Multiple reflection noise attenuation from seismic data using wavelet domain noise analysis algorithm

The reflection waves, which is reflected between the subsurface or free surface reflectors more than once before being received on the receivers, are called multiple reflections. Multiple reflections, often destructively interact with the primary reflections and reduce the quality of the seismic image. An inverse filter based on predictive deconvolution using the periodic feature is used to attenuate multiple reflections in the water. Multiple and primary reflections show different moveout and travel-times, This property is the basis of the theory of many multiple attenuation techniques such as CMP stacking, F-K filter, and Radon transform. Radon transform was first introduced by Johann Radon (1917) and for the first time, parabolic Radon conversion was used as a multiple attenuation technique by Hampson (1986). Since then, the Radon transform became one of the most widely used tools to suppress multiple noises. Goudarzi and Riahi (2013) presented WDGA method based on the data type, as an efficient way of attenuating various seismic noises. However, this approach, if there is a high level of random noise in the data, cannot well separate the coherent noise from the reflections. Here we try to introduce a new method to solve this problem. 

Methodology and Approaches:

The proposed method in this research is called wavelet domain noise analysis (WDNA) algorithm. Similar to WDGA, this method is based on data, but because of the use of the split Bergman iteration is less sensitive to random noise. It also reduces the time to reach an optimal solution and it has better convergence. These features enable better detection of the desired noise and better signal separation from the noise. The goal of this research is to apply the benefits of Radon transform, and at the same time, to use the DT-RADWT wavelet transform capabilities to provide high resolution. We take advantage of the split Bergman iterative algorithm to build a full multiple reflection model from initial multiple models (achieved from Radon filter). Finally, in the DT-RADWT domine, full model of multiple reflections would be subtracted from the input data, and thus, the filtered data would be obtained.

In this research, the WDNA algorithm has been introduced and its application in attenuating multiple reflections from seismic data has been investigated. The WDNA algorithm is based on the data and requires an initial noise model that is obtained from Radon transform (or any other suitable filter) to attenuate multiple reflections and in the dual-tree wavelet transform domain, it is used to produce a complete noise model with the Bergman iteration algorithm. Subtracting the full noise model from seismic input data yields almost no multiple reflection noise and the initial reflections are well maintained. The use of the DT-RADWT wavelet transform increases the frequency resolution and split Bergman algorithm helps to achieve a fast convergent solution that also causes insensitivity with random noise in the attenuation process of multiple reflections. The results of applying the WDNA method on synthetic and real data have resulted in better outputs than Radon and WDGA.