Improved Localization Accuracy Using Machine Learning: Predicting and Refining RSS Measurements

Abstract

Wireless localization methods are often subject to errors due to radio signal fluctuations that are used to estimate inter-device separation distances. We propose a novel method called MLRefine to counter these effects by refining RSS measurement data to obtain more accurate values that can enhance ranging and localization accuracies. MLRefine uses machine learning methods to model the relationship between accurate values and features extracted from in silico RSS values. MLRefine then applies the trained model to features extracted from real RSS measurement values to return a predicted set of refined RSS values. The refined RSS values are shown through computer simulations and real experiments to improve localization accuracy.