Deep Reinforcement Learning for UAV Trajectory Design Considering Mobile Ground Users.

Area of Science:

• Wireless communication networks
• Artificial intelligence in telecommunications
• Robotics and autonomous systems

Background:

• Next-generation communication systems increasingly utilize unmanned aerial vehicles (UAVs) as mobile base stations (UAV-BS).
• Optimal UAV-BS positioning is crucial for maintaining line-of-sight (LoS) links with ground users.
• Dynamic user movement necessitates adaptive UAV-BS deployment strategies.

Purpose of the Study:

• To propose a novel deep Q-network (DQN)-based learning model for optimal UAV-BS deployment.
• To enable dynamic UAV-BS trajectory optimization for moving ground users without model re-learning.
• To maximize the mean opinion score (MOS) for ground users by optimizing UAV-BS movement.

Main Methods:

• Development of a deep Q-network (DQN) model for UAV-BS trajectory optimization.
• Utilization of average channel power gain as a practical input parameter, avoiding individual user location tracking.
• Validation of the proposed model against a mathematical optimization solver.

Main Results:

• The proposed DQN model successfully determines optimal UAV-BS trajectories for moving users.
• The model achieves high practicality by using average channel power gain as input.
• The model's accuracy was validated through comparison with established mathematical optimization techniques.

Conclusions:

• The novel DQN-based model offers an efficient and practical solution for dynamic UAV-BS deployment.
• This approach enhances communication quality for mobile users in next-generation networks.
• The method provides a robust framework for optimizing UAV-BS networks in real-world scenarios.