The technology of radio frequency fingerprint identification based on deep learning for 5G application

Abstract

User Equipment (UE) authentication holds paramount importance in upholding the security of wireless networks. A nascent technology, Radio Frequency Fingerprint Identification (RFFI), is gaining prominence as a means to bolster network security authentication. To expedite the integration of RFFI within fifth-generation (5G) networks, this research undertakes the creation of a comprehensive link-level simulation platform tailored for 5G scenarios. The devised platform emulates various device impairments, including an oscillator, IQ modulator, and power amplifier (PA) nonlinearities, alongside simulating channel distortions. Consequent to this, a plausibility analysis is executed, intertwining transmitter device impairments with 3rd Generation Partnership Project (3GPP) new radio (NR) protocols. Subsequently, an exhaustive exploration is conducted to assess the impact of transmitter impairments, deep neural networks (DNNs), and channel effects on RF fingerprinting performance. Notably, under a signal-to-noise ratio (SNR) of 15 dB, the deep learning approach demonstrates the capability to accurately classify 100 UEs with a commendable 91% accuracy rate. Through a multifaceted evaluation, it is ascertained that the Attention-based network architecture emerges as the optimal choice for the RFFI task, serving as the new benchmark model for RFFI applications.