Federated learning for optimized resource allocation in power line communication systems

Abstract

This paper introduces the Priority-Aware Federated Resource Allocation (PAFRA) algorithm, a novel resource allocation strategy designed for power line communication (PLC) systems. Leveraging a federated learning framework, PAFRA optimizes the distribution of limited spectrum resources across multiple nodes within a residential environment. The algorithm integrates a priority-based time slot allocation to effectively manage subchannel conflicts and employs a Double Deep Q-Network for local training at each node. This training modality includes comprehensive state, action, and reward configurations to fine-tune transmission power and subchannel selections. Extensive simulations reveal that PAFRA substantially enhances system throughput, outperforming both traditional numerical optimization methods and contemporary machine learning approaches. Specifically, at a signal-to-noise ratio of 30 dB, PAFRA achieves a system throughput of approximately 950 Mbps, marking a 23% improvement over existing approaches. These findings highlight PAFRA's ability to significantly enhance network efficiency while complying with regulatory emission standards, demonstrating its potential to optimize dynamic PLC systems.