Channel estimation methods, apparatus, and computer equipment for multiple users

Abstract

This invention relates to the field of wireless communication technology, and in particular to a channel estimation method, apparatus, and computer device for multiple users. It fully exploits the spatial correlation of channels between geographically adjacent users and the narrow-angle spread characteristics in the angular domain, aiming to improve channel estimation accuracy under limited pilot resources. The method decomposes the channels of adjacent users into a common channel component shared by all users and a unique channel component for each user. By utilizing the characteristics of the common channel component in the angular domain, which has concentrated energy and a significant structure, the common channel component shared by all users is estimated first. This common channel component is used as prior information to assist in the recovery of the complete channel for each user.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a channel estimation method, apparatus, and computer device for multiple users. Background Technology

[0002] With the rapid development of mobile communication technology, very large-scale multiple-input multiple-output (MIMO) systems have become one of the core technologies for future sixth-generation wireless communication. This technology, by configuring hundreds or thousands of antennas at the base station to serve multiple user terminals on the same time-frequency resources, can greatly improve the system's spectral efficiency, energy efficiency, and transmission reliability. However, the full realization of the theoretical gains of MIMO systems heavily depends on accurate channel state information.

[0003] In time-division duplex systems, downlink channels can be obtained through uplink channel estimation based on channel reciprocity. To achieve multi-user channel estimation, each user needs to send orthogonal pilot sequences to the base station within a coherent time frame. Utilizing the orthogonality between pilots, various algorithms, such as least squares estimation, can be used to extract the channel for each user from the received signal.

[0004] However, as the number of users increases, the required length of orthogonal pilots increases accordingly, leading to a significant increase in pilot overhead. Furthermore, the time-varying nature of the channel limits the number of pilots available within the coherence time. When the number of users exceeds this limit, multiple users will be forced to use non-orthogonal pilots. In this case, channel information from multiple users is compressed and superimposed. Traditional least-squares estimation algorithms, relying on the invertibility of the pilot matrix, can only approximate the solution using pseudo-inverses under non-orthogonal conditions. This process cannot effectively separate the target channel from user interference, thus directly introducing interference components into the estimation results. Summary of the Invention

[0005] Based on this, the purpose of this invention is to provide a channel estimation method, apparatus, computer device, and storage medium for multiple users, which fully exploits the channel spatial correlation between geographically adjacent users and the narrow-angle spread characteristics in the angle domain. The aim is to improve the channel estimation accuracy under limited pilot resources. The method decomposes the adjacent user channels into a common channel component shared by all users and a unique channel component for each user. By utilizing the characteristics of the common channel component in the angle domain, which has concentrated energy and significant structure, the common channel component shared by all users is estimated first. This common channel component is used as prior information to assist in the recovery of the complete channel for each user.

[0006] In a first aspect, embodiments of this application provide a channel estimation method for multiple users, comprising the following steps:

[0007] The system obtains the superimposed communication signal and pilot sequence matrix received by the base station in the target area. The superimposed communication signal is the result of superimposing the uplink signals sent by all users through the uplink channel at the current time. The pilot sequence matrix is ​​used to represent the pilot sequences of all users.

[0008] The superimposed communication signal is subjected to angle-domain tensor transformation to obtain the angle-domain received tensor;

[0009] The angle domain receiving tensor is input into a preset common channel component extraction network to extract channel components and obtain common channel components.

[0010] Based on the common channel components and the pilot sequence matrix, the common channel component influence removal is performed on the angle domain receiving tensor to obtain the processed angle domain receiving tensor.

[0011] The processed angle domain receiving tensor is input into a preset user-specific channel component extraction network for channel component extraction to obtain a user-specific channel component representation, wherein the user-specific channel component representation includes the user-specific channel components of each user.

[0012] Complete channel construction is performed based on the user-specific channel component representation and the common channel component to obtain the complete channel for each user.

[0013] Secondly, embodiments of this application provide a channel estimation apparatus for multiple users, comprising:

[0014] The signal acquisition module is used to acquire the superimposed communication signal and pilot sequence matrix received by the base station in the target area. The superimposed communication signal is the superposition result of the uplink signals sent by all users through the uplink channel at the current time. The pilot sequence matrix is ​​used to represent the pilot sequences of all users.

[0015] The tensor conversion module is used to perform angle domain tensor conversion on the superimposed communication signal to obtain the angle domain received tensor.

[0016] The common channel component extraction module is used to input the angle domain receiving tensor into a preset common channel component extraction network to extract channel components and obtain common channel components.

[0017] The channel component processing module is used to remove the influence of the common channel component on the angle domain received tensor according to the common channel component and the pilot sequence matrix, so as to obtain the processed angle domain received tensor.

[0018] The user-specific channel component extraction module is used to input the processed angle domain receiving tensor into a preset user-specific channel component extraction network for channel component extraction to obtain a user-specific channel component representation, wherein the user-specific channel component representation includes the user-specific channel components of each user.

[0019] The channel component reassembly module is used to construct a complete channel based on the user-specific channel components in the user-specific channel component representation and the common channel components, thereby obtaining the complete channel for each user.

[0020] Thirdly, embodiments of this application provide a computer device, including: a processor, a memory, and a computer program stored in the memory and executable on the processor; when the computer program is executed by the processor, it implements the steps of the multi-user channel estimation method as described in the first aspect.

