5G Unmanned Aerial Vehicle Placement for Mountainous Environment using Deep Reinforcement Learning

Unmanned Aerial Vehicle-mounted Base Stations (UAV BSs) have emerged as a flexible and cost-effective solution for delivering wireless coverage in geographically challenging regions such as mountainous terrains, where existing terrestrial networks frequently fall short. UAV BSs possess essential mobility that enables adaptive repositioning in three-dimensional (3D) space to meet dynamic user demands; however, optimizing their 3D placement is an NP-hard problem, especially in environments with complex topography and obstruction induced shadowing. This study presents a reinforcement learning (RL)-based optimization framework employing Q-Learning (QL) and Deep Q-Network (DQN) agents, guided by a received signal strength (RSS)-based reward function, to optimize UAV BS deployment in mountainous environments with mobile user equipment (UE). A high-resolution ray tracing (RT) model is used to simulate realistic propagation conditions, capturing both Line-of-Sight (LOS) and Non-Line-of-Sight (NLOS) signal variations due to terrain, which are compared against the conventional Air-to-Ground (A2G) model to evaluate performance differentials. The framework allows dynamic adjustment of UAV altitude and positioning in real time to maximize coverage and signal quality. The study emphasizes the superiority of the RT-based RSS modeling in representing complex propagation phenomena, leading to improved convergence and deployment efficiency in both QL and DQN agents. Simulation results demonstrate that UAV placement optimized using the RT-based model achieves significantly better coverage and RSS performance compared to traditional methods. This work introduces a robust and generalizable deep reinforcement learning (DRL)-driven approach for UAV network deployment in extreme environments, offering critical insights for the design of next-generation, terrain-aware aerial communication systems.