Research On Channel Estimation Of Massive MIMO Systems Based On Deep Learning

Massive multiple-input multiple-output(MIMO)systems use a large number of antennas to improve the channel capacity and spectral efficiency of wireless communication systems.Orthogonal Frequency Division Multiplexing(OFDM)can effectively combat narrowband interference and multipath fading.OFDM technology and massive MIMO technology are the key technologies of the fourth generation mobile communication system and the fifth generation mobile communication system.In order to take full advantage of these technologies,base stations need to obtain downlink channel state information for precoding and beamforming,etc.The base station can obtain channel state information through downlink pilot training and uplink feedback.The overhead of downlink channel estimation and uplink feedback is proportional to the number of antennas of the base station which consumes a lot of resources.In recent years,deep learning technology has developed vigorously,providing new ideas for solving the channel estimation problem in wireless communication system.Based on the deep learning technology,we study the application of deep learning technology in channel estimation of wireless communication systems.Firstly,the conventional channel estimation algorithms: LS algorithm and MMSE algorithm are summarized,and two channel estimation algorithms are simulated and analyzed.The structure and characteristics of two classical network architectures: fully connected neural network and convolutional neural network are investigated;We study the application of deep learning in each module of wireless communication system and the architecture of end-to-end communication system based on autoencoder.Secondly,the channel estimation methods of OFDM baseband communication system are researched based on deep neural network.The system model is provided and the channel estimation algorithm based on deep neural network is studied.Aiming at the complexity of the network and the problem that the input signal in the deployment stage is an interpolated signal,we propose a channel estimation algorithm based on the interpolation trained deep neural network(ITDNN).The above channel estimation algorithms are simulated,analyzed and compared.The simulation results show that the channel estimation algorithm based on ITDNN has better robustness and higher estimation accuracy in different scenarios.Finally,the channel estimation methods for massive MIMO systems is investigated based on convolutional neural networks.The compressed sensing theory is summarized and the channel reconstruction model is provided.Based on the joint sparsity of the row vector of the channel matrix,we propose a method which uses channel state informationconvolutional neural network(CSI-CNN)to reconstruct the channel matrix estimated by the compressed sensing algorithm.In addition,we propose we also propose a method which employs residual noise convolutional neural network(RNCNN)to fit the residual noise of the channel matrix,which can remove the residual noise of the estimated channel matrix.The performances of different channel estimation algorithms are simulated and analyzed.The simulation results show that the channel estimation algorithms based on CSI-CNN and RNCNN can accurately estimate the channel matrix,which can effectively reduce the overhead of pilot time slots.Moreover,method based on RNCNN has a good generalization ability to the changes of channel propagation environment,which can ensure the performance of channel estimation.