| With the continuous development and popularization of 5G network,the maturity of mobile device browser technology and the enhancement of mobile performance,augmented reality(AR)technology began to become a new research content of global Internet manufacturers As an emerging computer vision technology,AR application has also been widely used in medical treatment,games,housing,trial installation,shopping and other aspects.With the help of mobile device cameras,it renders the content that could not be felt in the real world,so as to provide mobile users with a new interactive experience beyond reality.However,the current AR applications often rely on high-cost and nonportable professional devices,which cannot be well transplanted to users’ mobile devices.Although the new AR implementation scheme based on mobile web solves the difficulty of traditional AR application fast platform,there are still great challenges in the rational allocation of communication network resources in the distributed environment.Therefore,the research on how to use emerging technologies such as edge computing to provide efficient services for mobile Web AR applications in 5g network is worthy of our in-depth analysis and mining.By analyzing the characteristics of mobile Web AR service and using the deep reinforcement learning(DRL)and efficient adaptive learning ability,this paper designs the "device-edge-cloud" distributed collaborative computing framework,so as to meet the requirements of mobile users for service quality and service providers for deployment cost.This paper focus on how to combine DRL algorithm with distributed collaborative computing framework and how to use 5G edge server computing performance and memory resources for computing offloading and content caching.The main contents are as follows:Aiming at the problem of computing task offloading in the 5G network Web AR distributed scenario of "1-n-1",this paper first divides the Web AR computing tasks,abstracts them into a directed acyclic graph and flattens them to describe them,so as to provide the ability of partial offloading.Therefore,different subtask types can be packaged separately and sent to different computing devices for computing,which makes full use of the computing resources of the edge server and improves the computing efficiency.At the same time,a computing offload algorithm based on DRL is proposed to perform the distributed offload decision of Web AR computing tasks,making full use of the flexibility of separable computing tasks and the computing resources of edge servers.Aiming at the problem of uneven load of edge servers and low utilization of computing and storage resources in the 5G network Web AR distributed scenario of "n-n-1".This paper presents a distributed collaborative computing framework for complex network Web AR.Through the asynchronous and parallel advantages of DRL A3 C algorithm,the framework takes advantage of the way that multiple mobile users update the offloading decision network model at the same time to speed up the learning rate and change other DRL computing offloading decision algorithms,which have slow reasoning speed and occupy more storage resources,resulting in difficulties in deploying to mobile device browsers,In order to realize offloading decision,lightweight deployment in mobile devices and load balancing of edge servers.At the same time,using the large storage resources of the edge server and combined with the DRL learning algorithm,the virtual AR target cache mechanism is designed to store the virtual AR target with high matching popularity into the edge server cache,so as to further improve the computing efficiency of Web AR computing tasks as a whole,reduce the computing delay and energy consumption,and further improve the service quality and user experience of Web AR. |
