Wi-Fi Throughput Estimation for Vehicle-to-Network Communication in Heterogeneous Wireless Environments

Abstract

Vehicles increasingly need to be connected to networking infrastructure, to support applications such as over-the-air updates, edge computing, and even autonomous driving. The ubiquity of Wi-Fi networks makes them a good candidate for opportunistic vehicular access. However, that ubiquity also creates a problem of choice. In a heterogeneous Wi-Fi environment, with multiple different networks available, it becomes important for vehicles to be able to pick the best-performing one. Focusing on delay-insensitive traffic, we equate network performance with throughput, and aim to estimate it to inform network selection. Throughput estimation is traditionally done by injecting probe traffic, which induces congestion. We provide a solution that avoids this by using only passive measurements of variables such as signal strength to estimate throughput. Taking real-world training data collected in a diverse vehicular Wi-Fi communication scenario, with IEEE 802.11n, ac, and ad networks, we used Symbolic Regression (SR) and Unscented Kalman Filter (UKF) to develop a computationally inexpensive throughput estimation model, UKF-SR. Using a separate testing dataset, we compared the proposed UKF-SR model against traditional linear and support-vector regression, decision tree, random forest, and shallow neural network models. UKF-SR was competitive with even the most complex models. It yielded the lowest Root-Mean-Square Errors (RMSE) for 802.11n and ac, by 4.71% and 27.59 %, respectively, and was within 1% of the best-performing model for 802.11ad.