Implementation of algorithms for extracting tonal components in underwater noise

Abstract

Underwater target classification and tracking problems utilize the noise signals emanating from the targets as well as the target dynamics features. Target classification is carried out by extracting certain classification clues about the target such as its emission frequencies, and other target specific features from the self noise or active transmissions. The self noise generated by the targets being nonstationary in nature are to be analyzed with short term averaged data segments for extracting the tonal components using various spectral estimation techniques. Though classical spectral analysis techniques are computationally efficient, it suffers from several inherent limitations such as frequency resolution, performance degradation due to implicit windowing of the data, etc. In an attempt to improve the spectral resolution, several modern spectral estimation techniques, which utilize the procedures of indirect Fourier analysis by fitting the measured short data segments to an assumed model, have been evolved. This paper presents the development of procedures for the estimation of power spectral densities using modern techniques based on parameter estimation such as auto regressive, moving average, auto regressive moving average, maximum likelihood, etc. The performance of the estimator has been validated by computing the emission frequencies of a 50-foot vessel.