Research On Distributed Beam Selection Algorithm For Cell-Free Massive MIMO Systems

With the explosive growth of data traffic,the density of network infrastructure in current communication systems and the number of antennas equipped with each base station have also increased dramatically,resulting in severe inter-cell interference.The cell-free massive MIMO system can well solve this problem and is considered to be one of the key technologies for next-generation communications.The cell-free massive MIMO system eliminates the paradigm of cells,and the base stations are densely deployed and user-centric,and multiple base stations near the user cooperate to provide data transmission services for the user.However,cooperative transmission in this dense base station scenario also brings huge challenges,such as the interaction of a large amount of channel information between multiple base stations and the centralized unit brings unbearable fronthaul link pressure,the increase in the number of beam pairs between base stations and users leads to more serious beam collision problems,etc.Reasonable hybrid precoding design is the key to solving these problems.On the one hand,the hybrid precoding algorithm is required to be executed in a distributed manner,which will greatly reduce the data transmission pressure on the forward link? on the other hand,efficient analog beam selection is required,while reducing the interference between users and ensuring the equivalent beam strength.Therefore,this paper will focus on the distributed beam selection problem in a cell-free massive MIMO system,and design a reasonable and efficient distributed beam selection method.The main work of this paper is as follows:1.Two distributed beam selection algorithms based on potential games are proposed for the beam selection problem of cell-free massive MIMO systems.One is a potential game algorithm based on the Spatial Adaptive Play(SAP)learning algorithm,and the other is an algorithm based on a combination of potential game and linear search.The former has better stability of beam selection results with respect to the number of iterations,lower computational complexity,and can achieve larger sum-rate performance improvement than the traditional fully distributed algorithm? the latter achieves better sum-rate performance than the potential game algorithm based on the SAP learning algorithm,which is close to the effect of the centralized algorithm,but the computational complexity also increases.Compared with the traditional fully distributed and centralized algorithms,these two partially distributed algorithms based on potential games can guarantee better sum-rate performance with distributed execution.2.An unsupervised learning-assisted intelligent distributed beam selection scheme is proposed to address the complexity of beam selection algorithm due to the large beam space.In this scheme,each base station learns the compressed representation of the beam space based on the local received signal strength indication using an unsupervised learning method.Further,the constructed local collaborative base station set performs distributed beam selection among them by a potential game algorithm,and the local rate indicator corresponding to the beam selection is used to help train the base station local beam compression network.Based on the effectively compressed beam space,the potential game distributed beam selection algorithm can achieve higher sum-rate performance based on a small number of iterations compared to the original beam space,while significantly reducing the complexity of the algorithm execution.3.To address the problem of high overhead of acquiring complete channel state information(CSI)at the base station side in cellular-free large-scale MIMO systems,a Dense Net-based base station local supervised beam selection mapping network is proposed based on the unsupervised learning-assisted intelligent distributed beam selection scheme.By constructing a labeled data set of received signal strength indication and beam selection results locally at each base station,the supervised network is trained to output the probability that each code word in the codebook is selected,thus realizing effective beam selection by using local signal strength indication at the base station without the need of complete CSI and without CSI interaction between base stations.The experimental results show that the local beam selection mapping network can achieve a rate performance of more than 92% of the corresponding rate of tag beam selection.