Feature-based modulation classification using circular statistics

In this paper we present a new set of signal features that can be used for modulation classification of digital communication signals in a blind environment. These new features are based on circular summary statistics taken from the instantaneous frequency of the sampled signal. The instantaneous frequency of a sampled baseband digital signal is expressed in radians. We consider the sampled instantaneous frequency as a set of realizations of a circular random variable and apply circular summary statistics to extract classification features. In particular, we use trigonometric moments of the instantaneous frequency to create feature vectors. We address the problem of distinguishing between FSK-type signals and QAM-type signals; and subsequently, the problem of discrimination between the FSK-type signals. We show that in both problems the signal classes are well separated in the circular statistics feature space and that automatic classifiers can be defined with simple thresholds.