A Triplet Net-based High-Quality I/Q Data Collection Mechanism for 5G RFF Identification

The 5G Radio Frequency fingerprint (RFF) technology, by utilizing the unique physical layer radio frequency hardware characteristics of 5G terminals as a singular identity marker, can be adopted to verify the legitimacy of 5G terminals in the IoT domain. Research efforts predominantly employ deep learning methods to derive RFF features from I/Q data. However, the collection and utilization of high-quality I/Q data that aligns with real dynamic wireless channels is seldom discussed to improve identification accuracy. To fill this gap, this paper proposes a high-quality I/Q data collection mechanism based on the Triplet Net, named Cyclic Registration Mechanism (CRM). The CRM utilizes the registration mechanism of the Triplet Net to train a model that offers greater flexibility than traditional models, such as Convolutional Neural Networks (CNNs). In this context, the CRM suggests periodically collecting I/Q data to create training and registration datasets for RFF training and inference, thereby mitigating the effects of wireless channel dynamics. The CRM has been implemented in a real-world platform, where identification performance have been simulated in dynamic wireless channels. The results demonstrates that the CRM-based Triplet Net can still maintain good identification performance when the collection times of training and testing data are widely separated.