Caching with time-varying popularity profiles

This chapter considers the problem of content caching at the small-cell base stations (sBSs) in a heterogeneous wireless network. A cost function called the “offloading loss, " which measures the fraction of the requested files that are not available in the sBS caches, is used as a metric for evaluating the performance of the proposed caching scheme. In contrast to the previous approaches that consider time-invariant and perfectly known popularity profiles, caching with non-stationary and statistically dependent popularity profiles (assumed unknown, and hence, estimated) is studied from a learning-theoretic perspective. A high probability bound on the offloading loss difference referred to as a probably approximately correct (PAC) result is derived. Here, the offloading loss difference refers to the error between the estimated and the optimal offloading loss. The difference is shown to be a function of the Rademacher complexity, the β-mixing coefficient, the number of time slots, and a measure of discrepancy between the estimated and true popularity profiles. Using insights from this bound, a practical cache update algorithm is proposed. Simulation results are presented to show its superiority over periodic updates. This chapter also presents caching performance analyses for Bernoulli and Poisson request models.