Research On NOMA-based Computation Task Offloading Strategy Of Mobile Edge Computing

With the rapid development of the mobile Internet,the number of wireless terminals has grown rapidly,and a large number of applications with computational intensive and low latency requirements have emerged.Traditional mobile cloud computing is difficult to meet the latency requirements of these emerging applications.Mobile Edge Computing(MEC)makes computing services more sinking and closer to the user side by deploying computing resources at the edge of the network,so the latency of computing task offload-ing is significantly reduced,thereby meeting the user's latency requirements.However,MEC provides services for edge devices through wireless networks.Due to the scarcity of wireless spectrum resources,limited transmission bandwidth will become the bottle-neck of MEC system performance in dense access network scenarios.In order to improve the efficiency of wireless resource in the MEC system,non-orthogonal multiple access technology(NOMA)was introduced.NOMA allocates the same time-frequency resource to multiple users to increase the system capacity and spectrum efficiency.However,the introduction of NOMA couples the allocation of user grouping,power allocation,wire-less resources,and computing resources,which brings huge challenges to the modeling and solving of the task offloading problem in the MEC system.Therefore,this paper fo-cuses on user grouping,wireless resource allocation and task offloading decision-making in NOMA-MEC system.This thesis first studied the task offloading of multi-user NOMA-MEC in the sce-nario of low mobility of wireless devices and slow-changing channel.We model the task offloading problem in this scenario as a wireless device energy consumption optimization problem,while optimizing the task offloading amount of the wireless device,the wire-less device transmit power,the user grouping and the allocation of time slots.Due to the non-integer linear programming characteristics of the optimization problem,this thesis decouples the problem into two sub-problems.First,a user grouping algorithm based on the Hungarian algorithm is designed for user grouping,and then it is proved that when the user grouping is determined,the original problem can be transformed to a convex prob-lem.Finally,we verify the effectiveness of the model and the algorithm designed in this paper through numerical simulation.On this basis,we further studied the problem of task offloading when the channel of the wireless device changes dynamically and the task arrives randomly.In this scenario,the wireless device caches tasks that arrive randomly,and in each time slot,it needs to decide whether to offload the computing tasks or process computing tasks locally based on information such as channel,queue length,and computing task queuing time etc.We transform the task offloading decision problem of each wireless device into a Markov de-cision process,and use multi-agent reinforcement learning to make distributed decisions on the task offloading problem of wireless devices,so as to minimize the long-term en-ergy consumption of wireless devices.Finally,we conducted simulation verification to prove the convergence of the multi-agent reinforcement learning algorithm designed in this paper and the superiority of the single-agent and random algorithm in terms of energy consumption and task offloading.