Method, apparatus and related device for channel state information feedback in multi-user communication system

Abstract

This invention provides a method, apparatus, and related equipment for channel state information (CSI) feedback in a multi-user communication system. It extracts peak features of the channel signal in the angle-delay domain and regional correlation features of the channel signal in the angle-delay domain; fuses the peak features and regional correlation features; inputs the sparsity variation feature matrix of the low-Earth orbit (LEO) satellite channel as the satellite moves to different orbital positions into a channel state information (CSI) model to obtain a low-dimensional vector; transmits the low-dimensional vector to the satellite, so that the satellite can reconstruct the low-dimensional vector into a high-dimensional CSI matrix using the CSI model; wherein the CSI model is trained based on the high-dimensional CSI matrix. By compressing and reconstructing the CSI using the CSI model, the overhead of CSI feedback is reduced, improving system communication performance; the CSI model can extract changing data features from the channel matrix, improving reconstruction accuracy and further enhancing system performance.

Description

Technical Field

[0001] This invention relates to the field of satellite communication technology, and in particular to a method, apparatus and related equipment for channel status information feedback in a multi-user communication system. Background Technology

[0002] Phased array antennas have been widely used in low Earth orbit (LEO) satellite communication systems due to their high flexibility in beamforming, beam reconfiguration, and beam scanning technologies, as well as their strong anti-interference capabilities. For satellite internet, to leverage the advantages of phased array antennas, satellites need to accurately obtain Channel State Information (CSI). Current methods for obtaining CSI involve the satellite sending pre-agreed pilot signals to the user terminal (UT). The UT calculates the downlink CSI using the pilot signals and then sends it back to the satellite via a feedback link. CSI feedback techniques fall into two main categories: explicit feedback and implicit feedback. Implicit feedback is codebook-based, requiring both the satellite and the UT to pre-store codebooks before communication begins. Upon receiving the pilot signals from the satellite, the user obtains the downlink CSI using channel estimation techniques. Then, the user calculates the optimal codeword index for the CSI according to codeword selection rules and sends it to the satellite via the feedback link. The satellite receives the feedback codeword index and retrieves the corresponding codeword from its codebook—the user's feedback information. Although codebook-based channel state information feedback methods have low computational complexity, the application of large-scale phased array antennas on satellites leads to a dramatic increase in the number of antennas and an exponential increase in codebook size. Simultaneously, the computational cost and complexity of the optimal codeword search algorithm increase dramatically. Therefore, codebook-based channel state information feedback methods are not suitable for satellite communications with large-scale phased array antennas. Explicit feedback techniques include compressed sensing-based channel state information feedback. Due to the more complex channel characteristics of multi-beam, multi-user satellite phased array antennas, the sparse variation feature matrix commonly used in compressed sensing-based channel state information feedback is insufficient to characterize channel features, resulting in insufficient sparsity of the observed channel state information. Since the accuracy of channel reconstruction depends on the sparsity of the observed signal's representation of the channel state information, insufficient sparsity of the observed channel state information directly affects the accuracy of channel reconstruction, thus limiting the application of compressed sensing feedback methods in satellite internet communications. Summary of the Invention

[0003] This invention provides a method, apparatus, and related equipment for channel state information feedback in a multi-user communication system, which addresses the shortcomings of traditional LEO satellite multi-beam multi-user communication system channel state information feedback methods, such as their unsuitability for large-scale phased array antenna satellite communication and their low accuracy in channel reconstruction, which affects signal transmission quality.

[0004] This invention provides a method for channel state information feedback in a multi-user communication system, comprising:

[0005] Extract the peak characteristics of the channel signal in the angle-delay domain and the regional correlation characteristics of the channel signal in the angle-delay domain;

[0006] By fusing the peak features and the regional correlation features, a sparse variation feature matrix of the low-Earth orbit satellite channel as the satellite moves to different orbital positions is obtained.

[0007] The sparsity variation feature matrix of the low-orbit satellite channel as it changes with different orbital positions of the satellite is input into the channel state information model to obtain a low-dimensional vector.

[0008] The low-dimensional vector is sent to the satellite so that the satellite can reconstruct the low-dimensional vector into a high-dimensional channel state information matrix using the channel state information model.

[0009] The channel state information model is obtained by training a high-dimensional channel state information matrix.

[0010] According to the multi-user communication system channel state information feedback method provided by the present invention, the channel state information model includes an encoder;

[0011] The encoder is installed at the user end and is used to compress the sparse variation feature matrix in the input angle-delay domain, output the compressed low-dimensional channel state information vector, and feed it back to the satellite.

[0012] During the training process of the channel state information model, the sparse variation feature matrix in the angle-delay domain is passed through two convolutional layers and then through a lightweight self-attention feature extraction module to obtain a high-dimensional feature map. The high-dimensional feature map includes feature information of the peak value and the information around the peak.

[0013] The high-dimensional feature map is passed through two more convolutional layers and then subjected to dimensionality transformation to obtain a high-dimensional vector.