[0021] Fourthly, embodiments of this application provide a storage medium storing a computer program that, when executed by a processor, implements the steps of the multi-user channel estimation method described in the first aspect.

[0022] This application provides a channel estimation method, apparatus, computer device, and storage medium for multiple users. It fully exploits the channel spatial correlation between geographically adjacent users and the narrow-angle spread characteristics in the angle domain, aiming to improve channel estimation accuracy under limited pilot resources. The method decomposes the adjacent user channels into a common channel component shared by all users and a unique channel component for each user. By utilizing the characteristics of the common channel component in the angle domain, which has concentrated energy and significant structure, the common channel component shared by all users is estimated first. This common channel component is used as prior information to assist in the recovery of the complete channel for each user.

[0023] To better understand and implement this invention, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a multi-user channel estimation method provided in one embodiment of this application.

[0025] Figure 2 This is a flowchart illustrating step S2 in a multi-user channel estimation method provided in one embodiment of this application.

[0026] Figure 3 This is a flowchart illustrating step S3 in a multi-user channel estimation method provided in one embodiment of this application.

[0027] Figure 4This is a flowchart illustrating step S5 of a multi-user channel estimation method provided in one embodiment of this application.

[0028] Figure 5 A flowchart illustrating step S7 in a multi-user channel estimation method provided in another embodiment of this application;

[0029] Figure 6 A flowchart illustrating step S8 in a multi-user channel estimation method provided in yet another embodiment of this application;

[0030] Figure 7 This is a schematic diagram of the structure of a multi-user channel estimation device provided in one embodiment of this application;

[0031] Figure 8 This is a schematic diagram of the structure of a computer device provided in one embodiment of this application. Detailed Implementation

[0032] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0033] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0034] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0035] Please see Figure 1 , Figure 1 The following is a flowchart illustrating a multi-user channel estimation method provided in one embodiment of this application. The method includes the following steps:

[0036] S1: Obtain the superimposed communication signals and pilot sequence matrix received by the base station in the target area.

[0037] The execution entity of the multi-user channel estimation method of this application is an estimation device for multi-user channel estimation (hereinafter referred to as the estimation device). In an optional embodiment, the estimation device may be a computer device, a server, or a server cluster composed of multiple computer devices.

[0038] In this embodiment, the estimation device obtains the superimposed communication signal and pilot sequence matrix received by the base station in the target area. The superimposed communication signal is the superposition result of the uplink signals sent by all users through the uplink channel at the current time. The pilot sequence matrix is ​​used to represent the pilot sequences of all users.

[0039] S2: Perform angle domain tensor transformation on the superimposed communication signal to obtain the angle domain received tensor.

[0040] Since geographically adjacent users share most of the scatterers, the majority of energy in their angular domain channels is concentrated in the common channel component, and their non-zero energy clusters are located in roughly the same region. In contrast, each user's unique channel component has relatively weaker energy, and its non-zero values ​​are more sparsely distributed in the angular domain.

[0041] In this embodiment, the estimation device performs angle-domain tensor transformation on the superimposed communication signal to obtain the angle-domain received tensor.

[0042] Please see Figure 2 , Figure 2 The flowchart of S1 in a multi-user channel estimation method provided in one embodiment of this application includes steps S21 to S22, as follows:

[0043] S21: Multiply the superimposed communication signal with a preset unitary matrix to obtain the angle domain observation signal.

[0044] In this embodiment, the estimation device multiplies the superimposed communication signal with a preset unitary matrix to obtain the angle domain observation signal.

[0045] S22: Extract the real and imaginary parts of the angle domain observation signal to obtain the real and imaginary parts of the angle domain observation signal, and stack the real and imaginary parts of the angle domain observation signal to obtain the angle domain received tensor.

[0046] In this embodiment, the estimation device extracts the real and imaginary parts of the angle domain observation signal to obtain the real and imaginary parts of the angle domain observation signal. The real and imaginary parts of the angle domain observation signal are stacked to obtain the angle domain received tensor. By utilizing the characteristics of strong energy and significant structure of the common channel component and user-specific channel component in the multi-user angle domain channel, and the weak energy and high sparsity of the user-specific channel component, the accuracy of multi-user channel estimation is improved.

[0047] S3: Input the angle domain receiving tensor into a preset common channel component extraction network to extract channel components and obtain common channel components.

[0048] In this embodiment, the estimation device inputs the angle domain receiving tensor into a preset common channel component extraction network to extract channel components and obtain common channel components.

[0049] The common channel component extraction network includes a first data preprocessing unit, a common feature extraction unit, and a first feature distillation unit; the common feature extraction unit includes several stacked convolutional residual blocks and spatial attention blocks; please refer to... Figure 3 , Figure 3 The flowchart of S3 in a multi-user channel estimation method provided in one embodiment of this application is shown below, including steps S31 to S34, as follows:

[0050] S31: Input the angle domain receiving tensor into the first data preprocessing unit for nearest neighbor interpolation to obtain the expanded first angle domain receiving tensor; perform a linear transformation on the expanded first angle domain receiving tensor to obtain the first intermediate feature tensor; perform nonlinear feature deepening processing on the first intermediate feature tensor to obtain the first depth feature tensor.

