RGSVD distributed MIMO radar direction finding method based on ALS
By using ALS and RGSVD technologies in MIMO radar to process the echo signal of distributed MIMO radar, the problem of low direction finding accuracy under low signal-to-noise ratio is solved, and high-precision direction finding for multi-target angles is achieved. It is suitable for Multi-array system.
Patent Information
- Application Number
- CN202410007980.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-03
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-01-03
AI Technical Summary
The existing MIMO radar direction finding technology is low in direction finding accuracy under low signal-to-noise ratio conditions, and is only suitable for dual-base MIMO radar systems. The data information is insufficient to fully extract, making it difficult to achieve central fusion processing of multi-array receiving echo data, affecting the radar. location and tracking functions.
Using the RGSVD distributed MIMO radar direction finding method based on ALS, the matching filtering and vectorization processing of the received array echo signal of the distributed MIMO radar is constructed, and the covariance matrix is decomposed by RGSVD to obtain the signal The subspace is used to realize joint matrix decomposition through ALS iterative update, and the ESPRIT algorithm is substituted for joint estimation of multi-objective DOA and DOD.
Improves direction finding accuracy under low signal-to-noise ratio conditions, is suitable for distributed MIMO radar systems, making full use of the in-data correlation of multiple receiving arrays, achieving higher accuracy signal subspace decomposition and high accuracy of multi-target angles Direction finding.
Smart Images

Figure CN118962612A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of radar direction finding, and in particular relates to an ALS-based RGSVD distributed MIMO radar direction finding method. Background Art
[0002] With the rapid development of modern penetration means such as hypersonic vehicles, stealth aircraft, and cruise missiles, the national strategic early warning system has put forward new and higher requirements for radar detection technology. In order to meet the needs of radar anti-stealth, anti-interference and anti-destruction, MIMO radar based on distributed sub-platforms came into being and achieved great development.
[0003] From the perspective of signal processing, the functions of radar mainly include detection, estimation, tracking, imaging and recognition. Radar's direction finding technology for targets is a major aspect of parameter estimation and the basis for target positioning and tracking. Therefore, it is one of the core technologies of radar. Improving the direction finding accuracy of radar systems is a key direction for improving radar performance. Distributed array MIMO radar can effectively improve the joint estimation accuracy of target DOA and DOD under low signal-to-noise ratio through joint optimization of data.
[0004] From the perspective of existing MIMO radar direction-finding technology: for dual-base array MIMO radar, the algorithm for direction-finding using the classic ESPRIT algorithm is relatively close to that of this patent after the truncated singular value decomposition is used to process the mutual covariance matrix of the two arrays to obtain the signal subspace, but it is only applicable to dual-base MIMO radar systems; and because only the mutual covariance matrix is used, data information extraction is insufficient, and it is difficult to achieve central fusion processing and denoising of multi-array received echo data, resulting in low direction-finding accuracy under low signal-to-noise ratio conditions.
[0005] The following is a flowchart of implementing bistatic MIMO radar direction finding by processing the cross-covariance matrix through TSVD and then feeding it into the ESPRIT algorithm. Figure 1 As shown in the figure: This technology is a classic algorithm for joint DOA and DOD direction finding for dual-base MIMO radar. The algorithm performs singular value decomposition on the mutual covariance matrix of the echo signals of the two receiving arrays of the dual-base MIMO radar, obtains the signal subspace of each receiving array, and then uses the ESPRIT algorithm to measure the DOA of the target relative to the receiving array, and then permutes the receiving subspace to obtain the signal subspace of the transmitting array. Similarly, the DOD of the target relative to the transmitting array is calculated, and finally the multi-target angles are matched to complete the direction finding.
[0006] This implementation method can achieve high-precision estimation of target DOA and DOD under high signal-to-noise ratio conditions, but under low signal-to-noise ratio and small snapshot number conditions, the direction finding accuracy of the algorithm will be affected, which will affect the subsequent positioning and tracking functions of the radar. In addition, the algorithm is mainly applicable to dual-base MIMO radar systems, where the number of arrays is relatively limited and data information extraction is insufficient. Summary of the invention
[0007] In order to solve the above technical problems, the present invention proposes an ALS-based RGSVD distributed MIMO radar direction finding solution.
[0008] The first aspect of the present invention proposes an ALS-based RGSVD distributed MIMO radar direction finding method. ALS represents an alternating least squares algorithm, RGSVD represents a low complexity generalized singular value decomposition, and MIMO radar represents a multiple input multiple output radar; the method comprises:
[0009] Step S1: construct an echo signal model η of the distributed MIMO radar by performing matched filtering and vectorization processing on the receiving array echo signal of the distributed MIMO radar. (l) , and according to the echo signal model η (l) Calculate the covariance matrix R of the echo signal ηη ;
[0010] Step S2: Use RGSVD to calculate the covariance matrix R of the echo signal. ηη Decompose to obtain the receiving array signal subspace, use ALS to implement joint matrix decomposition through iterative update, and obtain the transmitting array signal subspace by permuting the matrix;
[0011] Step S3, substituting the calculated receiving array signal subspace and transmitting array signal subspace into the ESPRIT algorithm to perform joint estimation and matching of multiple DOAs and DODs for each target;
[0012] Wherein, DOA stands for angle of arrival, and DOD stands for angle of departure.
