W has elements which are unconstrained. in magnitude and phase. The ESA has weights that are constrained both The intermediate cases in which some elements of the weight vector are constrained is used in the branch and bound algorithm. The specific notation for the partition of the ~order weight vector W into its constrained and unconstrained parts is given by 16 where w~ = (W~ -c w~ -c = (wl, w2' ••• ' w~ = (w 9.+1' ... ' wN) -u W~) -u (2. 4) WQ, ) = constrained subvector of weights = unconstrained subvector of weights = level indicator ~ = 0, N = number of linear array elements 1, 2' ... ' N Further explanation of this notation will follow in the description of tree structures used in the branch and bound method. Equation 2.1 can now be restated as follows for the partially constrained weight vector l<~~t' w2-t) w*lz -q -u cw£t -c w£t) M (W£ ~u - -c ' SIR (W) = 2.3 (2.5) ~~) Tree Structure Equation 2.5 is used with the branch and bound algorithm for a variety of possible values for ~ and for each £, different weight values are assigned from the set A. Since any combination of 9- values and weight values may be needed, a complete enumeration of all possibilities is required. All such combinations can be visualized with the aid of a tree structure shown in Figure 2.1. Equation 2.4 that the level indicator Referring to Figure 2.1, £ =0 Q, It was mentioned in varies between 0 and N. refers to the top node of the tree. At this node the SIR is given for the totally unconstrained Nth order 17 NODE I =1 L • • • •• • =N SIR "'~---nr~~~==---.~~ \'!Y~ L Figure 2.1 Tree Structure for Branch and Bound 18 weight vector by Equation 2.1. level is denoted by ~ = 1. Continuing down the tree, the next The SIR is now a function of the partially 1 constrained weight vector w and is given below. 1 In this case the SIR (W , ~) is given by Equation 2.5. -c -u level the SIR is denoted by th At the k- ~) SIR (Wk -c ' -u where wk -c = (wl, w2' ..• ' wk) wk -u = (wk+l' · · ·' wN) The next level down is denoted by k+l and the following method is used st for constraining the (k+l)- element of w k+l. -c w k+l -c = (~ = (Wk -C wk+l = a2) = (Wk -c wk+l = aL) -c ' wk+l = al) or w k+l -C or w k+l -c Note that each node branches to L values for the next level in the tree. The unconstrained subvector changes in the following manner. At the kth level the vector is 19 = st Constraining the (k+l)-- element to ak, yields ~~+l Level ~=N (wk+ , ... , wN) of order N-k-1. 2 = denotes the bottom of the tree. The SIR only depends upon the fully constrained vector of weights and is denoted by SIR (vfJ) ~c In summary, at the top of the tree all of the elements in the weight vector are unconstrained. At each level down in the tree another element of the weight vector is constrained to a member of the discrete set A of allowable weight values. At the bottom of the tree, all of the elements of the weight vector are constrained to elements in the allowable discrete set. 2.4 A Fundamental Mathematical Relation The following relation is the central concept of the branch and bound method as presented here. It provides the justification for decreasing the number of points in the search space for the maximum Note that for two sets rl"CD' max f (y) > max f (y) ye:Q' ye:Q" This ;is due to the fact that fewer elements than (2.6) Q'. Q" is contained within Q' and fll! has SI~ 20 The inequality given in Equation 2.6 can be translated into tree notation as shown below. e:C SIR(WQ,+l max -c ' WQ,+l e:CN-Q,-1 N-Q, (2. 7) -u where CN-Q, = N-Q, dimensional complex space N-Q,-1 C = N-Q,-1 dimensional complex space and 2.5 cN-Q,-1 cc N-Q, Fathoming The SIR is evaluated at the bottom of the tree when all elements of the weight vector have been constrained to values from the discrete set. The SIR, a ratio of quadratic forms, takes on real non-negative values. Let SIR 1 = c1 , where ~ ) components are in A and c1 ~l is a vector such that all of its is a real constant. This constitutes a feasible solution to the problem stated in Equation 2.3. Note that c1 is actually a current lower bound (CLB) to the solution. Tracing the way down the tree, at a particular node the following inequality may hold. 1. 1 The specific equation used to calculate the maximum SIR for the partially constrained weight vector is derived analytically in Appendix B. Due to its complexity the equation and pertaining notation will not be presented here in order to avoid loss of continuity. 21 If this occurs, that node cannot lead to a feasible solution and it is declared fathomed. All succeeding nodes branching out below the fa- thorned node are automatically fathomed as well because of Equation 2.7. 2.6 Updating the Current Lower Bound In the event that a node is not fathomed during the search through the tree and the bottom of the tree is reached, the SIR is calculated. If then set SIR= c2 . This procedure is called updating the current lower bound for the maximum SIR. After this substitution occurs, the algo- rithm moves from the Nth level to level N-1. SIR for the next node at the ~level. It will then find the After all nodes at the~ level branching from the same node at level N-1 have been exhausted, the algorithm will move back to level N-2. It will then proceed to all nodes at level N-1 that branch from the same node at level N-2. I;t should be noted that when 1 < R. < N-1 the maximum SIR is calculated at each node using the equation in Appendix B, and it is this value which is compared to the current lower bound for fathoming. This process of tracing branches continues all the way back to the top of the tree. After all nodes have been exhausted in this manner, the current lower bound is the desired solution and the corresponding weight vee tor is W ~opt . Chapter 3 COMPUTER PROGRAM DESCRIPTION 3.1 Introduction to the Computer Program This chapter describes the computer program implementation of the equations given in Chapter 1 for the fully adaptive array and the branch and bound algorithm described in the previous chapter for the ESA. A functional block diagram for the computer program is given in Figure 3 .1. The program first calculates the noise and interference covariance matrix M, assuming that the jammer(s) power(s) and direction(s) are known and unchanging. It then calculates the quiescent or un- adapted weight vector W which is determined when the mainbeam pointing ~q direction is specified. With these two quantities the value for the maximum SIR for the fully adaptive array and the corresponding optimum weight vector are easily obtained from the equations given in Chapter 1. 2 jwt w*l = SIR max ~ ~g, wt MW (3 .1) where w= w ~opt = M- 1w ~q (3.2) The value for the maximum SIR is not used in the branch and bound algorithm since it represents the best performance for a fully adaptive array with weights having continuously varying magnitudes and phases. However, it is used to determine the relative performance of an ESA as 22 23 FULLY ADAPTIVE PROCESSING MWO I SIRCLB ESA PROCESSING PRE-BRANCH AND BOUND -·I SIRCLB -I BRANCH AND BOUND I A A FACT I APLOT Figure 3.1 Block Diagram for the Computer Program 24 the number of phase shifters bits increases. The ESA portion of the program begins with the third block in Figure 3.1. This part of the program uses a random number generator to select all of the discrete elements of the weight vector. Then the SIR is calculated using Equation 3.1 without using Equation 3.2. Many possible weight vectors are generated with the random number generator and the corresponding SIRs are compared to obtain the largest. value is stored as the current lower bound. This This process enables the branch and bound algorithm to converge more rapidly to the desired maximum SIR. The next block represents the branch and bound algorithm. theoretical concepts were introduced in Chapter 2. The A more detailed description follows in Section 3.4. The antenna patterns for the unadapted, fully adaptive and ESA are then plotted to determine jammer cancellation results. More accurate quantitative results are derived and tabulated from additional printed output. The description of the high level structure of the computer program was given above; the next sections will give detailed descriptions of the computer algorithms. 3.2 Fully Adaptive Array Processing The fully adaptive array processing includes two subroutines: MWQ which calculates the noise and interference covariance matrix M and the quiescent weight vector W and SIRCLB in which the maximum SIR for -q the fully adaptive array and the corresponding weight vector W t are -op determined. The following subsections will discuss each of these 25 routines in greater detail. 3.2.1 MWQ It was shown in Chapter 1 that the noise plus interference covariance matrix is given by +M --n M = M. ~ where M. represents the covariance matrix due ~ to interference alone and M denotes the covariance matrix for the --n noise alone, where with ~ = identity matrix of order k pn = thermal noise power in each and array element. The interference covariance matrix is given as ~= [ pi e j(k-£)(3.] l. (3 .3) 26 where pi = k, n N e.1 = interference power in each array element are the kth and Nnth c hanne1 , respect1ve · 1y direction of the interference For the case of multiple jammers the interference covariance matrix is given as a sum of terms similar to the right side of Equation 3.3. For the interference covariance in Equation 3.3 whenever m, where m is a fixed integer, we obtain identical elements. (k-~) = In each successive row, one column over to the right, the element value is repeated as shown in Figure 3.2 for N=4 elements. Thus all diagonals parallel to the main diagonal have identical elements. is called a Toeplitz matrix. This structure Note also that the elements on opposite sides of the main diagonal are complex conjugates. matrix is Hermitian and Toeplitz. If ~ Therefore, this = 1, the following formulation is. used to ohtain the elements in the first row of Equation 3.3. (3.4) Then the Toeplitz property is used to find the elements on the main diagonal and in the upper triangular portion of the matrix. Finally, th.e Hermitian property is used to fill the lower triangular portion. The total noise plus interference covariance matrix is now obtained by adding the noise power p n to the main diagonal elements of I 27 P. • 1 P. ~/3i P. ei213 i P. e-i/3 i pi ·1 P. ei/3 i P. ~ 2 13i P. e-i/3 i P.I • 1 P. ~/3i P. e-i/3 i pi. 1 I I I P. e-i 2 13 i I p .e I I -j3/3. 