Research On Intelligent Transceiver Technology In Mimo System

The fifth generation of mobile communication systems,represented by MIMO technol-ogy,is already commercially available in 2019,and discussions about the next generation of mobile communication systems are underway.It is widely believed that the theme of the next generation of mobile communication systems is to create digital twin worlds that enable seam-less connectivity control of physical and biological entities,thus enabling new mixed reality hyper-physical experiences.To achieve this vision,mobile communication systems need to be capable of ultra-large scale connectivity,ultra-high speed,ultra-low latency and high relia-bility.In recent years,intelligent communication solutions that combine artificial intelligence represented by deep learning with mobile communication are considered to be a major key to achieve the above requirements,and their application in MIMO systems has become one of the research hotspots in this field,gaining the attention of a large number of researchers.Therefore,in this paper,a series of studies on intelligent transceiver techniques in MIMO systems are con-ducted using deep learning techniques in artificial intelligence around some challenges existing in the physical layer of MIMO systems,such as low-complexity detection at the receiver side of MIMO systems,joint transceiver optimization of MIMO systems,and compressive recon-struction of precoding matrix information in MIMO systems.The key to MIMO detection technology is to find detection algorithms with good detec-tion performance and low computational complexity.We introduce classical MIMO detection algorithms and deep learning-based MIMO detection algorithms around the MIMO detection problem,and improve a low complexity iterative detection algorithm based on serial interfer-ence according to the depth unfolding method,and propose a model-driven deep MIMO detec-tion network called LCDNet.By adding trainable parameters to further eliminate multi-stream interference and improve the error propagation problem that often occurs in serial detection al-gorithms,LCDNet achieves better detection performance than the original algorithm and can be applied to a wider range of scenarios without increasing the computational complexity.Sim-ulation results show that the LCDNet algorithm can outperform the original iterative detection algorithm when the ratio of transmitting and receiving antennas is large and has significant advantages over the MMSE detection algorithm? when the ratio of transmitting and receiving antennas is small,the original iterative detection algorithm cannot work properly,but our pro-posed LCDNet network detection algorithm can still obtain better detection performance than the MMSE detection algorithm.Then,this paper focuses on the joint optimization technique of transceivers in MIMO sys-tems.The traditional sub-module design scheme can design and optimize each sub-module independently,but it is not globally optimal in terms of the whole system,and is limited by the difficulty of solving the optimization problem.The essence of the network is to implement the functions of transmitter,channel and receiver in traditional communication systems through neural networks,and to optimize the trainable parameters in the network to obtain jointly opti-mized MIMO transmitters and receivers under specified channels.The simulation results show that when the MIMO system size is small,the jointly optimized transceiver can achieve bet-ter system performance than the transceiver designed by the conventional sub-module design scheme,and when the system size increases,the system performance is still close to that of the independent optimal transceiver in the sub-module design scheme.Finally,this paper focuses on the precoding matrix information feedback technique in up-link MU-MIMO systems.Precoding for uplink MU-MIMO systems can effectively improve the system transmission performance,but the feedback of precoding matrix information from the base station to the user will occupy bandwidth resources,so the amount of system feedback should be reduced as much as possible.We propose a neural network-based PMINet network for the compression and reconstruction of the precoding matrix in combination with deep learning.Thus,the amount of system feedback is effectively reduced.The Simulation results show that both under the ideal simple channel or in a real channel scenario,our proposed PMINet network can effectively reduce the amount of system feedback with little impact on the system transmis-sion performance,and can achieve better system performance compared to the codebook-based indexed feedback scheme.