I did a first implementation following ATI's white paper around a year ago. The results ended up looking pretty good, as you can see on my old engine's page. I had troubles (and so do many people, it seems, from what i saw on some forums) to get a nice daylight blue sky. So i reviewed my old code and compared with the formulas found on white papers on the net.
I am completely amazed.
I didn't implement a single formula correctly.
The exponent exp(x) in the extinction term has been replaced "magically" by a pow(2, x). The units are not correct. The extinction term is modulated by the sun color, while it should actually be in the in-scattering term. Even the angle functions are not correct: a 1 + g^2 - 2.g.cos(thetha) had mysteriously been transformed into a 2.g.cos(thetha) + 1 + g.
I will try to reimplement everything correctly this week.