[0013] According to the method of the first aspect of the present invention, in step S1, the distributed MIMO radar includes m sub-transmitting arrays and n sub-receiving arrays, and each sub-array is arranged in a uniform linear array; wherein:
[0014] The number of array elements of each sub-transmitting array is M1, M2, ..., M m , the array element spacing is d t , then the total number of transmitting array elements is M = M1 + M2 + ... + M m The number of array elements of each sub-receiving array is N1, N2...N n , the array element spacing is dr , then the total number of receiving array elements is N = N1 + N2 + ... + N n ; and the array element spacing satisfies d t (d r )≤λ / 2, λ represents the wavelength of the transmitted signal;
[0015] All transmitted signals are received by the sub-receiving array after being reflected by P targets. The DOA of the p-th target with respect to the i-th receiving sub-array is expressed as φ ip , the DOD of the i-th transmitting subarray is expressed as θ ip , the target reflection coefficient is β p ; Then the vectors of DOD and DOA of P targets with respect to the i-th array are expressed as θ i =[θ i1 θ i2 …θ iP ] T and φ i =[φ i1 φ i2 …φ iP ] T ;
[0016] The steering vector of the i-th receiving sub-array with respect to the p-th target is:
[0017]
[0018] The manifold matrix of the i-th sub-receiver array is:
[0019] A ri (φ i )=[a ri (φ i1 ) ri (φ i2 )…a ri (φ iP )]
[0020] The steering vector of the i-th sub-transmitting array with respect to the p-th target is:
[0021]
[0022] The manifold matrix of the i-th sub-emission array is:
[0023] A ti (θ i )=[a ti (θ i1 ) ti (θ i2 )…a ti (θ iP )]
[0024] by represents the mth sub-emission matrix of the i-th sub-emission matrix i Each pulse contains K code elements, and its dimension is K×1. Then the transmission waveform of the i-th sub-transmitting array is an M i ×K matrix, and:
[0025]
[0026] The received signal matrix formed by each of the sub-receiving arrays is expressed as:
[0027]
[0028] in, It indicates the lth snapshot signal received by n sub-receiving arrays. Represents the reflection coefficient of P targets, which follows the distribution Z (l) Represents zero-mean complex Gaussian white noise, which obeys the distribution CN~(0,σ 2 I N ), transmit signal and:
[0029]
[0030]
[0031] After matching filtering and vectorization processing of the received signal, we get:
[0032]
[0033] in, ζ (l) represents the noise vector, represents the Kronecker product;
[0034] R ζζ =σ 2 I MN
[0035] Because A i1 ,A i2 ,...,A im contains the DOA information of the i-th sub-receive array. i The information of m matrices is combined to obtain:
[0036]
[0037] K i (φ i ) represents K i(φ i ,θ1,...,θ m ), combine the data vectors of the n sub-receive arrays, then:
[0038]
[0039] in, represents the combination of the received data vectors of the i-th receiving sub-array, and the covariance matrix of the received signal is:
[0040]
[0041]
[0042] Among them, R b =E[b (l) b (l)H ], represents the noise covariance matrix of the i-th sub-receive array.
[0043] According to the method of the first aspect of the present invention, in step S2, the covariance matrix of the received signal is divided into n matrices with the same number of columns:
[0044]
[0045] …
[0046]
[0047] H i It is expressed as:
[0048]
[0049] The above n matrices are tensors mode-3 slice, and J = max(MN1,...,MN n ), and fill the undefined elements with zeros, perform multi-threaded generalized singular value decomposition on the tensor and represent it in slice form as follows:
[0050]
[0051] Among them, k∈{1,...,n} is a tensor The kth slice of P is the number of targets to be detected, is a nonnegative diagonal matrix; the matrix Diagonal matrix C k The diagonal elements of are stacked as row vectors of matrix C, C k =diag(C(k,:)); then we get the optimization problem:
[0052]
[0053] Using the ALS algorithm, in C k and A is fixed to B k To optimize, in B k Fixed case C k and A are optimized, and the required signal subspace B is obtained by multiple iterations. k =A((MN k-1 +1):MN k ,:).
[0054] According to the method of the first aspect of the present invention, in step S2, the variables in the ALS algorithm are initialized, and A is initialized to The left singular matrix, C1,...,C k Both are initialized to I P , add the l1 norm constraint through the optimization problem expression, solve the following equation to update B k :
[0055]
[0056] The iterative shrinkage threshold algorithm yields:
[0057] B k =soft λt (B k (old) +α k (H k H -AC k (B k (old) ) H ) H AC k )
[0058] Among them, soft λt (·) is the soft threshold operation function, B k (old) Indicates B in the previous iteration k matrix, Indicates a fixed step value;
[0059] set up Fixed B k Update matrices A and C k The specific steps are as follows:
[0060] Use n Stack out tensors as mode-3 slices Tensor The constrained CP decomposition of the frontal slice of can be expressed as
[0061]
[0062] Tensor The mode-1 and mode-3 expansions satisfy:
[0063]
[0064]
[0065] Among them, the row vector of C contains the matrix C k The diagonal elements of the matrix A and C are expressed as follows:
[0066]
[0067]
[0068] When the number of iterations exceeds the preset maximum number of iterations or the residual If it is lower than the predefined threshold, it means it has entered the convergence state and the cycle stops. Represents the residual of the previous iteration, and obtains B k =A((MN k-1 +1):MN k ,:) is the signal subspace of the kth sub-receiver array.
[0069] The second aspect of the present invention proposes an ALS-based RGSVD distributed MIMO radar direction finding system. ALS represents an alternating least squares algorithm, RGSVD represents a low complexity generalized singular value decomposition, and MIMO radar represents a multiple input multiple output radar; the system includes a processing unit, and the processing unit is configured to execute:
[0070] Step S1: construct an echo signal model η of the distributed MIMO radar by performing matched filtering and vectorization processing on the receiving array echo signal of the distributed MIMO radar. (l) , and according to the echo signal model η (l) Calculate the covariance matrix R of the echo signal ηη ;
[0071] Step S2: Use RGSVD to calculate the covariance matrix R of the echo signal. ηη Decompose to obtain the receiving array signal subspace, use ALS to implement joint matrix decomposition through iterative update, and obtain the transmitting array signal subspace by permuting the matrix;
[0072] Step S3, substituting the calculated receiving array signal subspace and transmitting array signal subspace into the ESPRIT algorithm to perform joint estimation and matching of multiple DOAs and DODs for each target;
[0073] Wherein, DOA stands for angle of arrival, and DOD stands for angle of departure.
