Utilizing Deep Reinforcement Learning for Optimal UAV 3D Placement in NOMA-UAV Networks

Unmanned Aerial Vehicles (UAVs) have been extensively studied in recent years, for their role in providing wireless communication in future networks, alongside Non-Orthogonal Multiple Access (NOMA). In this context, UAVs need to dynamically and periodically change their placement due to the nature of high-speed wireless networks. Regarding this, Deep Reinforcement Learning (DRL) algorithms are implemented to optimize 3D UAV placement. Adopted are a set of DRL algorithms, including Deep Q-network (DQN), Advantage Actor-Critic (A2C), and Proximal Policy Optimization (PPO). Furthermore, a comparative study is conducted among these algorithms and fixed UAV placement. From the simulation results, PPO shows the best performance and the most stable learning compared to other approaches.