Learning distributed caching strategies in small cell networks

Caching has emerged as a vital tool in modern communication systems for reducing peak data rates by allowing popular files to be pre-fetched and stored locally at end users' devices. With the shift in paradigm from homogeneous cellular networks to the heterogeneous ones, the concept of data offloading to small cell base stations (sBS) has garnered significant attention. Caching at these small cell base stations has recently been proposed, where popular files are pre-fetched and stored locally in order to avoid bottlenecks in the limited capacity backhaul connection link to the core network. In this paper, we study distributed caching strategies in such a heterogeneous small cell wireless network from a reinforcement learning perspective. Using state of the art results, it can be shown that the optimal joint cache content placement in the sBSs turns out to be a NP-hard problem even when the sBS's are aware of the popularity profile of the files that are to be cached. To address this problem, we propose a coded caching framework, where the sBSs learn the popularity profile of the files (based on their demand history) via a combinatorial multi-armed bandit framework. The sBSs then pre-fetch segments of the Fountain-encoded versions of the popular files at regular intervals to serve users' requests. We show that the proposed coded caching framework can be modeled as a linear program that takes into account the network connectivity and thereby jointly designs the caching strategies. Numerical results are presented to show the benefits of the joint coded caching technique over naive decentralized cache placement strategies.