Large-scale MIMO downlink CSI feedback method based on deep learning
Through deep learning methods and PoolFormer module, the contradiction between CSI feedback accuracy and model complexity in large-scale MIMO systems is solved, and efficient CSI feedback performance is achieved, providing an efficient CSI feedback solution suitable for large-scale MIMO systems. .
Patent Information
- Application Number
- CN202410929746.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-11
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-07-11
AI Technical Summary
In large-scale MIMO communication systems, traditional CSI feedback algorithms need to increase model complexity to improve feedback accuracy, but this is difficult to accept in practical applications, and the computational complexity has increased dramatically affecting the real-time nature of the communication system.
Using the deep learning-based CSI feedback method, the COST2100 channel model is used to build a data set and a deep neural network model, including an encoder and a decoder. By compressing the CSI matrix and reconstructing it on the base station, combining PoolFormer module and residual learning, CSI is realized. Efficient feedback from the matrix.
Achieve higher feedback accuracy based on lower model complexity, greatly improving CSI feedback performance, and providing an efficient and feasible method for large-scale MIMO systems.
Smart Images

Figure CN119030579A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of wireless communication technology, and in particular to a large-scale MIMO downlink CSI feedback method based on deep learning. Background Art
[0002] Massive Multiple Input Multiple Output (MIMO) technology is a key technology for the next generation of wireless communication systems. By deploying large-scale antenna arrays at base stations, the spatial diversity gain of the system and the gain of the antenna array can be used to greatly reduce multi-user interference, and the spectrum efficiency and energy efficiency can be significantly improved. For massive MIMO systems, the premise for achieving these goals is that the base station can accurately obtain downlink channel state information. In time division duplex (TDD) massive MIMO systems, the reciprocity of uplink and downlink channels can be used to obtain channel state information (CSI). However, in frequency division duplex (FDD) mode, since the working frequency band of the uplink communication link is often different from that of the downlink communication link, the base station cannot obtain accurate CSI in real time through channel reciprocity. Therefore, the base station needs to first send a downlink pilot to the user end, and the user end obtains CSI by performing channel estimation on the pilot, and then transmits it back to the base station through the feedback link. The amount of CSI feedback data increases with the increase in the number of massive MIMO transceiver antennas, and the contradiction between feedback overhead and system performance gradually intensifies. The traditional CSI feedback algorithm based on compressed sensing (CS) requires a priori assumptions about the channel and is highly dependent on the sparse characteristics of the CSI channel. The difference between the prior knowledge and the actual situation greatly affects the accuracy of CSI feedback reconstruction. In addition, the CS recovery algorithm involves a large number of matrix iterative calculations. As the number of antennas increases, the computational complexity increases dramatically, which in turn affects the real-time performance of the communication system.
[0003] In recent years, deep learning (DL) has achieved great success in the fields of computer vision and natural language processing. The data-driven method based on deep learning provides new ideas for the design of CSI feedback algorithms. Many schemes have successfully introduced deep learning technology into the CSI feedback task. CsiNet is the first deep learning model designed for CSI feedback. By introducing an autoencoder structure, CSI is compressed at the user end and then reconstructed at the base station. Compared with traditional CSI feedback methods, CsiNet has better feedback accuracy and lower complexity. Based on CsiNet, researchers have proposed many DL-based CSI feedback models. For example, CRNet first introduced a multi-resolution convolution architecture, and used larger convolution kernels and residual learning, which improved the feedback performance to a certain extent. ACRNet uses network aggregation technology and PReLU activation function, and combines binarization with vector quantization for the first time to make the model have better feedback performance.
[0004] However, the above methods all have the disadvantage of not being able to take into account both model complexity and feedback performance. The method of improving feedback accuracy by increasing model complexity is unacceptable in large-scale MIMO communication systems. Summary of the invention
[0005] The purpose of the present invention is to provide a large-scale MIMO downlink CSI feedback method based on deep learning, which solves the problem that the method of improving feedback accuracy by increasing model complexity is unacceptable in a large-scale MIMO communication system.
[0006] To achieve the above object, the present invention provides a large-scale MIMO downlink CSI feedback method based on deep learning, comprising the following steps:
[0007] The COST2100 channel model is used to construct a data set, and the data set is preprocessed and divided into a training set, a validation set, and a test set;
[0008] Construct a deep neural network model, which includes an encoder and a decoder. The encoder is deployed at the user end and compresses the CSI matrix into codewords according to a preset compression rate and feeds it back to the base station through a feedback link. The decoder is deployed at the base station and reconstructs the received codewords into the original channel matrix.
