Análisis de medidas de potencia en interiores para su aplicación en sistemas de localización basados en la técnica del fingerprinting

Abstract

This thesis relates to the technique of indoor tracking called fingerprinting. Indoor tracking is having greater advances in the last years due to the ever increasing demand on services for added values in mobiles. Many of these services have to do with the contextual positioning of the user. The technique known as fingerprinting is one of the most widely used procedures in indoor tracking systems. This technique uses the relation between the levels of power being received in the mobile from different access points in a wireless network and the levels of power contained in a series of points, which are called fingerprints, the position of these points is definitely known. This array of points receives the name of radiomap. Therefore it is of paramount importance to possess precise values of power within the traces in order to achieve accurate precision while tracking. However and due to the propagation characteristics found indoors this aim could prove complicated as a result of the randomness of the received signal. In this thesis a rather small net for tracking formed by few access points was designed and a mobile device used. With this net various measurements at different locations and different environmental scenarios were carried out. A factor upon which special attention was focused was the human presence at the sites where the measurements occurred. The objective was to realize a study about the influence of different factors upon the variability of the levels of power measured at various positions or fingerprints, and by the same token how the tracking precision can vary in different situations. Clustering algorithms have been implemented in order to separate the values of power measured in groups or clusters. The idea behind it is to identify each particular cluster with a particular fingerprint. Among the applications this technique offers is the capability to test if the tracking system has been perfectly designed just by observing if the clusters coincide with the fingerprints. Another possibility is the design of a tracking algorithm based on the identification of the values of power received in the mobile with a particular cluster, and therefore with a fingerprint. For this reason the algorithms k-means and rek-means have been implemented and consecutively applied to a group of powers measured over a small radiomap, demonstrating that the algorithm rek-means is therefore the most effective one.