Double-panel intelligent metasurface assisted MU-MIMO communication system channel estimation method and system combining BLS algorithm and RECHANATTENTION neural network

Abstract

The invention provides a channel estimation method and a channel estimation system for a double-sided board intelligent super surface assisted MU-MIMO (Multiple User Multiple Input Multiple Output) communication system, and the channel estimation method and the channel estimation system are used for the double-sided board intelligent super surface assisted MU-MIMO communication system, wherein the channel estimation method and the channel estimation system are used for the double-sided board intelligent super surface assisted MU-MIMO communication system, and the double-sided board intelligent super surface assisted MU-MIMO communication system is combined with a bilateral least squares BLS (Bilateral Least Square) algorithm and a RECHANATTENTION neural network. The method comprises the following steps: firstly, acquiring state information of a double-reflection cascade channel through a phase conversion strategy, and processing and pre-estimating the state information of the double-reflection cascade channel by using a decomposition algorithm and a BLS algorithm; and then, the RECHANATTENTION neural network is used to carry out rapid and effective feature extraction on the channel state information pre-estimated by the BLS algorithm, so that more accurate channel state information is obtained. Simulation results show that the channel estimation method combining the BLS algorithm and the RECHANARRENTION neural network can perform feature extraction and fusion on the channel state information more effectively, the accuracy of channel estimation is improved, the calculation time consumption is lower, and the efficiency of channel estimation is improved.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to a channel estimation method and system for a double-panel intelligent metasurface-assisted MU-MIMO communication system combining a BLS algorithm and a RE_CHAN_ATTENTION neural network. Background Art

[0002] In recent years, reconfigurable intelligent surfaces (RIS) have become a key enabling technology for the next generation of wireless communication systems due to their outstanding advantages, including ease of deployment, low power consumption, low latency, and low cost. RIS, by configuring a large number of passive reflective elements, can alter the phase or amplitude of radio waves reflected by it without consuming additional energy. When deployed in a wireless network, RIS can provide additional reflection paths for base station signals, significantly improving the signal-to-noise ratio at the user end and enhancing network communication efficiency. RIS also plays a significant role in expanding network coverage and improving base station energy efficiency. Through extensive theoretical research and experimental verification by numerous scholars, RIS is now considered a highly promising technology for building intelligent, controllable wireless communication environments in the future.

[0003] Building an intelligent and controllable wireless communication environment relies on accurate channel state information. Compared to traditional wireless networks, RIS adds more reflection paths, significantly increasing the difficulty of channel estimation. Existing channel estimation research for RIS-assisted networks is mostly limited to single-RIS-assisted wireless communication systems. Most research on channel estimation for multi-RIS-assisted networks only considers the reflection link introduced by a single RIS. Even when considering the cascaded channel formed by the combined reflections of multiple RISs, current research generally assumes a single antenna at the user end to simplify the channel estimation. This shows that current channel estimation research for RIS-assisted networks has significant shortcomings. In particular, in multi-RIS-assisted wireless network scenarios, the network modeling is overly simplistic, making the proposed methods difficult to adapt to actual wireless environments and resulting in poor channel estimation accuracy. Summary of the Invention

[0004] The present invention is made to solve the above problems, and its purpose is to provide a double-panel intelligent metasurface assisted MU-MIMO communication system channel estimation method and system combining BLS algorithm and RE_CHAN_ATTENTION neural network.

[0005] The present invention provides a dual-panel intelligent metasurface-assisted MU-MIMO communication system channel estimation method combining a BLS algorithm with a RE_CHAN_ATTENTION neural network, which has the following characteristics: S1: a communication model construction step, constructing a dual-RIS-assisted uplink multi-antenna MU-MIMO wireless communication model, wherein there are a base station BS end, two RIS panels and a user equipment UE end, the BS end is equipped with multiple antennas, the number of reflection units of the two RIS panels is the same, and there are multiple multi-antenna users on the UE end. In the uplink multi-antenna MU-MIMO wireless communication, the base station BS end is equipped with multiple antennas, the number of reflection units of the two RIS panels is the same, and the UE end has multiple multi-antenna users. In the communication model, the communication channel from the UE to the BS is a direct channel, the communication channel reflected from the UE to the BS through a RIS panel is a single-reflection cascade channel, and the communication channel reflected from the UE to the BS through two RIS panels is called a double-reflection cascade channel; S2: Channel information extraction step, the UE transmits a pilot signal known to the BS to the BS in multiple time slots, and the pilot signal is received by the BS after passing through the direct channel, single-reflection cascade channel, and double-reflection cascade channel respectively. The received signal is the superposition of the signals passing through the direct channel, single-reflection cascade channel, and double-reflection cascade channel. Channel estimation is used Method, the channel information of the double-reflection cascade channel is extracted; S3: channel information matrix pre-estimation step, using the BLS algorithm to estimate the channel information matrix corresponding to the double-reflection cascade channel, to obtain the pre-estimated channel information matrix; S4: RE_CHAN_ATTENTION neural network model construction step, the pre-estimated channel information matrix and the real channel information matrix obtained by simulation constitute a feature label pair, and the data set composed of multiple feature label pairs obtained through multiple simulations is format-aligned, and the RE_CHAN_ATTENTION neural network is offline trained based on the format-aligned data set, and a RE_CHAN_ATTENTION neural network model capable of estimating the double-reflection cascade channel information matrix and the optimal weights of the corresponding network parameters are obtained by presetting the loss function and performing multiple rounds of optimization iterations; S5: double-reflection cascade channel information matrix estimation step, the pre-estimated channel matrix is ​​used as the input of the RE_CHAN_ATTENTION neural network model, and the RE_CHAN_ATTENTION neural network model and the corresponding optimal weights are used to perform online estimation of the double-reflection cascade channel information matrix, to obtain the double-reflection cascade channel information matrix estimation result.

