A novel special emitter identification method based on improved subclass discriminant analysis

Radio frequency fingerprinting (RFF) is used to uniquely identify individual radios by exploiting the radio frequency characteristics. Often attributing to the phase ambiguity, the features of an emitter may split into several clusters, in this case, the traditional feature extraction methods, such as Linear Discriminant Analysis (LDA), Subclass Discriminant Analysis (SDA), lose efficacy. This paper investigates the problem of feature extraction and presents an improved SDA method. By modifying the clustering algorithm and replacing the sample covariance matrix with within-subclass scatter matrix, our method can achieve better performance.