OFDM sparse channel estimation method based on joint correlation synchronization matching pursuit

Abstract

This invention discloses an OFDM sparse channel estimation method based on joint correlation synchronous matched pursuit. For OFDM baseband systems, the joint correlation synchronous matched pursuit algorithm is used to synchronously estimate the channel of multiple consecutive OFDM symbols through continuous iterative search. The joint correlation synchronous matched pursuit algorithm includes: (1) constructing a joint candidate set by comparing the residuals of the frequency domain response of each OFDM symbol with the correlation operation magnitude of each atom in the sensing matrix; (2) proposing a segmented pruning strategy for the joint candidate set to select the correct atom support; and (3) using the time correlation of channel multipath delay between consecutive symbols as the criterion for iteration stopping, proposing an iteration stopping condition that does not rely on prior information such as channel sparsity or channel noise energy. Simulation results show that this invention can achieve more robust sparse adaptive channel estimation with less time complexity.

Description

Technical Field

[0001] The invention relates to an OFDM sparse channel estimation method based on joint correlation synchronization matching pursuit. Background Art

[0002] OFDM is currently widely used in various wireless communication systems due to its efficient spectrum utilization and anti-multipath interference capabilities. In order to obtain more accurate channel state information (CSI), the traditional pilot-assisted channel estimation scheme based on the Nyquist-Shannon sampling theorem requires a large pilot overhead, which reduces the spectrum utilization efficiency of the OFDM system. A large amount of experimental data shows that most of the energy of the wireless channel is concentrated on very few channel taps, so it has a sparse characteristic. According to this characteristic, sparse channel estimation based on compressed sensing (CS) theory can save a lot of spectrum resources.

[0003] So far, many greedy pursuit sparse recovery algorithms with excellent performance and complexity advantages have been proposed. These greedy algorithms are all based on iterative loops. For the OFDM sparse channel estimation problem, their iteration stop conditions all depend on prior information such as channel sparsity or channel noise energy. These two parameters have a great impact on the performance of sparse recovery, but they are difficult to know in advance or accurately estimated by the receiver. In the equivalent baseband channel of wireless communication, the multipath delay of adjacent OFDM symbols has a stronger time correlation than the path gain. Although the path gain between adjacent symbols may be different, the path delay may remain relatively unchanged. Based on such time correlation characteristics, distributed compressed sensing (DCS) is applied to the multi-symbol joint recovery scenario of OFDM sparse channel estimation. DCS models the channel between several consecutive OFDM symbols as a sparse signal set with a common support set, which significantly improves the performance of OFDM sparse channel estimation. In some literatures, based on the DCS channel joint sparse model 2 (JSM2), the synchronous orthogonal matching pursuit (SOMP) algorithm gradually expands its common support set by finding the atom with the largest inner product sum with the current residual set in each iteration. Some literatures proposed an improved sparsity adaptive matching pursuit (IMSAMP) algorithm, which has better NMSE performance and sparsity adaptation ability than the SOMP algorithm.

[0004] However, the SOMP and IMSAMP algorithms still use the stopping iteration condition based on the residual energy threshold, and the sparsity adaptation capability of these algorithms depends on the accuracy of the channel noise energy prior information.

[0005] Therefore, it is necessary to design a new sparse channel estimation method. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide an OFDM sparse channel estimation method based on joint correlation synchronization matching pursuit, which can achieve more robust sparsity adaptive channel estimation with less time complexity.

[0007] The technical solution of the invention is as follows:

[0008] An OFDM sparse channel estimation method based on joint correlation synchronization matching pursuit is proposed. For OFDM baseband system, a joint correlation synchronization matching pursuit algorithm is used to perform synchronization estimation of channels of multiple consecutive OFDM symbols in a continuous iterative search mode.

[0009] The joint correlation synchronous matching pursuit algorithm includes:

[0010] (1) By comparing the residual of the frequency domain response on each OFDM symbol with the correlation operation size of each atom in the sensing matrix, a joint candidate set is constructed;

[0011] (2) A piecewise pruning strategy is proposed for the joint candidate set to select the correct atomic support;

[0012] (3) The time correlation of the channel multipath delay between consecutive symbols is used as the basis for determining the iteration stopping condition, and an iteration stopping condition that does not depend on prior information such as channel sparsity or channel noise energy is proposed.

