Conformal polarization array adaptive beam forming method based on sparse recovery

Through the adaptive beamforming method of sparse recovery and dynamic parameter norm regularization, the problem of signal polarization parameter changes and DOA mismatch in the conformal polarization array is solved, and the accurate signal output of the conformal polarization array is achieved and Interference suppression improves signal-to-noise ratio and robustness of signal output.

CN119093982APending Publication Date: 2024-12-06UNIV OF ELECTRONICS SCI & TECH OF CHINA
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411094318.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-09
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

The prior art fails to effectively deal with the problem of signal polarization parameter changes and DOA mismatch in conformal polarization arrays, resulting in a degradation of adaptive beamforming performance.

Method used

Adaptive beamforming method based on sparse recovery is adopted, and the wave angle of the incoming signal is roughly estimated by using singular value decomposition and orthogonal matching tracking algorithm (OMP), combined with iterative cycles of dynamic parameter norm regularization criterion, to accurately estimate the wave reach of the incident signal. The angle sum of incoming wave signal data is used to reconstruct the interference covariance matrix, and the projection matrix of the orthogonal complementary space of the interference signal subspace is obtained to suppress the interference signal.

Benefits of technology

In the case where the airspace-polarization domain parameters are unknown, the accurate output of the signal-to-noise ratio of the signal-to-noise ratio of the signal-to-noise is improved, and it has good robustness, strong adaptability and meets the actual needs of the engineering.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119093982A_ABST
    Figure CN119093982A_ABST
Patent Text Reader

Abstract

The invention relates to a radar communication technology, and provides a conformal polarization array adaptive beam forming method based on sparse recovery when spatial domain-polarization domain parameters are unknown. According to the method, an orthogonal matching pursuit (OMP) algorithm is firstly utilized to roughly estimate an incoming wave angle of a signal and an incident signal, then a dynamic parameter l1 norm regularization criterion is utilized to further accurately estimate a direction of arrival of the incident signal and incoming wave signal data, a covariance matrix of an interference signal is reconstructed, and an optimal weight vector is solved to perform beam forming. Polarization parameters of incoming wave signals do not need to be known, good robustness is achieved for DOA mismatch, and engineering actual requirements are better met.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to radar communication technology, and in particular relates to conformal polarization array adaptive beamforming technology. Background Art

[0002] Array optimization refers to the design of the position and excitation parameters of the antenna elements to obtain an antenna array that meets engineering needs. The periodic array beam pattern can be easily obtained through mathematical analysis, while the conformal polarization array elements point in different directions and there are factors such as the coupling of polarization factors and spatial factors. Therefore, the traditional adaptive beamforming method performs poorly when applied to conformal polarization arrays.

[0003] Kim proposed an adaptive beamforming method based on characteristic subspace, also known as principal component beamforming algorithm (ESB), (Kim JW, Un C KA robust adaptive array based on signal subspace approach [J]. IEEE Transactions on Signal Processing, 1993, 41: 3166-33171.) This method projects the desired signal steering vector into the reduced rank subspace composed of signal and interference, reducing the influence of the noise subspace, thereby improving the output signal to noise ratio. However, this method requires the number of known sources, and the performance will be reduced when the subspace division is not accurate enough. Vaskelainen optimizes the excitation of each antenna array element based on the least squares method to obtain the sum and difference beam patterns. (See the literature: Vaskelainen L. Constrained Least-Lquares Optimization in Conformal Array Antenna Synthesis [J]. IEEE Transactions on Antennas and Propagation, 2007, 55 (3): 859-867.) Wang proposed a robust beamforming algorithm based on diagonal loading (see the literature: Z. Wang, Y. Wang, Z. Li. An Improved Robust Adaptive Beamforming Algorithm Based on the Conformal Array [J]. 2015.) Compared with the generalized sidelobe cancellation and Capon beamforming algorithms, it can better suppress interference signals and has a lower sidelobe level.

[0004] The existing literature usually ignores the polarization parameters of the signal or regards them as known parameters. However, in actual scenarios, the polarization parameters of the signal may change after being scattered by the target, and the signal's incoming wave angle may also have DOA mismatch. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide an adaptive beamforming method for a conformal polarization array when spatial-polarization domain parameters are unknown.

