IRS channel estimation method, device, equipment and storage medium

Abstract

The present invention provides an IRS channel estimation method, apparatus, device and storage medium. The method obtains a received signal vector and a perception matrix; uses a list-based OMP algorithm to select n groups of support columns from the perception matrix in each iteration until the iteration is terminated, and determines the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration; obtains a priori vector based on the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration; obtains residual coefficients based on the received signal vector and the residuals corresponding to the n groups of support columns selected in the last iteration; determines initial hyperparameters based on the priori vector and the residual coefficients, and performs IRS channel estimation based on the initial hyperparameters and an SBL algorithm, thereby providing an effective priori for the SBL algorithm and reducing the computational complexity of the SBL algorithm.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to an IRS channel estimation method, apparatus, device and storage medium. Background Art

[0002] With the exploration and development of future network systems (such as 6G), intelligent reflecting surfaces (IRS) are finding more applications in millimeter wave (mmWave) communications. IRS can transmit signals to users whose direct channel to the base station (BS) is blocked, and its passive transmission characteristics make it easier to implement and maintain in practical applications.

[0003] Channel estimation (CE) plays a key role in communications. The channel state information (CSI) it provides is used in precoding and beamforming. Currently, many compressed sensing (CS) algorithms, such as sparse Bayesian learning (SBL), can achieve highly accurate channel estimation. However, traditional SBL algorithms are computationally very complex. Summary of the Invention

[0004] In view of the problems existing in the prior art, the present invention provides an IRS channel estimation method, apparatus, device and storage medium.

[0005] In a first aspect, the present invention provides an IRS channel estimation method, comprising:

[0006] Obtain received signal vector and perception matrix;

[0007] Using a list-based orthogonal matching pursuit (OMP) algorithm, n groups of support columns are selected from the perception matrix in each iteration until the iteration is terminated, and the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration are determined; n is an integer greater than or equal to 1;

[0008] Obtaining a priori vectors based on the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration; obtaining residual coefficients based on the received signal vector and the residuals corresponding to the n groups of support columns selected in the last iteration;

[0009] Based on the prior vector and the residual coefficient, initial hyperparameters are determined, and IRS channel estimation is performed based on the initial hyperparameters and a sparse Bayesian learning (SBL) algorithm.

[0010] Optionally, selecting n groups of support columns from the perception matrix in each iteration includes:

[0011] In the first iteration, n support columns corresponding to the smallest residuals are selected from the perception matrix;

[0012] In any iteration after the first iteration, determine the n groups of support columns selected in the iteration before the any iteration, and for any group of support columns, determine the n support columns with the smallest corresponding residuals from the perception matrix, and branch the any group of support columns into n groups of support columns based on the determined n support columns with the smallest corresponding residuals to obtain n*n groups of support columns, and then select n groups of support columns with the smallest corresponding residuals from the n*n groups of support columns.

[0013] Optionally, determining n support columns corresponding to the smallest residuals from the perception matrix, and branching the arbitrary group of support columns into n groups of support columns based on the determined n support columns corresponding to the smallest residuals, includes:

[0014] According to the residual of each support column other than the arbitrary group of support columns in the perception matrix corresponding to the arbitrary group of support columns, determining n support columns with the smallest corresponding residual from each of the other support columns;

[0015] Each support column in the n support columns is respectively combined with the arbitrary group of support columns to obtain n groups of support columns branched from the arbitrary group of support columns.

[0016] Optionally, obtaining the prior vector based on the n groups of support columns selected in the last iteration and the residuals respectively corresponding to the n groups of support columns selected in the last iteration includes:

[0017] Determine a priori vector based on the estimated matrix corresponding to the n groups of support columns selected in the last iteration and the first vector;

[0018] The first vector is determined based on the residuals corresponding to the n groups of support columns selected in the last iteration.

[0019] Optionally, determining the prior vector based on the estimated matrix corresponding to the n groups of support columns selected in the last iteration and the first vector includes:

[0020] Determine an initial prior vector based on the estimated matrix corresponding to the n groups of support columns selected in the last iteration and the first vector;

[0021] The initial a priori vector is determined as the a priori vector; or, a structural check is performed on the initial a priori vector to obtain the a priori vector.

[0022] Optionally, performing a structural check on the initial a priori vector to obtain the a priori vector includes:

[0023] Determine a set of row and column indices of all elements in the concatenated channel sparsity matrix that satisfy the first condition;

[0024] Determine a first index set based on the row and column index set;

[0025] Replacing the element corresponding to the first index set in the initial a priori vector with a first value, while keeping other elements in the initial a priori vector unchanged, to obtain the a priori vector;

[0026] The element in the cascaded channel sparsity matrix that satisfies the first condition means that the value of the element is less than a second value, and the second value is determined based on the structure coefficient for structure inspection, the cascaded channel sparsity matrix, the network structure index set, the number of IRS receive paths, and the number of IRS transmit paths;

[0027] The first value is determined based on the cascade channel sparsity matrix, the network structure index set, the number of IRS receive paths, the number of IRS transmit paths, and the number of steering vectors in the dictionary matrix at the IRS.

[0028] Optionally, obtaining residual coefficients based on the residuals corresponding to the received signal vector and the n groups of support columns selected in the last iteration includes:

[0029] A residual coefficient is obtained based on a ratio between an L2 norm of the received signal vector and a minimum value of the residuals respectively corresponding to the n groups of support columns selected in the last iteration.

[0030] Optionally, determining initial hyperparameters based on the prior vector and the residual coefficients includes:

[0031] For the element position with a value of 0 in the prior vector, determine that the element value at the corresponding position in the initial hyperparameter is 0;

[0032] For the element position in the prior vector whose value is greater than 0, the element value at the corresponding position in the initial hyperparameter is determined based on the value of the element at the element position with a value greater than 0, the residual coefficient, the smallest positive element in the prior vector, and the largest element in the prior vector.

[0033] In a second aspect, the present invention further provides an IRS channel estimation device, comprising:

[0034] An acquisition module, used to obtain a received signal vector and a perception matrix;

[0035] A first determination module is configured to use a list-based OMP algorithm to select n groups of support columns from the perception matrix in each iteration until the iteration is terminated, and determine the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration; n is an integer greater than or equal to 1;

[0036] A second determination module is configured to obtain a priori vector based on the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration; and obtain residual coefficients based on the received signal vector and the residuals corresponding to the n groups of support columns selected in the last iteration;

[0037] The third determination module is configured to determine initial hyperparameters based on the priori vector and the residual coefficients, and perform IRS channel estimation based on the initial hyperparameters and an SBL algorithm.

