Research On Massive MIMO Detection Algorithm Based On Deep Learning

Multiple Input Multiple Output(MIMO)is a promising technology in future communication systems,which has been widely concerned by researchers.As a crucial part of massive MIMO system,its detection technology is a direction with great research value at present.In order to solve the problem of high computational complexity or unsatisfactory performance of traditional detection algorithms in massive MIMO systems,this paper considers to introduce deep learning technology into the detection problem of massive MIMO systems,and improve the detection performance of massive MIMO systems by utilizing the huge advantages of deep learning in mining potential characteristics of data.The specific research content includes the following aspects:Firstly,this paper studies the detection performance of data-driven and model-driven intelligent detection algorithms in massive MIMO systems,deduces their algorithm principles in detail,and analyzes the influence of different channels,antenna configurations and modulation modes on the detection performance of these two algorithms in detail.The experimental results show that the data-driven detection algorithms have significant performance gains compared with the traditional Zero Forcing(ZF)and Minimum Mean Square Error(MMSE)algorithms in the static channel.However,they cannot work properly in large-scale systems under dynamic channels.Model-driven detection algorithms can better adapt to complex dynamic channel scenarios and have better detection performance and robustness than traditional iterative detection algorithms.Then,in order to reduce the cost of radio frequency chain,this paper also extends the research scope to massive Generalized Spatial Modulation MIMO(GSM-MIMO)system.The application of model-driven detection algorithm OAMPNet2 in the system is emphasized.In order to improve the detection performance of the algorithm in GSMMIMO systems,the prior probability of transmitting symbols on each antenna in the system is derived.In addition,two algorithms are proposed to avoid storing a large number of active antenna combinations in order to improve memory efficiency.Simulation results show that compared with the traditional MMSE and nonorthogonal Approximate Message Passing(OAMP)algorithms,the proposed algorithm can greatly improve the detection performance of GSM-MIMO systems,and for different scale GSM-MIMO systems,the algorithm shows strong robustness.Finally,the intelligent detection technology of Non-Orthogonal Multiple Access(NOMA)system is studied.An intelligent SIC sorting scheme based on deep learning is proposed to solve the problem of the Successive Interference Cancellation(SIC)algorithm in NOMA system.The simulation results show that the proposed scheme can improve the accuracy of the sorting.Thus,the detection performance of the system is greatly improved.