[0009] Use the training set to train the deep neural network model, and use the validation set to assist in adjusting the training strategy during the training process. Test the model performance of the trained deep neural network model on the test set.
[0010] Save the trained deep neural network model, deploy the encoder on the user side, and the decoder on the base station.
[0011] Wherein, a deep neural network model is constructed, the deep neural network model includes an encoder and a decoder, and the steps further include:
[0012] The encoder consists of a PF layer, a transposed convolution layer, a convolution layer, and a fully connected layer. The first PF layer of the encoder contains a PoolFormer module. After the PF layer, the feature map is upsampled using transposed convolution. After the transposed convolution layer, the feature map is restored using convolution, and then the second convolution is used for feature extraction. Finally, the codeword is generated through the fully connected layer.
[0013] Wherein, a deep neural network model is constructed, the deep neural network model includes an encoder and a decoder, and the steps further include:
[0014] The decoder consists of a fully connected layer, a PF layer, two convolutional layers and a Sigmoid layer. The fully connected layer reconstructs the received codeword into the original channel state matrix size and inputs the matrix into the first convolutional layer to extract features through convolution. After the first convolutional layer, a PF layer composed of three identical PoolFormer modules is used for feature extraction and CSI matrix reconstruction. After each PoolFormer module, residual learning is applied to improve the overall performance of the model. The processed matrix is subjected to feature fusion and reconstruction through the second convolutional layer composed of transposed convolution and convolution. Finally, the reconstructed matrix is output after being processed by the Sigmoid activation function.
[0015] Among them, the PoolFormer module consists of three different types of modules: normalization layer, average pooling layer and multi-layer perceptron.
[0016] Among them, the multilayer perceptron consists of two linear layers and a Gaussian error linear unit, and the Gaussian error linear unit is deployed between the two linear layers.
[0017] A large-scale MIMO downlink CSI feedback method based on deep learning of the present invention uses a COST2100 channel model component data set to generate a training set, a verification set and a test set; a deep neural network model is constructed, and the deep neural network model includes an encoder and a decoder. The encoder is deployed at the user end, and the CSI matrix is compressed into codewords according to a preset compression rate, and fed back to the base station through a feedback link. The decoder is deployed at the base station and the received codewords are reconstructed into the original channel matrix; the deep neural network model is trained with a training set, and the verification set is used to assist in adjusting the training strategy during the training process, and the trained deep neural network model is tested on the test set for model performance; the trained deep neural network model is saved, and the encoder is deployed at the user end, and the decoder is deployed at the base station. It can achieve higher feedback accuracy based on lower model complexity, greatly improve the CSI feedback performance, and provide an efficient and feasible method for downlink CSI feedback of large-scale MIMO systems. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.
[0019] Figure 1 It is a flow chart of the deep learning-based large-scale MIMO downlink CSI feedback method of the present invention.
[0020] Figure 2 It is a schematic diagram of the overall structure of the PFNet deep learning model of the present invention.
[0021] Figure 3It is a schematic diagram of the structure of the PoolFormer module of the present invention.
[0022] Figure 4 It is a schematic diagram of the structure of the MLP module of the present invention.
[0023] Figure 5 It is a step diagram of the deep learning-based large-scale MIMO downlink CSI feedback method of the present invention. DETAILED DESCRIPTION
[0024] Embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, but should not be construed as limiting the present invention.
[0025] The first embodiment of the present application is:
[0026] See also Figures 1 to 5 ,in, Figure 1 It is a flow chart of the deep learning-based large-scale MIMO downlink CSI feedback method of the present invention. Figure 2 It is a schematic diagram of the overall structure of the PFNet deep learning model of the present invention. Figure 3 It is a schematic diagram of the structure of the PoolFormer module of the present invention. Figure 4 It is a schematic diagram of the structure of the MLP module of the present invention. Figure 5 The present invention provides a method for providing CSI feedback for a large-scale MIMO downlink based on deep learning, which includes the following steps:
[0027] S101: Use the COST2100 channel model to build a data set and preprocess the data set into a training set, a validation set, and a test set;
[0028] Specifically, the COST2100 channel model is used to generate training sets, validation sets, and test sets for indoor and outdoor scenes, containing 100,000, 30,000, and 20,000 data points, respectively. The specific method is as follows:
[0029] The channel model used in this embodiment is deployed in the base station N t There are N transmission antennas, and one receiving antenna is deployed at the user end. c subcarriers. The signal received by the user end at the Nth subcarrier can be expressed as:
[0030]
[0031] In the formula They represent the channel vector and precoding vector respectively. In the FDD system, the base station designs the precoding vector based on the received channel matrix. Represent the data symbol of downlink transmission and the noise of the Nth subcarrier respectively. The downlink CSI matrix can be expressed as:
[0032]
[0033] The user end transmits the Feedback to the base station, the total number of feedback parameters is 2N t N c ,In order to reduce the feedback overhead, a two-dimensional discrete Fourier transform (DFT) compression method is used to make the channel matrix sparse, as shown below:
[0034]
[0035] In the formula and are two DFT matrices. For the angular delay domain channel matrix H, only the front The truncated matrix is denoted as H a , the total number of feedback parameters is reduced to 2N t N c , forming a size of N t *N c , a channel matrix with 2 channels. Through the above steps, feedback H a Instead of H, feedback overhead can be reduced. a The data generated by the COST2100 channel model is used as a dataset for model training and testing, where N t =32, N c =32.
