next up previous [pdf]

Next: Acknowledgments Up: Leader and Almomin: Encoding Previous: Phase Encoding With Random


Both phase encoding and random boundary propagation can be very effective in accelerating linearised inversion. We presented the individual benefits and how combining these methods can lead to a powerful inverse scheme with reference to inverse imaging on GPUs. We can conclude that these techniques can be used together, and acceptable images are obtained within the cost of a conventional RTM migration. However, to improve convergence we can see that a mute must be applied to avoid direct-arrival type artefacts in the Born-modelled data, and that we can improve convergence again by stacking separate gradient realisations per iteration. For future work we will explore the best method of cleaning the gradients between iterations and will update the non-linear solver to see if we can improve our convergence rate once we are within 50% data error.