Energy Efficient Power Control for Over-the-Air Federated Learning in Satellite Communications

We study to improve the energy efficiency for federated learning with the over-the-air aggregation mechanism in low- Earth orbit satellite backbones, which is a crucial issue given the limited onboard energy and communication resources of satellites. We first derive a concise relationship between the optimal satellite transmission power and the total energy consumption, and then propose an adaptive algorithm to obtain the optimal transmission power in a low-complexity manner. We numerically demonstrate that our approach consistently achieves higher energy efficiency than existing benchmarks with highly non-iid data distribution among satellites.