[0038] In a third aspect, the present invention further provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the IRS channel estimation method as described in the first aspect is implemented.

[0039] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the IRS channel estimation method described in the first aspect above.

[0040] The IRS channel estimation method, apparatus, device and storage medium provided by the present invention select n groups of support columns from a perception matrix through a list-based OMP algorithm to obtain a priori vectors and residual coefficients, and then obtain initial hyperparameters from the priori vectors and residual coefficients. The initial hyperparameters are used as the initial hyperparameters in an SBL algorithm, and channel estimation is performed through the SBL algorithm, thereby providing an effective prior for the SBL algorithm and reducing the computational complexity of the SBL algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 11 is a flow chart of the IRS channel estimation method provided by the present invention;

[0043] Figure 2 1 is a schematic diagram of simulation results of the IRS channel estimation method provided by the present invention;

[0044] Figure 3 1 is a schematic structural diagram of an IRS channel estimation device provided by the present invention;

[0045] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0046] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0047] The widely used orthogonal matching pursuit (OMP) algorithm often fails to meet estimation requirements. However, the more accurate SBL algorithm suffers from extremely high complexity and lacks performance at low signal-to-noise ratios (SNRs). Furthermore, its noise learning is not precise enough. Among these issues, the most pressing challenge is the excessive time complexity of the SBL algorithm. Several approaches have been proposed in the prior art to address this problem. For example, a Laplace prior is used to improve accuracy and efficiency; block sparsity is considered in millimeter-wave multiple-input multiple-output (MIMO) systems under quasi-static and time-selective channels to enhance CE performance; and support knowledge is incorporated into the prior. However, these approaches still exhibit high complexity and are not applicable to mmWave MIMO systems that exploit the unique channel structure exhibited by IRS assistance.

[0048] In an IRS-assisted MIMO system (where N t The transmitter Tx of the antenna contains N r The receiving end Rx of each antenna is equipped with a uniform linear array (ULA), and the IRS has a uniform planar array (UPA) with M=M x ×M yreflection units), the narrowband geometric channel model can be used to describe the IRS-Rx channel G and Tx-IRS channel R as formula 1:

[0049]

[0050] Where L1 and L2 are the number of IRS-Rx and Tx-IRS paths, and is the corresponding complex gain composed of the path loss, is the angle of arrival (AOA) at Rx, is the Azimuth of Departure (AOD) at Tx. and represent the (azimuth and altitude) AOD and AOA at the IRS, respectively. and Denote the normalized array steering vectors at Tx, Rx, and IRS, respectively. Therefore, G and R can be decomposed into Formula 2:

[0051]

[0052] in, and At Rx, IRS and Tx respectively and The dictionary matrix consists of steering vectors. Assuming that all spatial angles are on a uniform grid from -1 to 1, the cascaded channel can be defined as Equation 3:

[0053]

[0054] in, is the phase shift vector at the IRS.

[0055] During channel estimation, T pilot blocks are transmitted. For the t-th pilot block (t=1, 2, 3, ..., T), the IRS has a reflection vector Ψ t . and Beams are formed at Tx and Rx respectively therefore pilots are transmitted in one pilot block. Therefore, the received signal y t,p Corresponding to the pth Transmit beam and precoding vector ft,p It can be written as formula 4:

[0056]

[0057] Among them, W t is the merge matrix at Rx, s t,p is the pilot signal transmitted and satisfies |s t,p |=1, is Additive White Gaussian Noise (AWGN), H t is the corresponding IRS phase shift vector Ψ t Collect all channels. The transmitted signal and the received signal can be expressed in matrix form as Equation 5:

[0058]

[0059] in, and the noise matrix Therefore, Equation 5 can be vectorized into Equation 6:

[0060]

[0061] In (a), the concatenated channel in Equation 3 is substituted and In (b), the properties of Khatri-Rao product are exploited. In (c) and (d), G and R channels are decomposed according to Equation 2, exploiting the properties as well as and set up The first M of D G row, Λ is the merged version of J, as shown in Equation 7:

[0062]

[0063] in, is the set of indices that have the same row as the i-th row in D. Therefore, the received signal in Equation 6 can be simplified to Equation 8:

[0064]

[0065] The mixed product property of the Kronecker product is used After T pilot blocks, we get Formula 9:

[0066] y=Qλ+n (9)

[0067] in The dimension of y is the number of measurements Considering the sparsity of λ, a CS-based algorithm can be used. For any reflection vector Ψ used in data transmission, the cascade channel H can be calculated as Equation 10:

[0068]

[0069] The SBL algorithm aims to find the most likely concatenated channel, whose probability is defined as Equation 11:

[0070]

[0071] in, Hyperparameter γ=[γ1,γ2,…,γ G ], Γ = diag(γ). The expectation maximization algorithm (EM) can be used to iteratively update the hyperparameters and estimate the sparse vector λ.

[0072] The present invention uses a Bayesian framework-based orthogonal matching pursuit list-sparse Bayesian learning (OMPL-SBL) algorithm assisted by hyperparameter optimization lists. The algorithm consists of two phases: the OMPL phase, which generates a priori phase for channel properties that considers grid sparsity, and the SBL phase, which estimates the channel.

[0073] Figure 1 A flow chart of the IRS channel estimation method provided by the present invention is shown as follows: Figure 1 As shown, the method includes the following steps:

[0074] Step 100: Obtain a received signal vector and a perception matrix.

[0075] Specifically, the received signal vector can represent a vector composed of received signals corresponding to all pilot blocks transmitted during the IRS channel estimation period; the perception matrix can be the definition of the Q matrix described above during the IRS channel estimation period The derived Q matrix.

[0076] In one implementation, the relationship between the received signal vector and the sensing matrix can be expressed using the following formula:

[0077] y=Qλ+n

[0078] Where y represents the received signal vector, Q represents the perception matrix, λ represents the vector to be solved (also called a sparse vector), and n represents the noise vector.

[0079] Step 101: Use the list-based orthogonal matching pursuit (OMP) algorithm to select n groups of support columns from the perception matrix in each iteration until the iteration is terminated, and determine the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration; n is an integer greater than or equal to 1.

