Channel estimation method and device of MIMO system, electronic equipment and storage medium

Abstract

The application provides a channel estimation method and device of a MIMO system, electronic equipment and a storage medium, and relates to the technical field of wireless communication. The method comprises: inputting a quantized received signal of a receiving end of the MIMO system into a channel estimation model to obtain a predicted channel of the quantized received signal output by the channel estimation model; wherein the channel estimation model is used for denoising and correcting the quantized received signal to obtain a preprocessed quantized received signal; and iteratively calculating the predicted channel according to the preprocessed quantized received signal, an initial virtual angle basis matrix of a transmitting end, an initial virtual angle basis matrix of a receiving end, an initial residual error, an initial channel estimation value and an initial loss value. The application can effectively suppress nonlinear quantization distortion and improve the accuracy of channel estimation by iteratively solving the predicted channel. The statistical characteristics contained in the quantized received signal can be fully tapped by denoising and correcting the quantized received signal, and the quantization error can be eliminated.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a channel estimation method, apparatus, electronic device, and storage medium for a MIMO system. Background Technology

[0002] To meet the demands of ultra-high data rates and massive connectivity, next-generation wireless communication systems generally consider the integration of millimeter wave technology with massive MIMO (Multiple-Input Multiple-Output) technology as a core solution. The abundant spectrum resources of the millimeter wave band, combined with the high beamforming gain of large-scale antenna arrays, can significantly improve system performance. However, the surge in the number of antennas also leads to a sharp increase in the number of RF links, resulting in a dramatic increase in hardware costs and power consumption. Among these, the analog-to-digital converter (ADC) is the main source of power consumption; therefore, employing a low-resolution quantization ADC has become a key technical path to reduce system power consumption and cost. However, this introduces severe nonlinear quantization noise, leading to signal quality degradation and reduced accuracy in subsequent channel estimation.

[0003] Channel estimation in existing low-resolution ADC scenarios includes data-driven methods and model-driven methods. Data-driven methods typically utilize deep learning models (such as convolutional neural networks, autoencoders, etc.) to learn the complex nonlinear mapping relationship between the received quantized signal and the channel matrix.

[0004] This data-driven approach cannot meet the channel estimation accuracy requirements in low-resolution ADC scenarios and is difficult to meet the practical application requirements of millimeter-wave large-scale MIMO systems. Summary of the Invention

[0005] This invention provides a channel estimation method, apparatus, electronic device, and storage medium for a MIMO system, which addresses the shortcomings of low channel estimation accuracy in low-resolution ADC scenarios in the prior art and improves the channel estimation accuracy in low-resolution ADC scenarios.

[0006] This invention provides a channel estimation method for a MIMO system, comprising: The quantized received signal from the receiver of the MIMO system is input into the channel estimation model to obtain the predicted channel of the quantized received signal output by the channel estimation model. The channel estimation model includes a data-driven quantization error correction network and a learnable orthogonal matching pursuit network. The learnable orthogonal matching pursuit network consists of interconnected multilayers. The data-driven quantization error correction network is used to denoise and correct the quantized received signal to obtain a preprocessed quantized received signal; The current layer of the learnable orthogonal matching pursuit network is used to calculate the current layer's transmitter virtual angle basis matrix, current layer's loss value, current layer's residual, and current layer's channel estimate based on the preprocessed quantized received signal, the transmitter virtual angle basis matrix of the previous layer, the receiver virtual angle basis matrix of the previous layer, the loss value of the previous layer, the residual of the previous layer, and the channel estimate of the previous layer. The current layer is used as the previous layer to iteratively calculate the current layer's channel estimate. The channel estimate of the last layer is used as the predicted channel. The transmitter virtual angle basis matrix of the first layer is the initial transmitter virtual angle basis matrix, the receiver virtual angle basis matrix of the first layer is the initial receiver virtual angle basis matrix, the residual of the first layer is the initial residual, the channel estimate of the first layer is the initial channel estimate, and the loss of the first layer is the initial loss.

[0007] According to the channel estimation method for MIMO systems provided by the present invention, the channel estimation model is used for: For the current layer, the virtual angle basis matrix of the transmitter and the virtual angle basis matrix of the receiver of the previous layer are determined based on the virtual angle basis matrix of the transmitter of the previous layer, the virtual angle basis matrix of the receiver of the previous layer, and the loss value of the previous layer. Based on the virtual angle basis matrix of the transmitter and the virtual angle basis matrix of the receiver of the current layer, determine the sensing matrix of the current layer; Based on the residual of the previous layer, the perception matrix of the current layer, and the perception submatrix of the previous layer, the perception submatrix of the current layer is determined; the perception submatrix of the first layer is the initial perception submatrix. Based on the current layer's sensing submatrix, the channel estimate of the previous layer, and the preprocessed quantized received signal, determine the current layer's channel estimate and the current layer's residual; The loss value of the current layer is determined based on the error between the channel estimate of the current layer and the channel estimate of the previous layer.

[0008] According to the channel estimation method for MIMO systems provided by this invention, the current layer's sensing sub-matrix is ​​determined based on the residual of the previous layer, the current layer's sensing matrix, and the previous layer's sensing sub-matrix, including: Calculate the correlation between the residual of the previous layer and each column element in the perception matrix of the current layer, and take the column element with the strongest correlation as the perception column element of the current layer. Add the elements of the current layer's perceptual column to the perceptual submatrix of the previous layer to obtain the current layer's perceptual submatrix.

[0009] According to the channel estimation method for a MIMO system provided by the present invention, based on the current layer's sensing submatrix, the channel estimation value of the previous layer, and the preprocessed quantized received signal, the method determines the current layer's channel estimation value and the current layer's residual, including: The predicted received signal of the current layer is determined by multiplying the current layer's sensing sub-matrix with the channel estimate of the previous layer. Obtain the difference between the preprocessed quantized received signal and the predicted received signal of the current layer; perform least squares optimization on the difference to obtain the channel estimate of the current layer; The residual of the current layer is determined based on the channel estimate of the current layer, the sensing sub-matrix of the current layer, and the preprocessed quantized received signal.

[0010] According to the channel estimation method for the MIMO system provided by the present invention, the sensing matrix of the current layer is determined based on the virtual angle basis matrix of the transmitter and the virtual angle basis matrix of the receiver of the current layer, including: Based on the baseband precoding matrix and analog precoding matrix of the transmitter, determine the hybrid merging matrix of the transmitter; The sensing matrix of the current layer is determined based on the transmitter virtual angle basis matrix, the transmitter hybrid merging matrix, the transmitted symbol vector, the receiver analog merging matrix, and the receiver virtual angle basis matrix of the current layer.

