Intelligent reflecting surface channel estimation method and system based on feature reconstruction

Abstract

The application provides a smart reflecting surface channel estimation method and system based on feature reconstruction, relates to the technical field of wireless communication, and specifically includes the following steps: collecting a signal received at a user side to generate a random received signal matrix; inputting the random received signal matrix into a trained channel estimation neural network to generate and output estimated channel information; and using deep learning to extract global information and detail information, and performing feature reconstruction, so that the accuracy and generalization of channel estimation are improved, the calculation resources are saved, and the application is suitable for a wireless communication system assisted by RIS technology.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication technology, and in particular relates to a method and system for intelligent reflector channel estimation based on feature reconstruction. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In today's information age, Reconfigurable Intelligence Surfaces (RIS) have emerged as a promising new technology by reshaping the wireless transmission environment through signal reflection control, thereby improving bandwidth utilization and information transmission accuracy. RIS significantly improves wireless communication coverage and energy efficiency. Compared to traditional active repeaters, RIS has lower hardware costs and energy consumption, while supporting full-duplex communication. However, because RIS lacks transmit / receive and signal processing capabilities, the difficulty of practical channel estimation is further increased. Currently, RIS channel estimation still has shortcomings in terms of accuracy, and its performance needs further improvement. How to remove channel noise interference and improve channel estimation accuracy is one of the problems that needs to be solved.

[0004] Modeling RIS channel information as image information and using convolutional networks to extract image features can avoid the shortcomings of manual feature extraction and is widely used in the field of wireless communication. Currently, most channel estimation schemes based on convolutional networks utilize convolutional networks to extract RIS channel features.

[0005] The RIS channel estimation scheme based on deep learning suffers from the problem of image information loss after deep convolution. There are two ways to improve this: 1) design a scene classifier in the convolutional block, and 2) introduce an attention mechanism into the deep neural network model. However, the above improvements are difficult to achieve the best balance between computational complexity, channel estimation accuracy and generalization. Summary of the Invention

[0006] To overcome the shortcomings of the prior art, this invention provides a method and system for intelligent reflector channel estimation based on feature reconstruction. It utilizes deep learning to extract global and detailed information and perform feature reconstruction, which improves the accuracy and generalization of channel estimation while saving computational resources. It is suitable for wireless communication systems assisted by RIS technology.

[0007] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0008] The first aspect of this invention provides a smart reflector channel estimation method based on feature reconstruction.

[0009] A smart reflector channel estimation method based on feature reconstruction includes:

[0010] Collect the signals received by the user side and generate a random received signal matrix;

[0011] The random received signal matrix is ​​input into the trained channel estimation neural network to generate and output the estimated channel information.

[0012] The channel estimation neural network performs step-by-step downsampling on the random received signal matrix to extract channel information image features; through a branch-merge feature structure, it extracts details and contextual information from the channel information image features to obtain a channel image; it applies an attention mechanism to the channel image on the channel dimension, adaptively weights features, and adaptively adjusts the importance of different channels to generate reconstructed features; based on the reconstructed features, it obtains the estimated channel information through feature mapping.

[0013] Furthermore, it also includes data preprocessing after generating the random received signal matrix, specifically:

[0014] The random received signal matrix is ​​a complex matrix. The real part of the complex matrix corresponds to an image, and the imaginary part corresponds to an image, so that the random received signal matrix is ​​regarded as a 2-channel noise image.

[0015] Furthermore, the random received signal matrix is ​​downsampled in steps. Based on the convolutional layer, the max pooling layer, and the ReLU activation function, the input image is downsampled four times to obtain the channel information image features.

[0016] Furthermore, the branch-merge feature structure is specifically as follows:

[0017] Channel information image features are input into a two-branch neural network architecture, with each branch processing features at different scales to extract multiple features of the image;

[0018] Feature merging of multiple features in an image;

[0019] The merged features are then input into a single-layer convolutional network to reduce the number of channels.

[0020] The channel image is obtained by adding the shrunken feature map to the original channel information image features.

[0021] Furthermore, the attention mechanism applied to the channel image along the channel dimension specifically includes:

[0022] Calculate channel self-attention, compress and aggregate features, and generate channel description information;

[0023] The global spatial information is compressed into the aforementioned channel description information;

[0024] Each channel learns specific samples sequentially, and then uses global information to selectively emphasize informative features and suppress less useful features, thus obtaining reconstructed features.

[0025] Furthermore, the dataset used for training the channel estimation neural network, including the random received signal matrix and its corresponding channel information, is calculated based on the channel model.

