Jamming Echoes: On the Impact of Out-of-Band Interference on Radio Frequency Fingerprinting

Radio Frequency Fingerprinting (RFF) has recently emerged as a lightweight and efficient strategy for classifying wireless devices based on their Radio Frequency (RF) emissions at the physical layer. Such emissions contain device-specific distortions that, although not affecting the quality of the communication link, can be extracted from the received signals through capable hardware (Software-Defined Radios—SDRs) and be used to classify via Deep Learning (DL) techniques the specific transmitters in a pool of RF devices. Recent research has shown that, although promising, RFF is a fragile phenomenon whose performance is significantly affected by various phenomena, e.g., channel fluctuations, device reboot, and firmware reload operations. In this paper, we shed light on yet another phenomenon affecting the reliability and robustness of RFF, i.e., interfering out-of-band signals. Through an extensive real-world experimental campaign involving seven heterogeneous SDRs and state-of-the-art DL image-based RFF systems, we demonstrate that out-of-band interfering signals emitted on neighboring frequencies (less than 5 MHz apart from the main communication channel), independently from being malicious, reduce the accuracy of RFF up to a random guess of the transmitter, while not significantly impacting the Bit-Error Rate of the communication link. These results foster further research in the design of reliable and robust DL-based RFF systems, capable of mitigating real-world deployment factors.