Joint Optimization of Charging Time and Resource Allocation in Wireless Power Transfer Aided Federated Learning

As a promising methodology of distributed machine learning (ML) paradigm, federated learning (FL) protects data privacy and reduces communication cost by aggregating model parameters rather than raw data. However, training superb FL models incurs a lot of energy consumption, which is a significant challenge for energy-limited mobile devices (MDs). To address this challenge, this article proposes a wireless power transfer (WPT)-aided FL framework, where MDs train local FL models for base station (BS) and get corresponding payoff, while wireless charge provider (WCP) provides energy supplement for MDs and charges energy fees. Furthermore, we take into account the time-varying nature of MDs’ datasets, which affects their energy consumption and reward from BS. Then, we formulate the investigated problem to achieve joint optimization of WPT duration, computing resource allocation and the number of local iterations, with the goal of maximizing the total utility of all MDs throughout the whole FL process. The optimization problem is NP-hard and difficult to be solved by traditional optimization methods within limited timeframes. Therefore, we use Karush-Kuhn–Tucker (KKT) conditions and Lagrange dual method to analyze the problem, and propose a new improved lagrangian subgradient method (ILSM) as an efficient solution. Finally, extensive simulation experiments are conducted to demonstrate the effectiveness of the proposed scheme under various scenarios, and the results show that the proposed ILSM significantly outperforms other benchmarks in terms of the total utility of all MDs.