[0006] The dual-panel intelligent metasurface-assisted MU-MIMO communication system channel estimation method provided by the present invention, which combines the BLS algorithm with the RE_CHAN_ATTENTION neural network, may also have the following features: wherein, in S1, the two RIS panels are RIS1 and RIS2,

[0007] In S2, the received signal uses y kSpecifically, it means:

[0008] y k =BΦ1RΦ2F k x k +H 2k Φ2F k x k +BΦ1H 1k x k +D k x k +n k

[0009] Where, BΦ1RΦ2F k x k is the user signal reflected successively by RIS2 and RIS1, H 2k Φ2F k x k For user signals reflected only by RIS2, BΦ1H 1k x k is the user signal reflected only by RIS1, D k x k is the user signal not reflected by RIS, n k is the noise, here, represents the reflection coefficient matrix of RIS, diag(·) represents the diagonal matrix, Represents the reflection coefficient vector of RIS, each element represents the reflection coefficient of the nth reflection unit of RISi, β i,n ∈[0,1] is the amplitude value, is the phase value. k For UE k The pilot signal sent, is the noise received by the BS, satisfying

[0010] In the double-panel intelligent metasurface assisted MU-MIMO communication system channel estimation method of the combined BLS algorithm and RE_CHAN_ATTENTION neural network provided by the present invention, it can also have the following characteristics: wherein, in S2, the channel estimation method is specifically:

[0011] The phase conversion strategy is used to convert the phase shift of RIS to extract the relevant information of different channels from the received signal. Two phase conversion modes are set, mode 0 and mode 1. Mode 0 is the normal reflection mode, while mode 1 increases the phase value of the normal reflection mode by π. If the reflection coefficient vector of mode 0 is Then the reflection coefficient vector of mode 1 is Right now

[0012] Phase strategy 1: RIS1 is set to mode 0, RIS2 is set to mode 0, and the received signal is:

[0013]

[0014] Phase strategy 2: RIS1 is set to mode 0 and RIS2 is set to mode 1. The base station receives the signal as follows:

[0015]

[0016] Phase strategy 3: RIS1 is set to mode 1 and RIS2 is set to mode 0. The base station receives the signal as follows:

[0017]

[0018] Phase strategy 4: RIS1 is set to mode 1, RIS2 is set to mode 1, and the base station receives the signal:

[0019]

[0020] By adding the received signals of phase strategies 1 and 2, the received signal that only passes through the RIS1 single reflection channel and the direct channel can be obtained:

[0021]

[0022] By adding the received signals of phase strategies 1 and 3, the received signal that passes only through the RIS2 single reflection channel and the direct channel can be obtained:

[0023]

[0024] By adding the received signals of the four phase strategies, the received signal that only passes through the direct channel can be obtained:

[0025]

[0026] Receive signals from three categories {y k1 ,y k2 ,y k3 After obtaining the direct channel and the single reflection cascade channel in}, the received signal combination y is designed. k4 , to extract the channel information of the double-reflection cascade channel, expressed as:

[0027]

[0028] Where B, R, and F k are channels to be estimated.

[0029] In the double-panel intelligent metasurface assisted MU-MIMO communication system channel estimation method provided by the present invention, the BLS algorithm and the RE_CHAN_ATTENTION neural network are combined, and the following features may also be provided: in S3, the BLS algorithm is used to estimate the channel information matrix corresponding to the double-reflection cascade channel, and the method for obtaining the pre-estimated channel information matrix is:

[0030] The channel information corresponding to the double-reflection cascade channel is decomposed by dividing B into blocks. First, the received signal of each antenna at the BS is calculated, and in, At this time, the received signal of the mth antenna at the BS is expressed as follows:

[0031]

[0032] Where y k,m represents the received signal of the mth antenna at the BS end, represents the conjugate transpose of the reflection coefficient vector of RIS1, b m , R, F k are all channels to be estimated, Φ2 represents the reflection coefficient matrix of RIS2, n k is noise, b m , R is merged into And directly estimate R m With F k To estimate the double-reflection cascade channel, it can be expressed as:

[0033]

[0034] Where R m for b m , R after the merger, further, the pilot signal matrix is ​​composed of T pilot signals, and the test phase shift matrix of RIS1 composed of N test phase shift blocks is Set it as DFT matrix and multiply both ends of the equation And do vector processing on the equation,

[0035] K test phase shift blocks are used to form the test phase shift matrix Θ2 of RIS2. Combined with the Khatri-Rao product operation, the channel estimation problem is transformed into a minimization problem. The value of Θ2 is also taken as the DFT matrix. Then vectorization is performed to obtain an unconstrained closed-form solution and the estimated channel information matrix is ​​obtained.

[0036] In the dual-panel intelligent metasurface assisted MU-MIMO communication system channel estimation method of the combined BLS algorithm and RE_CHAN_ATTENTION neural network provided by the present invention, it can also have the following characteristics: wherein, in S4, the preset loss function is defined as:

[0037]

[0038] Among them, Ω is the parameter of RE_CHAN_ATTENTION neural network, Represents input The output of the RE_CHAN_ATTENTION neural network is used to minimize the loss function, and the parameters of the RE_CHAN_ATTENTION neural network are trained through random sampling and gradient back propagation method.

