Research On Radio Resource Management Strategy For Edge Computing

Mobile edge computing(MEC)sinks cloud services to the edge network close to the user side,meeting the requirements of high-speed,low-latency and low-power consumption for computing-intensive tasks such as virtual reality and smart home.Due to the limitation of edge network resources and the complexity of network environment,the design of reasonable and efficient radio resource management strategy can not only improve the quality of service(Qo S)of users,but also make full use of network resources,so as to improve the utilization of system resources.Therefore,it is necessary to study the radio resource management strategy in the edge network.Firstly,this dissertation introduces the research status of radio resource management in edge network,summarizes the key challenges faced by edge computing and discusses the feasible solutions.Secondly,aiming at the problem of system delay performance deterioration caused by underutilization of network resources in multi-user multi-server edge network model,a solution combining data compression and edge computing is proposed to minimize system delay.Firstly,the deep learning method is used to compress user's computing tasks on the mobile device side,so as to further reduce the uplink transmission time of data and improve the delay performance of the system.Secondly,considering the insufficient utilization of edge cloud resources caused by different ways and objects of task offloading,an adaptive task splitting ratio algorithm is designed.In order to make full use of the edge cloud resources and improve the delay performance of the system,part of the computing tasks is segmented for cloud computing.Thirdly,aiming at the problem of polarization of Qo S obtained by users in different network environments in multi-user single server edge network model,this dissertation proposes an online task offloading algorithm based on deep reinforcement learning(DRL).Firstly,convex optimization and other related mathematical theories are used to solve the optimal bandwidth allocation,local computing frequency and computing resource allocation of edge server under the known task unloading strategy,which realizes the relative fairness of resource allocation among users in different network environments.Secondly,for the task offloading decision-making problem,an improved DRL online task offloading algorithm is proposed.Different from the existing deep reinforcement learning algorithm,this algorithm uses multiple deep neural networks(DNN)to improve the?-greedy exploration strategy in deep reinforcement learning,which makes the algorithm convergence faster and accuracy higher.Finally,the main research work and existing problems of this dissertation are briefly summarized and future prospects are given.