Modulation identification for intercept sonar based on overcomplete dictionary

The purpose of intercept sonar is to detect the active transmissions from ships, submarines and torpedoes, using which fine target discrimination, recognition and classification is performed. Intercept signal detection and identification techniques continuously become more advanced and sophisticated. Signal's intelligence aims at extracting as much information as possible on detected sonar signals (number of paths, carrier frequency, modulation scheme, signal length etc). Either time or frequency representation is insufficient for the feature extraction. In this paper a modular approach is proposed which exploits the parsimonious time-frequency representation of the signal by applying different distributions such as Spectrogram, Wavelet transform, Wigner-Ville distribution (WVD) and Choi-Williams distribution (CWD), with emphasis on underwater scenario. The inherent characteristics of intercept signal are not completely exploited by representing them on a single set of basis vectors. We observe signal through known overcomplete dictionaries where each time-frequency representation reveals some characteristics and performance of different signals in each dictionary module.