Multi-Agent Reinforcement Learning for User-Router Assignment in Multi-Radio Multi-Channel Wireless Mesh Networks
International journal of intelligent engineering and systems · 2025
This paper addresses the challenging problem of user-router assignment in Multi-Radio Multi-Channel Wireless Mesh Networks (MRMC WMNs) using Multi-Agent Reinforcement Learning (MARL).Traditional assignment methods often fail to adapt to the dynamic nature of wireless environments, leading to suboptimal network performance.We propose three different MARL approaches: Epsilon-Greedy (EG), Greedy-Epsilon-Greedy (G-EG), and Greedy-Epsilon-Greedy-Q-Learning (G-EG-QL).We valuated our approach through simulations with synthetic mobility datasets in equally and non-equally distributed user environments.Our formulation includes detailed state representation based on user-router distance, user density, and buffer occupancy, along with a reward mechanism that accounts for Packet Delivery Ratio (PDR) and buffer load.We evaluated our approach Experimental evaluation conducted in both equally distributed and non-equally distributed user scenarios demonstrates that our RL-based methods outperform traditional greedy approaches, particularly in equally distributed environments.For the G-EG-QL approach, we observed improvements of 7.5% in PDR, 7.4% in throughput, and 11.2% lower end-to-end delay compared to greedy methods.The results provide strong evidence that MARL approaches are more suitable for dynamic user assignment in MRMC WMNs, offering better adaptability and performance optimization.