Channel estimation algorithm of RIS-assisted large-scale MIMO system based on variational Bayesian inference

Abstract

The invention relates to the technical field of wireless communication, in particular to a variational Bayesian inference-based RIS-assisted large-scale MIMO system channel estimation algorithm, which comprises the following steps: S1, establishing a single RIS-assisted large-scale MIMO system model through parameter setting, and determining the number of transmitting antennas, receiving antennas and RIS reflection elements of a large-scale MIMO system; according to the method, the uncertainty of the channel is modeled through the variational Bayesian inference technology, and the posterior distribution of the channel state information is deduced, so that the channel estimation precision is improved; meanwhile, an adaptive compressed sensing algorithm is introduced, a compressed sensing matrix is dynamically optimized by using channel sparsity characteristics, the data sampling amount and the calculation complexity are reduced, and compared with the prior art, the method has higher estimation precision, lower calculation complexity and higher system robustness in a high-noise and complex channel environment, and the method is suitable for being applied to the field of communication. The method is suitable for efficient deployment and application of a large-scale MIMO system in a next-generation wireless communication network.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to a channel estimation algorithm for a RIS-assisted large-scale MIMO system based on variational Bayesian inference. Background Art

[0002] With the rapid development of 5G and future 6G communications, Massive MIMO technology has become an important means to improve system capacity and spectrum efficiency. However, as the number of antennas increases, channel estimation faces challenges such as high dimensionality, sparsity, and computational complexity. Traditional channel estimation methods are difficult to meet the requirements of efficiency and accuracy. Especially in the millimeter wave frequency band, the high dimensionality and sparsity of the channel make the traditional methods limited in effectiveness.

[0003] In order to improve the channel estimation performance, the intelligent reflecting surface (RIS) technology came into being. RIS enhances signal transmission by adjusting the phase of the reflective unit, but its introduction makes the channel estimation problem more complicated. Current channel estimation algorithms mostly use variational Bayesian inference and compressed sensing technology. Variational Bayesian inference can handle the uncertainty in the channel and provide high-precision estimation, while adaptive compressed sensing uses channel sparsity to effectively restore channel information through a small amount of sampling and reduce computational overhead.

[0004] Although these two methods solve the accuracy and efficiency problems in channel estimation respectively, it is still difficult to combine them and apply them to the channel estimation of RIS-assisted massive MIMO systems. Therefore, the current methods cannot optimize both accuracy and efficiency at the same time, and the channel estimation performance in RIS-assisted massive MIMO systems is low, the computational complexity is high, and it cannot meet the real-time requirements of the system well. Therefore, we propose a channel estimation algorithm for RIS-assisted massive MIMO systems based on variational Bayesian inference to solve the above problems. Summary of the invention

[0005] 1. Technical issues to be solved

[0006] In view of the deficiencies in the prior art, the present invention provides a RIS-assisted channel estimation algorithm for a massive MIMO system based on variational Bayesian inference, which solves the problems raised in the above background technology.

[0007] (II) Technical solution

[0008] In order to achieve the above-mentioned purpose, the present invention specifically adopts the following technical solutions:

[0009] A channel estimation algorithm for RIS-assisted massive MIMO system based on variational Bayesian inference, comprising the following steps:

[0010] S1: Establish a system model of single RIS-assisted massive MIMO through parameter setting, and determine the number of transmit antennas, receive antennas and RIS reflection elements of the massive MIMO system;

[0011] S2: Based on the system model of single RIS-assisted massive MIMO, an objective function is established to represent the performance index of channel estimation;

[0012] S3: Use the adaptive compressed sensing algorithm to preliminarily estimate the channel to obtain the initial estimation result, and use the variational Bayesian method to perform posterior update based on the initial channel estimation result and the received signal to optimize the objective function;

[0013] S4: Based on the system model of a single RIS-assisted massive MIMO and the optimized objective function, the final channel estimation result is output to evaluate the effectiveness of the algorithm.

