Studies On Signal Detection Of Massive MIMO Based On Deep Learning

Massive multi-input multi-output(Massive MIMO),as one of the key technologies to improve the spectrum and energy efficiency of communication system,is the main research contents in future mobile communications.However,the traditional MIMO signal detection methods are difficult to make a trade-off between bit error rate and computational complexity when they are applied to the massive MIMO systems.Aiming to solve this problem,this thesis introduces deep learning to signal detection algorithm in massive MIMO systems,and focuses on the two detection algorithms including message passing(MP)algorithm combined with deep neural network(DNN)and convolutional neural network(CNN),respectively.1.In massive MIMO systems,considering that the traditional approximate message passing(AMP)algorithms are generally limited by the acquisition of transmitted signal prior message and fail in converging over spatially correlated channels,a low complexity AMP detection algorithm based on DNN is studied.Firstly,according to the traditional message passing algorithm,the proposed algorithm simplifies the message passing algorithm based on the factor graph model by approximate operations,and an AMP detection algorithm suiting for massive MIMO systems is derived accordingly.Furthermore,by selecting appropriate trainable parameters,the iteration process of AMP detection algorithm with training parameters is unfolded into a layer-by-layer connected DNN.Finally,the designed network is trained through a large amount of data and deep learning to determine the optimal parameters,and the signal detection is processed according to the optimal deep neural network using the above-mentioned optimal parameters.Theoretical analysis and simulation results show that the proposed algorithm can improve bit error rate performance compared with the traditional AMP detection algorithm,in the case of unknown signal prior message and the spatially correlated channel model.2.Aiming at the problem that the detection performance of the existing massive MIMO detection algorithm may be degraded in correlated noise channels,a message passing detection(MPD)algorithm based on CNN is studied in condition of correlated noise channels.The proposed algorithm concatenates a trainable CNN with a standard MPD detector.Firstly,the MPD algorithm based on channel hardening initially estimates the transmitted signal.Then,CNN is used to remove the estimation error of MPD detector to obtain more accurate channel noise,which provides a useful noise distribution for the subsequent MPD algorithm.Simulation results show that the iterative structure of MPD detector and CNN can improve the detection performance in conditions of correlated noise channels and fewer antennas.Compared with the traditional MPD algorithm,the proposed algorithm has a better detection performance.