A sensing-assisted semi-blind estimation method for massive MIMO time-varying channels

Abstract

The present invention relates to a sensing-assisted semi-blind estimation method for massive MIMO time-varying channels, comprising: a communication module of a synaesthesia dual-function MIMO base station receives uplink communication data from multiple users and utilizes the low-rank characteristics of the massive MIMO matrix to implement channel initialization based on the completeness of the low-rank matrix; a sensing module of the synaesthesia dual-function MIMO base station receives reflected echo signals from multiple targets and utilizes a root-finding MUSIC algorithm to effectively estimate the channel vector parameters AoA / AoD, thereby assisting in correcting the initial channel value; and the synaesthesia dual-function MIMO base station sequentially performs channel estimation and target sensing in a time-division multiplexing mode, and utilizes a semi-blind detection method to jointly estimate subsequent channel and payload data in combination with the channel correction value obtained in the previous cycle. Compared with the existing technology, the present invention can effectively reduce pilot overhead, reduce solution computational complexity, and improve channel estimation accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of communication-perception integrated signal processing, and in particular to a perception-assisted large-scale MIMO time-varying channel semi-blind estimation method. Background Art

[0002] Communication and perception integration refers to new information processing and service technologies that achieve synergy between communication and perception functions through the sharing of software and hardware resources or information. This technology, which combines communication and perception capabilities using the same equipment and spectrum, effectively improves system spectrum efficiency and hardware resource utilization, possessing enormous application value and practical significance. On the one hand, the entire network acts as a massive sensor, transmitting and receiving wireless signals and extracting information such as target distance, angle, and speed to provide target detection, identification, and tracking services. On the other hand, the high-precision positioning information obtained through perception can assist in correcting channel state estimates, thereby improving communication transmission and detection performance. Furthermore, to extract more path information and expand the spatial perception range, the use of large-scale multiple-input-multiple-output (MIMO) arrays to form multifunctional communication and perception beams is beneficial for achieving more accurate perception-assisted communication functions.

[0003] For large-scale MIMO integrated telemetry systems, obtaining accurate channel state information is a prerequisite for guiding subsequent beamforming optimization and efficient demodulation of data. In particular, for dynamically changing transmission environments, combining perception functions to assist in predicting and correcting time-varying channel estimates can avoid repeated channel estimation processes and improve communication reliability. However, with the continuous increase in antenna dimensions and carrier frequencies, the pilot overhead used for channel estimation will also increase accordingly, and the additional occupation of bandwidth resources by pilot information will limit spectrum efficiency. In addition, due to the low-rank nature of large-scale MIMO systems, traditional methods for solving low-rank matrix completion are mainly based on alternating minimization iterations, and a relatively high-complexity least squares calculation must be performed in each iteration, resulting in a very complex solution process. In order to speed up the solution, existing technologies often use compressed sensing methods, but this method has the problem of grid mismatch, resulting in poor estimation accuracy.

[0004] Therefore, how to utilize the inherent low rank of large-scale MIMO systems and the geometric sparsity of high-frequency channels, fully explore the structural characteristics of synaesthesia integrated signals, and use perception functions and a small number of pilots to achieve low-complexity dynamic tracking of random time-varying channels is the key to ensuring the coexistence of high-precision perception and high-speed communication, and is also a problem that needs to be solved urgently. Summary of the Invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a perception-assisted large-scale MIMO time-varying channel semi-blind estimation method, which can effectively reduce pilot overhead, reduce the solution calculation complexity, and improve the estimation accuracy.

[0006] The object of the present invention can be achieved by the following technical solution: a perception-assisted large-scale MIMO time-varying channel semi-blind estimation method, comprising the following steps:

[0007] S1. The communication module of the synaesthesia dual-function MIMO base station receives uplink communication data from multiple users and uses the low-rank characteristics of the massive MIMO matrix to achieve channel initialization based on the complete low-rank matrix.

[0008] S2. The perception module of the MIMO base station receives reflected echo signals from multiple targets and uses the root-finding MUSIC algorithm to effectively estimate the channel vector parameters AoA (Angle-of-Arrival) and AoD (Angle-of-Departure), thereby assisting in correcting the initial channel value.

[0009] S3. The synaesthesia dual-function MIMO base station performs channel estimation and target perception in time division multiplexing mode in sequence, combines the channel correction value obtained in the previous cycle, and uses the semi-blind detection method to realize the joint estimation of subsequent channel and load data.