[0036] S102: construct a deep neural network model, which includes an encoder and a decoder. The encoder is deployed at the user end, compresses the CSI matrix into codewords according to a preset compression rate, and feeds back the codewords to the base station through a feedback link. The decoder is deployed at the base station, and reconstructs the received codewords into the original channel matrix.
[0037] Specifically, we build a deep learning neural network model PFNet, which consists of two parts: an encoder and a decoder. The decoder is deployed at the user end to compress the CSI matrix H to generate codewords S and transmit them to the base station through a feedback link. The decoder is deployed at the base station to reconstruct the codewords sent from the user end into the original CSI matrix The specific method is as follows:
[0038] The encoder in the neural network model consists of a PF layer, a transposed convolution layer, a convolution layer, and a fully connected layer. a The encoder generates the feedback codeword s as a two-channel matrix input. The first PF layer of the encoder contains a PoolFormer module. After the PF layer, the feature map is upsampled using a transposed convolution layer with a kernel size of 4*4. After the transposed convolution layer, the feature map is restored using a convolution with a kernel size of 4*4, and then feature extraction is performed using a convolution with a kernel size of 5*5. In these operations, the channel of the feature map is kept as 2. Finally, the feedback codeword s is generated using a fully connected layer and then transmitted to the decoder of the base station. The encoder processing process is defined as follows:
[0039] s=f en (H a )
[0040] In the formula, f en and H a Represent the encoder model and input parameters respectively, and s is the codeword compressed by the encoder. The compression rate of the neural network η=S / 2N t N c .
[0041] The decoder consists of a fully connected layer, a PF layer, two convolutional layers, and a Sigmoid layer. The fully connected layer reconstructs the codewords transmitted from the encoder into the original channel state matrix size. The matrix is input to the first convolutional layer and a 5*5 convolution is performed for feature extraction. After the first convolutional layer, a PF layer consisting of three identical PoolFormer blocks is used for feature extraction and CSI matrix reconstruction. After each PoolFormer module, residual learning is applied to improve the overall performance of the model. Subsequently, the processed matrix is feature fused and reconstructed through the second convolutional layer consisting of a 4*4 transposed convolution and a 4*4 convolution. The CSI is continuously reconstructed through these processes, and finally the reconstructed CSI matrix is output through the Sigmoid layer. The decoder processing process is defined as follows:
[0042]
[0043] In the formula, f de and s are the function and input parameters of the decoder, The channel matrix reconstructed for the decoder.
[0044] PoolFormer module such as Figure 3 As shown in the middle right sub-figure, by specifying the average pooling operation as MetaFormer (such as Figure 3The token mixer (as shown in the left sub-figure) obtains PoolFormer, which consists of three different types of blocks: normalization layer (LayerNorm), average pooling layer (AvgPooling) and multi-layer perceptron (MLP). The MLP module is as follows Figure 4 As shown in the figure, it consists of two linear layers and a Gaussian Error Linear Unit (GELU), and the GELU is deployed between the two linear layers.
[0045] S103: train the deep neural network model using the training set, and use the validation set to assist in adjusting the training strategy during the training process, and test the model performance of the trained deep neural network model on the test set;
[0046] Specifically, the neural network is trained and the model compression rate η is set. In this embodiment, the compression rates are set to 1 / 4, 1 / 8, 1 / 16, 1 / 32, and 1 / 64 respectively. Hyperparameters such as learning rate, batch size, and number of training rounds are set. The AdamW optimization algorithm is used as the optimizer, and the model is continuously trained through back propagation to minimize the model loss function. The mean square error is used as the loss function Loss, which is specifically defined as follows:
[0047]
[0048] Where N is the number of training set data, H i is the i-th data in the training set, It is H i The network model output corresponding to the input.