[0039] In the double-panel intelligent metasurface assisted MU-MIMO communication system channel estimation method of the combined BLS algorithm and RE_CHAN_ATTENTION neural network provided by the present invention, it can also have the following characteristics: wherein, in S5, the pre-estimated channel matrix is ​​used as the input of the RE_CHAN_ATTENTION neural network model, specifically:

[0040] The input of the RE_CHAN_ATTENTION neural network model is the pre-estimated channel information matrix after BLS algorithm processing Divide it into two channels, which serve as the input channels of the real part and the imaginary part respectively. The dimensions of both input channels are N 2 L×M.

[0041] In the double-panel intelligent metasurface assisted MU-MIMO communication system channel estimation method of the combined BLS algorithm and RE_CHAN_ATTENTION neural network provided by the present invention, it can also have the following characteristics: wherein, in S5, the method for online estimation of the double-reflection cascade channel information matrix using the RE_CHAN_ATTENTION neural network model and the corresponding optimal weight is:

[0042] The real and imaginary input data first pass through a 3×3 convolution block, and then undergo normalization and activation function processing in sequence. After that, the data will be diverted to the upper and lower layers of convolution blocks of different sizes for processing. The upper layer uses a 3×3 convolution block, and the lower layer uses a 7×7 convolution block. After being processed by their respective convolution blocks, the data will be normalized and activated separately, and then residual connections will be performed. After the residual connection processing, the data of the upper and lower layers will be added. The result of the addition will pass through a 3×3 convolution block again and undergo normalization and activation function processing in sequence. After that, the data will pass through a multi-head attention layer to capture the features and correlation information in the data. Next, a convolution layer is used to process the data to restore it to the same shape as the input data. Finally, this processing result is residually connected with the original input data to obtain the double-reflection cascade channel information matrix estimation result.

[0043] The present invention provides a dual-panel intelligent metasurface-assisted MU-MIMO communication system channel estimation system that combines a BLS algorithm with a RE_CHAN_ATTENTION neural network. The system has the following features: a communication model construction module that constructs a dual-RIS-assisted uplink multi-antenna MU-MIMO wireless communication model, wherein there is a base station (BS) end, two RIS panels, and a user equipment (UE) end. The BS end is equipped with multiple antennas, the two RIS panels have the same number of reflection units, and there are multiple multi-antenna users on the UE end. In the uplink multi-antenna MU-MIMO wireless communication model, the communication channel from the UE end to the BS end is a direct channel, the communication channel reflected from the UE end to the BS end through a RIS panel is a single-reflection cascade channel, and the communication channel from the UE end to the BS end through the joint reflection of two RIS panels is called a double-reflection cascade channel.

[0044] In the channel information extraction module, the UE transmits a pilot signal known to the BS in multiple time slots. The pilot signal is received by the BS after passing through the direct channel, the single-reflection cascade channel, and the dual-reflection cascade channel. The received signal is a superposition of the signals passing through the direct channel, the single-reflection cascade channel, and the dual-reflection cascade channel. The channel information of the dual-reflection cascade channel is extracted using a channel estimation method.

[0045] The channel information matrix pre-estimation module uses the BLS algorithm to estimate the channel information matrix corresponding to the double-reflection cascade channel to obtain a pre-estimated channel information matrix;

[0046] The RE_CHAN_ATTENTION neural network model construction module forms a feature label pair by taking the pre-estimated channel information matrix and the simulation-obtained real channel information matrix as the feature label pair, aligns the data set composed of multiple feature label pairs obtained through multiple simulations in format, performs offline training on the RE_CHAN_ATTENTION neural network based on the data set aligned in format, obtains the optimal weight in the corresponding network parameters of the RE_CHAN_ATTENTION neural network model capable of estimating the double-reflection cascaded channel information matrix through a preset loss function and multiple rounds of optimization iteration, and the like.

[0047] The double-reflection cascaded channel information matrix estimation module takes the pre-estimated channel matrix as the input of the RE_CHAN_ATTENTION neural network model, uses the RE_CHAN_ATTENTION neural network model and the corresponding optimal weight to perform online estimation on the double-reflection cascaded channel information matrix, and obtains the double-reflection cascaded channel information matrix estimation result.

[0048] Effects of the application

[0049] The double-panel intelligent metasurface assisted MU-MIMO communication system channel estimation method and system involving the joint BLS algorithm and RE_CHAN_ATTENTION neural network have the following beneficial effects:

[0050] The application is aimed at the channel estimation problem of a double-RIS assisted MU-MIMO wireless communication system, converts the channel estimation problem into a denoising problem by using the additivity principle of noise, extracts the state information of the double-reflection cascaded channel by designing a reasonable phase conversion strategy, and processes and pre-estimates the extracted channel state information by designing a decomposition algorithm and a BLS algorithm. Then, the RE_CHAN_ATTENTION neural network is designed to quickly and effectively extract features from the channel state information pre-estimated by the BLS algorithm, so as to obtain more accurate channel state information. Compared with existing work, the application can more effectively extract and fuse channel state information, improve the accuracy of channel estimation, and has lower calculation time consumption, thereby improving the efficiency of channel estimation. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 is a schematic diagram of a double-RIS assisted uplink multi-antenna MU-MIMO wireless communication model in an embodiment of the application;

[0052] Figure 2 is a schematic diagram of an RE_CHAN_ATTENTION neural network model in an embodiment of the application;

[0053] Figure 31 is a comparison diagram of channel estimation simulation effects of different models in an embodiment of the present invention;

[0054] Figure 4 3 is a comparison chart of channel estimation calculation speeds of different models in an embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the technical means, creative features, objectives and effects achieved by the present invention easy to understand, the following embodiments, in conjunction with the accompanying drawings, specifically illustrate the channel estimation method and system of the double-panel intelligent metasurface-assisted MU-MIMO communication system of the present invention combined with the BLS algorithm and the RE_CHAN_ATTENTION neural network.

