Efficient Training Phase for Ultrawideband-Based Location Fingerprinting Systems

Abstract

Location fingerprinting utilizing ultrawideband (UWB) radio frequency (RF) signals is an attractive alternative to conventional positioning concepts based on range, angle, or received signal strength estimates. Such a location fingerprinting method proves particularly beneficial in indoor environments with dense multipath propagation and nonline-of-sight situations where conventional approaches would fail. The ultrawide bandwidth allows for location fingerprints with many degrees of freedom and thus gives the important advantage that a single anchor suffices for good localization performance. The downside is that a large amount of training data is usually required, which makes the training phase time-consuming and tedious. In this paper, we propose and study a novel and efficient training method which is based on the idea of spatial signal prediction. We develop a regional channel model which supports spatial signal prediction in the vicinity of a reference point. We show that the parameters of this regional channel model can be estimated from very few measured received signals with known transmitter and receiver positions. The prediction accuracy and the location fingerprinting performance are evaluated with measured channel impulse responses obtained in an anechoic chamber and in a typical office environment.