Research On CSI Feedback Technology In FDD Massive MIMO Systems

In recent years,mobile Internet is developing rapidly,increasing mobile data traffic brings severe challenges to cellular communication network.The 5th Generation(5G)is the latest generation of cellular mobile communication technology,which has become a hot topic of discussion and application in academia and industry.The performance goal of 5g is to improve data rate,reduce latency,reduce cost and increase system capacity,etc.Among the many key technologies of 5G,massive MIMO technology is one of them.Compared with traditional MIMO,massive MIMO has many advantages.However,to realize these advantages in massive MIMO,the base station(BS)needs to be able to obtain accurate downlink channel state information(CSI).In frequency division duplex(FDD)systems,uplink and downlink work at different frequencies,the BS cannot directly obtain the CSI of the downlink by using the reciprocity of the uplink and downlink channels.At this time,the user needs to feed it back to the BS through the uplink.However,in a massive MIMO system,the number of antennas on the BS side is large,and the system overhead is large.Therefore,in order to reduce the CSI feedback overhead and improve the CSI feedback accuracy in massive MIMO systems,this paper studies the CSI feedback technology in massive MIMO systems in FDD mode.The main work of this article is as follows:(1)The fading characteristics of the wireless channel are studied,and a massive MIMO system model is constructed.Based on this,some channel characteristics of massive MIMO,that is,the spatial correlation and the sparse characteristics of the channel matrix are analyzed.At the same time,some current classic CSI feedback technologies are summarized.(2)The massive MIMO CSI feedback method based on compressed sensing is studied.Firstly,the compressed sensing technology and the greedy-type signal reconstruction method are described in detail.Based on this,aiming at the shortcomings of under estimation or over estimation of sparsity adaptive matching pursuit(SAMP)due to fixed step size,this paper proposes an improved SAMP method.The proposed method combines the initial sparseness estimation,regularization and backtracking idea,and double threshold exponential variable step size idea,this method can improve the efficiency while ensuring the accuracy of signal reconstruction.Finally,using the sparsity of massive MIMO channel in the virtual angle domain,each greedy reconstruction method is applied to CSI feedback of massive MIMO systems.Compared with other greedy signal reconstruction methods,the proposed method effectively reduces the normalization Mean square error(NMSE)and increases system capacity.(3)The massive MIMO CSI feedback method based on deep learning is studied.Due to the strong spatial correlation of massive MIMO channels,this paper proposes a deep learning-based massive MIMO CSI feedback method,which is applicable to both singleuser and multi-user scenarios.The proposed method first uses the convolutional neural networks(CNN)to extract the channel feature vector,then uses the maxpooling to compress the CSI,finally,uses the bidirectional long short-term memory network(BiLSTM)network and bidirectional convolutional long short-term memory(BiConv LSTM)network for the single-user and the multi-user respectively to decompress and recover CSI.This method uses massive MIMO channel data to train the proposed deep learning network offline,and fully mines the channel information in the training samples to make the network learn the structural characteristics of massive MIMO channels.Simulation results show that compared with some other existing CSI feedback methods,the proposed method has less running time,higher feedback accuracy,and obvious system performance improvement in massive MIMO system.