Research On Distributed Edge Caching In Fog Radio Access Network

The rapid development of smart devices and mobile applications brings an unprecedented traffic pressure to the wireless networks.The large service delays have been unable to meet the upcoming real-time requirements,which has become the main technical bottleneck of the traditional Cloud Radio Access Network(CRAN).One promising improvement is through the dense deployment of caching resources at the network edges.At this point,fog radio access network(FRAN)has been proposed as an evolution form of heterogeneous cloud radio access networks.By exploiting computing and storage capabilities in the edge devices,FRANs can effectively alleviate the backhaul load and reduce the user request delay by prefetching and caching the right contents.To efficiently utilize the edge resources and avoid the extra communication with the control center,the distributed edge caching placement in FRAN has become a research hotspot.However,current related works either assume that the essential network parameters are known in advance,or fail to adapt to time-varying user requests.To address these problems,this thesis aims to investigate the distributed caching placement in FRAN.Firstly,a distributed caching scheme based on potential game is studied.By considering the subtle influence of network delay from network state,an edge caching optimization problem to minimize average network delay is investigated.For instance,when a user is served by multiple F-APs,the user can dynamically select the service F-AP when requesting different content,resulting in different network delays.When users have to obtain content through the backhaul link,considering the different quality of the F-APs' backhaul link,the resulting network delay also varies greatly.Considering the limit backhaul resources of F-APs,the release and update of every cache content need to stagger the peak traffic period to avoid increasing congestion.Furthermore,to minimize average delay in a distributed manner,a caching scheme based on potential game is proposed.A suitable potential function and utility function are constructed to cast the optimization problem into a potential game.And the existence of equilibrium point of the game is proved.Simulation results show that the proposed distributed caching scheme can effectively reduce the average network delay.Secondly,based on the above scheme,an improved distributed caching scheme based on potential game is studied.First,the topological relationship between F-AP and users,user wireless link quality,F-APs' backhaul link quality,and content popularity are all considered to affect the caching scheme.Then,inspired by a graph theory based dynamic programming algorithm,an improved distributed caching scheme based on potential game is proposed,which both reduce complexity and accelerate convergence.Simulation verifies the better performance of the new scheme.Finally,assuming that F-AP does not know any network information in advance,a fully distributed caching scheme based on reinforcement learning is put forth to maximize the long-term cache hit rate.First,a supplementary model is established to cast the optimization problem into a multiple agents reinforcement learning process.Secondly,a two-phase reinforcement learning algorithm is given.The convergence,stability and optimality of this algorithm are all proved.Finally,the simulation verifies the convergence of the algorithm.Also,the performance of the algorithm is compared with the classic caching scheme like Least Recent Used(LRU),Least Frequent Used(LFU),and First In First Out(FIFO),which further verify the superiority of the algorithm.