Channel estimation method suitable for super-large-scale MIMO communication

Abstract

The invention discloses a channel estimation method suitable for super-large-scale MIMO communication, and the method comprises the steps: decomposing a wave beam domain super-large-scale MIMO channel vector into an independent sparse part and an independent low-rank part, carrying out the prior modeling, forming an optimization target according to the Bayesian theory, fixing a channel sparse hyper-parameter and a low-rank hyper-parameter, and carrying out the priori modeling of the sparse part and the low-rank part. The method comprises the steps of calculating posterior expectations of a sparse part and a low-rank part of a channel, adjusting hyper-parameters and maximizing a likelihood function of the channel according to an existing estimation result, then judging whether the channel estimation result is converged or not according to whether the channel expectations of two iterations tend to be consistent or not, and finally outputting a beam-domain super-large-scale MIMO channel estimation result. According to the method, the low-rank performance and the sparsity of the channel are jointly utilized, different constraints are applied to the low-rank part and the sparse part of the channel matrix in the unified optimization criterion, and the channel estimation precision is remarkably improved.

Description

Technical Field

[0001] This application belongs to the field of channel estimation technology, and in particular relates to a channel estimation method suitable for ultra-large-scale MIMO communication. Background Technology

[0002] The innovation of information technology has driven the leapfrog evolution of mobile communication systems from the first generation (1G) to the fifth generation (5G). Currently, 5G communication systems have entered the stage of large-scale commercial use. They not only reshape the user experience through three core scenarios—enhanced mobile broadband, ultra-reliable low-latency communication, and massive machine-type communication—but also drive the digital transformation of vertical industries such as the Industrial Internet, smart healthcare, and the Internet of Vehicles. It is worth noting that although 5G has limitations in peak speed (20Gbps) and connection density (millions of terminals / km)... 2 While breakthroughs have been achieved in indicators such as 3D coverage, sub-millisecond latency, centimeter-level positioning accuracy, and integrated air-space-ground-sea networking, significant technical bottlenecks still exist in dimensions such as 3D coverage, sub-millisecond latency, centimeter-level positioning accuracy, and integrated air-space-ground-sea networking.

[0003] Looking towards the post-5G era, global industry, academia, and research institutions are actively developing sixth-generation (6G) communication systems, with a technological vision that can be summarized as "full coverage, full spectrum, full application, full sensory experience, full digitalization, and strong security." To achieve these goals, terahertz communication, intelligent metasurfaces, and integrated communication, sensing, and computing technologies have become key areas for breakthroughs. Among these, ultra-large-scale multiple-input multiple-output (MIMO) technology, through the deployment of hundreds of base station antenna arrays, demonstrates unique advantages in spatial multiplexing gain, beamforming accuracy, and energy efficiency optimization, and has now become a core solution for improving system capacity and spectral efficiency. To fully leverage the advantages of MIMO technology, obtaining accurate channel state information (CSI) is crucial in next-generation wireless communication systems.

[0004] Ultra-large-scale MIMO channels typically exhibit a sparse structure, meaning that most of the transmitted signal energy is concentrated in a few multipath components. As the antenna dimension of ultra-large-scale MIMO gradually increases, traditional least squares (LS) and minimum mean square error (MMSE) channel estimation algorithms fail to achieve an ideal balance between estimation accuracy and complexity because they do not utilize the channel's sparsity. Compressive sensing (CS)-based channel estimation algorithms, by leveraging the channel's sparsity, can reconstruct the channel with fewer samples, and have therefore attracted widespread attention. CS-based channel estimation algorithms can be divided into three categories: convex optimization algorithms, greedy algorithms, and Bayesian algorithms, each with advantages in estimation accuracy, efficiency, and robustness, respectively.

[0005] In related technologies, relying solely on the single characteristic of channel sparsity or low rank makes it difficult to accurately reconstruct the true channel state, resulting in poor channel estimation accuracy. Summary of the Invention

[0006] The purpose of this invention application is to provide a channel estimation method suitable for ultra-large-scale MIMO communication. This method can solve the technical problem in related technologies that rely solely on the single feature of channel sparsity or low rank, making it difficult to accurately reconstruct the real channel state, thus resulting in poor channel estimation accuracy.

