TIME-FREQUENCY APPROXIMATION AND FEATURE EXTRACTION \ud FOR RANGE-DEPENDENT UNDERWATER SOUND PROPAGATION

Sonar systems are used in localization, detection and classification of various objects in marine environments. Unlike the propagation of sound in air, which is largely unaffected by dispersion, the underwater channel can be highly dispersive, especially in shallow water environments. Such channels also introduce other significant propagation effects, including multipath and frequency-dependent energy attenuation due to interactions of the sound with the ocean surface and bottom. Compensating for these propagation effects is important with regard to classification of underwater objects based on their sonar backscatter, as the target “signature” will be different at different locations. Previous work in our lab has developed feature extraction methods for dispersion-invariant classification, and approximation methods to solve for dispersive propagation, in range-independent environments. Such environments, wherein the channel characteristics do not change with propagation distance, represent an idealistic assumption that generally does not hold for long-range propagation in underwater channels.\ud \ud In this work we concentrate on range-dependent guided wave propagation. We begin with an examination of the classification performance of the previously developed range-independent features in a range-dependent model, namely an ideal wedge waveguide. Motivated by the degradation in classification performance of these features, we derive new features that mitigate the range-dependent dispersion effects and show that the derived features outperform range-independent features in a wedge waveguide. We also derive the approximate Wigner distribution for a pulse propagating in this range-dependent environment, and highlight similarities and differences of this new result with a previously developed range-independent approximation. This approximation can be a useful tool for estimating the evolution of pulse propagating in a\ud range-dependent channel. \ud \ud Finally, we explore a second range-dependent model, namely the Parabolic Equation, which can be adapted to a wide array of propagation environments and media. We derive features that are invariant to dispersion and attenuation from this model