[0026] Furthermore, the channel model adopts the Saleh Valenzuela model to calculate the channel information h from the base station to the RIS end. B and RIS end-to-user channel information h R ;

[0027] The random received signal matrix is ​​expressed by the formula:

[0028] y k,q =Ψ k,q h k,q p k,q +n k,q

[0029] Where, p k,p This indicates the pilot signal sent by the base station. This represents the channel information h from the base station to the RIS end. B and RIS end-to-user channel information h R The cascaded channel, Ψ k,q Let n represent the reflection vector of the k-th user in time slot q at RIS. k,q Let represent the received noise of the k-th user in time slot q, which follows a Gaussian distribution.

[0030] A second aspect of the present invention provides an intelligent reflector channel estimation system based on feature reconstruction.

[0031] A smart reflector channel estimation system based on feature reconstruction includes a matrix generation module and a channel estimation module.

[0032] The matrix generation module is configured to: collect signals received by the user side and generate a random received signal matrix;

[0033] The channel estimation module is configured to: input a random received signal matrix into a trained channel estimation neural network, generate and output estimated channel information;

[0034] The channel estimation neural network performs step-by-step downsampling on the random received signal matrix to extract channel information image features; through a branch-merge feature structure, it extracts details and contextual information from the channel information image features to obtain a channel image; it applies an attention mechanism to the channel image on the channel dimension, adaptively weights features, and adaptively adjusts the importance of different channels to generate reconstructed features; based on the reconstructed features, it obtains the estimated channel information through feature mapping.

[0035] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps of a feature-reconstruction-based intelligent reflector channel estimation method as described in the first aspect of the present invention.

[0036] A fourth aspect of the present invention provides an electronic device including a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of a feature-reconstruction-based intelligent reflector channel estimation method as described in the first aspect of the present invention.

[0037] The above one or more technical solutions have the following beneficial effects:

[0038] This invention provides an intelligent reflector channel estimation scheme, which introduces a step-down sampling module and a feature reconstruction module to improve the accuracy of channel estimation. It has better normalized mean square error (NMSE) performance in quasi-static channels, and the feature reconstruction module improves the generalization of channel estimation. While improving the accuracy and generalization of channel estimation, it saves computing resources and is suitable for wireless communication systems assisted by RIS technology.

[0039] Advantages of additional aspects of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0040] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0041] Figure 1 This is a flowchart of the method in the first embodiment.

[0042] Figure 2 This is a block diagram of the RIS channel estimation method in the first embodiment.

[0043] Figure 3 This is a structural diagram of the RIS-assisted wireless communication system of the first embodiment.

[0044] Figure 4 This is a structural diagram of the step-down sampling module in the first embodiment.

[0045] Figure 5 The first embodiment features a reconstructed module structure diagram.

[0046] Figure 6 This is a structural diagram of the channel attention module in the first embodiment.

[0047] Figure 7 This is a structural diagram of the feature mapping module in the first embodiment.

[0048] Figure 8 This is a comparison chart of NMSE performance of the first embodiment on the Saleh Valenzuela channel model.

[0049] Figure 9 This is a comparison chart of the generalization performance of the first embodiment on the Saleh Valenzuela channel model. Detailed Implementation

[0050] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0051] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0052] Example 1

[0053] One embodiment of this disclosure provides a smart reflector channel estimation method based on feature reconstruction, such as... Figure 1 As shown, it includes the following steps:

[0054] Step S1: Collect the signals received by the user side and generate a random received signal matrix;

[0055] The random received signal matrix is ​​input into the trained channel estimation neural network to generate and output the estimated channel information.

[0056] The channel estimation neural network performs step-by-step downsampling on the random received signal matrix to extract channel information image features; through a branch-merge feature structure, it extracts details and contextual information from the channel information image features to obtain a channel image; it applies an attention mechanism to the channel image on the channel dimension, adaptively weights features, and adaptively adjusts the importance of different channels to generate reconstructed features; based on the reconstructed features, it obtains the estimated channel information through feature mapping.

[0057] As one embodiment, a channel estimation neural network is composed of three modules: step-by-step downsampling, feature reconstruction, and feature mapping. From the perspective of constructing and training the channel estimation neural network, the implementation process of a smart reflector channel estimation method based on feature reconstruction in this embodiment will be described in detail below. Figure 2 As shown:

[0058] Step 1: Using the Saleh Valenzuela model, establish a channel model to calculate the channel information h from the base station to the RIS. B and RIS end-to-user channel information h R .

