Research On MIMO Signal Detection Based On Deep Learning

The demand for high system capacity and high frequency spectrum utilization in 5G wireless communication is generally achieved by increasing the number of parallel streams and high-order modulation in MIMO systems,which makes signal detection in the physical layer a difficult problem to solve.In traditional signal detection schemes,linear detection is completely ineffective,while nonlinear detection has extremely high complexity.This has led to the introduction of deep learning(DL)to address the issues of poor performance and high complexity in traditional signal detection.In recent years,the application of deep learning in communication systems has not been fully studied,although it has made significant progress in many other fields.On the basis of the existing research results,this research has carried out in-depth research on the signal detection technology based on deep learning,which is a promising technology in current and future wireless communication.The main contents and research results of this paper are as follows:1.For the signal detection technology based on deep learning when CSI is known,this paper proposes a signal detection scheme based on random restart reactive Tabu search(R3TS)enhanced by FS-Net,called DL-Aided R3TS.In this scheme,the initial solution output by FS-Net effectively reduces the complexity of the overall scheme,and the rich and effective cutoff conditions in R3TS further reduce the complexity of the overall scheme.In addition,the accuracy of multiple initial solutions in R3TS is optimized to reduce complexity.The simulation results show that in the QPSK modulated mMIMO system,the DL-Aided R3TS scheme outperforms the current traditional and deep learning based schemes.Compared with existing DL-Aided TS schemes,the DL-Aided R3TS scheme reduces computational complexity by approximately 30%without sacrificing performance.2.For signal detection techniques based on deep learning when CSI is unknown,this paper proposes a ResNet based iterative soft interference cancellation(SIC)signal detection scheme called ResLight-DeepSIC.The scheme starts from the shortcomings of the existing DeepSIC scheme,introduces the residual idea to improve the performance of the deep network model,designs a simpler network structure to reduce the complexity,and optimizes the loss function to accelerate the training convergence.The simulation results show that in the mMIMO system,the ResLight-DeepSIC scheme outperforms current traditional schemes and deep learning based schemes.Compared with existing DeepSIC schemes,the ResLight-DeepSIC scheme reduces computational complexity by approximately 37%and 63%under 16QAM modulation and 64QAM modulation,with a performance gain of approximately 1dB.