Compressive sensing applied to full-waveform inversion

With the recent resurgence of full-waveform inversion, the computational cost of solving forward modeling problems has become–-aside from issues with non-uniqueness–-one of the major impediments withstanding successful application of this technology to industry-size data volumes. To overcome this impediment, we argue that further improvements in this area will depend on a problem formulation with a computational complexity that is no longer strictly determined by the size of the discretization but by transform-domain sparsity of its solution. In this new paradigm, we bring computational costs in par with our ability to compress seismic data and images. This premise is related to two recent developments. First, there is the new field of compressive sensing (CS in short throughout the paper, Cand‘es et al., 2006; Donoho, 2006)–-where the argument is made, and rigorously proven, that compressible signals can be recovered from severely sub-Nyquist sampling by solving a sparsity promoting program. Second, there is in the seismic community the recent resurgence of simultaneous-source acquisition (Beasley, 2008; Krohn and Neelamani, 2008; Herrmann et al., 2009; Berkhout, 2008; Neelamani et al., 2008), and continuing efforts to reduce the cost of seismic modeling, imaging, and inversion through phase encoding of simultaneous sources (Morton and Ober, 1998; Romero et al., 2000; Krohn and Neelamani, 2008; Herrmann et al., 2009), removal of subsets of angular frequencies (Sirgue and Pratt, 2004; Mulder and Plessix, 2004; Lin et al., 2008) or plane waves (Vigh and Starr, 2008). By using CS principles, we remove sub-sampling interferences asocciated with these approaches through a combination of exploiting transform-domain sparsity, properties of certain sub-sampling schemes, and the existence of sparsity promoting solvers.

https://slim.gatech.edu/Publications/Public/Conferences/EAGE/2009/Herrma...