Parameter estimation method based on multi-measurement matrix sparse reconstruction

Abstract

The invention discloses a parameter estimation method based on multi-measurement matrix sparse reconstruction, which is characterized in that joint sparse modeling is carried out on corresponding channel parameters of continuous symbols by utilizing time correlation joint estimation parameters of channels, and a two-stage block structure multi-measurement vector orthogonal matching pursuit algorithm is provided. The problem of signal reconstruction with a double-layer sparse model is solved. Simulation results show that compared with an existing block structure multi-measurement vector reconstruction method, the parameter estimation method based on multi-measurement matrix sparse reconstruction has higher parameter estimation precision.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a parameter estimation method based on sparse reconstruction of multiple measurement matrices. Background Art

[0002] Integrated Sensing and Communication (ISAC) is considered one of the key candidate technologies for 6G because it can meet higher transmission rates and lower latency. By sharing the same spectrum and most of the hardware platform, ISAC provides a solution to the problem of increasingly crowded wireless spectrum resources. At present, there are three main types of ISAC system designs, namely communication-centric design, radar-centric design, and joint design and optimization. Based on the widespread deployment of existing communication technologies and cellular networks, the communication-centric ISAC system is able to integrate sensing functions into mature wireless communication systems. Therefore, compared with the other two designs, the communication-centric ISAC system design is easier to implement by updating existing equipment, with fewer changes to the existing wireless communication network, and has the advantage of achieving dual functions at a lower cost.

[0003] As a typical communication-centric ISAC system, the Perceptive Mobile Network (PMN) incorporates perception capabilities into cellular networks and provides ubiquitous perception capabilities while providing lossless mobile communication services. Thanks to the widely deployed cellular networks, PMN can achieve large-scale perception coverage and excellent communication capabilities. There are mainly two types of perception in PMN: uplink perception and downlink perception. In uplink perception, the uplink communication signal of the user equipment transmitter is used to perceive the environment between the user and the Remote Radio Unit (RRU). Downlink perception can be further divided into active perception and passive perception, which respectively use the echo of its own transmission signal and the transmission signal of other RRUs to achieve perception.

[0004] For PMN, the parameters that need to be sensed include angle of arrival (AoA), angle of elevation (AoD), delay, Doppler frequency, etc., all of which are included in the channel state information. In other words, the PMN system not only focuses on the overall channel coefficient estimation, but also focuses on the parameter perception of the channel. However, the channel estimation technology of existing communication systems mainly focuses on channel coefficient estimation, and the parameters that can be estimated at the same time are very limited, which may not meet the needs of PMN.

[0005] Compressed Sensing (CS) is an effective and efficient technology. In channel estimation, it uses the inherent sparsity of channels in multiple domains due to limited scattering, such as the delay-Doppler domain and the angle domain, and has been widely used in traditional channel estimation and parameter perception. Although CS has made many achievements in channel estimation, there are still very few literatures on the research of PMN perception methods based on CS. In order to solve the positioning problem in PMN, the prior art proposes an iterative sparse recovery method to optimize the estimation accuracy of the covariance matrix and the power spectrum. The prior art uses the sparsity of the delay domain to establish a generalized delay quantization channel model for PMN. The perception parameters in this model are estimated by recovering the sparse signal of the block multiple measurement vector (MMV) and solved by block sparse Bayesian learning (BSBL). Although these studies have proposed perception methods with good performance, the computational complexity is high and cannot meet the efficiency requirements in PMN scenarios.

[0006] In addition, the above-mentioned documents all assume that the channel coefficients and channel parameters usually remain almost constant in certain consecutive symbol periods, and map this time correlation to the channel sparse domain to form a typical MMV model, in which multiple sparse signals share the same non-zero support position. However, due to the mobility of scattering, the channel cannot be completely unchanged, that is, the joint sparse coefficients of the channel in its sparse domain only share part of the non-zero support. The present invention refers to the corresponding model as atypical MMV. Some scholars have conducted research on the signal reconstruction problem under this atypical model. In the prior art, the MMV model of multi-user detection is converted into a single measurement vector model with a block structure. Using the intermediate reconstruction results, pruning is performed according to the noise threshold to eliminate the false active users corresponding to the reconstructed zero elements. The prior art uses a sparse Bayesian learning method to solve the atypical MMV problem, in which the prior information of row sparsity and cluster sparsity is utilized. These studies provide solutions for the recovery of atypical MMV models, but are only applicable to specific scenarios. Summary of the invention

