Task-Driven Interference Management for Over-the-Air Computation Assisted UAV Swarms

Unmanned Aerial Vehicle (UAV) swarms, which integrate sensing, communication, and computation, offer a transformative solution for Search and Rescue (SAR) in infrastructure-limited environments. These swarms must accommodate two types of services, task-driven sensing data aggregation and periodic communication. Federated Learning (FL) facilitates distributed data fusion, reducing transmission overhead while protecting privacy. To address spectrum resource scarcity, Over-the-Air Computing (AirComp) is introduced to synchronize computation and communication to improve spectrum utilization. However, the trade-off between aggregation performance and convergence speed restricts UAV participation in aggregation. Ad ditionally, the opposing requirements between reliability-focused communication and interference-reliant AirComp aggregation degrade signal quality. To tackle these challenges, we propose a joint slot scheduling and interference suppression architecture that uses intelligent resource allocation to support parallel communications and aggregation. Our Dual Time-scale Interference Management (DTIM) algorithm optimises UAV selection, slot scheduling, and beamforming to mitigate interference within a given Mean Square Error (MSE). Simulation results show that our algorithm maintains the network Age of Update (AoU) around 10 and controls the FL prediction accuracy to more than 93% of the upper bound under various communication loads.