Localisation d'usager dans un environnement interne en utilisant une technique pour contrer la variation temporelle du canal

Several researches suggest systems allowing indoor location estimation in no-time-varying channels. However, few of those studies take into account the case of time-varying channels. This thesis propose a location system allowing to counter time-varying channel in an indoor environment. The received signal strengths (RSS), in a WLAN system, are measured and used in a Fingerprinting technique in order to define the channel signature. The database is therefore created, in 3D, in no-time-varying channel at t 0. The Fingerprinting technique suffers degradations in location accuracy in time-varying channels. In that case, new measurements are carried out to create a new database. This thesis propose a method allowing the location in time-varying channel when it is difficult or impossible to collect measurements. The system generates a database with measurements of few reference points and a Data Mining technique called the Model Tree. The method uses a regression analysis to learn the temporal predictive relationship between the RSS values of the mobile and the reference points and generates a new database at tx. Afterwards, an artificial neural network (ANN) is applied to the database for location estimation. The method tree generates a degradation less than 9% within 3 meters accuracy. In order to validate the proposed system, 3 databases are generated at different rimes (t1, t 2 and t3) and used simultaneously as inputs for the ANN to estimate the position of the mobile. The distance location accuracy of 3 meters is obtained for 61.38% of the cases and of 5 meters for 78.22% of the cases. This causes a supplementary degradation of approximately 7% in 2D.;The 3D estimations adds a 3.5% supplementary degradation to the two previous ones because of the z variable which defines the mobile floor location (in a building). This approach allows a 3 meters accuracy in 57.06% of the cases and of 5 meters in 74.81% of the cases in 3D. The location system proposed in this thesis achieves acceptable results particularly in 2D. It is recommended when time and material means are insufficient or limited.