[0007] In order to solve the limitations and defects of the prior art, the present invention provides a parameter estimation method based on sparse reconstruction of multiple measurement matrices, comprising:

[0008] Step S1, obtain the matrix Y = [Y1, Y2, ..., Y T ] and the measurement matrix Φ are used to reconstruct the matrix X = [X1, X2, …, X T ];

[0009] Step S2: Set the first support set, the second column support set, and the second row support set of the matrix X. The first support set is The second layer support set is The second layer row support set is Where t=1,…,T,i=1,…,N1,M R The number of antenna units equipped for the radio remote unit, M U The number of antenna units equipped for the user, T is the number of orthogonal frequency division multiplexing symbols;

[0010] Step S3: In the first stage of the k-th iteration, the non-zero row block index of the matrix X is obtained according to the correlation coefficient between the matrix Y and the measurement matrix Φ. The expression of the non-zero row block index of the k-th iteration is as follows:

[0011]

[0012] Among them, R (k-1) is the residual after k-1 iterations, R (0) =Y, g=1,…,N1;

[0013] Step S4: In the second stage of the kth iteration, identify the non-zero sub-block X in the non-zero row block X[λ] according to the non-zero row block index λ. j [λ], j = 1, ..., T;

[0014] Step S5: according to time t Divided into c k Group, vector For recording c k The support set of the group, vector For recording c k The index of the group, for the i-th group (i=1,…,c k ), (k) (i) includes all subscript indices belonging to the i-th group, Including the support set of the i-th group, according to the row support set Vector Ξ (k-1) and vector Update support set;

[0015] Step S6: according to the updated vector (k) and vector The least squares method is used to reconstruct the signal for each group separately, and the expression is as follows:

[0016]

[0017] Among them, Y Ξ(i) is the joint signal of the i-th group;

[0018] Step S7: Reconstruct the signal and sub-block X t [g], estimate the perception parameters of the g-th multipath The expression is as follows:

[0019]

[0020] Among them, p and q represent X t The p-th column and q-th row of [g], Θ g is the row support of X[g], k=gmodK, g=1,…,N1.

[0021] Optionally, step S4 includes:

[0022] Defining vectors The jth element in is

[0023] Obtain the coefficient variance CV(Σ), the expression is as follows:

[0024] CV(Σ)=std(Σ) / avg(Σ),(7)

[0025] Where std(Σ) is the standard deviation of Σ, and avg(Σ) is the average value of Σ;

[0026] When CV(Σ)≥α, the sub-block is identified, where α is a preset threshold;

[0027] Arrange the elements in the vector Σ in descending order to obtain a new matrix

[0028] Pair Matrix Perform difference operation to obtain matrix D;

[0029] According to the maximum value v of matrix D D Divide the sub-blocks into non-zero sub-blocks and all-zero sub-blocks;

[0030] The non-zero sub-blocks X in the non-zero row block X[λ] j [λ], j=1,...,T, the indexes are collected into the row support set Θ λ .

[0031] Optionally, step S5 includes:

[0032] definition where i=1,…,c k-1 ,

[0033] For the range from 1 to c k-1 For different i, the support set is updated according to the following expression:

[0034]

[0035] Where λ is the non-zero row block index.

[0036] The present invention has the following beneficial effects:

[0037] The present invention provides a parameter estimation method based on sparse reconstruction of multiple measurement matrices, which jointly estimates parameters by using the time correlation of channels, performs joint sparse modeling on the channel parameters corresponding to continuous symbols, and proposes a two-stage block structure multi-measurement vector orthogonal matching pursuit algorithm to solve the signal reconstruction problem with a double-layer sparse model. Simulation results show that the parameter estimation method based on sparse reconstruction of multiple measurement matrices provided by the present invention has higher parameter estimation accuracy than the existing block structure multi-measurement vector reconstruction method. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic diagram of a perception scenario of a perception mobile network provided in Embodiment 1 of the present invention.

[0039] Figure 2 A schematic diagram of double-layer sparse structure provided in the first embodiment of the present invention.

[0040] Figure 3 This is a flow chart of a parameter estimation method based on sparse reconstruction of multiple measurement matrices provided in the first embodiment of the present invention.