[0006] The technical problem adopted to solve the above technical problem is a conformal polarization array adaptive beamforming method based on sparse recovery, comprising the steps of:

[0007] Step 1) using a snapshot number N to receive signals from K narrowband far-field signals incident on a conformal polarization array, and obtaining array received data X; performing singular value decomposition on X to obtain an array received signal data matrix X' after singular value decomposition;

[0008] Step 2) Divide the pitch angle θ of the entire airspace into S discrete angles [θ1, θ2, …θ S ], and the azimuth Divided into Q discrete angles Constructing an overcomplete dictionary matrix Set the number of repeated iterations to the number of incident narrowband far-field signals K, and the number of iterations variable k = 1, 2, ... K; in each iteration calculation, each atom in the overcomplete dictionary matrix is and the k-1th residual R k-1 Perform correlation operations and calculate the l2 norm to find the most correlated atom and will Add to the optimal set of atoms at the kth iteration Update the residual R of the kth iteration k ; After reaching the specified number of iterations K, the optimal atomic set is obtained and the residual R K ;

[0009] Step 3) Based on dynamic parameters The iterative operation of the norm regularization criterion obtains the final optimal atomic set is the set of arrival angles of K narrowband far-field signals;

[0010] Step 4) Reconstruct the interference covariance matrix

[0011] First, based on the optimal atomic set Estimated incident signal data matrix

[0012] Calculate the interference covariance matrix Wherein, the subscript i represents interference; each interference The optimal set of atoms The atom corresponding to the kth interference in is the incident signal data matrix The kth and k+1th rows of ;

[0013] Step 5) For the interference covariance matrix Perform eigendecomposition and take the interference eigenvector U corresponding to M-(K-1) small eigenvalues i Construct the corresponding projection matrix H represents conjugate transpose;

[0014] Step 6) Each snapshot received signal x(n) is projected through the matrix to obtain the interference-suppressed signal plus noise output y(n)=Dx(n), and the covariance matrix of y(n) is R y =DE{x(n)x H (n)}D H , the covariance matrix of y(n) is

[0015] Step 7) For the covariance matrix Perform eigenvalue decomposition and take the eigenvector corresponding to the largest eigenvalue as u s , and obtain the optimal weight vector w for adaptive beamforming opt , w opt =D H u s , using w opt Complete beamforming.

[0016] The applicant analyzed the common problems existing in the current field of conformal polarization array adaptive beamforming: the possibility that the polarization parameters of the signal will change after being scattered by the target is not considered, and the DOA mismatch is not considered. To address this problem, we established a sparse signal model and used the orthogonal matching pursuit (OMP) algorithm to roughly estimate the azimuth and elevation angles, and then further used dynamic parameters to calculate the DOA mismatch. The norm regularized DPL1 is iterated to obtain an accurate estimate of the incoming wave direction and reconstruct the interference covariance matrix. Based on the matrix, the projection matrix of the orthogonal complement space of the interference signal subspace is obtained. The array receiving data is passed through the projection matrix to suppress interference, and then filtered by the maximum signal-to-noise ratio criterion to obtain the optimal expected signal output when the spatial-polarization domain parameters are unknown.

[0017] The present invention discloses an adaptive beamforming algorithm based on sparse recovery for conformal polarization arrays in the case of unknown spatial and polarization domains. The algorithm first uses the orthogonal matching pursuit (OMP) algorithm to roughly estimate the signal's arrival angle and the incident signal, and then uses dynamic parameters to The norm regularization criterion further accurately estimates the angle of arrival of the incident signal and the incoming signal data, reconstructs the covariance matrix of the interference signal, and finds the optimal weight vector for beamforming. Compared with other algorithms, this algorithm does not need to know the polarization parameters of the incoming signal. At the same time, since the angle of arrival of the incoming signal is accurately estimated, it also has good robustness to DOA mismatch, which is more in line with actual engineering needs. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 is a flow chart of the method of the present invention;

[0019] Figure 2 It is a schematic diagram of a hemispherical conformal polarization array of the present invention;

[0020] Figure 3 It is the residual iteration graph of the DPL1 algorithm;

[0021] Figure 4 is the beam pattern based on the method of the present invention in Experiment 1;

[0022] Figure 5 It is a curve diagram of output SINR changing with signal-to-noise ratio;

[0023] Figure 6 The output SINR changes with the number of snapshots;

[0024] Figure 7 It is a curve diagram of output SINR changing with DOA mismatch (polarization matching);

[0025] Figure 8 This is a curve diagram showing the change of output SINR with DOA mismatch. DETAILED DESCRIPTION

[0026] The hemispherical conformal polarization array of the embodiment is as follows Figure 2 As shown, the adaptive beamforming method adapted to the conformal polarization array comprises the following steps:

[0027] Step 1) Perform singular value decomposition on the multi-snapshot data received by the array:

[0028] Assume that K narrowband far-field signals are incident on a conformal polarization array with M elements and N snapshots, then the array receives data X∈C M×N The definition is as follows

[0029]

[0030] is the orientation vector matrix, is the incident signal data matrix, is the noise matrix;

[0031]

