The effect of coverage gaps and measurement inaccuracies in fingerprinting based indoor localization

In this paper we estimate the effect of coverage gaps and inaccurate Received Signal Strength (RSS) values in fingerprinting based indoor localization using Wireless Local Area Networks (WLAN). The results are based on extensive measurement campaign including two multi-storey buildings with over 700 found WLAN access points in total. We introduce a novel randomized method to artificially create realistic coverage gaps in the database. It is further emphasized that a realistic fingerprint removal process for modeling coverage gaps cannot be based in uniformly distributed probability density function. User positioning performance between the original database and the partial database is compared using the well-known K-Nearest Neighbor (KNN) algorithm. In addition, we model RSS inaccuracies in the database originated from badly calibrated learning data or from a constant bias between learning data collection devices and the device used for positioning. The effect of coverage gaps and RSS inaccuracies on the user positioning accuracy is studied in terms of average horizontal positioning error and in average floor detection probability over several user tracks and randomized removal processes. The presented results and the provided methodology allow error dimensioning of collected learning data and assist in planning measurement campaigns in future indoor positioning studies.