Research On Dynamic User Association Based On Historical Data In Ultra Dense Networks

With the continuous development of communication technology and the widespread application of various smart devices on the user’s side,the wireless traffic generated by users increase dramatically.In order to cope with the rise for wireless traffic,ultra-dense base stations have begun to deployed by network operators to ensure the quality of service for users.However,new problems will be caused such as network congestion or load imbalance.If the future traffic of base station can be forecasted and then the user association strategy will be adjusted according to the prediction,the performance of system will be greatly improved.Hence,the problem about user association based on traffic prediction in ultra-dense networks which aims to achieve load balance is studied in this paper.Firstly,introductions about the basic knowledge of ultra-dense networks and convolutional neural networks are elaborated.The knowledge about the ultra-dense cellular network,user association,convolutional neural network and the convex optimization theory involved in this article are introduced.Secondly,in consideration of the two key characteristics about the temporal correlation and spatial correlation of the wireless traffic data,a densely connected convolutional neural network based on spatiotemporal feature is proposed to study the prediction for future wireless traffic in this paper.From the network architecture,there are two greatly significant characteristics for this strategy.One is that the convolutional layers are densely connected,and the other is that the correlation of the traffic data in the wireless cellular network in the time dimension and spatial dimension are considered jointly,and a fusion mechanism is proposed to fuse the characteristics of the traffic data about adjacency and periodic.Simulations on the real dataset provided by Telecom Italia show that the characteristics of the traffic data can be effectively captured by the algorithm and the performance of prediction is improved significantly.Finally,the problem about dynamic user association based on traffic prediction is studied.In order to achieve the overall load balance of the system as well as ensure the user’s quality of service to a certain extent,the problem about user association is modeled as maximizing the utility function of the load balancing index.It is more difficult to solve the problem directly because of multiple constraints.Hence,the original convex problem is transformed into the dual problem with the method of Lagrange duality,and a load-aware dynamic user association algorithm is proposed to solve the dual problem.Meanwhile,the proof that the solution to the dual problem equals to the approximate optimal solution of the original problem is given.The simulation results prove that the overall load balance of the system is realized and the long-term stability guarantee for the ultra-dense cellular network is also provided by the algorithm.