[0010] Furthermore, the step S1 specifically includes the following steps:

[0011] S11, a communication module of the synaesthesia dual-function MIMO base station receives uplink communication data from multiple users;

[0012] S12. Based on the low-rank property of massive MIMO matrix, the communication module constructs the low-rank matrix completion problem;

[0013] S13. Use the Alternating Steepest Descent (ASD) algorithm to solve the low-rank matrix completion problem and obtain the channel initial value, thus completing the channel initialization.

[0014] Furthermore, the low-rank matrix completion problem is specifically:

[0015]

[0016] Where U and V represent the channel and load data estimates, respectively. The elements at index set Ω are represented by Given, the rest are set to 0, represents a linear operator that preserves the non-zero elements in Ω, represents the Frobenius norm of the matrix, N a is the number of base station antennas, N t is the number of transmitting antennas, K is the number of users, T c is the channel coherence time.

[0017] Furthermore, the specific process of step S13 is as follows:

[0018] make and Represents functions respectively Relative to the gradients of U and V, the step size updated along the steepest gradient direction is:

[0019]

[0020] in

[0021]

[0022]

[0023] Based on this, for the t+1th iteration, U and V are updated alternately as follows until convergence:

[0024]

[0025]

[0026] Assume that the pilot and data transmission follow:

[0027] X≡[X p ,X d ],in And T p ∝KN t , then V=ΓX=Γ[X p ,X d ], since the pilot sequence usually satisfies the orthogonality, that is, The corresponding invertible matrix is:

[0028]

[0029] Among them, V(:,1~T p ) represents the 1~T of the matrix V p After solving the analytical value of the invertible matrix, the initial estimation of the channel can be calculated:

[0030] Furthermore, the specific process of step S2 is as follows:

[0031] When the dual-function MIMO base station performs the sensing function, it first sends an orthogonal sounding signal satisfy Then the reflected echo signals from K users are:

[0032]

[0033] Among them, β k,l and τ k,l represent the reflection coefficient and delay-Doppler effect of user k corresponding to the lth path, f s is the carrier frequency of the sensing signal, a t (θ k,l ) and a r (θ k,l ) are the base station transmitting and receiving antenna steering vectors respectively, represents the receive beamforming matrix of the base station;

[0034] Affected by the large-scale characteristics of the scattering environment, the signal angle changes relatively slowly. The AoA of the uplink signal in each transmission cycle is considered to be equal to the reception angle of the perceived signal, that is,

[0035] Afterwards, order:

[0036]

[0037]

[0038]

[0039] The covariance matrix of the perceived echo signal is calculated as follows:

[0040]

[0041] in, and According to the traditional MUSIC algorithm principle, the signal and noise subspaces based on eigenvalue decomposition are as follows:

[0042]

[0043] in represents a diagonal matrix with eigenvalues ​​in descending order, and The eigenvector matrices of the signal subspace and the noise subspace are respectively formed. Given the orthogonal relationship between the signal and noise subspaces, the MUSIC spectrum used to locate the K user azimuths is expressed as:

[0044]

[0045] Rewrite the denominator of the above formula as:

[0046]

[0047] Where ξ=2πd / λ,C l Represents the correlation matrix along The sum of the elements of the lth diagonal, if defined Then we simplify to get the polynomial:

[0048]

[0049] f(z)2(N r -1) roots correspond exactly to the extremes of the MUSIC spectrum:

[0050]

[0051] Select the KL roots closest to the unit circle to generate the estimated value of the target user's AoA / AoD;

[0052] After the AoA / AoD estimation value of the given channel is obtained, the finite scattering characteristics of the high-frequency channel are utilized to construct a sparse reconstruction problem and solve it to obtain the auxiliary correction matrix, which is then used to correct the initial channel value.

[0053] Furthermore, the estimated value of the target user's AoA / AoD is specifically:

[0054]

[0055] Furthermore, the corrected initial channel value in step S2 is specifically:

[0056]

[0057] in, is an auxiliary correction matrix, the elements of which include the scattering path gain α k,l and delay spread exponent ν k,l .

