Optimal action-selection optimization of wireless networks based on virtual representation

Modern smart homes are becoming increasingly complex due to the growing number of connected wireless devices and the need for a fast, stable Internet connection. Tri-band Wi-Fi 6E addresses these demands by reducing network congestion and enhancing performance across the 2.4, 5 and 6 GHz bands. However, improper channel management can lead to frequent interference and network instability. This work introduces a Wi-Fi management solution that uses reinforcement learning to automatically optimize network configurations based on the observed interference levels. The Wi-Fi network is mapped to a virtual environment to simulate different scenarios with real data acquired from the network, allowing the system to learn and improve without disrupting the actual network. Practical tests were conducted in four residential network layouts, including star and daisy chain topologies. The solution effectively reduces the number of hops, improves the throughput in longer paths from the gateway to clients, adapts to unforeseen network dynamics in the presence of Ethernet clients and adjusts the configuration as clients move within the home. The results demonstrate improved network capacity and responsiveness compared to default configurations. In the best cases, clients achieved rates of up to 1 Gbps, round-trip times as low as 10 ms, and efficient airtime distribution across the three frequency bands. The proposed approach provides a robust and automated solution to address the challenges of domestic wireless networking and paves the way to more sophisticated virtualization solutions for end users at home.