An improvement of blind deconvolution based on MIR to the nonstationary seismic data

An approach is proposed to improve the resolution of nonstationary seismic data by decomposing the data into segmentations which can be seen as quasi-stationary. In 2006, Anhony Larue et al., proposed a new blind deconvolution method based on the minimization of the mutual information rate (MIR). This algorithm can estimate any filter, minimum or not, and provide good results with better tradeoff between deconvolution quantity and noise amplification than existing methods. However, the method is relied on the hypothesis that the input record is stationary. Besides, it requires estimation of the signal probability density function (PDF) and score function which need large sample in the given method. Those restrict its application to the real seismic data. In this paper, we decompose the data into quasi-stationary segments according to its statistical properties: empirical distribution and entropy. In order to calculate the score function effectively based on small sample, generalized Gaussian distribution (GGD) is introduced. Subsequently, in each segment, respectively, a high-resolution result can be obtained after blind deconvolution based on MIR. Applications of these improvements to both synthetic and real data show that the proposed method works well for a general earth Q-model that varies with travel time, and can expand the frequency band of the nonstationary seismic trace effectively.