Learning-Based Caching With Unknown Popularity In Wireless Video Networks

Caching at the small base station (SBS) is a promising architecture to alleviate the highly-loaded wireless video networks. SBS can cache popular video files, thus serves mobile users without going through backhaul connection to the core network and provides content-level offloading. This paper proposes a novel method for the content caching problem that optimizes cache performance. Our proposed algorithm learns the content popularity profile by predicting the probability of files to be requested, and then refreshes the cache based on the learned content popularity. Popularity learning method runs in an online fashion and has no assumption of the file requests, thus it can be used for predicting either fixed or time-varying popularity. Our simulation results show that our proposed algorithm has similar performance compared to the traditional algorithms when the content popularity profile is fixed, and performs better than other algorithms when the content popularity profile is time-varying, which is more realistic.