Font Size: a A A

Channel Acquisition Technology Based On Deep Learning In Massive MIMO System

Posted on:2022-02-15Degree:MasterType:Thesis
Country:ChinaCandidate:Y D YangFull Text:PDF
GTID:2518306602493264Subject:Communication and Information System
Abstract/Summary:PDF Full Text Request
As one of the key technologies of the fifth generation wireless communication,the massive multiple-input-multiple-output(MIMO)greatly improves the system throughput by deeply developing the spatial dimension,and provides strong support for the high bandwidth,high reliability and low delay requirements of 5G network.As we all know,the performance of massive MIMO systems largely depends on the accuracy and efficiency of channel estimation.However,the performance bottleneck caused by the high computational complexity of traditional estimation methods has yet to be broken.The third wave of artificial intelligence not only brings dividends to computer vision and other fields,but also brings opportunities to solve problems in massive wireless communication.In this thesis,aiming at the problem that a large number of antennas in massive MIMO network bring about high complexity and low accuracy to obtain accurate channel state information,the uplink channel estimation and downlink channel extrapolation are the main research contents,and the uplink and downlink channel tracking algorithms based on deep learning are proposed respectively.The specific work is as follows:In the uplink channel estimation,in order to solve the problem that the traditional method of matrix inversion and other operations are too complicated in the massive MIMO system,this thesis mainly uses the simpler least square(LS)method and deep learning.Thought,a massive MIMO uplink channel estimation scheme based on graph neural network(GNN)is proposed.This scheme can use the graph model structure to effectively mine the spatial correlation of channels.At the same time,in order to capture the characteristics of fast timevarying channels,a time-correlation strategy for extracting time-varying channels is considered and designed.Based on the space-time correlation learning of the channel based on the model,more accurate massive MIMO time-varying uplink channel estimation can be completed,and both LS and deep learning have low complexity.The simulation results show that the performance of the GNN-based uplink channel estimation proposed in this thesis is better than that of traditional LS,feedforward neural network(FNN)and convolutional neural networks(CNN)methods.In addition,in response to the problem of high feedback that cannot be avoided in massive MIMO for downlink channel estimation,this thesis proposes a massive MIMO downlink channel estimation scheme based on deep learning.This scheme uses the deep neural network(DNN)to capture the correlation between the uplink and downlink channel data sets,and realizes the reconstruction of the downlink channel from the uplink channel,thereby avoiding the feedback process of the downlink CSI.Considering that there are only limited radio links in hybrid beamforming,the base station cannot obtain the uplink channel information of all antennas at the same time,so the antenna selection operation is particularly important before reconstruction based on DNN.That is to study the problem of antenna subset selection,and obtain the antenna subset that can extrapolate the downlink complete channel to the greatest extent.On this basis,the uplink channel of the selected antenna is used to complete the extrapolation of the downlink complete channel.The simulation results show that in the non-uniform antenna array and the uniform antenna array,the downlink channel extrapolation network based on antenna selection has better reconstruction performance than uniform sampling.In addition,when the frequency difference between the uplink and downlink channels increases and there is a certain error in the uplink estimation,the scheme still has better downlink channel extrapolation performance.
Keywords/Search Tags:massive MIMO, channel estimation, deep learning, graph neural network, antenna selection, channel extrapolation
PDF Full Text Request
Related items