Research On Antenna Selection And Hybrid Precoding Algorithm Based On Deep Learning

It is the key technology for 5g / 6G communication system to achieve low power consumption and high rate in the future to transmit and receive signals with massive multiple input multiple output(MIMO)antenna array in millimeter wave band.Without increasing bandwidth and transmission power,it can significantly improve the capacity of communication system and become an effective way to broadband green communication.Compared with the traditional MIMO system,the number of massive MIMO antenna array has doubled.If each antenna is connected with an independent RF link,it will lead to high hardware cost and computational complexity.Therefore,it is particularly important to find a method to reduce the hardware cost and computational complexity.On the one hand,the number of RF links can be greatly reduced by selecting some antennas for signal transmission through antenna selection algorithm,so as to reduce the computational complexity and hardware cost of the system;On the other hand,the hybrid precoding technology can also effectively reduce the cost and power consumption of the system,both of which meet the 5g energy-saving design vision.In this paper,antenna selection algorithm and hybrid precoding algorithm are studied:(1)An antenna selection algorithm is designed based on deep learning(DL).On the premise that the channel matrix is known,the objective function is to maximize the channel capacity.Firstly,the channel capacity of each channel submatrix is calculated by the exhaustive method as the input of the convolutional neural network(CNN),and the output is the optimal channel submatrix.The input and output pairs of training data are obtained.Then,the gradient descent method is used to train the designed CNN model,The trained model can get the corresponding output data according to the current input data after obtaining the channel matrix input.(2)A low complexity hybrid precoding algorithm based on singular value decomposition(SVD)is proposed.The SVD technique is used to calculate the all digital precoding matrix,and the eigenvectors with the same number of data streams are selected to capture the angle information of the analog phase shifter.By using the correlation information between the left singular matrix and the optimal all digital precoding matrix,the required residual eigenvectors are obtained,The resulting hybrid analog/ digital precoding is implemented in a low complexity manner.(3)In order to maximize the spectral efficiency,a hybrid precoding algorithm based on DL is proposed.Combined with DL based antenna selection algorithm and DL based hybrid precoding algorithm,a joint hybrid precoding algorithm for antenna selection is proposed.In this scheme,two CNN are designed.The output of the first CNN,that is,the channel sub matrix after antenna selection,is used as the input of the second CNN,and the output of the second CNN is mixed precoding.(4)A simulation platform is built to simulate the proposed algorithm and compare with the existing algorithms.The experimental results show that the proposed algorithm has the practical significance of research and can reduce the complexity,which verifies the effectiveness and robustness of the proposed algorithm.