[0051] In this embodiment, the estimation device inputs the angle domain received tensor into the first data preprocessing unit for nearest neighbor interpolation to obtain the expanded first angle domain received tensor, expanding the signal dimension to match the number of users, so as to adapt to subsequent network processing and construct a structurally regular input.

[0052] The estimation device performs a linear transformation on the extended first angle domain received tensor to obtain a first intermediate feature tensor. Specifically, the estimation device uses a fully connected layer with shared angle dimension weights to perform a linear transformation on the extended first angle domain received tensor to obtain a first intermediate feature tensor, so as to utilize the shared patterns between adjacent angles.

[0053] The estimation device performs nonlinear feature deepening processing on the first intermediate feature tensor to obtain a first depth feature tensor. Specifically, based on the first intermediate feature tensor and a preset nonlinear feature deepening algorithm, the estimation device maps the first intermediate feature tensor to 128 dimensions through a fully connected layer with shared angle dimension weights. Then, batch normalization and ReLU activation are applied, and the same operation is performed to map it back to 256 dimensions. Finally, it is mapped back to 128 dimensions to obtain the depth feature tensor, thus fully utilizing the narrow-angle diffusion characteristics of the angle domain to enhance feature representation capabilities.

[0054] S32: Input the first depth feature tensor into the common feature extraction unit, process it through several stacked convolutional residual blocks, and obtain the first convolutional residual feature tensor output by the last convolutional residual block.

[0055] In this embodiment, the estimation device inputs the first depth feature tensor into the common feature extraction unit, processes it through several stacked convolutional residual blocks, and obtains the first convolutional residual feature tensor output by the last convolutional residual block.

[0056] Specifically, the common feature extraction unit adopts a residual block structure that alternates between conventional convolutional residual blocks and vertical convolutional residual blocks. The first convolutional residual block is a conventional convolutional residual block, the second convolutional residual block is a vertical convolutional residual block, the third convolutional residual block is a conventional convolutional residual block, the fourth convolutional residual block is a vertical convolutional residual block, and the fifth convolutional residual block is a conventional convolutional residual block. The conventional convolutional residual block is responsible for capturing the spatial correlation between adjacent users, while the vertical convolutional residual block is specifically designed to enhance the narrow-angle diffusion characteristics between adjacent angles.

[0057] The estimation device uses the first depth feature tensor as the input tensor of the first convolutional residual block, which uses a 5×5 convolution. The estimation device expands the input first depth feature tensor from 2 channels to 32 channels and uses batch normalization and ReLU activation to obtain the first-level feature tensor of the main path of the regular convolutional residual block.

[0058] The estimation device then uses 5×5 convolution and batch normalization to process the first-level feature tensor of the main path of the conventional convolutional residual block, obtaining the second-level feature tensor of the main path of the conventional convolutional residual block. The estimation device performs residual concatenation between the second-level feature tensor of the main path of the conventional convolutional residual block and the first depth feature tensor after convolution alignment, to obtain the output tensor of the first convolutional residual block, in order to capture the local correlation patterns of angle and user two-dimensional plane, thereby enhancing the spatial correlation of shared channel features.

[0059] The estimation device uses the output tensor of the first convolutional residual block as the input tensor of the second convolutional residual block. The vertical convolutional residual block adopts a 5×1 vertical convolution. The estimation device performs a 5×1 convolution on the output tensor of the first input convolutional residual block, keeping the number of channels unchanged, and uses batch normalization and ReLU activation to obtain the first-level feature tensor of the main path of the vertical convolutional residual block.

[0060] The estimation device then uses 5×1 convolution and batch normalization to process the first-level feature tensor of the main path of the vertical convolutional residual block to obtain the second-level feature tensor of the main path of the vertical convolutional residual block. The estimation device then performs a residual concatenation between the second-level feature tensor of the main path of the conventional convolutional residual block and the output tensor of the first convolutional residual block to obtain the output tensor of the second convolutional residual block, thereby enhancing the narrow-angle diffusion characteristics of the features in the angular domain.

[0061] The estimation device uses the output tensor of the second convolutional residual block as the input tensor of the third convolutional residual block, and repeats the convolutional residual processing until the feature channel number is mapped back from 32 channels to 2 channels in the last convolutional residual block, thus obtaining the first convolutional residual feature tensor output by the last convolutional residual block.

[0062] S33: Based on the first convolutional residual feature tensor and the spatial attention block, perform spatial attention weight extraction and feature weighting to obtain a weighted common feature tensor; fuse the first depth feature tensor and the weighted common feature tensor to obtain a first fused feature tensor; based on the first fused feature tensor and the spatial attention block, perform spatial attention weight extraction and feature weighting to obtain a first refined feature tensor.

[0063] In this embodiment, the estimation device extracts spatial attention weights and performs feature weighting based on the first convolutional residual feature tensor and the spatial attention block to obtain a weighted common feature tensor.

[0064] Specifically, the estimation device uses the first convolutional residual feature tensor as the input tensor of the spatial attention block, employs a 7×1 convolution, aggregates features along the angular dimension of the input tensor, compresses the number of channels from 2 to 1, and passes the convolution output through a Sigmoid activation function to map the score at each position to the [0, 1] interval, obtaining a spatial attention weight map. The estimation device then broadcasts the generated spatial attention weight map along the channel dimension to the input tensor of the spatial attention block for element-wise multiplication, obtaining a weighted common feature tensor to adaptively enhance key angular regions. The spatial attention weight map is as follows:

[0065]

[0066] In the formula, C is the spatial attention weight map. Let X be a 7×1 convolution function, and let X be the input tensor of the spatial attention block.

