Resource Collaborative Research Of NOMA-MEC System Based On Learning Algorithm

With the development of 5G technology,the increasing data traffic has brought great challenge to the limited computing and spectrum resources.The emergence of mobile edge computing(MEC)and non-orthogonal multiple access(NOMA)technologies bring solutions to the above problems.By offloading the computing task to the edge server,MEC can reduce the computing delay and energy consumption.Non-orthogonal multiple access technology plays a great role in supporting more users' connection and improving spectral efficiency.At the same time,compared with traditional algorithms,reinforcement learning requires less model information and is more suitable for realistic scenarios.In this paper,the influence of computing resource,spectrum resource and model information are considered comprehensively,and based on reinforcement learning algorithm,mobile edge computing and non-orthogonal multiple access technologies are studied in depth.Firstly,in some applications,since calculation results are often used repeatedly by users,caching the popular calculation results can avoid the task being repeated,thus effectively reducing the delay.This part combines MEC and NOMA technologies to jointly optimize caching and computing decisions based on reinforcement learning algorithm.By extending the local caching of single server to collaborative caching of multiple servers,computation results can be shared among multiple servers.For the content cached by the server,the popularity is taken as the cache standard,and calculated results are placed on the corresponding server according to the popularity.According to the contents of caching,the multi-agent reinforcement learning algorithm is used to find the optimal caching and offloading decisions.Simulation analyses show that the collaboration between multiple servers is beneficial to users hit cache,and compared with other schemes,the proposed multi-agent reinforcement learning algorithm has significant advantages in reducing the delay.Next,with the development of real-time application equipment,how to maintain the freshness of information has become an urgent problem.In this part,the influence of MEC and NOMA technologies on information freshness are comprehensively considered.In the transmission stage,the device will be attacked by jamming.By optimizing the amount of uploading tasks and offloading power to resist jamming and reduce transmission time,the goal of minimizing the average update cost can be achieved.Then,considering the dynamic change of channel parameters in different time slots,multi-agent reinforcement learning algorithm is used to optimize the offloading factor and offloading power.The simulation results show that the partial offloading scheme has the best effect in reducing the average update cost.At the same time,the multi-agent reinforcement learning algorithm can effectively reduce the average update cost in the multi-device scenario.Moreover,reducing the number of devices are also means to reduce the average update cost.