1 P. e-i 2 13 i I I I I p j8 i3/3. I I I Fi;;ure 3.2 Interference Covariance Matrix ' 28 the interference covariance matrix. The quiescent or unadapted weight vector which is the optimum compromise between resolution, gain and low sidelobes is calculated from the signal direction angle 8 as shown in the following equations: s = quiescent weight vector where the kth element is denoted by = ak e -j (k-1) B s and 2IId A sin e s where d = A= es = distance between elements wavelength desired signal direction angle For this study an untapered array illumination is assumed for the unadapted pattern since sidelobe control in the unadapted pattern is not essential to demonstrate the adaptation capacity of the fully adaptive array or the ESA. Therefore, =1 - - - ~------- 29 3.2.2 SIRCLB The maximum SIR for the fully adaptive array and the correspond- ing optimum weight vector are calculated using the following equations from Chapter 1. wt w wt w SIR max = -q -q - (3.5) = wt w M wt M w where w = w -opt (3 .6) = Substituting Equation 3.6 into Equation 3.5 we obtain the following formula for the maximum SIR. wt M- 1 w wt M- 1 w SIR max = -q - ~q -q - -q wt M-l M M-l W -q-q w = wt -q -opt 3.3 ESA Processing --Pre-Branch and Bound (SIRCLB) In addition to obtaining the maximum SIR and the corresponding optimum weight vector for the fully adaptive array, the SIRCLB routine represented in Figure 3.3, performs another function, for the ESA which is described next. The largest current lower bound for the maximum SIR for an ESA 30 Wclb = 0.0 I= 0 1=1+1 WEIGHT w wtw• 12 l SIR=---- -q 'e{ t M '!Y N 1/lclb =SIR 't!clb = '!t, N Figure 3.3 Flowchart for SIRCLB ESA Processing 31 where all weights are constrained to discrete complex values is calculated in SIRCLB. This situation corresponds to a feasible solu- tion for the maximum SIR at the bottom of the tree for an ESA as described in Chapter 2. The discrete weight vector is calculated in a subroutine named WEIGHT which is called from SIRCLB. The following processing occurs in the WEIGHT subroutine as shown in Figure 3.4. First a sequence of N random numbers is generated and written into a vector R of order N where R = Then each element of the weight vector W is assigned a value based on the entries in R using the following inequality. < r If - Ll <- k- 2 b where L 1 = 1,2, ... , L b = number of phase shifter bits, and rk = th ~ = zb entry in R E [0,1) (3. 7) 32 RANDOM NUMBER GENERATOR k = k+1 N Fi.:;ure 3.4 Flowchart for Weight Subroutine, p. 1 33 N y RETURN Figure 3.4 Flowchart for Weight Subroutine, p. 2 f 34 th Then the k-- element in the weight vector w is given by wk = a k ejSk (3.8) where ak = 1 = sk -2'TT(1 ) 1 2b Equation 3.7 assigns a value to 1 varies from 0 to 1. 1 based on the value of rk, which This value for 1 1 is used in Equation 3.8 to determine the discrete phase for the weight vector. After the N elements of W have been chosen in this manner, the subroutine returns to SIRC1B. After the program has returned to ·the SIRC1B routine the SIR is calculated using the weight vector W computed in the WEIGHT subroutine and the following equation: SIR = lwt w*l ~ ,_.q wt M 2 w The SIR is always positive since the denominator is positive definite. This value is compared to the value for the current lower bound for the maximum SIR, ~clb' which is initialized to 0. for SIR is stored in ~clb and the weight If SIR vector~ >~clb' the value is stored in ~clb" The SIRC1B routine processes 2,000 weight vectors in this manner. The number 2,000 was selected because this should constitute a sufficient statistical sample. After this processing is concluded, control ' 35 returns to the MAIN program where the branch and bound logic is implemented. 3.4 Branch and Bound: Discrete Weights and Assignment Priority The flowchart for the computer program implementing the branch and bound concepts described in Chapter 2 is shown in Figure 3.5. The largest current lower bound for the maximum SIR is initialized to ~clb which is an output from the SIRCLB subroutine. The branch and bound program next computes the allowed discrete set of weight values: A -mr =lamrk : amrk 1, 2, ... , MDIM where NBITS is the number of phase shifter bits. The index I in the next block of the flowchart corresponds to the level indicator, L, plus 1. The reason for this convention is to avoid an illegal FORTRAN subscript of 0 (level 0, or the top of the tree). NPASS keeps track, at a given level L, of which tree branch of those emanating from one vertex, the program is processing. shown in Figure 2.1, there are~ MDL~= 2NBITS branches originating from a node (vertex) at the previous level. assigned to NPASS are 1, 2, •.. , MDIM. ~s Therefore, the values Since the level indicator Lis initialized to 0, the initial value for NML = N-1 is N. The program now calls two subroutines, ANGLE and RlR2 prior to constraining the elements in the weight vector to members of the allowable discrete set. the first level, L This sequence is performed one level at a time. At = 1, the first element in the weight vector is con- strained; at the second level, the second weight is constrained, etc. 36 INITIALIZATION BRANCH AND BOUND I SIRclb= VcJb I amr = k ej ((k-1) 27T/2NBITS) l I= 1 NPASS = 1 1 I NML=N I ANGLE I RIR2 I lJ Figure 3.5 Flowchart for Branch and Bound, p. 1 37 BRANCH AND BOUND LOOP PROCESSING 1=1+1 .NPASS = 1 1 L = 1- 1 NML = N -L YES RIR2 REORDR ANGLE SIMA X Figure 3.5 Flowchart for Branch and Bound, p. 2 38 BOTTOM OF THE TREE NSIMAX YES SIR clb = +max W -clb = B -cone Figure 3.5 Flowchart for Branch and Bound, p. 3 39 FATHOM VERTEX NPASSL = NPASSL + 1 NPASSL = 1 I= I- 1 L = L- 1 NML = NML + 1 RHIGH Figure 3.5 Flowchart for Branch and Bound, p. 4 40 In order to select the constrained discrete value, the program calls subroutine ANGLE which computes the angle, a, in the complex plane corresponding to the first element in the optimum weight vector W -opt using the following formula: a = tan This value for a a is now used in subroutine RlR2 to select the allowable discrete value for the first element in the weight vector. vectors, ~IM ~land ~ where MDL~ 2 , are defined in RlR2. = 2NBITS ~ , 3 Related vectors Two ~ , 4 •.. , are determined in another subroutine, RHIGH, which is described later. The method of selecting the discrete weight value is demonstrated with the aid of Figure 3.6, for NBITS = 3. Noting that the vector corresponding to angle a intersects the unit circle at a point denoted by p, which is closest to the discrete value a , the program will select 2 as the first entry in 2 closest value top is a . 1 point a in ~ 3 1. Therefore, the first entry in is the third closest value to p. 3 is 3. ~ The first entry in ~ , ~ 4 5 , ... , The second ~ 2 is 1. The Therefore, the first entry ~IM is determined in the same manner . In summary, RlR2 produces the vectors ~ and ~ 2 which contain the first and second closest allowable discrete values to be assigned to a given weight, and they are indexed by level L in the branch and bound tree. 3.5 Branch and Bound: Loop Processing As described in Chapter 2, the unconstrained portion of theweight 41 IM Figure 3.6 Complex Plane Representation of Discrete Set (NBITS =3) 42 vector is of order N-L, while the constrained portion is of order L. L will take on values from 0 to N, where N is the number of elements in the weight vector. Conversely, NML = N-L takes on values from N down to 0. The vector B represents the constrained subvector of weights, ~cone which is indexed by L. either ~ or ~ The discrete value vector, A ~mr 2 , depending on an indicator. ~l and ~ , is indexed by 2 are indexed by level L. I is compared to to determine if the processing has reached N+~ the bottom of the tree. If not, subroutine REORDR is called. This subroutine partitions all of the vectors and the noise and interference covariance matrix M which are used to calculate the maximum SIR for the partially cons trained weight vee tor, as shown in Appendix B. The following vectors are partitioned and reordered in this context: ao ~0 = = ao a ~ w ql 1 L w qL L (N-L) = "unconstrained" portion L = at "constrained" portion w qL+l tl = = N-L w qN N-L 43 where w ql w -q = = w qN bo B quiescent weight vector N b 1 = conc 1 = = -o bo b L conc L 1 constrained subvector of weights L The corresponding partition of the covariance matrix also occurs in REORDR, as follows: X 1 M L = [fa] X (N-L) (N-L) (N-1) x L X (N-L) NxN where M is the "unconstrained" submatrix. -c Note that only the constrained subvector of weights being constrained to discrete values. and ~ ~O The partitioning of is actually ~q is performed to be compatible in dimensionality with Similarly the partition of the ~ into ~O ~o· matrix is also done for this reason. The maximum SIR for the partially constrained weight vector is calculated in subroutine SIMAX. The following intermediate quantities are used in the computer program to evaluate the equations derived in Appendix B. 44 u = v = -1 M ~ -c M -l C .:::.0 ~0 -c NUMER = Bt DENOM = t ~0 ~0 Ci.l = -0 (Y B - -0 - ct V) =o - ~ - V' ~T NUMER DENOM Finally, the maximum SIR is calculated using the next equation. = ~ ~ + (DENOM)*/a1 SIMAX then returns to the MAIN routine, where the value for is compared to SIRclb' the current lower bound for the SIR. is greater than or equal to SIRclb' the node is not fathomed. If~ ~ max max As shown in Chapter 2, Section 2.3, only the first element in the unconstrained subvector of weights at level L = k is constrained to the (k+l)~ element in the constrained weight subvector at level L = k+l. Using the outputs from SIMAX, the first element in the optimum constrained weight vector can be obtained using the following formula: b opt = 1 alul - vl where ul is the first element in u vl is the first element in v 45 The derivation of the equation for b The value for b opt opt is given in Appendix B. 1 is used, in the branch and bound loop, as the input 1 to the ANGLE subroutine. The output, a, from ANGLE constitutes the input to RlR2. When the ~ level is reached (the bottom row of the tree) and all elements in the weight vector have been constrained to discrete values, the current lower bound for the SIR is calculated using a different formula from that in subroutine SIMAX. This is accomplished in a subroutine called NSIMAX, using the following familiar equation: 1/Jmax = where W = B -cone = W = quiescent M = noise -q vector of constrained weights weight vector and interference covariance matrix After returning to the MAIN program, a comparison is made to determine if the current lower bound for the SIR should be updated to this value for ,~.'+'max' If so, the corresponding weight vector ~clb is set to B -cone A vertex is considered to be fathomed if it is at the bottom of the tree or when the maximum SIR is less than the current lower bound. The following processing occurs. '. ·... 46 As described earlier, the values for NPASS are 1, 2, ..• , MDIM 2NBITS. NPASS (NPASS Therefore, at level L, where Lis the index for NPASS, is incremented by 1. 1 = 1 If some branches are yet to be processed < MDIM+l) a check is made to determine if NPASS 1 is equal to 2. If not, subroutine RHIGH is used to obtain the higher priority vectors 3 ~ , ~ 4 , ••• , ~IM· It should be noted that the following implementa- tion is used in RHIGH to save storage. ~3' ~5' ~7, ... , R2 +1 + Rl ···- n - ~2' ~4' ~6' ... , ~2n where n is an integer. +R -2 The fathoming loop closes at the calculation of B for the same level L. -cone If no branches are left to be processed (NPASS logic will back up the tree to the previous level. 1 = MDIM+l), the This is manifested in the program by resetting NPASS at level L to 1, decrementing L by 1, and incrementing NML by 1. If L is not zero, the top of the tree has not yet been reached and NPASS is incremented at level L so that a search of the other branches at that level can continue. When L is zero, the top of the tree has been reached and the value for SIRclb and ~clb are the solutions for the maximum SIR and the corresponding discrete optimum weight vector, respectively. These values are output at the printer and the antenna patterns for W lb' W t (the continuous -c -op optimum weight vector), and W (the quiescent weight vector) are -q plotted. 