I have repeatedly run into a focusing challenge with my C8: in the midst of a sequence, for some reason the focuser moves a certain distance away from focus, and settles on a doughnut pattern. This then stays there for several focusing sessions, and some (random) time later the scope returns to correct focus. So it is not just one, but 2-4 focusing attempts that settles on a doughnut - and then suddenly, not - back to pinpoint stars.
I really have no idea why the focuser moves from pinpoint stars to doughnuts in the first place - but it could be due to passing clouds or other intermittent stuff. However, the focus routine seems to accept the doughnuts more than I like. So my questions: are there strategies for handling this? TO minimize the risk that the focuser does not recognize the doughnut as a doughnut, settling for that as focus? Larger box size, more aggressive settings somewhere, or different algoritm? Or full field?
I'm using a DSLR: Canon 1000D. I've been experimenting with using darks in the focus routine, but it seems that it actually is MORE reliable without darks (easier identification of stars). Scope is C8 with reducer F 6.3.
I've observed somewhat similar behavior. If for whatever reason the first three (or so) measurements give very similar values without a good gradient it often starts doing an insanely big focus change. Fortunately (for me) with a refractor there's no doughnut, and the algorithm usually finds it's way back.
My guess would be the initial issue is the long focus jump, likely initiated by the polynomial extrapolation. Have you tried the other/old-fashioned algorithm?
IMHO some way to limit the (single) focus steps would be a nice option.
Hi!
No, I haven't yet tried other algoritms, as this is quite unpredictable, that is, I cannot force it to happen, and the polynominal seems to generally give me fastest and best results. I do also believe that the issue here is more that the focus mechanism does not recognize the doughnut as a doughnut... in that case, it would be as with your and find its way back - at least that is my thinking for the moment.
This is a known issue reported by several users. The current algorithm is fooled by doughnuts .. can you share several FITS images of the doughnuts at various focuses? I can then perhaps improve the algorithm.
Perhaps some machine learning as well? This is an interesting case for it.
If you can, please zip all the FITS files and share them.
I can easily produce fits-images of doughnuts. You mean from focuser, that is, short exposures, and at varying focuser positions, so the doughnut grows bigger and bigger? I can do that next time the clouds part...
Thanks! So I just tested so the default algorithm (Gradient) was able to identify correctly all the stars you sent. I think it would more useful if you turn Focus logging ON and then send me the logs if you see this issue so I can investigate it fully. Maybe you can also limit "Max Travel"?