[0049] Save the trained model and use the test set to test the model performance. The normalized mean square error (NMSE) and floating point operations (FLOPs) are used as evaluation indicators of feedback performance. The specific definition of NMSE is as follows:
[0050]
[0051] Where H i is the channel matrix of the ith sample in the dataset, H i The channel matrix is compressed and reconstructed by the encoder and decoder as the model input. After a large number of experiments, it is found that the performance of the deep learning-based neural network algorithm of PFNet in this example for downlink CSI feedback of large-scale MIMO systems is significantly better than the existing methods. As shown in Table 1, the PFNet proposed in the present invention has better feedback performance in indoor and outdoor scenarios with different compression ratios, which is specifically reflected in that PFNet has the lowest FLOPs and NMSE.
[0052] Table 1: Performance comparison of this embodiment and other CSI feedback algorithms
[0053]
[0054]
[0055] S104: Save the trained deep neural network model, and deploy the encoder at the user end and the decoder at the base station.
[0056] Specifically, the use of a trained deep neural network model has better performance and lower computational complexity than other deep learning-based feedback ideas, providing a high-performance method for downlink CSI feedback in massive MIMO communication systems.
[0057] What is disclosed above is only one or more preferred embodiments of the present application, and cannot be used to limit the scope of rights of the present application. Ordinary technicians in this field can understand that all or part of the processes of implementing the above embodiments and equivalent changes made according to the claims of the present application are still within the scope covered by the present application.
Claims
1. A large-scale MIMO downlink CSI feedback method based on deep learning, characterized in that: The following steps are involved: The COST2100 channel model is used to construct a data set, and the data set is preprocessed and divided into a training set, a validation set, and a test set; Construct a deep neural network model, which includes an encoder and a decoder. The encoder is deployed at the user end and compresses the CSI matrix into codewords according to a preset compression rate and feeds it back to the base station through a feedback link. The decoder is deployed at the base station and reconstructs the received codewords into the original channel matrix. Use the training set to train the deep neural network model, and use the validation set to assist in adjusting the training strategy during the training process. Test the model performance of the trained deep neural network model on the test set. Save the trained deep neural network model, deploy the encoder on the user side, and the decoder on the base station.
2. The method for massive MIMO downlink CSI feedback based on deep learning according to claim 1, characterized in that: Constructing a deep neural network model, the deep neural network model includes an encoder and a decoder, and the steps further include: The encoder consists of a PF layer, a transposed convolution layer, a convolution layer, and a fully connected layer. The first PF layer of the encoder contains a PoolFormer module. After the PF layer, the feature map is upsampled using transposed convolution. After the transposed convolution layer, the feature map is restored using convolution, and then the second convolution is used for feature extraction. Finally, the codeword is generated through the fully connected layer.
3. The method for massive MIMO downlink CSI feedback based on deep learning as claimed in claim 2, characterized in that: Constructing a deep neural network model, the deep neural network model includes an encoder and a decoder, and the steps further include: The decoder consists of a fully connected layer, a PF layer, two convolutional layers and a Sigmoid layer. The fully connected layer reconstructs the received codeword into the original channel state matrix size and inputs the matrix into the first convolutional layer to extract features through convolution. After the first convolutional layer, a PF layer composed of three identical PoolFormer modules is used for feature extraction and CSI matrix reconstruction. After each PoolFormer module, residual learning is applied to improve the overall performance of the model. The processed matrix is subjected to feature fusion and reconstruction through the second convolutional layer composed of transposed convolution and convolution. Finally, the reconstructed matrix is output after being processed by the Sigmoid activation function.
4. The method for massive MIMO downlink CSI feedback based on deep learning as claimed in claim 3, characterized in that: The PoolFormer module consists of three different types of modules: normalization layers, average pooling layers, and multi-layer perceptrons.
5. The method for massive MIMO downlink CSI feedback based on deep learning as claimed in claim 4, characterized in that: The multilayer perceptron consists of two linear layers and a Gaussian error linear unit, which is deployed between the two linear layers.
Citation Information
Patent Citations
CSI compression feedback method based on deep learning
CN115549742A
Coder and decoder decoupling CSI feedback method based on deep learning
CN115987339A
Large-scale MIMO channel state information feedback and reconstruction method based on deep learning
CN116248156A
Large-scale MIMO downlink CSI feedback method based on deep learning enhancement
CN116405070A
Channel state information (CSI) feedback method and system based on Transform
CN116800318A
