Coded Caching with Heterogeneous User Profiles

Coded caching has been proven to be a useful technique for reducing traffic in networks with point-to-multipoint links. The key idea is to pre-fetch popular content at the end users during off-peak hours, and encode transmissions when resources are scarce in such a way that different users can obtain different information from the same packet.Prior works have proposed placement and delivery algorithms to address a wide range of scenarios with varying file popularities, file sizes, and cache capacities. Both centralized and distributed algorithms have been proposed, and their performance limits have been characterized in terms of peak and average transmission rates. However, existing works have focused on the case where all the end users present an identical distribution of requests; in other words, the popularity of a given file is identical for all users. This assumption is overly simplistic in modern networks, where network operators often build detailed individual profiles on each user.This paper proposes and compares the peak rate of three coded caching schemes when the end users can be classified into distinct groups with different distribution of demands. Specifically, the first scheme ignores the differences between user profiles, the second performs independent caching and delivery of each class of files, and the third ignores the similarities between user profiles. The second scheme derives a method to partition the cache between widely popular files and files that may only be requested by a subset of the users. Our analysis yields some counter-intuitive results.