Geophysical Prospecting, 2013, 61 (Suppl. 1), 1–9 doi: 10.1111/j.1365-2478.2012.01090.x Coherent and random noise attenuation via multichannel singular spectrum analysis in the randomized domain Stephen K. Chiu ∗ ConocoPhillips, Houston, Texas, United States Received November 2011, revision accepted March 2012 ABSTRACT The attenuation of coherent and random noise still poses technical challenges in seismic data processing, especially in onshore environments. Multichannel Singular Spectrum Analysis (MSSA) is an existing and effective technique for random-noise reduction. By incorporating a randomizing operator into MSSA, this modification creates a new and powerful filtering method that can attenuate both coherent and random noise simultaneously. The key of the randomizing operator exploits the fact that primary events after NMO are relatively horizontal. The randomizing operator randomly rearranges the order of input data and reorganizes coherent noise into incoherent noise but has a minimal effect on nearly horizontal primary reflections. The randomizing process enables MSSA to suppress both coherent and random noise simultaneously. This new filter, MSSARD (MSSA in the randomized domain) also resembles a combination of eigenimage and Cadzow filters. I start with a synthetic data set to illustrate the basic concept and apply MSSARD filtering on a 3D cross- spread data set that was severely contaminated with ground roll and scattered noise. MSSARD filtering gives superior results when compared with a conventional 3D f-k filter. For a random-noise example, the application of MSSARD filtering on time- migrated offset-vector-tile (OVT) gathers also produces images with higher signal-to- noise ratios than a conventional f-xy deconvolution filter. Key words: Randomizing operator, Eigenimage, Deconvolution INTRODUCTION Coherent noise often coexists with random noise in seismic field data. Noise attenuation plays a key role in data pro- cessing to enhance the signal content of the data. The char- acteristics of coherent noise are considerably different than random noise. A general approach requires designing filtering techniques based upon the characteristics of noise and applies noise suppression to target each noise type separately. This is a standard denoising procedure in the industry. However, it is desirable to have a filtering method that can attenuate both coherent and random noise simultaneously. The main objec- ∗ E-mail: Stephen.K.Chiu@conocophillips.com tive of this paper is to introduce a new rank-based-reduction denoising algorithm to perform both filtering operations con- currently. Numerous filtering methods have been developed to sup- press various types of coherent noise; however, coherent- noise attenuation is still an active research area. For exam- ple, the suppression of surface-wave noise including ground roll and scattered noise is a difficult problem. Ground roll and near-surface scattered energy are the most troublesome forms of coherent noise. Ground roll is characterized by low- velocity events with high amplitude and dispersive low fre- quencies. Near-surface scattered energy often exhibits compli- cated diffraction patterns associated with the spatial positions of point scatters. Strong coherent noise often overwhelms and obscures primary reflections. Classical ‘global filtering’ C  2012 European Association of Geoscientists & Engineers 1