[0067] The estimation device fuses the first depth feature tensor and the weighted common feature tensor through skip connections to obtain a first fused feature tensor. The estimation device then extracts spatial attention weights and performs feature weighting based on the first fused feature tensor and spatial attention blocks to obtain a first refined feature tensor. Specific embodiments can refer to the above steps and will not be repeated here.

[0068] S34: Input the first refined feature tensor into the first feature distillation unit for nonlinear transformation and compression to obtain the common channel component.

[0069] In this embodiment, the estimation device inputs the first refined feature tensor into the first feature distillation unit for nonlinear transformation and compression, extracts the channel components common to all users, and obtains the common channel components.

[0070] Specifically, the estimation device inputs the first refined feature tensor into the first feature distillation unit, maps the first refined feature tensor to 1024 dimensions through a fully connected layer, and after batch normalization and ReLU activation, maps it back to 128 dimensions, performing batch normalization and ReLU activation again; then it is further compressed to 64 dimensions, again undergoing batch normalization and ReLU activation; finally, through a fully connected layer from 64 dimensions to 1 dimension, the final output is a common channel component shared by all users and consistent in the angle domain. Through the "weight sharing mapping" and "hybrid convolution decomposition" mechanisms, the fully connected layer with angle dimension weight sharing significantly reduces the number of parameters, forcing the network to focus on common patterns between angles; while the residual structure of alternating conventional convolution and vertical convolution captures the spatial correlation between adjacent users and the narrow-angle diffusion characteristics in the angle domain, respectively, achieving a good balance between estimation accuracy and computational complexity.

[0071] S4: Based on the common channel components and the pilot sequence matrix, remove the influence of the common channel components from the angle domain receiving tensor to obtain the processed angle domain receiving tensor.

[0072] In this embodiment, the estimation device performs common channel component influence removal on the angle domain received tensor based on the common channel component and the pilot sequence matrix to obtain a processed angle domain received tensor, wherein the processed angle domain received tensor is:

[0073]

[0074] In the formula, To receive the tensor in the processed angle domain, For receiving tensors in the angle domain, Let P be the common channel component and P be the pilot sequence matrix.

[0075] S5: Input the processed angle domain receiving tensor into a preset user-specific channel component extraction network to extract channel components and obtain user-specific channel component representation.

[0076] In this embodiment, the estimation device inputs the processed angle domain reception tensor into a preset user-specific channel component extraction network to extract channel components and obtain user-specific channel component representations, wherein the user-specific channel component representations include the user-specific channel components of each user.

[0077] The user-specific channel component extraction network includes a second data preprocessing unit, a user-specific feature extraction unit, and a second feature distillation unit; the user-specific feature extraction unit includes several stacked convolutional residual blocks and spatial attention blocks; please refer to... Figure 4 , Figure 4 The flowchart of S5 in the multi-user channel estimation method provided in one embodiment of this application is shown below, including steps S51 to S54, as follows:

[0078] S51: Input the angle domain receiving tensor into the second data preprocessing unit for nearest neighbor interpolation to obtain the extended second angle domain receiving tensor; perform a linear transformation on the extended second angle domain receiving tensor to obtain the second intermediate feature tensor; perform nonlinear feature deepening processing on the second intermediate feature tensor to obtain the second depth feature tensor.

[0079] In this embodiment, the estimation device inputs the angle domain reception tensor into the second data preprocessing unit for nearest neighbor interpolation to obtain the expanded second angle domain reception tensor.

[0080] The estimation device performs a linear transformation on the extended second angle domain received tensor to obtain a second intermediate feature tensor.

[0081] The estimation device performs nonlinear feature deepening processing on the second intermediate feature tensor to obtain a second deep feature tensor, increasing the intermediate hidden dimension of the nonlinear feature deepening to 2048 dimensions to enhance the modeling capability of user-specific complex structures. For specific implementation, please refer to step S31, which will not be repeated here.

[0082] S52: Input the second depth feature tensor into the user-specific feature extraction unit, process it through several stacked convolutional residual blocks, and obtain the second convolutional residual feature tensor output by the last convolutional residual block.

[0083] In this embodiment, the estimation device inputs the second depth feature tensor into the user-specific feature extraction unit, processes it through several stacked convolutional residual blocks, and obtains the second convolutional residual feature tensor output by the last convolutional residual block. Specifically, the stacked convolutional residual blocks in the user-specific feature extraction unit use five vertical convolutional residual blocks to extract local features only along the angular dimension. By mining the unique features of different users in the angular dimension through vertical convolution, feature interaction in the user feature dimension is avoided, thereby ensuring that the extracted features reflect the unique channel structure of each user. For specific implementation, please refer to step S32, which will not be repeated here.

[0084] S53: Based on the second convolutional residual feature tensor and the spatial attention block, perform spatial attention weight extraction and feature weighting to obtain a weighted user feature tensor; fuse the second depth feature tensor and the weighted user feature tensor to obtain a second fused feature tensor; based on the second fused feature tensor and the spatial attention block, perform spatial attention weight extraction and feature weighting to obtain a second refined feature tensor.