[0056] The channel estimation method of a double-panel intelligent metasurface-assisted multi-user multiple input multiple output (MU-MIMO) communication system using a combined bilateral least squares (BLS) algorithm and a RE_CHAN_ATTENTION neural network in this embodiment specifically includes the following steps:

[0057] Figure 1 FIG. 1 is a schematic diagram of a dual-RIS-assisted uplink multi-antenna MU-MIMO wireless communication model in an embodiment of the present invention.

[0058] Step S1: Communication model construction step. Figure 1 As shown in the figure, a dual-RIS-assisted uplink multi-antenna MU-MIMO wireless communication model is constructed. It consists of a base station (BS), two RIS panels (RIS1 and RIS2), and a user equipment (UE). The BS is equipped with M antennas, and the two RIS panels have the same number of N reflectors. The UE has K multi-antenna users, each equipped with L antennas.

[0059] In the uplink multi-antenna MU-MIMO wireless communication model, the communication channel from the UE to the BS is a direct channel. The communication channel reflected from the UE to the BS via a single RIS panel is a single-reflection cascade channel. The communication channel reflected from the UE to the BS via two RIS panels is a double-reflection cascade channel.

[0060] Specifically, record the kth user UE at the UE end k The transmitted pilot signal is By UE k Direct channel to BS Indicates that UE kDirect channel to RIS1 Indicates that UE k Direct channel to RIS2 Indicates that the direct channel from RIS2 to RIS1 is Indicates that the direct channel from RIS2 to BS is Indicates that the direct channel from RIS1 to BS is express.

[0061] Step S2 is the channel information extraction step. The UE transmits a pilot signal known to the BS in multiple time slots to the BS. The pilot signal is received by the BS after passing through the direct channel, the single-reflection cascade channel, and the dual-reflection cascade channel. The received signal is a superposition of the signals passing through the direct channel, the single-reflection cascade channel, and the dual-reflection cascade channel. Channel information for the dual-reflection cascade channel is extracted using a channel estimation method.

[0062] The BS receives the data from the user UE. k The received signal uses y k Specifically, it means:

[0063] y k =BΦ1RΦ2F k x k +H 2k Φ2F k x k +BΦ1H 1k x k +D k x k +n k

[0064] Where, BΦ1RΦ2F k x k is the user signal reflected successively by RIS2 and RIS1, H 2k Φ2F k x k For user signals reflected only by RIS2, BΦ1H 1k x k is the user signal reflected only by RIS1, D k x k is the user signal not reflected by RIS, n k is the noise, here, represents the reflection coefficient matrix of RIS, diag(·) represents the diagonal matrix, Represents the reflection coefficient vector of RIS, each element represents the reflection coefficient of the nth reflection unit of RISi, β i,n ∈[0,1] is the amplitude value, is the phase value.k For UE k The pilot signal sent, is the noise received by the BS, satisfying

[0065] The channel estimation method is specifically as follows:

[0066] The phase conversion strategy is used to convert the phase shift of RIS to extract the relevant information of different channels from the received signal. Two phase conversion modes are set, mode 0 and mode 1. Mode 0 is the normal reflection mode, while mode 1 increases the phase value of the normal reflection mode by π. If the reflection coefficient vector of mode 0 is Then the reflection coefficient vector of mode 1 is Right now

[0067] Four sets of phase conversion strategies are proposed for obtaining independent channels:

[0068] Phase strategy 1: RIS1 is set to mode 0, RIS2 is set to mode 0, and the received signal is:

[0069]

[0070] Phase strategy 2: RIS1 is set to mode 0 and RIS2 is set to mode 1. The base station receives the signal as follows:

[0071]

[0072] Phase strategy 3: RIS1 is set to mode 1 and RIS2 is set to mode 0. The base station receives the signal as follows:

[0073]

[0074] Phase strategy 4: RIS1 is set to mode 1, RIS2 is set to mode 1, and the base station receives the signal:

[0075]

[0076] By adding the received signals of phase strategies 1 and 2, the received signal that only passes through the RIS1 single reflection channel and the direct channel can be obtained:

[0077]

[0078] By adding the received signals of phase strategies 1 and 3, the received signal that passes only through the RIS2 single reflection channel and the direct channel can be obtained:

[0079]

[0080] By adding the received signals of the four phase strategies, the received signal that only passes through the direct channel can be obtained:

[0081]

[0082] Since there are a lot of studies on channel estimation for single RIS-assisted networks, direct channels and RIS single reflection channels can be directly used to receive signals from the three types of channels using existing methods. k1 ,y k2 ,y k3}, design the receiving signal combination y k4 , to extract the channel information of the double-reflection cascade channel, expressed as:

[0083]

[0084] Where B, R, and F k are channels to be estimated.

[0085] Step S3 is the channel information matrix pre-estimation step. The BLS algorithm is used to estimate the channel information matrix corresponding to the dual-reflection cascade channel, obtaining a pre-estimated channel information matrix. Compared to the traditional Least Square (LS) algorithm, the BLS algorithm offers higher accuracy and is therefore chosen to extract the channel information matrix for the dual-reflection cascade channel.

[0086] The channel information corresponding to the double-reflection cascade channel is decomposed by dividing B into blocks. First, the received signal of each antenna at the BS is calculated, and in, At this time, the received signal of the mth antenna at the BS is expressed as follows:

[0087]

[0088] Where y k,m represents the received signal of the mth antenna at the BS end, represents the conjugate transpose of the reflection coefficient vector of RIS1, b m , R, F k are all channels to be estimated, Φ2 represents the reflection coefficient matrix of RIS2, n k is noise. m , R is merged into And directly estimate R m With F k To estimate the double-reflection cascade channel, it can be expressed as:

[0089]

[0090] Where Rm For b m , the combined matrix. Further, a pilot signal matrix composed of T pilot signals, and a test phase shift matrix of RIS1 composed of N test phase shift blocks is set to a DFT matrix, and the same multiplication is performed on both ends of the equation and the vectorization processing is performed on the equation.

