CONCLUSIONS

We have described an algorithm for interpolating spatially-aliased seismic traces. This method can interpolate data even if the dip structure of data varies with respect to frequencies. Moreover, it does not require the amplitudes of the wavelets along an event to be constant. And finally, it finds correct prediction filters even when the data are completely aliased. Examples with synthetic and field data confirm that this method is superior to the existing algorithms in handling seriously aliased data.

Regarding its limitations, our algorithm does not, of course, work when data are random or when all events in the data have monochromatic wavelets. However, no algorithm works in these two situations. Therefore, it is to be expected that the results of our interpolation algorithm decrease in accuracy as the input data approach either of these conditions.

A less acceptable problem with our algorithm is its limited ability to handle a large number of dips. In our examples, we used prediction filters of less than 10 coefficients. Extraordinarily long filters may cause failures of the zero-searching routine and the neural net. These two parts of the algorithm need to be improved in the future.