[0080] Specifically, when using the list-based OMP algorithm (ie, OMPL algorithm), a total of k max Iterations (k max is an integer greater than or equal to 1).

[0081] Unlike the traditional OMP algorithm, which selects only one set of support columns in each iteration, the OMPL algorithm selects n sets of support columns from the perception matrix in each iteration (i.e., the traditional OMP algorithm is a special case of the OMPL algorithm when n = 1). The value of n can be adjusted based on different system requirements, balancing estimation accuracy and computational complexity. For example, a larger value of n results in higher estimation accuracy.

[0082] The conditions for OMPL algorithm iteration termination can be: the number of iterations reaches the maximum number of iterations, or the minimum value of the residuals corresponding to the n groups of support columns in a certain iteration result is less than the residual threshold.

[0083] After the iteration is terminated, the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration are determined.

[0084] Step 102: Obtain a priori vector based on the residuals corresponding to the n groups of support columns selected in the last iteration and the n groups of support columns selected in the last iteration; obtain residual coefficients based on the received signal vector and the residuals corresponding to the n groups of support columns selected in the last iteration.

[0085] Specifically, after determining the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration, a priori vector can be obtained by performing corresponding operations on the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration; and a residual coefficient can be obtained by performing corresponding operations on the received signal vector and the residuals corresponding to the n groups of support columns selected in the last iteration. The priori vector and the residual coefficient can be used in subsequent channel estimation steps.

[0086] Step 103: Determine initial hyperparameters based on the prior vector and the residual coefficients, and perform IRS channel estimation based on the initial hyperparameters and the sparse Bayesian learning (SBL) algorithm.

[0087] Specifically, after determining the prior vector and the residual coefficient, corresponding operations are performed on them to determine initial hyperparameters, which can be used as initial hyperparameters in the existing SBL algorithm, and then the IRS channel estimation is performed using the SBL algorithm.

[0088] The IRS channel estimation method provided by the present invention selects n groups of support columns from a perception matrix through the OMPL algorithm to obtain a priori vectors and residual coefficients, and then obtains initial hyperparameters from the priori vectors and residual coefficients. The initial hyperparameters are used as the initial hyperparameters in the SBL algorithm, and channel estimation is performed through the SBL algorithm, thereby providing an effective prior for the SBL algorithm and reducing the computational complexity of the SBL algorithm.

[0089] Optionally, n groups of support columns are selected from the perception matrix in each iteration, including:

[0090] In the first iteration, n support columns with the smallest corresponding residuals are selected from the perception matrix;

[0091] In any iteration after the first iteration, determine the n groups of support columns selected in the iteration before any iteration, and for any group of support columns, determine the n support columns with the smallest corresponding residuals from the perception matrix, and branch any group of support columns into n groups of support columns based on the determined n support columns with the smallest corresponding residuals to obtain n*n groups of support columns, and then select n groups of support columns with the smallest corresponding residuals from the n*n groups of support columns.

[0092] In one implementation, in the first iteration, the residual corresponding to each support column of the perception matrix is ​​first calculated.

[0093] After completing the residual calculation corresponding to each support column of the perception matrix, n support columns with the smallest corresponding residuals are selected as the results of the first iteration.

[0094] In any iteration after the first iteration, n groups of support columns selected in the iteration before any iteration can be determined, and for any group of support columns, n support columns with the smallest corresponding residual are determined from the perception matrix.

[0095] For example, in the second iteration, for each of the n support columns selected in the first iteration, the residuals corresponding to the G-1 support columns other than the support column can be calculated, and then n support columns with the smallest corresponding residuals can be selected.

[0096] Based on the determined n support columns with the smallest corresponding residuals, any group of support columns can be branched into n groups of support columns to obtain n*n groups of support columns, and then n groups of support columns with the smallest corresponding residuals are selected from the n*n groups of support columns.

[0097] For a certain iteration, the residual corresponding to any set of support columns can be calculated as follows:

[0098]

[0099] Among them, r i represents the residual corresponding to the i-th group of support columns selected in this iteration, y represents the received signal vector, ||||2 represents the L2 norm operation, D i represents the matrix composed of the column vectors in the matrix D that have the same column index as the i-th group of support columns selected in this iteration, Indicates D i The pseudo-inverse, U M Represents the dictionary matrix at IRS.

[0100] Optionally, determining n support columns corresponding to the smallest residuals from the perception matrix, and branching any one group of support columns into n groups of support columns based on the determined n support columns corresponding to the smallest residuals may include:

[0101] According to the residuals of each support column except any one group of support columns in the perception matrix corresponding to any one group of support columns, n support columns with the smallest corresponding residuals are determined from each of the other support columns;

[0102] Each support column in the n support columns is respectively combined with any group of support columns to obtain n groups of support columns branched out from any group of support columns.

[0103] In one embodiment, in an iteration subsequent to the first iteration, n groups of support columns selected in the iteration preceding the first iteration may be determined. For a particular group of support columns selected in the n groups, residuals corresponding to each support column in the perception matrix other than the selected group of support columns may be calculated, and then n support columns with the smallest corresponding residuals may be determined from the calculated residuals.

[0104] By taking the union of each support column in the n support columns and the group of support columns, we can obtain n groups of support columns branched from the group of support columns.

[0105] For example, let G = 20 and n = 4. In the third iteration, the four groups of support columns selected in the second iteration are first determined. Assuming that a certain group of support columns selected in the second iteration is the first and second columns of the perception matrix, the residuals of the third to 20th columns of the perception matrix corresponding to this group of support columns can be calculated respectively, and the four support columns with the smallest corresponding residuals can be determined from them. Assuming that the four support columns with the smallest corresponding residuals are the fifth, seventh, ninth, and eleventh columns of the perception matrix, this group of support columns can be branched into four groups of support columns. These four groups of support columns are: the first, second, and fifth columns of the perception matrix; the first, second, and seventh columns of the perception matrix; the first, second, and ninth columns of the perception matrix; and the first, second, and eleventh columns of the perception matrix.

