Hi Fred,

Thanks very much for the feedback and log. Very helpful.

For the FWHM AF run at 21:02 everything in the log looks good. Ekos was able to fit Gaussians to a lot of stars at each datapoint and the curve fit looks as expected.

Strangely at 22:07 things are quite different. The reason curvefitting didn't work is that at every datapoint Ekos is only managing to fit a very small number of Gaussians, sometimes 1 or 2. So the data going to curvefitting is very low quality. Curvefitting then gives up because the data it gets isn't very hyperbolic.

So why were so few stars processed? Looks like there were plenty of stars available so I don't think it was something like passing clouds. I have a number of control checks that are designed to throw out poor quality star fits and think something is probably going wrong there.

I'll work on this now.

Thanks John. Also, I can confirm clouds are likely not the problem. I was watching the autofocus process, including the frames from the camera - the skies were clear and the stars visible. Just going outside that evening and looking up provided some confirmation ; ).

Hi Fred,

I've made a tweak. It works on the Sim (but then so did the original parameter settings). I'll let you know when its merged.

John.

Hi Fred,
The change is merged.

Heads up that Hy has integrated Adaptive Focus into Analyse.

In the top Timeline section, anytime Adaptive Focus moves the focuser you'll get a short line in the Focus line. Clicking it will give you detail at the bottom. In the middle section you can select a new focus position metric that will graph the position of the focuser throughout the session as both Autofocus and Adaptive Focus change focus position.


(The Autofocus section in Analyse has been updated to include the graph as well as datapoint.)

This functionality has now been merged.

Hi John - I'm looking forward to testing the changes. The only way I know to download interim updates is to use the nightly build URL (binary-factory.kde.org/job/KStars_Nightly_macos/) but the last successful build was 10 days ago (#1888) and is the one I'm using. Is there another way to get updates that is more reliable? Thanks!

Same, I use the Windows Binary in my environment as well.

You can download the source code and build it yourself. You will need some development / IT skills to do this. If you've not got those skills then this is probably not a goer.

I don't have any experience of the environment that builds the nightlies but I'll ask the other developers if they know why they are broken and can fix the issue.

Some random notes:

• Some config values are not saved, like the focuser step size in μ, also whether weights are used or not - the latter is always swiched off again after a restart of kstars. Also if you change the type of measure. Some don't support weights, and after having selected one of those it's switched off also for those that support it
• Using FWHM instead of HFR: I have the impression it's more noisy, i.e., the error bars are larger. The focus position is correct though
• Using number of stars doesn't work for me. Seems it is clipped at 100? Is that in the code, or a "property" of my scope or target? It doesn't get a nice fit, and continues far out of focus. Strangely enough, it still came up with a reasonable focus..
• Fourier is much slower in processing. I didn't get a really nice curve, also hyperbole clearly is a bad fit function. To me it more looks like a lorentzian...
• I finally realized the yellow 'mustache' is the CFZ

That much for today, time to image

Thanks for the feedback Pit. Num Stars isn't limited to 100 but the underlying star detection (e.g. SEP) might be. If you look at the SEP editor for example, you'll see parameters to limit things like number of stars to keep.

Ah indeed! I had never looked at that page, as star detection had worked OOTB
It was set to finally keep 100 stars. I've now increased that to 180, and will check if/how that improves the procedure.

I've noticed that some parameter settings sometimes get lost. I need to try and track that down.

Measures that support weights are those that perform multiple measurements of what should be the same thing. For example, when Measure=HFR, Ekos will calculate the HFR of all the appropriate stars in the frame and calculate a robust location (think average) and scale (think variance). (It actually will calculate a sigma clipped average and use Rousseeuw and Croux' Sn scale estimator but that's just detail). The weighting for that datapoint is calculated from the scale estimator. This works for HFR, Adjusted HFR and FWHM.

For Num Stars and Fourier there are no multiple measurements so weighting is meaningless so I turn it off and disable the Use Weights checkbox for these Measures to be clear that weighting is not used. If you change Measures and want to use weights, you need to re-check the box.

FYI, I'm working on improving curve fitting (incl Gaussian PSF fitting) so this may make the FWHM "less noisy" - lets see.

Interesting comments on Fourier.