Testing for normality of UWB-based distance measurements by the Anderson-Darling statistic

摘要

The Ultra-Wideband measurement campaign promoted within NEW-COM++ is intended to provide real-world data to researchers. The set of measurements are composed of a number of distance estimates from several sensor nodes to anchor nodes with known positions. These estimates were based on Time-Of-Arrival estimates delivered by Timedomain PulseON 220 UWB nodes. The measurement campaign resulted in a database of measured distances along with an accurate measure of the actual locations of nodes and anchor nodes. Therefore, the experiment constitutes an appealing database for testing positioning algorithms in realistic conditions. One of the typical assumptions claimed by many of such algorithms is regarding Gaussianity of noise, and thus measurements. This paper studies the Gaussianity of such measurements by using the Anderson-Darling test for normality, which is known to be the most powerful tool to that aim. The test shows that even though some pairs of range measurements could be accepted to be Gaussian, the majority is not.