[0091] K test phase shift blocks are used to form a test phase shift matrix Θ2 of RIS2, and the channel estimation problem is converted into a minimization problem by combining the Khatri-Rao product operation. The value of Θ2 is also taken as a DFT matrix, and the vectorization processing is performed again to obtain an unconstrained closed-form solution, and the estimated channel information matrix

[0092] Specifically, considering a pilot signal matrix composed of T pilot signals, and a test phase shift matrix of RIS1 composed of N test phase shift blocks The received signal of the BS can be expressed as:

[0093]

[0094] In the formula, Y k,m is the result of block combination on y k,m , Θ1 is a test phase shift matrix, X k is a pilot signal matrix, N k is the result of block combination on n k . Here, we set Θ1 as a DFT matrix to simplify the subsequent Θ1 inverse operation, which is specifically:

[0095]

[0096] After multiplying the left side of the above received signal on both ends , it can be expressed as:

[0097]

[0098] Wherein, is the matrix multiplied by F k and , is the result of left multiplication operation on Y k,m , is the result of left multiplication operation on N k . The vectorization processing is performed on the above formula:

[0099]

[0100] Where vec(·) is the column vectorization operation, ◇ is the Khatri-Rao product operation, Φ2=diag(θ2) represents the reflection coefficient matrix of RIS2, and θ2 represents the reflection coefficient vector of RIS2. for The result after column vectorization. Similar to RIS1, the test phase shift matrix of RIS2 is formed by using K test phase shift blocks. The above formula can be rewritten as:

[0101]

[0102] in, for The result after block merging, for The result after block merging, Θ2 is the test phase shift matrix of RIS2. Combined with the Khatri-Rao product operation:

[0103]

[0104] in is the Kronecker product of the matrix, and the above formula can be expressed as:

[0105]

[0106] Among them, I N is the identity matrix. The channel estimation problem can be further transformed into the following minimization problem:

[0107]

[0108] Among them, H k,m yes Estimates of F is the F norm of the matrix. By solving the above problem, we can get H k,m The unconstrained closed-form solution of :

[0109]

[0110] in, For The result after processing, is the pseudo-inverse. Similar to Θ1, the value of Θ2 in the above formula can also be taken as the DFT matrix to simplify the subsequent Θ2 inverse operation, which is expressed as follows:

[0111]

[0112] Finally, for H k,m Perform vector processing and pre-estimate the channel based on the number of BS antennas M:

[0113]

[0114] h k,m For H k,m Take the result after column vectorization, for The transpose of For h k,m The above algorithm for solving the channel is called BLS algorithm. Can be regarded as the real channel information matrix H k Superposition with noise:

[0115]

[0116] Among them, N k is the noise information.

[0117] Step S4 constructs the RE_CHAN_ATTENTION neural network model. The pre-estimated channel information matrix and the actual channel information matrix obtained through simulation are combined to form feature label pairs. A dataset consisting of multiple feature label pairs obtained through multiple simulations is format-aligned. Offline training of the RE_CHAN_ATTENTION neural network is performed based on the aligned dataset. By presetting a loss function and performing multiple rounds of optimization iterations, a RE_CHAN_ATTENTION neural network model capable of estimating the dual-reflection cascade channel information matrix is ​​obtained, along with the optimal weights for the corresponding network parameters.

[0118] Specifically, after channel pre-estimation, With H k Constitute a feature label pair, set is a pair of training samples, K t The samples are combined into a data set, which can be expressed as:

[0119]

[0120] The default loss function is defined as:

[0121]

[0122] Among them, Ω is the parameter of RE_CHAN_ATTENTION neural network, Represents input The output of the RE_CHAN_ATTENTION neural network is used to minimize the loss function, and the parameters of the RE_CHAN_ATTENTION neural network are trained through random sampling and gradient back propagation method.

[0123] Step S5 is a dual-reflection cascade channel information matrix estimation step. The pre-estimated channel matrix is ​​used as the input of the RE_CHAN_ATTENTION neural network model. The RE_CHAN_ATTENTION neural network model and the corresponding optimal weights are used to perform online estimation of the dual-reflection cascade channel information matrix to obtain the dual-reflection cascade channel information matrix estimation result.

[0124] The input of the RE_CHAN_ATTENTION neural network model is the pre-estimated channel information matrix after BLS algorithm processing Divide it into two channels, which serve as the input channels of the real part and the imaginary part respectively. The dimensions of both input channels are N 2 L×M.

[0125] The real and imaginary input data first passes through a 3×3 convolutional block, followed by normalization and activation. The data is then split into two layers of convolutional blocks of different sizes for processing. The upper layer uses 3×3 convolutional blocks, while the lower layer uses 7×7 convolutional blocks. After processing by each convolutional block, the data is normalized and activated, followed by a residual connection. After the residual connection, the data from the upper and lower layers are added together. The result of the addition passes through a 3×3 convolutional block again, and then undergoes normalization and activation. The data then passes through a multi-head attention layer to capture complex features and correlations in the data. Next, a convolutional layer processes the data, restoring it to the same shape as the input data. Finally, a residual connection is performed with the original input data to further optimize network performance, ensuring that the network can more accurately learn and process channel information, resulting in a double-reflection cascade channel information matrix estimate.

[0126] This embodiment verifies the accuracy and speed of the method proposed in the present invention through simulation.

