A hybrid second- and higher-order statistical method for blind deconvolution of seismic data

Successful deconvolution of the seismic traces in exploration seismology relies on accurate wavelet estimation. We model the wavelet as the impulse response of a pole-zero model and allow it to be non-casual and/or non-minimum phase. We first employ second-order statistics of the data to construct spectrally equivalent models and then find the “best” model according to a fourth-order cumulant matching criterion. Both the lower variance property of the second-order statistics and the ability of the higher-order statistics to distinguish spectrally equivalent models are taken advantage of. Numerical examples illustrate the key points of the paper