[0032] ) represents the signal arrival angle, θ is the elevation angle of the signal arrival, is the azimuth of the signal arrival; (γ,η) represents the polarization phase descriptor of the signal, γ is the polarization angle, η is the polarization phase difference; represents the angle of arrival and polarization phase descriptor of the Kth narrowband far-field signal; the steering vector in is the spatial steering vector, is the steering vector in the polarization domain, and ⊙ represents the Hadamard product. It can be seen that the steering vector The spatial domain parameters are coupled with the polarization parameters, so the steering vector Rewritten as:

[0033]

[0034] in,

[0035]

[0036] [G] m,: =[g m,x g m,y g m,z ]

[0037]

[0038] in, is an intermediate quantity, diag is a diagonal matrix function, G is the polarization response of the array, and the response of each array element is only related to the array element position and has nothing to do with the signal incident angle. [G] m,: represents all elements on the mth row of G, is the conversion matrix from polarized coordinate system to rectangular coordinate system, e(γ,η) is the polarization parameter of the signal, and after decoupling, we can see the intermediate quantity It has nothing to do with the polarization parameters of the signal, but only with the angle of the signal. The multi-snap data X received by the array can be re-expressed as:

[0039]

[0040] in,

[0041]

[0042] in is a 2×1 dimensional vector.

[0043] The dimension of the received data matrix is ​​reduced by singular value decomposition (SVD), thereby reducing the amount of calculation for subsequent steps. Perform singular value decomposition on the received data matrix and assume that there are K large singular values. Perform singular value decomposition on the multi-snapshot data X as X = UΛV H , U is the left singular matrix of X, Λ is the singular value matrix of X, V is the right singular matrix of X, H represents conjugate transpose, then the array sparse signal model after reducing the dimension of the received data matrix X is The corresponding sparse incident signal is expressed as The right singular matrix D of the reduced dimension data X' K =[I K 0] T , I K is the K-dimensional identity matrix, 0 is the K×(NK)-dimensional zero matrix, T is transposed.

[0044]

[0045] in,

[0046]

[0047] [G] m,: =[g m,x g m,y g m,z ]

[0048]

[0049] Step 2) Use the OMP algorithm to roughly estimate the signal's angle of arrival and the incoming signal data matrix

[0050] The core of orthogonal matching pursuit is to continuously reconstruct the signal through iteration. Each iteration ensures that the vector corresponding to the selected spatial grid point has the strongest correlation with the residual. The atoms corresponding to the spatial grid point with the strongest correlation are added to the optimal atom set. The estimated value of the sparse signal is updated using the least squares principle. The residual is updated based on the estimated value of the sparse signal, and the iteration is stopped when the specified number K is reached.

[0051] The entire airspace pitch angle is divided into S discrete angles [θ1,…,θ s ,…,θ S ], and divide the azimuth into Q discrete angles Constructing an overcomplete dictionary matrix

[0052]

[0053] The array sparse signal model X' is expressed as follows:

[0054]

[0055] Among them, the sparse signal It is expressed as:

[0056]

[0057] The arrival angle of the airspace is The kth snapshot of the incoming signal containing polarization information is It is sparse. If there is an incident signal in this direction, then The vector represented by is non-zero, otherwise the position should be 0. The compressed sensing theory uses the l0 norm optimization solution, and reconstructs the original signal after sparse sampling. Its model is:

[0058]

[0059] ||·||0 is the l0 norm, the initial residual R is defined as the array sparse signal model X' after singular value decomposition, and the initial optimal atomic set A is an empty set. The specified number of iterations K is set according to the number of narrowband far-field signals incident on the array elements, and the iteration variable k=0,1,…,K. In the kth iteration:

[0060] In the kth iteration, each atom and the k-1th residual R k-1 Perform correlation operations and calculate the l2 norm, and select the atom with the strongest correlation as the corresponding solution

[0061]

[0062] Among them, ||·||2 is the l2 norm; arg max represents the variable when the output objective function is the maximum;

[0063] The signal estimate The corresponding atom is added to the optimal atom set A k middle:

[0064]

[0065] Solve the sparse incident signal based on the following formula:

[0066]

[0067] The optimal solution for sparse incident signals can be obtained using the least squares method:

[0068]

[0069] Update residual R k :

[0070]

[0071] The next iteration is performed until the specified number of iterations K is reached. The optimal atomic set A is obtained at this time. K And the residual matrix R at this time K And the estimated value of the incoming signal data matrix

[0072]

[0073] Step 3) Based on dynamic parameters The norm regularization criterion further estimates the angle of arrival of the incident signal and accurately reconstructs the incident signal.

[0074] It can be seen that the OMP algorithm has limited angle estimation accuracy and cannot handle the situation where the angle is out of grid. Therefore, the dynamic parameters The norm-regularized DOA estimation algorithm is applied to the conformal polarization array. The rough estimates of the azimuth and elevation angles estimated by the two-dimensional OMP algorithm are used as the initial values ​​of the iterative process. The norm regularization criterion is iteratively solved to obtain the accurate estimate of the incoming wave signal.

