Modeling data error during deconvolution

ALGORITHM

Before altering the old algorithm we need to be careful about a couple things. We may need different gain functions for $(d+n)\ast c$ and for $n$ . Sea swell is quite stationary in its physics, but the hyperbolic penalty function applies to the statistical perspective which is one where images are boosted in time from their physical form. We also need to be careful not to mix up $h(n)$ with $h(r=(d+n)\ast c)$ . We will need to scale the regularization with the fitting by experimentation.

We could update the old algorithm (Claerbout et al. (2011)) with the new noise parts. Alternately, we could follow the suggestion of Antoine and switch to the quasi Newton method. In either case we'll need to introduce a scale factor (learned from practice) to choose how much of $D$ ends out in $N$ .