Multicast-Aware Proactive Caching in Wireless Networks with Deep Reinforcement Learning

We consider mobile users randomly requesting contents from a single dynamic content library. A random number of contents are added to the library at every time instant and each content has a lifetime, after which it becomes irrelevant to the users, and a class-specific request probability with which a user may request it. Multiple requests for a single content are served through a common multicast transmission. Contents can also be proactively stored, before they are requested, in finite capacity cache memories at the user equipment. Any time a content is transmitted to some users, a cost, which depends on the number of bits transmitted and the channel states of the receiving users at that time instant, is incurred by the system. The goal is to minimize the long term expected average cost. We model the problem as a Markov decision process and propose a deep reinforcement learning (DRL)-based policy to solve it. The DRL-based policy employs the deep deterministic policy gradient method for training to minimize the long term average cost. We evaluate the performance of the proposed scheme in comparison to traditional reactive multicast transmission and other multicast-aware caching schemes, and show that the proposed scheme provides significant performance gains.