[0007] In a first aspect, embodiments of this application provide a channel estimation method suitable for ultra-large-scale MIMO communication, including: step S1, decomposing the beam-domain ultra-large-scale MIMO channel vector h into independent sparse parts h S and the lower-rank part h L And for the sparse part h of the channel S and the lower-rank part h L Prior modeling is performed separately, and optimization objectives are formed based on Bayesian theory; Step S2: Based on the optimization objectives, the channel sparsity hyperparameter α and low-rank hyperparameter β are fixed, and h is calculated respectively. S and h L posterior expectation and Step S3, according to and Adjusting the hyperparameters α and β to maximize h S and h L Likelihood function; Step S4, according to and Generate the current iteration channel estimation result, and determine whether the iteration has converged based on the current iteration channel estimation result and the previous iteration channel estimation result. If it has converged or the maximum number of iterations has been reached, execute step S5. If it has not converged and the maximum number of iterations has not been reached, return to step S2. Step S5: Output the beam domain ultra-large-scale MIMO channel estimation result.

[0008] In one possible implementation of the first aspect, step S1 specifically includes:

[0009] Step S101: Model the sparse part h of the channel using a complex Gaussian distribution. S and the lower-rank part h L Prior distribution:

[0010]

[0011] in, I represents a complex Gaussian distribution with mean μ and variance σ. M Let h represent an M×M dimensional unit vector, where M is the number of antennas at the base station, and α and β are hyperparameters controlling sparsity and low rank, respectively; let h be the sparse part of the channel. S and the lower-rank part h L They are independent, and their prior probability can be expressed as:

[0012] P(h)=P(h S )P(h L )

[0013] Step S102: Let y be the received signal of the pilot sequence, the pilot length be N, and the signal transmission model be:

[0014] y = Ah + n

[0015] Where A is the measurement matrix containing pilot data, and n is additive white Gaussian noise with variance σ. 2 According to Bayes' theorem, the relationship between the posterior probability P(h|y) of channel h, the prior probability P(h), and the likelihood function P(h|y) is as follows:

[0016] P(h|y)P(y)=P(y|h)P(h)

[0017] Step S103, since h = h S +h L The likelihood function of the channel satisfies the following distribution:

[0018]

[0019] Where ||·||2 represents the l2 norm, and exp(·) represents the natural exponent. According to Bayes' theorem, the posterior probability P(h|y) of the channel is calculated from the prior probability P(h) and the likelihood function P(h|y):

[0020]

[0021] Where ∝ represents proportional to;

[0022] Step S104: Using the nuclear norm as an approximation of the rank function, and the j1 norm to constrain sparsity, the problem of maximizing the posterior probability P(h|y), i.e. the optimization objective, can be transformed into:

[0023]

[0024] Where ||·||1 represents the l1 norm, and ||·|| * This represents the nuclear norm.

[0025] Optionally, in another possible implementation of the first aspect, step S2 specifically includes:

[0026] Step S201, h S Divided into G channel blocks, i.e. Represents the i-th channel block;

[0027] Fixed h L Then h S The optimization problem is expressed as:

[0028]

[0029] Where r S =y-Ah L For the sparse residual of the channel, A i It is the column in the measurement matrix that is associated with the i-th channel block;

[0030] Step S202: Employ the block sparse Bayesian learning method, h S The posterior expectation μ and variance ∑ are:

[0031] μ=ΓA H (AΓA H +σ 2 I) -1 r S

[0032] ∑=Γ-ΓA H (AΓA H +σ 2 I) -1 AΓ

[0033] where Γ=diag(γ1C1,γ2C2,...,γ G C G ) is the block correlation matrix, C i and γ i They are respectively The correlation matrix and weights; to avoid overfitting, the correlation matrix of each channel block is fixed at a constant value:

[0034]

[0035]

[0036] Where L is the length of a single channel block, then h S The estimated value is the posterior expectation:

[0037]

[0038] Step S203, Fixed channel sparse part h S h L The optimization problem is expressed as:

