Probability density function of ocean noise based on a variational Bayesian Gaussian mixture model.

Area of Science:

• Acoustics
• Signal Processing
• Statistical Modeling

Background:

• Ocean noise exhibits non-Gaussian characteristics, deviating from standard Gaussian distributions.
• This non-Gaussianity significantly affects the performance of sonar signal processing techniques.
• Accurate modeling of ocean noise amplitude distributions is crucial for effective sonar applications.

Purpose of the Study:

• To model the amplitude distribution of various ocean noise types using a Bayesian Gaussian mixture model (BGMM).
• To evaluate the effectiveness of BGMM in approximating complex, non-Gaussian noise distributions.
• To assess the goodness of fit for modeled noise probability density functions (PDFs).

Main Methods:

• Application of a Bayesian Gaussian mixture model (BGMM) for noise amplitude distribution modeling.
• Utilizing variational inference for the BGMM learning algorithm.
• Employing the one-sample Kolmogorov-Smirnov test to examine the goodness of fit for modeled PDFs.

Main Results:

• Ship noise and low-frequency ambient/typhoon noise in the South China Sea were found to be significantly non-Gaussian.
• High-frequency noise was observed to be closer to a Gaussian distribution.
• BGMM demonstrated a superior goodness of fit compared to standard Gaussian or Gaussian mixture models.
• Small mixing coefficient components in BGMM were vital for accurately describing non-Gaussian PDFs.

Conclusions:

• The Bayesian Gaussian mixture model (BGMM) provides a robust method for modeling complex non-Gaussian ocean noise.
• Accurate modeling of non-Gaussian noise is essential for improving sonar signal processing.
• BGMM's ability to automatically select components enhances its practical applicability in diverse acoustic environments.