Research On MIMO-OFDM Signal Detection Algorithms Based On Machine Learning

The combination of Multiple-Input Multiple-Output(MIMO)technology and Orthogonal Frequency Division Multiplexing(OFDM)technology is a research hotspot in 5G communication.The signal detection of MIMO-OFDM system is equivalent to the signal detection of MIMO system in flat fading channel.However,the optimal detection in MIMO system has high complexity,and suboptimal detection does not achieve the performance of ML detector.Therefore,research on MIMO detection algorithms with low complexity and good detection performance has been widely concerned.In recent years,MIMO detection algorithms based on machine learning have made some achievements,but it still exits some problems.This paper conducts further research based on the current basis of machine learning-based MIMO detection.The specific research contents include:1.The paper introduces principles of Belief Propagation(BP)algorithm and Damped BP algorithm firstly,and achieve the simulation verification.Simulation results show that the BP algorithm is a good MIMO detection algorithm,and the Damped BP algorithm can improve the convergence of the traditional BP algorithm.Then,With the BP algorithm and machine learning,this paper proposes a detection algorithm based on DNN-BP network.The DNN-BP algorithm selects the optimal damping factor through machine learning methods,which solves the problem that the damping factor is difficult to determine in traditional algorithms.Simulation results show that the DNN-BP detection algorithm can further improve the convergence of the BP algorithm and improve the BER performance.2.This paper introduces the principles of the Message Passing Detection(MPD)algorithm and the Damped-MPD algorithm in Massive MIMO.Combined with the principles of machine learning and MPD algorithm,and propose the DNN-MPD algorithm.DNN-MPD modified the log-likelihood ratio of the posterior probability during the MPD algorithm iteration process using machine learning,and sought the optimal damping factor.DNN-MPD can get better BER performance than MPD algorithm.3.This paper introduces the existing Det Net algorithm and its improved algorithm,Sc Net algorithm.They are the MIMO detection algorithms based on machine learning.On this basis,this paper propose the ZF-Sc Net detection algorithm,which is a combination of Zero-Forcing(ZF)detection algorithm and Sc Net algorithm.The ZF-Sc Net algorithm uses the ZF detection algorithm to initialize the Sc Net network,thereby improving the convergence speed of the original algorithm.The BER performance of ZF-Sc Net algorithm under perfect channel estimation and non-perfect channel estimation is simulated respectively.The simulation results show that when the channel estimate only contains Gaussian white noise and the normalized mean square error is less than 10^-4,the ZF-Sc Net algorithm can achieve BER performance close to the ideal performance.