Research On Robust Beamforming Technology Based On Millimeter Wave Communications

With the explosion of various intelligent devices and the increasing popularity of mobile Internet services,mobile data traffic around the world has grown dramatically.Massive data traffic demand is accompanied by the growing demand for ultra-high-speed wireless communication with full coverage,ultra-high reliability,and ultra-low latency.These emerging communication requirements promote the development of existing communication systems towards higher frequency millimeter wave to expand channel capacity and improve transmission speed.The high path loss of millimeter wave can be compensated effectively by beamforming technology combined with large-scale array antenna.However,with the increase of the number of antennas,the beam directivity is enhanced,the beam Angle becomes narrower,and the beam coverage becomes smaller.It becomes difficult to match millimeter wave channels with smaller beam alignment overhead,robust beam tracking,and accurate beamforming techniques.Aiming at the problems of high delay beam alignment,poor beam tracking robustness,and difficult to solve hybrid beamforming optimal in millimeter wave large-scale MIMO systems.Based on artificial intelligence algorithm,this paper explores intelligent solutions of beam alignment,beam tracking and hybrid beamforming in millimeter wave large-scale MIMO system with low delay beam alignment,robust beam tracking and high performance hybrid beamforming algorithm as the research target.The innovation points of this paper are summarized as follows:(1)A fast beam alignment algorithm driven by deep reinforcement learning is proposed.Based on sparse encoding and phaseless decoding,the algorithm first transforms the beam alignment problem into a beam grouping problem.Then the problem is modeled based on Markov decision process,and deep reinforcement learning algorithm is used to explore the fast beam alignment grouping strategy.Finally,based on the grouping strategy of reinforcement learning,a two-stage fast beam alignment algorithm is proposed,which greatly reduces the complexity of the algorithm without losing the performance of the algorithm.Simulation results show that compared with other beam alignment algorithms,this algorithm can reduce the search times by 60%and improve the success rate by about 20% in the same beam scan times.(2)Proposes an efficient and robust beam tracking algorithm based on channel feature map.Compared with the traditional scheme,the proposed algorithm does not need to rely on accurate position information,and uses the autoencoder neural network to build the channel feature map.Then,using the channel feature map coordinates instead of the real position information as an effective carrier of the channel angle characteristics,an intelligent beam tracking scheme based on the channel feature map is proposed,and the transceiver and receiver are adjusted in real time according to the current channel angle characteristics to adjust the direction and number of candidate beam sets,so as to achieve efficient,robust and intelligent beam tracking with less scanning overhead.Simulation results show that the algorithm reduces the number of beam scans by about 50% under the same tracking accuracy.(3)A high-performance neural network hybrid beamforming design framework is proposed.Under the condition of ensuring hardware constraints,the beamforming matrix joint optimization process is transformed into an end-to-end parameter update process of the autoencoder neural network.Based on this framework,Hybrid Beamforming(HB)algorithms are designed under different system settings,such as narrowband,wideband,limited resolution phase shifters,and low-bit digital-to-analog converters.Compared with the existing HB algorithm based on machine learning,it breaks through the performance boundary of the traditional linear matrix decomposition algorithm and maximizes the transmission rate of the hybrid beamforming system.Under the limitation of hardware structure,it realizes high-performance,adaptive and intelligent hybrid beamforming.The simulation results show that the performance gain of hybrid beamforming based on neural network is about 3d B.In addition,considering the existence of channel estimation errors in the actual system,it is difficult to obtain perfect channel state information.Under the same error factor,the performance loss of the proposed scheme is only 1/4 of that of the traditional scheme,which has higher robustness.