Refined characterization of RSSI with practical implications for indoor positioning

This paper regards the exploitation of RSS in localization techniques within UWB networks. Both fingerprinting and model based approaches are studied and evaluated using a real UWB measurement campaign. As for fingerprinting approach, SVM, KNN, and ANN techniques are proposed and compared. As for model based approach, refined RSS models are proposed in order to better characterize the RSSI-distance relation. We propose to use a different model for each reference station (AP, BS, Femtocell, etc.) instead of using a general path loss model for the whole scene. Those two proposed models are evaluated on the UWB measurement campaign. The obtained results show improvements in both RSSI-based ranging and localization. These improvements may make RSS based localization techniques, which are usually poor and imprecise because of the RSS fluctuations, more accurate and more reliable.