MobileNet for Differential Constellation Trace Figure

Abstract

Radio frequency fingerprint technology is of great significance to the security of the Internet of Things system. When the signal uses I/Q modulation, the demodulated signal can be drawn on a two-dimensional plane, that is, constellation diagram. Slight differences in the manufacture and use of transmitters can cause the constellation to shift. Differential Constellation Trace Figure (DCTF) with device characteristics can be generated by special differential processing of demodulation signal. In this paper, the DCTFs of QPSK signal are used as radio frequency fingerprints. Radio frequency fingerprints are classified using MobileNet, a lightweight network that can be used for mobile and embedded devices. In the final simulation experiment, the accuracy of MobileNet V2 and MobileNet V3 are both over 95% when SNR is low. Comparing with the accuracy of CNN, MobileNet V2 and MobileNet V3 are better choices.