[0074] According to the system of the second aspect of the present invention, in step S1, the distributed MIMO radar includes m sub-transmitting arrays and n sub-receiving arrays, and each sub-array is arranged in a uniform linear array; wherein:
[0075] The number of array elements of each sub-transmitting array is M1, M2, ..., M m , the array element spacing is d t , then the total number of transmitting array elements is M = M1 + M2 + ... + M m The number of array elements of each sub-receiving array is N1, N2...N n , the array element spacing is d r , then the total number of receiving array elements is N = N1 + N2 + ... + N n ; and the array element spacing satisfies d t (d r )≤λ / 2, λ represents the wavelength of the transmitted signal;
[0076] All transmitted signals are received by the sub-receiving array after being reflected by P targets. The DOA of the p-th target with respect to the i-th receiving sub-array is expressed as φ ip , the DOD of the i-th transmitting subarray is expressed as θ ip , the target reflection coefficient is β p ; Then the vectors of DOD and DOA of P targets with respect to the i-th array are expressed as θ i =[θ i1 θ i2 …θ iP ] T and φ i =[φ i1 φ i2 …φ iP ] T ;
[0077] The steering vector of the i-th receiving sub-array with respect to the p-th target is:
[0078]
[0079] The manifold matrix of the i-th sub-receiver array is:
[0080] A ri (φ i )=[a ri (φi1 ) ri (φ i2 )…a ri (φ iP )]
[0081] The steering vector of the i-th sub-transmitting array with respect to the p-th target is:
[0082]
[0083] The manifold matrix of the i-th sub-emission array is:
[0084] A ti (θ i )=[a ti (θ i1 ) ti (θ i2 )…a ti (θ iP )]
[0085] by represents the mth sub-emission matrix of the i-th sub-emission matrix i Each pulse contains K code elements, and its dimension is K×1. Then the transmission waveform of the i-th sub-transmitting array is an M i ×K matrix, and:
[0086]
[0087] The received signal matrix formed by each of the sub-receiving arrays is expressed as:
[0088]
[0089] in, It indicates the lth snapshot signal received by n sub-receiving arrays. Represents the reflection coefficient of P targets, which follows the distribution Z (l) Represents zero-mean complex Gaussian white noise, which obeys the distribution CN~(0,σ 2 I N ), transmit signal and:
[0090]
[0091]
[0092] After matching filtering and vectorization processing of the received signal, we get:
[0093]
[0094] in, ζ (l) represents the noise vector, represents the Kronecker product;
[0095] R ζζ =σ 2 I MN
[0096] Because A i1 ,A i2 ,...,A im contains the DOA information of the i-th sub-receive array. i The information of m matrices is combined to obtain:
[0097]
[0098] K i (φ i ) represents K i (φ i ,θ1,...,θ m ), combine the data vectors of the n sub-receive arrays, then:
[0099]
[0100] in, represents the combination of the received data vectors of the i-th receiving sub-array, and the covariance matrix of the received signal is:
[0101]
[0102]
[0103] Among them, R b =E[b (l) b (l)H ], represents the noise covariance matrix of the i-th sub-receive array.
[0104] According to the system of the second aspect of the present invention, in step S2, the covariance matrix of the received signal is divided into n matrices with the same number of columns:
[0105]
[0106] …
[0107]
[0108] H i It is expressed as:
[0109]
[0110] The above n matrices are tensors mode-3 slice, and J = max(MN1,...,MN n ), and fill the undefined elements with zeros, perform multi-threaded generalized singular value decomposition on the tensor and represent it in slice form as follows:
[0111]
[0112] Among them, k∈{1,...,n} is a tensor The kth slice of P is the number of targets to be detected, is a nonnegative diagonal matrix; the matrix Diagonal matrix C k The diagonal elements of are stacked as row vectors of matrix C, C k =diag(C(k,:)); then we get the optimization problem:
[0113]
[0114] Using the ALS algorithm, in C k and A is fixed to B k To optimize, in B k Fixed case C k and A are optimized, and the required signal subspace B is obtained by multiple iterations. k =A((MN k-1 +1):MN k ,:).
[0115] According to the system of the second aspect of the present invention, in step S2, the variables in the ALS algorithm are initialized, and A is initialized to The left singular matrix, C1,...,C k Both are initialized to I P , add the l1 norm constraint through the optimization problem expression, solve the following equation to update B k :
[0116]
[0117] The iterative shrinkage threshold algorithm yields:
[0118] B k =soft λt (B k (old) +α k (H k H -AC k (B k(old) ) H ) H AC k )
[0119] Among them, soft λt (·) is the soft threshold operation function, B k (old) Indicates B in the previous iteration k matrix, Indicates a fixed step value;
[0120] set up Fixed B k Update matrices A and C k The specific steps are as follows:
[0121] Use n Stack out tensors as mode-3 slices Tensor The constrained CP decomposition of the frontal slice of can be expressed as
[0122]
[0123] Tensor The mode-1 and mode-3 expansions satisfy:
[0124]
[0125]
[0126] Among them, the row vector of C contains the matrix C k The diagonal elements of the matrix A and C are expressed as follows:
[0127]
[0128]
[0129] When the number of iterations exceeds the preset maximum number of iterations or the residual If it is lower than the predefined threshold, it means it has entered the convergence state and the cycle stops. Represents the residual of the previous iteration, and obtains B k =A((MN k-1 +1):MN k ,:) is the signal subspace of the kth sub-receiver array.
[0130] The third aspect of the present invention discloses an electronic device. The electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, an RGSVD distributed MIMO radar direction finding method based on ALS according to the first aspect of the present disclosure is implemented.
[0131] The fourth aspect of the present invention discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the RGSVD distributed MIMO radar direction finding method based on ALS of the first aspect of the present disclosure is implemented.
[0132] In summary, this application designs a direction-finding algorithm for a distributed MIMO radar system from the perspective of improving the system direction-finding accuracy under low signal-to-noise ratio conditions. It mainly utilizes the generalized singular value decomposition technology to jointly decompose the echo signal covariance matrix of multiple receiving arrays of the system, and effectively applies the internal correlation of the data to achieve better denoising performance. The more accurately decomposed signal subspace is then introduced into the ESPRIT algorithm to realize the system's joint high-precision direction-finding of multiple target DOA and DOD. BRIEF DESCRIPTION OF THE DRAWINGS
[0133] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0134] Figure 1 The figure is a flow chart of TSVD in the prior art for joint direction finding of dual-base MIMO radar.
[0135] Figure 2 4 is a diagram of an echo signal model of a distributed MIMO radar system according to an embodiment of the present invention.
[0136] Figure 3 This is a flow chart of GSVD joint direction finding for distributed MIMO radar according to an embodiment of the present invention.
