Robust Trajectory and Resource Allocation for UAV Communications in Uncertain Environments With No-Fly Zone: A Deep Learning Approach

In this paper, we investigate robust trajectory design and resource allocation in unmanned aerial vehicle (UAV) enabled wireless networks to maximize the minimum average spectral efficiency (SE) among mobile nodes (MNs) on the ground while coping with uncertainties in trajectory. Our work specifically addresses practical challenges encountered during trajectory planning, namely: 1) the mobility of MNs, causing changes in their locations over time; 2) the positioning error of UAV, leading to deviations from its planned trajectory; and 3) the presence of no-fly zones (NFZs), which must be avoided during UAV flight. Taking these practical aspects into account, we propose a deep learning (DL) framework that integrates a deep neural network (DNN) structure with an unsupervised learning-based training methodology. The former enables efficient modeling of UAV trajectory and resource allocation, while the latter allows effective training of DNNs without labeled data. Through performance evaluations, we demonstrate that the proposed DL-based scheme outperforms the comparative baseline schemes in terms of the minimum average SE by optimizing trajectory and resource allocation with low computation time. Furthermore, we validate the robustness of our proposed scheme against the uncertainty associated with positioning errors.