Convergence Analysis and Optimization of SWIPT-Based Over-the-Air Federated Learning
Shaoshuai Fan, Shilin Tao, Wanli Ni, Hui Tian · IEEE Communications Letters · 2024
Federated learning (FL) are challenging for low-end Internet of Things (IoT) devices with limited energy storage. In this letter, to solve this difficulty, Base Station (BS) uses simultaneous wireless information and power transfer (SWIPT) to spread the global model and charge every device during each FL round. The convergence gap of SWIPT-based FL is derived to capture the effect of wireless communications on the learning performance. To speed FL convergence, a non-convex problem is formulated by jointly optimizing the transceiver beamforming and power-splitting ratio. Then, an alternating optimization algorithm is designed to obtain a sub-optimal solution. Simulation results show that our proposed scheme outperforms benchmarks in terms of prediction accuracy and convergence.