On A Deep Reinforcement Learning-Based Content Caching Strategy in 6G Space-Air-Ground Integrated Networks

Space-Air-Ground Integrated Networks (SAGINs) have been widely recognized as an important 6G hierarchical architecture for satisfying content caching and delivering demands of users. However, the complicated communication conditions and dynamic network topology pose great challenges for SAGINs to meet the diversified Quality of Service (QoS) requirements of users. Caching popular contents on Unmanned Aerial Vehicles (UAVs), which are closer to the ground end users, can efficiently improve the QoS and decrease the communication pressure of satellites. In this paper, targeting at the spatial-temporal popularity of the contents, we propose a Deep Reinforcement Learning (DRL)-based content selection approach to optimize the Cache Hit Ratio (CHR) using a Deep Q-network (DQN) algorithm. The simulation results demonstrate that the proposed approach can significantly improve the long-term CHR under various conditions compared with conventional caching strategies.