[0041] Figure 4a A schematic diagram of uplink perceived multipath delay estimation accuracy provided in the first embodiment of the present invention.

[0042] Figure 4b Another schematic diagram of uplink perceived multipath delay estimation accuracy provided in the first embodiment of the present invention.

[0043] Figure 4c A schematic diagram of another uplink perceived multipath delay estimation accuracy rate provided in the first embodiment of the present invention.

[0044] Figure 5a A schematic diagram of uplink perception parameter estimation performance provided in Embodiment 1 of the present invention.

[0045] Figure 5b Another schematic diagram of uplink perception parameter estimation performance provided by the first embodiment of the present invention.

[0046] Figure 5c A schematic diagram of another uplink perception parameter estimation performance provided in the first embodiment of the present invention.

[0047] Figure 6a This is a comparison diagram between the first uplink perception estimation value and the actual value provided in the first embodiment of the present invention.

[0048] Figure 6b This is a comparison diagram between the second uplink perception estimation value and the actual value provided in the first embodiment of the present invention.

[0049] Figure 6c This is a comparison diagram of the third uplink perception estimation value and the actual value provided in the first embodiment of the present invention.

[0050] Figure 6d This is a comparison diagram of the fourth uplink perception estimation value and the actual value provided in the first embodiment of the present invention.

[0051] Figure 6e This is a comparison diagram of the fifth uplink perception estimation value and the actual value provided in the first embodiment of the present invention.

[0052] Figure 6f This is a comparison diagram of the sixth uplink perception estimation value and the actual value provided in the first embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to enable those skilled in the art to better understand the technical solution of the present invention, the parameter estimation method based on sparse reconstruction of multiple measurement matrices provided by the present invention is described in detail below with reference to the accompanying drawings.

[0054] Embodiment 1

[0055] 1. General Introduction

[0056] In the prior art, the estimation model of perceptual parameters in perceptive mobile networks (PMNs) has been formulated as a generalized time-delay quantization channel model and solved as a block multiple measurement vector (BMMV) reconstruction problem. In order to further improve the reconstruction performance, this embodiment uses the time correlation of the channel to jointly estimate the parameters. The channel parameters corresponding to continuous symbols are jointly sparsely modeled, and a two-stage block multiple measurement vector orthogonal matching pursuit (TS-BMMV-OMP) is proposed to solve the signal reconstruction problem with a two-layer sparse model. Simulation results show that the algorithm has higher parameter estimation accuracy than the existing BMMV reconstruction algorithm.

[0057] 2. Parameter Estimation Scheme

[0058] This embodiment will Figure 1 The PMN system in RThe remote radio units (RRUs) work together to provide services for K users, who communicate based on orthogonal frequency division multiplexing (OFDM) signals. The RRUs and users are equipped with uniform linear antenna arrays, which consist of M R and M U The OFDM symbol period is T s , insert pilot symbols for channel and parameter estimation. Assuming B represents the signal bandwidth and P is the number of subcarriers, the interval between two subcarriers is f0 = B / P.

[0059] This embodiment considers uplink sensing, and the RRU receives multipath signals from all active users. Due to the complexity of the communication environment, it is assumed that the line-of-sight path from each user to the RRU is blocked. For a certain RRU, the tth pilot symbol y received on the pth subcarrier t,p The expression is as follows:

[0060]

[0061] in, is the pilot signal sent by the kth user, is Additive White Gaussian Noise (AWGN), L k is the number of multipaths from the kth user. For the lth multipath (l=1,…,L k ), β k,l is the amplitude, τ k,l is the propagation delay that depends on the propagation distance, and θ l is the angle of arrival (AoA) and the angle of elevation (AoD), f D,k,l is the Doppler frequency related to the target moving speed. a(M,φ)=[1,e -jπsinφ ,...,e -jπ(M-1)sinφ ] · is the array response vector of the ULA, where M is the number of antennas and φ is or θ l .