[0106] Similarly, in the kth iteration (k is greater than 1), for each group of support columns in the n groups of support columns selected in the k-1th iteration, the residuals corresponding to the G-k+1 support columns other than the k-1 support columns included in the group of support columns can be calculated. Then, n support columns with the smallest corresponding residuals are selected from them and combined with each group of support columns in the n groups of support columns selected in the k-1th iteration. Each group of support columns in the n groups of support columns selected in the k-1th iteration is branched into n groups of support columns, each of which includes k support columns, for a total of n*n groups of support columns. The residuals corresponding to each group of support columns in these n*n groups of support columns are compared, and the n groups of support columns with the smallest corresponding residuals are selected. The selected n groups of support columns with the smallest corresponding residuals are used as the result of the kth iteration. By branching each of the n groups of support columns selected in the previous iteration into n groups of support columns, the residuals corresponding to the n groups of support columns finally selected by the OMPL algorithm are made as small as possible, so that the obtained solution is closer to the actual value.

[0107] Optionally, based on the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration, a priori vector is obtained, including:

[0108] Determine the prior vector based on the estimated matrix and the first vector corresponding to the n groups of support columns selected in the last iteration;

[0109] The first vector is determined based on the residuals corresponding to the n groups of support columns selected in the last iteration.

[0110] Specifically, after determining the n groups of support columns selected in the last iteration, the estimated matrix corresponding to the n groups of support columns selected in the last iteration can be determined, and the prior vector can be further determined in combination with the first vector. The prior vector is determined by the support column group finally selected in the OMPL algorithm, so that the determined prior vector is more accurate.

[0111] In one embodiment, the estimated matrix corresponding to the n groups of support columns selected in each iteration can be determined by the following formula:

[0112]

[0113] Among them, M represents the estimated matrix corresponding to the n groups of support columns selected in any iteration, M(:, i) represents the i-th column of the estimated matrix corresponding to the n groups of support columns selected in the iteration, and D i represents the matrix composed of the column vectors in the matrix D that have the same column index as the i-th group of support columns selected in this iteration, Indicates D i The pseudo-inverse, U M Represents the dictionary matrix at IRS.

[0114] In one implementation, the first vector may be determined by the following formula:

[0115]

[0116] Among them, K represents the first vector, 1 n represents a full 1 row vector containing n elements, w represents the vector consisting of the residuals corresponding to the n groups of support columns selected in the last iteration, It means taking the reciprocal of each element in w.

[0117] Optionally, determining a priori vector based on the estimated matrix corresponding to the n groups of support columns selected in the last iteration and the first vector includes:

[0118] Determine the initial prior vector based on the estimated matrix and the first vector corresponding to the n groups of support columns selected in the last iteration;

[0119] The initial priori vector is determined as the priori vector; or, a structural check is performed on the initial priori vector to obtain the priori vector.

[0120] Specifically, after determining the estimated matrix and the first vector corresponding to the n groups of support columns selected in the last iteration, the initial prior vector can be determined based on the estimated matrix and the first vector corresponding to the n groups of support columns selected in the last iteration. By performing a structural check on the initial prior vector, the prior vector can be determined, so that the determined prior vector is more accurate.

[0121] In one implementation, the initial prior vector may be determined by the following formula:

[0122]

[0123] in, represents the initial prior vector, M represents the estimated matrix corresponding to the n groups of support columns selected in the last iteration, K represents the first vector, abs(MK T ) indicates MK T The vector obtained by taking the absolute value of each element in .

[0124] In one implementation, the initial priori vector may be directly determined as the priori vector.

[0125] Optionally, performing a structural check on the initial prior vector to obtain the prior vector may include:

[0126] Determine a set of row and column indices of all elements in the concatenated channel sparsity matrix that satisfy the first condition;

[0127] Determine a first index set based on the row and column index sets;

[0128] Replacing the element corresponding to the first index set in the initial prior vector with the first value, while keeping other elements in the initial prior vector unchanged, to obtain a prior vector;

[0129] The element in the cascade channel sparsity matrix that satisfies the first condition means that the value of the element is less than a second value, and the second value is determined based on the structure coefficient for structure inspection, the cascade channel sparsity matrix, the network structure index set, the number of IRS receive paths, and the number of IRS transmit paths;

[0130] The first value is determined based on the cascaded channel sparsity matrix, the network structure index set, the number of IRS receive paths, the number of IRS transmit paths, and the number of steering vectors in the dictionary matrix at the IRS.

[0131] In one embodiment, a first index set can be determined based on the row and column indices of all elements in the cascaded channel sparsity matrix whose values ​​are less than the second numerical value. For each pair of row and column indices in the first index set, the corresponding elements in the initial prior vector are determined based on the row and column indices, and these elements are replaced with the first numerical value, while the other elements in the initial prior vector remain unchanged, thereby obtaining a priori vector. By replacing the elements corresponding to the first index set in the initial prior vector with the first numerical value, while the other elements in the initial prior vector remain unchanged, the prior vector is obtained, so that the final prior vector is closer to the actual IRS channel, thereby improving the accuracy of the channel estimation result.

[0132] For example, the structure of the initial prior vector can be checked by the following formula:

[0133]

[0134]

[0135] in l=1,2,…,M G β∈[0,1) is the structural coefficient that controls the degree of structural inspection. M G Represents the number of steering vectors in the dictionary matrix at Rx, Tx and IRS, respectively, H S represents the cascade channel sparsity matrix, and Represents the grid structure index set, L1 and L2 represent the number of IRS-Rx and Tx-IRS paths, respectively.

[0136] Optionally, residual coefficients are obtained based on the residuals corresponding to the received signal vector and the n groups of support columns selected in the last iteration, including:

[0137] The residual coefficient is obtained based on the ratio between the L2 norm of the received signal vector and the minimum value of the residuals corresponding to the n groups of support columns selected in the last iteration.

[0138] Specifically, after determining the residuals corresponding to the n groups of support columns selected in the last iteration, the L2 norm of the residuals corresponding to the n groups of support columns selected in the last iteration can be calculated, combined with the received signal vector, and then the residual coefficient is obtained. The residual coefficient is obtained by taking the ratio between the L2 norm of the received signal vector and the minimum value of the residuals corresponding to the n groups of support columns selected in the last iteration, so that the value of the residual coefficient is closer to the actual IRS channel, thereby improving the accuracy of the channel estimation result.

[0139] In one embodiment, the residual coefficient can be obtained by the following formula:

[0140] α=||y||2 / min(w)

[0141] Among them, α represents the residual coefficient, y represents the received signal vector, and w represents the vector composed of the residuals corresponding to the n groups of support columns selected in the last iteration.