[0013] The segmented pruning strategy means that according to the time correlation of the channel multipath delay between consecutive symbols, multiple atoms with high correlation with the current residual set can be selected in some iterations, ensuring accuracy while improving the efficiency of atomic support selection. In addition, a significant feature of the iteration stop condition proposed by the present invention is that it does not rely on channel prior information such as channel sparsity or channel noise energy, so the iteration stop condition of the present invention does not require the receiver to estimate the prior information such as channel sparsity or channel noise energy.

[0014] The steps for constructing the joint candidate set are as follows:

[0015] Step (1): In the kth iteration of the algorithm, the residual between the nth OFDM symbol and the previous iteration is The indices of the first Q atoms with the largest correlation are defined as the candidate set of the OFDM symbol The construction of is expressed as:

[0016]

[0017] The function Max(a, b) returns a vector of indices corresponding to the first b largest elements in vector a; abs(a) returns the magnitude of all elements in vector a; where Ψ n Represents the sensing matrix on the nth OFDM symbol, and N consecutive sym The perceptual matrices on each symbol are different; the superscript H indicates the conjugate transpose, the initial residual is the observation vector formed by the frequency response obtained at the pilot position on the nth OFDM symbol; for a total of N participating in the joint correlation synchronization channel estimation sym consecutive OFDM symbols, each of which contains a 1≤n≤N sym ;

[0018] Step (2): Substitute the N obtained in step (1) into sym indivual Take the union operation to construct the joint candidate set:

[0019]

[0020] where |Δ (k) | represents the joint candidate set Δ (k) The number of elements in the joint candidate set is called N sym Candidate atomic indices on consecutive OFDM symbol channels; composed of real atomic indices and interference atomic indices;

[0021] Step (3): Based on N sym indivual Construct the joint candidate set matrix I (k) , and obtained by the following formula:

[0022]

[0023] The goal of the segmented pruning strategy is to select the target atom index from the joint candidate set to correctly construct the target atom support. The steps are as follows:

[0024] First, in the kth iteration of the algorithm, define is the joint correlation coefficient vector of each candidate atom index in the joint candidate set, and is obtained by the following formula:

[0025] Γ (k) = count(I (k) , Δ (k) ),

[0026] The function count(I (k) , Δ (k) ) returns a (k) Each element in the matrix I(k) Then, Γ (k) The maximum value in is recorded as the maximum joint correlation coefficient D (k) =max(Γ (k) ); Finally, according to D (k) The size of the cropping strategy is divided into the following two stages:

[0027] Phase 1: When D (k) ≥αN sym When Δ (k) All joint correlation coefficients in Not less than αN sym Index As the target atom index selected in the current iteration, denoted as λ (k) ,Right now:

[0028]

[0029] Where α is the clipping threshold of the target atom index corresponding to the high reliability path, and the value range of α is [0.7, 1]. The selection of α is based on Figure 1 A curve above -9dB is selected in combination with the Q value, and the value range is [0.7, 1]. The number of newly selected target atom indexes in this stage is greater than or equal to 1;

[0030] Phase 2: When βN sym ≤D (k) <αN sym When (0<β<α), the target atom index is selected as follows:

[0031]

[0032] Where β is the iteration stop threshold, the value range of β is [0.3, α), and the selection of β is based on Figure 1 A curve below -10dB is selected in combination with the Q value, and the value range is [0.3, α); ψ n，l Represents the perception matrix Ψ of the nth OFDM symbol n The lth atom of ; the superscript H indicates conjugated transposition, It means to find the function that satisfies the objective function f(ψ n，l ) takes the target atom index l with the maximum value; the number of newly selected target atom indexes in this stage is equal to 1.

[0033] When D (k) <βN symWhen the algorithm enters this stage in the first iteration, it returns the estimated result of the channel impulse response in the current iteration. This iteration stop condition does not depend on prior information such as channel sparsity or channel noise energy, so the use of the joint correlation synchronization matching pursuit algorithm of the present invention does not require the receiver to estimate prior information such as channel sparsity or channel noise energy.

