Just wanted to record a few things the focus routine may be improved with.

1. Setting for exposure time per filter for focusing. Useful for narrowband versus broadband.
2. Option to median/mean the entire frame hfr versus using a single star. (Could be tricky, but could also be useful for certain scopes).

Just wanted to make note of these. #1 seems the easiest, and likely the most beneficial.

I do wonder about option 2 a lot. I find it very impressive in SGP, but I can't really say it is more accurate per-se. I guess having an average of all stars is a bit less prone to errors given by a star that happens to be troublesome.

1. In Ekos, you select a "Focus Filter" which would be used for all focusing. Been working for me for the last 3 years without any issues and I only image in narrowband, but I select my "Focus Filter" to Luminescence/Clear
2. Not practical, at least for me, since a full frame download takes a LONG time. QSI takes 20 seconds to download, multiply that by the number of iterations needed to finish autofocus and then you'd be wasting 1/3 of the night time in "Focus" mode.

For #1, the critical focus zone (CFZ) changes when filters are changed. While the filters themselves are parfocal in some cases (like AstroDon for example) this does not mean that the CFZ is at the same place. This only means that the filters themselves are not changing the CFZ on their own.

Here is a good PDF written by Don Goldman on CFZ and Parfocal filters:
www.optcorp.com/pdf/Astrodon/parfocal.pdf

Your current narrowband images would likely improve from a change in the focusing logic, as you can see by reading that doc.

For #2 - It is not practical for me either, but I have heard that there are certain optical trains out there that would benefit from this in some manner. I like the method used today though, sans the CFZ case study above.

In Ekos, you can also:

1. Autofocus on Filter change, OR
2. Change relative focus offset on filter change

So both of these options should cover the filter-change-focus-issue

EDIT: Ok I just read the paper you linked (quickly). So still the proposed solutions above would account for this since you can auto-focus on filter change, or if you're already done your measurements, then simple change focus by a fixed relative offset. I think what I can do is that I can associate a filter with exposure duration.

I will have to explore the offset option, I wasn't aware it was there. That would likely solve the problem as the offset would be a constant, and I could determine that offset rather easily with a mask.

I do have the focus when filter change option enabled.

I'm not sure if option #2 is that problematic, I have pretty fast success with SGP doing it like that. Granted it takes 10 sec per picture, but I'm not focusing all the time, just once or twice in a night. I do notice a big difference in my routine for EKOS and in SGP, maybe it's because of settings, but in EKOS I can't get a nice V-curve, focusing outward looks ok, inwards it's staying rather flat until it rises. In SGP I get a very nice symmetrical curve... not sure what the difference really is, but I now use SGP for focusing.

Ekos never constructs a full V-curve, it iteratively steps until it crosses to the other side of the V-curve and then backtracks to find the solution.

Yes I did notice that, but even then, when it takes the first big steps, the points are pretty off in comparison to SGP. I wonder if it's due to my optical train, but it looks like it's not the best method for it.

Pretty off in what sense?

Well, in SGP the star-sizes go up in predictable ways on both focus in and out, getting a symmetrical V-curve. In EKOS I get sort of a nice straight steep slope on 1 side, on the other side the slope is much lower as if the focuser isn't getting out on that side with the same amount as in. True, EKOS doesn't do an entire V-curve, but still the points on 1 side should be rather similar as on the other.

Have you tried building a full V-Curve in Ekos yourself? Just use a fixed step-size and focus in/out and repeat until you see the full V-curve. I will try it next time with an actual star and post the results.