WRIST: Wideband, Real-Time, Spectro-Temporal RF Identification System Using Deep Learning
Hai Nguyen, Marinos Vomvas, Triet D. Vo-Huu, Guevara Noubir
Abstract
RF emissions’ detection, classification, and spectro-temporal localization are essential not only for understanding, managing, and protecting the radio frequency resources, but also for countering today's security threats such as jammers. Achieving this goal for wideband, real-time operation remains challenging. In this article, we present WRIST, a Wideband, Real-time, Spectro-Temporal RF Identification system. WRIST can detect, classify, and precisely locate RF emissions in time and frequency using RF samples of 100 MHz spectrum in real-time. The system leverages an <i>one-stage object detection</i> Deep Learning framework, and transfer learning to a multi-channel visual-based spectral representation. Towards developing WRIST, we devised an iterative training approach which leverages synthesized and augmented RF data to efficiently build a large dataset with high-quality labels. WRIST achieves over <inline-formula><tex-math notation="LaTeX">$99 \%$</tex-math></inline-formula> class detection accuracy, <inline-formula><tex-math notation="LaTeX">$94 \%$</tex-math></inline-formula> emission precision and recall, with less than 0.08 bandwidth and time offset ratios in a large anechoic chamber over-the-air environment. In the extremely congested in-the-wild environment, WRIST still achieves over <inline-formula><tex-math notation="LaTeX">$80 \%$</tex-math></inline-formula> precision and recall. WRIST currently supports five 2.4 GHz technologies (Bluetooth, Lightbridge, Wi-Fi, XPD, and ZigBee) and is easily extendable to others. We are making our curated dataset available to the whole community. It comprises over 10 million labelled RF emissions from off-the-shelf wireless radios spanning the five classes of technologies.