[0127] The simulation parameters for training are set as follows: the number of antennas for both the BS and UE users is set to M = L = 2, the number of reflection units N of RIS is set to 32, and all channel fading in the double-reflection cascade channel is assumed to be Ricean fading. For example, channel B can be expressed as Where α is the Rice fading factor, B LoS is a line-of-sight link, B NLoSIt is a non-line-of-sight link. The distance from the UE to RIS1 and the distance from RIS1 to the BS are both set to 16 meters, the distance from RIS1 to RIS2 is set to 80 meters, the distance from RIS2 to the BS and the distance from the UE to RIS1 are both set to 90 meters, and the distance from the UE to the BS is set to 100 meters. The UE side contains K=10 UEs, each UE measures 1500 data, of which 1000 are used to train the neural network and 500 are used for verification. Therefore, the total number of samples in the data set is 15000. The learning rate of the ADAM optimizer is set to 0.001, and the batch size of the training is set to 2. The optimal weights of the neural network are obtained after training

[0128] Figure 3 3 is a comparison chart of channel estimation simulation effects of different models in an embodiment of the present invention.

[0129] The baseline algorithm also uses the BLS algorithm as the pre-estimation algorithm, but uses the RE_CHAN neural network and SC_ATTENTION neural network in the refinement estimation stage. When the signal-to-noise ratio at the BS is changed, the normalized minimum mean square error (NMSE) of the channel estimation changes with the signal-to-noise ratio. Figure 3 As shown in FIG, the channel estimation accuracy of the RE_CHAN_ATTENTION neural network channel estimation method proposed in the present invention is much higher than that of the RE_CHAN neural network under low SNR conditions, and is comparable to that of the SC_ATTENTION neural network.

[0130] Figure 4 3 is a comparison chart of channel estimation calculation speeds of different models in an embodiment of the present invention.

[0131] The single-sample inference time of channel estimation is as follows Figure 4 As shown in Figure 2, the RE_CHAN_ATTENTION neural network has a significant advantage in inference speed compared to the SC_ATTENTION neural network, and is only slightly inferior to the RE_CHAN neural network. Simulation experiments show that the channel estimation method proposed in this paper significantly improves the channel estimation accuracy and speed of dual-RIS-assisted MU-MIMO wireless communications.

[0132] This embodiment also provides a dual-panel intelligent metasurface-assisted MU-MIMO communication system channel estimation system combining a BLS algorithm and a RE_CHAN_ATTENTION neural network, including:

[0133] A communication model construction module is used to implement step S1. A dual-RIS-assisted uplink multi-antenna MU-MIMO wireless communication model is constructed, which includes a base station (BS), two RIS panels, and a user (UE). The BS is equipped with multiple antennas, the two RIS panels have the same number of reflection units, and the UE has multiple multi-antenna users. In the uplink multi-antenna MU-MIMO wireless communication model, the communication channel from the UE to the BS is a direct channel, the communication channel reflected from the UE to the BS via a single RIS panel is a single-reflection cascade channel, and the communication channel from the UE to the BS via two RIS panels is a dual-reflection cascade channel.

[0134] The channel information extraction module is configured to implement step S2. The UE transmits a pilot signal known to the BS in multiple time slots. The pilot signal is received by the BS after passing through the direct channel, the single-reflection cascade channel, and the dual-reflection cascade channel. The received signal is a superposition of the signals passing through the direct channel, the single-reflection cascade channel, and the dual-reflection cascade channel. Channel information for the dual-reflection cascade channel is extracted using a channel estimation method.

[0135] The channel information matrix pre-estimation module is used to implement step S3 and use the BLS algorithm to estimate the channel information matrix corresponding to the double-reflection cascade channel to obtain a pre-estimated channel information matrix.

[0136] The RE_CHAN_ATTENTION neural network model construction module is used to implement step S4. Feature label pairs are formed from the pre-estimated channel information matrix and the actual channel information matrix obtained through simulation. The dataset consisting of multiple feature label pairs obtained through multiple simulations is format-aligned. Offline training of the RE_CHAN_ATTENTION neural network is performed based on the format-aligned dataset. By using a preset loss function and performing multiple rounds of optimization iterations, a RE_CHAN_ATTENTION neural network model capable of estimating the dual-reflection cascade channel information matrix and the optimal weights of the corresponding network parameters are obtained.

[0137] The dual-reflection cascade channel information matrix estimation module is used to implement step S5. The pre-estimated channel matrix is ​​used as the input of the RE_CHAN_ATTENTION neural network model. The RE_CHAN_ATTENTION neural network model and the corresponding optimal weights are used to perform online estimation of the dual-reflection cascade channel information matrix to obtain the dual-reflection cascade channel information matrix estimation result.

[0138] Functions and Effects of the Embodiments

[0139] The double-panel intelligent metasurface-assisted MU-MIMO communication system channel estimation method and system involving the combined BLS algorithm and RE_CHAN_ATTENTION neural network of the present invention have the following beneficial effects:

[0140] The present invention addresses the channel estimation problem of a dual-RIS-assisted MU-MIMO wireless communication system. The method utilizes the additivity principle of noise to transform the channel estimation problem into a denoising problem. The method extracts the state information of the dual-reflection cascade channel by designing a reasonable phase conversion strategy, and processes and pre-estimates the extracted channel state information by designing a decomposition algorithm and a BLS algorithm. Then, a RE_CHAN_ATTENTION neural network is designed to quickly and effectively extract features from the channel state information pre-estimated by the BLS algorithm, thereby obtaining more accurate channel state information. Compared with existing work, the present invention can more effectively extract and fuse features from channel state information, thereby improving the accuracy of channel estimation. Furthermore, the method also reduces computational time and improves the efficiency of channel estimation.

