Radio Frequency Fingerprint Identification of GMSK Modulated Signals Based on Eye Diagram Traces Deviation

Radio frequency fingerprint (RFF) technology has emerged as a prominent research direction within the field of wireless security because it uses the distinctive hardware characteristics of individual devices for identity authentication. In the context of the Gaussian Filtered Minimum Shift Keying (GMSK) modulation mode, this research introduces an RFF identification framework centered on the analysis of eye diagram, and proposes an RFF feature extraction method based on eye diagram traces deviation (EDTD). In practical applications, considering that the presence of synchronization errors leads to the divergence of the eye diagram traces (EDT) obtained through baseband demodulation, a frame folding synchronization algorithm is proposed, which enhances the generation of the EDT. The eye diagram traces density plot generated from EDT is further utilized to obtain the probability density curve with peak clustering characteristics. Furthermore, recognizing the significance of the time-bandwidth product (<inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\boldsymbol {BT}$</tex-math></inline-formula>) in shaping the GMSK baseband signal eye diagram, a <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\boldsymbol {BT}$</tex-math></inline-formula> parameter estimation algorithm based on probability density curves is proposed, and then an inversion calculation is applied to attain the ideal EDT. By scrutinizing the statistical attributes of the EDTD, which represents the disparity between the measured signal's EDT and the theoretically ideal EDT derived through inversion, a 27-dimensional RFF feature vector is constructed. This feature vector is combined with a lightweight classifier to accomplish the task of wireless device identification. To verify the effectiveness of our proposed method, two types of devices utilizing GMSK modulation methods were used to construct separate datasets, which are marine automatic identification system (AIS) devices and Internet of Things (IoT) modules utilizing the global system for mobile (GSM) communications technology. And it was tested and compared with intelligent feature extraction methods based on deep learning. The results show that with only 150 training samples per device, our solution can achieve 99% identification accuracy for AIS devices composed of independent radio frequency circuits, and over 80% for GSM IoT devices composed of integrated circuits.