Radio frequency fingerprinting based on the constellation errors

Radio frequency fingerprinting (RFF) is used to uniquely identify individual radios. This paper investigates the sources of RF fingerprints and proposes a RFF approach based on the constellation errors. The transmitter imperfections and their effect on constellations are explored, from which the conclusion is made that the constellation errors contains transmitter identity information. Then a machine-learning algorithm named subclass discriminant analysis (SDA) with a slight modification is introduced to extract fingerprint features from the constellation errors. The proposed approach provides richer discriminant information than previously proposed transient based, preamble based and modulation-metrics based approaches. Experimental results validate the superior performance of the proposed approach.