Conclusions

I emphasize that the mismatch between the limited number of live hydrophones in a marine-streamer survey and the pervasive number of live traces generated by modeling is essentially a form of crosstalk in multisource migration/inversion not seen in a fixed spread survey. To completely remove this marine crosstalk I propose a frequency selection encoding scheme similar to the ones used in the communications industry. This scheme demands that any crosstalk-prone sources should only emit signals in non-overlapping frequency bands. This scheme allows any receiver to selectively tune in to the valid source (the source that indeed has made a contribution to this receiver) and to selectively disregard potentially confounding sources. Such sources are grouped with the receiver only at the time of multisource modeling.

I show that my frequency selection technique in a multisource framework is similar in form to the well-studied stochastic optimization problem. In terms of computational cost, blending groups of $S$ sources together to form supergathers would reduce the subsequent computational cost by a factor of $S$ . Due to the weakened illumination capability, iterations are usually required to produce an image comparable to standard migration. Fast convergence with many fewer steps than $S$ yields an overall speed gain compared to conventional migration.

Numerical results with a 2D salt model and a marine survey show that crosstalk is completely removed with a multisource speedup nearly an order of magnitude faster than standard migration. In the 3D example with a fixed acquisition geometry (a fixed OBS geometry) a speed up of 40 was achieved compared to standard migration. In addition, better resolution was achieved.