Mobility-Aware Proactive Video Segment Caching Based on Deep Reinforcement Learning

Maintaining efficient and successive video streaming services in cellular networks is challenging due to user mobility and ever-increasing volume of data traffic. A promising solution is to cache popular contents at the edge of wireless networks. Although caching schemes have been widely discussed, few of them jointly considered the characteristics of streamed video data and user mobility. In this paper, we construct a dynamic caching decision framework based on Long Short-Term Memory (LSTM) and Deep Q-network (DQN). Based on this framework, a mobility-aware segment-level caching strategy is proposed to maximize the cache hit rate. Simulation results show that our proposed method can achieve 20% performance improvement by comparing with baseline caching algorithms.