Radio Frequency Fingerprint Identification (RFFI) exploits inherent hardware imperfections in devices for identification, and hardware imperfections generate Radio Frequency Fingerprint (RFF) including various features. The nonlinearity of the power amplifier (PA) is an important feature of RFF that has been widely used in radio frequency identification (RFFI) technology. In this paper, we constructed a testbed consisting of 60 IEEE 802.15.4 devices and a universal software radio peripheral (USRP) X310 software-defined radio (SDR) platform as the receiver for the estimation of PA parameters. Using this testbed, we modeled and estimated PA parameters using 60 real IEEE 802.15.4 devices to obtain real PA parameters and generate synthetic PA parameters for generation of synthetic devices. When the number of devices requiring identification or research is limited, this method can be employed to expand the device sample size to conduct identification model training or related studies. Experimental results demonstrated the feasibility of the synthetic data-based pre-training approach, whose performance approaches that of trained models without pre-training. In practical applications, legal device identity registration in device identification can be achieved with a low-cost fine-tuning compared to retraining.

In radio frequency fingerprint identification (RFFI) systems, mitigating channel interference remains a critical challenge. This paper introduces a robust RFFI system to tackle this issue effectively. Specifically, taking the IEEE 802.11 signal as the case study, a signal representation is designed based on the logarithmic spectrum, while an RFF extractor based on the U-Net neural network is employed which is guided by a proposed Channel2Channel (C2C) algorithm and powered by a designed data augmentation method. Furthermore, a collaborative identification mechanism is proposed based on a support vector machine (SVM) classifier, where a multi-frame RFF fusion method is designed to exploit the diversity across different frames of received signal. Extensive experimental evaluations are performed in various real-world scenarios using 7 mobile phones and a universal software radio peripheral (USRP) X310 receiver, where an average classification accuracy of 95.72% is obtained with a single frame of received signal, outperforming the neural network-based benchmarks, and an average accuracy of 99.46% is acquired with 10 signal frames based on the proposed collaborative identification method. In addition, the deployability of the system on a resource-constrained computing platform is also validated.

The radio frequency fingerprint (RFF) has gained significant traction in the identification of wireless Internet of Things (IoT) devices. However, RFFs extracted from wireless signals are inherently susceptible to noise, particularly for narrowband signals. Furthermore, the noisy domain adaptation (NDA) problem presents a substantial challenge for RFF identification due to the variable noise interference across different noisy domains. To address this, the squared cross power spectral density (SCPSD) as new device RFFs is derived theoretically as a function of signal-to-noise ratio (SNR). Combined with the proposed high-precision SNR estimation algorithm, SCPSDs under low SNR can be reconstructed to the same feature distribution as those under high SNR. Because of the interpretability, ten samples under high SNR from each device under test (DUT) and a shallow convolutional neural network (CNN) are trained for experimental evaluation on the NDA problem. Tested on 60 off-the-shelf ZigBee DUTs, the improvement of identification accuracy is around 26% for SNR between 5 dB and 10 dB, and the overall improvement is more than 20% compared to the baseline. It outperforms the three other compared methods across all testing SNR and is highly practical.

Radio frequency fingerprint (RFF) identification as a physical layer authentication technique by leveraging devices’ unique hardware-level imperfections in transmitted signals has been considered as a potential complement to combat spoofing attacks. In this paper, we develop a lightweight framework to identify long-term evolution (LTE) devices with intrinsic features extracted from Msg3. Channel-robust fingerprints from channel state information (CSI) in demodulation reference signal (DMRS) associated with physical uplink shared channel (PUSCH) are obtained. These modulation features are combined with transient-on features that are obtained from the cyclic prefix (CP). A hybrid feature matrix is constructed and fed into a shallow long short-term memory (LSTM) network, which improves the identification accuracy compared to using single feature representation. We carry out extensive experiments with five LTE devices in real-world environment via a pseudo base station. The robustness of our proposed scheme is evaluated by cross-scenario training and testing. Thanks to the hybrid information-rich feature matrix as input of network, the classification accuracy 90.90% obtained at 25 dB when training in static but testing in dynamic scenarios demonstrates that our scheme is channel-robust in the presence of channel variations.