Research On Optimal Resource Allocation Via Machine Learning In Coordinated Downlink Multi-cell OFDM Networks

With the advent of the 5th generation(5G)mobile communications,the number of mobile terminal devices connected to the wireless networks is increasing exponentially.At the same time,the needs of users are not limited to services such as text messaging and voice calls.Services such as social networking and multimedia are also important to users.So how to efficiently use limited spectrum resources and provide more flexible user access has become a crucial issue.Obviously,denser cellular deployment can increase user access.Besides,orthogonal frequency division multiplexing(OFDM)is one of the key techniques in wireless communications systems and has been adopted as the major access scheme for both 4G and 5G networks.Therefore,the wireless resources allocation in multi-cell OFDM systems has received extensive attention.However,the wireless communications environment is time-varying,especially in a high-mobility communication environment.Thus,the wireless resource allocation with real-time computing capability has become imperative.This dissertation will study the real-time wireless resource allocation scheme for the downlink of cooperative multi-cell OFDM systems.Considering a multi-cell OFDM downlink network,a basic problem is to perform resource allocation to maximize the spectral efficiency(SE).In this dissertation,we divide it into a user scheduling subproblem and a power allocation subprolem,and then adopt a resource allocation algorithm based on imperfect channel state information(CSI).Universal frequency reuse is considered,and the cochannel interference is dealt with via the cooperation of multiple base stations(BSs)sharing CSI but not user data.Since the wireless communications environment may change rapidly and need real-time computation,we then propose a deep neural network(DNN)approach to approximate the resource allocation algorithm,which greatly reduces then computation time and is capable of ”on-the-fly” adaptation to a time-varying environment.Simulation results verify the effectiveness of the DNN implementation,especially when the number of cells and subcarriers is large.On the other hand,Doppler shift leads to a loss of subcarrier orthogonality in OFDM systems,resulting in inter-carrier interference(ICI),especially in a high speed environment.We therefore solve the resource allocation problem by considering ICI caused by Doppler spread and imperfect channel state information(CSI)caused by estimation errors,quantization errors and feedback delay.However,the resultant resource allocation algorithm is so complicated that it may not be applicable to the wireless communications environment under high mobility since it may change rapidly and therefore needs real-time computation.As such we propose a deep neural network(DNN)approach to approximate the resource allocation algorithm,which greatly reduces the computation time while achieving very good prediction accuracy.Simulation results verify the influence of Doppler shift on the SE performance and the effectiveness of DNNs in terms of computing time.