A mobility-assisted protocol for supervised learning of link quality estimates in wireless networks

In this paper we propose MAPPLE, a novel method to learn link quality estimates in wireless networks. The method is a two-step process that combines a online distributed protocol, for gathering link quality measurements, with a supervised learning approach, for offline data processing and model building. The distributed protocol exploits channel probing and node mobility, while the offline learning is based on Support Vector Regression (SVR). The core idea is to use the online protocol to dynamically reshape a given network to generate a large number of different network configurations from which we sample, through the transmission of probe packets, link quality measures. Each measure is associated to a vector of network features, related to interference, traffic loads, and local topology, that jointly contribute to the definition of the observed link quality. Quality measures and network features are used to train the SVR model for link quality prediction. We validate our approach by extensive simulation tests, showing the good link quality prediction accuracy of the system, as well as its ability to generalize to networks much larger than the ones used to gather the training data.