Caching Strategy Based on Content Popularity Prediction Using Federated Learning for F-RAN

Abstract

Fog radio access network (F-RAN) is envisioned as a promising network architecture for edge computing-based content caching. In this paper, we propose a content caching strategy with content popularity prediction based on Federated Learning (FL) for F-RAN considering Device-to-Device (D2D) communication. More specifically, to obtain the most popular contents while also avoiding individual privacy disclosure, a content popularity prediction model based on FL is designed, where the user preference data of fog user equipment(F-UE) are only utilized in the user’s local model training process. Furthermore, aim at maximizing the cache hit rate, a distributed caching strategy is proposed based on the acquired popularity prediction results and Q-learning algorithm which further incorporates D2D communication in the caching process. Finally, by utilizing the real data set from MovieLens, simulation results demonstrate that the proposed content caching strategy can improve the cache hit rate compared with existing caching policies.