Coded Caching for a Large Number Of Users

TL;DR

This paper introduces a novel group-based centralized coded caching scheme that reduces delivery rates for large user populations, extends it to decentralized settings, and demonstrates superior performance through theoretical analysis and simulations.

Contribution

A new group-based centralized coded caching scheme is proposed, achieving lower delivery rates, and its extension to decentralized caching is also demonstrated to outperform existing methods.

Findings

Achieves smaller delivery rates than previous schemes.

Extends to decentralized caching with improved performance.

Validated through theoretical analysis and numerical simulations.

Abstract

Information theoretic analysis of a coded caching system is considered, in which a server with a database of N equal-size files, each F bits long, serves K users. Each user is assumed to have a local cache that can store M files, i.e., capacity of MF bits. Proactive caching to user terminals is considered, in which the caches are filled by the server in advance during the placement phase, without knowing the user requests. Each user requests a single file, and all the requests are satisfied simultaneously through a shared error-free link during the delivery phase. First, centralized coded caching is studied assuming both the number and the identity of the active users in the delivery phase are known by the server during the placement phase. A novel group-based centralized coded caching (GBC) scheme is proposed for a cache capacity of M = N/K. It is shown that this scheme achieves a smaller delivery rate than all the known schemes in the literature. The improvement is then extended to a wider range of cache capacities through memory-sharing between the proposed scheme and other known schemes in the literature. Next, the proposed centralized coded caching idea is exploited in the decentralized setting, in which the identities of the users that participate in the delivery phase are assumed to be unknown during the placement phase. It is shown that the proposed decentralized caching scheme also achieves a delivery rate smaller than the state-of-the-art. Numerical simulations are also presented to corroborate our theoretical results.