[0141] Those skilled in the art will appreciate that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A dual-panel intelligent metasurface-assisted MU-MIMO communication system channel estimation method combining the BLS algorithm and the RE_CHAN_ATTENTION neural network, characterized in that: The specific steps include: S1: a communication model construction step, constructing an uplink multi-antenna MU-MIMO wireless communication model assisted by a dual-RIS, wherein there is a base station BS, two RIS panels, and a user equipment UE. The BS is equipped with multiple antennas, and the two RIS panels have the same number of reflection units. There are multiple multi-antenna users on the UE. In the uplink multi-antenna MU-MIMO wireless communication model, the communication channel from the UE to the BS is a direct channel, the communication channel reflected from the UE to the BS through one RIS panel is a single-reflection cascade channel, and the communication channel from the UE to the BS through the joint reflection of two RIS panels is called a double-reflection cascade channel. S2: a channel information extraction step, wherein the UE transmits a pilot signal known to the BS to the BS in multiple time slots. The pilot signal is received by the BS after passing through the direct channel, the single-reflection cascade channel, and the dual-reflection cascade channel respectively. The received signal is a superposition of the signals passing through the direct channel, the single-reflection cascade channel, and the dual-reflection cascade channel. Channel information of the dual-reflection cascade channel is extracted using a channel estimation method. S3: a channel information matrix pre-estimation step, using a BLS algorithm to estimate the channel information matrix corresponding to the double-reflection cascade channel to obtain a pre-estimated channel information matrix; S4: A RE_CHAN_ATTENTION neural network model construction step, wherein the pre-estimated channel information matrix and the real channel information matrix obtained by simulation constitute a feature label pair, and a data set consisting of a plurality of feature label pairs obtained through multiple simulations is format-aligned, and the RE_CHAN_ATTENTION neural network is offline trained based on the format-aligned data set. By presetting a loss function and performing multiple rounds of optimization iterations, a RE_CHAN_ATTENTION neural network model capable of estimating the double-reflection cascade channel information matrix and the optimal weights of the corresponding network parameters are obtained; S5: Dual-reflection cascade channel information matrix estimation step, taking the pre-estimated channel matrix as the input of the RE_CHAN_ATTENTION neural network model, using the RE_CHAN_ATTENTION neural network model and the corresponding optimal weights to perform online estimation of the dual-reflection cascade channel information matrix to obtain a dual-reflection cascade channel information matrix estimation result.

2. The dual-panel intelligent metasurface-assisted MU-MIMO communication system channel estimation method combining the BLS algorithm and the RE_CHAN_ATTENTION neural network according to claim 1 is characterized by: in, In S1, the two RIS panels are RIS1 and RIS2. In S2, the received signal uses y k Specifically, it means: y k =BΦ1RΦ2F k x k +H 2k Φ2F k x k +BΦ1H 1k x k +D k x k +n k Where, BΦ1RΦ2F k x k is the user signal reflected successively by RIS2 and RIS1, H 2k Φ2F k x k For user signals reflected only by RIS2, BΦ1H 1k x k is the user signal reflected only by RIS1, D k x k is the user signal not reflected by RIS, n k is the noise, here, i∈{1,2} represents the reflection coefficient matrix of RIS, diag(·) represents the diagonal matrix, Represents the reflection coefficient vector of RIS, each element represents the reflection coefficient of the nth reflection unit of RISi, β i,n ∈[0,1] is the amplitude value, is the phase value. k For UE k The pilot signal sent, is the noise received by the BS, satisfying 3. The dual-panel intelligent metasurface-assisted MU-MIMO communication system channel estimation method combining the BLS algorithm and the RE_CHAN_ATTENTION neural network according to claim 2 is characterized by: in, In S2, the channel estimation method is specifically as follows: The phase conversion strategy is used to convert the phase shift of RIS to extract the relevant information of different channels from the received signal. Two phase conversion modes are set, mode 0 and mode 1. Mode 0 is the normal reflection mode, while mode 1 increases the phase value of the normal reflection mode by π. If the reflection coefficient vector of mode 0 is Then the reflection coefficient vector of mode 1 is Right now Phase strategy 1: RIS1 is set to mode 0, RIS2 is set to mode 0, and the received signal is: Phase strategy 2: RIS1 is set to mode 0 and RIS2 is set to mode 1. The base station receives the signal as follows: Phase strategy 3: RIS1 is set to mode 1 and RIS2 is set to mode 0. The base station receives the signal as follows: Phase strategy 4: RIS1 is set to mode 1, RIS2 is set to mode 1, and the base station receives the signal: By adding the received signals of phase strategies 1 and 2, the received signal that only passes through the RIS1 single reflection channel and the direct channel can be obtained: By adding the received signals of phase strategies 1 and 3, the received signal that passes only through the RIS2 single reflection channel and the direct channel can be obtained: By adding the received signals of the four phase strategies, the received signal that only passes through the direct channel can be obtained: Receive signals from three categories {y k1 ,y k2 ,y k3 After obtaining the direct channel and the single reflection cascade channel in}, the received signal combination y is designed. k4 , to extract the channel information of the double-reflection cascade channel, expressed as: Where B, R, and F k are channels to be estimated.

