Learning Distributed Coded Caching Strategy in a Cellular Network

The caching of popular contents in a cellular network is known to reduce the data load in the backhaul link, and have been an active area of research. This paper considers the problem of efficient distributed content coded caching in a small-cell Base Station (sBS) wireless network to improve the cache hit performance. The demands at each sBS across time and sBSs is assumed to be correlated, and is unknown. A new weighted (across time and sBS) caching strategy is proposed. A high probability lower bound on the cache hit is derived, which is obtained using the proposed strategy as a function of the cache hit of the optimal caching strategy. The bound is shown to depend on (i) the weighted average of cache hits, (ii) regret, and (iii) the discrepancy across time and sBSs (a measure of correlation of demands across time and sBSs). This provides the following insight on obtaining the caching strategy: (i) find a sequence of caching strategies by running regret minimization across time at each sBS, and (ii) maximize an estimate of the bound to obtain a set of weights. The insight is shown to result in an iterative distributed algorithm to obtain caching strategies at each sBS. The performance of the proposed caching strategy is shown to outperform Least Recently Frequently Used (LRFU) algorithm by a large margin.