[0014] The high-dimensional vector is reduced in dimensionality using a fully connected layer to obtain a low-dimensional vector.

[0015] According to the multi-user communication system channel state information feedback method provided by the present invention, the channel state information model includes a decoder;

[0016] The decoder is located at the satellite end and is used to reconstruct low-dimensional vectors;

[0017] During the training process of the channel state information model, the low-dimensional vector is converted into a high-dimensional vector through a fully connected layer;

[0018] The high-dimensional vector is transformed to obtain a high-dimensional feature matrix;

[0019] The high-dimensional feature matrix is ​​convolved through four convolutional layers, with the number of convolutional kernels in the four convolutional layers gradually increasing to obtain features from multiple channels.

[0020] The features of the multiple channels are aggregated and reconstructed into a high-dimensional channel state information matrix.

[0021] According to the multi-user communication system channel state information feedback method provided by the present invention, the extraction of peak features of the channel signal in the angle-time delay domain includes:

[0022] The vertical and horizontal information of the input matrix corresponding to the channel signal is extracted by pooling layers with different kernel sizes;

[0023] The information in the vertical and horizontal directions is fused into a feature map that simultaneously contains information in the angle domain and the time delay domain.

[0024] Extract the peak features of the feature map that simultaneously contains information in the angle domain and the time delay domain.

[0025] According to the multi-user communication system channel state information feedback method provided by the present invention, the extraction of regional correlation features of channel signals in the angle-delay domain includes:

[0026] The input matrix corresponding to the channel signal is convolved through two convolutional layers, and then the channel signal is processed through two channels. One channel expands the features after the convolution operation to obtain a first matrix, and the rows of the first matrix are summed to obtain a first column vector.

[0027] The second and third column vectors are obtained by performing a dimensional transformation through another channel. The second and third column vectors are then multiplied by the first column vector, and the results are concatenated along the third dimension to generate the second matrix.

[0028] The second matrix is ​​reconstructed to have the same dimension as the input matrix, and the regional correlation characteristics of the channel signal in the angle-delay domain are output.

[0029] According to the multi-user communication system channel state information feedback method provided by the present invention, the step of expanding the feature matrix after convolution operation to obtain the first matrix includes:

[0030] The feature matrix is ​​expanded by mirroring, adding two rows and two columns of elements.

[0031] Set a pre-defined aggregation region and use it to move row by row and column by column on the expanded matrix to perform raster scanning. Recombine the scanned region into a row vector with the number of elements equal to the product of the region height, region width and number of channels. Superimpose all the scanned and deformed row vectors to obtain the first matrix.

[0032] According to the multi-user communication system channel state information feedback method provided by the present invention, the fusion of the peak feature and the regional correlation feature includes:

[0033] A lightweight feature fusion layer is used to fuse the peak feature and the region correlation feature. The new mapping feature H4 output by the feature fusion layer is:

[0034] H4=ηH1+γH2+μH3

[0035] Wherein, H1 is the input matrix corresponding to the channel signal, H2 is the peak feature matrix, H3 is the regional correlation feature matrix, and η, γ, and μ are the weight parameters η, γ, and μ of the input matrix, peak feature matrix, and regional correlation feature matrix corresponding to the channel signal, respectively, which are obtained through model training.

[0036] The present invention also provides a channel state information feedback device for a multi-user communication system, comprising:

[0037] The extraction module is used to extract the peak features of the channel signal in the angle-time delay domain and the regional correlation features of the channel signal in the angle-time delay domain;

[0038] The fusion module is used to fuse the peak features and the regional correlation features to obtain a sparse variation feature matrix of the low-orbit satellite channel as the satellite moves to different orbital positions;

[0039] The input module is used to input the sparsity variation feature matrix of the low-orbit satellite channel as the satellite moves to different orbital positions into the channel state information model to obtain a low-dimensional vector;

[0040] The reconstruction module is used to send the low-dimensional vector to the satellite so that the satellite can reconstruct the low-dimensional vector into a high-dimensional channel state information matrix through the channel state information model.

[0041] The channel state information model is obtained by training a high-dimensional channel state information matrix.

[0042] 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 program to implement the multi-user communication system channel state information feedback method as described in any of the preceding claims.

[0043] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the channel state information feedback method for a multi-user communication system as described in any of the preceding claims.

[0044] The present invention provides a method, apparatus, and related equipment for channel state information (CSI) feedback in a multi-user communication system. This involves extracting peak features and regional correlation features of the channel signal in the angle-delay domain; fusing these features to obtain a sparse variation feature matrix of the low-Earth orbit (LEO) satellite channel as the satellite moves to different orbital positions; inputting this sparse variation feature matrix into a channel state information (CSI) model to obtain a low-dimensional vector; and transmitting this low-dimensional vector to the satellite, which then reconstructs the low-dimensional vector into a high-dimensional CSI matrix using the CSI model. The CSI model is trained based on the high-dimensional CSI matrix. By compressing and reconstructing CSI using the CSI model, the overhead of CSI display feedback is reduced, improving system communication performance. The CSI model can also extract changing data features from the channel matrix, improving the accuracy of CSI reconstruction and further enhancing system performance. Attached Figure Description

[0045] 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.