[0137] Figure 4 Schematic diagram of estimation results of θ1 and φ1 according to an embodiment of the present invention.
[0138] Figure 5 Schematic diagram of estimation results of θ2 and φ2 according to an embodiment of the present invention.
[0139] Figure 6 Schematic diagram of the matching results of φ1 and φ2 according to an embodiment of the present invention.
[0140] Figure 7 Schematic diagram of total root mean square errors of four algorithms at different signal-to-noise ratios according to an embodiment of the present invention.
[0141] Figure 8Schematic diagram of the root mean square error of φ1 according to an embodiment of the present invention.
[0142] Fig. 9 FIG. 4 is a schematic diagram of a root mean square error of φ2 according to an embodiment of the present invention.
[0143] Fig.10 is a schematic diagram of the root mean square error of θ1 according to an embodiment of the present invention.
[0144] Fig.11 FIG. 4 is a schematic diagram of a root mean square error of θ2 according to an embodiment of the present invention.
[0145] Fig.12 Schematic diagram of total root mean square errors of four algorithms under different snapshot numbers according to an embodiment of the present invention.
[0146] Fig.13 The figure is a structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION
[0147] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0148] Specific Example 1
[0149] Step 1: Establish a distributed MIMO radar echo signal model: First, the system echo signal model is obtained by matching filtering and vectorizing the distributed MIMO radar receiving array echo signal. Then, according to the echo signal model η (l) Get the covariance matrix R of the echo signal ηη ;
[0150] Step 2: Use the generalized singular value decomposition technique to decompose the echo signal covariance matrix to obtain the signal subspace U of each receiving array S In the process, the alternating least squares algorithm is used to iteratively update each sub-matrix to achieve joint matrix decomposition.
[0151] Step 3: Substitute the calculated signal subspace into the ESPRIT algorithm to perform joint estimation and matching of multiple DOAs and DODs of each target.
[0152] Building a signal model
[0153] First, a distributed MIMO echo signal model is established, such as Figure 2As shown in the figure, the distributed MIMO radar consists of m sub-transmitting arrays and n sub-receiving arrays, each of which is a uniform linear array. The number of array elements in each sub-transmitting array is M1, M2…M m , the array element spacing is d t , the total number of transmitting array elements is M = M1 + M2 + ... + M m , the number of array elements in each sub-receiving array is N1, N2…N n , the array element spacing is d r , the total number of receiving array elements is N = N1 + N2 + ... + N n , the array element spacing satisfies d t (d r )≤λ / 2, where λ represents the wavelength of the transmitted signal. All transmitted signals are received by the sub-receiving array after being reflected by P targets. The DOA of the p-th target with respect to the i-th receiving sub-array is expressed as φ ip , the DOD of the i-th transmitting subarray is expressed as θ ip , the target reflection coefficient is expressed as β p Therefore, the vectors of DOD and DOA of P targets with respect to the i-th array are expressed as θ i =[θ i1 θ i2 …θ iP ] T and φ i =[φ i1 φ i2 …φ iP ] T .
[0154] The steering vector of the i-th sub-receiving array with respect to the p-th target is:
[0155]
[0156] The manifold matrix of the i-th sub-receiver array is:
[0157] A ri (φ i )=[a ri (φ i1 ) ri (φ i2 )…a ri (φ iP )]#(2)
[0158] The steering vector of the i-th sub-transmitting array with respect to the p-th target is:
[0159]
[0160] The manifold matrix of the i-th sub-emission array is:
[0161] A ti (θ i )=[a ti (θ i1 ) ti (θ i2 )…a ti (θ iP )]#(4)
[0162] use represents the mth sub-emission matrix i A pulse transmitted by an array element, since each pulse has K code elements, it is a K×1 vector, and the transmission waveform of the i-th sub-transmitting array is an M i ×K matrix, that is:
[0163]
[0164] The received signal matrix of the receiving subarray can be expressed as:
[0165]
[0166] in: It indicates the lth snapshot signal received by n receiving arrays. Represents the reflection coefficient of P targets, which follows the distribution Z (l) Represents zero-mean complex Gaussian white noise, which obeys the distribution CN~(0,σ 2 I N ), transmit signal
[0167] After matching filtering and vectorization processing of the received signal, we get:
[0168]
[0169] in, ζ (l) represents the noise vector, represents the Kronecker product.
[0170] R ζζ =σ 2 I MN #(8)
[0171] Because A i1 ,A i2 ,...,A im contains the DOA information of the i-th sub-receiver array, so to extract φ i , combine these m matrices together to get the matrix:
[0172]
[0173] In the following text, K i (φ i ,θ1,...,θ m ) is abbreviated as K i (φ i ). Similarly, the data vectors of the n sub-receiving arrays can be combined respectively, so that the received signal in equation (7) is divided into the following forms:
[0174]
[0175] in represents the combination of the received data vector of the i-th sub-receive array. Therefore, the covariance matrix of the received signal can be expressed as:
[0176]
[0177] Substituting equations (7) and (9) into equation (11), we can derive:
[0178]
[0179] Where R b =E[b (l) b (l)H ], represents the noise covariance matrix of the i-th sub-receive array.
[0180] ALS-based R-GSVD algorithm (such as Figure 3 (shown)
[0181] Divide the covariance matrix of the received signal into n matrices with the same number of columns:
[0182]
[0183] From formula (12), we can see that H i It can be expressed as:
[0184]
[0185] The above n matrices can be regarded as tensors mode-3 slice, where J = max(MN1,...,MN n ), fill the undefined elements in equation (13) with zeros.
[0186] After multi-threaded generalized singular value decomposition of the tensor, it is represented by slicing as follows:
[0187]
[0188] Among them, k∈{1,...,n} is a tensor The kth slice of P is the number of targets to be detected, is a non-negative diagonal matrix. In addition, let the matrix Diagonal matrix C k The diagonal elements of are stacked as row vectors of matrix C, that is, C k =diag(C(k,:)).
