Throughput Analysis of Dense-Deployed WLANs Using Machine Learning

IEEE 802.11(Wi-Fi family of standards) have been widely adapted for internet access due to the ease of deployment, freedom of configuration and interoperability of the devices. However, in dense deployments of such Wireless Local Area Networks (WLANs) as in overlapped Basic Service Sets (OBSSs) and with the options to enhance the channel bandwidths to 80/160 MHz (for IEEE 802.11ac/ax), due to increased contention the performance suffers. This is manifested as reduction in the overall throughput much lower than that can be ideally achieved by the network. Thus, if the throughput of a network prior to the deployment is roughly known, the real-life deployment could be optimally chosen by reworking the deployment parameters. Machine Learning (ML) is a promising approach to capture the complex interactions among the AP/STA entities, learn the complex inter-relationships and be able to predict the performance parameter i.e. throughput of a given deployment with minimal or partial inputs. To do so, a large dataset of various scenarios of deployment is necessary, which is obtained through simulation using the network simulator Komondor. The ML model using the Graph Neural Network (GNN) captures the deployment details and is able to converge to the throughput values being predicted closely. By tuning the hyperparameters and varying the configuration parameters the ML model is refined to produce significant improvement in the accuracy of prediction of throughput. The results can be used to build and analyze the performance landscape of densedeployed WLANs for arbitrary or systematic deployments for discovery of optimal deployment.