Research On Deep Reinforcement Learning Based Resource Allocation For Multi-cell OFDMA Systems

Orthogonal frequency division multiple access(OFDMA)technology divides the transmission bandwidth into a series of mutually orthogonal sub-carrier sets,and assigns different sub-carrier sets to different users to achieve multiple access.It also dynamically allocates available bandwidth to users to achieve optimal utilization of resources and increase the data transmission rate.Therefore,for different application scenarios of the 5th generation mobile communication technology(5G),this thesis studies the resource allocation of multi-cell OFDMA system based on a multi-agent deep reinforcement learning algorithm under the premise of considering system transmission rate,user fairness and user service quality.The main contributions of this thesis are summarized as follows.Firstly,we study the sub-channel assignment and power allocation problem in a downlink multicell single-service OFDMA system aiming at transmission rate.However,traditional optimization algorithms face difficulties such as complex system models and high computational overhead in solving such complex problems.Therefore,we propose a multi-agent deep reinforcement learning resource allocation algorithm with centralized training and distributed execution mechanism.According to the channel state information of the corresponding cell,each agent first generates the sub-channel assignment scheme of each cell,then obtains the power allocation scheme of all users in each cell,and finally updates the resource allocation scheme based on the feedback reward information.The simulation results show that the multi-agent deep reinforcement learning algorithm effectively improves the transmission rate of the system.Secondly,we further propose a multi-agent deep reinforcement learning optimization algorithm based on a decision-driven mechanism to solve the complex resource allocation problem of enhanced mobile broadband(e MBB)and ultra-reliable and low-latency communication(URLLC)in downlink multi-cell OFDMA systems.The method monitors the channel state information of the multi-cell system,and executes the process of multi-agent deep reinforcement learning when necessary,thereby saving computing overhead and computing time.Simulations show that this optimization method based on the decision-driven mechanism can reduce unnecessary learning times while ensuring the service satisfaction level of e MBB and URLLC users,thereby optimizing the efficiency of the resource allocation method.