[0189] In order to minimize the impact of noise and obtain a more accurate signal subspace, the optimization problem to be solved is expressed as follows:
[0190]
[0191] This optimization problem is highly consistent with the PARAFAC2 model. Therefore, in order to solve the above optimization problem, the ALS algorithm can be used in C k and A is fixed to B k Optimize, then in B k Fixed case C k Optimize with A, iterate multiple times to finally solve the optimization problem of the function and find the required signal subspace B k =A((MN k-1 +1):MN k ,:).
[0192] First, initialize the variables in the algorithm: initialize A to The left singular matrix, C1,...,C k Both are initialized to I P Then, by adding the l1 norm constraint to equation (16), we solve equation (17) to update B k .
[0193]
[0194] This optimization problem can be solved by an iterative shrinkage threshold algorithm:
[0195] B k =soft λt (B k (old) +α k (H k H -AC k (B k (old) ) H ) H AC k )#(18)
[0196] Among them, soft λt (·) is the soft threshold operation function. B k (old) Indicates B in the previous iteration k matrix, Indicates a fixed step value.
[0197] set up Fixed B k Update matrices A and C k The specific steps are as follows:
[0198] Use n Stack out tensors as mode-3 slices Tensor The constrained CP decomposition of the frontal slice of can be expressed as:
[0199]
[0200] Therefore, the tensor The mode-1 and mode-3 expansions satisfy:
[0201]
[0202]
[0203] The row vector of C contains the matrix C k Therefore, the least squares solution of matrices A and C can be expressed as follows:
[0204]
[0205]
[0206] This process uses the direct fitting algorithm of the PARAFAC2 model solution. In the indirect fitting algorithm of PARAFAC2, the derived data is used to fit the original matrix, and the matrix B cannot be obtained. k , and through the above direct fitting algorithm, the actual data is used for fitting, and the matrix B can be directly calculated k In addition, when constraints are imposed on parameters (such as non-negative constraints, etc.) and when missing data are replaced with optimal model estimates, the process of the direct fitting algorithm is easier to adjust, and the direct fitting algorithm is easier to generalize to higher-order data sets and the algorithm is more efficient.
[0207] When the number of iterations exceeds the preset maximum number of iterations, or the residual is less than the predefined threshold, indicating that the algorithm converges and the loop stops. Represents the residual of the previous iteration. B obtained by applying this algorithm k =A((MN k-1 +1):MN k ,:) is the signal subspace of the kth sub-receive array. The main steps of the algorithm are summarized in Table 1:
[0208] Table 1 Specific process of R-GSVD algorithm based on ALS
[0209]
[0210]
[0211]
[0212] Specific Example 2
[0213] Taking dual-transmit dual-receive MIMO radar as an example, the algorithm performance is compared and analyzed:
[0214] Consider a dual-transmit dual-receive MIMO radar system, where both the transmitting array and the receiving array are uniform linear arrays, and the array element spacing is set to half the wavelength of the transmitted signal, that is, d t =d r =λ0 / 2, where λ0 is the wavelength of the transmitted signal. The reflection coefficient β of P targets p (p=1,...,P), modeled as a zero-mean Gaussian random variable with unit variance through the randn function, and the target angle estimation performance was tested through 500 Monte Carlo experiments. Different detection targets and snapshot numbers were set for different experiments. In the simulation experiment, the signal-to-noise ratio is defined as the ratio of the total power of the echo signal received by the receiving array to the ambient noise power.
[0215] The effect of joint estimation of DOA and DOD
[0216] The number of transmitting arrays and receiving arrays are set to M1 = 6, M2 = 7 and N1 = 4, N2 = 5 respectively. The number of iterations is set to no more than 5000 times and the residual predefined threshold is 1e -3 As the stopping criterion of the algorithm, the number of snapshots is set to L = 200, and a total of P = 6 detection targets are set. The DOD and DOA of these six targets are set to:
[0217] θ1=[-60 ° ,-45 ° ,30 ° ,50 ° ,35 ° ,60 ° ] T
[0218] θ2=[-45° ,-25 ° ,25 ° ,65 ° ,55 ° ,15 ° ] T
[0219] φ1=[-40 ° ,-20 ° ,30 ° ,40 ° ,60 ° ,70 ° ] T
[0220] φ2=[-55 ° ,-35 ° ,35 ° ,45 ° ,65 ° ,75 ° ] T
[0221] The estimated results of 500 Monte Carlo experiments are as follows Figures 4 to 6 As shown. Figures 4 to 6 It can be seen that in a spatial white noise environment with a signal-to-noise ratio of 5dB, the DOD of the target relative to the transmitting array and the DOA relative to the receiving array can be accurately and effectively identified, and the four DODs and DOAs of the six targets can be accurately matched. At the same time, the total root mean square error of the four angle estimates of the target measured by the simulation experiment is about 0.5204°. This algorithm shows good angle estimation performance and accuracy in a 5dB white noise environment.
[0222] Specific Example 3
[0223] Comparative analysis of algorithm performance
[0224] Performance comparison under different signal-to-noise ratios
[0225] In the range of signal-to-noise ratio SNR = [-10, -8, -6, -4, 0, 5, 10, 15, 20], the total root mean square error and the RMSE of the four angles are calculated using the R-GSVD algorithm, the cross-covariance matrix TSVD algorithm and the improved TSVD algorithm, and the performance is compared with the Cramer-Rao bound as a reference. In this experiment, the number of snapshots is set to 200, and three detection targets are set, namely:
[0226] θ1=[30 ° ,20 ° ,40 ° ] T ,θ2=[55 ° ,65° ,50 ° ] T
[0227] φ1=[30 ° ,40 ° ,50 ° ] T ,φ2=[65 ° ,55 ° ,75 ° ] T
[0228] The angle of the target relative to the transmitting array 1 and the receiving array 1 is not greater than 50 degrees, and the angle relative to the transmitting array 2 and the receiving array 2 is not less than 50 degrees. The results of 500 Monte Carlo experiments under each signal-to-noise ratio condition are as follows Figures 7 to 11 The dotted curve represents the R-GSVD algorithm, the lower triangular curve represents the cross-covariance matrix TSVD algorithm, the diamond curve represents the improved TSVD algorithm, and the left triangular curve represents the CRB of the system direction finding error under the experimental conditions.