47 3.6 ARFACT and APLOT The two routines ARFACT and APLOT compute the array factor and plot the resulting antenna pattern, respectively. Since they are used routinely for antenna applications, the programming details will not be presented here. 3. 7 Trench-Zohar Complex Toeplitz Hatrix Inversion Subroutine A subroutine called ZORN * was used to perform all of the complex Toeplitz matrix inversions in this program. It takes ~N 2 operations to invert an N-dimensional complex Toeplitz matrix as compared to the ~N 3 3.8 operations required by the Amdahl SCILIB routine. Program Listing A complete listing of the program appears in Appendix C. * ZORN was adapted from the real Toeplitz Hatrix Inversion Routine written by Captain Alan Paige, U.S. Air Force. Chapter 4 RESULTS 4.1 Introduction The branch and bound algorithm was tested with four and eight element arrays, to determine correctness and speed of convergence. Numerical results were obtained to determine quantitative jammer cancellation performance. Some of the cases converged very slowly and the runs were deliberately interrupted. obtained were, in many cases, unsatisfactory. The intermediate results These problems are further discussed at the end of the chapter. An untapered array illumination is assumed for the unadapted pattern since sidelobe control in the unadapted pattern is not essential to demonstrate the adaptation capacity of the ESA. 4.2 Results for Four Element Arrays The branch and bound method has been tested for correctness by comparing its results with those obtained using an exhaustive search through all possible phase settings. A four element linear array with element spacing of A/2 and three bit phase shifters has been used for this purpose. The results of the tests showed that the identical dis- crete optimal weight vectors were output from the branch and bound and the exhaustive search. The increased speed of convergence due to fathoming did not shorten the runs for these smaller cases, since the overhead costs were significantly higher for the branch and bound program. Results pertaining to the following special cases are shown. 48 49 Signal angle is always 40° from broadside. In each of the three configurations considered there is only one interfering signal 20 dB above array element thermal noise arriving at 20° (main beam) or -10° (first sidelobe) or -50° (second sidelobe) from broadside. Results have been obtained for arrays with 2, 3, and 4 bit phase shifters. Antenna patterns are given in Figures 4.1, 4.2 and 4.3 for all jammer configurations, for the array with 3 bit phase shifters. One pattern for the array with 4 bit phase shifters is shown in Figure 4.4. patterns: In each of Figures 4.1, 4.2 and 4.3, there are three (A) unadapted array, (B) adapted discrete or ESA (discrete adjustable phases, fixed gains) and (C) adapted continuous (continuously varying phases and magnitudes) . signal and jammer are indicated by arrows. The positions of We note invariably that both the ESA and the adaptive array are reducing the jamming severity by placing nulls in the vicinity of the jammer. As expected, the performance of (B) is worse than that of (C) as evidenced by less accurate placement of the jammer nulls, higher sidelobes elsewhere and more distortions in the main beam. Nevertheless, the performance of (B) is good enough to be of practical importance. In Figure 4.4, the same jamming case is covered as in Figure 4.1; the patterns displayed are type (C), type (B) 3 bit phase shifters and type (B) 4 bit phase shifters. over 3 bits. Note the improved performance achieved with 4 bits The array phase settings for the 4 bit phase shifters were obtained as an intermediate result in an incomplete branch and bound run. an increase More quantitative results of performance improvement with in the number of bits is given in Table 4 .1. One would 50 0 0 rl ('.1 I I :s; I (S:G) I!IVS ~I Cl Li'l 0 1.0 I I 0 r--.. 1 51 f- = I'<"\ 0 00 = f- ::;;: <:t: U) lJ..l ... L1J 0 I.D c:o = ..:...J LLJ = U) f- = u.. (/') z = LLJ ::=:: ""-- < -, = f- :;::: 0 ........ U) Iii V') = §. a; 0 N I 52 AOf~PTINCi fiGURE q,4 COMPfi.RISON OF 0 J~!·if·1ER PATTERNS \·llTil ff:GT' M-r-~-..., JS~ WLili 3 JTmPJI.~SI_UI!Hl~-~SfiTV};-_~_Bl l'"·'-·Jf---,;---..-DTT,~~;,:;•jl'1 tti'!lj:!i :nj'·l' 1,1-1: ,: 1·\litl'·l: :.~-,,!., '_,I 1. ·j1·-l·•·l· 11 1 '1-- I IIrll.ltr,\ 1!' 1!' ~~·I I . ,. 1 •. ~~ ~, ' 1 '!·''t.11-i - , I I •,, I· tf I' 1 ·f .,.g::::"';'-~·-·· I'I"""''T\i'''Ji'li-t:·· I' 1 •j ... , I ' ~--, ••,_-,-.... ' . ""'I:·. I [ l· I ' ' I I r.- ' I'I;,II··~~JN\,1ER, ' ' •,t_-, .: ': jt,' l r- 1: IN MAIN BEAM, Pjli\~F SfllfliRSa·---- --·r..-~ .. ,'· :.r!. ····-1 "'11-~-l;j · ·l""i" :·;--·'· I ·1-·· -~1 ..... ~.. !1.,.. l --~--.Q~ir.L ....,.__•1 l''f."l•~-··---,-1 1 1-, ,., . t·'Jl ... l- : I I'·,~ ' :1.14 ' . , .I ,.1\ f.." i I .. 1"·1-tl.'I''-·-rI"'1 i'"l·'-·-, ·.. r,-,:.,'I , ; I I , ( ,,•, ·~· (, . ' : ; · ' " , ~ : 'it-...: · .-~.· . I,-· ' . ·I• ' r·l•r·r , -Lj --:· • 1-'- l1 : ! · ' .. -, -nt: i.i,. ,· ··· •i·-' ··,, ·'·• ,·, ·:"-/ ·-a [•II, I .. ·-\.,~- ,-~--'-t-•~1 ~...':..:::_L'-•· - 10 r,-~---~--~·t - l -----l --~----~-- -·r'--'-~........,..., ~-. I i \ f ~- t t '- - - -l t '-1-i - -·---c-:-TT"-, ·--- ~ -1'1'.,.,!.:1, '\j .,:\ :1:' i: Iii'I III I: i I\t ~~I, 1-: ... ·l J·t· ,-j~-ll-1 •t-rl-i-j-L:~ I 'II . ·i!•t,I I . T,. 11·1' ':·Nil' ..... \!' I ') ' ''I I' 1·1.1I .~,·J: II,,I I 1·1: t· I'll., 1-ljlli 1-1 -1-J-,.;. ·-1I !-;:l.t.;.J H· \ !I ~IIJ jill .,I\ITI·t!-11 -I 'II '7"' I II I A ' ~ 'II' I.I.l'1' ' .-f ''\II I I j r' ' I' I : ' I . t· ..! ~.• 1 :,-·~···-···--,il·:·, \ :·,· 1 I l:r· I·· !,\. ·:!:1! !1·••-l-tll rlijl\/.' .• ... ,;~ , 1 !- [ l ~ I ' , I II II ' ' I ~---- t + ~ ~-. 1.. I . -,-1,:;~ • . .,. j 0 rT I •~,-·,·~... · ' • I ' ',, ... l ! I j I ~I \ I ·- I '·I I I I ' • 1 r-+ "I 1 -~ I 1 • r I 1 ' 1 .. I l I I ~ I 1 1 / ' -x- ------- -- ---m- - - -· • I I'!' 1 ' I I II'*' ' r' -• I" ·I. 1 •• - \. l \ I'' lm. I 1-J-,.~ I I ' I ; ' 1' I \ \t .. L I I . ' ' : ' . . . .. ,_ - .,_, f' 'I - I'- .._I , ' ' J I ' J r '-'I_ - :\· \' L, I I I I I 1 ' 1 r ,- ' 1 ·-' 1 I ' , 1·1 ' .. :--~·-,--: • • ' 1 • •• -?0 ~------,-.,-----------r,- ·----\~-~ --,-·--' .... I -. ; -. • I \-i I . -- __.....,_I , ' -·' --~-~ i ' ' ~~ ,---~--- ---~ 1 '. ~-I ' '-,1' ,--::I[. 'I;' \i j t , .. ;:, •·111'1: ·t:i'l i f,,,,,,SI 1'1/\L'· '·J "--]1, ij.: ,. . --,- .. !.I '.,;. I' r I : I ' , ' '•..t. ' 1., ! I II , , ' ' r II ,, ' I , . . ' . , ,. ~l . ' I I ,·, ; ; ; r· : 1 , .. ,' : ., ' . 1 I'' - t! II' I ' ' II I • I ' I I' I ' ' I -~-I1-1 I -I , 1 ! · 1 I , - ·! l I , 1 , I ~ 1t· . '~ i ,.: 'i ,., ' I; 1"1' ''I' i I . I' I i '). r I'! x·1 I I' I,I tl ! I' l I·-til; t I I' :·f I II- r ·J·' I f ,·! ;i·i-· '.,I; ·-·I-:' ~- r • '· '_, • I I I 1I I ill1 -·{' 1l •. -·--J.·--1-1 1 1 • 11 • f-, ---r 1 · 1 1 • r r I 1• 1-I • , ' 1 ! , 1• : • " i !' I 1 : 'J • r·< t • . ; - 30 1"1-1-' ~~ +++-i~·T' _._ - --'-+-'-+---~- wr-T- . ,I -'+ , ....u... ..----n-f-+--1·;-. ·tt:·;-.L~·r;· '-,r-t·-:·',, l,.ti·f·;[/, .. 1\''r\· t,' t' II'' 'l''lll 1.,!, 1 ' , --· ,-,-, --~---<-;---r-.--· l 1 ,., ' I ;·'·t I' .. 1 I ' • I IX . l ' I' I I ' -1' I ·I· I 'I' x---,' " I I .-~ --t--I.: I ' I -. I 1-t I • I I - • , '1- . ' 1- I [ ' I j-l·;;i:l11j·\,l\lit·nl~, /~· 1-1 1 'l . 1- •- I I ; I , ' ' I· - I '·- ' 1 r • I I : : - I ' I ' I I I I ' I l 1- 1 t I \ , I ' I I I ' I ., ' I I : I i· I :-1 - l_t I I 1 : I 1 I I - 1- I ' - ' - ' - I . ·-; • - • -I • 1 , . _, • -, : • r ; • -· J.. , I 1 • ·\:· 11 \-~-1! .''\1 I 1-1-1 r1 i-t -':.,.li•·: tr~··I·L.t'·'jki··J•'J I :"II· ··1· /:\1,,1 \ J· -,'·1.-'t I . , I J 1 1 I ' I ,I~:~~~~~· r[-~~~. 1·• 1 ~1-tl-hr: 1 1~,.1-•l , ,. I ' 1 I I ,~.X.ADAPTEDCOi·,TH.UQ.JS 1 1:-f.:.:::;·:-:.!. ;.(;~~~ ·T'i'iV'-'=I'':r... l~~ ;:rlt-lt'l :lmt·l-~~f-[i-:!·~~·Jffil-1·-;~1--: ~·~;~·Jiie· DISCRET·jt.:·:~~r;.:· --r · · I··: ! I -40 ,·1·11" "T/"1 ~ ~~-~·-;JI ;-,:' l'lfiiTl-;.. : !'] l-11-, .-n l:j~--:.:1\ 1: II'littl~-I i~'')I'll· ~-~-1-1"· I!,,1111~''+jl• IIi ''It' 1j ~-- r;+~-;:,l;-; 1 ·i--t--·;·· j··t-:··-it·--· -·lrlj~·t··- 1 ..J 1· 1 - ,. • ' ,.-~-~~-·-·~:r1 1 1. • - • 1 I'T''•. 1: i l-1' I I ' I I -1 r:-r ~-f·!~·'111i(-'·ijil-'\:. H+±+-1: :~ ~-~~ -l 1 t [i·j --H- 1 I ' I ,_\ h- • II- [Hi tl '1'1 1 lt-t·•!j-~! 1 .,. I ,. ' . ~ I ·1! t , \I' · \· 1 ' I· -1 - ·It, 1 1 I ,AnAP.TED IJ , 1, • , • • ,q .BITS' I:' --t-'- •' ... , • 1 1 1tml:.,:~ I lj_l' I -r-1' r·-1--M - , t.:·· .11·1•1.:. l•jl·j t1 • •• • ' ·i i' 'I ., .. r , .. , ,l,·-~1-·· · 1,.t-· ~ ,.l.!' I"'- _,...;..AD~PTED'D'ISCRCT'tl-r•·;! ·i i--:i 11-f; '3 ..BlT I -~·;r-111'1 I ~-\'11'1·\''' '\·,-·-~ I · ' 1'·1 '!" · . if"! ...I · · · I'I I.,• ·1·1 I I ' · I 1 1 r 1- ;1 -- \ ' 1· ·r~~-· ,-~1 ·t-:··-~r .. - · --f'r f ~ hl\-~i'-, 111,---·,l j·ll1'-:!-\, ·I ·I ., •I \! 1--1--H-·•-r··; f-,Jri-L ,, «' I·J ,·!-,-'1·1· ,,.!! 1'1 r, '-1)1'. .,_t_r __ ,~,:-1 0 ' '· r-11 · ' I I I · · •·• ·1 I T,\-•-1,--r+:, I ' i ,- ·' 1 .. · r-· • 'I ;: .. :,;_,,,,1,:.1,,1:· ··Ill·,;·1-:1:1 ,-1-J-IjJ ..I I, \,l,i'I 1 • 1i!·f:[;; ::·'·1--i-1-1; -50 --.., -~-j-t""----- -,.--........... ~~- .... ..J-:.-,n-- · --,..:...1.-1--~n- --n~r-r ~ ~~r;1· ~-___,...!...-'--r -------- t-1~·---r·),. 1-~-·:·:;--!L,i; . ,,;1' lj'r·li\: li!l. /r l· · 1111': ·~~ . ~~,, , '-i' • ·1' ,_,,·r .. 1.-:-,: 1 '1"1'1 .. ,. ' , ...... , I 'i-· .. . 1 --,1·.' I j , l ·'I I, I• .I ·I I .I 1~ .... ,~,1 :-,ri-1Tii1 ·.1 1·r'jl: 1'-\jrl· i Jl'i•IL 1 j )1\'-l i ~~ \!11 1\ 11'1·1• j 1-1· 1 !-1- .. t ' i-i- 1-,i\;-,- •-1 '"I;~--· . I I I . ' I r r 1 -- f I I t .. I I I I II· I - l I· . i t- : 11 !-l--t.-1 1·1 !·r t 1 i · -trl-,r,lll --1 ~~-II[' 1 I '-1·1 ~~~~~~~~ I'' t·lr· ·11·11 ''1·111 il-1-1 1 I I jl 1,--·(, -\J·J-~ ,_~.,,. -60 llifjf."r-1~·r-tl_Lr _.: ··r ~~t*~_:_ 1! ~- +hj_.:_r~ I ~~ ~_j__ l-_1___1-r' - ·_~J---~~L _l_l_ II !-l i·l- ·-i-t- -''-~ -~~_::.~-' •·· · ··' i 1.: · '· 1'1: · 1' lffi1 I • l•t 1 :1r II I,., i 1-' I r I:' 1''1 I··' 1' i I .1''·1 i1IT' ·-- 1 · ··-· · 1rl: .. ~r·:.... ,i:d·l!···l·l---!1]·I l .1t '' J!f•it ,J·j it-1: 1·j 1-:lli· 'r-,.. ,,,1i ··1 ,J.,: ··1'• 1 ., 1-1-I.J- •• I I 1, • • I I -I t ·I ' '-1 ~ 1 ' I '"!- ' 1 1,1,1 ::·-~r~,,~·,J 1 1-fl·· .l .. , .. ,li·I .L,J.rilll·-1'-:r!-.!, -ll·t ~ ~ 1!,.: ·-~-~'tl, 1 ~·-1,:11,-1. , 1 .,-i,-1-1.;.. l- -;-• 1-t·l.,- ,. I ·1---1 i I I f I r :' If I . I ' I ' I i ' I I II I I . ' I . I I I I 1 I I' I "I' I I ' I' I.,.,. '•r\'• I,-,'1 ll,_ I I II! I. II.,t1.. ,,I l·rl'j'll' ,,.I -j·;;-,-h ... 1-i-r r· ,1--<t,l·•: I ' It ·-r II ,-'j"l·'-1: j! \·-}I-1"1 ·t, t-- 1\II' ~I ' -'''l'i·•·\ I t\ I" I ,--"I ~-I I"" -·-··rt•L•,t .,~--..!1'1'' I '11'-1'' 1·1-ltl·ll J.1' · 1 ,\\·'·1-t.,.t.,l•lj·•tt·itl.l ·!·1+:-,,-, -70 tLtlJ::!ili_ ' I I ' 't -•·r I· 1 ' . t .. I --r' i 1'1 ' I I_ I ,• . ::·ill,·; I_! tr r.-,·j t . ···-,-,' ... , ..9 -- ~'[·\t'l-r·i I· " ·-·-· ' ;r•'·r•l · r· I •\-...:· ·!-t+[!'~-1,-H. : , • .. 1-! ·~·,: 1 • '1 • • •- L -I 1 1 1 1 I Jt~.ll PJ ~ -l~illfJ~1+ 1 t-~ ' ' l lt I 1 1 11I'' 1-.- - -60 J.,,.A .-r li·lt:lljJ-111 1 -30 I t -40 -20 8<DEGRffS) 0 20 I 1 t-d•-'rl- :f'' · rt r:--·' I''. .. ~~ 1JJ/I' II ~ :,!:II.! , r~'~i·l·l{j~~l I I ... I' !:lj'j·l·l''!'i,-jl:'![' I; 1 t1 1 I,jllll,l,-~ -H tt: i:c:; 11 •lr:'l'II 40 I 1 I ' 60 I ' ' I 1- I Irt 80 LJ1 w TABLE 4.1 PERFORMANCE IMPROVEMENT IN SIR CDB) WITH INCREASING NUMBER OF PHASE SHIFTER BITS 8·j 2 BITS 3 BITS 4 BITS ADAPTIVE ESA ESA ESA ARRAY i I " r:oo L -5.71 -1.24 3 ,Lb '"',.. 