Deep Learning Based Joint Channel Estimation And Antenna Selection Algorithm

With the continuous construction of commercialized 5G,the contradiction between communication performance and communication cost and energy consumption is increasing.As 5G is deployed in a higher frequency band,the increase of electromagnetic wave frequency increases the path loss at the same distance,and more base stations need to be deployed to cover the same area.Moreover,the base station in 5G network is equipped with massive antennas,and the required circuit power and signal processing power increase significantly.At present,in the research and development of B5G and 6G,Holographic MIMO technology and Ultra Massive MIMO technology further increase the number of antennas on the base station.Therefore,the control of base station construction cost and energy consumption is the focus of this thesis.This thesis proposes a joint channel extrapolation and antenna selection algorithm based on deep learning.This method combines communication technology with deep learning,and the extrapolation method serves the antenna selection algorithm,trying to control the construction cost and power consumption of the base station under the condition of ensuring high communication performance by using antenna selection technology.By evaluating the algorithm complexity and communication performance,the practical application value of the proposed algorithm is analyzed.Specific research contents are as follows:1.Step by step antenna selection algorithm based on deep learning.This thesis combines the deep learning algorithm with the optimal antenna selection algorithm,and proposes a method to obtain the optimal antenna selection solution in steps.The deep learning network is used to output the optimal solution set,and the optimal antenna selection algorithm is used to select the optimal solution from the optimal solution set.Based on ITU-R M.2412 standard channel simulation,different datasets are generated for the two antenna selection description methods with the optimization goal of maximizing channel capacity.The effectiveness of the proposed algorithm is verified through network training and performance analysis.2.Partial channel extrapolation algorithm based on deep learning.In this thesis,a method of channel extrapolation is introduced by preselecting part of the antenna to fix the extrapolation channel,and the effectiveness of the proposed channel extrapolation method will be verified using four partial antenna preselection methods.The open source data set generator is used to obtain channel data,and several data sets are constructed according to different extrapolation channel selection methods and different antenna numbers of extrapolation channel.The deep learning network is built to fit the channel extrapolation method.The constructed data sets are used for training respectively to obtain several different deep learning networks.Verify the availability of the proposed algorithm.3.Joint channel extrapolation and antenna selection algorithm.The antenna selection algorithm of base station needs complete downlink channel information which can be obtained by channel extrapolation.In this thesis,the channel extrapolation and antenna selection algorithm are combined,and the partial channel information of the extrapolation channel is used to obtain the accurate antenna selection result.In order to make the ideal simulation data closer to reality,this thesis proposes a channel estimation error simulation method for complex matrix,analyzes the error of channel extrapolation and antenna selection respectively and jointly.In addition,this thesis analyzes the instability of model performance that may be caused by error transmission in deep learning algorithm,and verifies the feasibility of the proposed joint algorithm.