Cache-Aided Networks with Shared Caches and Correlated Content under Non-Uniform Demands

In this paper, we propose a cache-aided delivery network with multiple shared caches and correlated content under non-uniform popularity demand. From an information theoretic perspective, we formulate the caching problem as a distributed source coding with side information at the decoder. To address the placement challenge, we propose an automatic clustering scheme considering the popularity and similarity of library content to extract the most efficient side information for caching. Next, we use a hybrid placement strategy, in which the popular side information is fully stored in all caches, while the clusters' side information is partially placed in different caches according to the coded caching strategy (CC). In the delivery phase, the server transmits coded multicast messages and encoded messages (refinement segments) so that users can reconstruct the cluster representatives and clustered files. Our simulation result demonstrates a significant improvement in the peak delivery rate of the system, which resulted from our perspective on problem formulation and the careful extraction of side information during the placement phase.