Degrees of Freedom of Cache-Aided Wireless Interference Networks

TL;DR

This paper characterizes the degrees of freedom in cache-aided wireless interference networks, demonstrating that a separation-based strategy is approximately optimal and revealing trade-offs between cache sizes and network performance.

Contribution

It provides a constant-factor approximation of the DoF for arbitrary network parameters and introduces the symmetric multiple multicast X-channel as a new communication problem.

Findings

Approximate DoF is achieved via a separation strategy of physical and network layers.

Increasing transmitter cache beyond full library storage offers limited additional DoF benefits.

A trade-off exists between receiver cache size and the number of transmitters needed for optimal performance.

Abstract

We study the role of caches in wireless interference networks. We focus on content caching and delivery across a Gaussian interference network, where both transmitters and receivers are equipped with caches. We provide a constant-factor approximation of the system's degrees of freedom (DoF), for arbitrary number of transmitters, number of receivers, content library size, receiver cache size, and transmitter cache size (as long as the transmitters combined can store the entire content library among them). We demonstrate approximate optimality with respect to information-theoretic bounds that do not impose any restrictions on the caching and delivery strategies. Our characterization reveals three key insights. First, the approximate DoF is achieved using a strategy that separates the physical and network layers. This separation architecture is thus approximately optimal. Second, we show that increasing transmitter cache memory beyond what is needed to exactly store the entire library between all transmitters does not provide more than a constant-factor benefit to the DoF. A consequence is that transmit zero-forcing is not needed for approximate optimality. Third, we derive an interesting trade-off between the receiver memory and the number of transmitters needed for approximately maximal performance. In particular, if each receiver can store a constant fraction of the content library, then only a constant number of transmitters are needed. Our solution to the caching problem requires formulating and solving a new communication problem, the symmetric multiple multicast X-channel, for which we provide an exact DoF characterization.