[0034] according to Figure 1 The logarithmic function property of the probability distribution curve is that the probability distribution has a large probability growth rate in the interval with a small Q. The Q used in constructing the joint candidate set is set to a number less than 10, and the typical value of Q is set to 5.

[0035] Preferably, Q=5 and α is set to 0.7.

[0036] When Q=5, β is set to 0.3.

[0037] Beneficial effects:

[0038] Sparsity adaptation and computational complexity are two important issues in sparse signal recovery. The greedy iterative tracking algorithm has been widely studied because of its low computational complexity and good sparse recovery performance. The present invention studies the sparse channel estimation of orthogonal frequency division multiplexing (OFDM) system based on the greedy iterative tracking algorithm, which usually adopts a stop condition based on channel noise energy to achieve sparsity adaptive recovery. However, due to the complex wireless channel environment, the noise energy prior information is not easy to be accurately known in advance by the receiving end. To this end, the present invention utilizes the time correlation of the channel path delay between adjacent OFDM symbols and proposes a joint correlation synchronization matching pursuit (JCSMP) algorithm under the framework of distributed compressed sensing (DCS). The algorithm does not require channel prior information such as sparsity and noise energy, and has lower computational complexity because of its parallel support set search capability. Simulation results show that compared with the traditional greedy iterative tracking algorithm, the JCSMP algorithm can achieve more robust sparsity adaptive channel estimation with less time complexity. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 Under the single-path channel with different signal-to-noise ratios, the candidate set c n Probability distribution of indices containing real atoms (M=40);

[0040] Figure 2 It is the flowchart of JCSMP algorithm;

[0041] Figure 3 Comparison of sparsity estimation errors (M=40);

[0042] Figure 4 NMSE performance comparison (M=40);

[0043] Figure 5 Comparison of sparsity estimation error under different numbers of pilots;

[0044] Figure 6 Comparison of NMSE performance under different numbers of pilots. DETAILED DESCRIPTION

[0045] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0046] Embodiment 1:

[0047] The JCSMP algorithm proposed in the present invention is based on the time correlation of the channel multipath delay between OFDM symbols, and its contributions are as follows: 1) Optimization of the iteration stop condition, with more robust sparsity adaptive performance. The time correlation of the channel multipath delay between consecutive symbols is used as the basis for determining the iteration stop condition, so that it does not rely on prior information such as channel sparsity and noise energy. 2) Lower computational complexity. A pruning strategy for multi-symbol joint candidate sets is proposed, so that the algorithm can select multiple atoms with high correlation with the current residual set in some iterations, ensuring accuracy while improving the efficiency of atomic support selection.

[0048] First assume an OFDM baseband system with N discrete Fourier transform points and a guard interval length of N. GI (N GI <N) and is greater than the maximum discrete delay in the channel. In the JSM2 framework, for sym OFDM symbol frame signal, the nth (1≤n≤N sym ) symbols can be expressed as:

[0049]

[0050] a n，l represents the complex gain of the lth path on the nth symbol. The channel path delay between each symbol of these OFDM frame signals is the same, while the complex gain of the path changes with the symbol. To this end, h n (k) is expressed as N GI Tap Finite Impulse Response Filter To describe the broadband channel in the time domain. In the actual wireless communication environment, h n is sparse, only at tap u i (1≤i≤S, 0≤u i ≤N GI -1) has a non-zero value S is the channel sparsity, and S<<NGI After discrete Fourier transform at the receiving end, the frequency response on the nth symbol can be expressed as:

[0051] f n =Gh n (2)

[0052] in is the first N of the N-th order Fourier matrices GI The sub-matrix consists of columns.

[0053] Assume that the number of pilots per OFDM symbol is M, and the pilot subcarrier position set p on the nth symbol is n ={p n，1 , p n，2 ,…,p n，M The pilot sets used by different symbols in an OFDM signal frame are different, that is, p a ≠p b (a, b∈{1, 2, ..., N sym}, and a≠b). Let the observation vector formed by the frequency response at the M pilot subcarrier positions on the nth symbol be According to the theory of compressed sensing, the observation vector of the original signal can be sparsely linearly represented by the atoms in the corresponding sensing matrix. n The sparse representation is:

[0054] y n =Ψ n h n +z n , (3)

[0055] The perception matrix is from the G matrix according to the set p n The submatrix composed of the row vectors extracted from the elements in . where the column vector ψ n，l It is called the perception matrix Ψ n atoms, l∈{0, 1, …, N GI -1} is the index corresponding to the atom. n is the random noise vector on the nth symbol. Formula (3) can also be expressed as:

[0056]

[0057] Known y n and n The core idea of ​​the sparse recovery algorithm based on greedy iterative tracking is to accurately search for the support set Λ={u1,…,u s}, and then estimate the corresponding path complex gain coefficient by least squares approximation Then recover the sparse signal h n .