[0014] Furthermore, the system model of the single RIS-assisted massive MIMO in S1 is:

[0015] The total angular response matrix Θ of the downlink transmission channel of the RIS-assisted massive MIMO system, the channel matrix F from the BS end to the RIS end, and the channel matrix H from the RIS end to the USERs end k , the user's channel matrix is ​​denoted as G k , specifically defined as:

[0016]

[0017] In which, the antenna response vectors from BS to RIS, the transmitting end and the receiving end are respectively x for When Y=NL, it represents the angle response matrix from the transmitter to the reflector. When θ x for When Y = NM, it represents the angle response matrix from the reflector to the receiving end. When θ x for When Y=NL, it represents the angle response matrix from USERs to RIS, NL represents the number of RIS reflection units, NM represents the number of antennas at BS, and N f It indicates the number of channels from BS to RIS. It represents the incident angle from BS to RIS. It represents the angle of emission from RIS to USERs, rand(1,N f ) represents the generation of a fA row vector of random numbers uniformly distributed in the range [0,1); represents the incident angle from USERs to RIS, rand(k,N hk ) represents the generation of a k-row N hk A row vector of random numbers uniformly distributed in the range [0,1). The superscript T represents the transpose operator.

[0018]

[0019] Where N(0,1) represents a normal distribution with a mean of 0 and a variance of 1; the superscript H represents the conjugate transpose operator;

[0020]

[0021] Among them, N hk It represents the number of channels from the RIS end to the USERs end; N(0,1) represents a normal distribution with a mean of 0 and a variance of 1;

[0022] G k =diag((H k ) H )*F

[0023] Among them, diag((H k ) H ) is expressed as (H k ) H A diagonal matrix with diagonal elements; the superscript H represents the conjugate transpose operator.

[0024] Furthermore, the objective function in S2 is:

[0025]

[0026] in, represents the estimated channel matrix; G represents the theoretical channel matrix; and the superscript H represents the conjugate transpose.

[0027] Furthermore, in S3, the adaptive compressed sensing algorithm is used to perform a preliminary estimation of the channel, and the specific process of obtaining the initial estimation result is as follows:

[0028] Ⅰ: A dictionary matrix Dmatrixo is used for sparse representation, expressed as:

[0029] x=Dmatrixo·s

[0030] Among them, x is the original signal, Dmatrixo is the dictionary matrix, and s is the sparse coefficient vector;

[0031] II: Different from traditional signal sampling, the adaptive compressed sensing algorithm uses random sampling, which is usually described by a measurement matrix, expressed as:

[0032] z bk =Dmatrixo·x

[0033] Among them, z bk is the compressed measurement vector, Dmatrixo is the dictionary matrix, and x is the original signal;

[0034] III: The matching pursuit algorithm is used in the signal sparse reconstruction process. The algorithm first selects the column vector with the highest correlation with the current residual to update the reconstruction set, and iteratively updates the residual until a certain condition is met. The iterative process is expressed as:

[0035] R k =z bk -Dmatrix·s k

[0036]

[0037] Among them, R k is the residual vector of the kth iteration, Dmatrix is ​​the normalized dictionary matrix, s k is the sparse coefficient vector of the kth iteration;

[0038] IV: During the iteration process, the stop condition of the iteration is controlled by dynamically adjusting the threshold η. The formula for dynamically reducing the threshold is:

[0039] η=max(η k ·decay_rate, current_residual_energy)

[0040] Among them, η k is the threshold of the current iteration, decay_rate is the decay rate of the threshold, and current_residual_energy is the current residual energy;

[0041] The formula for dynamically increasing the threshold is:

[0042] η=max(η k ·increase_rate, current_residual_energy)

[0043] Among them, η k is the threshold of the current iteration, decay_rate is the rate of increase of the threshold, and current_residual_energy is the current residual energy.