[0058] Furthermore, the semi-blind detection method in step S3 specifically includes the following steps:

[0059] S31. Considering the random and independent changes of channels in different transmission periods, the RF domain receiving signal model is expanded;

[0060] S32, based on the channel correction value of the previous cycle, combined with the channel change constraint H of the adjacent cycle t =H t-1 +ΔH t , obtain the preliminary estimated value of the subsequent cycle load data;

[0061] S33, using atomic norm to characterize the sparsity of channel AoA / AoD domain, by constructing a combination of atomic norm and The norm minimization problem is solved by the conjugate gradient method to obtain the subsequent period channel estimation value and the decoding error estimation value, and the improved estimation value of the load data is obtained based on the preliminary estimation value of the load data.

[0062] Furthermore, the combined atomic norm and The norm minimization problem is specifically:

[0063]

[0064]

[0065]

[0066]

[0067] in, and are the channel estimation value and decoding error estimation value respectively, Y t For a given observation signal, that is, the output signal after the RF domain received signal model is expanded, For a preliminary estimate of load data, is the atomic norm, represents the channel matrix between user k and the base station.

[0068] Furthermore, the load data improved estimated value is specifically:

[0069]

[0070] in, Improved estimates for load data.

[0071] Compared with the prior art, the present invention has the following advantages:

[0072] 1. Low pilot overhead: Traditional channel estimation schemes only use pilot information for channel estimation. As the antenna dimension and the number of terminal connections increase significantly, the pilot overhead used for channel estimation will also increase, which to some extent limits the further improvement of spectrum efficiency. This invention utilizes the inherent low rank property of massive MIMO to model the joint estimation of channel and payload data as a low-rank matrix completion problem. This reduces the pilot sequence length required by traditional channel estimation methods, which is proportional to the antenna dimension, to a length proportional to the number of terminals. This can significantly reduce the high pilot overhead required by traditional channel estimation methods.

[0073] 2. Low computational complexity: Traditional methods for solving low-rank matrix completion problems are mainly based on alternating minimization iterations, and require performing relatively complex least-squares calculations during each iteration. The present invention adopts the low-complexity ASD algorithm to solve the low-rank matrix completion problem, which can effectively avoid complex matrix inversion operations. The algorithm also uses a simple line search method to accurately update channel and data estimates, thereby reducing the computational complexity of factorization.

[0074] 3. High Estimation Accuracy: This invention fully exploits the structural correlation between communication and perception signals and utilizes the rapid detection characteristics of the perception function to effectively estimate the channel vector parameters AoA / AoD. In particular, it uses the root-finding MUSIC algorithm to convert the spectral peak quantization search of the traditional MUSIC algorithm into a polynomial solution, thereby helping to improve the accuracy of channel estimation. In addition, during the dynamic tracking of time-varying channels, this invention uses the atomic norm to describe the geometric sparsity of high-frequency channels and adopts the conjugate gradient descent method for rapid solution, avoiding the grid mismatch problem existing in traditional compressed sensing methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 Schematic diagram of the method flow of the present invention;

[0076] Figure 2 Schematic diagram of the application framework of the embodiment;

[0077] Figure 3 Schematic diagram of the hybrid transmission frame structure in the present invention;

[0078] Figure 4 Schematic diagram of the application process of the embodiment. DETAILED DESCRIPTION

[0079] The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0080] Example

[0081] like Figure 1 As shown, a sensing-assisted large-scale MIMO time-varying channel semi-blind estimation method includes the following steps:

[0082] S1. The communication module of the synaesthesia dual-function MIMO base station receives uplink communication data from multiple users and uses the low-rank characteristics of the massive MIMO matrix to achieve channel initialization based on the complete low-rank matrix.

[0083] S2, the perception module of the dual-function MIMO base station receives the reflected echo signals of multiple targets and uses the root-finding MUSIC algorithm to effectively estimate the channel vector parameters AoA / AoD, thereby assisting in correcting the initial channel value;

[0084] S3. The synaesthesia dual-function MIMO base station performs channel estimation and target perception in time division multiplexing mode in sequence, combines the channel correction value obtained in the previous cycle, and uses the semi-blind detection method to realize the joint estimation of subsequent channel and load data.

[0085] This embodiment applies the above technical solution to design a new hybrid transmission frame structure suitable for large-scale MIMO interawareness integrated system. Its main functional framework and corresponding hybrid transmission frame are as follows: Figure 2 and Figure 3 shown.