[0039]

[0040] Where r L =y-Ah S For the low-rank residual of the channel, the gradient descent method is used to calculate... The gradient is:

[0041]

[0042] According to the gradient descent criterion, the gradient value is gradually decreased in each iteration, and the estimation result obtained in the k-th iteration is:

[0043]

[0044] Where T is the iteration step size, and its value is matrix A. T The largest eigenvalue of A is to simultaneously minimize the regularization term β||h L || * By applying a soft threshold operation to impose a regularization constraint, the final estimation result of the k-th iteration is obtained:

[0045]

[0046] Where sign(·) represents the sign function. If the input value of the sign function is greater than 0, sign(·) returns 1; if it is equal to 0, it returns 0; if it is less than 0, it returns -1; threshold. Used for control matrix h L The low-rank nature;

[0047] For h L Perform singular value decomposition and use a soft threshold to constrain the number of singular values, setting singular values ​​smaller than a preset threshold to zero:

[0048]

[0049] Where σ i for h L The i-th singular value in matrix form, for h L After 100 iterations, the final low-rank component estimation result is output.

[0050] Optionally, in another possible implementation of the first aspect, step S3 specifically includes:

[0051] based on and The likelihood functions for hyperparameters α and β are as follows:

[0052]

[0053] Where N is the pilot length, M is the number of antennas at the base station, and h is the base station length. S and h L To maximize the likelihood function, the posterior expectation is obtained by adjusting the hyperparameters α and β, and their update formulas are as follows:

[0054]

[0055] Among them ||·|| F Represents the Frobenius norm. They are respectively The posterior weight, posterior mean, and posterior covariance of the i-th element, ∑ L yes The posterior covariance.

[0056] Optionally, in another possible implementation of the first aspect, step S4 specifically includes:

[0057] The estimation results of the sparse and low-rank components are superimposed to calculate the channel estimation result for the current iteration:

[0058]

[0059] Obtain the channel estimation results from the previous iteration. pass

[0060]

[0061] Determine whether the iteration has converged. If it has converged or the maximum number of iterations has been reached, proceed to step S5; if it has not converged and the maximum number of iterations has not been reached, return to step S2.

[0062] Optionally, in another possible implementation of the first aspect, step S5 specifically includes:

[0063] The beam-domain ultra-large-scale MIMO channel estimation result is approximately the superposition of the sparse part and the low-rank part at the end of the iteration, that is:

[0064]

[0065] This application proposes a channel estimation method suitable for ultra-large-scale MIMO communication. It decomposes the beam-domain ultra-large-scale MIMO channel vector into independent sparse and low-rank components, performs prior modeling on each, formulates an optimization objective based on Bayesian theory, and then fixes the channel sparsity hyperparameters and low-rank hyperparameters respectively. The posterior expectations of the sparse and low-rank components are calculated. Based on the existing estimation results, the hyperparameters are adjusted to maximize the channel likelihood function. The convergence of the channel estimation result is then determined by whether the channel expectations of two iterations converge. Finally, the beam-domain ultra-large-scale MIMO channel estimation result is output. This application jointly utilizes the low-rank and sparsity of the channel, applying different constraints to the low-rank and sparse components of the channel matrix under a unified optimization criterion, significantly improving the channel estimation accuracy. Attached Figure Description

[0066] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0067] Figure 1 A flowchart illustrating a channel estimation method for ultra-large-scale MIMO communication provided in an embodiment of this application;

[0068] Figure 2 A schematic diagram illustrating the sparsity of a beam domain channel for a super-large-scale MIMO provided in an embodiment of this application;

[0069] Figure 3 A schematic diagram illustrating the low-rank property of a beam domain channel for a very large-scale MIMO provided in an embodiment of this application;

[0070] Figure 4 This diagram illustrates the normalized mean square error performance of the LRSBE algorithm provided in one embodiment of this application compared with other contrasting channel estimation algorithms as a function of signal-to-noise ratio. Detailed Implementation

[0071] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0072] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0073] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0074] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0075] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0076] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0077] The following is a detailed description of a channel estimation method for very large-scale MIMO communication provided in this application, with reference to the accompanying drawings.

