Learning-Aided Collaborative Caching in Small Cell Networks

Caching enables reduction of peak backhaul load in modern wireless systems by allowing popular files to be pre-fetched and stored locally and closer to the network edge. With the shift in paradigm from homogeneous to heterogeneous cellular networks, data offloading to small cell base stations (sBS) has garnered recent attention. Caching at the sBS can help avoid bottlenecks in the capacity limited backhaul links to the core network. In this paper, we study collaborative caching strategies in a multi-sBS heterogeneous network with unknown file popularities from a reinforcement learning perspective. We present topology-aware uncoded and coded caching strategies under a learning-aided framework where the file popularities are dynamically learned over time by observing user requests. The uncoded collaborative cache placement problem is NP-hard and we propose a novel weighted graph-coloring and local searchbased approximation algorithm with an approximation ratio of ( 1 3 − ) for some > 0. Alternately, we formulate a coded cache placement strategy which is shown to be a linear program yielding an optimal solution. Through simulations we show that the uncoded approximate caching algorithm performs close to the optimal coded scheme when integrated with the learning framework in the multi-sBS setting. We also show that for network topologies of practical interest, the collaborative caching strategies outperform local caching strategies.