[0086] During the initial training cycle M0, the communication module and the perception module of the dual-function MIMO base station will receive uplink communication data from multiple users and perception echo signals from multiple targets, respectively. The communication module constructs a low-rank matrix completion problem based on the low-rank nature of the massive MIMO matrix and uses the low-complexity Alternating Steepest Descent (ASD) algorithm to solve the initial channel value. The perception module uses the root-finite MUSIC algorithm to estimate the channel's finite scattering parameters (AoA / AoD), thereby assisting in correcting the initial channel value and improving channel estimation accuracy.

[0087] During subsequent transmission cycles M1, M2, ..., the dual-function MIMO base station sequentially performs channel estimation and target sensing in a time-division multiplexing mode. By combining the sensing module's ability to dynamically track and promptly correct some channel parameters, and taking into account the sparsity of high-frequency channels and decoding errors, the communication module uses a semi-blind detection principle to achieve joint estimation of channel and payload data without the need for additional pilot information.

[0088] The specific application process is as follows Figure 4 As shown, the dual-function base station performs the entire joint estimation of the uplink channel and payload data, thereby reducing the actual power consumption on the user / terminal side. Furthermore, the rapid detection characteristics of the sensing function help improve channel estimation accuracy. Furthermore, by leveraging the inherent low-rank nature of massive MIMO systems and combining the low-complexity ASD algorithm with semi-blind detection principles, the pilot overhead and computational complexity used for channel estimation are significantly reduced.

[0089] The main process includes:

[0090] A. Channel initialization based on low-rank matrix completion

[0091] Assume that the base station configuration N a antennas and N r If N radio frequency (RF) links are used to receive uplink signals, t The RF domain received signal obtained by combining the K user transmission signals of the root antenna is:

[0092]

[0093] in, and denote the transmit beamforming matrix of user k and the receive beamforming matrix of the base station, and They represent the uplink transmission data stream of user k and the channel matrix between user k and the base station. In estimating the channel and load data, the present invention considers modeling the above signal model as a low-rank matrix completion problem. Specifically, from N a ×N a Randomly select N from the discrete Fourier matrix r The line generates W[t] to represent the analog phase shifter. Since in a massive MIMO system, the number of base station antennas is N a and channel coherence time T c Much larger than the number of transmitting antennas N t And the number of users K, that is, the channel and data are multiplied into a low-rank matrix: rank[HX]<<KN t <<min[N a ,T c ],in, If U and V represent the channel and load data estimates, respectively, the optimal joint estimate of the two can be given by:

[0094]

[0095] in, The elements at index set Ω are represented by Given, the rest are set to 0; represents a linear operator that preserves the non-zero elements in Ω, Represents the Frobenius norm of the matrix.

[0096] To solve the above low-rank matrix completion problem, the present invention adopts the low computational complexity Alternating Steepest Descent (ASD) algorithm, which combines an accurate line search method to alternately update the channel and data factors. Specifically, let and Represents functions respectively Relative to the gradients of U and V, the step size updated along the steepest gradient direction can be analytically expressed as:

[0097]

[0098] in

[0099]

[0100] Based on this, for the t+1th iteration, U and V can be updated alternately as follows until convergence:

[0101]

[0102] Although the formation of gradients and step sizes in the above alternating update process involves the product of residual terms and high-dimensional matrices, such operations only need to be calculated once at the beginning of each iteration to be updated efficiently. In addition, since the above method replaces the traditional least squares solution with a single-step line search, the computational cost of each iteration is much less than that of the traditional alternating minimization algorithm. It is worth noting that the UV obtained by the above solution is not unique. This is because there is an invertible matrix Satisfying H=UΓ and X=Γ -1 V, so a certain amount of pilot information is needed to eliminate the estimated ambiguity. Based on this, let the pilot and data transmission follow X≡[X p ,X d ],in And T p ∝KN t , then V=ΓX=Γ[X p ,X d Since the pilot sequence usually satisfies the orthogonality, that is, The corresponding invertible matrix can be derived as:

[0103]

[0104] Where V(:,1~T p ) represents the 1~T of the matrix V p After solving the analytical value of the invertible matrix, the initial estimate of the channel can be calculated:

[0105] B. Perception-Assisted Finite Scattering Parameter Estimation

[0106] For high-frequency transmission environments, the channel model has certain geometric characteristics, and the channel model constructed above cannot reflect environmental information. Based on this, the present invention uses a sensing module to send and receive echo signals, extracting limited scattering path information of the channel environment to assist in revising the above-mentioned initial channel estimate.