[0046] Figure 1 This is one of the flowcharts illustrating the channel state information feedback method for a multi-user communication system provided in this embodiment of the invention;

[0047] Figure 2 This is a schematic diagram of the channel state information model structure provided in an embodiment of the present invention;

[0048] Figure 3 This is a schematic diagram of the peak extraction process provided in an embodiment of the present invention;

[0049] Figure 4 This is a schematic diagram of the regional correlation extraction process provided in an embodiment of the present invention;

[0050] Figure 5 This is a schematic diagram of the unfolding operation process provided in an embodiment of the present invention;

[0051] Figure 6This is a schematic diagram of the lightweight feature fusion process provided in an embodiment of the present invention;

[0052] Figure 7 This is a second flowchart illustrating the channel state information feedback method for a multi-user communication system provided in this embodiment of the invention.

[0053] Figure 8 This is a functional structure diagram of the channel state information feedback device for a multi-user communication system provided in an embodiment of the present invention;

[0054] Figure 9 This is a functional structure diagram of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0055] 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.

[0056] Figure 1 The flowchart of the channel state information feedback method for a multi-user communication system provided in the embodiments of the present invention is as follows: Figure 1 As shown, the multi-user communication system channel state information feedback method provided in this embodiment of the invention includes:

[0057] Step 101: Extract the peak characteristics of the channel signal in the angle-delay domain and the regional correlation characteristics of the channel signal in the angle-delay domain;

[0058] Step 102: Fuse the peak features and the regional correlation features to obtain the sparsity variation feature matrix of the low-orbit satellite channel as the satellite moves to different orbital positions;

[0059] Step 103: Input the sparsity variation feature matrix of the low-orbit satellite channel as the satellite moves to different orbital positions into the channel state information model to obtain a low-dimensional vector;

[0060] Step 104: Send the low-dimensional vector to the satellite so that the satellite can reconstruct the low-dimensional vector into a high-dimensional channel state information matrix using the channel state information model;

[0061] The channel state information model is obtained by training a high-dimensional channel state information matrix.

[0062] Traditional channel state information (CSI) feedback methods for multi-user communication systems include implicit and explicit feedback. The overhead of traditional implicit feedback methods is proportional to the number of antennas; as the number of antennas increases, feedback incurs significant overhead, severely reducing the spectral efficiency of the communication system. With the application of large-scale spaceborne phased array antennas, the number of antennas is gradually increasing. Due to the excessively high feedback cost, obtaining accurate downlink CSI from satellites is becoming increasingly challenging. In other words, as the number of antennas increases, downlink CSI feedback occupies a large amount of bandwidth resources, and the contradiction between feedback overhead and system performance gradually intensifies. Therefore, researching CSI feedback techniques that ensure CSI feedback accuracy and reduce CSI feedback overhead is crucial. Large-scale antenna multi-beam multi-user channels exhibit sparsity characteristics in the angle-delay domain, providing support for the compressibility of the channel matrix. Unlike terrestrial communication, the peak energy of the channel in the angle-delay domain of satellite communication systems varies with satellite motion and azimuth angle changes, and the channel sparsity characteristics also change accordingly. Traditional methods of extracting features for compression using explicit feedback ignore the sparsity of the channel in the angular domain and the variation of the energy peak with the azimuth angle of the satellite motion. As a result, the compressed vector loses the information of the channel in the angular domain, which leads to a decrease in the accuracy of the reconstructed CSI matrix.

[0063] The multi-user communication system channel state information (CSI) feedback method provided in this invention extracts peak features and regional correlation features of the channel signal in the angle-delay domain. These peak features and regional correlation features are then fused to obtain a sparse variation feature matrix of the low-Earth orbit (LEO) satellite channel as the satellite moves to different orbital positions. This sparse variation feature matrix is ​​input into a channel state information (CSI) model to obtain a low-dimensional vector. The low-dimensional vector is then sent to the satellite, allowing the satellite to reconstruct a high-dimensional CSI matrix using the CSI model. The CSI model is trained based on the high-dimensional CSI matrix. By compressing and reconstructing the CSI using the CSI model, the overhead of CSI display feedback is reduced, improving system communication performance. The CSI model can also extract changing data features from the channel matrix, improving the accuracy of CSI reconstruction and further enhancing system performance.

[0064] Based on any of the above embodiments, the channel state information model includes an encoder;

[0065] The encoder is installed at the user end and is used to compress the sparse variation feature matrix in the input angle-delay domain, output the compressed low-dimensional channel state information vector, and feed it back to the satellite.

[0066] During the training process of the channel state information model, the sparse variation feature matrix in the angle-delay domain is passed through two convolutional layers and then through a lightweight self-attention feature extraction module to obtain a high-dimensional feature map. The high-dimensional feature map includes feature information of the peak value and the information around the peak.