[0011] According to the channel estimation method for MIMO systems provided by this invention, the data-driven quantization error correction network is used for: The quantized received signal is subjected to convolutional residual denoising to obtain the denoised quantized received signal. Fully connected correction is performed on the denoised quantized received signal to obtain the preprocessed quantized received signal.

[0012] According to the channel estimation method for MIMO systems provided by this invention, the channel estimation model is trained in the following manner: Obtain the sample quantized received signal from the receiver of the MIMO system; The sample quantized received signal is labeled based on the undistorted sample quantized received signal and the real virtual channel vector to obtain labeled training samples; The labeled training samples are input into the preset model for training, and the preprocessed sample quantized received signal and sample prediction channel output by the preset model are obtained; the preset model consists of an initial quantization error correction network and an initial orthogonal matching pursuit network. Calculate the first loss value of the preset model based on the preprocessed sample quantized received signal and the sample quantized received signal without quantization distortion. Calculate the second loss value of the preset model based on the sample predicted channel and the real virtual channel vector; The initial quantization error correction network is optimized based on the first loss value, and the initial orthogonal matching pursuit network is optimized based on the second loss value to obtain the channel estimation model.

[0013] The present invention also provides a channel estimation apparatus for a MIMO system, comprising: The channel estimation module is used to input the quantized received signal from the receiver of the MIMO system into the channel estimation model and obtain the predicted channel of the quantized received signal output by the channel estimation model. The channel estimation model includes a data-driven quantization error correction network and a learnable orthogonal matching pursuit network. The learnable orthogonal matching pursuit network consists of interconnected multilayers. The data-driven quantization error correction network is used to denoise and correct the quantized received signal to obtain a preprocessed quantized received signal; The current layer of the learnable orthogonal matching pursuit network is used to calculate the current layer's transmitter virtual angle basis matrix, current layer's loss value, current layer's residual, and current layer's channel estimate based on the preprocessed quantized received signal, the transmitter virtual angle basis matrix of the previous layer, the receiver virtual angle basis matrix of the previous layer, the loss value of the previous layer, the residual of the previous layer, and the channel estimate of the previous layer. The current layer is used as the previous layer to iteratively calculate the current layer's channel estimate. The channel estimate of the last layer is used as the predicted channel. The transmitter virtual angle basis matrix of the first layer is the initial transmitter virtual angle basis matrix, the receiver virtual angle basis matrix of the first layer is the initial receiver virtual angle basis matrix, the residual of the first layer is the initial residual, the channel estimate of the first layer is the initial channel estimate, and the loss of the first layer is the initial loss.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a channel estimation method for any of the MIMO systems described above.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a channel estimation method for any of the MIMO systems described above.

[0016] The present invention provides a channel estimation method, apparatus, electronic device, and storage medium for MIMO systems. By inputting the quantized received signal from the receiver of the MIMO system into a channel estimation model, the predicted channel of the quantized received signal output by the channel estimation model is obtained. The channel estimation model includes a data-driven quantization error correction network and a learnable orthogonal matching pursuit network. The learnable orthogonal matching pursuit network includes interconnected multiple layers. The data-driven quantization error correction network is used to denoise and correct the quantized received signal to obtain a preprocessed quantized received signal. The current layer of the learnable orthogonal matching pursuit network is used to determine the quantized received signal based on the preprocessed quantized received signal, the transmitter virtual angle basis matrix of the previous layer, and the receiver... This invention uses the preprocessed quantized received signal, the previous layer's transmitter virtual angle basis matrix, the previous layer's loss value, the previous layer's residual, and the previous layer's channel estimate to calculate the current layer's transmitter virtual angle basis matrix, the current layer's receiver virtual angle basis matrix, the current layer's loss value, the current layer's residual, and the current layer's channel estimate. The current layer is then used as the previous layer to iteratively calculate the current layer's channel estimate. The channel estimate of the last layer is used as the predicted channel. The transmitter virtual angle basis matrix of the first layer is the initial transmitter virtual angle basis matrix, the receiver virtual angle basis matrix of the first layer is the initial receiver virtual angle basis matrix, the first layer's residual is the initial residual, the first layer's channel estimate is the initial channel estimate, and the first layer's loss value is the initial loss value. This invention uses the preprocessed quantized received signal, the previous layer's transmitter virtual angle basis matrix, the previous layer's receiver virtual angle basis matrix, the previous layer's loss value, the previous layer's residual, and the previous layer's channel estimate to iteratively solve for the current layer's residual and channel estimate, thereby obtaining the predicted channel. This effectively suppresses nonlinear quantization distortion and improves the accuracy of channel estimation. This invention, by denoising and correcting the quantized received signal, can fully exploit the statistical features contained in the quantized received signal and eliminate quantization errors. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is one of the flowcharts illustrating the channel estimation method for the MIMO system provided by this invention.

[0019] Figure 2 This is a schematic diagram of the process for acquiring quantized received signals provided by the present invention.

[0020] Figure 3This is the second flowchart illustrating the channel estimation method for the MIMO system provided by this invention.

[0021] Figure 4 This is a schematic diagram of the process for calculating the channel estimate and residual of the current layer provided by the present invention.

[0022] Figure 5 This is a performance comparison chart of several channel estimation methods provided by this invention under different signal-to-noise ratios and ADC resolutions.

[0023] Figure 6 This is a performance comparison chart of several channel estimation methods provided by this invention under different numbers of receiving antennas and ADC resolutions.

[0024] Figure 7 This is a schematic diagram of the channel estimation device for the MIMO system provided by the present invention.

[0025] Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0027] The following is combined with Figures 1 to 8 The present invention describes a channel estimation method, apparatus, and electronic device for a MIMO system.

[0028] Figure 1 This is one of the flowcharts illustrating the channel estimation method for the MIMO system provided by this invention, such as... Figure 1 As shown, the channel estimation method for a MIMO system includes step S100.

[0029] S100: Input the quantized received signal from the receiver of the MIMO system into the channel estimation model to obtain the predicted channel of the quantized received signal output by the channel estimation model.