[0058]

[0059] in Indicates the number of cells extracted from Ψ using the element in A as the index. n The submatrix consisting of the columns of .

[0060] In formula (5), for a practical system, how to accurately search for the support set Λ in the case of unknown channel sparsity prior information S will directly affect the overall performance of sparse signal recovery.

[0061] Channel Estimation Search Algorithm

[0062] The greedy iterative algorithm obtains the target atom by comparing the residual with the relevant operation size of each atom in the perception matrix in a continuous iterative search manner, and the atom with the largest inner product value ψ n，λ is considered as a real atom, and its index is selected into the support set. The atom search process can be expressed as:

[0063]

[0064] in is the residual of the previous iteration of the observation vector on the nth OFDM symbol. However, due to the non-orthogonality between atoms and the influence of channel noise, the true atom does not always have the maximum inner product value with the residual, resulting in the problem of wrong atom selection. We define the index of the first Q atoms with the maximum correlation with the residual as the candidate set c n c n The construction can be expressed as:

[0065]

[0066] The function Max(a, b) returns a vector of indices corresponding to the first b largest elements in vector a. abs(a) returns the magnitude of all elements in vector a. Obviously, the candidate set c n The probability of containing real atoms increases with the increase of Q. To this end, the present invention performs a single-path channel analysis on c at different signal-to-noise ratios (SNRs). n The probability distribution of containing real atoms (F x (Q) = P{X≤Q}) was simulated and statistically analyzed. Figure 1As shown in Figure 1, the probability that the real atom has the maximum inner product value (Q = 1) decreases as the SNR decreases. As Q increases, the probability that the real atom is searched increases in a nonlinear manner. For example, for a channel path with SNR = -12dB, the probability that the real atom is ranked first (Q = 1) is 0.22. When Q = 5, c n The probability of containing a real atom increases to 0.47, and when Q=20, it increases to 0.75.

[0067] Considering the time correlation of channel multipath delay between adjacent symbols, the c of different symbols n Based on this time correlation, the JCSMP algorithm proposed in the present invention calculates the c of each symbol. n The union operation constructs the joint candidate set:

[0068]

[0069] where |Δ| represents the number of elements in the joint candidate set Δ. Let I be the candidate set c obtained by equation (7) n The constructed joint candidate set matrix:

[0070]

[0071] Based on formulas (8) and (9), we can statistically obtain Δ j The number of occurrences in I j (Γ j ∈{1,2,…,N sym}), and define it as Δ j The joint correlation coefficient is used to measure Δ j In N sym The degree of temporal correlation between symbols. Define Γ = [Γ1, ..., Γ j ,…,Γ |Δ| ] is the joint correlation coefficient vector, and is obtained by the following formula:

[0072] Γ=count(I,Δ), (10)

[0073] The function count(I, Δ) returns a vector consisting of the number of occurrences of each element in Δ in the matrix I. The joint correlation coefficient of the real atomic indexes in Δ varies under different SNR environments. Under the same Q, an increase in SNR will increase the joint correlation coefficient of the real atomic indexes. In addition, the value of the joint correlation coefficient is also related to the number of pilots used on each OFDM symbol.

[0074] Based on the joint correlation coefficient obtained above, the JCSMP algorithm proposes a phased pruning strategy for the joint candidate set and an iterative stop condition that does not rely on prior information such as channel sparsity and noise power. The pseudo code of the algorithm is shown in Algorithm 1, and its equivalent algorithm flowchart is shown in Figure 2 shown.