[0078] like Figure 1 As shown, a channel estimation method suitable for ultra-large-scale MIMO communication specifically includes the following steps:

[0079] Step S1: Decompose the beam domain ultra-large-scale MIMO channel vector h into independent sparse parts h S and the lower-rank part h L And for the sparse part of the channel g S and the lower-rank part h L Prior modeling is performed separately, and optimization objectives are formed based on Bayesian theory.

[0080] Specifically, in this embodiment, step S1 includes:

[0081] Step S101: Model the sparse part h of the channel using a complex Gaussian distribution. S and the lower-rank part h L prior distribution

[0082]

[0083] in, I represents a complex Gaussian distribution with mean μ and variance σ. M Let represent an M×M dimensional unit vector, where M is the number of antennas at the base station, and α and β are hyperparameters controlling sparsity and low rank, respectively.

[0084] As one possible implementation method, Figure 2 and Figure 3 The low-rank and sparsity of the beam domain and spatial domain channels are demonstrated when the base station is equipped with 32×32 antennas. Figure 2 The distribution of singular values ​​of the channel matrix is ​​described. In both the beam domain and the spatial domain, 99% of the energy of the singular values ​​of the channel matrix is ​​concentrated in the first 5, indicating that the matrix has only a small number of non-zero singular values ​​and exhibits obvious low-rank property. Figure 3 This shows the size distribution of channel elements. It can be seen that the spatial domain channel elements are evenly distributed in each interval and there is no obvious sparsity. However, the beam domain channel achieves energy focusing through two-dimensional Fourier transform, so that the effective information is concentrated on a few non-zero elements, which means it has obvious sparsity.

[0085] Let h be the sparse part of the channel. S and the lower-rank part h L It is independent; the prior probability of the entire channel can be expressed as the sparse part h. S and the lower-rank part h L The product of prior probabilities:

[0086] P(h)=P(h S )P(h L )

[0087] Step S102: Assume y is the received signal of the pilot sequence, the pilot length is N, and the signal transmission model is:

[0088] y = Ah + n

[0089] Where A is the measurement matrix containing pilot data, and n is additive white Gaussian noise with a mean of 0 and a variance of σ. 2 According to Bayes' theorem, the relationship between the posterior probability P(h|y) of channel h, the prior probability P(h), and the likelihood function P(y|h) is as follows:

[0090] P(h|y)P(y)=P(y|h)P(h)

[0091] Step S103, since h = h S +h L The likelihood function of the channel satisfies the following distribution:

[0092]

[0093] Where ||·||² represents the l² norm, and exp(·) represents the natural exponent. Therefore, according to Bayes' theorem P(y|h), the posterior probability P(y|h) of the channel can be calculated from the prior probability P(h) and the likelihood function P(y|h):

[0094]

[0095] Where ∝ represents proportional to.

[0096] It should be noted that, according to the above formula, maximizing P(y|h) is equivalent to minimizing

[0097] Step S104: Using the nuclear norm as an approximation of the rank function and the l1 norm to constrain sparsity, the problem of maximizing the posterior probability P(h|y)) is the optimization objective, which can be transformed into:

[0098]

[0099] Where ||·||1 represents the l1 norm, and ||·||* represents the nuclear norm.

[0100] Step S2: Based on the optimization objective, fix the channel sparsity hyperparameter α and the low-rank hyperparameter β, and calculate h respectively. S and h L posterior expectation and

[0101] Specifically, in this embodiment, step S2 includes:

[0102] Step S201: Based on the clustering of non-zero channel elements in the channel matrix, h S It is divided into G channel blocks for processing, that is Represents the i-th channel block. h is fixed. L Then h S The optimization problem can be expressed as:

[0103]

[0104] Where r S =y-Ah L Let Ai be the residual of the sparse part of the channel, and Ai be the column in the measurement matrix associated with the i-th channel block. Therefore, in h S The estimation process will not be affected by h L The impact.