[0044] Furthermore, the variational Bayesian method is used in S3 to perform posterior updating, and the specific process of optimizing the objective function is:

[0045] Ⅰ: Assume that the channel coefficient X follows a multivariate Gaussian distribution;

[0046]

[0047] Among them, m0 represents the prior mean vector, Λ0 represents the precision matrix of the prior distribution, that is, the inverse of the covariance matrix;

[0048] II: The above formula Y k As observation data, Y k The likelihood function is expressed as:

[0049]

[0050] Among them, Y k represents the observed signal matrix, represents the estimated value of the channel matrix, σ 2 represents the noise variance, and I represents the unit matrix; Among them, Y k represents the received signal matrix of user k; G k Let the channel matrix of user k be the channel matrix F from BS to RIS and the channel matrix H from RIS to USERs k Composition; V represents the reflection coefficient matrix at the RIS end; N k Expressed as the noise matrix, G k =diag((H k ) H )*F, where diag((H k ) H ) is expressed as (H k ) H The elements of are diagonal elements of the diagonal matrix; the superscript H represents the conjugate transpose operator, and the reflection coefficient matrix of RIS is represented by V. Among them, randn(NL,N bpilot ) means to generate a size of NL*N bpilot are independent and identically distributed standard normal random variables; j is represented by an imaginary unit; This is to ensure that the variance of the V element is 1, making it a standard complex Gaussian random variable; the noise matrix in the channel transmission process is in, represents the i-th row of matrix U, N npilot (j-1)+1 to Userpilot represents the variable to which the generated DFT matrix is ​​assigned; SNR = 10.^(SNR / 10), which is the signal-to-noise ratio; is to ensure that the variance of the noise meets the given signal-to-noise ratio; N npilot is the training sequence length of each user;

[0051] III: Through Bayes' theorem, the posterior distribution also obeys the multivariate Gaussian distribution, expressed as:

[0052] p(X|Y k )=N(X|m,Λ -1 )

[0053] Among them, m represents the mean vector of the posterior distribution, and Λ represents the posterior precision matrix;

[0054]

[0055] IV: Optimize the variational parameters by maximizing the variational lower bound (EvidenceLowerBound, ELBO), expressed as:

[0056] ELBO(m,Λ)=E q [logp(Y,X)]-E q [logq(X)]

[0057] Among them, q(X) represents the approximate posterior distribution, and p(Y,X) represents the joint distribution, including the likelihood function and the prior distribution.

[0058] The process of continuously optimizing the values ​​of m and Λ is equivalent to iteratively optimizing ELBO through variational inference, so that the approximate posterior distribution q(X) is closer to the true posterior distribution p(X|Y k ).

[0059] (III) Beneficial effects

[0060] Compared with the prior art, the present invention provides a RIS-assisted massive MIMO system channel estimation algorithm based on variational Bayesian inference, which has the following beneficial effects:

[0061] The present invention, by introducing the variational Bayesian inference (VBI) method, can more accurately estimate the channel state information, especially in environments with high noise or large channel uncertainty, significantly improves the accuracy of channel estimation, and is more capable of coping with complex propagation conditions than traditional methods.

[0062] The present invention can effectively handle signal sparsity and uncertainty in the channel estimation process by combining variational Bayesian inference with an adaptive compressed sensing algorithm. Variational Bayesian inference can accurately infer the posterior distribution of the channel state and reduce the estimation error, while the adaptive compressed sensing algorithm can improve the estimation accuracy while reducing sampling by effectively selecting the measurement matrix and sparse reconstruction strategy. The combination of the two helps to achieve high-precision and robust channel estimation in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flow chart for realizing the algorithm of the present invention;

[0064] Figure 2 A schematic diagram of a transmission scenario of a RIS-assisted massive MIMO system of the present invention;

[0065] Figure 3 This is a comparison curve of the normalized mean square error as the signal-to-noise ratio changes during the training process of the present invention. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] Example

[0068] like Figure 1-3 As shown, an embodiment of the present invention proposes a channel estimation algorithm for a RIS-assisted massive MIMO system based on variational Bayesian inference, comprising the following steps:

[0069] S1: Establish a system model of single RIS-assisted massive MIMO through parameter setting, and determine the number of transmit antennas, receive antennas and RIS reflection elements of the massive MIMO system;

[0070] Figure 2The invention shows a system model of a wireless communication network of a single RIS assisted massive MIMO, specifically determining the number of antennas, antenna array structure, antenna placement, and determining the direction of antenna transmission. The invention defines that the number of antennas at the transmitting end of the massive MIMO system is 256, and the number of antennas at the receiving end is 256; the antenna array structure is set to a uniform array structure, which can realize more complex beamforming, including beams in the horizontal and vertical directions, and the uniform array is a multi-dimensional structure with greater flexibility, which effectively increases the signal coverage area; the antenna placement adopts a uniform distribution method to achieve spatial diversity; the direction of antenna transmission is set to The random angle values ​​in the RIS are used to simulate random transmission channels. The RIS reflection elements are set to 8, 16, 24, 40, 72, and 128, respectively, which can support common phased array designs and help develop phased array algorithms. More RIS reflection elements can provide better channel diversity, thereby resisting multipath fading and improving communication reliability while maintaining the feasibility of system design.

[0071] The channel model of a single RIS-assisted massive MIMO wireless communication network, specifically, it consists of a BS with NM antennas, a RIS with NL passive reflective elements, and k Users. The total angular response matrix Θ of the downlink transmission channel of the RIS-assisted massive MIMO system, the channel matrix F from the BS end to the RIS end, and the channel matrix H from the RIS end to the USERs end k , the user's channel matrix is ​​denoted as G k , specifically defined as:

[0072]

[0073] In which, the antenna response vectors from BS to RIS, the transmitting end and the receiving end are respectively x for When Y=NL, it represents the angle response matrix from the transmitter to the reflector. When θ x for When Y = NM, it represents the angle response matrix from the reflector to the receiving end. When θ x for When Y=NL, it represents the angle response matrix from USERs to RIS, NL represents the number of RIS reflection units, NM represents the number of antennas at BS, and N f It indicates the number of channels from BS to RIS. It represents the incident angle from BS to RIS. It represents the angle of emission from RIS to USERs, rand(1,N f ) represents the generation of a f A row vector of random numbers uniformly distributed in the range [0,1); represents the incident angle from USERs to RIS, rand(k,N hk ) represents the generation of a k-row N hk A row vector of random numbers uniformly distributed in the range [0,1). The superscript T represents the transpose operator.

[0074]

[0075] Where N(0,1) represents a normal distribution with a mean of 0 and a variance of 1; the superscript H represents the conjugate transpose operator;

[0076]

[0077] Among them, N hk It represents the number of channels from the RIS end to the USERs end; N(0,1) represents a normal distribution with a mean of 0 and a variance of 1;

[0078] G k =diag((H k ) H )*F

[0079] Among them, diag((H k ) H ) is expressed as (H k ) H A diagonal matrix with diagonal elements; the superscript H represents the conjugate transpose operator.

[0080] S2: Based on the system model of a single RIS-assisted massive MIMO wireless communication network, an objective function is established to represent the performance index of channel estimation.

[0081] Specifically, the objective function is:

[0082]

[0083] in, represents the estimated channel matrix; G represents the theoretical channel matrix; and the superscript H represents the conjugate transpose.

[0084] S3: Based on the system model of a single RIS-assisted massive MIMO wireless communication network, an adaptive compressed sensing algorithm is used to preliminarily estimate the channel to obtain an initial estimation result. Based on the initial channel estimation result and the received signal, a variational Bayesian method is used to perform posterior updates to optimize the objective function.

[0085] The specific process of using the adaptive compressed sensing algorithm to preliminarily estimate the channel and obtain the initial estimation result is as follows:

[0086] Ⅰ: This paper uses a dictionary matrix Dmatrixo for sparse representation, expressed as,

[0087] x=Dmatrixo·s

[0088] Where x is the original signal, Dmatrixo is the dictionary matrix, and s is the sparse coefficient vector.

