Taming Subpacketization without Sacrificing Communication: A Packet Type-based Framework for D2D Coded Caching

TL;DR

This paper introduces a packet type-based framework for D2D coded caching that reduces subpacketization levels without sacrificing communication efficiency, making finite-length caching more practical.

Contribution

The paper proposes a novel PT design framework that strategically introduces asymmetry to reduce subpacketization while maintaining optimal communication rates.

Findings

Achieves significant subpacketization reduction compared to existing schemes.

Maintains optimal communication rates despite subpacketization reduction.

Provides an ILP formulation for designing rate-optimal caching schemes.

Abstract

Finite-length analysis is critical for bringing coded caching closer to practical deployment. In this work, we study the design of communication rate-optimal device-to-device (D2D) coded caching schemes with minimal subpacketization levels, a key bottleneck in finite-length settings. We present a novel \tit{packet type-based} (PT) design framework that (i) strategically introduces \tit{asymmetry} into file splitting through user grouping, and (ii) systematically exploits such asymmetry in both cache placement and multicast delivery to create subpacketization reduction opportunities. In particular, the induced asymmetry gives rise to two fundamental forms of subpacketization reduction gains: the \emph{subfile saving gain}, achieved by eliminating certain types of subfiles through careful user grouping and transmitter selection, and the \emph{further splitting saving gain}, attained by reducing the splitting granularity for the remaining subfiles. The combined effect of these two reduction gains yields an overall subpacketization improvement over the original Ji-Caire-Molisch (JCM) caching scheme~\cite{ji2016fundamental}, as well as various state-of-the-art schemes, while preserving optimal communication rates. Under the PT framework, we formulate the caching scheme design as an integer linear program (ILP), where each feasible solution corresponds to a valid rate-optimal D2D coded caching scheme with potentially reduced subpacketization relative to the JCM baseline.