[0067] The high-dimensional feature map is passed through two more convolutional layers and then subjected to dimensionality transformation to obtain a high-dimensional vector.

[0068] The high-dimensional vector is reduced in dimensionality using a fully connected layer to obtain a low-dimensional vector.

[0069] In this embodiment of the invention, low-Earth orbit (LEO) satellite channels are more complex than terrestrial channels, and their structural features in the angle-delay domain change with the satellite's azimuth angle. To improve the accuracy of the reconstructed CSI matrix, the sparsity features of the channel's structure in the angle domain that change with satellite motion can be extracted and compressed into a low-dimensional vector. This low-dimensional vector contains the channel's characteristic information in the angle-delay domain.

[0070] Based on any of the above embodiments, the channel state information model includes a decoder;

[0071] The decoder is located at the satellite end and is used to reconstruct low-dimensional vectors;

[0072] During the training process of the channel state information model, the low-dimensional vector is converted into a high-dimensional vector through a fully connected layer;

[0073] The high-dimensional vector is transformed to obtain a high-dimensional feature matrix;

[0074] The high-dimensional feature matrix is ​​convolved through four convolutional layers, with the number of convolutional kernels in the four convolutional layers gradually increasing to obtain features from multiple channels.

[0075] The features of the multiple channels are aggregated and reconstructed into a high-dimensional channel state information matrix.

[0076] In this embodiment of the invention, the channel state information model is as follows: Figure 2As shown, the encoder at the user end compresses the input CSI and outputs the compressed CSI vector, which is then fed back to the satellite via the uplink channel. The decoder at the satellite end takes the received compressed vector as input and outputs the reconstructed CSI matrix. The input to the model network is the sparse variation feature matrix H′ in the angle-delay domain. During network training, the input matrix H′ first passes through a convolutional layer with two 3×3 kernels, then through a lightweight self-attention feature extraction module to obtain a feature map containing both peak values ​​and surrounding information. Next, a convolutional layer with two 3×3 kernels is used to increase the encoder's nonlinear structure, and the output matrix is ​​transformed into a vector form through dimensionality transformation. Finally, a fully connected layer is used to simulate the projection function of the compressed sensing algorithm to reduce the dimensionality of the vector, resulting in a low-dimensional vector s, which is the codeword obtained during the compression process.

[0077] The compressed low-dimensional vector *s* is fed back to the satellite via the uplink, where the channel matrix is ​​reconstructed by the decoder. In the decoder, the low-dimensional vector *s* is transformed into a high-dimensional vector through a fully connected layer, and then transformed to the size of the CSI matrix through a dimensionality transformation. This matrix is ​​passed through four convolutional layers, each with a 3×3 kernel, and the number of kernels gradually increases to 2, 8, 16, and 32. After the final convolutional layer with 32 kernels, the matrix has 32 channels. To aggregate multiple channels, the last convolutional layer is set to have two 1×1 kernels. During this process, the reconstructed CSI matrix is ​​continuously refined. Due to the use of multiple convolutional layers, a residual structure is used to avoid the gradient vanishing problem caused by superimposed nonlinear transformations. Finally, the encoder output... The resulting reconstructed CSI matrix.

[0078] In this embodiment of the invention, the accuracy of the CSI feedback matrix can be improved by reconstructing the CSI matrix using the low-dimensional vector output by the model.

[0079] Based on any of the above embodiments, the extraction of peak features of the channel signal in the angle-delay domain includes:

[0080] Step 201: Extract the vertical and horizontal information of the input matrix corresponding to the channel signal through pooling layers with different kernel sizes;

[0081] Step 202: Fuse the information in the vertical and horizontal directions into a feature map that simultaneously contains information in the angle domain and the time delay domain;

[0082] Step 203: Extract the peak features of the feature map that simultaneously contains information in the angle domain and the time delay domain.

[0083] As the satellite moves, its azimuth and path delay constantly change, thus the location of the power peak in the CSI angle-time delay domain also continuously changes. Accurately identifying and extracting the peak location is crucial for generating feature maps and subsequent compression. In the satellite CSI structure, the vertical and horizontal directions correspond to the time delay and angle domains, respectively. Strip pooling kernels can be used to extract peak information separately. Using pooling kernels of sizes 44×1 and 1×26 to extract CSI information in the time delay and angle domains respectively avoids capturing irrelevant information from the other domain. The peak extraction block extraction process is as follows: Figure 3 As shown. This process includes:

[0084] First, pooling operations using two kernels are performed to extract information in the vertical and horizontal directions. The row vector l containing vertical time-delay domain information is calculated as follows:

[0085]

[0086] Where H1 is the input matrix of this module, [H1] i,j This represents the element in the i-th row and j-th column of the input matrix H1. j It is the j-th element of the output row vector l.

[0087] The column vector r containing horizontal angular domain information is calculated as follows:

[0088]

[0089] Where r i It outputs the i-th element of the column vector r. Then, the row vector l and column vector r are expanded along the columns and rows respectively to form H1 and matrices l′ and r′ of the same size.