[0089] II: Different from traditional signal sampling, the adaptive compressed sensing algorithm uses random sampling, which is usually described by a measurement matrix. It is expressed as:

[0090] z bk =Dmatrixo·x

[0091] Among them, z bk is the compressed measurement vector, Dmatrixo is the dictionary matrix, and x is the original signal.

[0092] III: The matching pursuit algorithm is used in the signal sparse reconstruction process. The algorithm first selects the column vector with the highest correlation with the current residual to update the reconstruction set, and iteratively updates the residual until a certain condition is met. The iterative process is expressed as:

[0093] R k =z bk -Dmatrix·s k

[0094]

[0095] Among them, R k is the residual vector of the kth iteration, Dmatrix is ​​the normalized dictionary matrix, s k is the sparse coefficient vector for the kth iteration.

[0096] IV: During the iteration process, the stop condition of the iteration is controlled by dynamically adjusting the threshold η. The formula for dynamically reducing the threshold is:

[0097] η=max(η k ·decay_rate, current_residual_energy)

[0098] Among them, η k is the threshold of the current iteration, decay_rate is the decay rate of the threshold, and current_residual_energy is the current residual energy.

[0099] The formula for dynamically increasing the threshold is:

[0100] η=max(η k ·increase_rate, current_residual_energy)

[0101] Among them, η k is the threshold of the current iteration, decay_rate is the rate of increase of the threshold, and current_residual_energy is the current residual energy.

[0102] Using the variational Bayesian method for posterior updating, the specific process of optimizing the objective function is:

[0103] Ⅰ: Assume that the channel coefficient X follows a multivariate Gaussian distribution;

[0104]

[0105] Among them, m0 represents the prior mean vector, and Λ0 represents the precision matrix of the prior distribution, that is, the inverse of the covariance matrix.

[0106] II: The above formula Y k As observation data, Y k The likelihood function is expressed as:

[0107]

[0108] Among them, Y k represents the observed signal matrix, represents the estimated value of the channel matrix, σ 2 represents the noise variance and I represents the identity matrix. Among them, Y k represents the received signal matrix of user k; G k Let the channel matrix of user k be the channel matrix F from BS to RIS and the channel matrix H from RIS to USERs k Composition; V represents the reflection coefficient matrix at the RIS end; N k It is represented as a noise matrix. G k =diag((H k ) H )*F, where diag((H k )H ) is expressed as (H k ) H The elements of RIS are diagonal matrices with diagonal elements; the superscript H represents the conjugate transpose operator. The reflection coefficient matrix of RIS is represented by V, Among them, randn(NL,N bpilot ) means to generate a size of NL*N bpilot are independent and identically distributed standard normal random variables; j is represented by an imaginary unit; This is to ensure that the variance of the V element is 1, making it a standard complex Gaussian random variable. The noise matrix during channel transmission is in, represents the i-th row of matrix U, N npilot (j-1)+1 to Userpilot represents the variable to which the generated DFT matrix is ​​assigned; SNR = 10.^(SNR / 10), which is the signal-to-noise ratio; is to ensure that the variance of the noise meets the given signal-to-noise ratio; N npilot is the training sequence length for each user.

[0109] III: Through Bayes' theorem, the posterior distribution also obeys the multivariate Gaussian distribution, expressed as:

[0110] p(X|Y k )=N(X|m,Λ -1 )

[0111] Among them, m represents the mean vector of the posterior distribution, and Λ represents the posterior precision matrix.

[0112]

[0113] IV: Optimize the variational parameters by maximizing the variational lower bound (EvidenceLowerBound, ELBO). Expressed as:

[0114] ELBO(m,Λ)=E q [logp(Y,X)]-E q [logq(X)]

[0115] Among them, q(X) represents the approximate posterior distribution, and p(Y,X) represents the joint distribution, including the likelihood function and the prior distribution.

