An Improved Swell Noise Attenuation Method Using Statistics

Using the characteristics of low frequency and high-amplitude, conventional swell noise attenuation methods basically identify the swell noise in the f-x domain according to a user-defined amplitude threshold. As it is difficult for the user to estimate the trace-and-frequency-dependent amplitude thresholds, this method cannot fully distinguish the variable swell noise from the signal. Based on a statistical technique, Bekara et al. gives a new swell noise attenuation method. Although this method can automatically estimate the trace-and-frequency-dependent amplitude thresholds, it doesn't work well when the condition of high amplitude swell noise is not satisfied. In this paper, we introduce the correlation of the signal and the randomness of the swell noise into swell noise identification. Using the correlation and randomness as weighting coefficients, our method can accurately attenuate swell noise without compromising the signal through a weighted expectation-maximization algorithm.