Decentralized resource allocation in UAV communication networks through reward based multi agent learning.

Area of Science:

• Wireless Communication Systems
• Artificial Intelligence in Networking
• Resource Management

Background:

• Unmanned aerial vehicles (UAVs) offer flexible aerial base stations (ABS) for on-demand wireless access.
• Dynamic resource allocation is crucial for maximizing performance in multi-UAV systems.
• Existing methods may require extensive information exchange, increasing system overhead.

Purpose of the Study:

• To investigate dynamic resource allocation strategies for multi-UAV-enabled communication systems.
• To maximize the long-term rewards and overall system performance.
• To develop a decentralized learning framework for UAV resource management.

Main Methods:

• Modeling the resource allocation problem as a stochastic game with UAVs as learning agents.
• Developing a reward-based multi-agent learning (RMAL) framework.
• Implementing an agent-independent strategy using a Q-learning-based framework with local observations.

Main Results:

• The proposed RMAL framework achieves effective resource allocation without requiring full information exchange between UAVs.
• Simulation results indicate that RMAL performance is sensitive to parameter tuning but offers a good trade-off.
• The method provides acceptable performance compared to systems with complete information sharing.

Conclusions:

• The RMAL framework offers an efficient approach to dynamic resource allocation in multi-UAV systems.
• Decentralized learning with local observations can achieve near-optimal performance while reducing communication overhead.
• This research contributes to the advancement of intelligent resource management in aerial communication networks.