[0062] By adding y t,p Rewriting it in matrix form, we get the following expression:

[0063]

[0064] in,

[0065] In the above expression, The lth row vector of A k (M U ,θ k ) is a(M U ,θ k,l ). B(M U ,θ) is therefore an M U K×L block diagonal matrix. D p and F t is a diagonal matrix of size L×L, where is the total number of multipaths for all users, and and

[0066] In order to achieve efficient parameter perception, we first need to use a delay quantization model to represent the received signal y t,p To obtain a specific representation, the following description is first given:

[0067] 1) Define the speed of electromagnetic waves as c, then the propagation delay of the lth multipath of the kth user is τ k,l =d k,l / c, where d k,l is the corresponding propagation distance;

[0068] 2) Assume (0,d max The propagation distance within the range is quantized into Q grid points, and the quantization interval is d0 = d max / Q. For y in formula (2) t,p The delay term When reconstructed at the receiving end, it is quantized as Where g0 = f0d0 / c, and the quantization grid is

[0069] 3) At this time, the delay term at the sending end When reconstructing at the receiving end, it can only be done by When Q is large enough (especially when Q>>L), the delay estimation error introduced by the quantization error can usually be ignored.

[0070] According to the above description, the received signal y t,p The expression of the grid model is as follows:

[0071]

[0072] in, and They are and Ft The matrices obtained by rearranging the delay quantization values ​​from small to large are M in size. R ×Q and Q×Q. is a mapping matrix used to map multipaths to signals sent by users, where only rows related to L multipaths have non-zero elements, and each row has only one 1, and the rest of the elements are 0.

[0073] The received signal y t,p When all subcarrier signals are combined and written in matrix form, the expression is as follows:

[0074]

[0075] Among them, Φ t The pth row of and

[0076] U l is the lth row of U, Z t =(z t,1 ,…,z t,p ,…,z t,P ) · is the noise matrix.

[0077] Assume that the pilot symbols transmitted on T consecutive orthogonal frequency division multiplexing (OFDM) symbols are the same, that is, Φ t =Φ.

[0078] make Where k = gmodK, g = 1, ..., N1 and N1 = QK, then expand Y t The expression of the joint model of multiple symbols is as follows:

[0079]

[0080] In the above expression, the matrix X=[X1,X2,…,X T ]like Figure 2 As shown in the figure, different colors are used to distinguish different users. For example, the red part marked by the red dotted line, namely X[i], represents the multipath from user 1. The part circled by the yellow box represents a sub-block, namely X t [i]. From this figure, we can see that X has double-layer sparsity. In order to distinguish different levels of sparsity, we define two levels of support sets: X = [X1, X2, …, X T The first-level support set of ] is The second layer defines the support set as two dimensions, namely X tThe column support set of The row support of X[i] is Where t=1,…,T, i=1,...,N1.

[0081] Figure 3 A flow chart of a parameter estimation method based on multi-measurement matrix sparse reconstruction provided in Example 1 of the present invention. For the above model, in order to reconstruct X, this embodiment proposes a two-stage block structure multi-measurement vector orthogonal matching pursuit (Two-Stage Block Multiple Measurement Vector Orthogonal Matching Pursuit, TS-BMMV-OMP) algorithm to restore the double-layer block sparse signal. The TS-BMMV-OMP algorithm is implemented through iteration, and the identification of non-zero positions in each iteration is divided into two stages. The goal of the first stage is to identify non-zero row blocks, that is, to select the first-level support set Ω, and the second stage is to further identify the non-zero sub-blocks therein, that is, to select the row support set Θ i The algorithm is described in detail as follows:

[0082] 1) Two-stage recognition for non-zero row blocks and non-zero sub-blocks

[0083] In the first phase of the kth iteration, the non-zero row block indices of X are selected by the correlation coefficient between Y and the measurement matrix Φ. Considering all X t The joint sparsity of , the index of the non-zero row block is selected at the kth iteration as follows:

[0084]

[0085] Among them, R (k-1) is the residual after k-1 iterations, and R (0) =Y.

[0086] Based on the selected row index λ, the goal of the second stage is to identify the non-zero sub-blocks X in the row block X[λ] j [λ], j = 1, ..., T. First, define the vector The jth element in In order to identify non-zero sub-blocks in the non-zero row X[λ], this embodiment uses the coefficient variance CV(Σ) to determine whether there is a zero sub-block in X[λ]. The coefficient variance is defined as follows:

[0087] CV(Σ)=std(Σ) / avg(Σ),(7)

[0088] In the above expressions, std(Σ) and avg(Σ) are the standard deviation and average value of Σ. In fact, for a certain X[λ], the Σ corresponding to the zero sub-block is jIt is obviously smaller than the non-zero sub-block, so this feature can be used to identify the zero sub-block. When CV(Σ) is large, the probability that there is a zero sub-block in X[λ] is also large. When CV(Σ)≥α, the operation of identifying the sub-block is performed, where α is the threshold. Then, by sorting the elements in Σ in descending order, the new matrix is ​​obtained Then through Perform a differential operation to obtain D, and take the maximum value v of D as D is regarded as the boundary between the non-zero sub-block and the all-zero sub-block. Finally, according to the partition result, the indexes of the non-zero sub-blocks in X[λ] are collected into the row support set Θ λ middle.