[0090] Then, l′ and r′ are fused into a new feature map RL that simultaneously contains information from the angle domain and the time delay domain, calculated as follows:

[0091] RL=l′+r′

[0092] Finally, the new feature map output by the peak extraction module H2 is calculated as follows:

[0093] H2 = RL ☉ H1

[0094] Where ⊙ represents the Hadamard product. Each position in the output new feature map H2 contains information about the corresponding position in H1 in both the angular and temporal domains.

[0095] Based on any of the above embodiments, the extraction of regional correlation features of the channel signal in the angle-delay domain includes:

[0096] Step 301: After the input matrix corresponding to the channel signal is convolved through two convolutional layers, the channel signal is processed through two channels. One channel expands the features after the convolution operation to obtain a first matrix, and the rows of the first matrix are summed to obtain a first column vector.

[0097] Step 302: Perform dimensional transformation through another channel to obtain the second column vector and the third column vector. Multiply the second column vector and the third column vector by the first column vector respectively, and connect the results in the third dimension to generate the second matrix.

[0098] Step 303: Reconstruct the second matrix to have the same dimension as the input matrix, and output the regional correlation characteristics of the channel signal in the angle-delay domain.

[0099] In this embodiment of the invention, considering the strong regional correlation around the peak, extracting regional correlation is also important for extracting channel features. Traditional self-attention models calculate the correlation coefficient of each pixel across the entire feature map, but ignore the positional relationships between different pixels. Considering the strong correlation in the region near the peak of the channel matrix, a new regional self-attention model is adopted. This model restricts the scope of self-attention to a local area, rather than the entire feature map. Since dimensionality reduction is performed in subsequent network processes, the embedding layer of the regional self-attention model is simplified. The simplified self-attention model α is calculated as follows:

[0100]

[0101] Here, R((i,j)) is the relevance region at position (i,j), indicating which elements will be aggregated into new features, and (i′,j′) is the position index belonging to R((i,j)). α is the aggregation function used to calculate [H1]. i,j The correlation between elements in region R((i,j)).

[0102] The process of generating region-related extraction blocks is as follows: Figure 4As shown, the input matrix H1 is convolved through two convolutional layers with 1×1 kernels, aggregating the two channels to obtain tensor1. This process adds non-linear features to the model. Then, tensor1 is unfolded to obtain a two-dimensional matrix tensor2, and the rows are summed to obtain a column vector tensor3. At this point, each element of tensor3 contains all the information of each element in H1 within the region R((i,j)). Simultaneously, tensor1 is reshaped into two column vectors tensor4 and tensor5. tensor4 and tensor5 are multiplied by tensor3 respectively and concatenated along the third dimension to generate a two-dimensional matrix tensor6. Finally, tensor6 is reshaped to the same dimension as H1, outputting matrix H3.

[0103] In this embodiment of the invention, the step of unfolding the features after the convolution operation to obtain the first matrix includes:

[0104] The feature matrix is ​​expanded by mirroring, adding two rows and two columns of elements.

[0105] Set a pre-defined aggregation region and use it to move row by row and column by column on the expanded matrix to perform raster scanning. Recombine the scanned region into a row vector with the number of elements equal to the product of the region height, region width and number of channels. Superimpose all the scanned and deformed row vectors to obtain the first matrix.

[0106] like Figure 5 As shown, the size of each feature matrix is ​​marked. First, the feature matrix (26*44*2) is expanded by mirroring, adding two rows and two columns of elements. The expanded row above the first row is the last row of the original matrix, and the expanded column to the left of the first column is the last column of the original matrix. The last row and last column are expanded in the same way. The expanded elements in the corners are the elements in the original matrix that are symmetrical to their diagonals. Next, a 3×3 aggregation region is set and used to move row by row and column by column on the expanded matrix to perform raster scanning. The purpose is to recombine the scanned region into a row vector with 3×3×2=18 elements. Finally, all the scanned and deformed row vectors are superimposed into a 1144×18 matrix. At this point, the original 26×44 matrix is ​​"expanded" into a first matrix of 1144×18.

[0107] In this embodiment of the invention, fusing the peak feature and the regional correlation feature includes:

[0108] A lightweight feature fusion layer is used to fuse the peak feature and the region correlation feature. The new mapping feature H4 output by the feature fusion layer is:

[0109] H4=ηH1+γH2+μH3

[0110] Wherein, H1 is the input matrix corresponding to the channel signal, H2 is the peak feature matrix, H3 is the regional correlation feature matrix, and η, γ, and μ are the weight parameters of the input matrix, peak feature matrix, and regional correlation feature matrix corresponding to the channel signal, respectively. η, γ, and μ are obtained through model training.

[0111] In this embodiment of the invention, after extracting the channel peak location and the correlation of the region surrounding the peak, a lightweight feature combination layer is used to fuse the two features together to generate a new mapping feature that simultaneously has the peak location and regional correlation.

