Assessing Learning-Based Resource Allocation Scheme for UAV-Assisted OTFS-NOMA System

Orthogonal time frequency space and non-orthogonal multiple access (OTFS-NOMA) protocol have increasingly become well-suited candidates for dynamic vehicle-to-everything (V2X) communications. However, the stochastic characteristics cannot be fully addressed for existing UAV-assisted communication systems. To address this problem, we propose a deep reinforcement learning method, known as Twin Delayed Deep Deterministic Policy Gradient (TD3), for allocating communication resources in unmanned aerial vehicle (UAV)-Assisted OTFS-NOMA System. In particular, we implement the gate recurrent units (GRUs) to learn the deterministic optimal time-varying actions with imperfect channel state information (CSI). Simulation results demonstrate the proposed TD3-based approach achieves a substantial reduction in total transmission delay and communication latency, while maintaining effective convergence under dynamic channel conditions. These results validate the practicality of TD3 in enhancing reliability and resource utilization for UAV-assisted OTFS NOMA networks.