Applying Random Forest and Multipath Fingerprints to Enhance TDOA Localization Systems

This letter proposes a method to apply the random forest (RF) algorithm for exploiting the multipath information and enhance the time difference of arrival (TDOA) localization systems performance. This new approach aims to enhance the performance of a location system in outdoor scenarios, where there are technical restrictions to deploy them in line-of-sight condition. A site-specific fingerprint is extracted with ray-tracing simulations to produce a multipath dataset. The signals are processed with a multidimensional cross-correlation function, called volume cross-correlation (VCC) function, to extract the multipath features from the TDOAs measurements. The machine learning framework searches the points in the scenario, trying to find the most likely to be the emitter position using the processed features and the radio map.