Federated Learning in SWIPT-Enabled Micro-UAV Swarm Networks: A Joint Design of Scheduling and Resource Allocation

Abstract

In this paper, we consider a micro unmanned aerial vehicle (UAV) swarm network enabled by simultaneous wireless information and power transfer (SWIPT). The network consists of one base station (BS) and multiple micro UAVs, where the BS aims to train a machine learning model over the data that resides on each UAV by utilizing federated learning (FL). Particularly, at each iteration of FL, the BS broadcasts both the global model and energy simultaneously to all UAVs, and each UAV relies on its harvested and battery-stored energy to train the received model and then upload it to the BS for global model aggregation. To improve the learning performance, we formulate a problem of maximizing the percentage of scheduled UAVs at each iteration of FL by jointly optimizing UAV scheduling and sub-channel assignment, as well as broadcasting/uploading time and power allocation. The formulated problem is a challenging mixed integer nonlinear programming problem and is NP-hard in general. By exploiting the special property of the problem structure, we develop a low-complexity suboptimal algorithm to efficiently solve the formulated problem. Numerical results show the advantages of our suboptimal algorithm.