[0075] According to the sparse signal model, the following optimization problem is established:

[0076]

[0077] ε represents the possible noise energy. Norm minimization alternative The norm is minimized, and the above problem is further transformed into:

[0078]

[0079] The rough estimates of the azimuth and elevation angles estimated by the OMP algorithm are: Corresponding dictionary matrix The overcomplete dictionary matrix Replace with A K , thereby reducing the dimension of the sparse signal. As the initial iteration value, the true angle is continuously approached through iterative optimization.

[0080] Introducing intermediate variables The above problem can be expressed as:

[0081]

[0082] Using the internal method, the optimization function is constructed as follows:

[0083]

[0084] Define the gradient vector w of K azimuth angles θ =[w θ,1 ,…,w θ,k ,…,w θ,K ] Τ , the gradient vector of K pitch angles In each iteration, for each azimuth The gradient w θ,k and each pitch angle Gradient The calculations are as follows:

[0085]

[0086] Therefore, we only need to solve as well as The gradient can be found.

[0087] First calculate and

[0088]

[0089] Among them, Re(·) represents the operation of taking the real part, and tr(·) represents the trace of the matrix. Based on the reweighting method, it can be approximately The norm is replaced by the weighted Norm in:

[0090]

[0091] Further:

[0092]

[0093] From this we can know and The solution depends on and Known Taking the derivatives of both sides with respect to azimuth and elevation, we get

[0094]

[0095] right Taking the derivative of both sides, we get and

[0096] Step 4) Reconstruct the interference covariance matrix

[0097] If based on dynamic parameters The number of iterations of the norm regularization criterion is T, and the accurate estimation of the signal azimuth and elevation angle is obtained. And the corresponding optimal set of atoms But at this time It is a sparse signal after SVD decomposition, so it is necessary to re-estimate the sparse signal based on the obtained optimal atomic set

[0098]

[0099] The optimal angle estimate is Among them, K-1 are interference signals. Calculate the sample covariance matrix of each interference, and then accumulate the covariance matrices of different interference signals.

[0100]

[0101] in, is the optimal set of atoms The atom corresponding to the kth interference in is the incident signal data matrix The 2k-1th and 2kth rows of .

[0102] Step 5) For the interference covariance matrix Perform eigendecomposition and obtain the projection matrix of the orthogonal complement space of the interference subspace.

[0103] Interference covariance matrix Perform eigendecomposition and take the eigenvector U corresponding to M-(K-1) small eigenvalues i The noise subspace of Zhang Cheng is also the orthogonal complement of the interference subspace, and the corresponding projection matrix D is constructed:

[0104]

[0105] Step 6) Let the received signal x(n) pass through the projection matrix to obtain the signal plus noise output y(n) after interference suppression, and its covariance matrix

[0106]

[0107] The autocorrelation matrix of y(n) is:

[0108] R y =E{y(n)y H (n)}=E{Dx(n)x H (n)DH}=DE{x(n)x H (n)}D H =DR x D H

[0109] E{·} means to find the expectation, R x =E{x(n)x H (n)};

[0110] Therefore, the covariance matrix of y(n) is estimated as:

[0111]

[0112] Step 7) Get the equivalent weight vector w opt

[0113] Pair Matrix Perform eigenvalue decomposition and take the eigenvector corresponding to the largest eigenvalue as u s , get the optimal weight vector w opt :

[0114] w opt =D H u s

[0115] Finally, using w opt Complete beamforming, the final output of adaptive beamforming z(n)

[0116]

[0117] The specific implementation manner and working principle of the present invention are further described in detail below with reference to the accompanying drawings.

[0118] like Figure 1 The flowchart of the method for adaptive beamforming of a conformal polarization array when spatial-polarization domain parameters are unknown is shown, which specifically includes the following steps:

[0119] Step 1: Perform singular value decomposition on the multi-snapshot data received by the array

[0120] Assume that K narrowband far-field signals are incident on a hemispherical conformal polarization array with M elements and N snapshots. Then the array receives data The definition is as follows:

[0121]

[0122] is the orientation vector matrix, is the incident signal data matrix, S is composed of the incident signal data s k Composition, each s kContains N snapshot data s k (n), k=1,2,..K, n=1,2,..N, is the noise matrix;

[0123]

[0124] Steering vector in is the spatial steering vector, is the polarization domain steering vector; steering vector The spatial domain parameters are coupled with the polarization parameters, so the steering vector Rewritten as: is the polarization parameter of the signal, The signal's arrival angle parameter.

