Link quality estimation in wireless multi-hop networks using Kernel based methods

A new approach for link modelling in wireless multi-hop networks is described for portable devices, based on Kernel Regression Statistics. A non-parametric estimation of errors in the wireless medium provides an efficient and accurate model of link errors between any two nodes. This estimation results from the analysis of the inter-arrival time between any periodically sent packets. The obtained results prove that it is possible to infer on link quality without having unrealistic assumptions or additional overhead, by using Kernel Methods. Moreover, similar performances were achieved for different scenarios, without requiring model recalculations. The presented results show that the proposed link quality estimation can be used in order to improve wireless connectivity and ubiquity in future networks.