MMSE Threshold-based Power Control for Wireless Federated Learning

We put forward a novel minimum mean square error (MMSE) threshold-based power control scheme for wireless federated learning in digital communication systems. The proposed approach uses pulse-amplitude modulation to attain digital over-the-air aggregation of local machine learning models. To reduce the communication cost, we design a novel threshold-based power control strategy that minimizes the mean squared error of parameter estimation and satisfies a constraint on the average power. Simulation results show that the proposed approach is superior to one-bit broadband digital aggregation (OBDA) in terms of the testing accuracy of machine learning and the power consumption of wireless communications. Furthermore, in comparison with broadband analog aggregation (BAA), the proposed approach reduces the power consumption of wireless communications without sacrificing the testing accuracy.