Introduction

Multiple suppression is one of the largest problems facing the seismic industry. One common technique are the family of approaches generally refered to as `model based' Berryhill and Kim (1986); Wiggins (1988). These methods work by first getting an estimate of the models through downard continuation Berryhill and Kim (1986), computing the first term of the Neuman series Ikelle et al. (1997), or some other method. Next, the primaries are estimated through some type of filtering operation using the estimated multiples. Recently, the problem has been formulated as a signal-noise separation problem in the frequency domain Bednar and Neale (1999); Spitz (1999). These methods operate in the f-x domain with the limiting assumption that the data are time-stationary.

Until recently the signal-noise method proposed by Spitz (1999) could not be formulated in the time domain because it involves dividing by a filter describing the multiple. Claerbout 1998 discovered that multi-dimensional PEFs can be mapped into 1-D, therefore making it possible to do inverse filtering in the time domain. The stationarity assumption inherent in PEF estimation can be overcome by estimating non-stationary filters Crawley et al. (1998). As a result, Spitz's 1999 method can be formulated to work with time domain PEFs Clapp and Brown (1999).

In this paper we show how the time domain formulation of Spitz's approach can effectively attenuate multiples. We apply the method to a 2-D synthetic dataset and show that it is effective in both simple and complex areas. We then apply it on a 2-D real CMP gather. We show that our technique is successful in the attenuating most of the multiple energy with little loss of primary energy.