[0116] The process of continuously optimizing the values ​​of m and Λ is equivalent to iteratively optimizing ELBO through variational inference, so that the approximate posterior distribution q(X) is closer to the true posterior distribution p(X|Y k ).

[0117] S4: Based on the system model of a single RIS-assisted massive MIMO wireless communication network and the optimized objective function, the final channel estimation result is output to evaluate the effectiveness of the algorithm.

[0118] Figure 3 This is a comparison curve of the normalized mean square error of a single RIS-assisted massive MIMO wireless communication network as the signal-to-noise ratio changes during the training process. Figure 3 It can be observed that when the number of RIS reflection elements is 128, the performance of the adaptive compressed sensing algorithm is the best.

[0119] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A channel estimation algorithm for RIS-assisted massive MIMO systems based on variational Bayesian inference, characterized in that , including the following steps: S1: Establish a system model of single RIS-assisted massive MIMO through parameter setting, and determine the number of transmit antennas, receive antennas and RIS reflection elements of the massive MIMO system; S2: Based on the system model of single RIS-assisted massive MIMO, an objective function is established to represent the performance index of channel estimation; S3: Use the adaptive compressed sensing algorithm to preliminarily estimate the channel to obtain the initial estimation result, and use the variational Bayesian method to perform posterior update based on the initial channel estimation result and the received signal to optimize the objective function; S4: Based on the system model of a single RIS-assisted massive MIMO and the optimized objective function, the final channel estimation result is output to evaluate the effectiveness of the algorithm.

2. The channel estimation algorithm for RIS-assisted massive MIMO system based on variational Bayesian inference according to claim 1, characterized in that: The system model of single RIS assisted massive MIMO in S1 is: The total angular response matrix Θ of the downlink transmission channel of the RIS-assisted massive MIMO system, the channel matrix F from the BS end to the RIS end, and the channel matrix H from the RIS end to the USERs end k , the user's channel matrix is ​​denoted as G k , specifically defined as: Θ=e (-jπsin(θ x )*[0:Y-1])T In which, the antenna response vectors from BS to RIS, the transmitting end and the receiving end are respectively x is θ nmAOAi When Y=NL, it represents the angle response matrix AoAF from the transmitter to the reflector. i ∈C NL×Nf When θ x is θ nmAODi When Y = NM, it represents the angle response matrix from the reflector to the receiving end. When θ x for When Y=NL, it represents the angle response matrix from USERs to RIS, NL represents the number of RIS reflection units, NM represents the number of antennas at BS, and N f It indicates the number of channels from BS to RIS. It represents the incident angle from BS to RIS. It represents the angle of emission from RIS to USERs, rand(1,N f ) represents the generation of a f A row vector of random numbers uniformly distributed in the range [0,1); represents the incident angle from USERs to RIS, rand(k,N hk ) represents the generation of a k-row N hk A row vector of random numbers uniformly distributed in the range [0,1). The superscript T represents the transpose operator. Where N(0,1) represents a normal distribution with a mean of 0 and a variance of 1; the superscript H represents the conjugate transpose operator; Among them, N hk It represents the number of channels from the RIS end to the USERs end; N(0,1) represents a normal distribution with a mean of 0 and a variance of 1; G k =diag((H k ) H )*F Among them, diag((H k ) H ) is expressed as (H k ) H A diagonal matrix with diagonal elements; the superscript H represents the conjugate transpose operator.

3. The channel estimation algorithm for RIS-assisted massive MIMO system based on variational Bayesian inference according to claim 1, characterized in that: The objective function in S2 is: in, represents the estimated channel matrix; G represents the theoretical channel matrix; and the superscript H represents the conjugate transpose.

