A Graphical Deep Learning Approach to RF Fingerprinting in the Time-Frequency Domain

The emergence of the Internet of Things (IoT) as a global infrastructure of interconnected network of heterogeneous wireless devices and sensors is opening new opportunities in myriad of applications. This growing pervasiveness of IoT devices, however, is leading to growing concerns regarding security and privacy. Radio Frequency (RF) fingerprinting techniques operating at the physical layer can be used to provide an additional layer of protection to ensure trustworthy communications between devices to address these concerns. We present a graphical deep learning approach in the time-frequency (TF) domain based on short-time Fourier transform (STFT) where the intensity information of the STFT is used to generate 2-D image inputs for training and testing of the deep learning models for RF fingerprinting and identification. The performance of the proposed approach is evaluated and compared with the baseline approach operating in the waveform domain, using the same neural network architecture based on over-the-air captured datasets from 12 Zigbee devices. The experimental results show that the proposed approach outperforms both baseline approach, achieving nearly 100% identification accuracy based on data captured in a makeshift RF chamber, and accuracy of 98% on the dataset which was captured under real-life conditions, demonstrating the robustness of the proposed approach. Furthermore, the impact of the STFT-parameter selection on the identification performance of the proposed approach is also evaluated.