[0107] Assume that there are L scattering paths between each user and the base station, then the channel model of user k can be expressed as:

[0108]

[0109] Among them, α k,l and ν k,ldenote the complex gain and delay spread of the lth scattering path, respectively. and They represent the angle of arrival (AoA)φ k,l and Angle of Departure (AoD) The antenna steering vector satisfies a(φ)=[1,e j2πd / λsin(θ) ,...,e j2πd(N-1) / λsin(θ) ] T When the dual-function MIMO base station performs the sensing function, it first sends an orthogonal detection signal satisfy Then the reflected echo signals from K users can be expressed as:

[0110]

[0111] Among them, β k,l and τ k,l represent the reflection coefficient and delay-Doppler effect of user k corresponding to the lth path, f s is the carrier frequency of the sensing signal, a t (θ k,l ) and a r (θ k,l ) are the steering vectors of the base station’s transmitting and receiving antennas, respectively. Affected by the large-scale characteristics of the scattering environment, the signal angle changes relatively slowly. In each transmission cycle, the AoA of the uplink signal can be considered equal to the reception angle of the perceived signal, i.e.

[0112] Therefore, by utilizing the fast detection characteristics of the perception module, effective estimation of channel AoA / AoD can be achieved. Specifically, let:

[0113] The covariance matrix of the perceived echo signal is calculated as follows:

[0114]

[0115] in, and According to the traditional MUSIC algorithm principle, the signal and noise subspaces based on eigenvalue decomposition are as follows:

[0116]

[0117] in represents a diagonal matrix with eigenvalues ​​in descending order, and They constitute the eigenvector matrices of the signal subspace and the noise subspace respectively. Given the orthogonal relationship between the signal and noise subspaces, the MUSIC spectrum used to locate the K user azimuths can be expressed as:

[0118]

[0119] Different from the traditional MUSIC algorithm that searches for peak angles through tedious spectrum searches, the present invention adopts a low-complexity and high-resolution root-finding MUSIC algorithm. Specifically, the denominator of formula (11) is rewritten as:

[0120]

[0121] Where ξ=2πd / λ,C l Represents the correlation matrix along The sum of the elements on the lth diagonal. If we define Then formula (12) can be further simplified to the following polynomial:

[0122]

[0123] Then f(z)2(N r -1) roots correspond exactly to the extreme points of the MUSIC spectrum:

[0124]

[0125] Selecting the KL roots closest to the unit circle produces the estimated value of the target user's AoA / AoD, that is,

[0126]

[0127] Finally, given the estimated AoA / AoD value of the channel, the finite scattering characteristics of the high-frequency channel are used to calculate the scattering path gain α k,l and delay spread exponent ν k,l It can be obtained by solving the sparse reconstruction problem, that is, the auxiliary correction matrix In summary, the present invention utilizes the fast detection characteristics of the dual-function base station sensing module and estimates the finite scattering parameters AoA / AoD through the root-finding MUSIC algorithm. The initial channel value of the previous part is corrected as follows:

[0128]

[0129] C. Dynamic tracking of time-varying channels based on semi-blind detection

[0130] Due to the high carrier frequency and random mobility environment, the duration of multi-user uplink signals is usually greater than the coherence time of the channel, resulting in time-domain fluctuations in channel conditions. In order to obtain channel state information in a timely manner, the present invention designs a time-varying channel estimation based on semi-blind detection, which uses a small number of pilot signals to achieve low-complexity dynamic tracking of channel characteristics. Specifically, considering the random and independent changes of channels in different transmission periods, the signal model of formula (1) is expanded to:

[0131]

[0132] Among them, t = 0 corresponds to the initial transmission period M0, at which time the transmission data X 0 Contains both payload data and a small amount of known pilots for channel initialization estimation; t=1,...,T corresponds to subsequent transmission periods M1,...,M T , at this time transmit data X t Only load data is included. Assume that the channel estimation value after correction with the assistance of the sensing function is And the adjacent period channel changes satisfy H t =H t-1 +ΔH t , using the channel estimation value of the previous cycle in sequence, a rough estimate of the load data of the subsequent cycles can be preliminarily given as follows:

[0133]

[0134] The decoding error caused by the channel time variation is defined as Since the demodulation error rate at the receiving end is usually controlled at 10 -6 Magnitude, it can be considered that E t In addition, the present invention adopts the atomic norm shown below to characterize the sparsity of the channel AoA / AoD domain to avoid the basis mismatch problem of the traditional gridding method, namely:

