A Ring Topology-Based Communication-Efficient Scheme for D2D Wireless Federated Learning

Federated learning (FL) over device-to-device (D2D) wireless networks faces significant communication bottlenecks due to the decentralized and resource-constrained nature of edge devices. In this paper, we propose a ring topology-based communication-efficient scheme for D2D wireless federated learning. By organizing participating devices into a ring structure, each device only communicates with its two neighbors, drastically reducing per-round communication overhead. We design a gradient compression and aggregation protocol tailored to the ring topology, and provide theoretical convergence guarantees under non-IID data distributions. Extensive simulations demonstrate that the proposed scheme achieves comparable model accuracy to centralized FL while significantly reducing total communication cost, making it well-suited for bandwidth-limited D2D environments.