In-Network Caching for Hybrid Satellite-Terrestrial Networks Using Deep Reinforcement Learning

Abstract

Large number of redundant requests in wireless networks have led to the hybrid satellite-terrestrial networks, where a satellite is used for content placement at edge caches at the base stations (BSs), thereby reducing backhaul link usage. In this paper, we consider in-network caching where an unavailable content at one BS can be fetched from the nearest BS in the network, before requesting from the content server. Obtaining optimal placement incurs exponentially huge computational overhead. Recent caching solutions are not scalable for large size of content library. Therefore, we propose a low-complexity approach using an action-coded deep deterministic policy gradient (AC-DDPG) algorithm towards optimizing the long-term average network delay. The proposed approach employs continuous valued popularity profiles rather than a fixed finite set in the literature. Simulation results demonstrate the successful application of proposed approach and the improvement over the most-popular content caching method.