Federated Learning-Driven Trajectory and Resource Optimization in UAV-enabled Integrated Sensing and Communication Systems

Nowadays, in an Unmanned Aerial Vehicle (UAV)- enabled Integrated Sensing and Communication (ISAC) System, a UAV can provide on-demand communication services to ground terminals (GTs) by cooperating with infrastructure networks. Furthermore, with radio frequency (RF) sensing capabilities enabled by the ISAC signal, UAVs can track the locations of GTs and thereby provide better communication services. However, due to the high mobility and SWAP (size, weight, and power) constraints of UAVs, as well as the complex working environment, it is difficult to achieve high-accuracy GT location sensing. As a result, UAVs cannot continuously support highdata-rate communication services for GTs. To address the above issues, we adopt Federated Learning (FL) between UAVs and a central server to improve the accuracy of GT location sensing, and jointly optimize UAV trajectory and resource allocation to ensure the Quality of Service(QoS) of communication services. FL is employed to perform distributed learning for the maximum likelihood estimation (MLE) of GT locations. We design a gametheoretic method to enhance UAV operations under the FL framework. This method ensures that each UAV contributes effectively to location sensing. On the other hand, the formulated joint trajectory and resource optimization problem is complex and subject to multiple non-convex constraints. We apply the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) method to solve the problem and obtain a sub-optimal solution. Numerical results validate that, with our proposed solution, UAVs can deliver communication services with enhanced QoS in terms of higher data rates, while also operating more energy-efficiently in the considered system compared to benchmark solutions.