[0105] Step S202: Employ the block sparse Bayesian learning method, h S The posterior expectation μ and variance ∑ are:

[0106] μ=ΓA H (AΓA H +σ 2 I) -1 r S

[0107] ∑=Γ-ΓA H (AΓA H +σ 2 I) -1 AΓ

[0108] where Γ=diag(γ1C1,γ2C2,...,γ G C G ) is the block correlation matrix, C i and γ i They are respectively The correlation matrix and weights; to avoid overfitting, the correlation matrix of each channel block is fixed at a constant value:

[0109]

[0110] Where L is the length of a single channel block. Then h S The estimated value is the posterior expectation:

[0111]

[0112] Step S203: Change to a fixed channel sparse part h S , so that in h L The estimation process will not be affected by h S The impact, h L The optimization problem can be expressed as:

[0113]

[0114] Where r L =y-Ah S For the low-rank residual of the channel, using the gradient descent method, the gradient of the first term in the above equation is first calculated as follows:

[0115]

[0116] Then, according to the gradient descent criterion, the gradient value is gradually decreased in each iteration, where the estimation result obtained in the k-th iteration is:

[0117]

[0118] Where T is the iteration step size, and to ensure the convergence of the iteration, it is taken as the value of matrix A. T The largest eigenvalue of A is to simultaneously minimize the regularization term β||h L || * By applying a soft thresholding operation to the regularization constraint on the results of each iteration, the final estimate for the k-th iteration is:

[0119]

[0120] Here, `sign(·)` represents the sign function. If the input value of the sign function is greater than 0, `sign(·)` returns 1; if the input value is equal to 0, it returns 0; and if the input value is less than 0, it returns -1. Additionally, the threshold... Used to control the low-rank property of the matrix, as β increases, h L More and more elements will be set to zero.

[0121] To further constrain h L The low-rank property of h L Perform singular value decomposition and again apply a soft threshold to constrain the number of singular values, setting smaller singular values ​​to zero:

[0122]

[0123] Where σ i for h L The i-th singular value in matrix form, for h L After 100 iterations, the final low-rank component estimation result is output.

[0124] Step S3, according to and Adjusting the hyperparameters α and β to maximize h S and h L The likelihood function.

[0125] Specifically, in this embodiment, step S3 includes:

[0126] Based on the estimation of the current iteration and The likelihood functions for hyperparameters α and β are as follows:

[0127]

[0128] Where N is the pilot length, M is the number of antennas at the base station, and h is the base station length. S and h L To maximize the likelihood function, the posterior expectation is obtained by adjusting the hyperparameters α and β, and their update formulas are as follows:

[0129]

[0130] Among them ||·|| F Represents the Frobenius norm. They are respectively The posterior weight, posterior mean, and posterior covariance of the i-th element. ∑ L yes The posterior covariance.

[0131] Step S4, according to and Generate the current iteration channel estimation result, and determine whether the iteration has converged based on the current iteration channel estimation result and the previous iteration channel estimation result. If it has converged or the maximum number of iterations has been reached, proceed to step S5. If it has not converged and the maximum number of iterations has not been reached, return to step S2.

[0132] Specifically, in this embodiment, step S4 includes:

[0133] The estimation results of the sparse and low-rank components are superimposed to calculate the overall channel estimation result for this iteration:

[0134]

[0135] Assumption Based on the channel estimation results of the previous iteration, through

[0136]

[0137] Determine if the iteration has converged. If it has converged or the maximum number of iterations has been reached, proceed to step S5. If it has not converged and the maximum number of iterations has not been reached, return to step S2.

[0138] Step S5: Output beam domain ultra-large-scale MIMO channel estimation results.

[0139] Specifically, in this embodiment, step S5 includes:

[0140] The beam-domain ultra-large-scale MIMO channel estimation result is approximately the superposition of the sparse part and the low-rank part at the end of the iteration, that is:

[0141]

[0142] This application provides a channel estimation method suitable for ultra-large-scale MIMO communication. It decomposes the beam-domain ultra-large-scale MIMO channel vector into independent sparse and low-rank components, performs prior modeling on each, formulates an optimization objective based on Bayesian theory, and then fixes the channel sparsity hyperparameters and low-rank hyperparameters respectively. The posterior expectations of the sparse and low-rank components are calculated. Based on the existing estimation results, the hyperparameters are adjusted to maximize the channel likelihood function. The convergence of the channel estimation result is then determined by whether the channel expectations of two iterations converge. Finally, the beam-domain ultra-large-scale MIMO channel estimation result is output. This application jointly utilizes the low-rank and sparsity of the channel, applying different constraints to the low-rank and sparse components of the channel matrix under a unified optimization criterion, significantly improving the channel estimation accuracy.

