Research On Content Popularity Prediction And Edge Caching In Fog-ran

With the explosive growth of wireless data traffic,the higher requirements of user service agility and realtime response,wireless networks have been confronted with tremendous challenges with limited resources.Due to capacity-limited fronthual links and centralized baseband unit pool,traditional centralized cloud radio access networks(C-RAN)may cause traffic congestion and communication interruptions especially at peak traffic moments.In order to meet the various requirements of future wireless communications,there is an urgent need for new technological innovations.By placing most popular contents as close as possible to the requesting users and localizing the user request,the edge caching technique in fog wireless access networks(FRAN)can effectively reduce fronthaul load and communication delay,which has become a feasible solution and has received extensive attention.However,due to the randomness and diversity of content requests and the temporal and spatial dynamic of content popularity,many problems have arisen for the implementation of effective edge caching technique.In this paper,we will systematically study the content popularity prediction methods,edge caching policy,and edge caching framework design to solve the above problems.Firstly,the content popularity prediction algorithm based on user preference learning is studied.Firstly,unlike the static approach,we redefine content popularity in terms of time and space from the perspective of regional users as the average request possibility of regional users for a certain content at the current moment.Secondly,we propose an online content popularity prediction algorithm based on the current user preference model,which can predict the future content popularity of a certain region in an online fashion without any restriction on content types and track the popularity change in real time.In addition,we propose an offline user preference learning algorithm,which can discover the user's own preference through its historically requested information.By monitoring the average prediction error in real time,it can be initiated automatically for relearning of user preference and continuous offline learning can thus be avoided.Finally,we theoretically analyze the performance of the algorithm,derive the upper bound of the popularity prediction error of our proposed online content popularity prediction algorithm,reveal the sub-linear relationship between the cumulative prediction error and the total number of content requests in a specific region.Secondly,the edge caching policy based on content popularity is investigated.Taking into account the randomness and diversity of content requests and the impact of mobile users on cache performance,the cache optimization problem that maximizes the cache hit rate is constructed combining with the characteristics of content requests,and a dynamic content caching and updating method is designed.We analyze the performance of our proposed edge caching policy.We first derive the upper bound of the popularity prediction error of our proposed online content popularity prediction algorithm,and reveal the sub-linear relationship between the cumulative prediction error and the total number of content requests.We then derive the regret bound of the overall cache hit rate of our proposed edge caching policy,and show through theoretical analysis that our proposed policy has the capability to achieve the optimal performance asymptotically.Finally,Numerical results also show the superior performance of our proposed edge caching policy in comparison with other traditional edge caching policy,and verify the asymptotical optimality of our proposed edge caching policy.Finally,learning-based edge caching framework design in fog wireless access network(F-RAN)investigated.According to the characteristics of F-APs and user equipments(UEs),two learning-based edge cache frameworks are designed.An F-APs-based edge caching framework is designed for situations where F-APs support computing and caching and the UEs does not have intelligent computing.It supports mobile user monitoring,collaboration between nodes,online popularity prediction,content caching,and intelligent updating.In the case that both F-APs and UEs support intelligent computing,an edge cache framework based on smart UEs is designed to place learning on the user equipment end,greatly reducing the signaling interaction and computational burden of F-APs,meanwhile,it ensures the privacy of users.