When using the linear autofocus routine, it does a really good job at calculating the best focus position, but then when the routine is dine, it chooses a position that is worse than the calculated one,  In fact, when using a long focal length scope, it seems to find the best focus position quickly, in about 7 or 8 exposures.   But then it keeps adding more and moire exposures - around 25 or 30 - which never ends well.  If EKOS had stopped after 10 exposures and just set the focuser to the best calculated position, it would have been fast and accurate.    But instead it spends much more time and fails to find a better solution.  This is a big issue with long focal lengths where the exposures need to be 7 to 10 seconds long.  It makes focusing take 10 minutes or longer, which is not ideal where I live as the temperature alway drops 50F (10C) every night.

Additionally I wish we could adjust some kind of  "resolution" of the focuser.  With the Esatto focuser, there are 240,000 steps!  But at 2,000mm focal length, you really cant see any difference between 160,000 and 160,500.  But EKOS tries to fine tune the focus back forth between 160,500 and 160,300 etc.  It's a waste of precious time. 160,000 to 160,400 are the same.  

But just solving the main issue - and having EKOS choose the already calculated best-focus position would be huge time saver.

Here are two screen shots from last night.  

The first shows the calculated focus position of 170,000.   But EKOS selected 172500 instead!  The HFR value was 1.25 arc seconds.

 

The second image shows that when I manually adjusted the focuser back to the calculated position, the HFR improved to 1.16.  The calculated position is almost always better than the one finally selected by EKOS.  It seems like an easy fix to just choose the already calculated focus when stopping the focus routine.  It's what I end up doing anyway manually, but it means I can't use autofocus unattended.  

 

 

Peter,

I'm happy to work with you and help figure this out,  but I'm not sure there is a simple solution as you suggest. The philosophy behind the Linear Focus scheme is that the focusing system's uncorrected backlash is comparable or greater than the critical focus zone and because of that the reported focuser positions are unreliable. That is, the assumption is that you can't just "go to a position", especially when changing direction, as you would be doing. If you could reliably go to a position, then, sure, after computing the V-curve, it could just go to the minimum and be done. Since it can't, what it does is (a) compute a V-curve, then (b) try and repeat a shortened procedure with smaller steps and stop when it is within tolerance% of he minimum HFR of the original v-curve (at whatever position the focuser is reporting). 

If you think your system has no backlash, or that it can reliably go to a position somehow, then you should use the Polynomial Focus algorithm instead and it would work as well or better than Linear and be much quicker. However, if you have some backlash, I'd expect Linear to do a much better job than Polynomial. Why don't you compare Linear and Polynomial and let me know how it goes?

WRT step size, you do have control over the step size. See the Mechanics sub-tab. "Initial Step Size" is the focus increment used in the first pass, and the 2nd pass uses 1/2 of that value. However,  your step size looks fine to me given your screen shots.

Honestly, to me, the backlash and/or the noise in HFR calculations shown in your plots (which by the way is typical in most systems including mine) seem to be the same order of magnitude as the focus difference between 1700000 and 172500. My belief is that the way to improve the system is to figure out how to get more consistent HFR values. I'm thinking about that, but don't expect anything in the short term.

I'd love to see a log file of yours where typical focus sessions last 10 minutes. If there were 25 images in the focus session (it looks yours had 18), then that would be 600/25= ~25seconds each. With your 7s exposures, that would mean 18s of overhead, which seems excessive.  FWIW, when I use Linear with an 8" RC and my ASI1600, I tend to complete focus in 2-3 minutes with similar 7s exposures. Typically I wind up with about 14 iterations per focus session, so that works out to say 12-14s/iteration, or say 5-7s of overhead on top of the exposure time. I'm using an RPi4 with 8MB and an SSD. 

BTW, you might try higher gain for the focus. Also, I'm surprised you're using 4x4 binning with your ASI 1600. I've never tried that much binning, does that work well? That should speed up things somewhat, I guess. 

Hy
 

I am under the impression that what Peter wants is a polynomial procedure with a customisable number of samples before getting to the resolved minimum.

-Eric