Direct inference of location-related context from wireless signal strength

In this dissertation we derive location-related context like mobility-states, co-mobility, speed and decelerations directly from the wireless signal strength information. The key insight is that the time-series of signal strength is robust to environmental factors that typically negatively affect the RSS-based localization systems. Therefore, inferring these physical properties directly from the time-series of wireless signal strength is more accurate than deriving them from location estimates. We apply correlation and time warping algorithms to the time series of wireless signals to infer these properties. Our trace-driven experimental approach shows that our inference techniques can work with minimal infrastructure, are computationally efficient, requires no explicit user participation and can produce higher accuracies than location-based systems. We have also experimentally identified the factors that limit the accuracy of indoor localization and have proved the existing assumptions behind theoretical lower bounds of indoor localization incorrect. Our results will enable new context aware applications, because accurate estimates of co-mobility and speed offer a richer set of primitives available to applications. Such applications can derive user mobility states like walking, running, driving or social states, such as if a user is in a meeting or alone.