Ambient sensing, human activity recognition and indoor floor mapping are three classical context detection problems that hackers try to crack using a variety of sources of side information. Other than overt signals such as microphones and cameras, covert channels such as WiFi, Bluetooth, assisted and coarse-grained Global Positioning System (GPS) signals have been exploited to date. Fine-grained or precise GPS information that is ubiquitously available to almost all Android devices as of today, has not been studied as a side channel for these problems. We present a novel set of attacks that exploit the 9 parameters that precise GPS provides to infer such contexts with an accuracy that often exceeds 99%. Specifically, this work reports a longitudinal study conducted over a period of more than a year in a 40,000 sq. km geographical region, where data was passively collected in all kinds of settings including open stadia, underground metros, flights, cruise ships, office spaces, dormitories and high-altitude locations. We propose a novel method, AndroCon , that combines linear discriminant analysis, unscented Kalman filtering, gradient boosting and random forest based learning to create a highly accurate context sensor. To the best of our knowledge, this study provides the most comprehensive characterization of all the information that we get from precise GPS signals, finds the parameters of interest, and uses them effectively to match the quality of results of other methods that rely on stronger channels such as WiFi. For instance, AndroCon can detect motions such as hand waving in front of the phone, which is known to be a very challenging use case.
Nag et al. (2026) studied this question.