ABSTRACT
Forecasting ambient noise levels in the ocean can be a useful way of characterizing the detection performance of sonar systems and projecting bounds on performance into the near future. The assertion is that noise forecasting can be improved with a priori knowledge of source positions coupled with the ability to resolve closely separated sources in bearing. One example of such a system is the large aperture research array located at the South Florida Test Facility. Given radar and Automatic Identification System defined source positions and environmental information, transmission loss (TL) is computed from known source positions to the array. Source levels (SLs) of individual ships are then estimated from computed TL and the pre-determined beam response of the array using a non-negative least squares algorithm. Ambient noise forecasts are formed by projecting the estimated SLs along known ship tracks. Ambient noise forecast estimates are compared to measured beam level data and mean-squared error is computed. A mean squared error as low as 3.5 dB is demonstrated in 30 min forecast estimates when compared to ground truth.
ABSTRACT
This paper presents an evaluation of the classical model for determining an ensemble of the broadband source spectra of the sound generated by individual ships and proposes an alternate model to overcome the deficiencies in the classical model. The classical model, proposed by Ross [Mechanics of Underwater Noise (Pergamon, New York, 1976)] postulates that the source spectrum for an individual ship is proportional to a baseline spectrum with the constant of proportionality determined by a power-law relationship on the ship speed and length. The model evaluation, conducted on an ensemble of 54 source spectra over a 30-1200-Hz to 1200-Hz frequency band, shows that this assumption yields large rms errors in the broadband source level for the individual ships and significantly overestimates the variability in the source level across the ensemble of source spectra. These deficiencies are a consequence of the negligible correlation between the source level and the ship speed and the source level and the ship length. The alternate model proposed here represents the individual ship spectra by a modified rational spectrum where the poles and zeros are restricted to the real axis and the exponents of the terms are not restricted to integer values. An evaluation of this model on the source spectra ensemble indicates that the rms errors are significantly less than those obtained with any model where the frequency dependence is represented by a single baseline spectrum. Furthermore, at high frequencies (400 to 1200 Hz), a single-term rational spectrum model is sufficient to describe the frequency dependence and, at the low frequencies (30 to 400 Hz), there is only a modest reduction in the rms error for a higher order model. Finally, a joint probability density on the two parameters of the single term model based on the measured histograms of these parameters is proposed. This probability density provides a mechanism for generating an ensemble of ship spectra.