4 ...,_ -10° -5.78 1.83 3.81 5.77 -50° 4.63 4.63 4.63 5.89 .ub ln .(:-- 55 expect the ESA performance (in the signal to interference plus noise sense) to improve as the number of bits increases without reaching the performance of the adaptive array. The behavior displayed by the data in Table 4.1 indicates that there is no definite number of bits where a performance plateau is reached. The ability of the ESA to combat interference is quantitatively described in Tables 4.2, 4.3 and 4.4. In Table 4.2, the signal to interference plus noise ratio is compared for a number of situations. It degrades by 10 to 20 dB for the fixed array as the environment changes from benign to the presence of one jammer. Allowing discrete phase-only adaptation recovers 8 to 17 of the formerly lost decibels. Finally, the additional margin required to meet the performance of the adaptive array is of the order of only 1.5 to 6 decibels. In Tables 4.3 and 4.4, information is given with regard to the ability of the ESA and the adaptive array to reject interference while still accepting the desired signal without appreciable degradation. The adaptive array reduces the jammer contribu- tion to the point where operation is thermal noise limited with the degradation in SNR resulting primarily from the slightly reduced desirable signal reception concomitant with adaptation. The ESA helps reduce the interference by about 20 dB which constitutes a very valuable contribution, but as a net effect, the residual degradation in SNR comes about both from lesser reception of the desired signal as well as jammer noise. 4.3 Results for Eight Element Arrays Many cases were tested using eight element arrays with four bit phase shifters. However, they could not reasonably be verified using IMPROVEMENT IN SIR <DB) DUE TO ADAPTATION IN THE PRESENCE OF ONE JAMMER TABLE 4.2 NO JAMMER O·J ~ ONE JAMMER, SIGNAL AT 40° ADAPTING PHft.SES ONLY, UN ADAPTING ARRAY UN ADAPTING ARRAY --------1...; ADAPTING PHASES AND GAINS, 3 BIT PHASE SHIFTERS CONTINUOUS I 20° 6.02 -14.38 -1.24 4.65 -10° 6.02 - 7.74 1.83 5.77 -50° 6.02 -4.93 -- ~---- 5.89 4.63 --·-----~----~-----~-- --------------- ~--~--------~--- INCOMPLETE SEARCH FOR 4 BIT PHASE SHIFTERS 1 20° 1 6.02 1 -14.38 1 3.26 1 1 4.65 ... \.n 0\ TABLE 4.3 JAMMER REJECTION GIVEN AS ADIFFERENCE BETWEEN ADAPTED AND QUIESCENT PATTERNS <DB) 1--4 ONE JAMMER, SIGNAL AT 40° to1 I Oj ADAPTING, PHASES ONLY, ADAPTING, PHASES AND GAINS, 3 BIT PHASE SHIFTERS CONTINUOUS 20° 19.87 -10° 22.34 52.09 -500 18.36 52.35 ' 39.20 . INCOMPLETE SEARCH FOR 4 BIT PHASE SHIFTERS 1 20° 1 3Q.48 1 39.20 1 Ln ....... TABLE 4.4 SIGNAL DEGRADATION GIVEN AS ADIFFERENCE BETWEEN ADAPTED AND QUIESCENT PATTERNS CDB) 1... (}· J ONE JAMMER, SIGNAL AT 40° .. ADAPTING, PHASES ONLY, ADAPTING, PHASES AND GAINS, 3 BIT PHASE SHIFTERS CONTINUOUS I i iI I I ! I 20° 4.00 2.75 -jQO 3.65 0.51 -50° 0.72 0.25 - -- -· ~------- - i -----------~----~--~-- INCOMPLETE SEARCH FOR 4 BIT PHASE SHIFTERS 1 20° [------~.~~- ---I 2.75 J • lJ1 00 59 an exhaustive search. Noting that there are LN points to be con- sidered (see Section 1.5), where N is the number of array elements and N=8, b=4, the search space consists of 16 8 points and for N2 =64 multiplications and additions per point, the computation would last twenty-four days on an ~~AHL 470/V7. Therefore, program validation consisted of running many cases while varying the jammer position and power. The antenna patterns were used to obtain quantitative results on the jammer cancellation performance. Convergence time was also measured during this testing procedure. It is important to note that none of these cases completely finished executing. The results were obtained by using the program in an interactive mode in which the most recent value for the SIR current lower bound (CLB) was saved and used to initialize the input for the next run. In this manner, acceptable adaptation performance was achieved with the final value for the CLB very close to the maximum SIR for continuously varying weights (adapted continuous) • A representative 8 element, 4 bit phase shifter case is shown in Figure 4.5. The adapting ESA pattern was obtained in an incomplete branch 'and bound run. Note, however, that the CLB (PSICLB = 7.65) is very close to the fully adaptive array optimum SIR (SIRMAX = 7.87). Both the ESA and fue adaptive array are reducing the jamming severity by placing nulls in the vicinity of the jammer. 4.4 Problems with Algorithm When running many four element cases with more than four bits in the phase shifters, convergence problems occurred. For five bits the program converged when used in the interactive fashion described in FIGURE 4.5 PATTERNS FOR EIGHT ELEMENT LINEAR ARRAY 0 C(l '"d - 30 2 :7i. -40 CJ b=4 JAMMER -10 -20 N-8 K .\ '.\ \\ ~1 1 r r, 1I l~,~\ 'i/,~~. II!' fl: ~ f 1X "'iH1/ ,' ,, !! 1': 1 1: i) 1 1 j 1 ~ . I I ~ I n ' II 1 ADAPTED DISCRETE -50 II -60 --- UNADAPTED Sl RI\~AX = 7.87 PSICLB = 7.65 \II• ~ -70~~--~~~~--~~--~~~~~~ -100 -80 -60 -40 -20 0 20 40 60 80 100 THETA (DEGREES) 0\ 0 61 the previous section. Results for the six element cases with three bits were also obtained in this manner. However, when 8 element arrays with 4 bit phase shifters were used, none of the cases completely finished executing. Using the program interactively led to results that were very close to the adapted continuous maximum SIR. In order to enable the program to start converging, the pre-ESA processing used for the smaller cases to select the current lower bound (CLB) for the SIR and the corresponding optimum weight vector was suspended. This was done because no advantage was perceived. Final judgment on this point, however, is reserved for later. Instead, the CLB for the SIR was initialized externally to a value smaller than the adapted continuous maximum SIR. of success. These runs had varying degrees Sometimes the program would update the CLB several times before reaching a plateau where the same value for the CLB would repeat. This type of behavior resulted in the antenna pattern shown in Figure 4.5. On the other hand, there were many cases where the input value for the CLB was~ updated after several minutes. In these cases, another guess was made for the initial value, and the program was rerun. The algorithm appears to be extremely sensitive to the initial value. The explanation for this behavior is that the SIR surface in WECN is rather flat; therefore a long computer run is "' required for convergence if the initial value is not close enough, from below, to the discrete solution. This amounts, practically, to guess- ing the maximum SIR for the discrete weights. The program is still valuable since it produces the required weights and provides performance verification. Results for larger cases (N=l6, b=4 and N=27, b=5) 62 were unsatisfactory; the SIRclb at the end of incomplete runs was small compared to the adapted continuous array performance. More work will be done to improve the program so that these larger cases can be studied. in the next chapter. Some specific ideas will be discussed Chapter 5 CONCLUSIONS AND FUTURE WORK A method has been described for obtaining optimal phase settings for jammer cancellation for an electronically scanning array whose only variable parameters are element phases which vary in discrete steps. The method has been tested by exploring a few examples. The jammer cancellation results in these cases have been encouraging. Results have been obtained for N=8 and b=4 which could not reasonably be obtained with an exhaustive search. Even though the amount of computation is reduced by orders of magnitude, running the program is still costly. As a minimum, the approach described provides upper bounds for adaptation performance for electronically scanning arrays that are controlled with digital phase shifters, for small cases. As was mentioned in the previous chapter, the results of the branch and bound were not always successful for larger cases. There- fore, the following programming improvements will be made to increase program efficiency and thereby reduce cost and generate more results: 1) Removing double precision on all computations, after verifying that the accuracy of the results will not be impaired. 22 Changing the output from the Toeplitz matrix inversion routine so that the scaling will no longer be done in the matrix formation where N2 multiplications must occur. 63 Instead, the matrix 64 will be multiplied by a vector and the resulting vector will be scaled. This will result in only N multiplications. 3) Removing debugging subroutine calls and executing the program on the optimized H-extended compiler, which will greatly increase the program's run-time efficiency. 4) Eventually running the program on a dedicated processor with low overhead cost. Additional conceptual improvements can be made to the branch and bound method itself. By constraining several weights at one time, the branch and bound tree will shrink vertically while expanding laterally. Constraining more than one weight as the algorithm proceeds from one level to another down the tree should result in more fathoming. would provide computational savings. This At one extreme, constraining all of the weights in one step is equivalent to solving the problem with the exhaustive search method, which is not practical for even moderately large N and b. Therefore, it is argued that the number of weights that are constrained at each step offers a computational effort tradeoff. Hopefully, the minimum execution time, as a function of the number of weights constrained at each step, is not a flat function. The method described previously whereby more than one weight is constrained at each level is a rigorous branch and bound technique. It is possible to approximate the branch and bound method in a way that 65 would lead, with high probability, to valid solutions "rith shorter execution times. Figure 5.1. One such example is described with the aid of The branch and bound is rigorously applied to subtree I. The vertex where the MAX SIR(W~,wt) occurs is chosen as the top Nt -c -u wi EC -U vertex of tree II. This procedure is repeated until the bottom of the tree is reached. Clearly, the method is imprecise since the trajectory leading to the true solution at the bottom of the tree may be outside the search path thus outlined. Yet, intuition indicates that much of the time a reasonable solution would be obtained with a lesser computational effort. Another method to test the branch and bound algorithm versus an exhaustive search will be implemented for larger cases. will be allowed to run for a fixed amount of time. Both programs At the end of this period, the resulting current lower bound for the SIR from the two methods will be compared. If the branch and bound value is larger for several cases, the algorithm will be shown to be superior to the exhaustive search, in addition to verifying the branch and bound algorithm without having the program run to completion. It is expected that using the improvements discussed in this chapter will enable the program to complete much larger cases and to verify the results. Additional development and testing is required to verify the method in terms of correctness and practicality for use in real time systems. 66 Figure 5.1 Approximate Branch and Bound Search References 1. B. Widrow, P. Mantey, L. Griffith and B. Goode, "Adaptive Antenna Systems," Proc. IEEE, Vol. 55, pp. 2143-2159, December 1967. 