Multi-agent reinforcement learning based dynamic self-coordinated topology optimization for wireless mesh networks

Wireless mesh network (WMN) technology enhances wireless communication coverage and increases end-to-end (E2E) delay. The delay in WMNs is influenced by various mesh topologies, which are shaped by transmission power factors. Due to the dynamic nature of WMNs, conventional offline topology optimization methods are ineffective. This paper presents a reinforcement learning (RL)-based dynamic collaborative optimization method to minimize E2E delay and power consumption in WMNs. First, we develop a numerical model that simulates real-world communication environments. Next, we design a baseline and a reward function tailored to this environment. We propose a novel dynamic self-coordinated topology optimization algorithm to address the challenges associated with high-dimensional state–action spaces and the need for coordinated actions among multiple terminal devices. By leveraging coordination signals from terminal agents and multiple optimization strategies, the algorithm enables automatic agent coordination. Experimental results and real-world simulations demonstrate that the proposed algorithm effectively reduces both E2E delay and device power consumption.