JOINT CONTENT POPULARITY PREDICTION AND CONTENT DELIVERY POLICY FOR CACHE-ENABLED D2D NETWORKS: A DEEP REINFORCEMENT LEARNING APPROACH

摘要

Compared with traditional device-to-device (D2D) communication networks, the users in the cache-enabled D2D communication networks can easily obtain their requested contents from the nearby users, and reduce the backhaul costs. In this paper, we investigate the caching strategy for the cache-enabled D2D communication networks, with the consideration of caching placement and caching delivery. The content popularity and user mobility are predicted by a machine learning approach of echo state networks (ESNs) in order to determine which content to cache and where to cache. Furthermore, a deep Q-learning network (DQN) algorithm is proposed to optimize the content delivery problem, with taking the delays and energy consumption into consideration. Simulation results show that the content hit rate and the traffic offloading can be remarkably improved with the proposed approach, compared to the random caching strategy.