2. S. P. Applebaum, "Adaptive Arrays," IEEE Trans. Antennas Propagat. Vol. AP-24, pp. 583-598, September 1976. 3. A. I. Cohen, "An Algorithm for Designing Burst Waveforms with Quantized Transmitter Weights," IEEE Trans. Aerospace and Electronic Systems, Vol. AES-11, pp. 56-75, January 1975. 4. W. F. Trench, ''An Algorithm for the Inversion of Finite Toeplitz Matrices," J. Soc. Indust. Appl. Math., Vol. 12, No. 3, pp. 515-522, September 1964. 5. S. Zohar, "Toeplitz Hatrix Inversion: The Algorithm of W. F. Trench," J. Ass. Comput. Hach., Vol. 16, pp. 592-601, October 1969. 6. L. J. Spafford, "Optimum Radar Signal Processing in Clutter," IEEE Trans. Information Theory, Vol. IT-14, pp. 734-743, September 1968. 7. D. K. Cheng and F. I. Tseng, "Haximization of Directive Gain for Circular and Elliptical Arrays," Proc. IEEE, Vol. 114, No.5, pp. 589-594, May 1967. 8. F. R. Gantmacher, Matrix Theory, Chelsea Publishing Co., New York, N.Y., Vol. 1, p. 322, 1960. 9. R. F. Harrington, Field Computation by Moment Methods, The Macmillan Co., New York, N.Y., pp. 191-193, 1968. 67 Appendix A MAXIMIZATION OF THE SIGNAL TO INTERFERENCE RATIO The results of Cheng and Tseng [7] in relation to the maximization of the ratio of two quadratic forms are presented below. Their proof establishes that w (Al) = maximizes the signal to interference ratio equation A2. ~I = SIR wt M 2 (A2) w Cheng and Tseng's results are applicable subject to~ being Hermitian and positive definite and the numerator of Equation A2 being a quadratic form of an Hermitian matrix. These features are established next. Expanding the numerator, Equation A3 is obtained. = w (A3) We will now use the following substitution T* = W (A4) ~q where W is the vector of unadapted weights representing the optimum ~q compromise between such performance factors as resolution, gain, and 68 69 low sidelobes. Another substitution is used in Equation AS. A= [ ~q wt -q J (AS) It should be noted that this is an outer product, or dyadic form which is an Hermitian matrix as shown in Equation A6. = (A6) A If the noise components are uncorrelated, the noise and interference covariance matrix M will be diagonal. nonzero entries in any position. = M . . ~n ~may [mij] . th and J-. th c h anne1 ~s . t h e ~-- (A7) , i=l, k, j=l, k . g~ven The covariance of the by m.. = E{n. * nj} ~ mji = E{n. * n.} = E{ni n. * } = m.. * ~ ~J have It is denoted by where k is the number of receiver channels. no~se In general, (AS) J J ~J (A9) This results in Equations AlO and All. * Mt = M (AlO) =M (All) Therefore, matrix M is Hermitian. It is also positive definite since 70 M is the covariance matrix for a stationary random process. Equation Al2 is greater than 0 whenever W t P n in 0. (Al2) Therefore, Equation A2 can be written as SIR = wt A w wt M w (A13) The proof of Cheng and Tseng makes use of the following theorem from Gantmacher [8]. THEOREM: Given n x n Hermitian matrices £ and D with Q positive definite the equation 1£- A Dl =0 has in general n distinct values for A (eigenvalues). Then: 1) The largest characteristic value AM is the maximum of the ratio of the quadratic forms A M = max wt D W and this maximum is assumed only for the eigenvector W associated with 71 the characteristic value ~, which satisfies the following homogeneous equation: CW= \i.l?..~ (Al4) The following properties enable the maximum obtainable SIR for a receiving array pointed in a specified direction to be determined, as well as the required set of weights to achieve it. a) The largest eigenvalue AM of the linear combination A AM is its only non-zero value and is given by = b) -1 W M ~q - W ~q (Al5) The eigenvector corresponding to AM is = -1 M W (Al6) ~q The following proof is used to establish the above properties. Proof that ~ - AQ has only one non-zero eigenvalue AM In order to show that the linear combination of matrices C - AD has only one non-zero eigenvalue ~' the following approach is used. THEOREM: If C and D are, respectively, m x n and n x m matrices, then (-l)m An det (C D - AI ) --m = (-l)nAm det (D C - AI ) ---n where I denotes them x m identity matrix. -m (A17) 72 Proof: (Al8) lrn [D - [f-lrn 0 ] -I = 0 --n A~] (Al9) Since the right sides of Equations Al8 and Al9 are the same, the determinants of the left sides are the same. Therefore, Equation A20 follows: A~] det det = (A20) det [: -~J AI det [ 0-m C ] D f.-~~ Noting that the determinants of the matrices in Equation A20 have the following properties det [-: det ~ ~] 0 l = (-l)m (A21) = (-l)n (A22) -I --n since both matrices are lower triangular and the determinants are 73 equal to the product of the elements on the diagonal and det det [c Q ['; - AI --m 0 c DC ,~] = An det (f. .!2. - AI ) --m (A23) Am det (D f. - AI ) -n (A24) AI] -n since both matrices are upper triangular and their determinants are equal to the product of the determinants of the (matrix) elements on the diagonal, we obtain Equation Al7: (-l)m An det (f.~- AI ) --m = (-l)n Am det (D C - AI ) -- -n Q.E.D. For the purpose of establishing property a) and Equation AlS the eigenvalues of the linear combination A- :.\Mare needed. These are the roots of the characteristic equation det (! - A M) =0 (A25) Since M is Hermitian and positive definite it has an inverse and Equation A25 can be written as (A26) leading to Equation A27. det (M-l A - AI ) - - -n = 0 (A27) ,, 74 Substituting C = Wt and D - -q into Equation Al7 results in (A28) With m =1 = N we and n obtain (A29) or (-l)(l-N)A (N-l) det (Wt M-l W -AI )=det (M-l W Wt- AI) -q -q 1 -q-q ~ Letting ... A= wW 1 -q-q (A30) and using Equation A27 gives the result A(N-1) det (Wt M-l W - AI ) = 0 1 -q-q (A31) Hence, N-1 of the N eigenvalues of A - AM are zero and the only non-zero eigenvalue is thus obtaining Equation AlS. Let an N X 1 column vector be equal to (A32) which can be written as W -q =M W (A33) ' 75 It follows that (A34) = since M is Hermitian. Substituting Equations A33 and A34 into Equation A15, the only nonzero eigenvalue is (A35) since M is Hermitian and positive definite. Thus, AM in Equation Al5 is the largest eigenvalue. We proceed next to establish property b) by proving that the eigenvector of A equal to M- A~ corresponding to its largest eigenvalue is 1w • Using C ~q = A and D = M in Equation Al4 we obtain AW (A36) which can be rewritten as (A37) using Equations AS and A32. From Equation AlS Therefore, Equation A37 can be written as 76 W AM=A W -q M -q Hence, Equation A32 is the optimum eigenvector W . -opt W -opt =k (A38) Therefore, -1 M W -q where k is an arbitrary scalar which we set equal to unity without affecting the results. Appendix B MAXIMUM SIR FOR A PARTIALLY CONSTRAINED WEIGHT VECTOR The SIR for the fully adaptive array is given by !wt w*l2 SIR = ~ ~g wtM w wt(w wt]w = - -g-g - (Bl) wtM w where the optimum weight vector which maximizes the SIR is Consider a partially constrained weight vector denoted by~· ~· which is The B vector is partitioned into an unconstrained and constrained subvector as follows: B = where b = [t] (B2) unconstrained subvector B = constrained subvector ~o The unconstrained subvector b that maximizes the SIR is derived next. For this purpose, the vector W and matrix M are partitioned ~q to be compatible with Equation B2 as follows: w -q (B3) = 77 78 M ~ • (B4) :] Substituting Equations B2, B3 and B4 into Bl yields ~o;t ~:~6 [~J ~ [!a] [" at aA t] [b t SIR Bt] -0 = [bt Bt] -0 ~~ C' .::.0 (BS) [aatb + aA tB [bt = -~ ~ --0 -0 . t ~o~T~ + ~o ~o~o Bt] -0 + fo~o] rb ctb + y --c- [b t .j. BI] -0 ~0 .::.0~ = where n, e, p,... and y are shorthand notations for the lengthy sions in the previous step. 2: b~ i SIR = J.. z b.* j J expres~ The equation can be rewritten as n. + e J.. = p. + y J N D (B6) 79 n, 8, p, andy are all functions of b. SIR is a real function of b. Let the dimension of b be k; then SIR is a function of the 2k real variables that constitute the real and imaginary parts of the components of b. To maximize SIR, it is necessary to take 2k partial derivatives with respect to the real and imaginary parts of the components of b and set them equal to 0. d(SIR) d[Real (b )] r = 0 Consider d(SIR) d[Imag (b )] and 0 (B7) r Equation B7 can be shown in [9] to be equivalent to d(SIR) ob r = and 0 ()(SIR) ob*r = 0 (B8) th where the r-- component of b, b r , and its complex conjugate, b*, are ~ r treated, for the purpose of taking the partial derivatives, as if they were real and independent variables. Taking the partial deriva- tives of Equation B6 yields ()(SIR) 0 () b*r D p which gives nr r (SIR) or, in vector notation, n = (SIR)p (B9) Substituting for n and p: (BlO) 80 The unknown in Equation BlO is the vector b. M b --c- = (SIR) Rearranging: a-~ ~0 (Bll) The expression .l. .l. Cl. = [~~~ + ~0 ~0) (Bl2) (SIR) is an unknown constant. Substituting Equation Bl2 into Equation Bll we obtain: b where g = ~ -1 (a.a - ~ ~0) = a.~ -1 a - g (Bl3) = ~-l~ ~o· Substituting Equation Bl3 into Equation BS yields (Bl4) Thus, the SIR is a function of only one complex scalar, a.. Maximi- zing SIR with respect to a. yields the value of a. which, when substituted in Equation Bl3, gives the optimum b, denoted in Chapter 3 by b-op.t When the same value for a. is substituted in Equation Bl4, the maximum SIR is obtained. fl 81 Carrying out the multiplications in Bl4 yields (Bl5) J Making the following substitutions, q = Equation Bl5 now reduces to SIR = 2 * * q-ag t aq+g t aaTg-g . t ap-p * aTg+p*p ' a*aq -a * qa t g+a'qp+ap - - ~ -- - - ~ (Bl6) To find the value of a required to maximize the SIR, the partial derivatives of Equation Bl6 are taken with respect to a* and a as if they were independent real variables, obtaining ~** = D[aq 2 - q~t~ + qp]-N[aq] (B17) =0 D[a *q 2 - qgt a + p *q] - N[a *q] ~- =0 The two above equations are complex conjugates and yield the same information. Equation Bl7 simplifies to D[aq- ~t~ + p] - Na =0 (Bl8) ' 82 From Equation Bl8 we obtain ~ = q- [~t~- p]/a SIR (Bl9) When the a corresponding to the minimum or maximum SIR is substituted in Equation Bl9, the corresponding extremum value for the SIR results. Substituting the values for N and D from Equation B6 into Equation Bl8, following some complicated algebraic manipulation, the following quadratic equation in a is obtained: 2 a q~* + a[8q-~~*] - ~8=0 (B20) The solutions for a are (C M -la ~/q a = .::.o:=c ~ (B21) = -(8/~*) (B22) Substituting Equation B21 in Equation Bl9 SIR . m1n ~ q- ..,.....-'..,.......,..