Improving Energy Efficiency in Federated Learning Through the Optimization of Communication Resources Scheduling of Wireless IoT Networks

Federated Learning (FL) has emerged as a promising paradigm for enabling distributed intelligence while preserving data privacy. Despite its advantages, deploying FL in practical environments remains challenging due to limitations related to communication efficiency, resource constraints, and data heterogeneity across participating devices. These challenges often lead to degraded model performance, unstable convergence, and inefficient resource utilization. To overcome these limitations, this paper proposes FL-E2WS, an energy-efficient FL framework that jointly optimizes device selection, uplink resource allocation, an aggregation stage based on the quality of local device data. The proposed approach integrates statistical and communication awareness by prioritizing devices with representative data distributions and favorable channel conditions. In addition, a mixed-integer linear programming formulation is employed to coordinate power and bandwidth allocation, while a data-quality-aware aggregation mechanism further enhances convergence. Simulation results show that FL-E2WS achieves significant improvements over baseline methods, with accuracy differences ranging from 4.73% to 20.09% and energy efficiency gains between 36.25% and 45.48%. These results highlight the effectiveness of the proposed framework in enhancing both learning performance and energy efficiency.