[0085] In this embodiment, the estimation device extracts spatial attention weights and performs feature weighting based on the second convolutional residual feature tensor and the spatial attention block to obtain a weighted user feature tensor. The estimation device fuses the second depth feature tensor and the weighted user feature tensor to obtain a second fused feature tensor. The estimation device extracts spatial attention weights and performs feature weighting based on the second fused feature tensor and the spatial attention block to obtain a second refined feature tensor. For a specific implementation, please refer to step S33, which will not be repeated here.

[0086] S54: Input the second refined feature tensor into the second feature distillation unit for nonlinear transformation and compression to obtain the user-specific channel component.

[0087] In this embodiment, the estimation device inputs the second refined feature tensor into the second feature distillation unit for nonlinear transformation and compression to obtain the user-specific channel components. Compared with the first feature distillation unit, the output dimension of the last fully connected layer of the second feature distillation unit is the number of users, so as to output the user-specific channel components for each user and construct the user-specific channel components.

[0088] S6: Construct a complete channel based on the user-specific channel component and the common channel component in the user-specific channel component representation to obtain the complete channel for each user.

[0089] In this embodiment, the estimation device constructs a complete channel based on the user-specific channel components in the user-specific channel component representation and the common channel components, thereby obtaining the complete channel for each user.

[0090] Specifically, the estimation device adds the user-specific channel components in the user-specific channel component representation to the common channel components element by element to obtain the complete channel for each user.

[0091] By fully exploring the channel spatial correlation between geographically adjacent users and the narrow-angle spread characteristics in the angle domain, this study aims to improve channel estimation accuracy under limited pilot resources. The channel of adjacent users is decomposed into a common channel component shared by all users and a unique channel component of each user. Taking advantage of the energy concentration and significant structure of the common channel component in the angle domain, the common channel component shared by all users is estimated first. This common channel component is used as prior information to help recover the complete channel of each user.

[0092] In an optional embodiment, step S7 is further included: training the common channel component extraction network to be trained; before step S3, please refer to [link to previous section]. Figure 5 , Figure 5 The flowchart of S7 in the multi-user channel estimation method provided in another embodiment of this application includes steps S71 to S74, as follows:

[0093] S71: Obtain several batches of pilot sequence matrices, uplink channel matrices, and common channel tag sets received by base stations in the target area.

[0094] In this embodiment, the estimation device obtains several batches of pilot sequence matrices, uplink channel matrices, and common channel tag sets received by base stations in the target area. The pilot sequence matrix includes pilot sequences of several sample users; the uplink channel matrix includes the uplink channels of each sample user; and the common channel tag set includes common channel tag components of each sample user.

[0095] Specifically, the estimation device adopts the single-ring model in the Geometric Random Channel Model (GSCM). Considering the characteristic that adjacent users share most of the scatterers in high-density user scenarios, the channel of each user is modeled as the sum of the common channel component from the shared scatterers and the exclusive channel component from each user's unique scatterers. That is, the uplink channel from sample user k to the base station can be represented as the superposition of various paths on other surrounding scatterers, as described below:

[0096]

[0097] In the formula, For the uplink channel of the k-th sample user, For the number of paths, Let be the complex gain at the base station for the i-th path of the k-th sample user. Let be the angle of arrival at the base station for the i-th path of the k-th sample user. This is the guidance vector for the i-th path of the k-th sample user.

[0098] S72: The signals are superimposed according to the pilot sequence matrix and uplink channel matrix of each batch to obtain the superimposed communication signal of each batch; the superimposed communication signal of each batch is transformed into an angle domain tensor to obtain the angle domain received tensor of each batch.

[0099] In this embodiment, the estimation device performs signal superposition based on the pilot sequence matrix and uplink channel matrix of each batch to obtain the superimposed communication signal of each batch, wherein the superimposed communication signal is:

[0100]

[0101] In the formula, Y represents the superimposed communication signal, and K represents the number of sample users. Let N be the pilot sequence of the k-th sample user, and let N be the additive white Gaussian noise at the base station receiver.

[0102] The estimation device performs angle domain tensor transformation on the superimposed communication signals of each batch to obtain the angle domain received tensor of each batch. For specific embodiments, please refer to steps S21 to S22, which will not be repeated here.

[0103] S73: Input the angle domain receiving tensor of each batch into the common channel component extraction network to be trained to extract channel components and obtain the common channel components of each batch.

[0104] In this embodiment, the estimation device inputs the angle domain received tensor of each batch to the common channel component extraction network to be trained for channel component extraction to obtain the common channel components of each batch. For specific implementation, please refer to steps S31 to S34, which will not be repeated here.

[0105] S74: Obtain the common mean square error loss value for each batch based on the common channel component, common channel label component, and preset common mean square error loss function for each batch; train the common channel component extraction network to be trained based on the common mean square error loss value for each batch to obtain the preset common channel component extraction network.

[0106] In this embodiment, the estimation device obtains the common mean square error loss value for each batch based on the common channel component, common channel tag component, and a preset common mean square error loss function. The common mean square error loss function is:

[0107]

[0108] In the formula, This is the common mean square error loss value. For the common channel component of the k-th sample user, For the common channel label component of the k-th sample user, is the regularization strength coefficient, and W is the total weight matrix of the common channel component extraction network, which represents the sum of the absolute values ​​of the parameters involved in the common channel component extraction network.