[0030] The channel estimation model includes a data-driven quantization error correction network and a learnable orthogonal matching pursuit network. The learnable orthogonal matching pursuit network consists of interconnected multilayers. The data-driven quantization error correction network is used to denoise and correct the quantized received signal to obtain a preprocessed quantized received signal; The current layer of the learnable orthogonal matching pursuit network is used to calculate the current layer's transmitter virtual angle basis matrix, current layer's loss value, current layer's residual, and current layer's channel estimate based on the preprocessed quantized received signal, the transmitter virtual angle basis matrix of the previous layer, the receiver virtual angle basis matrix of the previous layer, the loss value of the previous layer, the residual of the previous layer, and the channel estimate of the previous layer. The current layer is used as the previous layer to iteratively calculate the current layer's channel estimate. The channel estimate of the last layer is used as the predicted channel. The transmitter virtual angle basis matrix of the first layer is the initial transmitter virtual angle basis matrix, the receiver virtual angle basis matrix of the first layer is the initial receiver virtual angle basis matrix, the residual of the first layer is the initial residual, the channel estimate of the first layer is the initial channel estimate, and the loss of the first layer is the initial loss.

[0031] like Figure 2 As shown, a MIMO system includes a transmitter, a receiver, and a wireless channel. The transmitter is equipped with multiple antennas, multiple radio frequency links, and multiple independent data streams. Optionally, the transmitter is configured with... Root antenna, One radio frequency link, supporting Independent data stream transmission is required. Simultaneously, the constraints must be met. The transmitter has a built-in baseband precoding matrix. (For example, a 6×4 dimension) and analog precoding matrix (For example, with dimensions of 64×6), emission symbol vector After precoding with the baseband precoding matrix and the analog precoding matrix, through... The antenna array transmits signals in a high-gain, narrow-beam configuration to achieve directional signal transmission. The transmitted symbol vector satisfies the power normalization condition. ,in, For expectation operator, for The identity matrix.

[0032] The wireless channel uses the Saleh-Valenzuela (SV) model commonly used in millimeter-wave channels, with the number of multipaths set to L=2. The angle of arrival (AoA) and departure angle (AoD) of each path are randomly distributed in the interval [0, 2π]. The antenna element spacing is d=λ / 2 (λ is the signal wavelength), and the channel matrix H has dimensions of . (in, This refers to the number of antennas at the receiving end. (where the number of transmitting antennas is ), exhibiting significant angular sparsity.

[0033] The receiver is configured with multiple antennas and a multi-day radio frequency link. Optionally, the receiver can be configured with... Root antenna, One radio frequency link, satisfying the constraints. The receiving end contains an analog merging matrix. (For example, a 64×6 dimension) Baseband merging matrix (For example, a 6×4 dimension) and a low-resolution analog-to-digital converter (ADC). The quantization resolution of the ADC is set to 1-3 bits (to adapt to different energy efficiency requirements). After the received signal is analog-combined and baseband-combined, it is quantized into a digital signal by the ADC for subsequent channel estimation.

[0034] At the receiving end, the signal passes through an analog combining matrix. and baseband merging matrix After processing, the final received signal is obtained. .

[0035] ; in, The received signal at the receiving end. This refers to the number of antennas at the receiving end. This refers to the number of antennas at the transmitting end. The length of the emitted symbol vector. This is the baseband precoding matrix. To simulate the precoding matrix, for A dimensional channel matrix, For a mixed merge matrix, To simulate the merging matrix, This is the baseband merging matrix. For hybrid precoding matrices, For the emission symbol vector, This is the noise vector.

[0036] The final received signal is quantized by a low-resolution ADC to obtain the quantized received signal. The formula for calculating the quantized received signal is as follows.

[0037] ; in, To quantize the received signal, This is the baseband merging matrix. and Let these represent the Inverse Discrete Fourier Transform (IDFT) matrix and the Discrete Fourier Transform (DFT) matrix, respectively. To simulate the merging matrix, For a mixed merge matrix, For the emission symbol vector, For noise vectors, For the channel matrix, It is a uniformly ascending quantization function.

[0038] The quantized received signal of the present invention is obtained by performing low-resolution ADC quantization processing on the actual received signal at the receiving end.

[0039] The quantized received signal is input into the channel estimation model. The channel estimation model includes a Data-Driven Network for Quantization Error Correction (DDNet) and a Learnable Orthogonal Matching Pursuit Network (LOMP). DDNet denoises and corrects the quantized received signal to obtain a preprocessed quantized received signal.

[0040] Furthermore, LOMP iteratively solves for the predicted channel based on the preprocessed quantized received signal, the initial transmitter virtual angle basis matrix, the initial receiver virtual angle basis matrix, and the initial model parameters.

[0041] Angular domain sparsity transformation based on the SV (Saleh-Valenzuela) channel model. The channel matrix is ​​characterized using the SV model commonly used in millimeter-wave channels. The formula for calculating the channel matrix is ​​as follows.

[0042] ; in, This is the original channel matrix. This refers to the number of antennas at the receiving end. This refers to the number of antennas at the transmitting end. The total number of paths, Here is the transmitter angle basis matrix. Here is the angle basis matrix of the receiving end. For the first The complex gain corresponding to each path, For the first The angle of arrival (AoA) of the path. For the first The angle of departure (AoD) of each path, AoA and AoD are both randomly distributed within [0, 2π]. For analog channels, For wavelength, Let be the array response vector of the uniform linear array at the receiving end. Let be the array response vector of the uniform linear array at the transmitting end. The spacing between antenna elements can be set to half a wavelength.

[0043] Solving the analog channel using the above method It is quite complex. Utilizing the angular domain sparsity of millimeter-wave channels, the original channel matrix... In the corner domain via virtual channel Represented as ,in, This is the virtual angle basis matrix of the transmitter. This is the virtual angle basis matrix at the receiving end. and The matrix represents the unitary Discrete Fourier Transform (DFT) matrix, corresponding to the quantization of the arrival and departure angles in the virtual angular domain within the range [0, 2π], with quantization resolutions of respectively. and The formulas for calculating the quantized angle of arrival and the quantized angle of departure are as follows.

[0044] ; in, The quantized angle of arrival. The quantized departure angle. This refers to the number of antennas at the receiving end. This represents the number of antennas at the transmitting end.

[0045] Correspondingly, the quantized received signal can be rewritten as the following formula.

[0046] ; in, To quantize the received signal, This is the baseband merging matrix. and Let these represent the IDFT matrix and the DFT matrix, respectively. To simulate the merging matrix, For the perception matrix, This is the virtual angle basis matrix of the transmitter. This is the virtual angle basis matrix for the receiving end. For virtual channels, For noise vectors, This is the corrected noise vector. The length of the emitted symbol vector. For a mixed merge matrix, For vectorization operators, For predicting the channel (a vectorized form of the virtual channel). It is a uniformly ascending quantization function.

[0047] The problem of estimating the quantized received signal is transformed into the problem of estimating the channel prediction. The calculation formula and constraints for the channel prediction are as follows.