[0075]

[0076]

[0077] A. Phased pruning strategy

[0078] As shown in step 2 of Algorithm 1, Γ (k) The maximum value in is recorded as the maximum joint correlation coefficient D (k) . D (k) The larger the value, the more energy there is in the residual. As the iteration progresses, the energy of the path retained in the residual will gradually decrease. (k) The value of will also decrease accordingly. (k) The size of ,the pruning strategy is divided into the following two stages.

[0079] Phase 1: When D (k) ≥αN sym When Δ (k) All joint correlation coefficients in Not less than αN sym Index As support, select at the same time, namely:

[0080]

[0081] The joint correlation coefficient is close to N sym The index of the atom whose atom has not only a higher inner product value with the current residual, but also sym The symbols have high temporal correlation. Figure 1 It can be seen that when Q = 5, the channel path with a signal-to-noise ratio greater than or equal to -9dB has a probability of being searched for c close to 80%. n In N sym The corresponding atomic index of the continuous symbols is close to 0.8N sym The joint correlation coefficient of α is 0.7. At this time, multiple channel paths with a signal-to-noise ratio of -9dB or above can be selected simultaneously in a single iteration. In this stage, the algorithm uses the correlation between atoms and residuals, and the time correlation of path delays on consecutive symbols to synchronously select multiple high signal-to-noise ratio channel paths in a highly reliable manner.

[0082] Phase 2: When βN sym ≤D(k) <αN sym When (0<β<α), the energy of the path in the residual is small. To ensure the accuracy of the selected atoms, JCSMP adopts the same support selection method as SOMP, namely:

[0083]

[0084] This method only selects the single atomic index with the largest inner product with the current residual set as support. The above-mentioned piecewise pruning strategy of "loosening" first and then "tightening" not only has high search efficiency, but also effectively guarantees the accuracy of support selection.

[0085] B. Stop Condition

[0086] Based on the JSM2 framework, when the true support is searched, the residual becomes a non-sparse signal composed of noise components, and there is no common support set. Since the pilot pattern of each symbol is different, at this time Γ (k) The elements in are all close to 1. Therefore, when D (k) <βN sym When , it can be considered that there is no common support set in the residual, and the algorithm stops searching. Also assuming Q = 5, Figure 1 It can be seen that the channel path with a signal-to-noise ratio of -14dB has a 30% probability of being searched. sym In the consecutive symbols, the corresponding atomic index has 0.3N sym The joint correlation coefficient is about . At this time, setting β to 0.3 can end the support search under the premise of ensuring that the -14dB channel path has a high probability of being searched. This scheme uses the time correlation of the channel multipath delay in adjacent symbols to determine when to stop iteration. It does not rely on the prior information in the channel. Compared with the stop condition based on the channel noise energy threshold, it has better sparsity adaptability and robustness.

[0087] C. Selection of search range Q

[0088] Figure 1 The simulation results provide a basis for the selection of the three parameters Q, α, and β. The value of Q directly affects the size of the candidate set. The larger the candidate set, the greater the probability that it contains real atoms. However, the increase in interfering atoms will also affect the performance of the pruning and stopping conditions of the joint candidate set, thereby affecting the accuracy of the overall JCSMP algorithm's atom selection. Figure 1 As shown in , the probability distribution of the candidate set containing real atoms is similar to the properties of the logarithmic function. When Q is relatively small (Q < 10), the probability increases rapidly. As Q increases, the increase rate slows down. Therefore, choosing a smaller Q value can obtain a higher pruning benefit of the joint candidate set. According to Figure 3 and Figure 4 The simulation results show that when Q is selected as 5, good performance can be obtained.

[0089] Simulation results:

[0090] This section will verify the improvement of the proposed JCSMP algorithm's sparsity adaptation capability and its advantages in supporting search efficiency. As a reference for performance comparison, the SOMP and IMSAMP algorithms use the same stop condition based on the noise energy threshold as in [4], that is, the iteration is stopped when the energy of the current residual is less than the channel noise energy prior information. The step size of the IMSAMP algorithm is set to 1. At the same time, the present invention will also discuss the performance of SOMP and IMSAMP when there is a 3dB deviation in noise energy, and verify the impact of noise energy prior information on sparse recovery. The simulation adopts a random multipath channel model, in which the number of paths and discrete path delays satisfy uniform distribution, and the path gain obeys Rayleigh distribution. The main parameters of the pilot-assisted OFDM system simulation are shown in Table 1.