[0135]

[0136] Based on the above, if the observation signal Y is given t and a rough estimate of load data Then about H t and E t The estimate of can be constructed as a combination of atomic norms and Norm minimization problem:

[0137]

[0138] Finally, a rough estimate of the load data The following improvements can be made:

[0139]

[0140] In fact, due to the finite scattering characteristics of the channel, the optimization problem corresponding to Equation (20) can be further converted to an unconstrained optimization problem and directly solved using the low-complexity conjugate gradient descent method. In this way, based on the semi-blind detection principle and the atomic norm constraint, by alternately estimating the channel and payload data in each transmission cycle, only a small number of pilot signals in the initial cycle can be used to achieve low-complexity dynamic tracking of the time-varying channel in subsequent cycles.

[0141] In this embodiment, Figure 3 Taking the initial training cycle M0 as an example, it is assumed that K single-antenna users are randomly distributed and a simplified Saleh-Valenzuela multipath scattering channel model is used: where α l,k and θ l,k are the complex scattering coefficient and azimuth angle of the lth path of user k relative to the base station, and it is assumed that the user is parallel to the antenna array, that is, the AoA and AoD are consistent. If the perception module has used the root-finding MUSIC algorithm to obtain all the AoA / AoD information of the target user and the estimated azimuth angle set The channel estimation value correction scheme for sensing aided massive MIMO is transmitted to the communication module, and the algorithm flow is shown in Table 1.

[0142] Table 1 Algorithm flow of perception-assisted massive MIMO channel estimation correction

[0143]

[0144] In summary, this technical solution proposes a semi-blind estimation scheme for large-scale MIMO time-varying channels assisted by perception. First, the communication module of the synaesthesia dual-function base station receives uplink communication data from multiple users, and uses the low rank property of the large-scale MIMO matrix to implement channel initialization based on the complete low-rank matrix; then, the perception module receives the detection echo signals of multiple targets, and uses the root-finding MUSIC algorithm to achieve effective estimation of the channel vector parameters AoA / AoD, thereby assisting in correcting the initial channel value; on this basis, the dual-function base station performs channel estimation and target perception in a time-division multiplexing mode, and combines the channel correction value obtained in the previous cycle, and uses a semi-blind detection method with the help of a small number of pilots to achieve low-complexity dynamic tracking of subsequent channel characteristics. The expected results of this patent will further promote the rapid development of synaesthesia integration theory research, and is expected to be applied to many fields such as autonomous driving, intelligent control, and industrial monitoring.

Claims

1. A perception-assisted semi-blind estimation method for massive MIMO time-varying channels, characterized by: The following steps are involved: S1. The communication module of the synaesthesia dual-function MIMO base station receives uplink communication data from multiple users and uses the low-rank characteristics of the massive MIMO matrix to achieve channel initialization based on the complete low-rank matrix. S2, the perception module of the dual-function MIMO base station receives the reflected echo signals of multiple targets and uses the root-finding MUSIC algorithm to effectively estimate the channel vector parameters AoA / AoD, thereby assisting in correcting the initial channel value; S3. The synaesthesia dual-function MIMO base station performs channel estimation and target perception in time division multiplexing mode in sequence, combines the channel correction value obtained in the previous cycle, and uses the semi-blind detection method to realize the joint estimation of subsequent channel and load data.

2. The method for semi-blind estimation of a perception-assisted massive MIMO time-varying channel according to claim 1, wherein: The step S1 specifically includes the following steps: S11, a communication module of the synaesthesia dual-function MIMO base station receives uplink communication data from multiple users; S12. Based on the low-rank property of massive MIMO matrix, the communication module constructs the low-rank matrix completion problem; S13. Use the alternating steepest descent algorithm to solve the low-rank matrix completion problem and obtain the channel initial value, thus completing the channel initialization.

3. The method for semi-blind estimation of a perception-assisted massive MIMO time-varying channel according to claim 2, wherein: The low-rank matrix completion problem is specifically: Among them, U and V represent the channel and load data estimates respectively, The elements at index set Ω are represented by Given, the rest are set to 0, represents a linear operator that preserves the non-zero elements in Ω, represents the Frobenius norm of the matrix, N a is the number of base station antennas, N t is the number of transmitting antennas, K is the number of users, T c is the channel coherence time.

