Preconditioned least-squares reverse-time migration using random phase encoding

Discussion and Conclusions

By examining the results using the Marmousi model, we see that encoding the sources in LSRTM is more efficient than the conventional method. This is true despite the fact that each iteration requires an additional forward operator and the stepper is not as efficient.

When estimating the Hessian matrix, encoding both the source and receiver sides is more efficient than encoding the receiver side only. Moreover, the source-based Hessian is more accurate than the conventional source intensity function. As a preconditioner, the source-based Hessian is the most efficient due to its cheaper computational cost compared to estimating both sides of the Hessian matrix. However, using this source-based Hessian in different acquisition geometries needs to be further tested.