[0229] from Figure 5 It can be seen that under higher signal-to-noise ratio, the performance of R-GSVD algorithm is basically the same as that of improved TSVD algorithm, and is slightly better than cross-covariance matrix TSVD algorithm. Under low signal-to-noise ratio conditions, the error of R-GSVD algorithm is slightly smaller than that of improved TSVD algorithm, and the error is about 0.5° less when the signal-to-noise ratio is -5dB.
[0230] Depend on Figures 5 to 9 It is known that the R-GSVD algorithm has better estimation performance for φ1, and its error is about 0.5° less than that of the TSVD-based algorithm. However, under low signal-to-noise ratio conditions, the R-GSVD algorithm's estimation performance for φ2 is similar to that of the TSVD algorithm, or even slightly larger. As the signal-to-noise ratio increases, the performance of the cross-covariance matrix TSVD algorithm decreases rapidly, and the performance of the other two algorithms is similar. This is due to the error in the signal subspace decomposition caused by the cross-covariance matrix TSVD algorithm's insufficient application of the correlation between the received data of the two receiving arrays. The R-GSVD algorithm's estimation performance for DOD is significantly better than the two TSVD-based algorithms. Under -5dB conditions, the estimation error for θ1 is about 0.6° less, and the estimation error for θ2 is about 0.7° less. As the signal-to-noise ratio increases, the performance of the other two algorithms is slightly better than that of the cross-covariance matrix TSVD algorithm. It can be seen that the R-GSVD algorithm designed in this patent makes full use of the correlation between all received data, thereby achieving better estimation performance.
[0231] Performance comparison under different snapshot numbers
[0232] Under the condition of 0dB signal-to-noise ratio, the target design is the same as Section 5.2.1, and the RMSE of the three algorithms is calculated when the number of snapshots L = [50, 100, 150, 200, 250, 300, 350, 400, 450, 500]. The results of 500 Monte Carlo experiments under different snapshot numbers are shown in the figure. Fig.12 .
[0233] Depend on Fig.10 It can be seen that as the number of snapshots increases, the internal correlation of the signal in the covariance matrix increases, the influence of noise on the direction finding accuracy decreases, and the direction finding errors of the three algorithms will decrease. The R-GSVD algorithm designed in this patent makes more full use of the internal correlation of the data in the covariance matrix, which is slightly better than the other two algorithms. Therefore, within a certain range, as the number of snapshots increases, the estimation error of the R-GSVD algorithm decreases faster than the other two algorithms.
[0234] Specific Example 4
[0235] Update B in the algorithm k The part of B can be updated by minimizing formula (24) k :
[0236]
[0237] When there is a constraint B k ·B k H =I, for B k Optimizing so that minimizing equation (24) is equivalent to maximizing equation (25):
[0238]
[0239] The solution process of the optimization problem is as follows: First, Perform singular value decomposition to obtain its left singular matrix U k and the right singular matrix V k , then we can get B that maximizes equation (25) k , which can be expressed as:
[0240]
[0241] In the actual optimization process of formula (25), since the number of detection targets is P, After performing singular value decomposition, the singular vectors corresponding to the first P largest singular values are taken to form U k and V k , and then obtain B k The optimization result is:
[0242]
[0243] However, the algorithm complexity of updating using this method is higher and the calculation time is longer.
[0244] The second aspect of the present invention proposes an ALS-based RGSVD distributed MIMO radar direction finding system. ALS represents an alternating least squares algorithm, RGSVD represents a low complexity generalized singular value decomposition, and MIMO radar represents a multiple input multiple output radar; the system includes a processing unit, and the processing unit is configured to execute:
[0245] Step S1: construct an echo signal model η of the distributed MIMO radar by performing matched filtering and vectorization processing on the receiving array echo signal of the distributed MIMO radar. (l) , and according to the echo signal model η (l) Calculate the covariance matrix R of the echo signal ηη ;
[0246] Step S2: Use RGSVD to calculate the covariance matrix R of the echo signal. ηη Decompose to obtain the receiving array signal subspace, use ALS to implement joint matrix decomposition through iterative update, and obtain the transmitting array signal subspace by permuting the matrix;
[0247] Step S3, substituting the calculated receiving array signal subspace and transmitting array signal subspace into the ESPRIT algorithm to perform joint estimation and matching of multiple DOAs and DODs for each target;
[0248] Wherein, DOA stands for angle of arrival, and DOD stands for angle of departure.
[0249] The third aspect of the present invention discloses an electronic device. The electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, an RGSVD distributed MIMO radar direction finding method based on ALS according to the first aspect of the present disclosure is implemented.
[0250] Fig.13 is a structural diagram of an electronic device according to an embodiment of the present invention, such as Fig.13As shown, the electronic device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the electronic device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, near field communication (NFC) or other technologies. The display screen of the electronic device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the electronic device can be a touch layer covered on the display screen, or a button, a trackball or a touch pad set on the housing of the electronic device, or an external keyboard, touch pad or mouse, etc.
[0251] Those skilled in the art will understand that Fig.13 The structure shown in the figure is only a structural diagram of the part related to the technical solution of the present disclosure, and does not constitute a limitation on the electronic device to which the technical solution of the present application is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0252] The fourth aspect of the present invention discloses a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the RGSVD distributed MIMO radar direction finding method based on ALS of the first aspect of the present disclosure is implemented.
[0253] In summary, this application designs a direction-finding algorithm for a distributed MIMO radar system from the perspective of improving the system direction-finding accuracy under low signal-to-noise ratio conditions. It mainly utilizes the generalized singular value decomposition technology to jointly decompose the echo signal covariance matrix of multiple receiving arrays of the system, and effectively applies the internal correlation of the data to achieve better denoising performance. The more accurately decomposed signal subspace is then introduced into the ESPRIT algorithm to realize the system's joint high-precision direction-finding of multiple target DOA and DOD.
[0254] Please note that the technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. The above embodiments only express several implementation methods of the present application, and their descriptions are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, without departing from the concept of the present application, several variations and improvements can be made, which all belong to the scope of protection of the present application. Therefore, the scope of protection of the patent in this application shall be based on the attached claims.
