Research On Channel Estimation Technology For Massive MIMO Systems

With the rapid development of virtual reality,HD video,Internet of things,public security and other industries,the capacity of 4G network is unable to bear the demand in various high-speed application fields.At present,the research on 5G communications is carried forward steadily.Massive MIMO is one of its key technologies,which can greatly improve the communication capacity by enlarging the number of BS antennas and taking advantage of beamforming.However,its superiority is based on the accurate channel state information.With the increasing number of antennas and users,the dimension of the channel matrix becomes larger and more pilot resources are needed.Channel estimation in massive MIMO faces enormous challenges so that traditional algorithms cannot fulfil the estimation task.Therefore,in this thesis,we exploit the characteristics of massive MIMO channels and propose appropriate algorithms.Targeting the low frequency wideband channel estimation for downlink in massive MIMO systems,we proposed a channel estimation method which can adaptively learn channel sparse features to reduce the pilot overhead and improve the accuracy of estimation.First of all,we consider to solve the channel estimation problem under the Bayesian framework based on the theory of compressed sensing.Due to the spatial nonstationary characteristics of massive MIMO channels,we can design a two-layer Gaussian prior model to capture the sparse feature.This model includes two kinds of hyper-parameters,one type of hyper-parameter is associated with the whole antenna array which can capture the common non-zero positions of all antennas,and the other is independently associated with each single antenna which can capture the unique non-zero positions of one antenna.Finally,estimating the model parameters by means of variational Bayesian inference,and then the channel estimation can be realized.What's more,the proposed method can adaptively estimate channels with unknown sparsity levels of signals.For the channel estimation problem in millimeter-wave massive MIMO systems,in order to achieve accurate channel parameters,we proposed a two-stage algorithm based on optimization theory and the sparsity of millimeter-wave channels in angular domain.In the first stage,under the framework of variational Bayesian inference,we can obtain the channel parameters within the feasible region.In the second stage,based on the Newton optimization method,the relevant channel parameters are iteratively updated by maximizing the objective function.We can obtain accurate channel parameters after the iteration is terminated,and then reconstruct the channel according to the parameters.At the same time,the proposed algorithm can greatly reduce the training overhead and has moderate computational complexity.We used MATLAB simulation software to verify the effectiveness of the two proposed algorithms from different views,and also compared our method with various counterparts and the ideal benchmark.Simulation results are shown in this thesis,which can verify the feasibility and superiority of the proposed algorithms.