- (~/q) = 0 Substituting Equation B22 in Equation Bl9 gives the maximum SIR 83 SIR max q ]J]J* + -e- or, in terms of the original quantities, (B23) SIR max This quantity is denoted in Chapter 3 by ~ max . Appendix C COMPUTER PROGRAM LISTING 84 onooooto (\I)Q(.I{)(.Ii-;0 8 •*** MA~SIR (N,L ARE USED FOR VARIABLE DIMENSIONS) ~·C iHIS PROGRAM CALCULATES THE LARG~ST LOW~R ROIJND FOR AlL WEIGHTS ONSTHAtNEDA THE~ IT US~~ A BRANCH AAO ~OUND ~~OCEOliR~ TO C ElECT fH€ uPT~~tJM PHASF8 IN OROER TO HAXl~tzr T~~ SJ~NAL TO C ~t@AFERE~C£ R TIOa IT ~~lb THEN PLOT PATTERN§ ,oq 6HHAYS WITH C V4RIETY O' JAMAEK CON, G RATIONS. · 8 DIMENSION + + • Ot10000 ~\) ()('Q()(.I()lllj {:lt:)Q8(\0SC 000 OOttO OQt'Ot'<! "II) 000\Hl()BQ oooooo'fo ~0Qt)f) A(64) 1 RETAJ(\Ol(EXP(b4), M~b4tb4)tHfNVRS,ba,~Ol 1 ()00001t'.O 1 W(~a), w PTCb4) WQ(h4) HNC&b\ ~ (6a,~4)~PJ(101 o('o~o1 :~o • WTCONMtbi~,WCLaf~q,,THfTAJ(lO~,PJOS(10l,PROO(b4,64) DIMENSION tO RET .1 Ct 0 l Ex:p C6q) + M &Ut~4), lN~RSt64 64)~~J(~4,b4) 1 • P (10 ,W(f>4 WOPT 4l WCHn4l RN(b4 E 8R4NC~ + . WT ONH(b~~,wctAfb«~,THETA~(tQ) ~~08ttO),PROD(b4 b4) AND BOUND o•T• ~ c DIMENIION CON(b4) UNCON(b4) 8CONC(b4) ' Of~EN tON MC(h4J~&, cOCbn,~&\ 1 cocTCbh,b4),V(64,o4),NPAS8Cb4) o ME~ JQN AT(~a Ao ~4) Bo(~«l OIME;~~ XON flt(t,IJ :R2 ~~.t);RnPT(~4) OXMENStON MCtNV(bU ~a~ 1 MriNCnCb4 64), + VOUT l ( b4 l, VOIJT ~ t ~;~to, Y:H 64), Uf t,a), V ( 64) ~ A!SUHE TH! LARGEST NUMBER OF ~ITS fOR PHASE SHtFTERS IS b E c DIMENSION jHR(b4) c oiMENSI8N CON(&4,,UNCONC~4~,acoNccool o0 M~NS~ N MCib4l~q)tCOC~a,bq) 1 COCTC~hlb4lrV(b4,b4) M~NS ON AT bU , AUC64)l AO(b4, 1 NPAS~C&«J DXM~~s gN Rt ha ,R2(h4) HOPTjhiJ) OIMENSX N Mt NV ba 1 b~\ 1 ~CINCO(o4 64), . + VOUTl(b4 ,VftJTl(~:tilJ,VlCb4),Ufo4),V(f:.4) c COMPLE~*tb EXP,M 1 MJNVR8 1 WOPT,WQ,W,WCL8,WTCONM 1 8Nt(bU,b4) cOMPLE.)(•R WSOL c PRECISION A 1 8ETAJ,8ETAS,PJ,PQ,SIRNu~,SIROEN,THETAS,RAD, *DOUBLE THETAJ,PJ0~ 1 PSJC\.~ 1 AIP. REAL•A PStMA~,SICL~ COHPLEX•t& MCJCO,COCT 1 V. 1 AT,Ao,~o,AMR~BOPT~8C0NC COMPLEX•t• ~C NV,MtiN~O,VOUTt(VOUTl,v3,U,vrAl COM"LEX•tl, wO)T 1 wQ 1 wc:b~,M,wTr.JNM INTEGER At R2 CuN UNC ~ C •N,L ARE U~EO ,oQ'VARlA6L£ DIMENSIONING COM~QN N,l. INPUT OATA OEFtNtTIONS E ** N,L DETERMINE OlHENStON FOR SEVERAL VECTOR~ AND ARRAYS c c c ~ c Alb 1 1 1 0() 00\HH•tl () 1 0 0 (:U:it,H lll) OtHlOOtSO (IC')Ot'O ~hi,) (';Of.\(1()170 000001AO 000001(~0 OOOOO:>OO (\000('~ 1 Q t)OOOOt-~?.0 OOO()O?.SO O()OQO?.AQ 1)(.\()I)OrC:.O ooooo;.·t,l) Q(H)OO? 10 o~ooo~P.o ()00002Q() ('~I)Qo:~oo 11)0000310 ()00C03;.lO I)()() I)(!:~ ~I) coo~,o·:;oo ()0 (}I} 0 ~ :~id) 0 Q (HH) : 1) ~>~I) (}~00()~~ 1Q OfH'tOO!,go !+ I) (H) Q .S '0 0 OOOQOl.IOI) 000001H)5 00(\CII.)IJ l 0 () Q (H) (;< 1.1,?. 9 OOOO!Jll ~() 1\')Q(',IQIJIJLI\) ('0~()1)/.ar;.o OUOOOIH,i) OO~OOI.i'IO o<>ooonf.l.o 00,)001100 IJO')QOSOO (){'(,)00510 oooooe.r.o tHlOOOS~O ~0(\0Q5H() 00 l/1 c € c ~ ~ c ~ ~ c c N•NUMBER OF Et..Et-4ENTS IN THE ADAPTIVE ARRAY onooosso LsN\)MRER OF JU1MF:RS 00Q005"0 0 0 0 /J 0 ~.:; 'I 0 ~a4PER4TURE ILLU~JNATION VECTOR (IF A IS •LL ttSt THFAE UNIFORM WF.:tGHf ANO NO RIDf.tiJ!iE TTENlJATlON) OOOiH•~;HQ THE TAJ•JAMMtNG SIGNAL DJREC T ION ANCl.E: fROM Mf CH. eor.t:;HGHT 000005QQ N OfGRf[S ( L NOICATES LT~ J~MHERl 00000601.) THETAS•O SI~ED SIGNAL DtRECiiON ANGLE FROM MECHe BORESIGHT 00000~10 N OtGRF.f.S ()(\()0(}1;)20 PJDRsNOt ~ POWER OF EACH JAMMER IN OECI8EL8 (0 TO 100) OQ?Q~P;..~O ('()')() O"tt () t1N0f.X ~NOICAT[S lTH JAMMER) 000(~\)t:,~~., PQaNflJSE P W\R OUTPUT Or EACH ELf.MENT (QUIESCENT NOIS~ POWER) AS A RATI, OF SIGNAL, TO NOISE . ()00006.&,(} NBITS •· NUMA(R OF PHASE SHIFTER BITS (USED IN WEIGHT SUBROUTINE) ()00011610 ~ANG~ 2 TO b ijJTS OtHjOOhP.O 1s Ib NAMELI8TINPUTt/N,L,A,THETAJ,THETAS,PJOB,PQ,NSITS 1 LIMNB,TEMCLR, + I<CMA)(,t<C~A)CN OATA OLAMIOt5/,P!Il,t4lSQ2~54/,0EG/51,2QbltRA0/0~0l7~53/ INITIALJZATJON' PJ(t) l'!'LF~~M ol Th~A I J 8 • 8 ~c c I)0()007qo ('00008~0 O*P *0 AM*DS N(THETAJ( )•RAO) RA,IO OF SJGA4L TO JAMMER •to.o••CPJoB<Y,-o.t) g *** REVISED AND SIMPLIFtfO BRANCH AND BOUND E N JSTS TH'NUMRf.R OF ARRAY EL~HENTS i9 THE NUHR~R 0~ PH~S~ SH_FTER BITS c NB OPTt~UM W~t~YT VECTO~ t'00QOA10 oooon~lfi 0001)\'HUO OOI.)Oi"!,'HlO OtH)()(Ig~>O CALCULATE THE QUIESCENT WEIGHT VECTOR AND THE JAMMER COVARIANCE lNO tfS 1NV£~SE CALL MWQ(ALBEJAJ,B!TA5,E~P,M 1 MINVRS,MJ,PJ,PQ,wQ, + PHOO CALCULATE lHE LAR~EST CURRENT LOW!R BOUND WHE~£ ALt wEIGHTS AR~ CONSTR tNEO Tu A DISC~~TE SET OF VALUES (~SIC 8) THE ASSOCIATED WEIGHT VECTOR IS ALSO OUTPUT 10 CALL 81RCL8(WtWQ,HSMINVRS,WTCONM,WOPT,WCLB,NBITS,RN,A, + P~ltL8 WO Y IS THE ooooo·rso ro E c 0000073\l l)()()(.\()7t10 000l)07~0 5 CONTINUE 8 0000070~1 ()00()Q'1l0 {I(HJ007i?O {11).1.)007,4,0 00()QQ7 READ (S 1 NPUT1 1 ENDa2le8) BETAS 12,0 * PI•OLAM•DSIN(THETAS•RAD) no s ~ TAJ c CONY~RT (IQ(')()O~t)\) ()00Q0"7<t1 PRoGR•M OVER THE CONTINUUM (t()fJQQA~O O·~?()O.a. IQ MATRt~COOi)r:\~0 000()0.1."\!1() ooooo;1o~ ('0~100~1 Q 00000'720 -:-OvQV<i.11•0 ~i)Q(i'.)'H10 ooooo~so ('QOOOq~O 0 () (1 0 0 l? '1 () oooonq~o oooo~q~o oooo cno OOO(l Ott') (100 1H O?Q t.~c.oolo:~o (}0 0 01 0 ll 0 ooootoso I)OOC~Ot~O 0000~070 (\000 l OEHJ 00 0\ c c c ~ c 8c WQ IS THE QUIESCENT (UNJAMMEO) WEIGHT VECTOR WQ IS PARTITIONED INTO AT(UNCONSTRAtNEO) ANO A0(t0~5TRAJNf0) BCONC IS CONSTRAINED OPTIMAL. WEIGHT VECTOR~ ONE ELEMENT IS DEfi~ED f CH TIME CHSW IS EXECUTED. BOPT IS lH~ UNCONSTRAINED OPTtMAL.WEIGHT VE.CTOR• IT IS CALCULATED FROM PAR M TERS DF.FTNEO IN THE PS!MA)( SUBROHtY~JF~ PSICL~ IS ti~ ~APt;f.:A~ CURRfiNT tOWf.R f1011NO fl_,;i, 1 ·~~ 51R WC~R IS TH ~E GHT V.CTOR ~nnq SPO~OlNG TO P~Trl" M s tHE J MHE nv• tAN E MAT x Sl MAX IS OHE OP~IMU~ St~ OVER 'HE CONTINUUM••NOT USED IHE ~ c ~ ~ THIS c ** IS THE BRANCH AND BOUND PORTION 0, THE PROGRAM • c c c c c c c c I(.Ot II~ H· N EX * ~u~ 1. . . wtlL W L RUN lk usEo ROM TO N+l · ~oq olMf.NStONING THE ARRAY! ~ 00 50 J•lt"OIM . . 6MR(J) • ~£XPCCo.o,t.O)•(J•t)•RRAO) CO~TINUE C WCL8, PSICLR IS THE SOLUTION FoA TH[ UNCONSTRAtNED CASE. c c NHt • N onooit"'o {'IQOQ~it.,O ()(ii)Q117Q C00011Ar-J IJ~t(ll)11 r,;o ()<)00~200 GOGO 12 lt') OOO<H;?..·.o 00001?:!14.) ()0('>1)1,z'"'!IQ 1)0(\Q~?kO C00Q~2'70 c ~ FOR FIRST TIME THROUGH JHE BRANC~ RULE, M IS NOT PA~TtTIONEO AND C SOPT ~ ~OPt &~D PSJMA~ S PSJCt.O c 0000 11.~0 00t'Ol1UO 1)000 l r!SS WRITE (~,.) KCMAX I n l K22CNT • 0 NPASS(I\ • I J~tN9XfS,NE~2) GO TO 20 AIC~Rs"S CLo GO 0 30 . 20 1~(. ICL8AGTAPSICL8) GO TO 30 20 AtCU~:TEMCl." wRIT~ (~,~) SICLB 0 CONTINUE ALCULAfE THE VECTOR FOR THE WEIGHT VALUES B•SED ON THE NUMAER oF HAS€ SHIFTER BITS HDIM • 2 ** NBlTS RRAD • 2.0 * Pt/FLOAT(HDIH) 50 G'~001.UO 0(10011~0 00tl01?50 kt<~NTN•O N •O ~ 000010QO ~000 11 0.0 CA L ANGLECWOPT,NHL,ANGL,PJ) 00001?.~0 oooot?Qo 00001~"0 t\000 L!.05 Q0~0131Q Q(lOOLH.? 00~0 1 ~~?Q 00001~JO oooot:~t-40 ooo"tr5"' (\f\0()1 .,-',() olHltH :no n.ooo~~fio 00001 t~HS 000Q13t~t') QQO?lt~flO 1}00011!10 0000lL120 OOOQliJ,~O 1.)00014~0 0 0 0 0 11J r~ 0 000014b() (1000lt.t10 OQ001UAO 0(Hl0 14'Ht Ot'?Q1500 '.'O~H) 151!) (\(}()0\'?2() tH!OQ1).AIO ~00015,40 Q!J~0~5SO ,_ 5~0 oooot':)so ~Jo-:•o t'OOQ~~7() 0000151')0 (X) '-I c ~ c E c c c E CALCULAT~ O~T[R~IN A(R) TO THE INDEX INTO THE ALLOWED VALUES '0R THE WEI~HT VECTOR Rft 1 l)t THE INOEX cnqRESPONnlN' TO THE Cl~~~~ST V~LUE HE W~ ~HT(tl) DUE TO THE NUHBER OF PHASl SHI~rfR SITS CALL RtR2(1,HDJM,ANGL 1 Rt,Ri,Pl,SEQ 1 IN0 1 N) ()0001td0 0 I) , ~! 1.1 () ('\1~1 100 NP I •• ~+~ • (. WRY' ~HL a 0000 lMHJ QOOQl"qt) •. I 1,*, t,NPASS(t) 'JII)I}(;I 1)01)0~ 1.HI 1)00011S() N•L 0000\i"S 1 II:'~~TN aKCNT~+I NT+! OT II I(~OT • t .· ' Jf(~~Nl,EQ 1 25000) WRJTE (6,*) KCNT,K80T,S1Ct6,WCLB WRI bt*J t PS~MAX SICLA TF . S. HA~ G.'s C A)'GO TO 2t0 k JF(~CN N.FO.~eHl~b) 1 fF(KCNTN GO 10 IIOu ()00017~0 Q(l()Q17"10 ()0001780 OOQO 1 7f~O (l0()0 1 p,Ot) ()()f)') C l5S RCONC(L)IAMR(R2(L)) c lbO CONTINUE c tFCI.GEefN+t)) GO TO ~00 c C REORDER AND PARTITION THE MATRICES USED IN PSIMAX c CALt RE. ORPRCNHL 1 N~BCONC~MtMCtC0 1 COCT 1 V.,ATtAO,R0 1 L,wQ!CON,UNCON) CAl StM~~(NML,~~MC(COt O~T,v 1 Al,AO,~O,MCINV,PRuD,MC1NCO,VOUT1, + voUT2 1 V1,u,v,A ,P, MA~,ANt) I<CNT • KCNT+j WRITE (~'*' KCNTrKROT,SICLB J f(KtNT.eQ.KCHA~~ F(KCNT 9 EQ 1 (KCM.~+t)) KCMAX•KC~AW~JOOO wqtTe (ol*J PSIMAXtSlClB c tF (PS!M XlLT.AICL~) ~0 TO 400 ROPT(lJ • l*U(l) • V\l) .AL. N (AOPl NH •NGL,P ) ~AL~ t2~~J,MDIA,•~6L 1 Rt,RI,PI,SEQ,INO,N) GO ~0 10t) C ZOO CAlL NSIHA~CACONC,WQ 1 M,WTCONM,N,P5lMA~) I( 171 () (';()00\7(,() SELECT VALUE 'OR NEWLY CONSTRAINED COMPllNfNT 0, WEIGHT VECTOR 150 t'(INOrNElO) GO TO t~S K NT a f'J(}(i0,~111) (;t()001i;110 (10001701) aao~gc 1; MR(RI(L)) c ('0(1Q1~10 000016~0 oo Qt• ~ etf.o L • l•l ~ OOOOtbOO wRITE ~Q.(KCMA~N+t)) (b~•) KCNTN,KBOT,siCLR 1 KCMAXN kCM-~N•KCMAXN+JOOO ZlO JF(PSIM4X.EQ.SlCLB) K22CNT•K22CNT+l 1~ 10 ()()(101~.;.?0 000() l P.lO oooot~ao OOOQtF."!-0 (\0001~~0 (\t)Q 1HA'7() OGOO~P.~() ~0001H90 ooootQno ~Q~O'-f';)tO OOOOl'JeO oooo~<;,;o O~OQ,</~1) ool>o 1q~~o ()00019~1 00001!?t>.() OO(HH ()10 t;'OQO~t:U~O 0(}001t;!Qi) (H)C(.Ii!()t'l('! (•{'()OrO 1 Q 000(}~~020 :s (I (' ".· 0 ~~. Q 0 1)000?0~5 (1(1()(,?(\tl{l tHrOOt~O~I) (I(IQ(}t,ll)~() ~;')00?07Q ()()(l()t?(i$1 0 -., 0 Q () 11 QA l 0(h~OiiOA2 (I(}Q(!.!!QAl OOOOi'OAtl 00 00 WRITE (bt*l PSlMAX,StCLB,K22CNT S I Cl. ~ • P !\ 1 M4 X 00 2ti0 J•l N Q()()(J2(IO(l ~CLBlJ) • ~CONC(J) tF(K?.2C~T.EQ,l0) GO TO t9qq ** FATHOM VERTEX Z50 ~ ~ CIJ~TINliE l 400 NPAfS(L) 1 NPA8$1l) + JF NPASStt) 1 G~ 1 ~DIM+ )) GO TO 500 IF . NPAS$( ),N~, .) GO 0 4~0 I~D • no To so c Q00021 ;;;s ~Q<HJ.? 1 J1) OOOOi"~ll(l 00001?150 OOOQe'tt...l) 0000a'1'10 00002180 OQ00.?.1t?Q 450 I • t•l COOO?.t~?O ()QQ()t';~3\) trci+t · GH(JNO,S!Q,Rt,R2,MOIM,J,N) ~~t=~ML•t GO. TO 150 500 NPASS(L) e l J • l•l t. • L•1 NML • NML+I NO a 0 F i'~~T.t) GO TO GOO lqqq r.ON t~UE 00 .000 Kt<Kat 1 MOIM WSOL • A"'IHI<K,c,) WRIT~ (b,*l WSOL I aooo coNTaNtJ£ 00 ?.tOO t(t<t<•t 1.N wSOL • WCt,I!\(KI'\1() wRit~ (~,•) WSOL 2100 CONT NtE WRIT ~b,*) 8ICL8rNBtTS,KCNT,KBOT CALc. ARR•v ''CTOR &No PLOT GAIN FOR ALL THREE WF.tGHT VECTORS CALL ARFACT(~ 1 ~CLS,WOPT~WQITHET.J 1 THETAS) tFinaiTS.(Of 1"~8) GO Tu 2 6~ N!JITS • N'\I •t. r,o rn to 21b8 CONptlNtJE STO 0[13UQ 8U8CHK c c co9oi1to 01l00.?.1?0 ('0()0~?-l'H) c•LL RR E c 0000..'.100 l ~~t:AHLtt c nooo?o~s ENO SUBROUTINE MWQ(A 1 8£T&J,BfTjS 1 EXP,H 1 MINVRS 1 MJ,PJ 1 ooooc?10 IJOOOl;:tiO 00(H) i?.250 OOOOl?ii\Q I)(H)£)?21(> OOI)Q~~~() ooOo22CJO 0 MH; I. 3 o() ?OOQ?~tO OOQO?.:~rl) v \J t";)l •. ,) u 0 ""•"' ''" ~ ~~ oQoo.r .-,ao oooo?::r;o <HH>023"8 ('100(1? -~I oooo21"1) OOOQJ!~1t! "000i?'i.1S ~ooo2:n6 00fJQ2~77 f.IOIJOr3f\?. 00002~VW 0000~-~B6 oooo~~AP. 0000t?3it:;O 0000?3~.?. t)t.'00i?Yi3 00 00~ ~r'J~i ~J~OOOi.l,Qb (\0(\0t:'LHJ~t OO()O,l/.l.tO 1)00021A?O l)()()(ii'lAif.() OOOQ~fJ.tlO ()0002/J'!H) OOO{l2tH.() 0 0 0 () ; l.i)' 0 IJ000?.ll8() OQQO#!.IAqO (X) ~ + PQ,WQ,PP.OD) 1)0002'500 ~ HI~ ROUTINE CALCUL.TES THE QUIESCENT WEIGHT VECTOR c I NO TH~ J M~E~ CONFIGURATION (COVlRIANCt MATRIX) PLUS ItS tNVE~SE ~ c 00002'510 t'Cit:~o?s;:o 00 1l0?5.~,Q Ch'.l 0 ();;»5!1 0 t'OOO~S':~o <:*f..HJ02~~~o DIMENSION A!N) 8fTAJtl) EXP(N) M N 1 ~),~lNV~StN,~),H~(N,N),PJ(L),WQ(N), + PROO(N, ) COMPLEX*Ib EXP,M,MINVRS,MJ,wQ,SUM,PROD,TMCt00) 1 RC100) DOUBL! PR£Ct810N AtBETAJ,B!TAB,PJ,PQ 0.00()?~1'0 + c c c ~ C0~t-10N N L CALCULATE THE QUIESCENT WEIGHT VECTOR (WQaA(TPANSPOSE) * OOOO;?Ci€\0 ~ooo;s·~o (l01)Qi?f)~0 ()(i()i)fJI, 1 0 Oil002b . ~o oooot?.-,>o EXP) oo too t<•trN E~P(~) • ~uEXP(COaOlpl,O)•CK•l)•BETAS) WQ(K~ • A,M) * EX~(K) 100 CONTINUE E c ~ THIS LOGIC FILLS IN THE fiRST ROW OF MJ DO 200 y•t,N SUM e ( ,O,Q,O) DO ISO •t,L · ~UM • SUM t PJ(I)*CDEXP((0.0 1 1eO)*(J•l)*BETAJ(I)) 150 CONTINUE MJ(lfJ) • SUM 200 Cl,NT ~llJE fiLL IN THE MAIN DIAGONAL ELEMENTS C DO 300 1<•2 N MJ(kiK) • HJ(t,t) g E 300 CONT t-WE FILL IN UPPER IRIANGULAR PORTION OF THE HJ MATRIX (TOEPLITZ) Hf.R~I1'JAN I"1Pt. f:S M(J,t)a"4(J 1 ,J)• NHt • N•l ~M?. a N•i! 00 700 l(o2,NMl S~T5~0 ~~!~NM2 tF (JKf) l~l,SOO ()t)Q()i,llf:o,LI" ~>ooo2b•;o ()1000;.>~60 (1(\0Q;!b't(J (If..~ \) f) ;> 6 ~ 0 ()I.Hl()?hG!) 1Ji)00i HHJ oooo27to 0000~720 (){100~730 1)()()()27&0 000027~0 (}000~7h0 tJI00027i'O o~,oo;-780 000027Q(} {\0Q0?800 .,OOQi?fH 0 00002.A?.O oooo;..".})O ()()QQ?P./Jt) OO()~);!'F\t;~t.) OOOQ?P.~10 oo6o . u~·~Jo o:.?og:?.B. e,£ .?A~•J oooc.!?.qoo (100 ()000~CJ1Q (H) QJ. •? ;1!o <P 0 I) 0 ?.li OOOOi?