Research On Intelligent Algorithm In Millimeter Wave MIMO Communication System

In order to meet the communication scenarios of enhanced mobile bandwidth,massive machine intercon-nection,ultra-high reliability and low delay,the 5th generation mobile communication system puts forward higher requirements in terms of communication rate,energy efficiency,delay and communication reliability,and extends its communication frequency range to the millimetre wave band.Combined with large-scale MIMO and beamforming technology,millimeter wave MIMO communication system can not only make full use of the rich spectrum resources of millimeter wave band,but also resist the serious problems of millimeter wave path loss and attenuation.However,the use of narrow beam also increases the complexity of beam design in millimeter wave MIMO communication system.In order to design codebook more efficiently and reduce the time required for beam alignment,this paper proposes to solve the above problems by combining intelligent algorithms such as deep learning and reinforcement learning.The main work of this paper is as follows:1.In order to make full use of the environment and channel characteristics of the communication system,a codebook design algorithm based on deep learning is studied.The traditional codebook design method does not take advantage of the characteristics of the environment in which the communication system is located,and can not achieve the best performance.In this chapter,the algorithm learns the environment and channel characteristics of the communication system through neural network,and then designs the appropriate codebook to improve the beam gain and the receiving power.The simulation results show that the codebook design algorithm based on deep learning not only improves the beam gain in codebook,but also meets the coverage requirements.In addition,the hardware constraints of the phase shifter are considered,and each precoding vector is a constant modulus value.2.Aiming at the problem that too many codewords in the codebook of millimeter wave MIMO commu-nication system lead to too much time cost of finite search algorithm,a beam selection algorithm based on deep learning is studied.Under ideal channel conditions,the base station can predict the optimal codeword only by using the pilot signal sent by the user,which greatly reduces the time of beam selection.In the uplink,the user sends the pilot sequence to the base station.The base station receives and analyzes the pilot signal through omni-directional beam,and uses the trained neural network model to predict the optimal beam.The received signal of the base station contains the scattering and reflection of the environment,and the channel information between the base station and the user.Therefore,the neural network model can learn the mapping relationship between the historical optimal beam and the received signal,and then predict the optimal beam in the current channel environment.Compared with the finite search algorithm,the beam selection algorithm based on deep learning does not need to scan the codebook,but predicts the optimal beam according to the received pilot signal.The simulation results show that when the channel quality is good,the beam alignment probability of the beam selection algorithm based on deep learning is comparable to that of the finite search algorithm.As the time of beam selection is greatly reduced,the neural network can predict the optimal beam to obtain better effective achievable rate.3.Aiming at the problem that millimeter wave MIMO communication system needs frequent beam scan-ning to obtain the optimal beam in mobile scene,a beam training algorithm based on reinforcement learning is studied.The main idea of the algorithm is to regard the beam training in mobile scene as a Markov decision-making process,explore the channel environment through reinforcement learning method,accumulate the experience of previous beam training,determine the optimal sub codebook space,and reduce the number of codewords to be scanned.In the simulation phase,the system only needs to scan the optimal codebook sub-space selected by reinforcement learning algorithm and determine the analog precoding vector according to the maximum received power principle.In the digital precoding stage,the equivalent channel is determined according to the selected optimal analog precoding vector,and then the optimal digital precoding vector is iteratively solved to complete the digital analog mixed precoding.The simulation results show that the beam training algorithm based on reinforcement learning can effectively select the optimal sub codebook space and greatly save the beam training time.Compared with the finite search algorithm,this algorithm can achieve higher effective reachable rate and improve the overall performance of the system.