Power allocation and communication resource scheduling for federated learning in wireless IoT networks

The article addresses the critical challenges of power allocation and communication resource scheduling in federated learning (FL) within wireless Internet of Things (IoT) networks. It highlights the limitations of centralized cloud-based approaches due to data privacy concerns and the inefficiencies of direct data collection from IoT devices. The proposed solution, DFed-w (Data Federated Learning Wireless Optimizer with Dynamic Power Allocation), is an advanced FL algorithm designed to optimize device selection and communication resource scheduling. DFed-w employs the Earth Mover’s Distance (EMD) metric to ensure data diversity and connection quality, dynamically allocating uplink resource blocks (RBs) to minimize energy consumption while maintaining efficient communication. Simulation results demonstrate that DFed-w achieves a higher number of successful model transmissions with lower transmission power, maintaining competitive global model accuracy. The article also provides a detailed comparison with existing algorithms, showcasing the superior performance of DFed-w in terms of energy efficiency and transmission success. Additionally, it discusses the mathematical models used for network simulation, ML model training, energy consumption, and communication, offering a comprehensive framework for implementing FL in wireless IoT networks.