~t.ao o ~ 1 H J+ 1 J+~ U MJ KJ M~~J•k,J•t • ocb~~G(MJCJ+t,J+K)) t~ CtJ•~).r.Q.N) r.O TO hO~ 500 CONTI~UE hOO MJCKtll a DCONJGCMJ(t,K)) 700 CONTJ~UE HJ(N,l) e OCONJGCMJ(t,N)) .H> OOI)0,~9r~o ()0{11)i"7~Q 00002\)/'t;J ()000.?.1?1'-H) Ot'OQ?'f"'O 0000l(H')O Q c \) I) ;~ 0 t 0 OCH)0~02€J 0000 :so:so () 0 IJ Q 30_£1 () \0 0 c c c c 0000305() ** c c CALCULATE THE TOTAL COVARIANCE MATRIX t H • MO + MJ MQ ONLV CONTRIBUTE$ • PQ TO THf D.LAGONAL Et.f.MfNTS AND ITS I~VE~SE 00 C)OO Jat,N 09 800 K•ttN c :gg M~~fK) 8 M,J(J,K) ~o~n I~~~ lORN~R,MINVRS,N, 29N!gAt~l ~i~f~aHlNVRS(lZR,JZC)/M(t 1 l) 1012 00 01 R11t,N t6NTJNIJ ~tOtt CONTtNU CALL CO XMP(M,N,MJNVRS,N,N,N,N,PROO,N) RETURN nfaUG SU8CHt< 8 g c ~ ~ c c E~O SUBROUTINE SlRCLB(w,wQf~'MINVR8 1 WTCONM,WOPT,WCLB,NBITS,RN, + A,PS CLR) THIS ROUTIN! WILL CALC THE LA~GEST C~RRENT LOWER BOUND FOR THE c•S£ ~HERE l~L WElaHTS lRE CONStRt ~EO TO OlSCRETE VALUES. T~E ASSOCIATED ~EIGHT VECTOR IS ALSO 5 VEOe IN AOOITION~ THE ASSOCIATED OPTIMUM WEIGHT VECTOR ANO HAXIMUR SIGNAl. TO INTERFERENC~ RATIO &RE CALCULATED 4ND PRINTED ~(NLN~ 1 MtNVRSCNLN)~W(N~ 1 WCL8(N) 1 WQPT(N) 1 ~Q(N) oOfMENSJ8N MENS N WTCu~RlN) 1 A(Nl 1 HN(N) COMPb~••t6 MtHINVRStWCLA~WOPT 1 W~wQlWTtONM{.Wt 1 W2 I)OIJS PRECtttiON A,~JRNUH,SIRuEN,SIR,~SIC 8 CO~M ~,L ooooJt OOOOJ1?.0 1)1)(1()31.~0 (1000llUO lH'P)O~,~O DO 1000 J•ltN · M(JtJl • M(J,J) + PQ 1000 CONTINUE c oo Into INDEXat 1 N TH( NOF.X)cM(JNOt~rtliM(t,l) I 0 l 0 r. miT I NUE "4Htf:aN•t 00 1020 NtNOat 1 NHt! N~NOPt•NHto+l R NlNO)•fH(N!NDPt) 1020 CCNTJNUE . CAlL ooco.,o~o (it)I)Q;\QJII) CHHJ 0 .~ 0 A() t:' 0 Q 0 ~ lYHl (>t.) ~0')•).~110 t'OQO~H~o (I ()() 0 -~ l 'f 0 oono:-\HIO (II)Q 0 j l ::t (.l 0 0 ~ I) ~It?. fH) (10(103~\ 0 00003220 tjr,-()0~~~10 1)00\).~~?f.~Q tf0003?cso '.){H)•)l2~0 ('OOQ'V'IQ OOOOJ(!AO 1)000;\;(i)\) oooo~.1no 0000;\1!0 ~OOQ:lil?.O oooo:\:no oogg;~ ""fl \1 "l.t.tQ ~~-··1·{1 • .~ .:iv ('OQ\1~~ . ~~0 0 1)()0().~~ ~ 00003:\~0 0000!~~9() 1)000:~40(; OOOO.~IH Q ooooltl~O C01i03a3o OQQ0!.4LI') 0~00:4811)(> Ou0031J60 OOCO;~.t.! IQ OQ(}O :i!.IJ8(1 OIJOO:~tJIQ() oooo:;soo 000Ct.~51 0 0000 ~f)?, I) OGOO~~:do 00003540 0000!5~~? 00(ll):a;$h() 0000l510 , 0 Q 0 :\ 5 ·!H'~ 0000:\5()() \0 1-' C INJTIALIZ! 8EED 'OR RN GENERATOR IN W!IGHT SUBROUTINE c c 4)001)1600 (\I)') r,<;, t Q ooCliL~ f;: .2 0 15 • 5 _,OC1 ~:;f:L~0 t<R • 2 PSICL8 • o.o c 00 700 I11 1 KR c C GENERATE A WEIGHT VECTOR FOR ALL WEIGHTS CONSTR4INEO C' T.. lS ROUTINE WILL G~NFfhTJ:: ONE ~EIGHT VECTOR W CAlL Wf.IGHT(.,N,~BIT~,RN,tS,w) CALCULATE THE CURRENT LOWER BOUND FOR THE SIR * * * CALC. WCTRANSPOSE•CONJUr.ATE,•WQ AND C WQ(TRANSPOSE•CONJUGATE)•W oooo:o\hHO 0000!,.1-,~0 ~,..1 nt'lo"\',.:. c"' . , . . . vv .. l-~.,~t.:'ly ovoO:!·t::.'iv OOOOJf,RO ')0-::'03tf.,(JO ~ oQc<~ "(u) 0001).~1tO 8 o~I(I('I:~;P,o ~Jooo ~J.o oono:~TtHl OOi)rJ:~ '1t:•O .n E c c ~ c E c ~ E c ~ f c 11111 • ('0003'1~1) oooo:.;7·7o co,o,o,o) <:t 0 0 <L~ i' !Hl 0 (1 (l 0 .~ i' f) 0 (i (,\ I) (I .HHI 0 (1000 !~Fq 0 00 100 llat· N . . Wt ~ Wt + ofONJGCWClt))•WQ(It) 100 CONTINUE NUMERATOR 0' SIR • Wt * W2 (R!AL) SINCE M IS HERMITIAN SIRNUM • WI * DCONJG(Wl) CALCULATE THE DENOHtNATOR ,OR SIRCLB • W(TRANSPOSE•CONJUGATE) •M ~0\)0Hl~Q OOOO~oAJO 0000 :".P. t!O OOOOJASO t'O 00 ~H\60 * ()000,~·~'10 w ~(H)03ABU Ot100 ~P.90 1)0("J3':;oo CALCULATE W(TRANSPOSE•CONJUGATE) on 2oo K•t,N wTCONM(~) 1(0.0,0 200 CONTINUE oo 00 4no 300 *H oooo .'~(? 3t') (,\() 0 (} J t; 1 0) t<at N 12t~~(,.N WTCONHCK) a wTCONHCK) + 0CONJG(W(I2))•H(I2,K) 300 cONT}NtJ! AOI) CONT NJE CALCULATE W(TRANSPOSE•CONJUGATE, * M * W • SIR DENOMINATOR CREAL) S!RCE M IS HERMITIAN SIROEN I o.o 00 500 K•t!N ~O(H)!,q\ () 0(\ I) 0 .~.::h: {) SIRD~N • 81RDEN + WTCONH(~)*W(K) 500 CO~TINUE E CALCULATE SIR(CURRENT LOWER AOUND) CREAL) SINCE M HERMITIAN ~H) OOQQV;l~'d) 0000~~~() ooo'.>!· 9 ·~o ,.,~o:r.c:~o ~OQO.!i'990 oooo«·.H>Q 0001)1.1()10 OQQ()IJQ(!O l)nooao:t;o 00().()4{ilfA0 OOQOI.IOSO OOOOi.iQ~O 00001!0"10 ('00QIJOfiO ~4)00I!OOQ OOt'JQtq hO ~oooat,o ('00014 120 t'OOOIJL~O ooooat«O \D N c SIR • SIRNU~ISIROEN ~ fiNO LARGEST CURRENT LOWER BOUND roR SIR AND THE CORR• WCLA C I' CSJR.LEIPStCLB) GO TO 700 c PSICLR • S R oo w LAolot() K•t,.N • w( K 600 c8Nt NUE ) no q~g K• WOPT~K)J• ~OPT(K) + ~8~lt~H~ 00 ~ 1·, NN :38 t)() • 0 0 (} 0 iA t 'I 0 (\(H'IQI.l~@.Q (10(! 0 IH ~0 {II()QI)IlJoo 0 tl 0 {) '~ if. i I) OQO Oll;?~IJ ()0(.~~1.1~31.' 000()/J;.'l/l(p ooo?4.t:•:;o C 700 CONTINUE ~ CALCULATE THE OPTIMUM WEIGHT VECTOR , WOPT • MINVRS 00 A~O Kat N WOPT K) • fo,o,O.O) 800 CONt NU£ c ocoo~t ~.o l.)()(q)I,\~~Q oo * WQ Q() (!. ~hQ (\0\')0l.tP, ]l'j 0 0 ('H') 1..1 ;;> Pl 0 (\ {) (i I) /.J. t~ q 0 OOOOI.t!{-00 <H'~!I).~t3 1)00(1t13].0 '··· ;.) ··:< \• HINVRS(K 1 1)•WQ(J) it' ()()0(11,\?.i!() fA l ao {\ 0 I) 0 1A3 5 0 ,'• ;"l(l 0 /.A -~ ~ 0 QO\H,IB./0 ()o()OQI.A3::\() l){n)OLI3c::IQ oOOOtJIJ.O() (\C)\)QlJI..ItQ CALCULATE THE MAXlMYM StGNAL TO INTER,ERENCE RATIO OVER THE CONTINUUM~.oo~aa?.o UPPER SOUND (NOT C UALLV USEO) oOOOLI~JJ<> f..'()Q(I.:A/Hl() () 0 I) 0 tAIJ '5 (1 SIRMAX • o.o c OOI'IOIJ.IJ~O DO lOOO ~at N 000?4a7o SIRA4X • SlAMAX + DCONJG(WQ(~))•~OPT(~) OOOCLia8o 1000 CONTINUE ooooau~o ~QOOLIS~O PRINT OUT WCL8,PSICLB, AND WOPT AND SlRMAX ~OOOLI5io nooou~:?o WRITE l) PS!CLB StRHAX oCOQUS~O t FOA:M~T IH t'OUTPuJ rROM stRCLB ROUTtNE•-.•*'''' tHh)O'.!~·iLIO + PS CL8•' ~16 t0,5XlfSIRMA~al,Fth tn,/1 1 • WCLB't}OX tWOPTt~oooaS~O + L sox,•~Q•,tox,fM•,tnx,•N•,to~,,NRtYs•,t~x,•PsiCL~ ,t&x, ooooasho f ~ ~ ~ c c c c (t • 'StRMAXl,l) wRITE WRIT€ <t•*) ( '*) 00004510 WCLB,WOPT 1 WG,N,N8JT5 1 PSfCL8 1 SIRMAX WC~e,WOPT 1 WQ,N,N~ITS,PS C~Q 1 SlRMA~ RETURN t'tRUG SUBCH~ E~O ('l()O(lilSP.Q oOooas~o 00004600 OOO()L161 0 c.t.h) o L\ 6 .:> o tl () () 0 1..\l:d 0 ('()Q()L!6tlQ (! () 0 0 IJ As Q SUBROUTINE WEIQHT(A,N 1 NAlTS 1 RN,IS 1 W) 000()1At,~() OOfJQI~,;~/0 t)OOOl!i:\f\0 0000/J~QO 1.0 w c c THIS RO¥TINE CALCULATES ONE SOLUTION VECTORi w1 FRf.IM THE BOTTOM e or THE ~EEA EACH ElEMENT tN W fAKES 0~ A 0 SCRETEr COMDLfX VALUE c NAilS • Nu. ()F Pt-iASE' ~HifTER RITS c c c ~c ~ ~ c ~ c 00(H)L1170 !)00Ql1'1P..Q () Q 1,) 0 tA I' 9 I) 00 {HJ (,;\ fH'; 0 OOOQtH~ 000()1..\P,~I) ~ " 0 0 J,q l\ (l t) 0 tH) IH\ ':· \l OOOQ4B-':<~) (\ I) ()I) I! fi; 'j' ~ w 000(11!.'\80 00 ZOO K•trN CHOOSE W(K)•AM CAM TAKES ON VALUES BETWEEN I AND 2 DO lnO ~t•trNVAb ·· JFCRN(K .~f..CFL ATCLt)IYAL)) GO TO 150 100 CONTtNU Q()t~OilAGO ** N8ITS) c ~ c Q0t)OI49t')0 (' ('HHJ lJ t) 1 Q Ot-QOlit?.~o 000 ()liGoJO (I () {I 4J iA q. !Ht '='Q()('Jt.ie:p'SQ oooo~4~qo I)()')Otl")\JO {'II.)QQt:.060 150 RETAK ••CFLOAT(L12 * 2eO*PI)/FLOATCNVAL) W(k) • A(K) * CO~XP((0 1 0rl 1 0)•8ETAK) 200 CO~TJNUE Oti(lvSoP.o 300 CONTINUE ()t)OQ~ ('t 0 (} 0 ~~ ~ 7 'J OOO<;:.:;(J9Q oooor;.too 1t 0 (J00U5\?.0 f)tAUG SUBCHt< c t0 oooot.1.e2o R(TlJRN c nco ('(,1()()/J '7';~\ (H)()()/l'i't!-tiJ' DIMENSION A(N) 1 RNCN;, W(N) COMPLEX W•1& OOUBLE PRECISION BETAK 1 4 0-TA Pti!.I415Q2bS4/ NVAL • 2 ** NHITS VAL • NVAL GENERATE 4 SEQUENCE OF N RANDOM NUMBERS CALL RNOUNCIS, RN, N; CALCULATE N EL~M£NTS OF THE VECTOR 0000l1700 0()00471'.' (' (\!)(~ IJ_ (' ;r IJ '~ f.A l s. n i) 0 t) 0 /J i !J 1) oQOotit:~O E~O ()()~I.)S1.Ll(\ 0000':'11?;1) SUBROUTINE ANGLEC80PT,NML 1 ANGLrPI) OO~H)t:,1 AQ OCHiOS17t~ COMPLEX*lb ~OPT(NML) ONLV ONE WEIGHT IS USED AT 4 TIME JF(REAL(AOPT(~)~ 1 Nf. 0 1 0) GO TO 50 F(A MaGCSOPT ' J L u O) AN~ • 1 5*P ~F (1JMAC(80P ( )!,!t,6 1 0) ANhL a & JF(AIHAGCf\OPT 1 1etO.u,u) WRITE (o, 1) r,o ro too 50 ANGlL a AtAN(AIMAG(BOPT(l))/REAL(BOPT(l))' _ IF (AtMAf.(B0P't(1)).LT.O.O) ,.~H'.>. (Rf.Al.(fHIPT(l)).LT.O,O)) ANGL* + ANr.L+PI , JF((AJMAG(BOPT(t)).GT.O.O).ANO,(REAL(BOPTCt;).LT.O.O)) ANGLa + ANGL.+Pl s.JI (tCH}Qt;~~() coo or;~ fJIJ 0 !'H1 0 t~ ? 0 0 ()~\()Qt"·2i () 0000~220 ~1.)0(1\),?.j() OOOC';~2~0 O')t~O';?.St;t OO()Ot;}~O tH) 0 !> S 2 'I 0 t;'OOO~i~ll.l) oocos;;Hjo ooooS.~(to ~ooo~:.to 1)0005320 \.0 -P- c c ~ C C c C ~ c c ~ ~ c tF (ANGLeLT.O~O) ANGL•ANGL+2•Pl tOO CONTINUE l FORHAT(tH ,•REAL AND IMAG. PARTS Of WOPT BOTH ZERO••ANGLE RTNe') ~!TURN O(BUG SUBCHK 00005310 SUBROUTINE RlR2(t,MDtM,ANGL,Rt,RZ,PI,SEQ,INO,N) THIS ROUTJNE CALCULATES THE R'S FOR ACR,(Ra1,,~. MDlH) R IS THE NOE~ INTO THf Al.b0W£D VAl.Uf!J f"W>. THt Wt,1 IGHTS OU£ TO PHASE SHIFTER BITS P. IS ROlJNOI!D P trJ tNTF.GER VALUE . SEQ DENOTES THE: OIREt:TION (•taCLOC~WlSf: •NO +t~ncOUNTE'R CLOCKWISE) FOR THf A8SjGN~ENT 0~ VALUES FOQ HIGHE~ ORDER Nt5 THESF. VALUF. ALTERNATE Cw AND CCW ROUND THE CIRCLE FORMED BV TH~ VALUES OR A(R) INTEGER RtUU 1 R2(N),SEQ C • FLOAT(MD!H)/f2.0•PI) Rl(f) GIVfS THE CLO,EST AblQWED x•.bUE A(R) TO tHE w~IGHT W(lt) 0~(\2~/J 10 ooovS420 E~O R2( ) GtVF.S tH£ S~c~•NO CL. SESf V l. E A(A) Tfl THE WEIGI-lT ~(I1) Rt(Ii•AfNT(C*ANGL + 0 5' + t IF (RI( ).GE,(MOJM+t)J Rl(l)•Rl(t)•MDIM IF((C*ANGL)•(Rt(l)•l;) 210 1 220,220 ~ THE AltOWED VALU£ FOR WEtGHT W 19 LESS THAN THE CLOSEST INTEGER C HULTIP E 0' THE FU~OAMENTAL ANGLE DETERMINED 8V THE PH~SE S~lFTERS ItO R2(Il (11• l. F. • 8 t F" ( R2.. AEQ • • so·to 3 C 2ZO c ~ ('1)00~:~~0 C000t;~10 00005~HHJ "O\JO'lii~ 0 0 OOOO'?ltiO(I 0 0 (,!> () ~ fj §. 0 oo~osaao 0 0 (} 0 ~~ IJ ~; \) 0 0 ~)(\!;i !.! 1, Q I) 0 0 ~)!)In'() OOOC~I~AO OO<'O"ii!I'JQ 0000~500 OO()()~~HO oooo~s~o COOOS"i$(' I) 0 I) (} '-" '5 ll () oooo~~s!Sc ()I) IJ (} S ')~ 0 oooor~\l)i(} ()~OOSSMO OOOl)"S"'O c c ~ OOOOI:t3!!0 ('()('Q!d5() R2(t~ Rt(l)•t ~ ) Ri C1 ) 1 R2 CI ) +MD I H o • RtCl)+l . IF (R2(J).GE.(MOJH+t)) R2(1)1R2(1)•MDJM At:Q • l ]00 tND • 0 OOO(]r-.~C'>O 0000~6,0 OOML6?0 tJoOOf>r•;~.:!io l)()i)Q~"~fl.,l,l~) ooocf:~,r;o () 0 0 () S~f)O t,)QQr,..; '1(,• Sooo•"lhAO O«)OOS~!:jiJ QOOQ~;..ioO OQ 1l0~7t0 \HHH.I ~ 7 21) Q00()'\7~0 1)0('10'-i i' «O 001'iOr.:i7C:•O ()OOOS 'lbt) .,()1)(),710 noooS7Au I) t) (II) r:; ·; C'>t) oooot;Aoo 0000~~11) ()(\i)Qe:./320 RETURN OO?OSR10 nERUG SU8CHK OOOO~Rao END o~~o5A50 sueROUTINE REOROR(NHL,N,BCONC 1 M,MC,C0 1 COCT,y,AT,AO,eo,L,WQ,coN,UNC~O~Q~~h0 + ON) · 00005870 \0 lro c OOOOSSRO (I(~ () 0 r; p, r 'rr) PARTITION SURROUT!NE **** OO~'QI!'.?<'l(} T~Is ROUTINE PARTiliONS AND REORDERS THE COVARIANCE MATRIX AND THE A ('H)OOI';J~ t 0 8 8A5EO ~ OUTPUTS FROM VfCTO~ ~ COMPLE~•t6 COMPL~~*tb ON R CONCATENAT~D) THIS ROUTINE ARE Ht, co, COCT, y, AT, AO, O~H)()I'.i<7?0 BO M(NtN) 1 MC(NML 1 NML) 1 COCNMLeL) 1 COCT(L,NMl), VCL,l) AT(NHL)r AO(LJ,BO(L),Br,ONCtij) tQMPLt'.X*H• WCHN) c c JNTEG€R CON(N),UNCON(N) oo too c OOOi'l~-i,:;:•p.() t.)QQ()~(ftJQ 0000~0~0 AO(JloWQ(.JJ RO(J~aACONC(J) O(j () 0 6 0 ll \) t<t•t oooQ;,o '~<> t'OOQ~'~O.~O 100 CtlNTJ.NllE OO'J~~~\) 150 0{!001iQ~•O t.;OI)Qiol()8Q OOOOilOilO l)~)t)t;h ]00 DO 5()0 t<tat,~ oo aoo l<i!•t 400 ~b~t~~~.l I ~ Kt,K2) 1500 CONTtNU~ 00 700 l<l•t,NMt 00 bOO K?.•t NM M8(Kt,K~~) -·~(UNCON(Kl),UNC0N(K2)) bOO CNT NU oo eoo c (\()(H)'}I'?MQ oooocieno OO(H)b()lO J•trL 100 C 'NT NU 00 ~PO K~·t,NML cl ooou~~?so ooi)O~~oo I(P•L•t · no 200 t<•I<PSN ATCt(t)•WY(~ UNCON 0< I) aK Kl1:1Kl+l · 200 CONT NUE c ('!()0(1:~9~~0 0 () () () 5 q l4 Q K .at CO(~t,l<2 •'k(UNCONCI<t),K2) 800 CONTINUE qoo CONTINUE DO 1100 Kl•f•L 00 1000 K2• NML , COCT(Kt~l<2) ' DCONJG(C0(~2,Kt)) ooo tO~TPW 100 CUNTlNU RETURN n ~RUG 0 C::HUG Et-H"' SURCHK, INIT(HC,COCT,C0 1 Y,AT,AO,BO,CON,UNCON) SUBCHI< SU6ROUTINE SIMAX(NML 1 L,MC,CO~COCT 1 Y 1 ATlAO,an,MCINV,PROD, + MC NCO,VOUT~ 1 VOUT~ 1 V3,U,v 1 At 1 P8XMA~,aNt) 1 0() ()(H~\Jb 11 0 Oi.Hi0~1t)O () (1 {)I) b l .'.¥: I) (11)(\i)bl t!() 0')1,.'11)1, ~so 00001,,~·1,) (}(1(){)()1 'i" 0 OCH!Oll'IU1 0 0 CHH~ ~ 1 q t'l (t00\H•200 0~)('(}~2' 0 O:')(HJ,..•??.O ('Ql'Ob2' F·O ()0 () ()r:O,j? lll) ()00062':~0 0 0\10 ~.