Research On Channel Estimation For Massive MIMO Systems Based On Sparse Bayesian Learning

With the rapid development of mobile communication,a variety of new applications have sprung up,which requires high system-capacity.Massive multiple-input multiple-output(MIMO)can greatly enhance the wireless communication capacity,which is suitable to meet high throughput demands of the 5-th generation(5G)and even of the 6-th generation(6G).According to the channel state information,the base station can deploy channel adaptive technologies such as precoding and resource allocation in order to improve the frequency efficiency and energy efficiency of massive MIMO systems.Therefore,channel state information is essential.The orthogonal pilot overhead required by traditional channel estimation methods is proportional to the number of antennas in the base station.Considering the amount of antennas in massive MIMO systems,it will cause heavy pilot overhead.The channel model is established according to the sparsity of the channel,and the compressive sensing method is used here to accurately recover the high-dimensional sparse channel from the low-dimensional observation vector.Meanwhile,considering different massive MIMO scenarios,this thesis establishes different sparse channel models,and proposes channel estimation methods based on Bayesian learning framework.For massive MIMO systems equipped with lens antenna array,the first research proposes a channel estimation scheme based on non-uniform cluster model in Bayesian learning framework.The challenge is that for massive MIMO systems,it is unrealistic for each antenna unit to be equipped with a radio frequency(RF)link in terms of hardware and power consumption;Secondly,compared with the independence between non zeros of existing sparse channels,non-uniform cluster sparsity can grasp the inherent sparsity properties of actual channels.In order to solve the above problems,the massive system with lens antenna array is adopted in this thesis,and the number of RF links required is greatly reduced through beam selection;Gaussian mixture model is used to describe the sparse characteristics,and a coupling model is set between adjacent points to describe the correlation.In the first research,the correlation and sparsity characteristics between adjacent points of sparse channel with common support and non-uniform cluster structure are reflected in sparse Bayesian framework.Finally,simulation experiments are shown to verify the effectiveness of the algorithm.For the channel estimation problem of massive MIMO systems in high-speed scenarios,the second research proposes a channel estimation scheme using orthogonal time frequency space(OTFS)modulation.The core problem is that the performance of OFDM modulation will not be ideal any more because of inter-carrier interference caused by Doppler frequency offsets.For Doppler frequency offset,OTFS modulation is adopted in this dissertation.In addition,this dissertation grasps the channel Doppler and the angular domain characteristics of massive MIMO,and applies local beta process(LBP)to provide the inherent structure description of 2 dimensional sparse signal.Finally,the Doppler angular channel sparse model of the channel in high-speed scene is established,and the channel estimation scheme affected by Doppler frequency offset is proposed.Simulation results demonstrates that the scheme can effectively realize the channel recovery of massive MIMO systems in high-speed scenarios.