[0125] The multi-snap data received by the array is re-expressed as:

[0126]

[0127] in, The angle parameters of the K signals constitute, Each element in For K signals Polarization parameter e(γ k ,η k ) and N snapshot signal data s of K incident signals k (n) is a 2×1 dimensional vector.

[0128] The dimension of the received data matrix is ​​reduced by singular value decomposition (SVD). The received data matrix is ​​subjected to singular value decomposition and has K large singular values. Sparse incident signal Where D K =[I K 0] T .

[0129] Step 2: Use the OMP algorithm to roughly estimate the signal's angle of arrival and the incoming signal data matrix

[0130] The spatial pitch angle is divided into S discrete angles [θ1,θ2,…θ S ], azimuth angle divided into Q discrete angles Calculate the atoms corresponding to each angle Constructing dictionary matrix s=1,2,..S,q=1,2,..Q.

[0131] The array sparse signal model is expressed as follows:

[0132]

[0133] Among them, the sparse signal Elements The arrival angle of the airspace is The kth snapshot of the incoming signal containing polarization information.

[0134] The specified number of iterations K is set according to the number of narrowband far-field signals incident on the array elements, and the iteration variable k=0,1,…,K. In the kth iteration:

[0135] 2-1) Each atom in the overcomplete dictionary matrix and the k-1th residual R k-1 Perform correlation operations and calculate the l2 norm, and select the atom with the strongest correlation as the signal estimate of the kth iteration

[0136]

[0137] 2-2) The signal estimation value The corresponding atom Add to the optimal atom set A of the kth iteration k middle:

[0138]

[0139] 2-3) Using the optimal atomic set A k Update the estimated value of the incoming signal data matrix for the kth iteration

[0140]

[0141] 2-4) Update the residual R of the kth iteration k :

[0142]

[0143] 2-5) Before reaching the specified number of iterations K, update k = k + 1 and repeat steps 2-1) to 2-4). When the specified number of iterations K is reached, the optimal atomic set A is obtained. K And the residual matrix R at this time K And the estimated value of the incoming signal data matrix

[0144]

[0145] Step 3) Based on dynamic parameters The norm regularization criterion further estimates the angle of arrival of the incident signal and the estimated value of the incoming signal data matrix

[0146] The residual R after the OMP algorithm iteration K As a dynamic parameter Initial value of the residual of the norm regularization criterion estimate The corresponding optimal atomic set A K Iteration initial value Incoming signal data matrix estimate As the initial value of the iteration Set the iteration to be repeated T times, with iteration variable t = 1, 2, ..., T. In the tth iteration, perform the following steps for the kth estimated angle:

[0147] 3-1) Calculate the optimal atomic set at the kth estimated angle at the tth iteration The partial differential and

[0148]

[0149] Then there is

[0150]

[0151] because

[0152]

[0153] therefore

[0154]

[0155] Same reason

[0156]

[0157] 3-2) Calculate the estimated value of the incoming signal data matrix at the kth estimated angle at the tth iteration The partial differential and

[0158]

[0159] 3-3) Calculate the residual at the kth estimated angle at the tth iteration The partial differential and

[0160]

[0161] 3-4) Calculation of Partial differential of norm and

[0162]

[0163] in, for norm, Re is the real part, tr is the trace of the matrix; intermediate quantity I2 is the 2×2 identity matrix, represents the Kronecker product;

[0164] 3-5) Calculation of Partial differential of the square of the norm and

[0165]

[0166]

[0167] 3-6) Calculate the gradient w θ,k and

[0168]

[0169] ε is an introduced constant;

[0170] 3-7) Update the k-th angle estimate of the t+1-th iteration

[0171]

[0172] γ is the step size of each iteration, and |·| is the absolute value.

[0173] 3-8) For each angle, perform steps 3-1 to 3-7 until the t+1th estimated value of K angles is updated.

[0174] 3-9) Update the t+1th optimal atomic set

[0175]

[0176] 3-10) Update the estimated value of the data matrix of the t+1th incoming wave signal

[0177]

[0178] 3-11) Update t = t + 1 and repeat steps 3-1) to 3-10) until the number of iterations t reaches T, and the optimal atomic set is obtained. Go to step 4).

[0179]

[0180] Step 4: Reconstruct the noise covariance matrix

[0181] 4-1) Based on the optimal atomic set To estimate the incident signal data matrix

[0182]

[0183] 4-2) Calculate the covariance matrix of the interference signal

[0184] in, is the optimal set of atoms The atom corresponding to the kth interference in is the incident signal data matrix The kth and k+1th rows of .

[0185] Step 5: Interference covariance matrix Perform eigendecomposition and take the eigenvector U corresponding to the small eigenvalue i Zhang Cheng's noise subspace is also the orthogonal complement of the interference subspace, and the corresponding projection matrix is ​​constructed.