[0142] Optionally, based on the prior vector and the residual coefficients, initial hyperparameters are determined, including:

[0143] For the element position with the value of 0 in the prior vector, determine that the element value of the corresponding position in the initial hyperparameter is 0;

[0144] For element positions with values ​​greater than 0 in the prior vector, the values ​​of the elements at the corresponding positions in the initial hyperparameters are determined based on the values ​​of the elements at the element positions with values ​​greater than 0, the residual coefficients, the smallest positive element in the prior vector, and the largest element in the prior vector.

[0145] Specifically, when the value of an element in the prior vector is 0, the value of the element at the corresponding position of the initial hyperparameter is also 0; when the value of an element in the prior vector is greater than 0, the value of the element at the corresponding position of the initial hyperparameter can be determined based on the value of the element, the residual coefficient, the smallest positive element in the prior vector, and the largest element in the prior vector. By using the prior estimated by the OMPL algorithm to determine the initial hyperparameters of the SBL algorithm, the computational complexity of the SBL algorithm is reduced.

[0146] In one implementation, the initial hyperparameters may be determined using the following formula:

[0147]

[0148] Where e=1,2,…,G, M G Represents the number of steering vectors in the dictionary matrix at Rx, Tx and IRS respectively, represents the initial hyperparameters, represents the e-th element in the initial hyperparameters, α represents the residual coefficient, represents the prior vector, Represents the e-th element in the prior vector, A represents the smallest positive element in the prior vector, and B represents the largest element in the prior vector.

[0149] The method provided by the present invention is illustrated below by using specific application scenarios.

[0150] The present invention uses a hyperparameter optimization OMPL-SBL algorithm based on a Bayesian framework to implement IRS channel estimation. The algorithm consists of two stages: an OMPL stage for generating a priori phase of channel properties considering grid sparsity and a SBL stage for estimating the channel.

[0151] Algorithm 1 gives the whole process of the OMPL stage. The maximum number of iterations k is set max Greater than the number of nonzero elements in λ (the number of nonzero elements in λ L1 and L2 are the number of IRS-Rx and Tx-IRS paths).

[0152] Algorithm 1 (OMPL stage) specifically includes the following steps:

[0153] Input: y (received signal vector), Q (perception matrix).

[0154] initialization:

[0155] 1. r = y;

[0156] 2. for k = 1, 2, ..., k max do

[0157] 3. if k=1 then

[0158] 4. Branch: Initialize Ξ with n support columns corresponding to the minimum residual ||r||2 (1) ;

[0159] 5. else

[0160] 6. Branch: Calculate residual r i,j for 1≤i≤n, 1≤j≤G-k+1;

[0161] 7. Merge: Update using formula 14

[0162] 8. Merger: Contains n smallest || r i,j ||2;

[0163] 9. end if

[0164] 10. All Θ (k) Estimated composition of

[0165] 11. The residuals corresponding to all M constitute w;

[0166] 12. if min(w)≤r th then

[0167] 13. break;

[0168] 14. end if

[0169] 15. end for

[0170] 16.

[0171] 17. To MK T As Structural inspection;

[0172] 18. α = ||y||2 / min(w);

[0173] Output: Estimated prior Residual coefficient α.

[0174] List update (steps 2-15): Unlike the traditional OMP algorithm, which greedily selects only one column support in one iteration, the OMPL algorithm selects a list of n column supports (column support is a column of the perception matrix Q, which corresponds to an element of the sparse vector λ). The list update in the OMPL stage can be performed as an iterative, repeated branching and merging to update the index of the column support (the index of the column of the perception matrix Q). Define Θ (k) is the sequence number of the n different selected support columns in the kth iteration. (k) It can be expressed by formula 12:

[0175]

[0176] Among them, the collection (i=1,2,…,n) contains k different support column numbers selected in each iteration. In the branch operation, the first iteration (step 4) calculates G residuals and selects the smallest n of them as in the traditional OMP algorithm, which can be recorded as In the kth iteration (k greater than 1) (step 6), Θ in the previous iteration (k-1) Each support column Branch into n sets. For the set Add a new support column number (1≤i≤n, 1≤j≤G-k+1) and calculate the corresponding residual ||r i,j ||2. The newly selected support column number corresponding to the minimum residual is recorded as Therefore, n from The branched set can be expressed as Formula 13:

[0177]

[0178] In the merge operation of step 7, the results of the n branches in formula 13 can be merged into a new list, expressed as formula 14:

[0179]

[0180] Because in There may be duplications in The number of elements may be greater than n 2 Small. In step 8 The n elements with the smallest residual in will form the iteratively updated list Θ (k) , in step 10, indexing is performed using column support knowledge In the kth iteration with the i-th group of column supports, the estimated matrix for each group of column supports can be calculated as i=1,2,…,n。D i Represents the matrix composed of column vectors in matrix D that have the same column index as the i-th group of support columns selected in this iteration.

[0181] Estimate (steps 16-18): The iterations of branching and merging are repeated until the maximum number of iterations k is reached. max Or in step 12, the residual threshold r th The estimation is terminated when the estimation result is accurate enough. The likelihood of each set of column supports is related to the residual in step 16. The estimated prior is combined with the n estimates in step 17 according to the likelihood. After this, the channel structure check is applied because the beam space cascade channel Has grid sparsity, that is, non-zero elements are located in certain rows and columns. Consider the merged version As H S , which also has the sparsity of the grid, can be expressed as Equation 15:

[0182]

[0183] in, Estimated grid structure indicator set and is calculated by formula 16:

[0184]

[0185]

[0186] The prior of the SBL stage is defined as And it is initialized to Check using formula 17:

[0187]

[0188] in l=1,2,…,M G β∈[0,1) is the structure coefficient that controls the degree of structure checking. A larger β makes it more likely that true non-zero elements will be marked as non-zero elements. In particular, when β=0, no structure checking is performed. The residual coefficient α is calculated in step 18 by dividing the L2 norm of the received signal by the minimum residual min(w). This coefficient of the SBL stage will be used in equation 18 to initialize its hyperparameters.

[0189] The core of the OMPL phase is to iteratively update the list of column support indices by first expanding Θ (k-1) (k>1) to Then from Select the best n sets to form Θ(k) Afterwards, the prior of the SBL phase is calculated and the channel structure is checked. It can be concluded that the traditional OMP algorithm is a special case of OMPL when n=1.

[0190] The overall OPML-SBL algorithm structure is given in Algorithm 2, which specifically includes the following steps:

[0191] Input: y (received signal vector), Q (perception matrix).

[0192] Initialization: λ b =0 G×1 .

