Getting the autofocus to run reliably with my short focal length AT65 was easy and it works every time. Not so when trying to image with my SCT. Can anyone give me some ideas how to get it work reliably?
It’s an 8" EdgeHD SCT with .7 reducer so focal length is 1422 mm. Camera is a Canon T2i DSLR. Focuser is a Rigel System. SGP is newest beta of 2.4.
Last night I did the initial focus with a Bahtinov mask. I have steps at 120. That resulted in a focus curve that was close to what the mask got but the curve HFR values went from a maximum of 6.5-7 to a minimum of 5.2 at the bottom of the curve. If I set step size higher, the stars go so out of focus that SGP doesn’t detect them anymore and the values drop very low. If I lower the step size, I get more or less a flat line.
AF sample size is set to 100 and nebula rejection is towards the weak side but I have moved it all over and it doesn’t seem to change anything.
I have noticed that SGP 2.4 seems to detect less hot pixels when the stars are out of focus but I also notice that it is using very few stars overall, maybe 20 are boxed in the image (I didn’t actually count but it is not close to 100 as AF sample size would suggest). It still has problems with the stars as they move farther out of focus. When I got up this morning all of my later images were so out of focus that the stars were big dim circles. It was in focus when I started.
Maybe I’ll have to just use temp compensation like I do with the Ha filter? I don’t have FocusMax or other software to try.
This won’t help much, but I can relate. I’m using an 11" Edge SCT, microtouch, and sometimes the .7 reducer.
I notice the same behavior though maybe not quite as bad as I can get a fairly flat U. Agree that it seems to be using much fewer stars – I’m not sure if this is the case, so this may be completely inaccurate: suppose it’s using 20 stars in the field. I can visually see maybe twice or three times as many stars, so not sure why these other stars are not getting used (nebula rejection minimum). On the next focus step, it’s picking potentially completely different stars. In theory it shouldn’t matter, but, with a small sample size it’s creating enough variability.
Previously SGP ran autofocus about once or twice an hour as a sort of temp compensation, so I had the same thought of using temp compensation.
I need to save the autofocus packages and maybe even screen record everything. I admit I’m grasping at straws a bit.
Since the moon is so bright, I’ve got a Bahtinov mask sitting on the SCT with it focused on Deneb. I’ll figure out the temp compensation over the next few hours. Thing is… I usually set SGP to autofocus every hour to compensate for mirror flop. Using temp compensation will not allow this to happen.
I tried to autofocus the SCT again last night. The first time it worked and focus was good, When it changed targets and focused, it was also good. An hour later it ran again due to temp drop and took it so far out of focus that the stars were large, faint rings. When I look at autofocus images it is choosing 20 or less stars to focus. I forgot to check off the save autofocus box.
I too have an Edge 8 HD with the same focal reducer and I have not been able to get the auto-focus to work as expected. I also find that it is very easy to have the auto focus routine get the scope so far out of focus that stars are big fat donuts. I have tried steps in the 75 to 100 range and I have set the backlash compensation for the focuser. However, I have yet to be able to get a decent graph that I can trust. I can set focus quite cleanly with a Bahtinov mask but so far SCP auto-focus has been quite frustrating. I have noticed that when the focus is at the fat donut stage, SCP thinks it is finding valid stars at the edges of the donuts. It appears the SCP focus algorithm is not good at detecting grossly out of focus situations. I know I’m not being very concrete with this issue but it is a challenge to be out in the cold trying to get auto-focus to work and at the same time trying to gather actionable data. However, I did want to add another voice to indicate that you are not the only one having these issues.
Our AF routines do not work so well with long FL and a central obstruction. This is a known issue. Our star detection algorithm only works because it is extremely fast (even on slow computers). The math involved to distinguish fat stars from donut stars is exceedingly slow and not realistic for whole image routines.
In 2.5, we may look at adding alternate focus methods to SGPro, like single star AF. Meaning… you tell us where the centroid is and we’ll give you a better HFR. This may involve auto slewing to a bright star and then back to the target… all automated of course.
Ken, any ideas on what changed between 2.3 and 2.4 to be more problematic for SCTs?
It may not be SGP but other conditions/factors; focus has always been a bit tricky with an SCT but had it working quite well with a reasonable v curve.
