Deep Learning Assisted Multi-Point Cooperative Beamforming Method Under 5G Massive MIMO

The millimeter wave(mm Wave)wireless communication system can support high mobility to realize various important applications,such as vehicle communication,wireless virtual reality,and augmented reality.However,several challenges need to be overcome to achieve in practice.First,the use of narrow beams and the sensitivity of mm Wave signals to blocking greatly affect the coverage and reliability of high-mobility links.Secondly,users with high mobility in dense mm Wave deployments need to switch frequently between base stations(BS),which leads to beam training overhead and delay problems.In addition,identifying the best beamforming vector in a large antenna array millimeter wave system requires a lot of training overhead,which greatly affects the efficiency of mobile users.This article will solve this problem with a machine learning collaborative beamforming solution and optimize the machine learning model.The main work and innovations are as follows:1.In view of the high mobility users in dense mm Wave deployments that need to switch frequently between base stations(BS),a large amount of waste of resources and waiting time overhead caused by this,this paper coordinates multiple base stations to be the same at the same time in a multi-point cooperative manner.Users provide services.In this method,many base stations using RF links are connected to a central cloud processor with baseband processing to jointly serve a user.The uplink signal is first pre-coded on the central processor to reduce interference between data streams.Then the analog beamforming in the analog domain is used at the base station.This article details the training and design of the central processing unit and base station RF beamforming vector.The goal is to find the best beamforming vector with minimal overhead and maximize the effective reachable rate.In this way,a balance is achieved between the beamforming training overhead and the achievable rate obtained by using the designed beamforming vector.2.Aiming at the problem that identifying the best beamforming vector in the high-speed mobile mm Wave system will cause a lot of training overhead,this paper proposes a new type of machine learn-ing cooperative beam method to replace the traditional method to estimate the downlink channel,and learn the uplink through training Obtain downlink beamforming vector information.The user only needs to send an omnidirectional or quasi-omnidirectional beam pattern,and multiple distributed base stations working in cooperation jointly receive the uplink pilot sequence.These received signals not only draw signs of location information for the user's location,but also draw signs for interaction with the surrounding environment.The model learns how to use these signatures to predict the beamforming vector at the base station.The machine learning model uses a fully connected deep neural network.In the simulation stage,the deep learning cooper-ative beamforming method is compared with the basic cooperative beamforming method of traversal search.The basic cooperative beamforming method is designed to design the central processing unit and base station RF beamforming vector Depending on the uplink training,the base station first selects its RF beamforming vector from a predefined codebook.Then,the central processing unit designs its baseband beamforming to ensure the user's coherent combination.The proposed machine learning cooperative beamforming method can greatly reduce training overhead and realize high-speed mobile mm Wave applications.3.In the method of deep learning cooperative beamforming,this paper further optimizes the model,and proposes a convolutional neural network cooperative beamforming algorithm and a Dense Net-based convo-lutional neural network cooperative beamforming algorithm.These two network architectures can achieve lower training overhead while predicting the RF beamforming vector of the base station,and at the same time improve data efficiency,have better feature extraction capabilities,and achieve better results.In addition,it can achieve wide coverage and low-delay cooperative beamforming gain with lower coordination overhead,and realize high-mobility millimeter wave applications.The simulation stage is based on accurate ray tracing experiments,showing that the cooperative beamforming method based on these two learning models can find the best RF beamforming vector of the base station with lower training overhead,and its performance is close to the optimal genie- The achievable rate of the Aided(without considering training overhead)solution.The above methods in this article can achieve better performance than the basic cooperative beamforming method.Finally,the shortcomings of the above methods and the next improvement direction are analyzed.