Propagation-invariant classification of sonar signals

Underwater sound waves can undergo frequency‐dependent effects, such as dispersion and damping, that cause the wave to change as it propagates. In active sonar, these changes can be detrimental to automatic classification because the observed backscatter depends on the propagation environment and how far the wave has traveled. We aim to develop signal processing techniques to enable propagation‐invariant classification. Toward that end, we have developed a local phase space view of propagation with dispersion and damping that is physically illuminating and suggests new features for classification. In this talk, we will present moment‐like features extracted from the backscattered wave that are invariant to dispersion and damping, per mode. Environmental knowledge is not needed, beyond knowing the general form of the damping. Two forms of damping are considered. Classification results on simulations of the backscatter from different steel shells propagating in a Pekeris waveguide with damping and random variations will be presented. A dispersion‐invariant matched‐filter receiver approach will also be presented. [Supported by ONR Grant No. N00014‐06‐1‐0009.]