Optimal UAV-Trajectory Design in a Dynamic Environment Using NOMA and Deep Reinforcement Learning

Effective deployment of cellular-connected UAV networks necessitates efficient techniques to minimize mutual interference between UAVs and ground users. Moreover, the existing sub-6 GHz band suffers from extreme congestion, making it challenging to allocate unused resource blocks (RBs) for UAVs. This paper presents a learning-based UAV-path planning approach at the Base Station (BS) side, leveraging Non-Orthogonal Multiple Access (NOMA) and Deep Q-Network (DQN) methodologies to address massive connectivity and air-to-ground interference. The proposed NOMA-DQN learning approach optimizes UAV-transmission power and RB allocation jointly, taking into account the UAV-location. Additionally, it devises an interference-aware path for the UAV, considering its limited battery capacity. Simulation results demonstrate the efficacy of our proposed approach in terms of maximizing the total sum rate of aerial and ground users in a shared RB, as well as enhancing UAV energy efficiency, as compared to shortest path, orthogonal multiple-access (OMA), and random selection schemes.