Scalable video coding (SVC) has been widely used in video-on-demand (VOD) service, to ef-ficiently satisfy users' different video quality require-ments and dynamically adjust video stream to time-variant wireless channels. Under the 5G network structure, we consider a cooperative caching scheme inside each cluster with SVC to economically utilize the limited caching storage. A novel multi-agent deep reinforcement learning (MADRL) framework is pro-posed to jointly optimize the video access delay and users' satisfaction, where an aggregation node is in-troduced helping individual agents to achieve global observations and overall system rewards. Moreover, to cope with the large action space caused by the large number of videos and users, a dimension de-composition method is embedded into the neural net-work in each agent, which greatly reduce the com-putational complexity and memory cost of the rein-forcement learning. Experimental results show that:1) the proposed value-decomposed dimensional net-work (VDDN) algorithm achieves an obvious perfor-mance gain versus the traditional MADRL;2) the pro-posed VDDN algorithm can handle an extremely large action space and quickly converge with a low compu-tational complexity.

Youjia Chen;Yuekai Cai;Haifeng Zheng;Jinsong Hu;Jun Li

Fujian Key Lab for Intelligent Processing and Wireless Transmission of Media Information,College of Physics and Information Engineering,Fuzhou University,Fuzhou 350000,China;School of Electronic and Optical Engineering,Nanjing University of Science and Technology,Nanjing 210000,China

中国通信（英文版）

[1]Youjia Chen;Yuekai Cai;Haifeng Zheng;Jinsong Hu;Jun Li-.Cooperative Caching for Scalable Video Coding Using Value-Decomposed Dimensional Networks)[J].中国通信（英文版）,2022(09):146-161

Decomposed,ficiently,VDDN

Cooperative,Caching,Scalable,Video,Coding,Using,Value,Dimensional,Networks,coding,SVC,has,been,widely,demand,VOD,service,ef,satisfy,users,different,quality,require,ments,dynamically,adjust,stream,variant,wireless,channels,Under,network,structure,we,consider,cooperative,caching,scheme,inside,each,cluster,economically,utilize,limited,storage,novel,multi,deep,reinforcement,learning,MADRL,framework,jointly,optimize,access,delay,satisfaction,where,aggregation,node,troduced,helping,individual,agents,global,observations,overall,system,rewards,Moreover,cope,large,space,caused,by,number,videos,composition,method,embedded,into,neural,which,greatly,reduce,putational,complexity,memory,cost,Experimental,results,show,that,proposed,value,decomposed,dimensional,algorithm,achieves,obvious,perfor,mance,gain,versus,traditional,can,handle,extremely,quickly,converge,low,compu

0.625715