CovertBand: Activity Information Leakage using Music

This paper contributes a novel method for low-cost, covert physical sensing and, by doing so, surfaces new privacy threats. We demonstrate how a smartphone and portable speaker playing music with embedded, inaudible signals can track multiple individuals’ locations and activities both within a room and through barriers in 2D space. We achieve this by transforming a smartphone into an active sonar system that emits a combination of a sonar pulse and music and listens to the reflections off of humans in the environment. Our implementation, CovertBand, monitors minute changes to these reflections to track multiple people concurrently and to recognize different types of motion, leaking information about where people are in addition to what they may be doing. We evaluated CovertBand by running experiments in five homes in the Seattle area, showing that we can localize both single and multiple individuals through barriers. These tests show CovertBand can track walking subjects with a mean tracking error of 18 cm and subjects moving at a fixed position with an accuracy of 8 cm at up to 6 m in line-of-sight and 3 m through barriers. We test a variety of rhythmic motions such as pumping arms, jumping, and supine pelvic tilts in through-wall scenarios and show that they produce discernibly different spectrograms from walking in the acoustic reflections. In tests with 33 subjects, we also show that even in ideal scenarios, listeners were unlikely to detect a CovertBand attack.