A Study of RF Fingerprint Feature Dimensionality Expansion for I/Q Signal Data Extraction

Security research based on RF fingerprints has received extensive attention in recent years. Traditional RF fingerprint identification technology is first based on feature engineering method to extract signal key features and then uses machine learning method to identify and classify, but the method relies on expert experience and high computational complexity and lacks universality in feature selection and judgment criterion; while the RF fingerprint identification research based on deep learning that has emerged in recent years is usually obtained by extracting statistical features from signal sample sequences and often require colossal storage and arithmetic power. This paper proposes an RF fingerprint feature expansion dimensional recognition method to save storage and computational power and improve the universality and robustness of RF fingerprint recognition technology. The method preserves some results of traditional RF fingerprint recognition feature engineering and then builds a dimensional data table using the time-frequency map image features converted from the sample sequence of the extracted signal as the extended dimension. Finally, using the feature dimension data as input, machine learning and deep learning models are used for fingerprint recognition, respectively.