[0143] The advantages of this application will be illustrated below with an example.

[0144] In this embodiment, a 16×16 base station antenna array and 10 randomly generated users are used for estimation, and a pilot sequence of length 5 is used to complete 15 random samplings. In the urban microcell scenario, the performance of the proposed LRSBE algorithm is compared with that of other typical channel estimation algorithms, including the classic greedy algorithm Orthogonal Matching Pursuit (OMP), the Iterative Soft Thresholding Algorithm (ISTA) based on convex optimization, the Sparse Bayesian Learning Estimation (SBE) algorithm [7] and the Block SBE (BSBE) algorithm. Figure 4The results show that the proposed LRSBE algorithm outperforms other algorithms in terms of NMSE under different SNR conditions. Since the greedy algorithm is sensitive to noise, OMP performs relatively poorly under low SNR conditions. In contrast, SBL has a lower NMSE due to the introduction of prior assumptions, while BSBE further utilizes the channel's block structure, thus improving estimation accuracy. ISTA, through l1-norm regularization, can suppress the influence of noise, and its accuracy is even better than BSBE, but it still faces a performance bottleneck due to the use of a fixed soft threshold. However, the above algorithms only utilize the channel's sparsity, ignoring the channel's low-rank property in the low SNR region. By simultaneously fusing the channel's low-rank and sparsity, the proposed LRSBE significantly improves estimation accuracy. At an SNR of -15dB, its NMSE is reduced by approximately 3dB compared to the best-performing comparison algorithm ISTA; at an SNR of 15dB, its NMSE is reduced by approximately 1dB compared to the best-performing comparison algorithm BSBE.

[0145] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0146] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A channel estimation method suitable for ultra-large-scale MIMO communication, characterized in that, Includes the following steps: Step S1: Decompose the beam domain ultra-large-scale MIMO channel vector h into independent sparse parts g. S and the lower-rank part g L And for the sparse part of the channel g S and the lower-rank part h L Prior modeling is performed separately, and optimization objectives are formed based on Bayesian theory; Step S2: Based on the optimization objective, fix the channel sparsity hyperparameter α and the low-rank hyperparameter β, and calculate h respectively. S and h L posterior expectation and Step S3, according to and Adjusting the hyperparameters α and β to maximize h S and h L The likelihood function; Step S4, according to and Generate the current iteration channel estimation result, and determine whether the iteration has converged based on the current iteration channel estimation result and the previous iteration channel estimation result. If it has converged or the maximum number of iterations has been reached, proceed to step S5. If it has not converged and the maximum number of iterations has not been reached, return to step S2. Step S5: Output beam domain ultra-large-scale MIMO channel estimation results.

2. The channel estimation method for ultra-large-scale MIMO communication according to claim 1, characterized in that, Step S1 specifically includes the following steps: Step S101: Model the sparse part h of the channel using a complex Gaussian distribution. S and the lower-rank part h L Prior distribution: in, I represents a complex Gaussian distribution with mean μ and variance σ. M Let h represent an M×M dimensional unit vector, where M is the number of antennas at the base station, and α and β are hyperparameters controlling sparsity and low rank, respectively; let h be the sparse part of the channel. S and the lower-rank part h L They are independent, and their prior probability can be expressed as: P(h)=P(h S )P(h L ) Step S102: Let y be the received signal of the pilot sequence, the pilot length be N, and the signal transmission model be: y = Ah + n Where A is the measurement matrix containing pilot data, and n is additive white Gaussian noise with variance σ. 2 According to Bayes' theorem, the relationship between the posterior probability P(h|y) of channel h, the prior probability P(h), and the likelihood function P(y|h) is as follows: P(h|y)P(y)=P(y|h)P(h) Step S103, since h = h S +h L The likelihood function of the channel satisfies the following distribution: Where ||·||2 represents the l2 norm, and exp(·) represents the natural exponent. According to Bayes' theorem, the posterior probability P(h|y) of the channel is calculated from the prior probability P(h) and the likelihood function P(y|h): P(y|h)=P(h S ,h L |y)∝P(y|h S +h L )P(h S )P(h L ) Where ∝ represents proportional to; Step S104: Using the nuclear norm as an approximation of the rank function, and the l1 norm to constrain sparsity, the problem of maximizing the posterior probability P(h|y), i.e. the optimization objective, can be transformed into: Where ||·||1 represents the l1 norm, and ||·|| * This represents the nuclear norm.

