Cache-Aided Combination Networks with Interference

TL;DR

This paper investigates cache-aided combination networks with interference, proposing three schemes to minimize delivery time based on cache sizes and fronthaul capacity, highlighting the impact of fronthaul on scheme performance.

Contribution

It introduces new achievable NDT schemes for different cache configurations and fronthaul capacities in cache-aided combination networks with interference.

Findings

Fronthaul capacity influences the optimal caching scheme.

Soft-transfer scheme performs better with higher fronthaul capacity.

Achievable NDT is characterized for multiple cache and network configurations.

Abstract

Centralized coded caching and delivery is studied for a radio access combination network (RACN), whereby a set of $H$ edge nodes (ENs), connected to a cloud server via orthogonal fronthaul links with limited capacity, serve a total of $K$ user equipments (UEs) over wireless links. Each user, equipped with a cache of size $\mu_R N F$ bits, is connected to a distinct set of $r$ ENs each of which equipped with a cache of size $\mu_T N F$ bits, where $\mu_T$, $\mu_R \in [0,1]$ are the fractional cache capacities of the UEs and the ENs, respectively. The objective is to minimize the normalized delivery time (NDT. Three coded caching and transmission schemes are considered, namely the\textit{ MDS-IA}, \textit{soft-transfer} and \textit{ zero-forcing (ZF)} schemes. The achievable NDT for MDS-IA scheme is presented for $r=2$ and arbitrary fractional cache sizes $\mu_T$ and $\mu_R$, and also for arbitrary value of $r$ and fractional cache size $\mu_T$ when the cache capacity of the UE is above a certain threshold. The achievable NDT for the soft-transfer scheme is presented for arbitrary $r$ and arbitrary fractional cache sizes $\mu_T$ and $\mu_R$. The last scheme utilizes ZF between the ENs and the UEs without the participation of the cloud server in the delivery phase. The achievable NDT for this scheme is presented for an arbitrary value of $r$ when the total cache size at a pair of UE and EN is sufficient to store the whole library, i.e., $\mu_T+\mu_R \geq 1$. The results indicate that the fronthaul capacity determines which scheme achieves a better performance in terms of the NDT, and the soft-transfer scheme becomes favorable as the fronthaul capacity increases.