[0109] The estimation device trains the common channel component extraction network to be trained based on the common mean square error loss value of each batch, and obtains the preset common channel component extraction network.

[0110] In an optional embodiment, step S8 is further included: training the user-specific channel component extraction network to be trained; before step S4, please refer to [link to previous section]. Figure 6 , Figure 6 The flowchart of S8 in the multi-user channel estimation method provided in another embodiment of this application includes steps S81 to S84, as follows:

[0111] S81: Obtain the complete channel label set of each batch of users and the common channel components of each batch output by the common channel component extraction network.

[0112] In this embodiment, the estimation device obtains the complete channel label set of each batch of users and the common channel components of each batch output by the common channel component extraction network, wherein the complete channel label set of users includes the complete channel labels of each sample user.

[0113] S82: Based on the common channel components extracted from the network output of each batch and the pilot sequence matrix, the common channel component influence removal is performed on the angle domain received tensor of each batch to obtain the processed angle domain received tensor of each batch.

[0114] In this embodiment, the estimation device extracts the common channel components and pilot sequence matrix of each batch of network output based on the common channel components, and removes the influence of the common channel components on the angle domain received tensor of each batch to obtain the processed angle domain received tensor of each batch. For specific implementation, please refer to step S4, which will not be repeated here.

[0115] S83: Input the processed angle domain receive tensor of each batch into the user-specific channel component extraction network to be trained for channel component extraction, and obtain the user-specific channel component representation of each batch.

[0116] In this embodiment, the estimation device inputs the processed angle domain reception tensor of each batch into the user-specific channel component extraction network to be trained for channel component extraction, thereby obtaining the user-specific channel component representation of each batch. For specific implementation, please refer to steps S51 to S54, which will not be repeated here.

[0117] S84: Based on the common channel components, user-specific channel component representations, user complete channel label sets, and preset user complete channel mean square error loss functions output by the common channel component extraction network for each batch, obtain the user complete channel mean square error loss value for each batch; based on the user complete channel mean square error loss value for each batch, train the user-specific channel component extraction network to be trained to obtain the preset user-specific channel component extraction network.

[0118] In this embodiment, the estimation device extracts the common channel components (CMS) of each batch output by the network, the user-specific channel component representations of each batch, the user complete channel label set, and a preset user complete channel mean square error loss function based on the target common channel components, to obtain the user complete channel mean square error loss value for each batch. The user complete channel mean square error loss function is:

[0119]

[0120] In the formula, This represents the user's complete channel mean square error loss value. For the common channel component of the k-th sample user, For the k-th sample user's dedicated channel component, This is the complete channel label for the k-th sample user.

[0121] The estimation device trains the user-specific channel component extraction network to be trained based on the mean square error loss value of the complete channel for each batch of users, thereby obtaining the preset user-specific channel component extraction network. A phased, progressive strategy is adopted to train the common channel component extraction network and the user-specific channel component extraction network. Taking advantage of the concentrated energy and salient structure of common channel components in the angular domain, priority is given to estimating common channel components shared by all users and more easily learned by the network, thus achieving collaborative optimization.

[0122] Please refer to Figure 7 , Figure 7 This is a schematic diagram of a multi-user channel estimation device provided in one embodiment of this application. The device can be implemented entirely or partially through software, hardware, or a combination of both. The multi-user channel estimation device 7 includes:

[0123] The signal acquisition module 71 is used to acquire the communication superimposed signal and pilot sequence matrix received by the base station in the target area, wherein the communication superimposed signal is the superposition result of the uplink signals sent by all users through the uplink channel at the current time received by the base station; the pilot sequence matrix is ​​used to represent the pilot sequences of all users;

[0124] Tensor conversion module 72 is used to perform angle domain tensor conversion on the communication superimposed signal to obtain angle domain received tensor;

[0125] The common channel component extraction module 73 is used to input the angle domain receiving tensor into a preset common channel component extraction network to extract channel components and obtain common channel components.

[0126] The channel component processing module 74 is used to remove the influence of the common channel component on the angle domain receiving tensor according to the common channel component and the pilot sequence matrix, so as to obtain the processed angle domain receiving tensor.

[0127] User-specific channel component extraction module 75 is used to input the processed angle domain receiving tensor into a preset user-specific channel component extraction network for channel component extraction to obtain user-specific channel component representation, wherein the user-specific channel component representation includes the user-specific channel components of each user.

[0128] The channel component reassembly module 76 is used to construct a complete channel based on the user-specific channel components in the user-specific channel component representation and the common channel components, thereby obtaining the complete channel for each user.

[0129] In this embodiment, a signal acquisition module obtains the superimposed communication signal and pilot sequence matrix received by the base station in the target area; a tensor transformation module performs angle-domain tensor transformation on the superimposed communication signal to obtain an angle-domain received tensor; a common channel component extraction module inputs the angle-domain received tensor into a preset common channel component extraction network for channel component extraction to obtain common channel components; a channel component processing module removes the influence of common channel components from the angle-domain received tensor based on the common channel components and the pilot sequence matrix to obtain a processed angle-domain received tensor; a user-specific channel component extraction module inputs the processed angle-domain received tensor into a preset user-specific channel component extraction network for channel component extraction to obtain a user-specific channel component representation, wherein the user-specific channel component representation includes the user-specific channel components of each user; and a channel component reconstruction module constructs a complete channel based on the user-specific channel components of each user in the user-specific channel component representation and the common channel components to obtain the complete channel of each user. By fully exploring the channel spatial correlation between geographically adjacent users and the narrow-angle spread characteristics in the angle domain, this study aims to improve channel estimation accuracy under limited pilot resources. The channel of adjacent users is decomposed into a common channel component shared by all users and a unique channel component of each user. Taking advantage of the energy concentration and significant structure of the common channel component in the angle domain, the common channel component shared by all users is estimated first. This common channel component is used as prior information to help recover the complete channel of each user.