4. The method for semi-blind estimation of a sensing-assisted massive MIMO time-varying channel according to claim 3, wherein: The specific process of step S13 is as follows: make and Represents functions respectively Relative to the gradients of U and V, the step size updated along the steepest gradient direction is: in Based on this, for the t+1th iteration, U and V are updated alternately as follows until convergence: Assume that the pilot and data transmission follow: X≡[X p ,X d ],in And T p ∝KN t , then V=ΓX=Γ[X p ,X d ], since the pilot sequence usually satisfies the orthogonality, that is, The corresponding invertible matrix is: Among them, V(:,1~T p ) represents the 1~T of the matrix V p After solving the analytical value of the invertible matrix, the initial estimation of the channel can be calculated:

5. The method for semi-blind estimation of a sensing-assisted massive MIMO time-varying channel according to claim 4, wherein: The specific process of step S2 is: When the dual-function MIMO base station performs the sensing function, it first sends an orthogonal sounding signal satisfy Then the reflected echo signals from K users are: Among them, β k,l and τ k,l represent the reflection coefficient and delay-Doppler effect of user k corresponding to the lth path, f s is the carrier frequency of the sensing signal, a t (θ k,l ) and a r (θ k,l ) are the base station transmitting and receiving antenna steering vectors respectively, represents the receive beamforming matrix of the base station; Affected by the large-scale characteristics of the scattering environment, the signal angle changes relatively slowly. The AoA of the uplink signal in each transmission cycle is considered to be equal to the reception angle of the perceived signal, that is, Afterwards, order: The covariance matrix of the perceived echo signal is calculated as follows: in, and According to the traditional MUSIC algorithm principle, the signal and noise subspaces based on eigenvalue decomposition are as follows: in represents a diagonal matrix with eigenvalues ​​in descending order, and The eigenvector matrices of the signal subspace and the noise subspace are respectively formed. Given the orthogonal relationship between the signal and noise subspaces, the MUSIC spectrum used to locate the K user azimuths is expressed as: Rewrite the denominator of the above formula as: Where ξ=2πd / λ,C l Represents the correlation matrix along The sum of the elements of the lth diagonal, if defined Then we simplify to get the polynomial: f(z)2(N r -1) roots correspond exactly to the extremes of the MUSIC spectrum: Select the KL roots closest to the unit circle to generate the estimated value of the target user's AoA / AoD; After the AoA / AoD estimation value of the given channel is obtained, the finite scattering characteristics of the high-frequency channel are utilized to construct a sparse reconstruction problem and solve it to obtain the auxiliary correction matrix, which is then used to correct the initial channel value.

6. The method for semi-blind estimation of a sensing-assisted massive MIMO time-varying channel according to claim 5, wherein: The estimated values ​​of the target user's AoA / AoD are specifically:

7. The method for semi-blind estimation of a sensing-assisted massive MIMO time-varying channel according to claim 5, wherein: The corrected initial channel value in step S2 is specifically: in, is an auxiliary correction matrix, the elements of which include the scattering path gain α k,l and delay spread exponent ν k,l .

8. The method for semi-blind estimation of a perception-assisted massive MIMO time-varying channel according to claim 7, wherein: The semi-blind detection method in step S3 specifically comprises the following steps: S31. Considering the random and independent changes of channels in different transmission periods, the RF domain receiving signal model is expanded; S32, based on the channel correction value of the previous cycle, combined with the channel change constraint H of the adjacent cycle t =H t-1 +ΔH t , get the preliminary estimated value of the subsequent cycle load data; S33. Atomic norms are used to characterize the sparsity of the channel AoA / AoD domain. By constructing a combined atomic norm and l1 norm minimization problem, the conjugate gradient method is used to solve the subsequent periodic channel estimation value and decoding error estimation value, and the improved load data estimation value is obtained based on the preliminary load data estimation value.

9. The method for semi-blind estimation of a sensing-assisted massive MIMO time-varying channel according to claim 8, wherein: The combined atomic norm and l1 norm minimization problem is specifically: in, and are the channel estimation value and decoding error estimation value respectively, Y t For a given observation signal, that is, the output signal after the RF domain received signal model is expanded, For a preliminary estimate of load data, is the atomic norm, represents the channel matrix between user k and the base station.

10. The method for semi-blind estimation of a sensing-assisted massive MIMO time-varying channel according to claim 9, wherein: The load data improved estimated value is specifically: in, Improved estimates for load data.

Images