4. The channel estimation algorithm for RIS-assisted massive MIMO system based on variational Bayesian inference according to claim 1, characterized in that: In S3, the adaptive compressed sensing algorithm is used to perform a preliminary estimation of the channel, and the specific process of obtaining the initial estimation result is as follows: Ⅰ: A dictionary matrix Dmatrixo is used for sparse representation, expressed as: x=Dmatrixo·s Among them, x is the original signal, Dmatrixo is the dictionary matrix, and s is the sparse coefficient vector; II: Different from traditional signal sampling, the adaptive compressed sensing algorithm uses random sampling, which is usually described by a measurement matrix, expressed as: z bk =Dmatrixo·x Among them, z bk is the compressed measurement vector, Dmatrixo is the dictionary matrix, and x is the original signal; III: The matching pursuit algorithm is used in the signal sparse reconstruction process. The algorithm first selects the column vector with the highest correlation with the current residual to update the reconstruction set, and iteratively updates the residual until a certain condition is met. The iterative process is expressed as: R k =z bk -Dmatrix·s k Among them, R k is the residual vector of the kth iteration, Dmatrix is ​​the normalized dictionary matrix, s k is the sparse coefficient vector of the kth iteration; IV: During the iteration process, the stop condition of the iteration is controlled by dynamically adjusting the threshold η. The formula for dynamically reducing the threshold is: η=max(η k ·decay_rate,current_residual_energy) Among them, η k is the threshold of the current iteration, decay_rate is the decay rate of the threshold, and current_residual_energy is the current residual energy; The formula for dynamically increasing the threshold is: η=max(η k ·increase_rate,current_residual_energy) Among them, η k is the threshold of the current iteration, decay_rate is the rate of increase of the threshold, and current_residual_energy is the current residual energy.

5. The channel estimation algorithm for RIS-assisted massive MIMO system based on variational Bayesian inference according to claim 2, characterized in that: In S3, the variational Bayesian method is used for posterior updating, and the specific process of optimizing the objective function is as follows: Ⅰ: Assume that the channel coefficient X follows a multivariate Gaussian distribution; Among them, m0 represents the prior mean vector, Λ0 represents the precision matrix of the prior distribution, that is, the inverse of the covariance matrix; II: The above formula Y k As observation data, Y k The likelihood function is expressed as: Among them, Y k represents the observed signal matrix, represents the estimated value of the channel matrix, σ 2 represents the noise variance, and I represents the unit matrix; Among them, Y k represents the received signal matrix of user k; G k Let the channel matrix of user k be the channel matrix F from BS to RIS and the channel matrix H from RIS to USERs k Composition; V represents the reflection coefficient matrix at the RIS end; N k Expressed as the noise matrix, G k =diag((H k ) H )*F, where diag((H k ) H ) is expressed as (H k ) H The elements of are diagonal elements of the diagonal matrix; the superscript H represents the conjugate transpose operator, and the reflection coefficient matrix of RIS is represented by V. Among them, randn(NL,N bpilot ) means to generate a size of NL*N bpilot are independent and identically distributed standard normal random variables; j is represented by an imaginary unit; This is to ensure that the variance of the V element is 1, making it a standard complex Gaussian random variable; the noise matrix in the channel transmission process is in, represents the i-th row of matrix U, N npilot (j-1)+1 to Userpilot represents the variable to which the generated DFT matrix is ​​assigned; SNR = 10.^(SNR / 10), which is the signal-to-noise ratio; is to ensure that the variance of the noise meets the given signal-to-noise ratio; N npilot is the training sequence length of each user; III: Through Bayes' theorem, the posterior distribution also obeys the multivariate Gaussian distribution, expressed as: p(X|Y k )=N(X|m,Λ -1 ) Among them, m represents the mean vector of the posterior distribution, and Λ represents the posterior precision matrix; IV: Optimize the variational parameters by maximizing the variational lower bound (EvidenceLowerBound, ELBO), expressed as: ELBO(m,Λ)=E q [logp(Y,X)]-E q [logq(X)] Among them, q(X) represents the approximate posterior distribution, and p(Y,X) represents the joint distribution, including the likelihood function and the prior distribution; The process of continuously optimizing the values ​​of m and Λ is equivalent to iteratively optimizing ELBO through variational inference, so that the approximate posterior distribution q(X) is closer to the true posterior distribution p(X|Y k ).

Images