Claims
1. An RGSVD distributed MIMO radar direction finding method based on ALS, characterized in that: ALS denotes alternating least squares algorithm, RGSVD denotes low complexity generalized singular value decomposition, and MIMO radar denotes multiple-input multiple-output radar; The method comprises: Step S1: construct an echo signal model η of the distributed MIMO radar by performing matched filtering and vectorization processing on the receiving array echo signal of the distributed MIMO radar. (l) , and according to the echo signal model η (l) Calculate the covariance matrix R of the echo signal ηη ; Step S2: Use RGSVD to calculate the covariance matrix R of the echo signal. ηη Decompose to obtain the receiving array signal subspace, use ALS to implement joint matrix decomposition through iterative update, and obtain the transmitting array signal subspace by permuting the matrix; Step S3, substituting the calculated receiving array signal subspace and transmitting array signal subspace into the ESPRIT algorithm to perform joint estimation and matching of multiple DOAs and DODs for each target; Wherein, DOA stands for angle of arrival, and DOD stands for angle of departure.
2. The RGSVD distributed MIMO radar direction finding method based on ALS according to claim 1, characterized in that: In step S1, the distributed MIMO radar includes m sub-transmitting arrays and n sub-receiving arrays, and each sub-array is arranged in a uniform linear array; wherein: The number of array elements of each sub-transmitting array is M1, M2, ..., M m , the array element spacing is d t , then the total number of transmitting array elements is M = M1 + M2 + ... + M m The number of array elements of each sub-receiving array is N1, N2...N n , the array element spacing is d r , then the total number of receiving array elements is N = N1 + N2 + ... + N n ; and the array element spacing satisfies d t (d r )≤λ / 2, λ represents the wavelength of the transmitted signal; All transmitted signals are received by the sub-receiving array after being reflected by P targets. The DOA of the p-th target with respect to the i-th receiving sub-array is expressed as φ ip , the DOD of the i-th transmitting subarray is expressed as θ ip , the target reflection coefficient is β p ; Then the vectors of DOD and DOA of P targets with respect to the i-th array are expressed as θ i =[θ i1 θ i2 … θ iP ] T and φ i =[φ i1 φ i2 … φ iP ] T ; The steering vector of the i-th receiving sub-array with respect to the p-th target is: The manifold matrix of the i-th sub-receiver array is: A ri (f i )=[a ri (f i1 ) a ri (f i2 ) … a ri (f iP )] The steering vector of the i-th sub-transmitting array with respect to the p-th target is: The manifold matrix of the i-th sub-emission array is: A ti (i i )=[a ti (i i1 ) a ti (i i2 ) … a ti (i iP )] by represents the mth sub-emission matrix of the i-th sub-emission matrix i Each pulse contains K code elements, and its dimension is K×1. Then the transmission waveform of the i-th sub-transmitting array is an M i ×K matrix, and: The received signal matrix formed by each of the sub-receiving arrays is expressed as: in, It indicates the lth snapshot signal received by n sub-receiving arrays. Represents the reflection coefficient of P targets, which follows the distribution Z (l) Represents zero-mean complex Gaussian white noise, which obeys the distribution CN~(0,σ 2 I N ), transmit signal and: After matching filtering and vectorization processing of the received signal, we get: in, ζ (l) represents the noise vector, represents the Kronecker product; R ζζ =s 2 I MN Because A i1 ,A i2 ,...,A im contains the DOA information of the i-th sub-receive array. i The information of m matrices is combined to obtain: K i (φ i ) represents K i (φ i ,θ1,...,θ m ), combine the data vectors of the n sub-receive arrays, then: in, represents the combination of the received data vectors of the i-th receiving sub-array, and the covariance matrix of the received signal is: Among them, R b =E[b (l) b (l)H ], represents the noise covariance matrix of the i-th sub-receive array.
3. The RGSVD distributed MIMO radar direction finding method based on ALS according to claim 2, characterized in that: In step S2, the covariance matrix of the received signal is divided into n matrices with the same number of columns: … H i It is expressed as: The above n matrices are tensors mode-3 slice, and J = max(MN1,...,MN n ), and fill the undefined elements with zeros, perform multi-threaded generalized singular value decomposition on the tensor and represent it in slice form as follows: Among them, k∈{1,...,n} is a tensor The kth slice of P is the number of targets to be detected, is a nonnegative diagonal matrix; the matrix Diagonal matrix C k The diagonal elements of are stacked as row vectors of matrix C, C k =diag(C(k,:)); then we get the optimization problem: Using the ALS algorithm, in C k and A is fixed to B k To optimize, in B k Fixed case C k and A are optimized, and the required signal subspace B is obtained by multiple iterations. k =A((MN k-1 +1):MN k ,:).
4. The RGSVD distributed MIMO radar direction finding method based on ALS according to claim 3, characterized in that: In step S2, the variables in the ALS algorithm are initialized, and A is initialized to The left singular matrix, C1,...,C k Both are initialized to I P , add the l1 norm constraint through the optimization problem expression, solve the following equation to update B k : The iterative shrinkage threshold algorithm yields: B k =soft λt (B k (old) +α k (H k H -AC k (B k (old) ) H ) H AC k ) Among them, soft λt (·) is the soft threshold operation function, B k (old) Indicates B in the previous iteration k matrix, Indicates a fixed step value; set up Fixed B k Update matrices A and C k The specific steps are as follows: Use n Stack out tensors as mode-3 slices Tensor The constrained CP decomposition of the frontal slice of can be expressed as Tensor The mode-1 and mode-3 expansions satisfy: Among them, the row vector of C contains the matrix C k The diagonal elements of the matrix A and C are expressed as follows: When the number of iterations exceeds the preset maximum number of iterations or the residual If it is lower than the predefined threshold, it means it has entered the convergence state and the cycle stops. Represents the residual of the previous iteration, and obtains B k =A((MN k-1 +1):MN k ,:) is the signal subspace of the kth sub-receiver array.
