Research On Radio Fingerprints Localization Techniques Based On Cluster Analysis

The mobile Internet has dual characteristics of mobile and Internet, and location-based services (LBS) have become its standard configuration. With the rapid development of mobile Internet, the increasing LBS applications have higher requirement for the geographic information location, so the wireless location system is becoming a hot topic.Radio frequency (RF) fingerprint-based WiFi positioning technology provides a new real-time location tracking solutions for its easy deployment and low cost advantages. However, this technology also faces many problems, such as large real-time searching overhead, the time-varying characteristics of received signal, the signal fluctuations of different devices at the same location caused by device heterogeneity. For the series of problems arisen in the positioning process, this thesis adopts a cluster-based RF fingerprinting positioning technology and analyzes the location characteristics, time-varying characteristics of heterogeneous devices based on the statistical characteristics of radio frequency fingerprint. The thesis uses clustering algorithms for signal data analysis and processing the positioning algorithm validation in real environments. First, the thesis uses the DBSCAN algorithm to eliminate signal data of lager errors or deviations to build a high-quality fingerprint database. Second, for signal’s time-varying characteristics, this thesis uses the Affinity Propagation clustering algorithm to solve the problem of migration time of the signal effectively. In location clustering, by considering multiple floors and single floor layout, the thesis uses clustering methods based on AP sets and similarities of fingerprints respectively, which effectively reduce the localization computational overhead and improve the positioning speed. In addition, different devices in the same position received varying signal strength which is seriously affecting the positioning accuracy. For this, this thesis adopts a method to cluster the heterogeneous devices and then makes mappings between clusters to improve the positioning accuracy and reduce the sampling overhead.This thesis makes contributions to clustering analysis in the aspect of dealing with the positioning problems and designs an approach of building a high quality fingerprint database, clustering methods based on location, time, device and other dimensions, which improve localization accuracy and reduce the cost.