Migration is a convolution. So why not following experiment:
model space m: regularly sampled with 2 or 3 impulses in 2d image. (alternatively, random) two operators: B binning from regular to irregular 2d C convolution of regular with L shape impulse responses
data: d = B C m Two solution approaches: estimate (B C)(-1) what are good preconditioner etc invert binning m' = B(-1) d, then invert convolution m = C(-1) Is one better than the other. More stable, robust? How can you build in velocity analysis.