Low-complexity Architecture Achieves Multi-access Coded Caching with Arbitrary User Access

Multi-access coded caching (MACC) systems represent a promising approach to meeting the increasing demands on network delivery, yet current models often assume simplified and unrealistic network topologies. Ting Yang, Minquan Cheng, and Robert Caiming Qiu, from Huazhong University of Science and Technology, alongside Xinping Yi and Giuseppe Caire from Technische Universität Berlin, address this limitation by investigating MACC under arbitrary user-cache access topologies. Their research introduces a novel framework that models the delivery problem using a coloring approach, linking transmission load to the number of colours required. This work is significant because it moves beyond restrictive network designs, offering a low-complexity solution applicable to diverse and large-scale caching scenarios, and demonstrating performance comparable to existing, computationally expensive methods. The team’s learning-based framework efficiently constructs coded multicast transmissions, paving the way for more practical and scalable MACC implementations.

Graph Coloring Optimises Coded Caching Access

Scientists demonstrate a significant advancement in multi-access coded caching (MACC) systems, achieving a low-complexity architecture capable of operating with arbitrary user-cache access topologies. This research extends existing MACC models which traditionally rely on highly structured network configurations, offering a more versatile solution for real-world deployments. The team achieved this breakthrough by formulating the MACC delivery problem as a graph-based framework, effectively modelling decoding conflicts as a ‘conflict graph’ and reducing the design process to a graph coloring problem. Lower transmission loads are directly correlated with using fewer colours in this graph representation, providing a clear optimisation target.

The study unveils a universal approach to MACC, enabling efficient delivery schemes regardless of the underlying network’s access topology and fixed cache placement. Researchers propose a novel framework where the complexity of delivery design is reduced to a graph coloring problem, with the number of colours directly influencing transmission load. While the established DSatur algorithm effectively solves this coloring problem and provides a benchmark for performance, its computational demands become impractical for large-scale networks. To address this limitation, the work introduces a learning-based framework utilising graph neural networks (GNNs) to efficiently construct near-optimal coded multicast transmissions.

This innovative GNN-based framework generalises across diverse access topologies and varying numbers of user nodes, significantly reducing computational time compared to traditional methods. Furthermore, the research extends the index-coding converse bound, a theoretical limit on achievable performance, to MACC systems with arbitrary access topologies, and proposes a low-complexity greedy approximation that closely matches this bound. Numerical results confirm that the proposed scheme achieves transmission loads comparable to both DSatur and the index-coding converse, while dramatically decreasing processing time. Experiments show the proposed learning-based scheme delivers transmission loads on par with the DSatur algorithm and the index-coding converse bound, but with a substantial reduction in computational demands.

This makes the system particularly well-suited for large-scale MACC deployments where traditional algorithms become computationally prohibitive. The research establishes a pathway towards more flexible and efficient caching solutions for future wireless networks, paving the way for improved performance and scalability in diverse network environments. The work opens new possibilities for optimising content delivery in complex network scenarios, moving beyond the limitations of rigid, pre-defined topologies. By leveraging the power of graph neural networks, scientists prove the feasibility of a low-complexity architecture that adapts to arbitrary user-cache access, promising significant gains in network efficiency and user experience. This advancement is poised to impact a range of applications, from edge computing and content distribution networks to future 6G wireless systems.

Graph Colouring for Coded Caching Delivery

The study addresses the multi-access coded caching (MACC) problem, moving beyond traditional models that depend on highly structured network connections. Researchers engineered a system comprising a single server, multiple cache nodes, and multiple user nodes, where each user accesses a variable subset of cache nodes to obtain content. The core innovation lies in a universal graph-based framework for modelling MACC delivery, translating decoding conflicts between requested packets into a conflict graph, effectively reducing the delivery design to a graph coloring problem. A lower transmission load is achieved by minimizing the number of colours used in this graph representation.

To establish a performance benchmark, scientists initially employed the DSatur greedy coloring algorithm, demonstrating transmission loads approaching the index-coding converse bound. However, recognizing the prohibitive computational cost of DSatur for large-scale networks, the team developed a learning-based framework utilizing graph neural networks (GNNs). This approach efficiently constructs near-optimal coded multicast transmissions, generalizing effectively across diverse access topologies and varying user numbers. The GNN operates directly on graph instances, enabling a single network to adapt to heterogeneous topologies and user demands without requiring environment-specific retraining, a significant advancement over reinforcement learning methods.

Furthermore, the research extended the index-coding converse bound to accommodate uncoded cache placement within arbitrary access topologies, alongside a low-complexity greedy approximation technique. Experiments demonstrate the learning-based scheme achieves transmission loads comparable to both DSatur and the index-coding converse bound, while drastically reducing computational time. This methodology enables practical implementation of coded caching in complex, real-world network scenarios previously limited by computational constraints and topological irregularities. The work pioneers a scalable solution for MACC delivery, offering a robust framework for future advancements in content distribution networks.