Studies On Signal Detection Of Massive MIMO Based On Deep Learning

With the development of wireless communication technology,the number of connected devices and mobile data traffic are also increasing with each passing day,which places higher demands on future wireless communication systems.At present,Massive MIMO is the one of the key technology of future wireless communication standards.Massive MIMO technology provides higher spectral efficiency and link reliability by deploying massive antenna arrays at the receiver and transmitter.But at the same time,with the increase of the number of antennas,the difficulty of signal detection at the uplink base station side is also improved greatly.To solve this problem,Massive MIMO signal detection algorithms have attracted extensive attention of late years,and various excellent algorithms emerge in endlessly,there is still room for further improvement in the performance and complexity of the detection algorithm.With improvement of machine learning theory,the application of deep learning technology to wireless communication,such as signal detection,is also deepening.Based on the existing Massive MIMO detection algorithms,this thesis focuses on the application of deep learning in signal detection.Firstly,this thesis introduces the typical MIMO detection algorithms and the basic theory of deep learning technology,and then applies deep learning technology to Simplified Approximate Message Passing(SAMP)detection algorithm and Sphere Decoding(SD)algorithm in massive MIMO system.The main innovations of this thesis are as follows:1.Based on the in-depth analysis of the principle of SAMP algorithm,a signal detection scheme is proposed based model-driven deep learning,which is named SAMP-FCNet.This scheme adds trainable parameters as iterative step for the delivery message in SAMP algorithm,and uses a fully connected neural network(FC-NN)as the threshold network to replace the imperfect threshold function of the SAMP algorithm,which free SAMP from the dependence on the prior information of the transmitted signal.Afterwards,based on the model-driven theory,the SAMP algorithm is expanded into the SAMP-FCNet neural network,Then with the help of deep learning tools,SAMP-FCNet is trained to fit the optimal threshold function and get a reasonable iterative step size.Theoretical analysis and simulation results show that,compared with the SAMP algorithm,SAMP-FCNet has a effective improvement in the bit error rate performance,and can achieve the performance of MMSE algorithm in less iterations.2.Aiming at the problem that the complexity of FSD detection algorithm in Massive MIMO systems is too high,a detection scheme based on deep learning is proposed which named SN-FSD.This scheme uses SAMP-FCNet which based on iterative algorithm to obtain the initial solution of the signal detection problem.According to the initial solution,pruning thresholds are designed for the full expansion(FE)layers and single expansion(SE)layers of FSD algorithm respectively.At the same time,when all candidate paths are discarded,the initial solution is output as the final solution.Simulation results show that in large-scale MIMO systems,SN-FSD detection scheme can achieve similar detection performance to FSD and significantly reduce the complexity of the algorithm.