[0112] The structure of the lightweight feature fusion layer is as follows: Figure 6 As shown, the features from the two feature extraction modules are fused using a three-bit trainable parameter. The resulting new feature map contains both peak information and correlation information of the region surrounding the peak.

[0113] In this embodiment of the invention, the energy peak and regional correlation information of the channel in the angle-delay domain are first extracted through the attention mechanism. Then, the information is fused through a lightweight feature fusion layer to form a feature map with peak information and regional correlation information. Finally, a low-dimensional vector is obtained by projection through a fully connected layer, thereby improving the information content of the low-dimensional vector.

[0114] Based on any of the above embodiments, the embodiments of the present invention include two stages: initial access and information transmission, such as... Figure 7 As shown, the dashed lines represent the initial access phase, and the solid lines represent the information transmission phase. The initial access phase includes five steps: the user sends user channels to the satellite; the satellite integrates the user channels into a system channel and sends it to the gateway station; the gateway station trains the network; the gateway station sends the network structure and parameters to the satellite; and the satellite sends encoder parameters from the network to the user. The information transmission phase includes five steps: user channel processing; the user compresses the CSI into a low-dimensional vector using the encoder; the user feeds back the low-dimensional vector to the satellite; the satellite reconstructs the low-dimensional vector into a high-dimensional CSI matrix using the decoder; and the satellite integrates the reconstructed user matrix into a system channel matrix.

[0115] Initial access phase:

[0116] Step 1: Users send user channels to the satellite, and each UT sends its own space-frequency domain user channel matrix h to the satellite. k .

[0117] Step 2: The satellite integrates the user channels into system channels and sends them to the gateway station. The satellite receives the channel matrix h of these users. k Combine them into the system channel matrix Hsf :

[0118] H sf = [h1;h2;...;h K ]

[0119] The system channel matrix H sf Send to the mail gateway.

[0120] Step 3: The gateway station trains the network by transmitting H signals from the satellite. sf The network is trained using the training data. The network consists of two parts: an encoder and a decoder.

[0121] Step 4: The gateway station sends the network structure and network parameters to the satellite. The gateway station then sends the network structure and network encoder parameters Θ. en Decoder parameter Θ de Send to the satellite.

[0122] Step 5: The satellite sends encoder parameters from the network to the user, and the satellite sends decoder parameters Θ. de Send to the user; the initial access phase ends.

[0123] Information transmission stage:

[0124] Step 1: User channel processing, transforming the user channel matrix from the space-frequency domain to the angle-delay domain using a two-dimensional Fourier transform:

[0125] h′ k =F M h k F c

[0126] Then, the user's near-zero values ​​in the angle-delay domain are truncated.

[0127] Step 2: The user compresses the CSI into a low-dimensional vector using an encoder. The user inputs the channel matrix after channel processing into the encoder, and the output is the compressed low-dimensional vector s. k :

[0128] s k =f en (h′ k ,Θ en )

[0129] Step 3: The user feeds back a low-dimensional vector to the satellite. The user sends the low-dimensional vector s output by the encoder. k Feedback is sent to the satellite.

[0130] Step 4: The satellite reconstructs the low-dimensional vector into a high-dimensional CSI matrix using a decoder. The satellite then receives the low-dimensional vector s from the feedback. kInput is fed into the decoder, and output is the reconstructed user CSI matrix.

[0131] Step 5: The satellite integrates the reconstructed user matrix into a system channel matrix, and then transforms the reconstructed user channel matrix into the space-time delay domain h′ using a two-dimensional Fourier transform. k :

[0132]

[0133] Integrate the user channel matrix to form the system channel matrix

[0134]

[0135] This refers to the feedback CSI obtained by the satellite.

[0136] Traditional CSI feedback techniques compress the CSI channel using conventional compression algorithms, but these algorithms can only capture linear relationships in the data. Since satellite channels exhibit many nonlinear variations that cannot be captured by traditional compression algorithms, their compression ratios are relatively low.

[0137] The channel state information (CSI) feedback method for multi-user communication systems provided in this invention can reduce the CSI feedback burden of satellite massive MIMO (Multi-Beam Multi-User) systems and improve the communication system capacity. It also improves the accuracy of satellites obtaining reconstructed CSI. By capturing nonlinear data relationships in CSI data through machine learning algorithms, more efficient compression is achieved, reducing the amount of CSI feedback data and increasing system communication capacity.

[0138] The channel state information feedback device for a multi-user communication system provided by the present invention is described below. The channel state information feedback device for a multi-user communication system described below can be referred to in correspondence with the channel state information feedback method for a multi-user communication system described above.