3. The channel estimation method for ultra-large-scale MIMO communication according to claim 2, characterized in that, Step S2 specifically includes the following steps: Step S201, h S Divided into G channel blocks, i.e. Represents the i-th channel block; Fixed h L Then h S The optimization problem is expressed as: Where r S =y-Ah L For the sparse residual of the channel, A i It is the column in the measurement matrix that is associated with the i-th channel block; Step S202: Employ the block sparse Bayesian learning method, h S The posterior expectation μ and variance ∑ are: μ=ΓA H (AGA H +s 2 I) -1 r S ∑=Γ-ΓA H (AGA H +s 2 I) -1 AG where Γ=diag(γ1C1,γ2C2,...,γ G C G ) is the block correlation matrix, C i and γ i They are respectively The correlation matrix and weights; to avoid overfitting, the correlation matrix of each channel block is fixed at a constant value: Where L is the length of a single channel block, then h S The estimated value is the posterior expectation: Step S203, Fixed channel sparse part h S h L The optimization problem is expressed as: Where r L =y-Ah S For the low-rank residual of the channel, the gradient descent method is used to calculate... The gradient is: According to the gradient descent criterion, the gradient value is gradually decreased in each iteration, and the estimation result obtained in the k-th iteration is: Where T is the iteration step size, and its value is matrix A. T The largest eigenvalue of A is to simultaneously minimize the regularization term β||h L || * By applying a soft threshold operation to impose a regularization constraint, the final estimation result of the k-th iteration is obtained: Where sign(·) represents the sign function. If the input value of the sign function is greater than 0, sign(·) returns 1; if it is equal to 0, it returns 0; if it is less than 0, it returns -1; threshold. Used for control matrix h L The low-rank nature; For h L Perform singular value decomposition and use a soft threshold to constrain the number of singular values, setting singular values ​​smaller than a preset threshold to zero: Where σ i for h L The i-th singular value in matrix form, for h L After 100 iterations, the final low-rank component estimation result is output.

4. The channel estimation method for ultra-large-scale MIMO communication according to claim 3, characterized in that, Step S3 specifically includes the following steps: based on and The likelihood functions for hyperparameters α and β are as follows: Where N is the pilot length, M is the number of antennas at the base station, and h is the base station length. S and h L To maximize the likelihood function, the posterior expectation is obtained by adjusting the hyperparameters α and β, and their update formulas are as follows: Among them ||·|| F Represents the Frobenius norm. They are respectively The posterior weight, posterior mean, and posterior covariance of the i-th element, ∑ L yes The posterior covariance.

5. The channel estimation method for ultra-large-scale MIMO communication according to claim 1, characterized in that, Step S4 specifically includes the following steps: The estimation results of the sparse and low-rank components are superimposed to calculate the channel estimation result for the current iteration: Obtain the channel estimation results from the previous iteration. pass Determine whether the iteration has converged. If it has converged or the maximum number of iterations has been reached, proceed to step S5; if it has not converged and the maximum number of iterations has not been reached, return to step S2.

6. The channel estimation method for ultra-large-scale MIMO communication according to claim 1, characterized in that, Step S5 specifically includes the following steps: The beam-domain ultra-large-scale MIMO channel estimation result is approximately the superposition of the sparse part and the low-rank part at the end of the iteration, that is:

Images