[0048] ; in, To predict the channel (a vectorized form of the virtual channel), and to ensure that the predicted channel is a sparse matrix, we need to solve... The minimum norm (the minimum sum of squares of each element). To solve The minimum norm of , To quantize the received signal, This is the baseband merging matrix. and Let these represent the IDFT matrix and the DFT matrix, respectively. For the perception matrix, For uniformly ascending quantization functions, To set a threshold.

[0049] Based on the calculation formula and constraints of the predicted channel, and combined with the preprocessed quantized received signal, the initial transmitter virtual angle basis matrix, the initial receiver virtual angle basis matrix, and the initial model parameters, the predicted channel is iteratively solved. .

[0050] LOMP consists of multiple interconnected, learnable deep neural networks. The input to the current layer of the deep neural network is the preprocessed quantized received signal, the transmitter virtual angle basis matrix of the previous layer, the receiver virtual angle basis matrix of the previous layer, the loss value of the previous layer, the residual of the previous layer, and the channel estimate of the previous layer. The output of the current layer of the deep neural network is the transmitter virtual angle basis matrix of the current layer, the receiver virtual angle basis matrix of the current layer, the loss value of the current layer, the residual of the current layer, and the channel estimate of the current layer. The input of the current layer of the deep neural network serves as the output of the next layer of the deep neural network. The channel estimate of the current layer is iteratively calculated by using the current layer as the previous layer; the channel estimate of the last layer is used as the predicted channel.

[0051] The residual is the error between the predicted received signal and the quantized received signal calculated in each layer of the deep neural network.

[0052] The loss value is the error between the channel estimate of the current layer and the channel estimate of the previous layer calculated by each layer of the deep neural network.

[0053] Furthermore, the predicted channel will be... Convert to virtual channel Further based on The original channel matrix is ​​obtained by inverse solving. .

[0054] The channel estimation method for MIMO systems provided in this invention obtains the predicted channel of the quantized received signal output by the channel estimation model by inputting the quantized received signal from the receiver of the MIMO system into the channel estimation model. The channel estimation model includes a data-driven quantization error correction network and a learnable orthogonal matched pursuit network. The learnable orthogonal matched pursuit network includes multiple interconnected layers. The data-driven quantization error correction network is used to denoise and correct the quantized received signal to obtain a preprocessed quantized received signal. The current layer of the learnable orthogonal matched pursuit network is used to determine the quantized received signal based on the preprocessed quantized received signal, the transmitter virtual angle basis matrix of the previous layer, and the receiver virtual angle of the previous layer. Using the base matrix, the loss value, the residual, and the channel estimate of the previous layer, the current layer's transmitter virtual angle base matrix, receiver virtual angle base matrix, loss value, residual, and channel estimate are calculated. The current layer is then used as the previous layer to iteratively calculate the current layer's channel estimate. The channel estimate of the last layer is used as the predicted channel. The transmitter virtual angle base matrix of the first layer is the initial transmitter virtual angle base matrix, the receiver virtual angle base matrix of the first layer is the initial receiver virtual angle base matrix, the residual of the first layer is the initial residual, the channel estimate of the first layer is the initial channel estimate, and the loss value of the first layer is the initial loss value. This invention uses the preprocessed quantized received signal, the transmitter virtual angle base matrix of the previous layer, the receiver virtual angle base matrix of the previous layer, the loss value, the residual, and the channel estimate of the previous layer to iteratively solve for the residual and channel estimate of the current layer, thereby obtaining the predicted channel. This effectively suppresses nonlinear quantization distortion and improves the accuracy of channel estimation. This invention, by denoising and correcting the quantized received signal, can fully exploit the statistical features contained in the quantized received signal and eliminate quantization errors.

[0055] Based on the above embodiments, the data-driven quantization error correction network is used for: The quantized received signal is subjected to convolutional residual denoising to obtain the denoised quantized received signal. Fully connected correction is performed on the denoised quantized received signal to obtain the preprocessed quantized received signal.

[0056] like Figure 3As shown, the Data-Driven Quantization Error Correction Network (DDNet) comprises a hybrid architecture combining a convolutional residual denoising module and a fully connected correction module, balancing local noise feature capture with global residual distortion compensation. The convolutional residual denoising module maps the single-channel input to a high-dimensional feature space in its first L-1 layers and models the complex quantization noise distribution using the ReLU activation function. The final convolutional layer maps the high-dimensional features back to the single channel, outputting the predicted quantization noise residual. The fully connected correction module maps the input to a high-dimensional feature space in its first fully connected layer, capturing the global statistical characteristics of the quantization error. The second fully connected layer maps the high-dimensional features back to the original dimension, outputting the final preprocessed quantized received signal. The calculation formula for the preprocessed quantized received signal is as follows.

[0057] ; in, For the preprocessed quantized received signal, To quantize the received signal, For convolutional residual denoising networks, It is a fully connected correction network.

[0058] This invention embeds a data-driven quantization error correction network before LOMP to fully exploit the statistical characteristics contained in the acquired quantized received signal, thereby effectively eliminating quantization errors.

[0059] Based on the above embodiments, the channel estimation model is used for: For the current layer, the virtual angle basis matrix of the transmitter and the virtual angle basis matrix of the receiver of the previous layer are determined based on the virtual angle basis matrix of the transmitter of the previous layer, the virtual angle basis matrix of the receiver of the previous layer, and the loss value of the previous layer. Based on the virtual angle basis matrix of the transmitter and the virtual angle basis matrix of the receiver of the current layer, determine the sensing matrix of the current layer; Based on the residual of the previous layer, the perception matrix of the current layer, and the perception submatrix of the previous layer, the perception submatrix of the current layer is determined; the perception submatrix of the first layer is the initial perception submatrix. Based on the current layer's sensing submatrix, the channel estimate of the previous layer, and the preprocessed quantized received signal, determine the current layer's channel estimate and the current layer's residual; The loss value of the current layer is determined based on the error between the channel estimate of the current layer and the channel estimate of the previous layer.

[0060] A perceptual submatrix is ​​a submatrix composed of the column elements in the perceptual matrix that have the strongest correlation with the residuals.

[0061] The transmitter virtual angle basis matrix and receiver virtual angle basis matrix of the current layer are determined based on the transmitter virtual angle basis matrix, receiver virtual angle basis matrix, and loss value of the previous layer. Optionally, the transmitter virtual angle basis matrix of the current layer is calculated based on the transmitter virtual angle basis matrix, the LOMP preset learning rate, the gradient operator of the transmitter virtual angle basis matrix, and the loss value of the previous layer. The receiver virtual angle basis matrix of the current layer is calculated based on the receiver virtual angle basis matrix, the LOMP preset learning rate, the gradient operator of the receiver virtual angle basis matrix, and the loss value of the previous layer. The calculation formulas for the transmitter virtual angle basis matrix and receiver virtual angle basis matrix of the current layer are as follows.