[0186] Step 6: Let the signal pass through the projection matrix D to obtain the signal plus noise output y(n) after interference is suppressed. The autocorrelation matrix of y(n) is R y , the covariance matrix of y(n) is

[0187] Step 7: Perform eigenvalue decomposition, take the eigenvector corresponding to the maximum eigenvalue, and obtain the expected signal subspace u s The weight vector is The optimal weight vector of the filter is Final output of filter adaptive beamforming

[0188] In order to make the purpose, technical scheme and technical effect of the present invention clearer, the present invention is further described in detail through simulation experiments.

[0189] Simulation experiment 1: The conformal array model is set to a hemispherical surface, the array element uses a matrix microstrip antenna, the hemispherical radius R = 4λ, the arc length between adjacent circular rings l = 0.5λ, and the adjacent interval of the same circular ring is d c =0.5λ, λ is the signal wavelength, and the total number of array elements N = 420. Assume that three incoherent polarization signals are incident on the spherical conformal array, and the parameters of the desired signal are The relevant parameters of the two interference signals are Signal-to-noise ratio SNR = 15dB, interference-to-noise ratio INR = 20dB. Number of snapshots N s = 2000. The sampling intervals of the azimuth and elevation angles of the overcomplete dictionary matrix in the OMP algorithm are both 1°. The gradient descent iteration step size γ in the DPL algorithm is 0.005, the number of iterations is 200, and the residual iteration diagram is as follows: Figure 3 As shown, the beam pattern is Figure 4 shown.

[0190] The results of angle estimation using the OMP rough estimation and DPL estimation are shown in the following table:

[0191]

[0192] It can be seen from the table that the OMP method can roughly estimate the angle of the incoming wave signal, but due to the limited accuracy of the OMP sampling interval, the angle is out of grid. The DPL algorithm is used to further estimate the incoming wave angle based on the OMP estimation result. From the estimation results, it can be seen that the angle estimation result of DPL is very close to the actual incoming wave angle. From the residual iteration diagram, it can be seen that the DPL algorithm has converged after 100 iterations and has good convergence. From the beam diagram, it can be seen that an obvious null is formed in the direction of the interference. Although the beam pointing is somewhat deviated from the angle of the expected signal, the gain in the direction of the expected signal is much greater than the interference gain.

[0193] Simulation experiment 2: In order to verify the impact of SNR on algorithm performance, the following simulation experiment was conducted. The signal-to-noise ratio range was -10dB to 30dB, the interference-to-noise ratio INR = 10dB, and the number of snapshots N s =3000. The ideal polarization parameters of the desired signal and the interference signal are (γ i ,η i )=(0,0),i=1,2,3, the actual polarization parameters of each signal And the azimuth and elevation angles of the actual incoming signal angle and the expected signal angle have a random deviation of [-5,5]. The adaptive beamforming algorithm proposed in this paper is denoted as "proposed beamformer" and compared with other algorithms. The sampling intervals of the azimuth and elevation angles of the overcomplete dictionary matrix in the OMP algorithm are both 1°. In the DPL algorithm, the gradient descent iteration step size γ=0.005, the number of iterations is 200 times, and 100 independent simulation experiments are performed. The output SINR changes with the input SNR curve as shown in the figure. Figure 5 shown.

[0194] From the results, we can see that at low signal-to-noise ratios, the SINR of the MVDR and DL methods is very low due to the unknown spatial-polarization domain parameters. As the signal-to-noise ratio increases, the expected signal of the MVDR and ESB algorithms is treated as interference suppression, and signal self-cancellation occurs, so the SINR decreases as the SNR increases. Although the DL output SINR increases as the signal-to-noise ratio increases, it is still not as good as the optimal beam performance. This algorithm reconstructs the interference covariance matrix and does not need to know the polarization parameters of the signal. In addition, since the signal arrival angle is accurately estimated, the angle deviation can be well handled, so the output SINR is very close to the SINR of the optimal weight vector.

[0195] Simulation experiment 3: In order to verify the impact of the number of snapshots on the performance of the algorithm, the following simulation experiment was conducted with the signal-to-noise ratio SNR = 15dB, the interference-to-noise ratio INR = 20dB, and the number of snapshots N s The range of variation is 500 to 5000. The ideal polarization parameters of the desired signal and the interference signal are (γ i ,η i )=(0,0),i=1,2,3, the actual polarization parameters of each signal The proposed method is denoted as "proposed beamformer" and compared with other algorithms. The sampling intervals of the azimuth and elevation angles of the overcomplete dictionary matrix in the OMP algorithm are both 1°. The gradient descent iteration step size γ in the DPL algorithm is 0.005, the number of iterations is 200, and 100 independent simulation experiments are performed. The output SINR changes with the number of input snapshots as shown in the figure. Figure 6 shown.

