Blind 2D Convolved Mixtures Separation of Primaries and Multiples with Non-Gaussian Filters

Adaptive multiple subtraction based on blind 2D convolved mixtures separation exploits non-Gaussian maximization constraint on primaries to estimate the 2D matching filter. The 2D filter looks like a dip filter. However, there exist spurious or noisy filter impulse causing residual multiples or distorted primaries. Given the filter’s coefficients satisfy non-Gaussian distribution, we introduce the non-Gaussian maximization constraint of filters into the blind 2D convolved mixtures separation based method. The iterative reweighted least squares (IRLS) algorithm is used to estimate the 2D matching filter. Since the proposed method estimates the 2D matching filter with less noisy filter impulse, it can better remove residual multiples and preserve primaries compared with the blind 2D convolved mixtures separation based method and traditional least squares method.