5. An ALS-based RGSVD distributed MIMO radar direction finding system, characterized in that: ALS denotes alternating least squares algorithm, RGSVD denotes low complexity generalized singular value decomposition, and MIMO radar denotes multiple-input multiple-output radar; The system comprises a processing unit configured to perform: Step S1: construct an echo signal model η of the distributed MIMO radar by performing matched filtering and vectorization processing on the receiving array echo signal of the distributed MIMO radar. (l) , and according to the echo signal model η (l) Calculate the covariance matrix R of the echo signal ηη ; Step S2: Use RGSVD to calculate the covariance matrix R of the echo signal. ηη Decompose to obtain the receiving array signal subspace, use ALS to implement joint matrix decomposition through iterative update, and obtain the transmitting array signal subspace by permuting the matrix; Step S3, substituting the calculated receiving array signal subspace and transmitting array signal subspace into the ESPRIT algorithm to perform joint estimation and matching of multiple DOAs and DODs for each target; Wherein, DOA stands for angle of arrival, and DOD stands for angle of departure.
6. The ALS-based RGSVD distributed MIMO radar direction finding system according to claim 5, characterized in that: In step S1, the distributed MIMO radar includes m sub-transmitting arrays and n sub-receiving arrays, and each sub-array is arranged in a uniform linear array; wherein: The number of array elements of each sub-transmitting array is M1, M2, ..., M m , the array element spacing is d t , then the total number of transmitting array elements is M = M1 + M2 + ... + M m The number of array elements of each sub-receiving array is N1, N2...N n , the array element spacing is d r , then the total number of receiving array elements is N = N1 + N2 + ... + N n ; and the array element spacing satisfies d t (d r )≤λ / 2, λ represents the wavelength of the transmitted signal; All transmitted signals are received by the sub-receiving array after being reflected by P targets. The DOA of the p-th target with respect to the i-th receiving sub-array is expressed as φ ip , the DOD of the i-th transmitting subarray is expressed as θ ip , the target reflection coefficient is β p ; Then the vectors of DOD and DOA of P targets with respect to the i-th array are expressed as θ i =[θ i1 θ i2 … θ iP ] T and φ i =[v i1 φ i2 …φ iP ] T ; The steering vector of the i-th receiving sub-array with respect to the p-th target is: The manifold matrix of the i-th sub-receiver array is: A ri (f i )=[a ri (f i1 ) a ri (f i2 ) … a ri (f iP )] The steering vector of the i-th sub-transmitting array with respect to the p-th target is: The manifold matrix of the i-th sub-emission array is: A ti (i i )=[a ti (i i1 ) a ti (i i2 ) … a ti (i iP )] by represents the mth sub-emission matrix of the i-th sub-emission matrix i Each pulse contains K code elements, and its dimension is K×1. Then the transmission waveform of the i-th sub-transmitting array is an M i ×K matrix, and: The received signal matrix formed by each of the sub-receiving arrays is expressed as: in, It indicates the lth snapshot signal received by n sub-receiving arrays. Represents the reflection coefficient of P targets, which follows the distribution Z (l) Represents zero-mean complex Gaussian white noise, which obeys the distribution CN~(0,σ 2 I N ), transmit signal and: After matching filtering and vectorization processing of the received signal, we get: in, ζ (l) represents the noise vector, represents the Kronecker product; R ζζ =s 2 I MN Because A i1 ,A i2 ,...,A im contains the DOA information of the i-th sub-receive array. i The information of m matrices is combined to obtain: K i (φ i ) represents K i (φ i ,θ1,...,θ m ), combine the data vectors of the n sub-receive arrays, then: in, represents the combination of the received data vectors of the i-th receiving sub-array, and the covariance matrix of the received signal is: Among them, R b =E[b (l) b (l)H ], represents the noise covariance matrix of the i-th sub-receive array.
7. The ALS-based RGSVD distributed MIMO radar direction finding system according to claim 6, characterized in that: In step S2, the covariance matrix of the received signal is divided into n matrices with the same number of columns: … H i It is expressed as: The above n matrices are tensors mode-3 slice, and J = max(MN1,...,MN n ), and fill the undefined elements with zeros, perform multi-threaded generalized singular value decomposition on the tensor and represent it in slice form as follows: Among them, k∈{1,...,n} is a tensor The kth slice of P is the number of targets to be detected, is a nonnegative diagonal matrix; the matrix Diagonal matrix C k The diagonal elements of are stacked as row vectors of matrix C, C k =diag(C(k,:)); then we get the optimization problem: Using the ALS algorithm, in C k and A is fixed to B k To optimize, in B k Fixed case C k and A are optimized, and the required signal subspace B is obtained by multiple iterations. k =A((MN k-1 +1):MN k ,:).
8. The ALS-based RGSVD distributed MIMO radar direction finding system according to claim 7, characterized in that: In step S2, the variables in the ALS algorithm are initialized, and A is initialized to The left singular matrix, C1,...,C k Both are initialized to I P , add the l1 norm constraint through the optimization problem expression, solve the following equation to update B k : The iterative shrinkage threshold algorithm yields: B k =soft λt (B k (old) +α k (H k H -AC k (B k (old) ) H ) H AC k ) Among them, soft λt (·) is the soft threshold operation function, B k (old) Indicates B in the previous iteration k matrix, Indicates a fixed step value; set up Fixed B k Update matrices A and C k The specific steps are as follows: Use n Stack out tensors as mode-3 slices Tensor The constrained CP decomposition of the frontal slice of can be expressed as Tensor The mode-1 and mode-3 expansions satisfy: Among them, the row vector of C contains the matrix C k The diagonal elements of the matrix A and C are expressed as follows: When the number of iterations exceeds the preset maximum number of iterations or the residual If it is lower than the predefined threshold, it means it has entered the convergence state and the cycle stops. Represents the residual of the previous iteration, and obtains B k =A((MN k-1 +1):MN k ,:) is the signal subspace of the kth sub-receiver array.
9. An electronic device, characterized in that: The electronic device includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements the ALS-based RGSVD distributed MIMO radar direction finding method described in any one of claims 1-4.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the ALS-based RGSVD distributed MIMO radar direction finding method according to any one of claims 1 to 4 is implemented.
Citation Information
Patent Citations
Bistatic MIMO radar angle estimation method based on four-linear decomposition
CN108594194A
Double-base MIMO radar angle estimation method in colored noise background
CN108983143A
Nested MIMO radar angle estimation method and device based on tensor structure
CN112269172A
Cited By
Coherent signal DOA estimation method based on tensor decomposition
CN119828067A