4. The dual-panel intelligent metasurface-assisted MU-MIMO communication system channel estimation method combining the BLS algorithm and the RE_CHAN_ATTENTION neural network according to claim 3 is characterized by: in, In S3, the channel information matrix corresponding to the dual-reflection cascade channel is estimated using the BLS algorithm to obtain the pre-estimated channel information matrix: The channel information corresponding to the double-reflection cascade channel is decomposed by dividing B into blocks. First, the received signal of each antenna at the BS is calculated, and in, At this time, the received signal of the mth antenna at the BS is expressed as follows: Where y k,m represents the received signal of the mth antenna at the BS end, represents the conjugate transpose of the reflection coefficient vector of RIS1, b m , R, F k are all channels to be estimated, Φ2 represents the reflection coefficient matrix of RIS2, n k is noise, b m , R is merged into And directly estimate R m With F k To estimate the double-reflection cascade channel, it can be expressed as: Where R m for b m , R after the merger, further, the pilot signal matrix is ​​composed of T pilot signals, and the test phase shift matrix of RIS1 composed of N test phase shift blocks is Set it as DFT matrix and multiply both ends of the equation And do vector processing on the equation, K test phase shift blocks are used to form the test phase shift matrix Θ2 of RIS2. Combined with the Khatri-Rao product operation, the channel estimation problem is transformed into a minimization problem. The value of Θ2 is also taken as the DFT matrix. Then vectorization is performed to obtain an unconstrained closed-form solution and the estimated channel information matrix is ​​obtained.

5. The dual-panel intelligent metasurface-assisted MU-MIMO communication system channel estimation method combining the BLS algorithm and the RE_CHAN_ATTENTION neural network according to claim 1 is characterized by: in, In S4, the preset loss function is defined as: Among them, Ω is the parameter of RE_CHAN_ATTENTION neural network, Represents input The output of the RE_CHAN_ATTENTION neural network is used to minimize the loss function, and the parameters of the RE_CHAN_ATTENTION neural network are trained through random sampling and gradient back propagation method.

6. The dual-panel intelligent metasurface-assisted MU-MIMO communication system channel estimation method combining the BLS algorithm and the RE_CHAN_ATTENTION neural network according to claim 4 is characterized by: in, In S5, the pre-estimated channel matrix is ​​used as the input of the RE_CHAN_ATTENTION neural network model, specifically: The input of the RE_CHAN_ATTENTION neural network model is the pre-estimated channel information matrix after BLS algorithm processing Divide it into two channels, which serve as the input channels of the real part and the imaginary part respectively. The dimensions of both input channels are N 2 L×M.

7. The dual-panel intelligent metasurface-assisted MU-MIMO communication system channel estimation method combining the BLS algorithm and the RE_CHAN_ATTENTION neural network according to claim 6 is characterized by: in, In S5, the method for online estimation of the double-reflection cascade channel information matrix using the RE_CHAN_ATTENTION neural network model and the corresponding optimal weight is: The real and imaginary input data first pass through a 3×3 convolution block, and then undergo normalization and activation function processing in sequence. After that, the data will be diverted to the upper and lower layers of convolution blocks of different sizes for processing. The upper layer uses a 3×3 convolution block, and the lower layer uses a 7×7 convolution block. After being processed by their respective convolution blocks, the data will be normalized and activated separately, and then residual connections will be performed. After the residual connection processing, the data of the upper and lower layers will be added. The result of the addition will pass through a 3×3 convolution block again and undergo normalization and activation function processing in sequence. After that, the data will pass through a multi-head attention layer to capture the features and correlation information in the data. Next, a convolution layer is used to process the data to restore it to the same shape as the input data. Finally, this processing result is residually connected with the original input data to obtain the double-reflection cascade channel information matrix estimation result.

8. A dual-panel intelligent metasurface-assisted MU-MIMO communication system channel estimation system combining the BLS algorithm and the RE_CHAN_ATTENTION neural network, characterized by: include: A communication model construction module is provided for constructing an uplink multi-antenna MU-MIMO wireless communication model assisted by a dual-RIS, wherein there is a base station (BS), two RIS panels, and a user equipment (UE). The BS is equipped with multiple antennas, and the two RIS panels have the same number of reflection units. The UE has multiple multi-antenna users. In the uplink multi-antenna MU-MIMO wireless communication model, the communication channel from the UE to the BS is a direct channel, the communication channel reflected from the UE to the BS through one RIS panel is a single-reflection cascade channel, and the communication channel from the UE to the BS through the joint reflection of two RIS panels is a dual-reflection cascade channel. a channel information extraction module, wherein the UE transmits a pilot signal known to the BS in multiple time slots, the pilot signal being received by the BS after passing through the direct channel, the single-reflection cascade channel, and the dual-reflection cascade channel, respectively; the received signal is a superposition of signals passing through the direct channel, the single-reflection cascade channel, and the dual-reflection cascade channel; and the channel information of the dual-reflection cascade channel is extracted using a channel estimation method; A channel information matrix pre-estimation module estimates the channel information matrix corresponding to the double-reflection cascade channel using a BLS algorithm to obtain a pre-estimated channel information matrix; A RE_CHAN_ATTENTION neural network model construction module, which forms a feature label pair from the pre-estimated channel information matrix and the real channel information matrix obtained by simulation, and performs format alignment on a data set consisting of multiple feature label pairs obtained through multiple simulations. The RE_CHAN_ATTENTION neural network is trained offline based on the format-aligned data set. By presetting a loss function and performing multiple rounds of optimization iterations, a RE_CHAN_ATTENTION neural network model capable of estimating the dual-reflection cascade channel information matrix and the optimal weights of the corresponding network parameters are obtained; The dual-reflection cascade channel information matrix estimation module uses the pre-estimated channel matrix as the input of the RE_CHAN_ATTENTION neural network model, uses the RE_CHAN_ATTENTION neural network model and the corresponding optimal weight to perform online estimation of the dual-reflection cascade channel information matrix to obtain a dual-reflection cascade channel information matrix estimation result.

Images