[0062] ; in, For the current layer, For the previous layer, This is the virtual angle basis matrix of the transmitter in the current layer. This is the virtual angle basis matrix of the transmitter in the previous layer. The learning rate (preset by LOMP). The gradient operator for the virtual angle basis matrix of the transmitter (preset by LOMP). The gradient operator for the virtual angle basis matrix at the receiver (preset by LOMP). This is the loss value of the previous layer. This is the virtual angle basis matrix for the receiver of the current layer. This is the virtual angle basis matrix of the receiver in the previous layer.

[0063] Calculate the sensing matrix of the current layer based on the virtual angle basis matrix of the transmitter and the virtual angle basis matrix of the receiver of the current layer. .

[0064] Based on the residual of the previous layer The perception matrix of the current layer and the perceptron of the previous layer Determine the perceptual submatrix of the current layer. .

[0065] Based on the current layer's perceptual submatrix Channel estimation value of the previous layer and preprocessed quantized received signal Determine the channel estimate for the current layer. and the residual of the current layer .

[0066] Based on the error between the channel estimate of the current layer and the channel estimate of the previous layer, the loss value of the current layer is determined. .

[0067] The current layer's perceptron submatrix is ​​used as the perceptron submatrix of the previous layer, the current layer's channel estimate is used as the channel estimate of the previous layer, the current layer's residual is used as the residual of the previous layer, and the current layer's loss value is used as the loss value of the previous layer. The channel estimate of the current layer is calculated iteratively until the current layer's loss value tends to stabilize. The channel estimate at the last moment is then used as the predicted channel.

[0068] Optionally, the current layer deep neural network sends the calculated current layer perceptron matrix, current layer channel estimate, current layer residual, current layer loss, current layer transmitter virtual angle basis matrix, and current layer receiver virtual angle basis matrix to the next layer deep neural network. The next layer deep neural network is then used as the current layer deep neural network (or the current layer deep neural network is used as the previous layer deep neural network) to iteratively calculate the current layer channel estimate until the channel estimate of the last layer is calculated, thus obtaining the predicted channel.

[0069] This invention achieves iterative solution of channel estimation through the orthogonal matching pursuit algorithm, realizes iterative optimization of the predicted channel, and improves the accuracy of determining the predicted channel.

[0070] Based on the above embodiments, the sensing matrix of the current layer is determined based on the virtual angle basis matrix of the transmitter and the virtual angle basis matrix of the receiver of the current layer, including the following steps: Based on the baseband precoding matrix and analog precoding matrix of the transmitter, determine the hybrid merging matrix of the transmitter; The sensing matrix of the current layer is determined based on the transmitter virtual angle basis matrix, the transmitter hybrid merging matrix, the transmitted symbol vector, the receiver analog merging matrix, and the receiver virtual angle basis matrix of the current layer.

[0071] like Figure 4 As shown, the sensing matrix of the current layer is determined based on the virtual angle basis matrix of the transmitter and the virtual angle basis matrix of the receiver. The formula for calculating the sensing matrix of the current layer is as follows.

[0072] ; in, This is the perception matrix of the current layer. For the current layer, This is the virtual angle basis matrix of the transmitter in the current layer. For the emission symbol vector, This is the virtual angle basis matrix for the receiver of the current layer. For a mixed merge matrix, To simulate the merging matrix, This is the baseband precoding matrix. This is for simulating the precoding matrix.

[0073] This invention combines the virtual angle basis matrix of the transmitter at the current layer, the hybrid merging matrix of the transmitter, the transmitted symbol vector, the analog merging matrix of the receiver, and the virtual angle basis matrix of the receiver at the current layer to achieve accurate calculation of the perception matrix of the current layer.

[0074] Based on the above embodiments, the perceptual submatrix of the current layer is determined based on the residual of the previous layer, the perceptual matrix of the current layer, and the perceptual submatrix of the previous layer, including the following steps: Calculate the correlation between the residual of the previous layer and each column element in the perception matrix of the current layer, and take the column element with the strongest correlation as the perception column element of the current layer. Add the elements of the current layer's perceptual column to the perceptual submatrix of the previous layer to obtain the current layer's perceptual submatrix.

[0075] The formula for calculating the perceptual submatrix of the current layer is as follows.

[0076] ; in, The perception matrix for the current layer is: For the current layer's perception column elements, For each column of the perception matrix of the current layer, The column element with the strongest correlation. The residual of the previous layer, This is the perception submatrix of the current layer. This is the perceptual submatrix of the previous layer.

[0077] This invention calculates the perceptual column elements of the current layer and adds them to the perceptual submatrix of the previous layer to obtain the perceptual submatrix of the current layer, thus achieving accurate iterative updates of the perceptual submatrix of the current layer.

[0078] Based on the above embodiments, the channel estimate and residual of the current layer are determined based on the current layer's sensing sub-matrix, the channel estimate of the previous layer, and the preprocessed quantized received signal, including the following steps: The predicted received signal of the current layer is determined by multiplying the current layer's sensing sub-matrix with the channel estimate of the previous layer. Obtain the difference between the preprocessed quantized received signal and the predicted received signal of the current layer; perform least squares optimization on the difference to obtain the channel estimate of the current layer; The residual of the current layer is determined based on the channel estimate of the current layer, the sensing sub-matrix of the current layer, and the preprocessed quantized received signal.

[0079] The formulas for calculating the channel estimate and residual of the current layer are as follows.

[0080] ; in, This is the channel estimate for the current layer. This is the channel estimate from the previous layer. For the residual of the current layer, For the preprocessed quantized received signal, This is the perception submatrix of the current layer. For the predicted received signal of the current layer, This is an optimization for the least squares method.

[0081] This invention optimizes the difference between the preprocessed quantized received signal and the predicted received signal of the current layer using the least squares method, thereby iteratively solving for the channel estimate of the current layer and subsequently iteratively solving for the residual of the current layer. This invention achieves high-precision and robust channel estimation in low-resolution ADC millimeter-wave massive MIMO systems while maintaining low hardware and power consumption, thus improving the overall performance of the MIMO system.

[0082] The channel estimation method for MIMO systems provided by this invention can effectively suppress nonlinear quantization distortion introduced by low-resolution ADCs, while balancing channel estimation accuracy, generalization ability, and interpretability. Under the same buffer capacity, same bandwidth, and different SNR and ADC resolution scenarios, it can improve channel estimation accuracy (reduce NMSE), maintain robust performance, and thus improve the transmission reliability and energy efficiency of millimeter-wave massive MIMO systems.

