Time-domain sparsity-promoting least-squares migration with source estimation

Traditional reverse-time migration (RTM) gives images with wrong amplitudes and low resolution. Least-squares RTM (LS-RTM) on the other hand, is capable of obtaining true-amplitude images as solutions of $\ell_2$-norm norm minimization problems by fitting the synthetic and observed reflection data. The shortcoming of this approach is that solutions of these $\ell_2$ problems are typically smoothed, tend to be overfitted, and computationally too expensive because it requires compared to standard RTM too many iterations. By working with randomized subsets of data only, the computational costs of LS-RTM can be brought down to an acceptable level while producing artifact-free high-resolution images without overfitting the data. While initial results of these "compressive imaging" methods were encouraging various open issues remain including guaranteed convergence, algorithmic complexity of the solver, and lack of on-the-fly source estimation for LS-RTMs with wave-equation solvers based on time-stepping. By including on-the-fly source-time function estimation into the method of Linearized Bregman (LB), on which we reported before, we tackle all these issues resulting in a easy-to-implement algorithm that offers flexibility in the trade-off between the number of iterations and the number of wave-equation solves per iteration for a fixed total number of wave-equation solves. Application of our algorithm on a 2D synthetic shows that we are able to obtain high-resolution images, including accurate estimates of the wavelet, for a single pass through the data. The produced image, which is by virtue of the inversion deconvolved with respect to the wavelet, is roughly of the same quality as the image obtained given the correct source function.

https://slim.gatech.edu/Publications/Public/Conferences/SEG/2016/yang201...