Propagation-Invariant Classification of Signals in Channels with Dispersion and Damping

We consider the classification utility of time- domain moment-like features of a wave that are invariant to dispersion and damping. Simulations of the sonar backscatter from metallic shells in a dispersive, damped waveguide are conducted for single mode and multi-mode propagation. Random variations in the waveguide parameters are also investigated. Results show our invariant moment features have greater discriminability, as quantified by ROC curves, at low false alarm rates when compared to ordinary temporal moments, and the invariant features perform best for single mode propagation.