[0139] Figure 8 This is a schematic diagram of a channel state information feedback device for a multi-user communication system provided in an embodiment of the present invention, as shown below. Figure 8 As shown, the multi-user communication system channel state information feedback device provided in this embodiment of the invention includes:

[0140] Extraction module 801 is used to extract the peak features of the channel signal in the angle-time delay domain and the regional correlation features of the channel signal in the angle-time delay domain;

[0141] The fusion module 802 is used to fuse the peak features and the regional correlation features to obtain a sparse variation feature matrix of the low-orbit satellite channel as the satellite moves to different orbital positions;

[0142] Input module 803 is used to input the sparsity variation feature matrix of the low-orbit satellite channel as the satellite moves to different orbital positions into the channel state information model to obtain a low-dimensional vector;

[0143] The reconstruction module 804 is used to send the low-dimensional vector to the satellite so that the satellite can reconstruct the low-dimensional vector into a high-dimensional channel state information matrix through the channel state information model.

[0144] The channel state information model is obtained by training a high-dimensional channel state information matrix.

[0145] The multi-user communication system channel state information feedback device provided in this invention extracts the peak features and regional correlation features of the channel signal in the angle-delay domain. The peak features and regional correlation features are fused to obtain a sparse variation feature matrix of the low-Earth orbit (LEO) satellite channel as the satellite moves to different orbital positions. This sparse variation feature matrix is ​​input into a channel state information model to obtain a low-dimensional vector. The low-dimensional vector is then sent to the satellite, allowing the satellite to reconstruct a high-dimensional channel state information matrix using the channel state information model. The channel state information model is trained based on the high-dimensional channel state information matrix. By compressing and reconstructing the channel state information (CSI) using the model, the overhead of CSI display feedback is reduced, improving system communication performance. The channel state information model can extract changing data features from the channel matrix, improving the accuracy of CSI reconstruction and further enhancing system performance.

[0146] Figure 9 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 9As shown, the electronic device may include: a processor 910, a communication interface 920, a memory 930, and a communication bus 940, wherein the processor 910, the communication interface 920, and the memory 930 communicate with each other through the communication bus 940. The memory 930 includes a computer program, an operating system, and acquired data. The processor 910 can call logical instructions in the memory 930 to execute a multi-user communication system channel state information feedback method. This method includes: extracting peak features of the channel signal in the angle-delay domain and regional correlation features of the channel signal in the angle-delay domain; fusing the peak features and the regional correlation features to obtain a sparse variation feature matrix of the low-Earth orbit satellite channel as the satellite moves to different orbital positions; inputting the sparse variation feature matrix of the low-Earth orbit satellite channel as the satellite moves to different orbital positions into a channel state information model to obtain a low-dimensional vector; sending the low-dimensional vector to the satellite so that the satellite can reconstruct the low-dimensional vector into a high-dimensional channel state information matrix through the channel state information model; wherein, the channel state information model is trained based on the high-dimensional channel state information matrix.

[0147] Furthermore, the logical instructions in the aforementioned memory 930 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 related technologies, or a portion 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.

[0148] On the other hand, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a method for feedback of channel state information in a multi-user communication system provided by the methods described above. The method includes: extracting peak features of the channel signal in the angle-delay domain and regional correlation features of the channel signal in the angle-delay domain; fusing the peak features and the regional correlation features to obtain a sparse variation feature matrix of the low-Earth orbit satellite channel as the satellite moves to different orbital positions; inputting the sparse variation feature matrix of the low-Earth orbit satellite channel as the satellite moves to different orbital positions into a channel state information model to obtain a low-dimensional vector; and sending the low-dimensional vector to the satellite so that the satellite can reconstruct the low-dimensional vector into a high-dimensional channel state information matrix through the channel state information model; wherein the channel state information model is trained based on the high-dimensional channel state information matrix.

[0149] 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.

[0150] 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 parts that contribute to the related technology, can be embodied in the form of software products. 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.

[0151] 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 method for channel state information feedback in a multi-user communication system, characterized in that, include: Extract the peak characteristics of the channel signal in the angle-delay domain and the regional correlation characteristics of the channel signal in the angle-delay domain; By fusing the peak features and the regional correlation features, a sparse variation feature matrix of the low-Earth orbit satellite channel as the satellite moves to different orbital positions is obtained. The sparse variation feature matrix of the low-Earth orbit satellite channel as it changes with different orbital positions is input into the channel state information model, which includes an encoder. The encoder is set at the user end and is used to compress the sparse variation feature matrix in the angle-delay domain, output the compressed low-dimensional channel state information vector, and feed it back to the satellite. During the encoder training process of the channel state information model, the sparse variation feature matrix in the input angle-delay domain is passed through two convolutional layers and then through a lightweight self-attention feature extraction module to obtain a high-dimensional feature map. The high-dimensional feature map includes feature information of the peak and surrounding information. The high-dimensional feature map is then passed through two more convolutional layers for dimensionality transformation to obtain a high-dimensional vector. The high-dimensional vector is then dimensionality-reduced using a fully connected layer to obtain a low-dimensional vector. The channel state information model includes a decoder; the decoder is located at the satellite end and is used to reconstruct low-dimensional vectors; The low-dimensional vector is sent to the satellite so that the satellite can reconstruct the low-dimensional vector into a high-dimensional channel state information matrix using the channel state information model. During the decoder training process of the channel state information model, the low-dimensional vector is converted into a high-dimensional vector through a fully connected layer; the high-dimensional vector is transformed by dimension to obtain a high-dimensional feature matrix; the high-dimensional feature matrix is ​​convolved through four convolutional layers, with the number of convolution kernels in the four convolutional layers gradually increasing to obtain features for multiple channels. The features of the multiple channels are aggregated and reconstructed into a high-dimensional channel state information matrix.

