A method for constructing wireless communication precoder based on generalized singular value decomposition

Abstract

The present invention discloses a method for constructing a wireless communication precoder based on generalized singular value decomposition, comprising the following steps: S1: simulating a wireless channel using Matlab and generating a data set based on the number of transmitting and receiving antennas; S2: importing the data set into the TensorFlow deep learning framework for preprocessing, dividing the data set into a training set and a test set, and transforming them into a vector form; S3: building a neural network using the TensorFlow deep learning framework; S4: controlling the convergence of the neural network model using a loss function, calculating the network loss and accuracy, and deriving the model weights after the network converges; S5: loading the trained model weights and the preprocessed test set data for detection, obtaining an output matrix, and calculating the system rate index. Compared with similar methods for solving the system precoding matrix, the present invention can achieve fewer channel inputs, thereby reducing training overhead.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless channel encoders, and in particular to a method for constructing a wireless communication precoder based on generalized singular value decomposition. Background Art

[0002] Currently, there are many methods for designing precoders for wireless channels, which can be mainly divided into traditional optimization algorithms and deep learning-based methods.

[0003] Traditional optimization algorithms primarily design optimal or suboptimal precoders by optimizing or jointly optimizing relevant technical indicators of wireless communication systems (such as system bandwidth, energy efficiency, and transmit power). These algorithms primarily formulate and solve optimization problems based on convex optimization theory, including a range of optimization methods such as fractional programming.

[0004] Traditional algorithms have achieved remarkable results to date, and the technology is relatively mature. However, since solving most problems requires a series of solutions to large amounts of data, such as exhaustive search or alternating optimization, to obtain the optimal or suboptimal solution to the problem, the computational complexity is too high for large-scale antenna scenarios, resulting in large computational overhead and placing high demands on the computing power of the equipment. Summary of the Invention

[0005] In view of the existing problems, the purpose of the present invention is to provide a method for constructing a wireless communication precoder based on generalized singular value decomposition to solve the above problems through TensorFlow deep learning.

[0006] The present invention provides the following technical solutions:

[0007] A method for constructing a wireless communication precoder based on generalized singular value decomposition is characterized by comprising the following steps: S1: simulating a wireless channel through Matlab and generating a data set based on the number of transmitting and receiving antennas; S2: importing the data set into the TensorFlow deep learning framework for preprocessing, dividing the data set into a training set and a test set, and transforming the data set into a vector form; S3: building a neural network through the TensorFlow deep learning framework; S4: controlling the convergence of the neural network model through a loss function, calculating the network loss and accuracy, and deriving the model weights after the network converges; S5: loading the trained model weights and the preprocessed test set data for detection, obtaining the output precoding matrix, decoding matrix, and diagonal matrix of the associated power allocation factor, and calculating the system rate index.

[0008] The data set in step S1 is the multi-user channel matrix H(b,u,h,w).

[0009] The preprocessing of step S2 is specifically as follows: dimensionality reduction processing is performed on the multi-user channel matrix from H(b,u,h,w) to H(b,h,u*w); the real part and imaginary part of the reduced multi-user channel matrix H(b,h,u*w) are separated to obtain two matrices H_real and H_imag, and H_real and H_imag are spliced ​​in the fourth dimension to obtain a training set H_train and a test set H_test of dimension (b,h,u*w,2).

[0010] The last dimension of the matrix dimensions (b,h,u*w,2) for the training and test sets represents the number of separation matrices.

[0011] Step S3 is specifically as follows: feature extraction of the data set is performed through two convolutional layers, followed by down-sampling through a pooling layer, and then through a Dropout layer to prevent network overfitting; the extracted feature data is input into a branch network in parallel for further feature extraction, and the feature extraction of the precoding matrix P, decoder R, and generalized singular value diagonal matrix C is completed through the three-branch network structure; the outputs of the above-mentioned parallel branch networks are combined into a matrix form for output, and the output data is applied in the process of calculating the power allocation factor, system and rate.

[0012] Step S4 is performed by the function:

[0013]

[0014] Calculate network loss and accuracy, where H k represents the set of channel matrices for multiple users, represents the set of predicted multi-user channel matrices, P is the precoding matrix output by the network, Ck is a diagonal matrix and can be expressed as a power allocation factor, and Dk is the decoding matrix output by the network.

[0015] The beneficial technical effects of the present invention are:

[0016] 1. It can complete the training task well when the input is only two channels (including the real and imaginary parts of the channel matrix);

[0017] 2. Based on the convolutional neural network structure, the parallel branch network design idea enables the network to output three types of matrices with different properties;

[0018] 3. Compared with similar methods for solving the system precoding matrix, this method can achieve fewer channel inputs, thereby reducing training overhead, and can also adapt well to the current multi-user communication environment;

[0019] 4. Compared with traditional algorithms under the same computing power conditions, the prediction process complexity is lower after network training is completed. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 1 is a flow chart of a method for constructing a wireless communication precoder based on generalized singular value decomposition provided by an embodiment of the present invention;

[0021] Figure 2 It is a structural diagram of a neural network in a method for constructing a wireless communication precoder based on generalized singular value decomposition provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The following embodiments of the present invention are described in detail with reference to the accompanying drawings. The following embodiments are implemented based on the technical solutions of the present invention, and detailed implementation methods and specific operating procedures are given. However, the scope of protection of the present invention is not limited to the following embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0023] Reference herein to an "embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it refer to independent or alternative embodiments that are mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments unless there is a conflict.

