Conclusion

Three different methods to regularize seismic data with a wavenumber based AMO operator are described. The adjoint implementation is the most efficient for creating a Common Azimuth but inefficient for creating multi-azimuth data. The adjoint approach also shows noticeable amplitude dimming and brightening due to acquisition geometry. Formulating the regularization problem as an inverse problem and then approximating the Hessian with a diagonal operator provides better results. The cost and memory requirements are significantly increased but multi-azimuth data comes for free. Estimating the model with a conjugate gradient solver produces the best results. Amplitude artifacts are virtually eliminated and the frequency content is noticeably improved.