[0193] 1. Calculate using Algorithm 1 α;

[0194] 2. Initialize hyperparameters using formula 18

[0195] 3. fork = 1, 2, ..., k max do

[0196] 4. Hyperparameters pruning;

[0197] 5. Use formula 19 to iteratively update the SBL algorithm and

[0198] 6.

[0199] 7. break;

[0200] 8. end if

[0201] 9. end for

[0202] 10. fori = 1, 2, ..., Gdo

[0203] 11.

[0204] 12.

[0205] 13. else

[0206] 14.

[0207] 15. end if

[0208] 16. end for

[0209] 17. Estimate using formula 10 in

[0210] Output: Estimated channel

[0211] Step 1 uses the estimation results in Algorithm 1, and then in step 2 the hyperparameter prior The i-th element of is determined by formula 18:

[0212]

[0213] Where e=1,2,…,G, M G Represents the number of steering vectors in the dictionary matrix at Rx, Tx and IRS respectively, express The smallest positive element in , express The largest element in , the residual coefficient α, represents the reliability of the prior in the OMPL stage, as a larger α corresponds to a smaller final residual, indicating that the support selection tends to be more accurate. Therefore, the hyperparameter prior can be set to a larger variance to achieve faster convergence. Steps 3 to 9 follow the traditional SBL algorithm. The EM method for parameter update in step 5 can be expressed as Equation 19:

[0214]

[0215] where Q r is the effective perception matrix, where The hyperparameters are updated to i=1,2,…,G. In this process, the hyperparameters Gradually sparse, noisy The pruning operation sets a threshold and discards elements smaller than the threshold to reduce the amount of computation. This is of great significance to the proposed algorithm because the prior given in the OMPL stage Compared to the non-informative priors that initialize all non-hyperparameters with non-zero values, they are already sparse. Therefore, the dimension of the hyperparameters is much smaller, which not only makes each iteration more efficient, but also reduces the total number of iterations. Normally, the termination condition of the SBL phase iteration is that the change in the hyperparameters is less than the threshold ε (step 6) or the maximum number of iterations k is reached. max Steps 10 to 16 finally give the results of using hyperparameters to For those corresponding to a value less than the threshold μ th For very small elements, we can Otherwise, The elements in are set to the corresponding average values ​​shown in step 14. In step 17, the estimated result of the cascaded channel is calculated.

[0216] The computational complexity and runtime comparison of the OMPL-SBL algorithm are shown in Table 1. The simulation is run on macOS using the Apple M1 chip. The system parameters used for runtime evaluation are shown in Table 2.

[0217] Table 1: Comparison of computational complexity

[0218]

[0219]

[0220] 1. K is the average number of iterations in BP-ADMM.

[0221] 2. q and Q are the number of iterations of the SBL-EM and SBL stages of OMPL-SBL, respectively, and q<<Q.

[0222] 3. is the effective size of the SBL stage.

[0223] Table 1 clearly shows that the complexity of the above algorithms can be ranked as OMP<OMPL-SBL<BP-ADMM<<SBL-EM.

[0224] The priors calculated by OMPL play a key role in channel estimation. The informative priors should be sufficiently accurate to enable the SBL algorithm to work well. It's important to note that the informative priors generated by OMPL don't necessarily distinguish all zero-valued entries. Instead, they set most zero-valued entries to zero, carefully avoiding any non-zero elements. If a value is set to zero, it will be removed during the SBL phase. Setting most zero-valued elements to zero significantly speeds up SBL convergence.

[0225] The performance of the OMPL algorithm itself is comparable to that of the OMP algorithm, as the introduced list ensures that the residual is smaller than that of the OMP algorithm, alleviating the problem of OMP's greedy support selection. Furthermore, even at low signal-to-noise ratios, OMPL achieves relatively satisfactory performance in support selection, making OMPL-SBL stable at low signal-to-noise ratios.

[0226] A key feature of the proposed algorithm is that it is applicable to almost all SBL algorithms. An important flexibility of the OMPL-SBL method is that its parameters can be adjusted to meet different requirements. The OMPL list size n and the channel structure check coefficient β involve a trade-off between estimation accuracy and computational complexity.

[0227] Obviously, a larger OMPL list size n and channel structure check coefficient β will lead to more non-zero elements in the SBL prior, that is, a larger g. Therefore, the complexity increases with the increase of n and β. Here we mainly consider the estimation performance with respect to n and β. The prerequisite for accurate estimation is that all supports are selected without omission in the OMPL stage. OMPL provides a non-greedy estimation with support selection, and a larger n will result in better support selection. Originally, b supports were missing. After the channel structure check, the error P for missing supports is e The probability upper bound can be approximated as formula 20:

[0228]

[0229] It is worth noting that when the value b is fixed, P e Increases with n, but the larger n is, the smaller E[b] can be. In general, larger β and larger but not too large n help better support selection.

[0230] From the perspective of the SBL phase, the prior provided by OMPL should meet certain requirements. Intuitively, a larger n makes OMPL less greedy, while n = 1 reverts to the traditional OMP, which is greedy and therefore less reliable for support selection. When the prior g for the number of nonzero elements is large enough and the corresponding effective sensing matrix satisfies the Uniform Random Projection ensemble (URP), the upper bound on the number of local minima of SBL in an ideal noise-free scenario is given by Equation 21:

[0231]

[0232] It increases with the increase of g. Therefore, the upper bound of the number of local minima of SBL of this method is smaller than that of the traditional SBL algorithm. The smaller the number of local minima, the smaller the convergence error. Therefore, if no support is missed in the OMPL stage, the performance can be improved. However, in order to ensure a reliable support and Q r For the URP, g should not be too small, thus n and β need to be large enough.

[0233] Here, we use C++ with the Armadillo linear algebra library to simulate performance. The system parameters used in the simulation are shown in Table 2. In OMPL-SBL, we choose the appropriate parameters as n = 16 and β = 0.5.