[0196] From the results, we can see that each algorithm has good convergence under a small number of snapshots. Since DPL accurately estimates the angle of arrival of the signal, this algorithm can also handle angle deviations very well, and its performance is very close to the optimal algorithm. The performance of other algorithms is inferior to this algorithm due to unknown spatial-polarization domain parameters.

[0197] Simulation experiment 4: In order to verify the impact of DOA mismatch on algorithm performance, the following simulation experiment is conducted. The DOA mismatch range is 0° to 4°, with an interval of 0.25°. SNR = 5dB, INR = 10dB, and the number of snapshots N s =3000. The actual polarization parameters of the signal are consistent with the expected parameters The proposed method is denoted as "proposed beamformer" and compared with other algorithms. The sampling intervals of the azimuth and elevation angles of the overcomplete dictionary matrix in the OMP algorithm are both 1°. The gradient descent iteration step size γ in the DPL algorithm is 0.005, the number of iterations is 200, and 100 independent simulation experiments are performed. The output SINR curve with DOA mismatch is shown in the figure. Figure 7 shown.

[0198] From the results, it can be seen that when the polarization is consistent with the actual parameters, MVDR is very sensitive to angle error, and the output SINR decreases significantly with the angle error. As the mismatch angle increases, it can be seen that the other algorithms also show varying degrees of decline. This algorithm accurately estimates the angle of arrival of the signal, and its performance is very close to the optimal algorithm, which verifies the robustness of this algorithm to DOA mismatch.

[0199] Simulation experiment 6: In order to verify the impact of DOA mismatch on algorithm performance under unknown polarization conditions, the following simulation experiment was conducted. The DOA mismatch range was 0° to 4°, with an interval of 0.25°. SNR = 5dB, INR = 10dB, and the number of snapshots N s =3000. The ideal polarization parameters of the desired signal and the interference signal are (γ i ,η i )=(0,0),i=1,2,3, the actual polarization parameters of each signal The proposed method is denoted as "proposed beamformer" and compared with other algorithms. The sampling intervals of the azimuth and elevation angles of the overcomplete dictionary matrix in the OMP algorithm are both 1°. The gradient descent iteration step size γ in the DPL algorithm is 0.005, the number of iterations is 200, and 100 independent simulation experiments are performed. The output SINR curve with DOA mismatch is shown in the figure. Figure 8 shown.

[0200] From the results, we can see that the impact of unknown polarization is much greater than that of DOA mismatch. Even in the absence of DOA mismatch, the SINR of other algorithms is far inferior to that of this algorithm, and the performance decreases with the increase of DOA angle mismatch. Since this algorithm accurately estimates the signal arrival angle and interference covariance matrix, its performance is very close to the optimal algorithm.

[0201] In summary, it can be seen that for the unknown spatial-polarization domain parameters, this method uses the orthogonal matching pursuit algorithm (OMP) and the DPL algorithm to accurately estimate the signal's arrival angle, and uses the least squares criterion to estimate the signal arrival data based on the estimated angle to reconstruct the interference covariance matrix. From the simulation results, it can be seen that this algorithm shows good robustness when DOA mismatch and polarization parameters are unknown, and its performance is far better than other algorithms and close to the performance of the optimal weight vector.

[0202] The above description is only a specific implementation mode of the present invention. Any feature disclosed in this specification, unless otherwise stated, can be replaced by other alternative features that are equivalent or have similar purposes; all disclosed features, or all steps in the methods or processes, except for mutually exclusive features and / or steps, can be combined in any way; any non-essential additions and substitutions made by technicians in this field based on the technical features of the technical solution of the present invention belong to the protection scope of the present invention.

