Research On Pilot Compression And Channel Estimation For Massive MIMO Under Quasi-sparse Channel Environment

Due to its advantages of high spectral efficiency and energy efficiency,the massive MIMO system is regarded as one of the key technologies of future mobile communication.However,its potential gain depends heavily on the accuracy of channel state information(CSI)obtained at the base station(BS).As the number of antennas increases,the pilot overhead required to obtain CSI also doubles,which greatly reduces spectral efficiency.Existing compressed sensing and deep learning methods are generally done under the assumption that the number of antennas is huge and the channel is strictly sparse.In the actual system,due to the limited number of antennas and the influence of the rich scattering environment,the channel is not strictly sparse,resulting in a sharp drop in system performance.Therefore,this thesis is oriented to the quasisparse channel environment,taking the massive MIMO system as the research object,and proposes two innovative pilot compression and channel estimation schemes based on deep learning,which can effectively reduce the pilot overhead while ensuring estimation accuracy.First,inspired by the powerful learning ability of deep learning in extracting the main features of the target,this paper proposes a multi-stage training optimization scheme based on the Approximate Message Passing(AMP)algorithm to match the quasi-sparse multi-antenna channel environment and achieve effective pilot compression and accurate channel estimation.The entire system is regarded as a deep neural network(DNN),and the channel adopts the3 GPP CDL-C quasi-sparse model.In the first stage,the sensing matrix of the AMP algorithm is optimized through training to match the quasi-sparse channel.In the second stage,optimize the linear coefficient and nonlinear shrinkage parameters of AMP to further improve the performance.In the third stage,jointly optimize the trainable parameter set to avoid the system from falling into a bad local optimum and make the whole system reach the optimum.Simulation results demonstrate that the proposed architecture is capable of accurately estimating the channel while effectively compressing pilots in massive MIMO systems under quasi-sparse channel environments.Secondly,based on the above work,this paper proposes a massive MIMO-OFDM compressed channel estimation scheme based on joint spatio-temporal optimization,which aims to fully exploit the sparse characteristics of the time delay domain,and use deep learning to carry out an optimization strategy that is first independent and then combined to achieve efficient pilot compression and accurate channel estimation.To fully exploit the sparsity properties of the spatial angle domain and the time delay domain,this method transforms the signal data into the spatio-temporal domain through a specially designed MIMO-OFDM system and then trains the neural network through an “independent-joint” method.For the training of the neural network,regard the entire system as a DNN.Considering that the pilot design and resource allocation are based on RB as the basic unit,we regard the sensing matrix on each RB as a layer of linear neural networks that are trained individually.In the joint training stage,based on a large number of spatio-temporal sparse domain channels and observation data,unite all RBs to learn AMP algorithm parameters using a specific layer-by-layer training scheme.After the training is completed,the optimized sensing matrix and parameters obtained from the training can be used for pilot transmission and channel estimation.This can not only make full use of the sparsity of the channel in the beam domain and the delay domain but also improve the performance of sparse signal reconstruction.At the same time,the layer-by-layer training scheme can fine-tune the trained neural network layer by layer,thereby avoiding bad local optima caused by overfitting.Simulation results show that the proposed architecture can achieve high-precision channel estimation in massive MIMO-OFDM systems under quasisparse channel environments,while effectively compressing pilots.