Physical-layer fingerprinting of LoRa devices using supervised and zero-shot learning

Abstract

Physical-layer ngerprinting investigates how features extracted from radio signals can be used to uniquely identify devices. is paper proposes and analyses a novel methodology to ngerprint LoRa devices, which is inspired by recent advances in supervised machine learning and zero-shot image classi cation. Contrary to previous works, our methodology does not rely on localized and low-dimensional features, such as those extracted from the signal transient or preamble, but uses the entire signal. We have performed our experiments using 22 LoRa devices with 3 di erent chipsets. Our results show that identical chipsets can be distinguished with 59% to 99% accuracy per symbol, whereas chipsets from di erent vendors can be ngerprinted with 99% to 100% accuracy per symbol. e ngerprinting can be performed using only inexpensive commercial o-the-shelf so ware de ned radios, and a low sample rate of 1 Msps. Finally, we release all datasets and code pertaining to these experiments to the public domain.

Physical layer identification of LoRa devices using constellation trace figure

“…The LoRa terminal and USRP are 5 m apart without obstacles and 1500 symbols are collected for the six terminals with sample rate of 5 Msps. The comparisons in Table 2 are consistent with the results in [33], and the proposed method obtains a high accuracy under the current experimental conditions. The next experiment is the effect of sample rate.…”

“…They used the proposed method to identify and authenticate ZigBee devices effectively. For the research of radio frequency fingerprint of LoRa signal, in 2017, Robyns et al [33] proposed a new method of fingerprint recognition based on machine learning. Unlike other methods, this method does not extract the local features of the signal, but takes the pre-processed data of the signal as the whole object of recognition and the input data of machine learning.…”

“…Unfortunately, most of the current methods are designed for specific wireless terminals and cannot be directly used for the LoRa module. The supervised machine learning (SML) method in [33] applied to LoRa terminals is compared with the proposed method. In general, these two methods belong to different types and the main differences are summarized as follows.…”

Securing the Internet of Things in the Age of Machine Learning and Software-Defined Networking

“…Bluetooth + NFC) to ensure secure pairing [45]. Another approach to device identification is device fingerprinting, which leverages imperfections of hardware components (i.e., clock skew, RF signature, phase noise, and so on) to uniquely identify different wireless devices [46]. Given its paramount role toward an effective and efficient IoT, research on IoT authentication has gained a lot of traction in the research community over the last years.…”

I Send, Therefore I Leak

“…We assume that the attacker has a priori knowledge regarding the application associated with an end device transmission. It has already been shown that multiple LoRa end devices can be uniquely identified using physical layer fingerprinting techniques [27]. Furthermore, the packet header information is transmitted in the clear.…”