[0091] Table 1

[0092] The main parameters of the simulation

[0093] Meaning (unit) symbol size Number of effective subcarriers 700 Fast Fourier Transform Length N 1024 Number of pilots M 30 / 40 / 50 The number of symbols in one OFDM frame <![CDATA[N sym ]]> 14 Channel path number random range [5，12] Channel discrete delay random range [0，60]

[0094] A. Sparsity Adaptability and NMSE Performance

[0095] Figure 3 and Figure 4 The channel sparsity estimation bias and NMSE performance of the three algorithms are compared. The definition of channel sparsity estimation bias is:

[0096]

[0097] Where S and |Λ| represent the actual and estimated channel sparsity, respectively. The formula of NMSE is defined as Referring to the path with a signal-to-noise ratio of -14dB in Fig.1, the performance of the JCSMP algorithm when α=0.7 and Q and β take three different values ​​is simulated. As can be seen from the figure, when there is a 3dB error in the noise energy prior information, the sparsity estimation accuracy and NMSE performance of the SOMP and IMSAMP algorithms are greatly reduced. Even when the noise energy prior information is assumed to be accurate, the performance of the JCSMP algorithm under three different parameter combinations is significantly better than the previous two algorithms in the SNR range of 5dB to 15dB. However, in a high signal-to-noise ratio environment, when Q is 10 and 15, the sparsity estimation deviation of the JCSMP algorithm becomes significantly larger, and the corresponding NMSE performance is also reduced. Therefore, the selection of Q should not be too large.

[0098] B. Robustness to the number of pilots

[0099] Figure 5 and Figure 6 The comparison between JCSMP and SOMP algorithms with different numbers of pilots per OFDM symbol is shown. JCSMP uses the parameter configuration of Q = 5, α = 0.7 and β = 0.3. Figure 4 It can be seen that whether the number of pilots is reduced by 25% to 30 pilots, or increased by 25% to 50 pilots, the NMSE advantage and sparsity adaptation advantage of JCSMP over SOMP remain unchanged. Therefore, the JCSMP algorithm has good robustness to the number of pilots.

[0100] C. Complexity Analysis

[0101] The total computational complexity of the SOMP algorithm is O(K s MN GI )[8]. The total computational complexity of IMSAMP is consistent with that of SAMP, which is O(K I MN GI The number of iterations of the JCSAMP algorithm can be expressed as O(K J MN GI ), where K S , K I and K J represents the number of iterations required for SOMP, IMSAMP and JCSMP respectively. The SOMP algorithm based on noise energy threshold selects only one atom at a time, so K s ≈S. The IMSAMP algorithm with a step size of 1 requires additional iterations to determine the update of the sparsity estimate, so K I ≈2S. The JCSMP algorithm selects at least one atom in each iteration, and multiple atoms can be selected in some iterations, so the number of iterations required is K J <K s Table 2 shows the average operation time of the three algorithms under SNR = 25dB environment. The simulation software runs on an Intel i5-7300HQ CPU with 16GB memory. The data shows that the operation time of IMSAMP is about twice that of SOMP. The operation time of JCSMP is nearly 20% lower than that of SOMP.

[0102] Table 2: Average operation time (seconds) when SNR is 25dB

[0103] algorithm JCSMP SOMP IMSAMP time 0.00506 0.00618 0.01340

[0104] In the case of unknown channel prior information such as channel sparsity or noise power, how to achieve good recovery performance is crucial for sparse channel estimation in actual OFDM systems. The JCSMP algorithm proposed in the present invention effectively improves the iterative stopping conditions and support selection efficiency of the greedy iterative algorithm under the DCS framework. Simulation results show that the algorithm is more robust to channel prior information and has better support set search efficiency. It is mainly reflected in: 1) The optimal stopping condition design based on the time correlation of the path delay between consecutive OFDM symbols improves the robustness of the algorithm to channel prior information; 2) The pruning strategy for the joint candidate set enables the algorithm to synchronously select multiple supports with high reliability in some iterations, greatly improving the support set search efficiency.