[0059] The channel information from the base station to the RIS terminal is represented by equation (1):

[0060]

[0061] Where L1 represents the number of multipath paths from the base station to the RIS, M represents the number of antennas at the base station, and N represents the number of elements at the RIS. This represents the path complex gain at position l1 on the base station. and This represents the azimuth (elevation) array response at the RIS and the azimuth (elevation) array response at the base station.

[0062] The channel information from the RIS terminal to the user terminal is represented by equation (2):

[0063]

[0064] Where L2 represents the number of multipaths from RIS to the user, This represents the complex gain of the path at position l2 on the RIS side. This represents the azimuth (elevation) array response at the path at position l2.

[0065] The array response vector α(γ,μ) of a uniform planar array N = N1 × N2 is further expressed as equation (3):

[0066]

[0067] Where n1 = [0, 1, ..., N1-1]T n² = [0, 1, ..., N²-1] T λ represents the carrier wavelength, and d represents the antenna spacing.

[0068] Step 2: Consider RIS-assisted narrowband wireless communication systems, such as Figure 3 As shown, assuming the signal travels through a quasi-static channel, based on the channel information h from the base station to the RIS obtained in step 1... B and RIS end-to-user channel information h R The received signal in the k-th user's q-th time slot, i.e., the random received signal matrix, is calculated using the following formula:

[0069] y k,q =Ψ k,q h k,q p k,q +n k,q (4)

[0070] Where, p k,p This indicates the pilot signal sent by the base station. This represents the channel information h from the base station to the RIS end. B and RIS end-to-user channel information h R The resulting cascaded channel, Ψ k,q Let n represent the reflection vector of the k-th user in time slot q at RIS. k,q Let represent the received noise of the k-th user in time slot q, which follows a Gaussian distribution.

[0071] In the RGB color model, a color image can be divided into three channels: red, green, and blue. The real and imaginary parts of the RIS channel matrix are highly correlated. After data preprocessing, this matrix is ​​treated as a 2-channel noise image, and an image neural network is built to utilize this matrix.

[0072] Step 3: The channel estimation neural network is deployed on the user side for downlink cascaded channel estimation.

[0073] The channel estimation neural network establishes a nonlinear mapping relationship from the received pilot signal to the cascaded channel as shown in equation (5):

[0074]

[0075] Among them, f θ This represents a nonlinear mapping function with weights θ. This represents the channel information output by the channel estimation neural network.

[0076] Step 4: Build the step-by-step downsampling module as follows Figure 4As shown, the process is divided into four steps, with the feature map of size H×W×2 being downsampled four times to obtain a feature map of (H / 16)×(W / 16)×128.

[0077] Step 4-1 passes the input image through a 5×5 convolutional layer and a 3×3 max pooling layer, mapping the input feature map of size H×W×2 to a feature map of size (H / 2)×(W / 2)×64.

[0078] Step 4-2: The feature map output in Step 4-1 is processed through a 5×5 convolutional layer and a 3×3 max pooling layer to obtain a (H / 4)×(W / 4)×128 feature map.

[0079] Step 4-3 takes the feature map output from Step 4-2 and passes it through two 3×3 convolutional layers and the ReLU activation function. Then, it is downsampled through a 2×2 max pooling layer to obtain a feature map of (H / 8)×(W / 8)×128.

[0080] Step 4-4 takes the feature map output from Step 4-3 and passes it through three 3×3 convolutional layers and the ReLU activation function. Then, it is downsampled through a 2×2 max pooling layer to obtain a feature map of (H / 16)×(W / 16)×128.

[0081] By using a step-by-step downsampling module, the spatial resolution of the feature map is reduced, thus reducing the computational burden while extracting higher-level features.

[0082] Step 5: Build the feature reconstruction module as follows Figure 5 As shown, a channel information image of size H×W×64 is processed... Figure 4 A medium-depth convolution module is used to obtain a reconstructed image of H×W×64.

[0083] Step 5-1 processes the channel information image features through a two-branch neural network architecture. Each branch can process features at different scales, extracting multiple features of the image. This helps to capture information at different scales in the channel image, improving the accuracy of channel estimation and the generalization ability in different scenarios.

[0084] Step 5-2 combines the dual-branch H×W×64 channel image information output from Step 5-1 to obtain an H×W×128 channel image.

[0085] Step 5-3: Input the H×W×128 channel image output from Step 5-2 into a single-layer convolutional network to reduce the number of channels, and obtain an H×W×64 channel image.

[0086] Step 5-4 adds the channel image output from step 5-3 to the original channel image, which helps the model better preserve the details and contextual information of the original image. At the same time, residual connections can alleviate the gradient vanishing problem.

[0087] Steps 5-1 to 5-4 constitute the branch-merge feature structure.

