On the Reliability of Radio Frequency Fingerprinting

TL;DR

The paper addresses the reliability of Radio Frequency Fingerprinting (RFF) and demonstrates that FPGA image reloads in Software Defined Radios induce fingerprint mutations, challenging the assumption of a persistent, unique PHY fingerprint. It introduces a graph-based framework to abstract fingerprint mutations and a measurement methodology using five USRP X310 SDRs, converting I-Q samples to 224×224 images and evaluating pairwise similarity with CNN classifiers to produce a dissimilarity index $\delta$. Key findings show pronounced mutability across FPGA reloads: only about $9\%$ of measurements are consistently linked, roughly $40\%$ exhibit a different fingerprint, and $51\%$ are ambiguous, with a median authentication requiring observing at least $24\%$ of fingerprint measurements to exceed $P>0.9$. The work provides a generalizable tool for assessing RFF reliability, underlining a security-privacy trade-off and guiding future research and practical deployment beyond channel-induced variability.

Abstract

Radio Frequency Fingerprinting (RFF) offers a unique method for identifying devices at the physical (PHY) layer based on their RF emissions due to intrinsic hardware differences. Nevertheless, RFF techniques depend on the ability to extract information from the PHY layer of the radio spectrum by resorting to Software Defined Radios (SDR). Previous works have highlighted the so-called ``Day-After-Tomorrow'' effect, i.e., an intrinsic issue of SDRs leading to a fingerprint mutation following a radio power cycle. In this work, we extend such a study by demonstrating that fingerprint mutations appear every time a new FPGA image is reloaded, i.e., when the SDR initiates a new communication. In this context, we provide an in-depth analysis of the reliability of RFF over multiple FPGA image reloading operations, highlighting its ephemeral and mutational nature. We introduce a methodology for abstracting fingerprint mutations into a graph and provide a theoretical framework for assessing fingerprint reliability. Our results show that the common assumption of considering the RF fingerprint as unique and always persistent is incorrect. By combining real-world measurements, high-performance SDRs, and state-of-the-art deep learning techniques, we experimentally demonstrate that radio devices feature multiple fingerprints that can be clustered according to shared features. Moreover, we show that the RF fingerprint is a time-independent probabilistic phenomenon, which requires the collection of multiple samples to achieve the necessary reliability.

Figures (14)

• Figure 1: RFF overview: the receiver can infer the transmitting source by just eavesdropping a raw signal through a SDR. Differences in electronics and software configuration reflect in the over-the-air signal (features), and the receiver can detect and extract these features, thus identifying the signal source.
• Figure 2: Our methodology involves three phases: (i) Measurement collection, (ii) Dissimilarity analysis of the measurements, and finally, (iii) Analytics extraction.
• Figure 3: We consider two opposite cases: measurements 6 and 11 are characterized by high accuracy (a), while measurements 24 and 25 cannot be distinguished by the classifier (b). We infer that the fingerprint is different for case (a) while it is the same for case (b).
• Figure 4: Accuracy (dissimilarity index) between consecutive measurements for five transmitters. The color map spans between blue ($\delta = 0.5$) and red ($\delta = 1$).
• Figure 5: Sorted hit rate (dissimilarity index - $\delta$) while considering all ($300 \cdot 5 = 1500$) measurements. We identify three regions: R1 where $\delta = 1$ (FPGA image between measurements is different), R3 where $\delta = 0$ (measurements have the same fingerprint), and finally, R2 where fingerprint detection is challenging.

Theorems & Definitions (1)

• Remark