Estimating AP Location Using Crowdsourced Wi-Fi Fingerprints with Inaccurate Location Labels

Abstract

The fundamental property of Wi-Fi signals for Wi-Fi positioning is that the signal strength of radio signals decreases as radio signals propagate. Since access points (APs) are sources of each Wi-Fi signal, knowing AP locations can help to exploit Wi-Fi fingerprints. However, investigating AP locations is a demanding task. Measuring absolute coordinates of AP locations is necessary for the extensive investigation, but manual measurement of absolute coordinates costs excessive human labor. In this paper, we propose an AP location estimating method using Wi-Fi fingerprints labeled with coordinates obtained from commercial location providers on smartphones. Both Wi-Fi fingerprints and location labels are implicitly collected from smartphone users. We apply nonlinear regression with log-distance path loss (LDPL) model on collected location-labeled Wi-Fi fingerprints to estimate AP locations. Estimated AP locations can accelerate radio map matching, simplify radio map construction, and assist AP location-based Wi-Fi positioning system.