November 16th 2017 par B. Berruet : Indoor localization performance analysis with CSI-based feature extraction algorithms
The expansion of location-based applications for monitoring the crowd and giving specific services in personal assitance activities requires the development of solutions working with the ambient connectivity and without industrial installations. That requirement is widely covered by the global navigation satellite systems (GNSS) present all around the world but the GNSS signals is severely attenuated in the urban canyon or the indoor environments. The wireless communication systems such as the WiFi or the bluetooth low enery (BLE) become then for the industrial and the academic researches as an alternative solution to the GNSS in these complex scenarios. Hence, many propositions emerge from the ashes based on different localization approaches. My thesis looks for developing a system respecting the IoT and ambient connectivity paradigms: Battery saving, low-cost and fast deployement. From this, the fingerprinting approach based on the channel state information (CSI) completes the above requirements. CSI reveals the influence of the channel propagation on the transmitted signal such as the multiple paths taken by this one or the fading. The fingerprinting approach consists in acquiring the CSI at regularly spaced positions in the study field. In this way, the system try to determine a function from the signal space to the spatial space. One solution is to implement machine learning techniques. My last works was to study the spectral methods such as the principal component analysis (PCA) in order to improve the machine learning algorithm.