[0130] Please refer to Figure 8 , Figure 8 This is a schematic diagram of the structure of a computer device provided in one embodiment of this application. The computer device 8 includes: a processor 81, a memory 82, and a computer program 83 stored in the memory 82 and executable on the processor 81; the computer device can store multiple instructions, which are adapted to be loaded and executed by the processor 81. Figures 1 to 6 For the method steps and specific execution process, please refer to [link / reference]. Figures 1 to 6 Specific details will not be elaborated here.

[0131] The processor 81 may include one or more processing cores. The processor 81 connects to various parts of the server using various interfaces and lines. It executes instructions, programs, code sets, or instruction sets stored in the memory 82, and calls data from the memory 82 to perform various functions and process data for the multi-user channel estimation device 7. Optionally, the processor 81 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 81 may integrate one or a combination of several of the following: a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), and a modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content displayed on the touch screen; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor 81.

[0132] The memory 82 may include random access memory (RAM) or read-only memory. Optionally, the memory 82 may include a non-transitory computer-readable storage medium. The memory 82 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 82 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch instructions), instructions for implementing the various method embodiments described above, etc.; the data storage area may store data involved in the various method embodiments described above, etc. Optionally, the memory 82 may also be at least one storage device located remotely from the aforementioned processor 81.

[0133] This application embodiment also provides a storage medium that can store multiple instructions, which are adapted to be loaded and executed by a processor as described above. Figures 1 to 6 For the method steps and specific execution process, please refer to [link / reference]. Figures 1 to 6 Specific details will not be elaborated here.

[0134] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0135] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0136] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the algorithm. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0137] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / terminal devices and methods can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0138] 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0139] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0140] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms.

[0141] This invention is not limited to the above-described embodiments. If any modifications or variations to this invention do not depart from the spirit and scope of this invention, and if such modifications and variations fall within the scope of the claims and equivalent technologies of this invention, then this invention also intends to include such modifications and variations.

Claims

1. A channel estimation method for multiple users, characterized in that, Includes the following steps: The system obtains the superimposed communication signal and pilot sequence matrix received by the base station in the target area. The superimposed communication signal is the result of superimposing the uplink signals sent by all users through the uplink channel at the current time. The pilot sequence matrix is ​​used to represent the pilot sequences of all users. The superimposed communication signal is subjected to angle-domain tensor transformation to obtain the angle-domain received tensor; The angle domain receiving tensor is input into a preset common channel component extraction network to extract channel components and obtain common channel components. Based on the common channel components and the pilot sequence matrix, the common channel component influence removal is performed on the angle domain receiving tensor to obtain the processed angle domain receiving tensor. The processed angle domain receiving tensor is input into a preset user-specific channel component extraction network for channel component extraction to obtain a user-specific channel component representation, wherein the user-specific channel component representation includes the user-specific channel components of each user. Complete channel construction is performed based on the user-specific channel component representation and the common channel component to obtain the complete channel for each user.

2. The channel estimation method for multiple users according to claim 1, characterized in that, The step of performing angle-domain tensor transformation on the superimposed communication signal to obtain the angle-domain received tensor includes the following steps: The superimposed communication signal is multiplied by a preset unitary matrix to obtain an angle domain observation signal; The real and imaginary parts of the angle domain observation signal are extracted to obtain the real and imaginary parts of the angle domain observation signal. The real and imaginary parts of the angle domain observation signal are stacked to obtain the angle domain received tensor.

3. The channel estimation method for multiple users according to claim 2, characterized in that: The common channel component extraction network includes a first data preprocessing unit, a common feature extraction unit, and a first feature distillation unit; the common feature extraction unit includes several stacked convolutional residual blocks and spatial attention blocks; The step of inputting the angle domain receiving tensor into a preset common channel component extraction network for channel component extraction to obtain common channel components includes the following steps: The angle domain receiving tensor is input into the first data preprocessing unit for nearest neighbor interpolation to obtain an expanded first angle domain receiving tensor; the expanded first angle domain receiving tensor is linearly transformed to obtain a first intermediate feature tensor; the first intermediate feature tensor is nonlinearly feature-deepening processed to obtain a first depth feature tensor. The first depth feature tensor is input into the common feature extraction unit and processed by several stacked convolutional residual blocks to obtain the first convolutional residual feature tensor output by the last convolutional residual block. Spatial attention weights are extracted and features are weighted based on the first convolutional residual feature tensor and the spatial attention block to obtain a weighted common feature tensor; the first depth feature tensor and the weighted common feature tensor are fused to obtain a first fused feature tensor; Based on the first fused feature tensor and the spatial attention block, spatial attention weights are extracted and features are weighted to obtain the first refined feature tensor; The first refined feature tensor is input into the first feature distillation unit for nonlinear transformation and compression to obtain the common channel component.

