Abstract

With the growing adaptation of Wi-Fi and the increased possibilities of complementing it with 5G, there is a need to exploit the fullest potential of the IEEE 802.11ac/ax and higher Wireless Local Area Network (WLAN) standards, especially in densely deployed scenarios. Machine learning techniques can be used to predict the performance of WLANs with the vast training data obtained through contemporary network simulators. They are quite useful to predict the throughput in crowded and dynamic deployments of WLANs where hand-crafted solutions may not be feasible. This paper presents a novel, data-driven approach that can contribute to improving the performance of next-generation WLANs. In particular, we employ a Graph Neural Network (GNN) model to predict the performance of Wi-Fi deployments by exploiting topology information and capturing complex wireless interactions. The network simulator, Komondor, is used to simulate different real-life scenarios for generating comprehensive datasets for training the model. Our approach addresses challenges related to energy efficiency, latency, and data rate in WLANs, and the regression model can be used to predict the throughput of a Basic Service Set (BSS) before it is deployed, allowing for better network design and optimization. The findings of this study demonstrated that GNNs can accurately forecast the throughput of BSSs in WLAN deployments in a given region with minimal information. Overall, our proposed approach can significantly influence the choice of topology for deployment, leading to optimal performance in crowded and dynamic WLAN scenarios.