[0083] Figure 2 The dual-drive network architecture LOMP proposed in this invention is a serial structure, consisting of an input layer, a data-driven quantization error correction network (DDNet), a model-driven learnable orthogonal matching pursuit network (LOMP), and an output layer, from top to bottom. The functions of each layer work together to achieve high-precision channel estimation. The specific structure and workflow are as follows: The input layer receiver outputs the quantized received signal. After standardization, the data is input into DDNet. The DDNet module consists of a convolutional residual denoising module and a fully connected correction module. The convolutional residual denoising module is configured as a 3-layer one-dimensional convolutional network (the first two layers are feature extraction layers using the ReLU activation function, and the last layer is the residual output layer). First, it starts from... Extract the quantized noise features and output the noise residual, then... Obtain the denoised quantized received signal The fully connected correction module consists of two fully connected layers (with 256 hidden neurons using the LeakyReLU activation function) to process the denoised quantized received signal. Global feature fusion and residual distortion compensation are performed to output the final preprocessed quantized received signal. .

[0084] DDNet output The input is LOMP. LOMP expands the traditional OMP algorithm into a T=5-layer learnable network. Each layer corresponds to one OMP iteration process. It adaptively optimizes the transmitter virtual angle basis matrix and the receiver virtual angle basis matrix of the current layer, updates the receptive submatrix of the current layer and the residual of the current layer, and gradually approximates the real virtual channel vector. The original channel matrix is ​​obtained. This completes the entire channel estimation process.

[0085] Based on the above embodiments, the channel estimation model is trained in the following manner: Obtain the sample quantized received signal from the receiver of the MIMO system; The sample quantized received signal is labeled based on the undistorted sample quantized received signal and the real virtual channel vector to obtain labeled training samples; The labeled training samples are input into the preset model for training, and the preprocessed sample quantized received signal and sample prediction channel output by the preset model are obtained; the preset model consists of an initial quantization error correction network and an initial orthogonal matching pursuit network. Calculate the first loss value of the preset model based on the preprocessed sample quantized received signal and the sample quantized received signal without quantization distortion. Calculate the second loss value of the preset model based on the sample predicted channel and the real virtual channel vector; The initial quantization error correction network is optimized based on the first loss value, and the initial orthogonal matching pursuit network is optimized based on the second loss value to obtain the channel estimation model.

[0086] Generate sample quantized received signals with different signal-to-noise ratios and ADC resolutions. Label the sample quantized received signals based on the undistorted sample quantized received signals and the real virtual channel vector to obtain labeled training samples.

[0087] Set the initial values ​​for the parameters of the convolutional layers and the parameters of the fully connected layers in the initial quantization error correction network, as well as the first loss function.

[0088] Set the initial transmitter virtual angle basis matrix and the initial receiver virtual angle basis matrix for the initial orthogonal matching pursuit network. Set the learning rate, number of iterations T, gradient operators for the transmitter virtual angle basis matrix and receiver virtual angle basis matrix, and the second loss function for the initial orthogonal matching pursuit network.

[0089] First, the initial quantization error correction network (DDNet) is independently pre-trained. The received signal is then quantized using samples. Using the pre-processed sample quantized received signal (without quantization distortion) as input and the sample quantized received signal (without quantization distortion) as label, the initial convolutional residual denoising module and fully connected correction module of DDNet are trained to give DDNet preliminary quantization noise reduction capability. After pre-training, their parameters are fixed. During training, the first loss value is calculated for the pre-processed sample quantized received signal and the sample quantized received signal (without quantization distortion). Based on the first loss value, the parameters of the convolutional layers and fully connected layers of DDNet are initially adjusted.

[0090] Subsequently, dual-module joint training was conducted: the sample was quantized to receive the signal. Input the pre-trained DDNet and output the pre-processed sample quantized received signal. .Will The initial Orthogonal Matching Pursuit (LOMP) network is input, and the sample predicted channel is output after T iterations. The second loss value (Normalized Mean Square Error, NMSE) between the sample predicted channel and the real virtual channel vector is calculated. The DDNet parameters are simultaneously fine-tuned and the initial transmitter virtual angle basis matrix, the initial receiver virtual angle basis matrix, the learning rate, the number of iterations T, the gradient operator of the transmitter virtual angle basis matrix, and the gradient operator of the receiver virtual angle basis matrix are updated through backpropagation until the second loss value converges and tends to stabilize, thus completing the training of the entire network.

[0091] To verify the effectiveness of the LOMP proposed in this invention, simulation experiments were conducted based on a millimeter-wave massive MIMO system simulation platform. The specific experimental setup and results analysis are as follows.

[0092] (1) Simulation Platform and Parameter Settings: The simulation was built using MATLAB and PyTorch. The system bandwidth was 200MHz, the carrier frequency was 28GHz (millimeter wave band), and the ADC quantization resolution was set to 1-bit, 2-bit, and 3-bit scenarios. The signal-to-noise ratio (SNR) ranged from 0 to 20dB. The training dataset contained 10,000 training samples under different channel scenarios, different SNRs, and different ADC resolutions. The test dataset contained 2,000 independent samples. Training parameter settings: learning rate =0.001, training epochs=100, batch size=64, loss function is NMSE.

[0093] (2) Comparison scheme: Four mainstream channel estimation methods were selected for comparison: traditional orthogonal matching pursuit algorithm (OMP), model-driven learnable approximate message passing network (LAMP), data-driven channel estimation network (CNN), and single model-driven deep unfolded OMP network (LOMP).

[0094] (3) Simulation Results and Analysis: The simulation uses Normalized Mean Square Error (NMSE) and algorithm robustness as the core evaluation indicators. The specific simulation results are as follows: Figure 5 and Figure 6 As shown in the figure. The results show that, in terms of NMSE performance, the LOMP of this invention performs best under different ADC resolutions and SNR scenarios. When the ADC resolution is only 1 bit and the SNR is 20 dB, the NMSE of LOMP is reduced by 29 dB compared to traditional OMP, 24 dB compared to CNN, 7 dB compared to LAMP, and 18 dB compared to LOMP, which fully demonstrates that the dual-drive fusion architecture of this invention can effectively suppress quantization distortion and improve estimation accuracy.

[0095] In terms of robustness, when the ADC resolution is reduced from 3 bits to 1 bit and the SNR is 20dB, the NMSE fluctuation of the LOMP of this invention is only 0.2dB, which is much lower than other comparative schemes (fluctuation amplitude 1dB-15dB), indicating that it can still maintain stable performance in low-precision quantization scenarios.

