Proximity Verification Based on Acoustic Room Impulse Response

Device proximity verification has a wide range of security applications such as effortless zero-interaction authentication, multi-factor authentication, group-membership management and many more. To achieve high ease-of-use, a recently proposed class of solutions exploit contextual information captured by onboard sensors including radio (Wi-Fi, Bluetooth and GPS receivers), ambient sounds (microphones), movement (accelerometers) and physical environment (light, temperature and humidity) to facilitate the verification process with minimal user involvement. Active acoustic methods have some advantages over many others: they work indoors, they can take the shape of the enclosure and barriers into account, they don’t require pre-installed infrastructure and they are relatively fast. In this paper we propose R-Prox, an approach for proximity (copresence) verification based on acoustic Room Impulse Response (RIR). In R-Prox, one device actively emits a short, wide-band audible chirp and all participating devices record reflections of the chirp from the surrounding environment. From this impulse response signal, we extract features on different frequency bands and compare them for a copresence verdict. This RIR method is less dependent on pre-existing sound than passive ambient sound methods and more resilient to relay attacks since it is, by its very nature, dependent on the physical surroundings. We evaluate our method by collecting RIR data with various Commercial Off-The-Shelf (COTS) mobile devices in different rooms. We present an application to collect data from multiple devices simultaneously. We then train a binary classification model to determine copresence using RIR features. In our experiments we show R-Prox to be sensitive, with false negative verdicts can be as low as 0.059 of the true copresence cases. Although R-Prox’s false positive rate is not as low (in case rooms are likely having similar acoustic RIR), we show that it can be effectively combined with schemes like Sound-Proof (which suffers from high false positive rates in some adversarial settings) so that the resulting system has high accuracy.