Deblending using normal moveout and median filtering in common-midpoint gathers

Introduction

The technique of simultaneous-source (sometimes called multisource) acquisition means firing more than one shot at nearly the same time regardless of their interference. In conventional acquisition, however, either the temporal shooting intervals or the spatial sampling intervals are large enough so that the interference between successive shots can be ignored. The multisource technique can reduce the acquisition period and at the same time can improve data quality because of the decreased spatial sampling interval (Berkhout, 2008). The benefits from simultaneous-source acquisition are compromised by the challenges in removing strong blending interference. Because of its economic benefits and technical challenges, this technique has attracted the attention of researchers in both industry and academia (Huo et al., 2012; Mahdad et al., 2011).

In blended acquisition, more than one source is shot simultaneously, regardless of their interactive interference. The term source denotes a shot array, which can contain all the shots in a conventional acquisition system. When more than one source is involved in acquisition, either a denser or a wider shot coverage can be obtained for a given constant acquisition period. Figure 1a depicts two simultaneous sources shooting from the same position towards the same direction. In this case, a two-times denser coverage can be obtained. Figure 1b depicts two simultaneous sources far from each other shooting along the same direction, in which case, a two-times wider shot coverage can be obtained. The number of simultaneous sources can be even larger, yielding a much denser and wider shot coverage. The shooting sequence of the shots and the direction of each source can also be variable. It's natural, that the observed data will contain strong interference, and the more simultaneous sources involved, the more severe the interference will be. For the blended acquisition geometry as shown in Figure 1b we acquire blended data with strong interference, as shown in Figure 2b. Figure 2a shows the acquired data using conventional acquisition, supposing that the two sources in Figure 1b are fired with large-enough time interval.

demo1,demo2
Figure 1. Demonstration for denser shot coverage (a) and wider shot coverage (b). Red points denote shot positions for source 1. Green points denote shot positions for source 2. Blue points denote receiver positions. Red and green strings denote the shooting rays. Arrows denote the shooting directions.

datademo,blenddemo
Figure 2. (a) Unblended data. (b) Blended data.

There are two main ways to deal with the challenges posed by simultaneous-source acquisition. The first is to use a first-separate and second-process strategy (Chen et al., 2013), which is also known as "deblending" (Doulgeris et al., 2012). The other is to use direct imaging and waveform inversion by applying some constraints to attenuate the artifacts caused by interference (Verschuur and Berkhout, 2011; Dai et al., 2012). Although the direct imaging approach has achieved some encouraging results, the preferable way so far is still to focus on the separation of blended data into individual sources as if acquired conventionally.

Different filtering and inversion methods have been used previously to deblend seismic data. Filtering methods utilize the property that the coherency of the simultaneous-source data is not the same in different domains, thus we can get the unblended data by filtering out the randomly distributed blending noise in a particular domain, in which one source record is coherent and the other is not (Hampson et al., 2008; Huo et al., 2012; Mahdad et al., 2012). One choice is to transform seismic data from the common-shot domain to common-receiver, common-offset or common-midpoint domain. Inversion methods treat the separation problem as an estimation problem that aims at estimating the desired unblended data. Because of the ill-posed property of such estimation problems, a regularization term is usually required (Doulgeris and Bube, 2012). Moore et al. (2008), Akerberg et al. (2008) and Moore (2010) use a sparsity constraint in the Radon domain to regularize the inversion. A sparsity constraint is also used by Abma et al. (2010) to minimize the energy of incoherent events present in the blended data. Bagaini et al. (2012) compared two separation techniques for dithered slip-sweep (DSS) data using the sparse inversion method Moore (2010) and $f$ -$x$ predictive filtering (Chen and Ma, 2014; Canales, 1984), and found the advantage of inversion methods over random noise attenuation techniques. van Borselen et al. (2012) proposed to distribute all energy in the simultaneous shot records by reconstructing the individual shot records at their respective locations. Mahdad et al. (2012) introduced an iterative estimation and subtraction scheme that combines the properties of filtering and inversion methods and exploits the fact that characteristics of blending noise differs in different domains. In order to deal with the aliasing problem, Beasley et al. (2012) proposed the alternating projection method (APM), which chooses corrective projections to exploit data characteristics and claims to be less sensitive to aliasing than other approaches.

Median filtering is notable for its ability to remove spiky noise and is also suitable to remove blending noise. However, median filtering can be applied only to seismic profiles containing horizontal events, otherwise it will harm much of the useful energy. In this paper, we propose to implement median filtering to attenuate blending noise after normal moveout in common-midpoint (CMP) gathers. The deblending can be inserted into a conventional processing workflow . The benefits of the proposed deblending approach are its easy implementation and efficient improvement for the migrated image without use of iterative deblending, which is extremely time-consuming. Synthetic and field data examples demonstrate the effectiveness of the deblending method and improvement for the final migrated image.

Deblending using normal moveout and median filtering in common-midpoint gathers