Research On Data Transmission And Processing For Mobile Edge Computing

With the rapid development of Internet of Things(Io T)and 5G communications,the field of mobile computing has gradually shifted from centralized mobile cloud computing(MCC)to mobile-edge computing(MEC).Unlike conventional cloud computing,MEC is promising technology to enable real-time information transmission and intensive computing at the resource-limited mobile devices by migrating mobile computing,network control and storage to the network edges,which can provide low-latency as well as flexible computing and communication services for Io T devices.In this dissertation,by utilizing game theory and reinforcement learning,we investigate the data transmission and processing for MEC systems,together with jointly optimizing the radio and computation resource allocation.Moreover,based on extensive simulations,we compare the binary offloading and partial offloading schemes in detail in terms of energy and latency.Our main work and contributions are detailed below:1.We propose a price-based computation offloading for a multi-user MEC system.In order to manage the offloaded computation tasks from users,the interaction between the edge cloud and users is modeled as Stackelberg game.Specially,the edge cloud sets prices to maximize its revenue subjected to its finite computation capacity,and for given prices,each user locally makes offloading decision to minimize its own cost which is defined as latency plus payment.Depending on the edge cloud's knowledge of the network information,we develop the uniform and differentiated pricing algorithms,which can both be implemented in distributed manners.2.We investigate joint task offloading decision and transmit power allocation for MEC systems with multiple independent tasks.In order to minimize the weighted sum of energy and delay subjected to the finite energy constraint at the edge cloud,we propose a deep reinforcement learning based online offloading algorithm,which learns from historical offloading decisions based on different channel conditions and task attributes.It can automatically improve its offloading actions generated by a deep neural network(DNN),completely avoiding the complex mixed integer programming problem.In addition,we also consider a variety of traditional optimization algorithms for performance comparison.Simulation results validate the effectiveness of the proposed deep reinforcement learning based algorithm.3.We compare the binary offloading and partial offloading schemes of the MEC systems under the same system setup and extensive simulations.Binary offloading scheme has a better performance in terms of energy consumption,while partial offloading scheme can usually achieve a superior latency performance due to its flexible task segmentation and parallel computing.