[0096] The channel estimation apparatus for the MIMO system provided by the present invention is described below. The channel estimation apparatus for the MIMO system described below can be referred to in correspondence with the channel estimation method for the MIMO system described above.

[0097] like Figure 7 As shown, the present invention provides a channel estimation device for a MIMO system, comprising: The channel estimation module 701 is used to input the quantized received signal from the receiver of the MIMO system into the channel estimation model and obtain the predicted channel of the quantized received signal output by the channel estimation model. The channel estimation model includes a data-driven quantization error correction network and a learnable orthogonal matching pursuit network. The learnable orthogonal matching pursuit network consists of interconnected multilayers. The data-driven quantization error correction network is used to denoise and correct the quantized received signal to obtain a preprocessed quantized received signal; The current layer of the learnable orthogonal matching pursuit network is used to calculate the current layer's transmitter virtual angle basis matrix, current layer's loss value, current layer's residual, and current layer's channel estimate based on the preprocessed quantized received signal, the transmitter virtual angle basis matrix of the previous layer, the receiver virtual angle basis matrix of the previous layer, the loss value of the previous layer, the residual of the previous layer, and the channel estimate of the previous layer. The current layer is used as the previous layer to iteratively calculate the current layer's channel estimate. The channel estimate of the last layer is used as the predicted channel. The transmitter virtual angle basis matrix of the first layer is the initial transmitter virtual angle basis matrix, the receiver virtual angle basis matrix of the first layer is the initial receiver virtual angle basis matrix, the residual of the first layer is the initial residual, the channel estimate of the first layer is the initial channel estimate, and the loss of the first layer is the initial loss.

[0098] The channel estimation apparatus for a MIMO system provided in this invention obtains the predicted channel of the quantized received signal output by the channel estimation model by inputting the quantized received signal from the receiver of the MIMO system into the channel estimation model. The channel estimation model includes a data-driven quantization error correction network and a learnable orthogonal matching pursuit network. The learnable orthogonal matching pursuit network includes multiple interconnected layers. The data-driven quantization error correction network is used to denoise and correct the quantized received signal to obtain a preprocessed quantized received signal. The current layer of the learnable orthogonal matching pursuit network is used to determine the quantized received signal based on the preprocessed quantized received signal, the transmitter virtual angle basis matrix of the previous layer, and the receiver virtual angle of the previous layer. Using the base matrix, the loss value, the residual, and the channel estimate of the previous layer, the current layer's transmitter virtual angle base matrix, receiver virtual angle base matrix, loss value, residual, and channel estimate are calculated. The current layer is then used as the previous layer to iteratively calculate the current layer's channel estimate. The channel estimate of the last layer is used as the predicted channel. The transmitter virtual angle base matrix of the first layer is the initial transmitter virtual angle base matrix, the receiver virtual angle base matrix of the first layer is the initial receiver virtual angle base matrix, the residual of the first layer is the initial residual, the channel estimate of the first layer is the initial channel estimate, and the loss value of the first layer is the initial loss value. This invention uses the preprocessed quantized received signal, the transmitter virtual angle base matrix of the previous layer, the receiver virtual angle base matrix of the previous layer, the loss value, the residual, and the channel estimate of the previous layer to iteratively solve for the residual and channel estimate of the current layer, thereby obtaining the predicted channel. This effectively suppresses nonlinear quantization distortion and improves the accuracy of channel estimation. This invention, by denoising and correcting the quantized received signal, can fully exploit the statistical features contained in the quantized received signal and eliminate quantization errors.

[0099] All relevant content of each step involved in the above method embodiments can be referenced from the functional description of the corresponding functional module, and will not be repeated here.

[0100] Figure 8 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 8 As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a channel estimation method for a MIMO system. This method includes: inputting the quantized received signal from the receiver of the MIMO system into a channel estimation model, and obtaining the predicted channel of the quantized received signal output by the channel estimation model; wherein the channel estimation model includes a data-driven quantization error correction network and a learnable orthogonal matching pursuit network, the learnable orthogonal matching pursuit network including interconnected multilayers; the data-driven quantization error correction network is used to denoise and correct the quantized received signal to obtain a preprocessed quantized received signal; the current layer of the learnable orthogonal matching pursuit network is used to determine the quantized received signal based on the preprocessed quantized received signal, the virtual angle basis matrix of the transmitter in the previous layer, and... The receiver virtual angle basis matrix, loss value, residual, and channel estimate of the previous layer are used to calculate the transmitter virtual angle basis matrix, receiver virtual angle basis matrix, loss value, residual, and channel estimate of the current layer. The current layer is used as the previous layer to iteratively calculate the channel estimate of the current layer. The channel estimate of the last layer is used as the predicted channel. The transmitter virtual angle basis matrix of the first layer is the initial transmitter virtual angle basis matrix, the receiver virtual angle basis matrix of the first layer is the initial receiver virtual angle basis matrix, the residual of the first layer is the initial residual, the channel estimate of the first layer is the initial channel estimate, and the loss value of the first layer is the initial loss value.

[0101] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0102] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a channel estimation method for a MIMO system provided by the methods described above. This method includes: inputting a quantized received signal from the receiver of the MIMO system into a channel estimation model, and obtaining a predicted channel of the quantized received signal output by the channel estimation model; wherein the channel estimation model includes a data-driven quantization error correction network and a learnable orthogonal matching pursuit network, the learnable orthogonal matching pursuit network including interconnected multiple layers; the data-driven quantization error correction network is used to denoise and correct the quantized received signal to obtain a preprocessed quantized received signal; the current layer of the learnable orthogonal matching pursuit network is used to perform denoising and correction on the preprocessed quantized received signal. Given the signal, the transmitter virtual angle basis matrix of the previous layer, the receiver virtual angle basis matrix of the previous layer, the loss value of the previous layer, the residual of the previous layer, and the channel estimate value of the previous layer, calculate the transmitter virtual angle basis matrix of the current layer, the receiver virtual angle basis matrix of the current layer, the loss value of the current layer, the residual of the current layer, and the channel estimate value of the current layer; use the current layer as the previous layer to iteratively calculate the channel estimate value of the current layer; use the channel estimate value of the last layer as the predicted channel; the transmitter virtual angle basis matrix of the first layer is the initial transmitter virtual angle basis matrix, the receiver virtual angle basis matrix of the first layer is the initial receiver virtual angle basis matrix; the residual of the first layer is the initial residual; the channel estimate value of the first layer is the initial channel estimate value; the loss value of the first layer is the initial loss value.

[0103] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0104] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention 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; and these 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 the present invention.

