Throughput and Delay Estimator for 2.4GHz WiFi APs: A Machine Learning-Based Approach

This paper reports our recent result in designing a function for autonomous APs to estimate throughput and delay of its clients in 2.4GHz WiFi channels to support those APs' dynamic channel selection. Our function takes as inputs the traffic volume and strength of signals emitted from nearby interference APs as well as the target AP's traffic volume. By this function, the target AP can estimate throughput and delay of its clients without actually moving to each channel, it is just required to monitor IEEE802.11 MAC frames sent or received by the interference APs. The function is composed of an SVM-based classifier to estimate capacity saturation and a regression function to estimate both throughput and delay in case of saturation in the target channel. The training dataset for the machine learning is created by a highly-precise network simulator. We have conducted over 10,000 simulations to train the model, and evaluated using additional 2,000 simulation results. The result shows that the estimated throughput error is less than 10%.