Radio frequency fingerprint identification towards statistical and deep learning features: Review, recent results and future directions

In the context of next-generation wireless communication and heterogeneous Internet of Things (IoT) systems, the security of IoT based on cryptographic mechanisms and security protocols presents significant vulnerabilities. The radio frequency fingerprinting (RFF) method based on the physical layer of signals is considered an effective and reliable solution to address these issues. RFF identification utilizes the subtle differences in radio frequency signals generated by emitters to distinguish between different individuals, making it difficult to clone or forge. RFF identification extracts features from collected radio signals through signal processing to achieve specific emitter identification (SEI) tasks, with RFF features being the key elements for realizing SEI. To promote research development in this field, this paper comprehensively reviews the identification methods of RFF from the perspectives of statistical features and deep learning features. Firstly, this paper introduces the fundamental theory of RFF, including its origin, formation mechanism, properties, an analysis of its development trends, and a compilation of available datasets. Secondly, based on the general model of emitter identification, it reviews current RFF identification methods based on both statistical and deep learning features and conducts a comparative study of these two approaches. Finally, it highlights several development challenges and potential research directions in the field of intelligent RFF identification, aiming to guide future research and applications in this domain.