[0089] 2) Update support set and signal reconstruction

[0090] Based on the above results of non-zero position identification, According to the time t, it is divided into c k groups, and reconstruct the signal for each group. Vector and vector Used to record the c k The support set and index of the group. For the i-th group (i=1,…,c k ), (k) (i) Contains all subscript indices t belonging to the group. Contains the support set of the i-th group, that is, In the kth iteration, according to the obtained Ξ (k-1) and Update the support set. First, for i=1,…,c k-1 ,definition Then, according to the k-1 For different i in the range, the update process is performed according to the following expression:

[0091]

[0092] Based on the updated (k) and Each sub-block is reconstructed in groups through the least squares (LS) process, as shown below:

[0093]

[0094] Among them, Y Ξ(i) represents the joint signal of the i-th group.

[0095] This embodiment provides a parameter estimation method based on sparse reconstruction of multiple measurement matrices, utilizes the time correlation of the channel to jointly estimate the parameters, performs joint sparse modeling on the channel parameters corresponding to continuous symbols, and proposes a two-stage block structured multiple measurement vector orthogonal matching pursuit algorithm to solve the signal reconstruction problem with a double-layer sparse model.

[0096] 3) Perception parameter estimation

[0097] According to the Multiple Measurement Vector (MMV) model, the estimation of perceptual parameters is based on the reconstructed signal According to the estimated sub-block X t [g], the perception parameter of the g-th multipath Calculated by the following expression:

[0098]

[0099] Among them, Θ g is the row support of X[g].

[0100] 3. Simulation Results

[0101] This embodiment gives the simulation results in the uplink scenario of the Perceptive Mobile Network (PMN) to verify the effectiveness of the TS-BMMV-OMP algorithm. In the simulation, there is a remote radio unit (RRU) in the PMN system, which is connected to 4 active users through multiple input multiple output (MIMO). The RRU is equipped with 6 antennas, and each user is equipped with 4 antennas. The carrier frequency is set to 2.35GHz and the signal bandwidth is 100MHz. The number of uplink subcarriers is 256, and the path loss factor is 2. The number of symbols estimated jointly is T=10. The multipath signals of each user are randomly generated in clusters to simulate the reflected / scattered signals from objects. For each cluster, the parameters are set as follows, where each parameter variable is set to be uniformly distributed within its value range: the range of the number of multipaths is [10, 15], the range of the incoming wave direction is [0°, 45°], the range of the distance is [0, 50] meters, and the range of the Doppler frequency is [0, 600] Hz. For the grid model, this embodiment sets w=2, Q=200, and the threshold is α=0.2.

[0102] Table 1 Average operation time of the reconstruction algorithm in simulation

[0103] <![CDATA[L c ]]> TS-BMMV-OMP BSBL BOMP 1 0.8576s 4.3659s 0.2378s 3 1.0027s 5.0663s 0.2674s 5 0.9867s 4.5513s 0.2601 s 7 0.9750s 4.7028s 0.2511s

[0104] First, this embodiment intends to evaluate the algorithm from three aspects: reconstruction performance, time consumption, and perception performance. In addition, this embodiment intends to select BOMP and BSBL algorithms to further compare with the proposed algorithm to reflect the superiority of the proposed algorithm. Figure 4a-4c The accuracy of correctly estimating multipath delay is given, where the horizontal axis L c Indicates that among all multipaths of a user, there are L c The state of each multipath changes over time. For example, Figure 2 In the example, the multipath of user 1 in the second grid remains unchanged from t=1 to 5, while the multipath in the Qth grid changes with time. The path appears at t=2 and disappears at t=5. At this time, L c =1. Figure 4a-4c The results show that when the sparsity is unknown, the proposed algorithm achieves better performance than the comparison algorithm. Table 1 shows the average time consumption of the algorithm to reconstruct the signal. The results show that TS-BMMV-OMP has a shorter running time than BSBL; due to the additional LS operation introduced by group reconstruction, it takes a little longer than the traditional BOMP algorithm. Figure 4a-4c It can be seen that TS-BMMV-OMP has better reconstruction performance with only a slight increase in complexity.