2. The method for feedback channel state information in a multi-user communication system according to claim 1, characterized in that, The extraction of peak features of the channel signal in the angle-delay domain includes: The vertical and horizontal information of the input matrix corresponding to the channel signal is extracted by pooling layers with different kernel sizes; The information in the vertical and horizontal directions is fused into a feature map that simultaneously contains information in the angle domain and the time delay domain. Extract the peak features of the feature map that simultaneously contains information in the angle domain and the time delay domain.

3. The method for feedback channel state information in a multi-user communication system according to claim 1, characterized in that, The extraction of regional correlation features of the channel signal in the angle-delay domain includes: The input matrix corresponding to the channel signal is convolved through two convolutional layers, and then the channel signal is processed through two channels. One channel expands the feature matrix after the convolution operation to obtain the first matrix, and the rows of the first matrix are summed to obtain the first column vector. The second and third column vectors are obtained by performing a dimensionality transformation on the feature matrix after the convolution operation through another channel. The second and third column vectors are then multiplied by the first column vector, and the results are concatenated along the third dimension to generate the second matrix. The second matrix is ​​reconstructed to have the same dimension as the input matrix, and the regional correlation characteristics of the channel signal in the angle-delay domain are output.

4. The channel state information feedback method for a multi-user communication system according to claim 3, characterized in that, The process of expanding the feature matrix after convolution to obtain the first matrix includes: The feature matrix is ​​expanded by mirroring, adding two rows and two columns of elements. Set a pre-defined aggregation region and use it to move row by row and column by column on the expanded matrix to perform raster scanning. Recombine the scanned region into a row vector with the number of elements equal to the product of the region height, region width and number of channels. Superimpose all the scanned and deformed row vectors to obtain the first matrix.

5. The method for feedback channel state information in a multi-user communication system according to claim 1, characterized in that, The fusion of the peak feature and the regional correlation feature includes: A lightweight feature fusion layer is used to fuse the peak feature and the region correlation feature. The new mapping feature H4 output by the feature fusion layer is: H4 = ηH1 + γH2 + μH3; Wherein, H1 is the input matrix corresponding to the channel signal, H2 is the peak feature matrix, H3 is the regional correlation feature matrix, and η, γ, and μ are the weight parameters of the input matrix, peak feature matrix, and regional correlation feature matrix corresponding to the channel signal, respectively. η, γ, and μ are obtained through model training.

6. A channel state information feedback device for a multi-user communication system, characterized in that, include: The extraction module is used to extract the peak features of the channel signal in the angle-time delay domain and the regional correlation features of the channel signal in the angle-time delay domain; The fusion module is used to fuse the peak features and the regional correlation features to obtain a sparse variation feature matrix of the low-orbit satellite channel as the satellite moves to different orbital positions; An input module is used to input the sparse variation feature matrix of the low-Earth orbit satellite channel, which varies with the satellite's orbital position, into a channel state information model. The channel state information model includes an encoder. The encoder, located at the user end, compresses the input sparse variation feature matrix in the angle-delay domain, outputs a compressed low-dimensional channel state information vector, and feeds it back to the satellite. During encoder training of the channel state information model, the input sparse variation feature matrix in the angle-delay domain is passed through two convolutional layers, then through a lightweight self-attention feature extraction module to obtain a high-dimensional feature map. The high-dimensional feature map includes peak values ​​and surrounding information. The high-dimensional feature map is then passed through two more convolutional layers for dimensionality transformation to obtain a high-dimensional vector. A fully connected layer is used to reduce the dimensionality of the high-dimensional vector to obtain a low-dimensional vector. A reconstruction module is used to send the low-dimensional vector to the satellite, so that the satellite can reconstruct the low-dimensional vector into a high-dimensional channel state information matrix through the channel state information model. The channel state information model includes a decoder. The decoder is set at the satellite end and is used to reconstruct the low-dimensional vector. During the training process of the decoder of the channel state information model, the low-dimensional vector is converted into a high-dimensional vector through a fully connected layer. The high-dimensional vector is transformed by dimensionality to obtain a high-dimensional feature matrix. The high-dimensional feature matrix is ​​convolved through four convolutional layers, and the number of convolutional kernels of the four convolutional layers gradually increases to obtain features of multiple channels. The features of the multiple channels are aggregated and reconstructed into a high-dimensional channel state information matrix.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the channel state information feedback method for a multi-user communication system as described in any one of claims 1 to 5.

8. A non-transitory readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the channel state information feedback method for a multi-user communication system as described in any one of claims 1 to 5.

Images