[0088] Step 5-5 passes the channel image with the branch-merge feature structure through the channel attention module, which is introduced as follows: Figure 6 As shown, adaptive feature weighting enhances the deep learning model by adaptively adjusting the importance of different channels.

[0089] Calculating channel self-attention involves: the feature U output from the branch-merge feature structure is passed through F. s Compression involves aggregating feature maps across spatial dimensions H×W×C to generate channel description information 1×1×C, thereby compressing global spatial information into the aforementioned channel description information. This can be expressed by the formula:

[0090]

[0091] Each channel learns specific samples sequentially, then uses global information to selectively emphasize informative features and suppress less useful features; it employs a fully connected (FC) layer, a ReLU function, and another FC layer, using a Sigmoid activation function; the adaptively adjusted channel weights F... e The mapping is represented as shown in equation (7):

[0092] s = F e (z,W) (7)

[0093] Where W is the parameter of the fully connected layer and s is the adaptively adjusted weight of each channel.

[0094] Step 6: Build the feature mapping module as follows Figure 7 Output channel image.

[0095] Step 6-1, Adaptive Average Pooling, pools the input feature map in the spatial dimension, reducing the feature map to a 1×1 size.

[0096] Step 6-2 builds a neural network module containing 3 linear layers and 2 ReLU activation functions, which will convert the features obtained from the feature extraction part into the final task output.

[0097] Step 7: Train the three modules of stepwise downsampling, feature reconstruction, and feature mapping using the 18,000 training datasets in the dataset.

[0098] The goal of training is to minimize the loss function by optimizing the weights θ, and the training loss function is L. θ As in equation (8):

[0099]

[0100] Where D represents the size of the training dataset, and other training parameter settings are shown in Table 1. Save the parameters after the network is trained.

[0101] Table 1 RIS Channel Estimation Training Parameter Settings

[0102]

[0103]

[0104] Step 8: When performing data estimation on the RIS channel, the complete process is as follows: Figure 2 The process shown outputs the predicted value of the channel.

[0105] To demonstrate that this embodiment achieves higher channel estimation accuracy, its channel estimation performance is verified in a quasi-static channel. The method of this embodiment is compared with the traditional least squares (LS) method and other deep learning schemes (DCENet, EAMNet, STransNet, and MobileNet). This embodiment introduces a step-by-step downsampling module and a feature reconstruction module to improve the accuracy of channel estimation. Figure 8 The normalized mean square error (NMSE) performance of six methods in quasi-static channels is shown, demonstrating that the channel estimation method in this embodiment has a lower NMSE.

[0106] To demonstrate the generalization ability of this embodiment, experimental data were divided into different regions in the angular domain, and training and testing were performed on different datasets. This embodiment introduces a feature reconstruction module to improve the generalization ability of channel estimation. Figure 9 The generalization performance of the six methods in a quasi-static channel is compared, demonstrating the excellent generalization performance of this embodiment.

[0107] To demonstrate the low computational complexity of this embodiment, Table 2 presents a comparison of the method in this embodiment with other deep learning schemes (DCENet, EAMNet, STransNet, and MobileNet). As can be seen from Table 2, the method in this embodiment significantly reduces computational complexity compared to the STransNet method.

[0108] Based on feature reconstruction-based deep learning methods, this embodiment proposes a RIS channel estimation method, which has higher channel estimation accuracy and better generalization performance than LS methods and deep learning schemes (DCENet, EAMNet, STransNet and MobileNet), and has lower computational complexity than STransNet. Applying this method can improve the channel estimation accuracy and operating efficiency of wireless communication systems.

[0109] Table 2. Computational Complexity of RIS Channel Estimation

[0110]

[0111] Example 2

[0112] One embodiment of this disclosure provides an intelligent reflector channel estimation system based on feature reconstruction, including a matrix generation module and a channel estimation module:

[0113] The matrix generation module is configured to: collect signals received by the user side and generate a random received signal matrix;

[0114] The channel estimation module is configured to: input a random received signal matrix into a trained channel estimation neural network, generate and output estimated channel information;

[0115] The channel estimation neural network performs step-by-step downsampling on the random received signal matrix to extract channel information image features; through a branch-merge feature structure, it extracts details and contextual information from the channel information image features to obtain a channel image; it applies an attention mechanism to the channel image on the channel dimension, adaptively weights features, and adaptively adjusts the importance of different channels to generate reconstructed features; based on the reconstructed features, it obtains the estimated channel information through feature mapping.

[0116] Example 3

[0117] The purpose of this embodiment is to provide a computer-readable storage medium.

[0118] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in a feature-reconstruction-based intelligent reflector channel estimation method as described in Embodiment 1 of this disclosure.

