UAV-Enabled Over-the-Air Federated Learning: A Hierarchical Aggregation Approach

作者：Xiangyu Zhong, Chenxi Zhong, Xiaojun Yuan, Ying–Jun Angela Zhang · 发表于：IEEE Transactions on Wireless Communications · 年份：2025 · DOI：10.1109/twc.2025.3635287 · 被引用次数：1 · 研究领域：Privacy-Preserving Technologies in Data、UAV Applications and Optimization、IoT and Edge/Fog Computing

With explosive increase of data at the mobile edge, federated learning (FL) emerges as a promising technique to reduce data transmission costs and privacy leakage risks. Nevertheless, the huge communication overhead for an increasing volume of edge devices still restricts the FL performance. Over-the-air computation (AirComp) is viable for alleviating the communication burden in FL systems. However, there consequently appears a straggler issue restraining the performance of the over-the-air FL (OA-FL) framework, which is even worse especially when devices training a machine learning model are distributed over a relatively large service area. In this paper, we propose an unmanned aerial vehicle (UAV) enabled OA-FL scheme, where the UAV acts as a parameter server (PS) to aggregate the local gradients hierarchically for global model updating. The global aggregation frequency is tunable in the hierarchical aggregation approach, enabling it to balance the resource consumption between communication and learning. Building on this approach, we carry out a gradient-correlation-aware FL performance analysis and jointly optimize the trajectory of UAV-PS, the device selection state, and the aggregation coefficients. An algorithm based on alternating optimization (AO) is developed to solve the formulated problem, where successive convex approximation (SCA) and fractional programming (FP) are utilized for the convexification of the non-convex problem. Numerical simulation results demonstra...