[0024] Example

[0025] like Figure 1 As shown, the method for constructing a wireless communication precoder based on generalized singular value decomposition provided by an embodiment of the present invention includes the following steps: S1: simulating a wireless channel through Matlab and generating a data set based on the number of transmitting and receiving antennas; S2: importing the data set into the TensorFlow deep learning framework for preprocessing, dividing the data set into a training set and a test set, and transforming the data set into a vector form; S3: building a neural network through the TensorFlow deep learning framework; S4: controlling the convergence of the neural network model through the loss function, calculating the network loss and accuracy, and deriving the model weights after the network converges; S5: loading the trained model weights and the preprocessed test set data for detection, obtaining the output precoding matrix, decoding matrix, and diagonal matrix of the associated power allocation factor, and calculating the system rate index.

[0026] The data set in step S1 is the multi-user channel matrix H(b,u,h,w).

[0027] The preprocessing of step S2 is specifically as follows: dimensionality reduction processing is performed on the multi-user channel matrix from H(b,u,h,w) to H(b,h,u*w); the real part and imaginary part of the reduced multi-user channel matrix H(b,h,u*w) are separated to obtain two matrices H_real and H_imag, and H_real and H_imag are spliced ​​in the fourth dimension to obtain a training set H_train and a test set H_test of dimension (b,h,u*w,2).

[0028] The last dimension of the matrix dimensions (b,h,u*w,2) for the training and test sets represents the number of separation matrices.

[0029] Step S3 is specifically as follows: Figure 2 As shown, the feature extraction of the data set is performed through two convolutional layers, and then down-sampling is performed through a pooling layer, and then the network is prevented from overfitting through a Dropout layer; the extracted feature data are input into a branch network in parallel for further feature extraction, and the feature extraction of the precoding matrix P, the decoder R, and the generalized singular value diagonal matrix C is completed through the three-branch network structure; the outputs of the above parallel branch networks are combined into a matrix form for output, and the output data is applied in the process of calculating the power allocation factor, system and rate.

[0030] Step S4 is performed by the function:

[0031]

[0032] Calculate network loss and accuracy, where H k represents the set of channel matrices for multiple users, represents the set of predicted multi-user channel matrices, P is the precoding matrix output by the network, Ck is a diagonal matrix and can be expressed as a power allocation factor, and Dk is the decoding matrix output by the network.

[0033] The above-mentioned embodiment of the present invention can achieve fewer channel inputs than similar methods for solving the system precoding matrix, thereby reducing training overhead.

[0034] The above describes in detail the preferred embodiments of the present invention. It should be understood that numerous modifications and variations based on the concepts of the present invention can be made by those skilled in the art without inventive effort. Therefore, any technical solution that can be derived by those skilled in the art based on the concepts of the present invention through logical analysis, reasoning, or limited experimentation based on the existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for constructing a wireless communication precoder based on generalized singular value decomposition, characterized in that: The following steps are involved: S1: Use Matlab to simulate the wireless channel and generate a data set based on the number of transmitting and receiving antennas; S2: importing the dataset into the TensorFlow deep learning framework for preprocessing, dividing the dataset into a training set and a test set, and transforming the dataset into a vector form; S3: Building a neural network using the TensorFlow deep learning framework; S4: Control the convergence of the neural network model through the loss function, calculate the network loss and accuracy, and derive the model weight after the network converges; S5: Load the trained model weights and the preprocessed test set data for testing, obtain the output precoding matrix, decoding matrix, diagonal matrix of associated power allocation factors, and calculate the system rate index; The step S3 is specifically as follows: extracting features of the data set through two convolutional layers, then performing down-sampling through a pooling layer, and then performing over-fitting of the network through a Dropout layer; inputting the extracted feature data into a branch network in parallel for further feature extraction, and completing feature extraction of the precoding matrix P, decoder R, and generalized singular value diagonal matrix C through three branch network structures; combining the outputs of the branch networks into a matrix form for output, and applying the output data in the process of calculating the power allocation factor, system and rate.

2. The method for constructing a wireless communication precoder based on generalized singular value decomposition according to claim 1, characterized in that: The data set in step S1 is a multi-user channel matrix H(b,u,h,w).

3. The method for constructing a wireless communication precoder based on generalized singular value decomposition according to claim 2, characterized in that: The preprocessing described in step S2 is specifically as follows: performing dimensionality reduction processing on the multi-user channel matrix to convert it from H(b,u,h,w) to H(b,h,u*w); separating the real part and the imaginary part of the reduced multi-user channel matrix H(b,h,u*w) to obtain two matrices H_real and H_imag, and splicing H_real and H_imag in the fourth dimension to obtain the training set H_train and the test set H_test of dimension (b,h,u*w,2).

4. The method for constructing a wireless communication precoder based on generalized singular value decomposition according to claim 3, characterized in that: The last dimension of the matrix dimensions (b, h, u*w, 2) of the training set and the test set represents the number of separation matrices.

5. The method for constructing a wireless communication precoder based on generalized singular value decomposition according to claim 2, characterized in that: The step S4 is performed by the function: Calculate network loss and accuracy, where represents the set of channel matrices for multiple users, represents the set of predicted multi-user channel matrices, , P is the precoding matrix output by the network, is a diagonal matrix and can be expressed as a power allocation factor, is the decoding matrix output by the network.

Images