[0119] Example 4

[0120] The purpose of this embodiment is to provide an electronic device.

[0121] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in a feature-reconstruction-based intelligent reflector channel estimation method as described in Embodiment 1 of this disclosure.

[0122] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart reflector channel estimation method based on feature reconstruction, characterized in that, include: Collect the signals received by the user side and generate a random received signal matrix; The random received signal matrix is ​​input into the trained channel estimation neural network to generate and output the estimated channel information. The channel estimation neural network performs step-by-step downsampling on the random received signal matrix to extract channel information image features; through a branch-merge feature structure, it extracts details and contextual information from the channel information image features to obtain a channel image; it then applies an attention mechanism to the channel image along the channel dimension, adaptively weights the features, and adaptively adjusts the importance of different channels to generate reconstructed features; based on the reconstructed features, it obtains the estimated channel information through feature mapping. The branch-merge feature structure is specifically as follows: Channel information image features are input into a two-branch neural network architecture, with each branch processing features at different scales to extract multiple features of the image; Feature merging of multiple features in an image; The merged features are then input into a single-layer convolutional network to reduce the number of channels. The channel image is obtained by adding the shrunken feature map to the original channel information image features.

2. The intelligent reflector channel estimation method based on feature reconstruction as described in claim 1, characterized in that, This also includes data preprocessing after generating the random received signal matrix, specifically: The random received signal matrix is ​​a complex matrix. The real part of the complex matrix corresponds to an image, and the imaginary part corresponds to an image, so that the random received signal matrix is ​​regarded as a 2-channel noise image.

3. The intelligent reflector channel estimation method based on feature reconstruction as described in claim 1, characterized in that, The random received signal matrix is ​​downsampled in steps. Based on the convolutional layer, the max pooling layer, and the ReLU activation function, the input image is downsampled four times to obtain the channel information image features.

4. The intelligent reflector channel estimation method based on feature reconstruction as described in claim 1, characterized in that, The attention mechanism applied to the channel image along the channel dimension is specifically as follows: Calculate channel self-attention, compress and aggregate features, and generate channel description information; The global spatial information is compressed into the aforementioned channel description information; Each channel learns specific samples sequentially, and then uses global information to selectively emphasize informative features and suppress less useful features, thus obtaining reconstructed features.

5. The intelligent reflector channel estimation method based on feature reconstruction as described in claim 1, characterized in that, The dataset used for training the channel estimation neural network includes a random received signal matrix and its corresponding channel information, which is calculated based on the channel model.

6. The intelligent reflector channel estimation method based on feature reconstruction as described in claim 5, characterized in that, The channel model uses the Saleh Valenzuela model to calculate the channel information from the base station to the RIS. and RIS end-to-user channel information ; The random received signal matrix is ​​expressed by the formula: in, This indicates the pilot signal sent by the base station. This indicates the channel information from the base station to the RIS end. and RIS end-to-user channel information The cascaded channels formed This represents the reflection vector of the k-th user in time slot q at RIS. Let represent the received noise of the k-th user in time slot q, which follows a Gaussian distribution.

7. A smart reflector channel estimation system based on feature reconstruction, characterized in that, Includes a matrix generation module and a channel estimation module: The matrix generation module is configured to: collect signals received by the user side and generate a random received signal matrix; The channel estimation module is configured to: input a random received signal matrix into a trained channel estimation neural network, generate and output estimated channel information; The channel estimation neural network performs step-by-step downsampling on the random received signal matrix to extract channel information image features; through a branch-merge feature structure, it extracts details and contextual information from the channel information image features to obtain a channel image; it then applies an attention mechanism to the channel image along the channel dimension, adaptively weights the features, and adaptively adjusts the importance of different channels to generate reconstructed features; based on the reconstructed features, it obtains the estimated channel information through feature mapping. The branch-merge feature structure is specifically as follows: Channel information image features are input into a two-branch neural network architecture, with each branch processing features at different scales to extract multiple features of the image; Feature merging of multiple features in an image; The merged features are then input into a single-layer convolutional network to reduce the number of channels. The channel image is obtained by adding the shrunken feature map to the original channel information image features.

8. An electronic device, characterized in that it comprises: Memory is used to store computer-readable instructions in a non-transitory manner. as well as Processor, for executing the computer-readable instructions, When the computer-readable instructions are executed by the processor, they perform the method described in any one of claims 1-6.

9. A storage medium characterized in that it non-transitory stores computer-readable instructions, wherein, When the non-transitory computer-readable instructions are executed by a computer, the instructions of the method according to any one of claims 1-6 are executed.

Images