4. The channel estimation method for multiple users according to claim 3, characterized in that: The user-specific channel component extraction network includes a second data preprocessing unit, a user-specific feature extraction unit, and a second feature distillation unit; the user-specific feature extraction unit includes several stacked convolutional residual blocks and spatial attention blocks; The step of inputting the processed angle-domain receive tensor into the user-specific channel component extraction network for channel component extraction to obtain the user-specific channel component representation includes the following steps: The angle domain receiving tensor is input into the second data preprocessing unit for nearest neighbor interpolation to obtain an expanded second angle domain receiving tensor; the expanded second angle domain receiving tensor is linearly transformed to obtain a second intermediate feature tensor; the second intermediate feature tensor is nonlinearly feature-deepening processed to obtain a second depth feature tensor. The second depth feature tensor is input into the user-specific feature extraction unit and processed through several stacked convolutional residual blocks to obtain the second convolutional residual feature tensor output by the last convolutional residual block. Spatial attention weights are extracted and features are weighted based on the second convolutional residual feature tensor and the spatial attention block to obtain a weighted user feature tensor; the second depth feature tensor and the weighted user feature tensor are fused to obtain a second fused feature tensor; Based on the second fused feature tensor and the spatial attention block, spatial attention weights are extracted and features are weighted to obtain the second refined feature tensor; The second refined feature tensor is input into the second feature distillation unit for nonlinear transformation and compression to obtain the user-specific channel component.

5. The channel estimation method for multiple users according to claim 4, characterized in that, Before inputting the angle domain receiving tensor into a preset common channel component extraction network for channel component extraction and obtaining the common channel components, the method further includes the following steps: Train the common channel component extraction network to be trained; The training of the common channel component extraction network to be trained includes the following steps: The system obtains several batches of pilot sequence matrices, uplink channel matrices, and common channel tag sets received by base stations in the target area. The pilot sequence matrix includes pilot sequences from several sample users; the uplink channel matrix includes the uplink channels of each sample user; and the common channel tag set includes common channel tag components of each sample user. Signals are superimposed based on the pilot sequence matrix and uplink channel matrix of each batch to obtain the superimposed communication signal of each batch; angle domain tensor transformation is performed on the superimposed communication signal of each batch to obtain the angle domain received tensor of each batch. The angle domain receiving tensor of each batch is input into the common channel component extraction network to be trained for channel component extraction to obtain the common channel components of each batch. Based on the common channel components, common channel label components, and a preset common mean square error loss function for each batch, the common mean square error loss value for each batch is obtained; based on the common mean square error loss value for each batch, the common channel component extraction network to be trained is trained to obtain the preset common channel component extraction network.

6. The channel estimation method for multiple users according to claim 5, characterized in that, Before inputting the processed angle domain receive tensor into a preset user-specific channel component extraction network for channel component extraction to obtain the user-specific channel component representation, the method further includes the following steps: The user-specific channel component extraction network to be trained is used for training. The training of the user-specific channel component extraction network to be trained includes the following steps: Obtain the complete channel label set of each batch of users and the common channel components of each batch output by the common channel component extraction network, wherein the complete channel label set of users includes the complete channel labels of each sample user; Based on the common channel components and pilot sequence matrix of each batch output by the common channel component extraction network, the common channel component influence removal is performed on the angle domain received tensor of each batch to obtain the processed angle domain received tensor of each batch. The processed angle domain receive tensor of each batch is input into the user-specific channel component extraction network to be trained for channel component extraction, so as to obtain the user-specific channel component representation of each batch. Based on the common channel components output by the common channel component extraction network for each batch, the user-specific channel component representation for each batch, the user complete channel label set, and the preset user complete channel mean square error loss function, the user complete channel mean square error loss value for each batch is obtained; based on the user complete channel mean square error loss value for each batch, the user-specific channel component extraction network to be trained is trained to obtain the preset user-specific channel component extraction network.

7. A channel estimation device for multiple users, characterized in that, include: The signal acquisition module is used to acquire the superimposed communication signal and pilot sequence matrix received by the base station in the target area. The superimposed communication signal is the superposition result of the uplink signals sent by all users through the uplink channel at the current time. The pilot sequence matrix is ​​used to represent the pilot sequences of all users. The tensor conversion module is used to perform angle domain tensor conversion on the superimposed communication signal to obtain the angle domain received tensor. The common channel component extraction module is used to input the angle domain receiving tensor into a preset common channel component extraction network to extract channel components and obtain common channel components. The channel component processing module is used to remove the influence of the common channel component on the angle domain received tensor according to the common channel component and the pilot sequence matrix, so as to obtain the processed angle domain received tensor. The user-specific channel component extraction module is used to input the processed angle domain receiving tensor into a preset user-specific channel component extraction network for channel component extraction to obtain a user-specific channel component representation, wherein the user-specific channel component representation includes the user-specific channel components of each user. The channel component reassembly module is used to construct a complete channel based on the user-specific channel components in the user-specific channel component representation and the common channel components, thereby obtaining the complete channel for each user.

8. A computer device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the multi-user channel estimation method as described in any one of claims 1 to 6.

9. A storage medium, characterized in that: The storage medium stores a computer program that, when executed by a processor, implements the steps of the multi-user channel estimation method as described in any one of claims 1 to 6.

Images