Collaborative Coded Caching for Partially Connected Networks

TL;DR

The paper addresses scalable content delivery in partially connected networks with distributed MIMO and coded caching. It introduces a two-phase delivery framework—partitioning and transmission—that jointly exploits cache diversity and spatial multiplexing. The partitioning phase uses a least-cost branch-and-bound algorithm to minimize the number of groups per cache profile, while the transmission phase multicast-codes to partitions via MIMO precoding to satisfy all requests. Through simulations, it demonstrates that the proposed scheme approaches the fully connected optimum as connectivity improves and outperforms greedy and collision-model heuristics, showing practical potential for edge caching with reduced subpacketization and improved sum-DoF, quantified by $d_{ ext{sum}} = rac{K(1- heta)}{T}$.

Abstract

Coded caching leverages the differences in user cache memories to achieve gains that scale with the total cache size, alleviating network congestion due to high-quality content requests. Additionally, distributing transmitters over a wide area can mitigate the adverse effects of path loss. In this work, we consider a partially connected network where the channel between distributed transmitters (helpers) and users is modeled as a distributed multiple-input-multiple-output (MIMO) Gaussian broadcast channel. We propose a novel delivery scheme consisting of two phases: partitioning and transmission. In the partitioning phase, users with identical cache profiles are partitioned into the minimum number of sets, such that users within each set can successfully decode their desired message from a joint transmission enabled by MIMO precoding. To optimally partition the users, we employ the branch and bound method. In the transmission phase, each partition is treated as a single entity, and codewords are multicast to partitions with distinct cache profiles. The proposed delivery scheme is applicable to any partially connected network, and while the partitioning is optimal, the overall delivery scheme, including transmission, is heuristic. Interestingly, simulation results show that its performance closely approximates that of the fully connected optimal solution.

Figures (6)

• Figure 1: A partially connected network with $E=2$ transmitters/helpers, $K=7$ users and $L = 3$ cache profiles.
• Figure 2: Example subnetwork where each user has the same cache profile.
• Figure 3: Least cost branch and bound solution of the subnetwork in Figure \ref{['fig:subnetwork']}. Each vertex corresponds to assigning a user to a helper, while an edge corresponds to the number of partitions after such an assignment.
• Figure 4: Hexagonal Placement of the Helpers, $r=1.2$, $r_u=2.7$.
• Figure 5: Sum-DoF for different values of $L$, with $H=4$, $u/L=\frac{4}{(1.2)^2\pi}$, $\gamma=0.1$, $r=1.2$.

Theorems & Definitions (2)

• Example 1
• Example 2