Cluster-Aware Multi-Round Update for Wireless Federated Learning in Heterogeneous Environments

TL;DR

The paper addresses the challenge of data and communication heterogeneity in wireless FL by introducing a cluster-aware multi-round update (CAMU) that treats device clusters as update units. It combines dual-stage clustering (based on SNR and data distribution) with a cluster contribution threshold to control local update frequency, and proves a convergence upper bound with a contraction factor $A$ that guides resource allocation. A PPO-based framework then jointly optimizes cluster-level transmit power and local update frequencies under a total energy budget to minimize the convergence gap, backed by theoretical guarantees (Theorem 1 and Corollary 1) and extensive simulations on MNIST and Fashion-MNIST that show substantial improvements over baselines. The approach yields practical benefits for heterogeneous, resource-constrained wireless FL, offering improved aggregation quality, reduced update bias, and efficient resource utilization in real-world deployments.

Abstract

The aggregation efficiency and accuracy of wireless Federated Learning (FL) are significantly affected by resource constraints, especially in heterogeneous environments where devices exhibit distinct data distributions and communication capabilities. This paper proposes a clustering strategy that leverages prior knowledge similarity to group devices with similar data and communication characteristics, mitigating performance degradation from heterogeneity. On this basis, a novel Cluster- Aware Multi-round Update (CAMU) strategy is proposed, which treats clusters as the basic units and adjusts the local update frequency based on the clustered contribution threshold, effectively reducing update bias and enhancing aggregation accuracy. The theoretical convergence of the CAMU strategy is rigorously validated. Meanwhile, based on the convergence upper bound, the local update frequency and transmission power of each cluster are jointly optimized to achieve an optimal balance between computation and communication resources under constrained conditions, significantly improving the convergence efficiency of FL. Experimental results demonstrate that the proposed method effectively improves the model performance of FL in heterogeneous environments and achieves a better balance between communication cost and computational load under limited resources.

Figures (7)

• Figure 1: The hierarchical FL architecture based on device clustering, where $K$ devices are divided into $C$ clusters. Local updates are performed by transmitting parameters from cluster members to the leaders, and global updates are performed from the leaders to the BS.
• Figure 2:
• Figure 3:
• Figure 5: Distribution of the participating devices.
• Figure 6: FL convergence performance based on each aggregation scheme under different degree of data heterogeneity.