[0234] Table 2: System parameters in simulation

[0235]

[0236] The OMP algorithm is a low-complexity channel estimation scheme widely used in IRS-assisted millimeter-wave systems. Since the second stage of the OMPL-SBL algorithm is SBL, it is also compared with the SBL algorithm with EM update method. SBL is proved to be equivalent to the iterative reweighted L1 minimization algorithm. Therefore, a commonly used L1 minimization-based algorithm Basis Pursuit (BP) using the Alternating Direction Method of Multipliers (ADMM) solver is also compared. The channel estimation performance is measured by the normalized mean squared error (NMSE), which is expressed as

[0237] Figure 2 The following is a schematic diagram of the simulation results of the IRS channel estimation method provided by the present invention, as shown in the figure. Figure 2 (a) Comparison of the NMSE vs. SNR relationship between the OMPL-SBL algorithm and traditional SBL, BP-ADMM, and OMP methods. The OMPL algorithm, with an OMPL list size of n = 16, achieves a performance gain of approximately 10 dB over the traditional SBL algorithm at high SNRs (greater than 10 dB), approaching the Cramér-Rao bound for Oracle LS. OMP, BP-ADMM, and SBL-EM, on the other hand, fail to achieve high estimation accuracy.

[0238] NMSE performance and pilot overhead The relationship as Figure 2 Even with a small number of pilots, the NMSE performance of the OMPL-SBL algorithm proposed in this paper is also outstanding.

[0239] Figure 2 In (c), the bit error ratio (BER) performance of the MMSE MIMO detector for 16-QAM modulated signals is also evaluated. The results show that the proposed OMPL-SBL algorithm can achieve a low BER with perfect channel information (CSI) close to that of Oracle LS.

[0240] Simulation results also show that the proposed method converges much faster in the SBL phase, with a total of approximately 8 to 16 iterations. In comparison, the traditional SBL algorithm takes longer to converge due to the larger number of total iterations and longer computation time per iteration.

[0241] The main features of the OMPL-SBL algorithm provided by the present invention are summarized as follows:

[0242] 1. The extended greedy algorithm OMP considers the grid sparsity structure unique to the IRS auxiliary channel and provides an effective prior for SBL.

[0243] 2. This algorithm has higher accuracy than the L1 minimization-based algorithm and the Bayesian algorithm, and its computational complexity is significantly reduced (by more than 95%) compared to the traditional SBL algorithm.

[0244] 3. The parameters in OMPL-SBL can be adjusted according to different system requirements, making a trade-off between estimation accuracy and computational complexity.

[0245] The IRS channel estimation apparatus provided by the present invention is described below. The IRS channel estimation apparatus described below and the IRS channel estimation method described above can refer to each other.

[0246] Figure 3 A schematic diagram of the structure of the IRS channel estimation device provided by the present invention is shown as follows: Figure 3 As shown, the device includes:

[0247] An acquisition module 300 is configured to acquire a received signal vector and a perception matrix;

[0248] A first determination module 301 is configured to use a list-based orthogonal matching pursuit (OMP) algorithm to select n groups of support columns from the perception matrix in each iteration until the iteration terminates, and to determine the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration; n is an integer greater than or equal to 1;

[0249] A second determination module 302 is configured to obtain a priori vectors based on the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration; and obtain residual coefficients based on the received signal vector and the residuals corresponding to the n groups of support columns selected in the last iteration.

[0250] The third determination module 303 is configured to determine initial hyperparameters based on the prior vector and the residual coefficients, and perform IRS channel estimation based on the initial hyperparameters and a sparse Bayesian learning (SBL) algorithm.

[0251] Optionally, n groups of support columns are selected from the perception matrix in each iteration, including:

[0252] In the first iteration, n support columns with the smallest corresponding residuals are selected from the perception matrix;

[0253] In any iteration after the first iteration, determine the n groups of support columns selected in the iteration before any iteration, and for any group of support columns, determine the n support columns with the smallest corresponding residuals from the perception matrix, and branch any group of support columns into n groups of support columns based on the determined n support columns with the smallest corresponding residuals to obtain n*n groups of support columns, and then select n groups of support columns with the smallest corresponding residuals from the n*n groups of support columns.

[0254] Optionally, n support columns corresponding to the smallest residuals are determined from the perception matrix, and any one group of support columns is branched into n groups of support columns based on the determined n support columns corresponding to the smallest residuals, including:

[0255] According to the residuals of each support column except any one group of support columns in the perception matrix corresponding to any one group of support columns, n support columns with the smallest corresponding residuals are determined from each of the other support columns;

[0256] Each support column in the n support columns is respectively combined with any group of support columns to obtain n groups of support columns branched out from any group of support columns.

[0257] Optionally, based on the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration, a priori vector is obtained, including:

[0258] Determine the prior vector based on the estimated matrix and the first vector corresponding to the n groups of support columns selected in the last iteration;

[0259] The first vector is determined based on the residuals corresponding to the n groups of support columns selected in the last iteration.

[0260] Optionally, determining a priori vector based on the estimated matrix corresponding to the n groups of support columns selected in the last iteration and the first vector includes:

[0261] Determine the initial prior vector based on the estimated matrix and the first vector corresponding to the n groups of support columns selected in the last iteration;

[0262] The initial priori vector is determined as the priori vector; or, a structural check is performed on the initial priori vector to obtain the priori vector.

[0263] Optionally, a structural check is performed on the initial prior vector to obtain a prior vector, including:

[0264] Determine a set of row and column indices of all elements in the concatenated channel sparsity matrix that satisfy the first condition;

[0265] Determine a first index set based on the row and column index sets;

[0266] Replacing the element corresponding to the first index set in the initial prior vector with the first value, while keeping other elements in the initial prior vector unchanged, to obtain a prior vector;

[0267] The element in the cascade channel sparsity matrix that satisfies the first condition means that the value of the element is less than a second value, and the second value is determined based on the structure coefficient for structure inspection, the cascade channel sparsity matrix, the network structure index set, the number of IRS receive paths, and the number of IRS transmit paths;

[0268] The first value is determined based on the cascade channel sparsity matrix, the network structure index set, the number of IRS receive paths, the number of IRS transmit paths, and the number of steering vectors in the dictionary matrix at the IRS. Optionally, the residual coefficients are obtained based on the residuals corresponding to the received signal vector and the n groups of support columns selected in the last iteration, including:

[0269] The residual coefficient is obtained based on the ratio between the L2 norm of the received signal vector and the minimum value of the residuals corresponding to the n groups of support columns selected in the last iteration.

[0270] Optionally, based on the prior vector and the residual coefficients, initial hyperparameters are determined, including:

[0271] For the element position with the value of 0 in the prior vector, determine that the element value of the corresponding position in the initial hyperparameter is 0;

[0272] For element positions with values ​​greater than 0 in the prior vector, the values ​​of the elements at the corresponding positions in the initial hyperparameters are determined based on the values ​​of the elements at the positions with values ​​greater than 0, the residual coefficients, the smallest positive element in the prior vector array, and the largest element in the prior vector.