·2 'i-·(1 (,H)i')0~2ll) 00006?.110 0000~?.'91) o<H~o~:;!"'o cooon:H o 02\'t~~~;.;o Ov00t15 ••·0 Q()(H)~~iiQ 0¢ lj (ttl:, 3t:;Q 0 (I 0 0 t,:~ ~n') OOOOI'l·~·;o I) I) 0 0 I? r•}'l 0 000 \~~ }~,n 0 0 00 (I ~LHH) 000064~0 oooo~ur.o 1.0 0'\ C ~ c c c ** I PSIMAX CALCULATION SUBROUTINE HIS ROUTINE CALCULATE~ THE MAXIMUM SIGNAL TO INTFRFF.RENCE RATtO T ALSO OUTPUTS U,V, ~ND At TO AE USED IN HOPT CALt,CSUBAOUTINE): REAL•& PSIMA)( CUMI't.f)(•lh M~(NMLpNML' 1 COCNM~rL) 1 COCTCL,Nt-4t ),VCt.'-L)l + A f NML ) , A 0 u, ) , f.\ oc ) r u t N~i L ) , vCN "'L ) , Dp., ot>1 , . 1, NuMF. R, • ATCU TM(tOll flNi(f'H-ii..,M-1~) R too, COMPLE,~Ib MCfNV NMLtNML (~Q(O(NHL,NML),MCINCO(NHL,L), ~* c + VOlJTt L),vOliT .li,.),V:Ht,.),SOCTAO,vCTAl INVERT MC AND VERIFV MC ~ MC(lNVERSE) • I JFCL.NE,(NML+L•l)) Go TO 101~ ~ ~~x(,qi) a 1,/MC(J,I) 8 l0l5 IF lL,NE (NM)+L•~)) GO TO !000 O(NI~aMCf!'' *~Ctit2)•Mt.Ctr2)•MC(2,t) MCINV!t,i eM,(2 ~ /OrNIM M NV it ••MC( .)/0 ~IM M~INv ~:, ·-~cc 't)/o~NtM ~C NV ~t2 •MC(l, )/DENIM GO qq 1000 00 ~ 0 NOEXat NHL TM( NAFxJ.,..c,IN6tx,t~IMC(t,ai 1010 (":ONTtNUE Nfottl:cNML•t 00 1020 NlNO•J,NMlE N~NOP211NINO+t R ~IN \cTM(NINDPt) 1010 CCNftJUf. CALL ZORN~R,MCINV,NML) on 101 7.Rct,NM Jo 0000~/J~O oovoe£4 'fQ OOOO..,IHH.l ()()(HJ~ltQ(l OOOOb'SOt} (1000~5·1 0 ()e)(){)!'-it.?o (I I) \) () ,(~ ~, ··~ I) Otlt\0~~,)~0 OO{H)t,";,t;o 00006~~6 'I OC0\,6St,?. l)(ii)Qh5~oJ3 0 ~H.I 0 t, S61J QOOQ~Sk6 000Q61)~'1 00006~!'.:;13 000t;'lk5~Q ('J0t)Q6'571 00tJQ.,S1?. O{H.)\)P..~;·;tJ Ot)006SFiO ()Q0t\bS 1~0 OOijQ':.~OQ 0000h610 QOOIJi:l~?.O 0000f.·fi:~.i) OOOOoouo ~OOQt,~SO OOQOQ6b0 0 ~l ~o~or~tJLIO oooo!-.r..1so oo ooe-.~)h 1) 11 ~ p,rh8lf~~;~z1.S~~~~NY(IZR,IZC11MC(I,l) 012 C. NTimJ 0 I C lNT NU ()0006430 (:IO()l,)/>:.6'1(> 001J0~~6AO OOOOM-,1'?0 · CALL CO XMP(HC,NML,MCINV,NML,NML,NML,NML,PRoDrNMLl C **CALCULATE U•MC(INVERSE) •AT (ORDER NMLXl) c qq CALL CDMXMP(MCJNV,NMLrATtNML,NML,NML,l•U,NMtl c C ** CALCULATE V•MC(INvf.RSE)*CO•BO c CALL COM~MP(MCINV,NML,CO,NML,NML,NML,L,MCINCO,NHL) c CALL COMXMP(MCINCO,NML,B0 1 L,NML 1 L1 t 1 V,NML) ~ ** CALCULATE NUMER•BO(CONJ• TRANSPOSE)• (V•BO • COCT•V) c CALt COHXMP(COCTtL,V,NHt,L 1 NMbtltVOUTI,L~ CAL COMXMP(V,L,BO,L,L, ,t,VO 2,L, ('(.\006100 ('10<t()h7t0 ()0{)0.07~(i ()O,:>o~7:~o OO'J0~1t&O ()CH)Ob7SO 0000~1~') 1)01)0~110 1)000~1~0 0000ib7Q() ('QI;('t,~,()() (l()('O~R' 0 4)00Qei)Aj-~1) 00 0 O.f:·R :; 0 ()t){}(ihbtit) OO')Qt.f)f',SO ()00Qot,gf,Q 0000_6Pt70 \,() -....j c 00 t 0 0 K t • t 1. L Vl(~~) • c c c VOuT2(~1) oooo~BfH> • VOUTt(Kt) 100 r.ONT NUE NUME~ • co.o,o~o) DO ~00 Kl•ltl . NUH.R • Nu~t:R + DCONJG(BO(I<l)) " V3(t<l) 200 CON INUE c~ ** CALCULATE DENOMIBO(CONJe TRANSPOSE)*AO •V(CONJ• TRANSPOSE)•AT AOCTAO ~C0 1 0r0~0) VCTAT •(O.u,o.u) c DO ]00 Kl•ltl 80ClAO • BO~TAO + OCONJG(BO(Kl))•AO(KI) 300 CONtlNU[ c DO 400 Kl•t NML . VCTAT • Vr.tlT + DCONJG(V(KI))•AT(Kt) 400 CONTINUE c DENOM 1 80CTAO • VCTAT CALCULATE ALPHAt c ** Al • NU~ERIDENOM c cc ** CALCULATE PSIHAX ATCU • Co.o,O.O) ~ c c c c c 00 500 Kf•t 1 NML 4TCU • 4 CU + OCONJG(AT(Kt))•U(Kl) 500 CONTINUE PSIMAX • ATCU + DCONJG(OENOH)/Al CONTINUE RETURN OtAUG 8U8CHK F. NO c c cc c c ** ().,QOt'~!J,r,,o t) '' 0 1) fi''H> 0 ~0006910 .,(){10~9~~ Q()l)(\~~ :!- 8 ()0006?«-H) ~0Q()t;9~0 oooo~qhCt OOO<H.•~"HO <H' f.l' ('H· q eo 01)01)~'-,t~QO 0 0 0 tff 0 (H) 00001010 000010?.0 t:'~HlQ7QY•O OOOO'IOLIO ooooio~o M>vo1o~o OOC070'70 ~09070$() (H)01')70Q() 0000?100 (\0()1)7 10 ()0007!?.0 oOIJOi' L·~O 000071«0 QOQQ715Q "0•!)071E-lO 00101~"10 00ir01HHl t~OQ07~90 oooo7t-~no ooOI)7?to 00001?~0 OOQ07?!•0 ()(.\007 2t.IO 000fl7?.~S{) oo_ oo72!.. Q 00001210 00007280 oooo7t-!.:)O 00tllj130Q 1.100(17~10 OOQO 1.~?.0 ()O(.I07ll0 ,()()O'I:ott!O Qi')(I0135t) SURROUTINE RHIGH(IND,SEQ,Rt 1 R2,MDt~,I,N) ~0007~~0 00007310 THIS ROUTINE CALCULATES THE HIGHER ORDER R VECTOR~·•R3,R4, ••• ,RMOIMe0001. ~BO OUtPUT GOES TO Rl IF tNDn~ ~ND R2 IS FILLED tF tND~t. 00007Jry0 (t()t)l)1ll00 INTEGER SEQ,RI(N),R2CN) (\Q0()71.110 000071.1?.0 1.0 00 c tFCINO.£Q.I) GO TO 300 INOIO••FILL R2 VECTOR t ** !NO a t rFCS~Q.EQ GO TO tOO ~;~~~(~):lt.HotM~ R2CI)•R2Cl)•MDIH GO TO !;Oo c 100 R2(1) • R2(1)•1 JF(R2(f)eLE.O) R2(J)•R2(J)+MDIM ~ •fJ ~ c GO TO 500 ~ ** JNO • l••PtLL Rl VECTOR JOO tND • 0 tFSS~Q~EQ· GO TO qoo ~~~Atttl!&r!AotM, Rltl)•Rt(J)•MDtH J) c r.o TO ~oo R1(l)1Rl(l)•t IF(Rt(tJ. f..O) Rt(I)•Rt(I)+MDIM C 500 CONTINUE c RETURN 400 ~f:fWG SUBCHK c ~c ~ r.NO SURROUTINE N$1MAX(BCONC 1 WQ 1 M1 WTCONM 1 N1 PStMAx) THIS ROUTINE WILL CALC THE LARGEST CURRENT LoWER ROYND ~OR T~E CASE WH~RE 'LL WEI~HTS •RE CO~STRAINEO TO Ol8CR~IE VALUES. REAL*& PStMAX StRNUM StROEN COMPLEX•t~ M(~,N),BC6NC(N),WQ(N),WTCONM(N),wt,W2 ~ CALCULATE THE CURRENT LOwER BOUND FOR THE SIR * * * WCTRANSPQSE•CONJUGATE)•WQ AND c~ CALCe WQ(TR4N$rUSE•CONJUGATf.)•W c w1 • <o,o,o.o) c DO 100 ll•ltN WI n WI • O~ONJG(BCONCClt~)*WQ(Il) lOO CONTINUE oono?n~~ ooouta~o ooon?~•o 0000/G~? ~0001110 600Q7U80 oooo?a?o ~OOQ1Sno ~0~07~10 000075~0 00007530 00001~00 onuo7cno o6oo7~~o 00?07~70 oooo75Ao on~~1s0o 00007h00 0~007Qt0 00007&~0 00007~~0 00007&00 00007~~0 coryo7~~o 00007b70 Q~001~AO 00007hOO 00001700 00007710 000077~0 oooo77Jo 000077UO 00007750 OOOQ77~~ 00007770 00001?~0 000077QO OC.Oo1~oo oooo1GIO ~0001R~Q 000078~0 ooo~1aoo 00007R~O 00007S&O 0~007870 oooo7eeo 00d07A10 00007QOO 00007?10 000079~0 000079JO 000079«0 ~00079~0 oooo7q~o oooo7Q7o \0 1.0 c c c E ~ ~ O{t007()80 NUMERATOR OF 8lR a Wl t W2 (REAL) SINCE H IS HERMITIAN (' OQ 0 79 <:· 0 ooQoe, MHJ SIRNUM a WI * OCONJGCWI) OOOOi~O l <> t"I?IJt)AO?.O CALCULATE TH£ D!NOMINATOR FOR 8tRCL8 • ~CTRANSPOSE•CONJUGATE) tM * W (1(\(i(jf;l):~t) OOOOP.Ol!O OOOQBQSQ ()000~0f10 CALCULATE W(TRANSPOSE•CONJUG4TE) * M 0¢00·0.\)"fi) 0000AI)P,0 ~HiOQ8Q(}O ooooAt<>o 2~c~~~ t<•t,H C iOO CONi~N~~) a(o.o,O.O) 00 4.,8 l<•t N 12ai,N WICONM(i) a WTCONH(I<~ + DCONJG(BCONCCIZ))•M(l2,~) 300 CO~TINU 000 CONTINO . CALCULATE WCTRANSPOSE•CONJUGATE~ * M * W • SIR DENOMINATOR (REAL) SINCE M IS Hf.RMITIAN SIRDEN a OeO 00 500 K•tfN SIROfN a S RDEN + WTCONH(I<)tBCONC(I<) 00 lO ~ ~ c 0000A110 o~tJ\n~ soo cO~TINlJE c c CALCULATE SIR(CURRENT LOWER ROUND) (REAL) c c c c ~ SINCE H HERMITIAN PStMAX a SJRNUM/SIROEN RETURN Of::AU~ SU8CHt< END SUBROUTINE Z0RN(R,RNt 1 NPt) HERMITIAN MATRIX TOEPLJTZ INVERSION (COMPt..EX) COMPLEXtt& X(20t,,8NJ(NPl 1 NPt),R(lOO),RG,GAt,UGIJ ~-NPl•l N2Pl~t2*N+l X~I~t.• •~O 8S(R(t))•t2 )( N)II•R l) ~ N+?)a CONJG(X(N)l NHtBN .. t 4 . 12:0 OOtH;M1 J.O (:000All10 (~OfrO!:Ii ~~0 0~0{,1~~60 c () 0 1) i~ i 'i' 0 OOOO~U~Q 0000~1CJO 0090A2CQ OOOOP•,~U) (IQQOi~??.O I)IJOQ~I?. ~0 0000~·2!10 0000A2SO 000082~(1 0000P..?10 ()0008?~1) OOQ OP!?'1Q OOOOA300 OOU(H~.\\t 0 (I()O~R:l\?.0 1)0¢(il~fd0 0000$o\:Ht0 oooos:,so OOOQPr~·~:; (IO<H;~>>:i·hQ OOOOP..;~s (' 0 Q0 tJ. 3'1\) {'P)I/0~S7 1 ) (P,IiQ~3~tl OO(HjP> §90 _,OOO~H!.OO r.•OOO~tl11j r,H'J 1J () P. 1.1 ?, Q OOOOR 4.H.I OOOOP.IH!O 00 3 Int,NMt RGa~O. 00 4 Kat,I 0 t: of'H'! -~bO ~GaR(K)tX(NPt•L)+RG 00 1)0~490 Lwl•K+t OOOOflll~O (.\ () Q 0 1'\ 1.1. ., () OOOOf:WRO 1-" 0 0 JPt•l+l OAX••CIHtPt )+PG) ~(NPt•tPtl•GAJIXLI f>(J S J•lrl L•t•J+t ooooesoo . 5 ~~NPt•J~•XCNPt•J~+X(NPl+L)•GAIIXLI DC o J• ,JPt b ~ NPt+J aOCONJG(X(NPt•J)) 3 X~ll1Xl.f•(C0ASSCG•Il••2li)(LI X NPt)•l(Lt X N~~l)•OCONJG(X(l)) R~t trJ)atAIX(N+t) oo .o x~s, ... JaN t•I ANtil•~,t~aX(J~IX(N+tl 20 ANt l 1 +J).•OCONJG(8Nt(l+t 1 t)~ 00 2 •t,Nt.4l tMteY•t no 22 ,hat r t UGJJ~XtN•Jui)*X!N+t+J)•X~2*N+2•~)*X(J) 22 24 6~J~I•t,J•l ~Ni J+I 1 I+~. •AN CitJ\+UG JIX(N• ) •DC NJ~(RNt( +t,J+l ) oooo~:;(;~o 00il0.111·600 ('(10(1861 1) ()(}00"'~}~~0 0 0 6 0 gflt :~. (t (' I) Q {) 8 £dH.) (!OOO~h~·!) cno~H~6ho ooooQtj~,·~o ., tHl p, eo ~~., OOOOAI,'~') (.I.(H)QH'I(i() END Q0008?7() SU8ROUTIN~~ ARFACTtN~wCLA,WOPTcWQIJHETAJtTHfTAS) (\\:l(i(IW;''~~ FTP9 tAI)tFfPT 1etl,FTPQ(I8 ,~,LAMADl GD Rt~OMTt~n ~ CUHPLEX•Jb ~U~RtsVMTLSUMQ 1 TIAtTlT 1 TIQ COMPLEX*16 WCLSlNI,WuPT(NJ,~QtN) L~MJ3()A•l 1802'> Plw3 1 t41~q~65t.t ~O•lsO.OIPo~ 0~•1 OlRO RlAL•~ (}0()€)~720 00t'()87.JQ 04)•30Pli'40 onooR7So 000()~7~() (!OOOR]f'•l) ij i) 0 0 8'1 '? s 0000P.7lHl ~)00087BS ()i){l()P.'/92 00001~ ?95 0 Q 0 (l r-~ ·'HI () ('H>v08i:i.1 t' ooot;RP.r"V 0 0 <HHHi :!? 0 t<ra2,3•PJILAM8DA 001J()P.i1,t10 tPt-~so PH0::~~0 0 tlt.i 0 ~HV~';J 0 10 Pl-i~O u FMA~~D0e0 J'M~XT II 0.0 GOtlO~H~IbO oooo~Rgo 0001l~P.95 Of)0(lH900 I)IJOO~«J!tJJ QO~OH'?O'i 'MA)(~ • 0.0 ()(JJ l () J •• ' 181 onOOiYH {) THO~Fl.OAT(JJ) 000 QP.I') H' J •J•t c OOQ08S~I) •i•"'"' ~~lt};J~=~~A~~~t;~rl3,1,, · RETURN oo a R~AL c QO(!OR'S~Q () () 0 (} "". t; .,, 0 000()"11 0 OEAUG SUBCHK c Q0 !J I) ,(1, ~; t () 0 0 () 0 ~:\ ' :) i? !) I\€) OOP.5 JQ ... ' ,,~ (' p, ·~ !.1 ·\) A(> I; -.gr~"-~.r. '1.1 ,, ,, \! . :;) ., v THaTHO•OR suMa.co,o,o,o) SUMT AUMQ 11 ,o 1 0r0t.O) a(O.u,o,u) (\{HJO~<J;tO OOOO~{';)f.h) OOOOW)'~O 000()H'7h0 OOOOA?bS 00008'HO I-' 0 I-' CTaCOSCTH) 00 15 Nl•l,N 000()F.IQAO Qf\(~4)~~?~0 N~IJNt•l Tt8=W8LBCNt)•~~OS(PltNN•CTl+(O ,t )•S~NCP~•NN•CT)) Ttf•W PT(Nt)• OS(PI•NN•CT +Co• ,t:\•S N(P *NN~CT,) TI~•WQ(N•)•(Cf (PI•NN•CT)+ o., 1•)•SIN PI• NtCT)) StJ~T a StJHT +TtT SUMQ • SUMQ • TtQ 15 SUMA~tSm1"•TtR JF(THD 1 NE.C~OaO•THF.TAJ)) GO 34 ~OBAw2u.~•OLOb10tCDARS(SUHH~ GDHTcZO,O•DLUGlO(COARSCSUHT ) GORQ•ZO O•Ol0GtOCCDARS~5~1MQ ~ W~lTF C~t*) TH!TAJ~Gn9n~GO~T GDBQ 34 JF(tHO,NE.C90•0•TH~xAS) ~0 fo 3S G0BAw2u,O•Dt0h10(~. RSi UH~)) GO~ 8~0 O•D OGIO A~S AUHT) JO c 35 GDB~a2Qi'O•DLOGto~ BAR~ SUMQ)~ ~~~T~ C '*) THF.TAS,G0~8,GOBT,GDBQ CONTINU FTP8(J)•C0-8S!SUMB) FtPT J)wCDA~S SU~T, FTPQ J)•COAA~ SUMO FMax1•AMAX1CF ~~AXf•AMAXt(FHAXT,FTPT ~ ,MAXO•AMAXt(FM4XO,FTPQ S 10 CONTINIJf ,~~N a 4MAXf(FMA~BtFMAMT) FMXN a AMAX CFMXN,fMAX~) WRITE (~r2) FMA~ij ~HAXY FMAXQ 1 FMXN , . 2 'ORM4T (1~ 1 tFMA~~at,f't~.a,ax,tFMAXTat,FJ2.4,tX 1 tFMAXQQt,Ft2.4 1 + 'M trM~~-t Fl2 hl CALL AP Of(FTP8 1 F~XN,,TPT,,MXN,FTPQ,FHXN,PHO) .•xs,ltPBi~,~ S t:Ot..ITINU ~ETUAN Of.:RUG SUBCHK (Nf) c ~UBROUT~NE APLOT(VB VBMAX VJ VTMAXtVO,VQMAX PHOl PEAL VA 1bi),Vf(~btS,YQC3Lt !THO(l8I),F8CtA(),rTC1At),FQ(181\ C4L! on oPt..Jr:tTt t8t THO J)zl~6AT(J•~t) TF ~ .J A l?., A ~~f~~~·~t•O*AlOGIO(YB(J)/VBMAX) tr(FR~t~~·J +70,0) 2,t2,tO FACt~ "J)•• o.o U£. 1o CONTI oo ?.ll J•~,tet tF(VT J) 1~,24 t& lh fT('8~!~J •~o,c•lLOGlO(VT(J)/VTMAX) trCFT IB2•J)•70.0) 2a,~4,20 8 2 24 FT(t8."J\••70.0 2.0 CONTI UE 00<>v'';)oon ~!il)(l.YI) 11) ''V0(,9!H J 0000901?. Q0QQ9(H :; 0000q011' ooooqo2o I) () () Q ? I) U -~ 00009<)1.14 I)OOO'IOl.lS O(){)(}tf(.ll.lh OOtlO"~OlJ'/ t)i(}{II;)<:J(IIJP, (q)Q()9C~·<'t OOOCOifJ'H no 0 ()'?(l Sr! 1 0(Hi 0 it 0 G 3 i)OOQl.JC:O'!i' 1 (\(j(l(l(J{.It"l\\ ()()4)Q9070 0000'H17.? 000()q01~ I)I)QO<~o<?IQ 00009M~~:: OOOQQ(l~tJ fj()Q()'?(iry() t.HlOO"'?(i~P~ ~t)vO"i0:)9H OtHJ2'~(l 0(\ (}0lHJC;I11€> "OO()q~~~ Ol)liv'i''A ~ s 01HliJ(~1?0 i)~{)()(/1$(~ ij;(li.)Q'?L~S OOOO<i1Lit'l (\{}(l(l<f ~ C_;(; 00(' Qt;~ ~ hO OOOOCii '/() OC•OO~ 1 "1'5 ('(,0()9 ~ <'l(l 0000'1~?00 O(H)t~02 t (l (!0009?~0 O<H)OC>i?!lO 00009?!lQ OOQI)9?HO ooGo9e:a 1 oooor-:•,,..p,i!, 000092·.o.L! OI':IO()r,::~?H~i· (\()t)()(/,?P:.\f., (I Q? t~ p, 7 oo ....... 0 N oo too J•t.tel ~~~VQ(J)) 80, 10 ooooqzqt AO OOOO'i2~3 80 tQ t8~•J)a1-~,n•AtOGIO(VQ(J)/VQMAX) tF frh tEI?.•J •7o,o> 120 FQ(lA2•~)•• 0.0 10 ,NTINU ~o, OOOOQ2QQ 120, t.no O+~~LLqCP 0T,~J~~~Dif0t3f'1~ITA COEGREfS)•,ts,7.o,tGAIN CALL f~L~f>-ci,TH6,t6.Q,fTHF.TA CDEGREES)t,t5,7.0,'GAIN + 91t8t r:AS CALL ~PLOf~it,THD,tO;o,•THETA COEGREF.S)t,t5,7.0,~GAIN + 9tl8t F ) . CALL ~PlOf2 t 1 THD,IO,O,•THETA CD£GR£!S)t,t5,7.o,tGAlN + q,_tet,F , 1000 RETURN I.'E:OUG SU8CHt< END (tt)OQ9;?Qfj OOOOG'.?.Qb ()()Q ~Q~'-)f (08)' 1 I)I.H>093?0 COB)•, (0B)t 1 CDB)t, ()OOQq~;i~ tHJOf.l9'~2 t t'OOQ9;l2'~ (';0(!(19!,~ I 00009~78 ()()(t()Q ~HO ()i'JQQQ3)t OOOOQJJ:S OOl•OQIJOO OOOOQtHl l 1-' 0 w