Throughput prediction in wireless networks using statistical learning

Abstract

The focus of this work is on the estimation of throughput in wireless networks, more specificaly on IEEE 802.11. Our proposal is based on active measurements and statistical learning tools. We present a methodology where the system is trained during short periods with application flows and probe packets bursts. We learn the relation between throughput obtained by the application and the state of the network, which is inferred from the interarrival times of the probe packets bursts. As a result we obtain a continuous non intrusive methodology that allows to determine the maximum throughput of a wireless connection only knowing some characteristics of the network. We use Support Vector Machines (SVM) for regression and we show results obtained by simulations.