The Application of Machine Learning Techniques on Channel Frequency Response Based Indoor Positioning in Dynamic Environments

Traditional IPS uses triangulation based on signal strength but its accuracy is impaired in non-line-of-sight (NLOS) situations. Among the available wireless technologies for indoor positioning, WiFi is a good candidate since it is supported by existing mobile devices and infrastructure indoors, and it can operate under both LOS and NLOS conditions. One of the cutting-edge WiFi-based localization techniques exploits time-reversal resonating strength (TRRS) of coherent channel frequency responses (CFR). The basic concept of CFR-based positioning is based on the similarity measure between the testing CFR and the pre-recorded CFR fingerprints. A common assumption in previous works is that the wireless channel is time invariant. In this paper, we study CFR-based positioning in a dynamic indoor environment. Using the collected channel response fingerprints for both LOS and NLOS scenarios, we exploit supervised machine learning techniques to enhance the processing speed while achieving high positioning accuracy under the effect of dynamic wireless channels.