SVM-based loss differentiation mechanism in mobile ad hoc networks

In MANET (Mobile Ad hoc NETwork) packet loss at the receiver can be due to wireless random errors, the route change loss induced by node mobility and network congestion. While the traditional TCP-Friendly Rate Control protocol attributes all the packet losses to congestion, and in turn takes congestion control actions to reduce the sending rate, that causes the channel utilization of MANET network extremely low. This paper proposes an end-to-end machine learning algorithm to cognize the loss reasons. Through extracting the statistical characteristics of the lost packet the receiver can train a support vector machine (SVM) model, which can be used to discriminate the three major loss reasons: congestion, route change and wireless error. The unique merit of our solution is that it can cope with this puzzled problem intelligently without subjective or artificial thresholds to been determined. Extensive simulation results demonstrate that the differentiation scheme performs effectively under most scenarios.