Research On Iterative MIMO Detection Algorithm Based On Deep Learning Technology

Future mobile communication system applications have high requirements for system delay,spectrum efficiency,quality of service,system reliability and other indicators.In order to meet these requirements,the 5th generation cellular mobile communication system(5G)introduces multiple input multiple output(MIMO)system,millimeter wave and ultra-dense heterogeneous network technology.In recent years,with the rise of artificial intelligence technology,for example,deep learning technology,researchers have paid more and more attention to the application of deep learning technology in the field of traditional mobile communication.MIMO detection field has become one of the research hotspots in this field because of its inherent technical characteristics.However,the existing MIMO detection technologies based on deep learning still have potential in terms of detection performance,and the trade between detection performance and computational complexity also needs to be improved.Aiming at the above problems of traditional MIMO detection technology based on deep learning,this paper studies the iterative detection method based on deep learning,in order to provide better detection performance and better detection performance computational complexity conversion by constructing iterative detection structure.In this paper,the research status of MIMO detection field is briefly introduced,and the bottleneck in the development of traditional MIMO detection technology and the challenges of MIMO detection technology based on deep learning are pointed out.At the same time,the innovation and main work of this paper are introduced.In Chapter 2,the MIMO detection theory,the MIMO channel model used in the simulation and some classical detection methods such as ZF,MMSE,OAMP are described in detail.Then the software and hardware environment settings and frame structure used in the simulation are introduced in detail.In Chapter 3 and Chapter 4,two typical MIMO detection networks based on deep learning,Det Net and LISA,are described in detail.On this basis,three iterative detection models are constructed,which are iterative Det Net detection(IDD)algorithm,separation training detection(STD)algorithm,and joint training LISA(JT-LISA)algorithm.Among them,IDD algorithm and STD algorithm are based on Det Net model while JT-LISA algorithm is based on LISA model.The IDD algorithm aims at the improvement of detection performance,the STDalgorithm aims at the better detection performance computational complexity conversion while the JT-LISA algorithm could both effectively improve the detection performance and reduce the computational complexity.At the same time,this paper also introduces the detailed process of simulation and training of these deep MIMO detectors,while the simulation results are given to prove that the three iterative detectors have their own advantages.At the end of these two chapters,the generalization ability of five depth detectors for channel correlation is analyzed.The simulation results show that the IDD algorithm has a certain performance improvement compared with the Det Net algorithm.The gain is obvious in low-order modulation and lower in highorder modulation.The STD algorithm can reduce the computational complexity by about half with a performance loss of about nearly 0.1d B or even lower,while the JT-LISA detector can reduce the computational complexity on the premise of improving the performance.In the region with high SNR,the computational complexity of JT-LISA detector is even lower than that of Det Net detector whose performance is far worse.Thus JT-LISA algorithm could provide the optimal detection performance as well as the trade of detection performance to computational complexity.In addition,the above five deep MIMO detectors all have certain generalization ability,for the detector trained in uncorrelated channel has good detection performance in correlated channel.