Content Caching and Delivery over Heterogeneous Wireless Networks

TL;DR

This paper proposes a coded caching and delivery scheme for heterogeneous wireless networks with dense access points and macro-cell base stations, optimizing content delivery by balancing transmission, storage, and user access costs.

Contribution

It introduces a dynamic, popularity-aware coded caching scheme tailored for heterogeneous networks, demonstrating near-optimal performance and practical benefits through numerical evaluation.

Findings

Scheme achieves near information-theoretic optimality.

Dividing content into few popularity classes captures most benefits.

Numerical results validate trade-offs between transmission, storage, and access costs.

Abstract

Emerging heterogeneous wireless architectures consist of a dense deployment of local-coverage wireless access points (APs) with high data rates, along with sparsely-distributed, large-coverage macro-cell base stations (BS). We design a coded caching-and-delivery scheme for such architectures that equips APs with storage, enabling content pre-fetching prior to knowing user demands. Users requesting content are served by connecting to local APs with cached content, as well as by listening to a BS broadcast transmission. For any given content popularity profile, the goal is to design the caching-and-delivery scheme so as to optimally trade off the transmission cost at the BS against the storage cost at the APs and the user cost of connecting to multiple APs. We design a coded caching scheme for non-uniform content popularity that dynamically allocates user access to APs based on requested content. We demonstrate the approximate optimality of our scheme with respect to information-theoretic bounds. We numerically evaluate it on a YouTube dataset and quantify the trade-off between transmission rate, storage, and access cost. Our numerical results also suggest the intriguing possibility that, to gain most of the benefits of coded caching, it suffices to divide the content into a small number of popularity classes.