Claims

1. A channel estimation method for a MIMO system, characterized in that, include: The quantized received signal from the receiver of the MIMO system is input into the channel estimation model to obtain the predicted channel of the quantized received signal output by the channel estimation model. The channel estimation model includes a data-driven quantization error correction network and a learnable orthogonal matching pursuit network, wherein the learnable orthogonal matching pursuit network includes interconnected multilayers. The data-driven quantization error correction network is used to denoise and correct the quantized received signal to obtain a preprocessed quantized received signal. The current layer of the learnable orthogonal matching pursuit network is used to calculate the current layer's transmitter virtual angle basis matrix, receiver virtual angle basis matrix, loss value, residual, and channel estimate based on the preprocessed quantized received signal, the transmitter virtual angle basis matrix of the previous layer, the receiver virtual angle basis matrix of the previous layer, the loss value of the previous layer, the residual of the previous layer, and the channel estimate of the previous layer; the current layer is used as the previous layer to iteratively calculate the channel estimate of the current layer; and the channel estimate of the last layer is used as the predicted channel. The transmitter virtual angle basis matrix of the first layer is the initial transmitter virtual angle basis matrix, the receiver virtual angle basis matrix of the first layer is the initial receiver virtual angle basis matrix, the residual of the first layer is the initial residual, the channel estimate of the first layer is the initial channel estimate, and the loss of the first layer is the initial loss.

2. The channel estimation method for a MIMO system according to claim 1, characterized in that, The channel estimation model is used for: For the current layer, the transmitter virtual angle basis matrix and the receiver virtual angle basis matrix of the current layer are determined based on the transmitter virtual angle basis matrix of the previous layer, the receiver virtual angle basis matrix of the previous layer, and the loss value of the previous layer. The sensing matrix of the current layer is determined based on the virtual angle basis matrix of the transmitter and the virtual angle basis matrix of the receiver of the current layer. Based on the residual of the previous layer, the perception matrix of the current layer, and the perception sub-matrix of the previous layer, the perception sub-matrix of the current layer is determined; the perception sub-matrix of the first layer is the initial perception sub-matrix. Based on the current layer's sensing sub-matrix, the channel estimate of the previous layer, and the preprocessed quantized received signal, determine the current layer's channel estimate and the current layer's residual; The loss value of the current layer is determined based on the error between the channel estimate of the current layer and the channel estimate of the previous layer.

3. The channel estimation method for a MIMO system according to claim 2, characterized in that, The step of determining the current layer's perceptual submatrix based on the residual of the previous layer, the current layer's perceptual matrix, and the previous layer's perceptual submatrix includes: Calculate the correlation between the residual of the previous layer and each column element in the perception matrix of the current layer, and take the column element with the strongest correlation as the perception column element of the current layer. The perceptual column elements of the current layer are added to the perceptual submatrix of the previous layer to obtain the perceptual submatrix of the current layer.

4. The channel estimation method for a MIMO system according to claim 2, characterized in that, The step of determining the channel estimate and residual of the current layer based on the current layer's sensing submatrix, the channel estimate of the previous layer, and the preprocessed quantized received signal includes: The predicted received signal of the current layer is determined based on the product of the current layer's sensing sub-matrix and the channel estimate of the previous layer. Obtain the difference between the preprocessed quantized received signal and the predicted received signal of the current layer; perform least squares optimization on the difference to obtain the channel estimate of the current layer; The residual of the current layer is determined based on the channel estimate of the current layer, the sensing sub-matrix of the current layer, and the preprocessed quantized received signal.

5. The channel estimation method for a MIMO system according to claim 2, characterized in that, The step of determining the sensing matrix of the current layer based on the transmitter virtual angle basis matrix and the receiver virtual angle basis matrix of the current layer includes: Based on the baseband precoding matrix and analog precoding matrix of the transmitter, determine the hybrid merging matrix of the transmitter; The perception matrix of the current layer is determined based on the transmitter virtual angle basis matrix of the current layer, the hybrid merging matrix of the transmitter, the transmitted symbol vector, the analog merging matrix of the receiver, and the receiver virtual angle basis matrix of the current layer.

6. The channel estimation method for a MIMO system according to claim 1, characterized in that, The data-driven quantization error correction network is used for: The quantized received signal is subjected to convolutional residual denoising processing to obtain the denoised quantized received signal; The denoised quantized received signal is subjected to full-connection correction to obtain the preprocessed quantized received signal.

7. The channel estimation method for a MIMO system according to claim 1, characterized in that, The channel estimation model was trained in the following manner: Obtain the sample quantized received signal from the receiver of the MIMO system; The sample quantized received signal is labeled based on the undistorted sample quantized received signal and the real virtual channel vector to obtain labeled training samples; The labeled training samples are input into a preset model for training, and the preprocessed sample quantization received signal and sample prediction channel output by the preset model are obtained. The preset model consists of an initial quantization error correction network and an initial orthogonal matching pursuit network; Based on the preprocessed sample quantized received signal and the sample quantized received signal without quantization distortion, calculate the first loss value of the preset model; Calculate the second loss value of the preset model based on the sample predicted channel and the real virtual channel vector; The initial quantization error correction network is optimized based on the first loss value, and the initial orthogonal matching pursuit network is optimized based on the second loss value to obtain the channel estimation model.

8. A channel estimation device for a MIMO system, characterized in that, include: The channel estimation module is used to input the quantized received signal from the receiver of the MIMO system into the channel estimation model and obtain the predicted channel of the quantized received signal output by the channel estimation model. The channel estimation model includes a data-driven quantization error correction network and a learnable orthogonal matching pursuit network, wherein the learnable orthogonal matching pursuit network includes interconnected multilayers. The data-driven quantization error correction network is used to denoise and correct the quantized received signal to obtain a preprocessed quantized received signal. The current layer of the learnable orthogonal matching pursuit network is used to calculate the current layer's transmitter virtual angle basis matrix, receiver virtual angle basis matrix, loss value, residual, and channel estimate based on the preprocessed quantized received signal, the transmitter virtual angle basis matrix of the previous layer, the receiver virtual angle basis matrix of the previous layer, the loss value of the previous layer, the residual of the previous layer, and the channel estimate of the previous layer; the current layer is used as the previous layer to iteratively calculate the channel estimate of the current layer; and the channel estimate of the last layer is used as the predicted channel. The transmitter virtual angle basis matrix of the first layer is the initial transmitter virtual angle basis matrix, the receiver virtual angle basis matrix of the first layer is the initial receiver virtual angle basis matrix, the residual of the first layer is the initial residual, the channel estimate of the first layer is the initial channel estimate, and the loss of the first layer is the initial loss.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the channel estimation method for the MIMO system as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the channel estimation method for the MIMO system as described in any one of claims 1 to 7.

Images