Content Caching Optimization Based On Learning In Mobile Radio Access Network

Mobile data traffic experiences explosive increase because of multime-dia services,ubiquitous social networks and computation-intensive appli-cations.Huge traffic brings about bandwidth burden and resource alloca-tion problem in mobile radio access networks.According to statistics,most of the traffic originates from some popular content requested by multiple mobile users,which causes duplicated data transmission due to frequent download of those popular contents.Therefore,deployment of content caching on radio access network equipment and mobile user equipment has become a promising solution for Mobile Edge Computing.However,only some of the content can be cached on edge nodes and/or user devices be-cause of limited storage capacity compared with increasing large amount of content.Therefore,the optimal placement of content caching in mobile radio access network has become a hot research direction.In this thesis,the optimization problem of content caching in IoP-ori-ented mobile wireless access network is investigated.The main research work includes the following aspects:(1)In the scenario IoP-oriented mobile wireless access network,con-sidering that a mobile user is associated with multiple terminal devices and some mobile users via social networks,the social-aware proactive content caching optimization problem is formulated by maximizing the social re-ward of devices.In order to solve the above optimization problem,four learning-based optimization algorithms are proposed,including independ-ent reinforcement learning based algorithm,joint action reinforcement learning based algorithm,joint shared Q-value reinforcement learning based algorithm and joint back reward reinforcement learning based algo-rithm.Simulations results show the efficiency of the performances of the four proposed algorithms in terms of average social rewards,cache hit ratio and cooperative hit ratio.(2)In the scenario of mobile radio access network,the problem of con-tent popularity prediction is addressed,and two content popularity predic-tion algorithms are proposed,including content popularity prediction algo-rithm based on integrated features of user,and content and content popu-larity prediction algorithm based on integrated features of user and content by federated learning.Based on the predicted content popularity,a global content placement optimization approach based on local content popularity is presented to maximize long-term cache hit ratio of the system.The per-formances of the proposed algorithms are evaluated based on MovieLens dataset.Simulation results show that the proposed content popularity pre-diction algorithms have good performance in terms of prediction accuracy,and the proposed federated learning-based algorithm outperforms in terms of global hit ratio and transmission cost compared with the baseline algo-rithm.