Research On Beam Optimization Algorithm For Cell-Free Massive MIMO System

Driven by the exponential growth of mobile data traffic and the requirements of the intelligent connection of all things,mobile communication continues to evolve and develop from 5G to 6G.In order to meet the needs of more diverse business scenarios and better system performance,the user-centric cell-free massive MIMO network was proposed.It flexibly deploys a large number of access points(APs)and provides cooperative services to users within the interest area.Because of its unique net architecture,cell-free massive MIMO system obtains high network throughput and user coverage rate,and receives extensive attention in 5G and6 G research.However,due to the increase of the number of APs and the increase of the antenna scale,the increasing network complexity has brought many challenges.For example,the large number of channel state information interactions in the network lead to excessive transmission link pressure,numerous network parameters make it hard to construct an effective beamforming scheme or optimize the network.To solve those questions,there are two main strategies.Firstly,Received Signal Strength Indicator(RSSI)is usually used as a substitute for complete channel information to reduce the amount of information.And Secondly,artificial intelligence technology is used to guide beamforming design and give network optimization scheme,so the needs of green and efficient communication can be achieved.This thesis focuses on reducing network cost and inproving system performance through reasonable hybrid precoding scheme and beam optimization strategies.The main work of this thesis are as follows:Firstly,a two-stage hybrid beamforming design method based on unsupervised learning is proposed,and it can solve the problem of high beam design difficulty and high system overhead in a cell-free massive MIMO system.As for system modeling,the system transport model is constructed and the hybrid beamforming problem is simplified into an analog beam selection problem based on the Discrete Fourier Transform(DFT)codebook,so the calculation of the system sum-rate is clear.In the first stage,based on RSSI of all users,an unsupervised deep learning network is used to compress the analog codeword selection space,which reduces the complexity of beam selection.In the second stage,each base station uses the strongest beam first algorithm with beam collision control strategy to select best codewords from the compressed codebook space,the algorithm is effective to reduce the interference between users and maximize the system transmission rate.The simulation shows that the proposed beamforming method can achieve the performance of the full codeword space search scheme,while greatly reducing the computational complexity and time overhead,so it can meet higher communication requirements.Secondly,an RSSI data augmentation method for system performance improvement is proposed,and it can make up insufficient data volume for model training and limited beam parameter configuration.RSSI is important because it affects the system sum-rate by guiding the design of the beamforming scheme,and the training of the deep learning model is also greatly affected by the RSSI training data.Therefore,The RSSI data augmentation model combines Conditional Variational Auto-Encoder(CVAE)and Mixture Density Networks(MDN),it learns the probability distribution of the system RSSI according to the inadequate RSSI data sampling from several location with limited scanning beam configuration parameters.It uses the mixed Gaussian distribution to improve the accuracy of the CVAE network and achieve more effective and more flexible user RSSI prediction.The simulation shows that the model proposed in this thesis can give out RSSI that is closer to the real distribution than a single CVAE or MDN network.Thirdly,an intelligent beam optimization method for statistical port selection is proposed and a scanning beam optimization framework for a cell-free MIMO system is designed.In system modeling part,the beam vector optimization problem is transformed into a codebook-based port selection optimization problem.And in the beam optimization framework,an RSSI data augmentation module,a rate mapping module,and a beam parameter optimization module are built.According to the actual collected data,the relationship model between beam parameters and statistical average system sum-rate is established.After that,genetic algorithm is used to search the port selection space to find best port selection parameters that maximize the average system sum-rate.The simulation shows that the system optimization framework proposed in this thesis can effectively improve the average system sum-rate.