Learning-Based Content Caching with Update Strategy for Fog Radio Access Networks

Aiming at improving the edge caching efficiency of the fog radio access network (F-RAN), this paper proposes a distributed content caching scheme based on user preference prediction and content popularity prediction. Under the constraint that storage capacity of each user is limited, we formulate the optimization problem to maximize the caching hit rate. Then, by taking users' selfishness into consideration, user preference and content popularity are predicted through popular topic models. Finally, the Q-learning based content caching algorithm is applied to get the optimal content caching matrix with the predicted user preference and content popularity. Moreover, we also propose a content update policy, so that the proposed algorithm can track the variations of contents popularity in a timely manner. Simulation results demonstrate that the proposed algorithm achieves better caching hit rate compared with existing algorithms.