Moment feature variability in uncertain propagation channels

In this study, moments of a signal as features for classification in active sonar systems are considered. The authors analyse the impact of propagation effects on the moment features, including the effect of random variability in certain channel parameters, such as target distance. Since the propagation model includes frequency-dependent effects that induce non-stationarities in the propagating signal, the authors use a time-frequency based approximation technique to analyse the moments. The results show how particular random channel effects increase the variability in the moment features, and thus provide some insight on the possible degradation in classification performance of the moment features.