A Comparative Survey of WLAN Location Fingerprinting Methods

Abstract—The term “location fingerprinting” covers a wide va-riety of methods for determiningreceiver position usingdatabasesof radio signal strength measurements from different sources.In this work we present a survey of location fingerprintingmethods, including deterministic and probabilistic methods forstatic estimation, as well as filtering methods based on Bayesianfilter and Kalman filter. We present a unified mathematicalformulation of radio map database and location estimation,point out the equivalence of some methods from the literature,and present some new variants. A set of tests in an indoorpositioning scenario using WLAN signal strengths is performedto determine the influence of different calibration and locationmethod parameters. In the tests, the probabilisticmethod with thekernel function approximation of signal strength histograms wasthe best static positioning method. Moreover, all filters improvedthe results significantly over the static methods. I. I NTRODUCTION Location-aware services have become popular with thedevelopment of modern communication technology. The in-creased variety of commercial applications has established thedemand for indoorlocalization services. Weak signal receptionand missing line-of-sight between the user and the satellitescauses Global Positioning System (GPS) to perform poorlyindoors and thus different indoor localization systems havebeen developed.Systems such as the Active Badge, the Cricket, the Batand the Ekahau positioning engine (EPE) rely highly on aninfrastructure that is specially designed for indoor localization[1, 2]. However, these kind of purpose-built systems can beexpensive and thus hard to implement on a world-wide scale.There are also other dedicated indoor positioning systems, seee.g. survey [3].Localization systems can exploit different kinds of mea-surements; systems based on the angle of arrival (AOA), timeof arrival (TOA) and the time difference of arrival (TDOA)have been proposed [4]. However, the reliability of thesemeasurements suffers from the complex signal propagationenvironments [5].The increased deployment and the popularity of wireless lo-cal area networks (WLAN) have opened a new opportunity forlocation-aware services. Although WLAN was not designedfor localization, it can be used for location estimation byexploiting the received signal strength indicator (RSSI) value.RSSI also allows the utilization of the existing infrastructure,because no additional hardware is needed. Signal to noise ratio(SNR) is also available, but it is often omitted because RSSIhas stronger correlation with the location than SNR [5].Location fingerprinting differs from other localization prin-ciples. Instead of determining the distances between the userand the transmitting access points (AP) and triangulating theuser’s location, the location of the user is determined bycomparing the obtained RSSI values to a radio map. The radiomap is constructed in an offline phase and it contains themeasuredRSSI patternsat certainlocations.Thisway thechar-acteristics of the signal propagationin indoorenvironmentsarecaptured and the modeling of the complex signal propagationis avoided. However, the offline phase is quite laborious andthe radio maps have to be stored in memory.Many of the existing location fingerprinting methods lacka proper mathematical formulation and theoretical basis. Thefirst purpose of this work is to present the mathematicalformulation of the location fingerprinting methods covered inthis paper. The second goal is to apply Bayesian and Kalmanfilters to location fingerprinting. The third objective in thiswork is to implement the different algorithms and to test themin varying circumstances.This paper is organized as follows. In Section II, themathematical formulation of the radio map is presented. InSection III, the location estimation phase is covered fromthe deterministic (III-A) and the probabilistic (III-B) pointof view. In Section IV, the traditional location fingerprintingis extended to computating the location estimates in timeseries by applying different filters. Different state models arecombined with the static location estimation algorithms. Thetest results are presented in Section V. Section VI summarizesthe results and suggests guidelines for the future work.II. R