A novel similarity measure for WLAN fingerprints using signal fluctuation matrix

Abstract

WLAN fingerprint-based localization is most spotlighted technique for location-based services. However, the WLAN fingerprint-based localization technique cannot provide precise positioning information without considerable calibration efforts of collecting fingerprints at a target area. Since the indoor space usually has walls and obstacles, and some APs installed in the space may share the same channel, the fluctuation and missing of WLAN signals frequently happen in collecting the WLAN fingerprints. Refinement of fingerprints by repeated measurement of fingerprints at the same measure point is the most general approach to address this problem. However, it results in high calibration cost. Many trials have been made to reduce the calibration efforts, and war-walking is known as the most spotlighted techniques to date. However, since it is hard to collect measurements collected at the same point without considerable calibration efforts, a necessity of a new similarity measure for WLAN fingerprints that can produce a good accuracy even with incomplete WLAN approach is being magnified. In this paper, we introduce a Signal Fluctuation Matrix (SFM) for the new similarity measure. The SFM is a kind of the substitution matrix used for sequence alignment of amino acids. The SFM is constructed with the observed and expected frequency of signal fluctuation from offline measurements. Then the similarity between fingerprints are measured based on the SFM. SFM can impose appropriate weight according to RSS, can provide plausible values for missing signals, and reflects LDPL model. Thus, accurate similarity can be computed can be computed more accurately with SFM regardless of incomplete fingerprint involvement. For the evaluation of the method, we compared the accuracies of the proposed method with existing similarity measures in three different environments, E-mart, KAIST Library and N1 building. And the proposed method demonstrated improvement of classification and posit...