Channel Characterization and Object Classification in Non-Stationary and Uncertain EnvironmentsThiruneermalai Gomatam, Vikram (2016) Channel Characterization and Object Classification in Non-Stationary and Uncertain Environments. Doctoral Dissertation, University of Pittsburgh. (Unpublished)
AbstractClassification of SONAR targets in underwater environments has long been a challenging problem. These are mainly due to the presence of undesirable effects like dispersion, attenuation and self-noise. Furthermore, we also have to contend with range dependent environments, like the continental shelf/littoral regions, where most of the human and aquatic life's activities occur. Our work consists of analyzing the propagation in these environments from a pulse-evolution perspective. We look at cases where characterizing wave propagation using conventional Fourier-spectral analysis is infeasible for practical applications and instead resort to a phase-space approximation for it. We derive the phase-space approximations for a variety of propagating waves and limiting boundary conditions. We continue our past work on invariant features to enhance classification performance; we simulate the derived features for waves with cylindrical spreading. Another area of our work includes looking at the equation governing the wave propagation from a phase space perspective. It has been shown before that reformulating the classical wave equation in the phase-space provides interesting insights to the solution of the equation. It has been posited that this would be especially useful for non-stationary functions, like the ones governing SONAR propagation underwater. We perform classification of real world SONAR data measured by the JRP ( DRDC-Atlantic, NURC, ARL-PSU, NRL) program. We use a 'classic' MPE classifier on the given non-stationary and contrast its performance with an MPE classifier augmented by a Linear Time Varying (LTV) filter, to assess the impact of adding a time-varying pre-filter to a classifier (MPE) deemed optimal for stationary additive white Gaussian noise. We show that the addition of the time-varying pre-filter to augment the standard MPE classifier does increase the performance of the classifier. Finally, we look at the self-noise problem that is commonly present in the littoral regions of the ocean, which also happens to be the region where most of shallow water sound propagation occurs. We look at phase-space approach to the stochastic models that simulate the effect of signal dependent noise reverberations and attempt to design time-varying estimators that would mitigate the problem at hand. We perform simulations that corroborate our premise. Further directions in the aforementioned areas are also presented. Share
Details
MetricsMonthly Views for the past 3 yearsPlum AnalyticsActions (login required)
|
![[img]](http://d-scholarship.pitt.edu/style/images/fileicons/application_pdf.png)
![[feed]](/images/feed-icon-32x32.png){kind=icon}