Long period multiple suppression by predictive deconvolution in the x-t domainl1

There are two forms of systematic error in conventional deconvolution as applied to the problem of suppressing multiples with periodicities longer than a hundred milliseconds. One of these is the windowing effect due to the assumption that a true autocorrelation function can be computed from a finite portion of data. The second form of error concerns the assumption of periodicity, which is strictly true only at zero offset for a 1D medium. The seriousness of these errors increased with the lengthening of the multiple period. This paper describes and illustrates a rigorous 2D solution to the predictive deconvolution equations that overcomes both of the systematic errors of conventional 1D approaches. This method is applicable to both the simple or trapped system and to the complex or peg-leg system of multiples. It does not require that the design window be six to ten times larger compared to the operator dimensions and it is accurate over a wide range of propagation angles. The formulation is kept strictly in the sense of the classical theory of prediction. The solution of normal equations are obtained by a modified conjugate gradient method of solution developed by Koehler. In this algorithm, the normal equations are not modified by the autocorrelation approximation. As with all linear methods, approximate stationary attitude in the multiple generating process is assumed. This method has not been tested in areas where large changes in the characteristic of the multiple-generating mechanism occur within a seismic spread length.