Claims

1. A conformal polarization array adaptive beamforming method based on sparse recovery, characterized in that: The following steps are involved: Step 1) K narrowband far-field signals are incident on a conformal polarization array with M array elements and N snapshots, and array received data X is obtained; singular value decomposition is performed on X to obtain a singular value decomposed array received signal data matrix X'; Step 2) Use the orthogonal matching pursuit (OMP) algorithm to roughly estimate the signal's angle of arrival and the incoming signal data matrix The pitch angle θ of the entire airspace is divided into S discrete angles [θ1, θ2, …θ S ], and the azimuth Divided into Q discrete angles Constructing an overcomplete dictionary matrix Set the number of repeated iterations to the number of incident narrowband far-field signals K, and the number of iterations variable k = 1, 2, ... K; in each iteration calculation, each atom in the overcomplete dictionary matrix is and the k-1th residual R k-1 Perform correlation operations and calculate the l2 norm to find the most correlated atom and will Add to the optimal set of atoms at iteration k Update the residual R of the kth iteration k ; After reaching the specified number of iterations K, the optimal atomic set is obtained and the residual R K ; Step 3) Based on the rough estimation of the signal angle using the OMP algorithm, the optimal atomic set of the final incoming wave signal is obtained by iterating T times based on the dynamic parameter l1 norm regularization criterion. is the set of arrival angles of K narrowband far-field signals; Step 4) Reconstruct the interference covariance matrix First, based on the optimal atomic set Estimated incident signal data matrix Calculate the interference covariance matrix set up The first K-1 of them are interference signals, so the interference covariance matrix is Wherein, the subscript i represents interference; each interference The optimal set of atoms The atom corresponding to the kth interference in is the incident signal data matrix The 2k-1th row and the nth column of the 2kth row; Step 5) For the interference covariance matrix Perform eigendecomposition and take the interference eigenvector U corresponding to M-(K-1) small eigenvalues i Construct the corresponding projection matrix H represents conjugate transpose; Step 6) The received signal x(n) is obtained through the projection matrix to obtain the signal plus noise output y(n)=Dx(n) after interference suppression. The covariance matrix of y(n) is R y =DE{x(n)x H (n)}D H =DR x D H , so the covariance matrix of y(n) is estimated to be Step 7) Estimate the covariance matrix Perform eigenvalue decomposition and take the eigenvector corresponding to the largest eigenvalue as u s , and obtain the optimal weight vector w for adaptive beamforming opt , w opt =D H u s , using w opt Complete beamforming.

2. The method according to claim 1, characterized in that: The array received signal data matrix X'=UΛD after singular value decomposition K , U is the left singular matrix, Λ is the singular value matrix, D K is a right singular matrix, D K =[I K 0] T , I K is the K-dimensional identity matrix, 0 is the K×(NK)-dimensional zero matrix, and T is the transpose.

3. The method according to claim 1, characterized in that: The iterative calculation in step 2) defines the initial residual R as the array received signal data matrix X' after singular value decomposition, and the initial optimal atomic set A is an empty set.

4. The method according to claim 1 or 3, characterized in that: Step 2) The specific steps for performing the kth iteration are: 2-1) Each atom in the overcomplete dictionary matrix and the k-1th residual R k-1 Perform correlation operations and calculate the l2 norm, and select the atom with the strongest correlation as the signal estimate of the kth iteration Among them, ||·||2 is the l2 norm; 2-2) The signal estimation value The corresponding atom Add to the optimal atom set A of the kth iteration k middle: 2-3) Using the optimal atomic set A k Update the estimated value of the incoming signal data matrix for the kth iteration 2-4) Update the residual R of the kth iteration k : 2-5) Repeat steps 2-1) to 2-4) before reaching the specified number of iterations K. When the specified number of iterations K is reached, the optimal atomic set A is obtained. K :

5. The method according to claim 1, characterized in that: In the iterative operation in step 3), the residual R K As the residual initial value of the dynamic parameter l1 norm regularization criterion estimation, the optimal atomic set A K is the initial value of the optimal atomic set.

6. The method according to claim 1 or 5, characterized in that: In step 3), the iteration operation based on the dynamic parameter l1 norm regularization criterion is set to be repeated T times, and the iteration variable t=1,2,…,T; In the tth iteration, the following steps are performed for the kth estimated angle: 3-1) Calculate the pitch angle at the kth estimated angle and azimuth The optimal set of atoms at iteration t The partial differential and Where diag is the diagonal matrix function, G is the polarization response of the array, For the pitch angle and azimuth The above is the spatial guidance vector, For the pitch angle and azimuth The transformation matrix from the polar coordinate system to the rectangular coordinate system; 3-2) Calculate the estimated value of the incoming signal data matrix at the kth estimated angle at the tth iteration The partial differential and in, is the residual of the tth iteration; 3-3) Calculate the residual of the tth iteration Partial differential with respect to the kth angle and 3-4) Calculation The partial differential of the l1 norm with respect to the kth angle and Among them, ||·1|| is the 1 norm, Re is the real part, tr is the trace of the matrix; the intermediate quantity I2 is the 2×2 identity matrix, represents the Kronecker product; 3-5) Calculate the residual The partial differential of the square of the l2 norm of with respect to the kth angle 3-6) Calculate the gradient w of the kth angle θ,k and ε is an introduced constant; 3-7) Update the k-th angle estimate of the t+1-th iteration γ is the step size of each iteration, and |·| is the absolute value; 3-8) For each angle, perform steps 3-1) to 3-7) until the t+1th estimated values ​​of K angles are updated. Then proceed to step 3-9); 3-9) Update the t+1th optimal atomic set 3-10) Update the estimated value of the data matrix of the t+1th incoming wave signal 3-11) Update t = t + 1 and repeat steps 3-1) to 3-10) until the number of iterations t reaches T, and the optimal atomic set is obtained. Go to step 4):