ASA 125th Meeting Ottawa 1993 May

5aUW9. Gaussian mixture statistical model of arctic noise.

R. Myers

B. Sotirin

Ocean and Atmospheric Sci. Div., Code 541, NRaD, San Diego, CA 92152-5000

Arctic ambient noise is often stationary and Gaussian for significant periods of time. However, during periods of local ice activity, the ambient noise can become highly non-Gaussian, nonstationary, and exhibit substantial variability in spectral shape. During these periods, the task of characterizing the statistics of the ambient noise is greatly complicated. In many cases, however, the distribution of the noise during the periods of intense ice activity can be modeled as a two-state Gaussian mixture, where the mixture parameters are estimated from the data by searching for the maximum of the likelihood function. A statistical analysis of selected ambient noise collected over a 10-day period during which long-range ambient noise was interspersed with intense, brief periods of local ice noise indicates that the ambient noise was highly Gaussian except during the periods of ice activity. During the periods of ice activity, the distribution of the ambient noise was modeled quite well as a two-state Gaussian mixture.