Non-Gaussian Likelihood Detectors for Broadband Active Sonar

"In shallow water environments or using broadband processing, sonar signals can exhibit highly non-Gaussian noise statistics due to the discrete nature of the background returns from different clutter elements in different range/bearing resolution cells.Optimal likelihood detectors rely crucially on accurate noise statistics estimation, and poor fits to an assumed Gaussian PDF can lead to increased false alarm rates when the true PDF is heavy tailed. We will implement both empirical and physics-basedmodels of non-Gaussian clutter statistics, in order in improved likelihood detectors. For example, physics-based forms for bottom reverberation statistics may be derived from assumed forms for the scale-dependent surface roughness spectra, which alsoprovide models for statistical correlation between different range cells. SIRV or Gaussian mixture model parameterizations will be used to construct rapidly computable analytic or semi-analytic forms for the PDF. Our approach smoothly interpolates betweenthe low frequency regime (100 Hz to 1000 Hz) to mid and high frequency ranges. This is because our acoustic models have natural low, middle and high frequency implementations, and because of the adaptive nature of our models for the clutter returns. Wealso propose to develop criteria to optimize active waveforms for improved suppression of clutter interference. The improved detection performance of broadband active sonar systems equipped with a