Uncovering the Portability Limitation of Deep Learning-Based Wireless Device Fingerprints

Abstract--Recent device fingerprinting approaches rely on deep learning to extract device-specific features solely from raw RF signals to identify, classify and authenticate wireless devices. One widely known issue lies in the inability of these approaches to maintain good performances when the training data and testing data are collected under varying deployment domains. The same also happens when considering other varying domains, like channel condition and protocol configuration. We will next demonstrate how the limited portability of fingerprints can impact device fingerprinting I. Recently, there has been considerable interest in adopting deep A. Testbed and Data Collection Setup learning-enabled device fingerprinting in automated network To explain these challenges, we used our IoT fingerprinting authentication mechanisms for emerging large-scale wireless testbed [7] to run several experiments under different devices (e.g., 6G, IoT, vehicular, etc.) [1], [2]. In essence, varied domains, by training and testing the deep learning device fingerprinting relies on deep learning techniques to models on data collected on different days, using different extract device-specific features and signatures, solely from raw receivers, and/or under different protocol configurations.