Claims

1. An OFDM sparse channel estimation method based on joint correlation synchronization matching pursuit, characterized in that: For OFDM baseband system, the joint correlation synchronization matching pursuit algorithm is used to estimate the channel synchronization of multiple consecutive OFDM symbols in a continuous iterative search manner; The joint correlation synchronous matching pursuit algorithm includes: (1) By comparing the residual of the frequency domain response on each OFDM symbol with the correlation operation size of each atom in the sensing matrix, a joint candidate set is constructed; (2) A piecewise pruning strategy is proposed for the joint candidate set to select the correct atomic support; (3) The time correlation of the channel multipath delay between consecutive symbols is used as the basis for determining the iteration stopping condition, and an iteration stopping condition that does not depend on the channel sparsity or the prior information of the channel noise energy is proposed; The steps for constructing the joint candidate set are as follows: Step (1): In the kth iteration of the algorithm, the residual between the nth OFDM symbol and the previous iteration is The indices of the first Q atoms with the largest correlation are defined as the candidate set of the OFDM symbol The construction of is expressed as: The function Max(a, b) returns a vector consisting of the indices corresponding to the first b largest elements in vector a; abs(a) returns the magnitudes of all elements in vector a; where Ψn represents the perception matrix on the nth OFDM symbol, and N consecutive sym The perceptual matrices on each symbol are different; the superscript H indicates the conjugate transpose, the initial residual is the observation vector formed by the frequency response obtained at the pilot position on the nth OFDM symbol; for a total of N participating in the joint correlation synchronization channel estimation sym consecutive OFDM symbols, each of which contains a Step (2): Substitute the N obtained in step (1) into sym indivual Take the union operation to construct the joint candidate set: where |Δ (k) | represents the joint candidate set Δ (k) The number of elements in the joint candidate set is called N sym Candidate atom indices on consecutive OFDM symbol channels; Step (3): Based on N sym indivual Construct the joint candidate set matrix I (k) , and obtained by the following formula: The goal of the segmented pruning strategy is to select the target atom index from the joint candidate set to correctly construct the target atom support. The steps are as follows: First, in the kth iteration of the algorithm, define is the joint correlation coefficient vector of each candidate atom index in the joint candidate set, and is obtained by the following formula: Γ (k) =count(I (k) ,Δ (k) ), The function count(I (k) , Δ (k) ) returns a (k) Each element in the matrix I (k) Then, Γ (k) The maximum value in is recorded as the maximum joint correlation coefficient D (k) =max(Γ (k) ); Finally, according to D (k) The size of the cropping strategy is divided into the following two stages: Phase 1: When D (k) ≥αN sym When Δ (k) All joint correlation coefficients in Not less than αN sym Index As the target atom index selected in the current iteration, denoted as λ (k) ,Right now: Where α is the clipping threshold of the target atom index corresponding to the high reliability path, and the value range of α is [0.7, 1]. The number of newly selected target atom indexes in this stage is greater than or equal to 1; Phase 2: When βN sym ≤D (k) <αN sym When 0<β<α, the target atom index is selected as follows: Where β is the iteration stop threshold, and the value range of β is [0.3,α), ψ n,l The sensing matrix Ψ represents the nth OFDM symbol n The lth atom of ; the superscript H indicates conjugated transposition, It means to find the function that satisfies the objective function f(ψ n，l ) takes the target atom index l with the maximum value; the number of newly selected target atom indexes in this stage is equal to 1.

2. The OFDM sparse channel estimation method based on joint correlation synchronization matching pursuit according to claim 1, characterized in that: When D (k) <βN sym When , the joint correlation synchronization matching pursuit algorithm stops iterating and returns the estimation result of the channel impulse response in the previous iteration; When the algorithm enters this phase in the first iteration, the estimated result of the channel impulse response in the current iteration is returned.

3. The OFDM sparse channel estimation method based on joint correlation synchronization matching pursuit according to claim 1, characterized in that: The Q used when constructing the joint candidate set is set to a number less than 10.

4. The OFDM sparse channel estimation method based on joint correlation synchronization matching pursuit according to claim 3, characterized in that: Q=5, α is set to 0.

7.

5. The OFDM sparse channel estimation method based on joint correlation synchronization matching pursuit according to claim 4, characterized in that: Q=5, β is set to 0.3.

Images