Performance Analysis And Traffic Offloading Optimization Of Mobile Edge Computing System

With the emergence of emerging mobile applications,the demand for data transmission and processing of mobile communication network users is growing rapidly.Mobile Edge Computing(MEC)technology which refers to the deployment of computing servers at the edge of the network close to users,can provide users with low latency data processing services.At the same time,the dense deployment of base stations in Ultra-dense Networks(UDN)can greatly improve the transmission capacity of wireless access network.The combination of MEC and UDN brings convenience for new services with large data transmission requirements and low delay data processing requirements,and has become a mobile communication scenario with great concern.However,the dense deployment of small base stations has brought serious interference problems.How to set the density of base stations and configure MEC servers reasonably to achieve the matching of transmission performance,computing service capacity and user demand,and how to select the appropriate access node in the dense deployment of small base stations to achieve efficient offloading of user computing tasks have become urgent problems to be solved.At the same time,how to design an appropriate MEC service solution for specific business has become a hot topic for researchers.In order to solve the above problems,this paper first studies the performance analysis of MEC enhanced UDN and the design of task offloading node selection algorithm,and takes the short video task application supported by MEC as an example,the paper further designs a joint optimization algorithm of radio bearer control and short video quality selection for user experience.The specific work are as follows:(1)Based on the statistical analysis of uplink interference of small base stations,the uplink space traversal capacity of MEC enhanced UDN is explored,and the configuration of small base stations is designed.Firstly,the spatial Poisson point process is used to model the user distribution of interference sources,and take the multi antenna and small-scale fading characteristics of the wireless channel into consideration.the statistical characteristics and change rules of the uplink user signal to interference ratio(SIR)are analyzed with the deployment density of the smallbase station as the variable.Then,the constraint relationship between the spatial traffic intensity and the computing capacity of the MEC server is calculated according to the queuing theory.Finally,the correctness of the theoretical relationship between SIR and base station density is verified by the numerical simulation,the convergence trend of spatial ergodic capacity is obtained.An example of configuration design of small base station is given.(2)In MEC enhanced UDN,a Deep Reinforcement Learning(DRL)based Traffic Offloading Node Selection(DRL-TONS)algorithm for multi-user and multi base stations is designed.In this study,the transmission and computing service delay of users is taken as the optimization objective,and considering wireless channel resources and MEC computing resources,node selection problem of task offloading access is modeled as Markov Decision Process(MDP).Based on the Policy Gradient algorithm in DRL,DRL-TONS algorithm is designed to solve the training problem.The training results show that the performance of DRL-TONS algorithm is better than that of contrast algorithm.The test results also show that,when the task arrival rate changes between 0.4 and 1.0,the deep strategy network trained at the task arrival rate of 0.8 still has adaptability.(3)In the scenario of MEC supporting short video service,a joint optimization algorithm of Policy Gradient based Quality selection and Radio bearer control(PG-QR)is proposed.In this paper,the cost function of long-term video quality benefit,wireless bearer cost and task delay penalty is taken as the joint optimization objective,which is modeled as MDP,and PG-QR algorithm is proposed based on the strategy gradient algorithm in DRL.The simulation results show that the training performance of PG-QR algorithm is better than that of the contrast algorithm with different weight coefficient settings.