Research On MIMO Antenna Selection Algorithm Based On Machine Learning

With the development of the fifth generation(5G)wireless communication technology and the proliferation of communication data,the service quality and communication demands of users for wireless communication are continuously increasing.As one of the key technologies of the 5G wireless communication system,the multiple-input-multiple-output(MIMO)communication technology has excellent performance in ensuring the high transmission bandwidth and high reliability of the system,but it is increasing the complexity of the communication system.In order to maintain the superior performance of the MIMO communication system while controlling the complexity of the MIMO system,the antenna selection(AS)technology has caused extensive research.However,with the increase in the number of antennas in massive MIMO systems,the complexity of existing optimal antenna selection algorithms based on traditional mathematical optimization methods has also increased significantly,which greatly limits the antenna selection technology in massive MIMO systems.With the increasing application of machine learning(ML),there have also been great breakthroughs in the field of communications in recent years.It is of great significance to apply machine learning to antenna selection to improve system performance and reduce computational complexity.Therefore,this paper mainly introduces the background and significance of the MIMO communication system and machine learning,and also analyzes the research status.Combining the relevant characteristics of MIMO communication system and machine learning,the MIMO antenna selection algorithm based on machine learning is mainly studied and studied.The implementation of deep learning based antenna selection(DLBAS)on software-defined radios(SDR)platform has also been implemented.The main work content is briefly described as follows:1)For the antenna selection scenario of the receiver or transmitter of the MIMO communication system,the antenna selection problem is regarded as a multi-classification problem.For the first time,transmitter antenna selection algorithm based on the deep neural network(DNN)is proposed.Coherent space-time shift keying(CSTSK)channel state information(CSI)of a MIMO system is calibrated by multi-class labels to construct a data set and train the transmitter antenna selection model.Simulation analysis shows that the antenna selection model has a classification accuracy rate of 96.4%,and the channel capacity under the CSTSK MIMO system is very close to the optimal,and its selection complexity is also lower than the optimal exhaustive antenna selection algorithm.2)Aiming at the antenna selection scenario of the joint transmitter and receiver of a massive MIMO communication system,a joint antenna selection algorithm based on a convolutional neural network(CNN)was first proposed.Firstly,the CSI matrix of the massive MIMO system is obtained,and the data set is constructed by multi-label calibration.Secondly,the multi-channel classification model is designed according to the characteristics of the CNN.Finally,the neural network classification model is obtained by training the neural network through the data set.Simulation analysis shows that the classification accuracy of the joint antenna selection model is 87.9%,and the channel capacity in a large-scale MIMO communication system is very close to the optimal.However,compared to the optimal exhaustive algorithm,its selection complexity is greatly reduced,thereby reducing the loss of system hardware.3)Based on the research of the above two algorithms,a new SDR communication platform based on the assistance of the deep learning antenna selection algorithm is proposed.Firstly,the channel data set is constructed based on the measured data and the DLBAS classification model is trained.The deep learning decision server and the DLBAS classification are constructed.The deployment of the model then realizes the communication between the MIMO SDR platform and the deep learning decision server.Finally,the research and analysis are performed on the measured data,which further verifies the correctness and effectiveness of the algorithm proposed in this paper.