5G Massive MIMO Signal Detection

Massive MIMO is one of the key technologies in the fifth generation mobile communication system due to its high spectral efficiency and link reliability.However with the increase of antenna number in Massive MIMO system,signal detection algorithm will face the challenge of high computational complexity.Therefore,designing a signal detection algorithm with low complexity and high performance is an urgent issue for Massive MIMO systems.The main work of this thesis is as follows:The message passing detection(MPD)algorithm is widely studied in Massive MIMO system because of its good detection performance,but the complexity of MPD algorithm increases with the increase of modulation order and antenna number.The Probability Approximation Message Passing Detection(PA-MPD)algorithm based on MPD algorithm selects part of symbol probability to calculate the messages passed during each iteration.Although the complexity is reduced,there are unnecessary iterations,which brings high overhead.Therefore,based on the PA-MPD algorithm,an early termination iteration strategy is introduced to design the Improved PA-MPD(IPA-MPD)algorithm,and the optimal iteration number of the algorithm is determined according to the user symbol probability convergence speed,thereby avoiding unnecessary iterations.The complexity of the IPA-MPD algorithm is reduced.The simulation results show that the complexity of the IPA-MPD algorithm can be reduced to 52%-77% of the PA-MPD algorithm under different single-antenna user configurations without loss of the detection performance of the algorithm.Due to deep learning has made great achievements in other fields,more and more researches have been made on the application of deep learning in communication systems.OAMP-Net uses the idea of OAMP detection algorithm,while adding adjustable parameters,and using deep learning optimization method to improve its detection performance.However,each layer of OAMP-Net needs matrix inversion,which brings high complexity.In this thesis,we simplify the structure of the neural network based on OAMP-Net,and ensures the IOAMP-Net detection performance by increasing the correction factor and the training parameters of the neural network with the method of DNN-MPD.By comparing the performance and complexity of OAMP-Net and IOAMP-Net,the simulation results show that the complexity of the IOAMP-Net can be reduced to 12%-17% of the OAMP-Net while ensuring the performance.