Statistical wavelet filtering for impulsive noise mitigation

Impulsive noise sources in acoustic data can represent a significant amount of contamination that is not easily addressed by spectral methods alone. In ocean acoustics, such noise sources can be biological in nature, e.g., marine mammal localization clicks, or even mechanical noise induced on an instrument itself. While the broadband character of an impulse does not lend itself to frequency domain filtering, the wavelet domain is more appropriate and has been used to analyze impulsive sources of interest in the past. We present a method for filtering unknown nuisance impulses using basic statistics of the continuous wavelet transform (CWT), with minimal disturbance to the remaining signal(s) of interest. Examples from both simulation and collected ocean acoustic data are presented.