Channel Estimation For Millimeter Wave Massive MIMO Based On Compressed Sensing

Millimeter wave massive multiple input multiple output(MIMO)is one of the important development trends of mobile communication.Targeting the increasing de-mand for high-speed and low-latency communications,non-blind channel estimation for millimeter wave massive MIMO systems has become the key to communication system design.The overhead of pilot sequence or training sequence required in non-blind channel estimation is positively related to the size of the antenna at the transceiver.So the current non-blind channel estimation scheme for massive MIMO systems has a problem of excessive pilot consumption.In order to solve this problem,this disserta-tion models the channel estimation problem in the form of compressed sensing(CS)by using the unique angular sparsity of the system,and further studies the millimeter wave massive MIMO uplink(UL)channel estimation problem.In the system modeling process,in order to avoid the problem of high computational complexity due to small quantization angle spacing and poor sparse recovery performance due to increased cor-relation between the columns of the sensing matrix,the proposed schemes all adopt an off-grid model.Therefore,this dissertation focuses on the channel estimation prob-lem in the case of grid mismatch,and explores the single-user and multi-user millimeter wave massive MIMO uplink channel estimation schemes based on compressed sensing.In the study of single-user uplink channel estimation,a two-dimensional MIMO off-grid system model by Taylor expansion is first developed.And according to the Bayesian framework,an uplink channel estimation scheme based on sparse Bayesian learning(SBL)is proposed.At the same time,its fast implementation version is also provided to reduce computational complexity.This scheme has the advantages of low complexity and high accuracy.Simulation results show that the performance of the off-grid channel estimation algorithm based on sparse Bayesian learning is better than conventional algorithms.In the study of multi-user uplink channel estimation,a channel estimation frame-work based on recurrent neural networks(RNN)is established by the neural network's noise reduction capability.The framework can switch between offline training and online prediction to adapt to small changes in the channel,and has the advantages of low complexity and simple deployment.Simulation results show that the performance of channel estimation is greatly improved at low signal to noise ratio(SNR).Further considering the wide applications of channel angular information in communication systems,we transform the angular domain channel estimation problem into a multiple-measurement-vectors(MMV)compressed sensing problem by taking the advantage of the sparsity correlation among multi-user uplink channels.Simulation results also show that the proposed scheme has more advantages at low SNR.