Characterization and feature extraction of transient signals using multifractal measures

This paper focuses on multifractal characterization and feature extraction of nonstationary temporal signals, particularly the turn-on transients for radio transmitter classification purposes. A transient signal contaminated by noise can be considered as an output from a chaotic dynamical system. The trajectory of such a system in phase space is often attracted to a bounded fractal object called strange attractor. Transient signals are preprocessed to obtain the corresponding strange attractor in phase space. Multifractal measures based on a generalized entropy are then applied to the strange attractor and features are extracted in terms of a fractal dimension spectrum. Experimental results show that there exists a strange attractor for the transient signal and the fractal dimension spectrum can characterize a transient signal uniquely, even with a relatively short time duration.