[0273] It should be noted here that the above-mentioned device provided by the present invention can implement all the method steps implemented in the above-mentioned method embodiment and can achieve the same technical effects. The parts and beneficial effects that are the same as those in the method embodiment will not be described in detail here.

[0274] Figure 4 A schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 4 As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other via the communication bus 440. The processor 410 may call the logic instructions in the memory 430 to execute any of the IRS channel estimation methods provided in the above embodiments.

[0275] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0276] It should be noted here that the electronic device provided by the present invention can implement all the method steps implemented in the above method embodiment and can achieve the same technical effects. The parts and beneficial effects that are the same as those in the method embodiment will not be described in detail here.

[0277] On the other hand, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when executed by a processor to perform any of the IRS channel estimation methods provided in the above embodiments.

[0278] It should be noted here that the non-transitory computer-readable storage medium provided by the present invention can implement all the method steps implemented in the above-mentioned method embodiment and can achieve the same technical effects. The parts and beneficial effects that are the same as those in the method embodiment will not be described in detail here.

[0279] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0280] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0281] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for intelligent reflecting surface IRS channel estimation, characterized in that: include: Obtain received signal vector and perception matrix; Using a list-based orthogonal matching pursuit (OMP) algorithm, n groups of support columns are selected from the perception matrix in each iteration until the iteration is terminated, and the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration are determined; n is an integer greater than or equal to 1; Obtaining a priori vectors based on the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration; obtaining residual coefficients based on the received signal vector and the residuals corresponding to the n groups of support columns selected in the last iteration; Based on the prior vector and the residual coefficient, initial hyperparameters are determined, and IRS channel estimation is performed based on the initial hyperparameters and a sparse Bayesian learning (SBL) algorithm.

2. The IRS channel estimation method according to claim 1, wherein In each iteration, n groups of support columns are selected from the perception matrix, including: In the first iteration, n support columns corresponding to the smallest residuals are selected from the perception matrix; In any iteration after the first iteration, determine the n groups of support columns selected in the iteration before the any iteration, and for any group of support columns, determine the n support columns with the smallest corresponding residuals from the perception matrix, and branch the any group of support columns into n groups of support columns based on the determined n support columns with the smallest corresponding residuals to obtain n*n groups of support columns, and then select n groups of support columns with the smallest corresponding residuals from the n*n groups of support columns.

3. The IRS channel estimation method according to claim 2, wherein: The determining n support columns corresponding to the smallest residuals from the perception matrix, and branching the arbitrary group of support columns into n groups of support columns based on the determined n support columns corresponding to the smallest residuals, comprises: According to the residual of each support column other than the arbitrary group of support columns in the perception matrix corresponding to the arbitrary group of support columns, determining n support columns with the smallest corresponding residual from each of the other support columns; Each support column in the n support columns is respectively combined with the arbitrary group of support columns to obtain n groups of support columns branched from the arbitrary group of support columns.

4. The IRS channel estimation method according to claim 1, wherein: The prior vector is obtained based on the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration, including: Determine a priori vector based on the estimated matrix corresponding to the n groups of support columns selected in the last iteration and the first vector; The first vector is determined based on the residuals corresponding to the n groups of support columns selected in the last iteration.

5. The IRS channel estimation method according to claim 4, wherein: The determining of the prior vector based on the estimated matrix and the first vector corresponding to the n groups of support columns selected in the last iteration includes: Determine an initial prior vector based on the estimated matrix and the first vector corresponding to the n groups of support columns selected in the last iteration; The initial a priori vector is determined as the a priori vector; or, a structural check is performed on the initial a priori vector to obtain the a priori vector.

6. The IRS channel estimation method according to claim 5, wherein: The performing a structural check on the initial a priori vector to obtain the a priori vector includes: Determine a set of row and column indices of all elements in the concatenated channel sparsity matrix that satisfy the first condition; Determine a first index set based on the row and column index set; Replacing the element corresponding to the first index set in the initial a priori vector with a first value, while keeping other elements in the initial a priori vector unchanged, to obtain the a priori vector; The element in the cascaded channel sparsity matrix that satisfies the first condition means that the value of the element is less than a second value, and the second value is determined based on the structure coefficient for structure inspection, the cascaded channel sparsity matrix, the network structure index set, the number of IRS receive paths, and the number of IRS transmit paths; The first value is determined based on the cascade channel sparsity matrix, the network structure index set, the number of IRS receive paths, the number of IRS transmit paths, and the number of steering vectors in the dictionary matrix at the IRS.

7. The IRS channel estimation method according to claim 1, wherein: The obtaining of residual coefficients based on the received signal vector and the residuals corresponding to the n groups of support columns selected in the last iteration includes: A residual coefficient is obtained based on a ratio between an L2 norm of the received signal vector and a minimum value of the residuals respectively corresponding to the n groups of support columns selected in the last iteration.

8. The IRS channel estimation method according to claim 1, wherein: The determining of initial hyperparameters based on the prior vector and the residual coefficients includes: For the element position with a value of 0 in the prior vector, determine that the element value at the corresponding position in the initial hyperparameter is 0; For the element position in the prior vector whose value is greater than 0, the element value at the corresponding position in the initial hyperparameter is determined based on the value of the element at the element position with a value greater than 0, the residual coefficient, the smallest positive element in the prior vector, and the largest element in the prior vector.

9. An IRS channel estimation device, characterized in that: include: An acquisition module, used to obtain a received signal vector and a perception matrix; A first determination module is configured to use a list-based orthogonal matching pursuit (OMP) algorithm to select n groups of support columns from the perception matrix in each iteration until the iteration is terminated, and to determine the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration; n is an integer greater than or equal to 1; A second determination module is configured to obtain a priori vector based on the n groups of support columns selected in the last iteration and the residuals corresponding to the n groups of support columns selected in the last iteration; and obtain residual coefficients based on the received signal vector and the residuals corresponding to the n groups of support columns selected in the last iteration; The third determination module is configured to determine initial hyperparameters based on the priori vector and the residual coefficients, and perform IRS channel estimation based on the initial hyperparameters and a sparse Bayesian learning (SBL) algorithm.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the IRS channel estimation method according to any one of claims 1 to 8 is implemented.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the IRS channel estimation method according to any one of claims 1 to 8 is implemented.

Images