Particle Filters and Position Tracking in Wi-Fi Networks

Abstract

In this work we quantify the usefulness of particle filters applied to the problem of mobile device tracking in WiFi networks, under the assumption of log-normal fading. Our principal aim was to determine if a real-time deployment of a particle filter was possible while still providing factor two gains in the prediction performance relative to a stand-alone optimal Wi-Fi positioning algorithm. We conclude that the required gains are achieved in our adopted filter algorithm when the particle number is set to the relatively small number of 300, meaning that a real-time deployment is possible. In addition, we quantify the performance gain of the particle filter when intermittent GPS information is available to the mobile device. We propose the fusion of the GPS information be implemented as a renormalization of the particle cloud. Finally, we probe the limits of the filter performance under biased-error distributions. Our simulations show that tracking of people, vehicles and robotic devices in an outdoor Wi-Fi network, where non-linear and nonGaussian conditions exist, can be significantly enhanced by the pragmatic real-time particle filter algorithm presented here.