[0105] Considering that the ultimate goal of TS-BMMV-OMP is parameter estimation, Figure 5a-5c The performance of parameter estimation when SNR varies is shown in That is, the probability of correctly estimating the delay, the arrival angle and the estimated value NMSE of the Doppler frequency. Figure 5a-5c The results show that TS-BMMV-BOMP has better non-zero sub-block identification and parameter estimation performance than other algorithms. In order to verify the performance gain of the proposed algorithm in more detail, c =7 and SNR = 20dB, Figure 6a-6f The actual and estimated values ​​of the arrival angle and Doppler frequency are given, which are represented by circles and stars, respectively. Comparing the distances between each pair of circles and stars, it can be seen that the stars estimated by TS-BMMV-OMP are closer to the corresponding circles, that is, the estimated values ​​of the proposed algorithm are closer to the actual parameter values.

[0106] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present invention, but the present invention is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A parameter estimation method based on sparse reconstruction of multiple measurement matrices, characterized in that: include: Step S1, obtain the matrix Y = [Y1, Y2, ..., Y T ] and the measurement matrix Φ are used to reconstruct the matrix X = [X1,X2,…,X T ]; Step S2: Set the first support set, the second column support set, and the second row support set of the matrix X. The first support set is The second layer support set is The second layer row support set is Where t=1,…,T,i=1,…,N1,M R The number of antenna units equipped for the radio remote unit, M U The number of antenna units equipped for the user, T is the number of orthogonal frequency division multiplexing symbols; Step S3: In the first stage of the k-th iteration, the non-zero row block index of the matrix X is obtained according to the correlation coefficient between the matrix Y and the measurement matrix Φ. The expression of the non-zero row block index of the k-th iteration is as follows: Among them, R (k-1) is the residual after k-1 iterations, R (0) =Y, g=1,…,N1; Step S4: In the second stage of the kth iteration, identify the non-zero sub-block X in the non-zero row block X[λ] according to the non-zero row block index λ. j [λ], j = 1, ..., T; Step S5: according to time t Divided into c k Group, vector For recording c k The support set of the group, vector For recording c k The index of the group, for the i-th group (i=1,…,c k ), (k) (i) includes all subscript indices belonging to the i-th group, Including the support set of the i-th group, according to the row support set Vector Ξ (k-1) and vector Update support set; Step S6: according to the updated vector (k) and vector The least squares method is used to reconstruct the signal for each group separately, and the expression is as follows: Among them, Y Ξ(i) is the joint signal of the i-th group; Step S7: Reconstruct the signal and sub-block X t [g], estimate the perception parameters of the g-th multipath The expression is as follows: Among them, p and q represent X t The p-th column and q-th row of [g], Θ g is the row support of X[g], k=gmodK, g=1,…,N1.

2. The parameter estimation method based on sparse reconstruction of multiple measurement matrices according to claim 1, characterized in that: The step S4 comprises: Defining vectors The jth element in is Obtain the coefficient variance CV(Σ), the expression is as follows: CV(Σ)=std(Σ) / avg(Σ),(7) Where std(Σ) is the standard deviation of Σ, and avg(Σ) is the average value of Σ; When CV(Σ)≥α, the sub-block is identified, where α is a preset threshold; Arrange the elements in the vector Σ in descending order to obtain a new matrix Pair Matrix Perform difference operation to obtain matrix D; According to the maximum value v of matrix D D Divide the sub-blocks into non-zero sub-blocks and all-zero sub-blocks; The non-zero sub-blocks X in the non-zero row block X[λ] j [λ], j=1,...,T, the indexes are collected into the row support set Θ λ .

3. The parameter estimation method based on sparse reconstruction of multiple measurement matrices according to claim 2, characterized in that: The step S5 comprises: definition where i=1,…,c k-1 , For the range from 1 to c k-1 For different i, the support set is updated according to the following expression: Where λ is the non-zero row block index.

Images