If I can, I will try to set up a second machine with 2.3 and run through it to compare.
Nothing except for auto continue. Your V should be the same if you did not change any AF params. The one thing that is exactly the same about 2.3 and 2.4 is that we move out by half of (steps * size) take an image grade it, then move in by steps, take an image repeat. In 2.3 we would stop when we reached your steps value and make a best guess at focus, now 2.4 will continue if it wants more data on the “left” side of the “V”. This is really the only difference. Unfortunately, when you continue, you may see the HFR dip back down because the HFR of the circumference of donut stars is smaller than that of big fat stars.
Because of this, we might add an option to disable auto continue (and expand) for folks with gear like yours.
Got it – thanks Ken. I’ll pay closer attention to this when I can get out and try to run some tests while saving the data this time. It seemed like the AF was picking much fewer stars than before, but that could be due to any number of things.
I still have 2.3 running on my little Venue 8 Pro (no perf beast but really great as an astro mini controller!) so I’ll try to do a quick comparison for sanity sake. I just need clouds to break (feature request).
I will add that I think that autofocus for my SCT has become more of a problem with 2.4. It was always a bit tricky but unless I’m watching it now I can’t rely on it.
Changes I’ve noticed in AF behavior are:
the new AF sample size slider - this doesn’t seem to have an effect on anything that I can tell. I’ve tried changing from min to max and anywhere in between and it doesn’t change star detection. I’ve seen this with both my small refractor and the SCT.
Far fewer stars detected and used in AF. I have also noticed this with my AT65 where AF works beautifully. I get no more than 20 stars detected in any frame, even in a dense star field, less with the SCT. SGP 2.3 had a problem of detecting too many hot pixels as stars and putting HFR values very low. Now it seems to not use enough stars in the field. My focus curve bottoms out around 5-6 with the SCT, max is around 7 so the curve is really flat. Any larger step size and the stars go to circles and HFR values go to very low, turning V into sort of an M. I have nebulosity rejection set to weak (full left).
The “use same stars” check box appears to be gone. Maybe it would help the SCT’s to use the same stars every time, especially if the sample size is smaller? Changing to different stars every time may alter the results, depending on the stars chosen?
Is this AF problem limited only to SCT’s or do other long focal length scopes have the same issue? Pardon my ignorance but why does the central obstruction matter? It can’t be seen in the frame. Does it change the way stars look when out of focus?
I’ve never had much trouble with my (non-Edge) C11 and SGP 2.3, but I haven’t tried 2.4 yet. I do have one suggestion, from a rare bad AF run the other night. In that run (which happened at 3 am, so this is from the logs), the number of stars detected suddenly dropped from 25 to 5, just as HFR dramatically improved. I assume SGP switched from using stars to using donut edges.
If that “discontinuity” is typical, perhaps SGP can be instructed to back out farther and repeat the focus run when it encounters this? At least if the user has told SGP that the scope is an SCT.
BTW, does autocontinue in 2.4 only address situations where HFR continuously declines, rather than continuously rises? Can it be made to deal with the latter situation?
I just verified that this does have an effect on AF. Here is a sample AF frame using M13 with the slider on 15:
Here is the same frame with the slider on 100:
As you can see from the screen shots above (from 2.4), AF will select as many stars as you allow it to. Are you certain that your sample size slider was allowing enough stars to be selected? Maybe you need to increase exposure time in that area?
The bottom line is that we cannot really help you without logs and focus packs… everything you or I say without them is just speculation.
Very interesting. Are there any negative side effects if the slider value is too large (i.e. 15 vs 100)? In other words, how would we know what’s the correct slider value for different images?
I’ll save some autofocus packs, I forgot to turn that on again. I have the AF slider on 100 all the time. I have never seen that many stars selected in 2.4. I have my AF exposures set to 30 sec at iso6400. There are lots of stars around but only a few are selected by SGP.
Anyway, I’ll try to get some AF packs saved tonight if its clear.
You adjust the slider only to remove inclusion of hot pixels and noise in the HFR calculation. If no hot pixels or noise is detected, leave the sample size at 100. Otherwise, leave it as high as possible.
I have some data here – not quite sure if it will be helpful. I went back and forth between 2.3 and 2.4 with my SCT using the same settings. It is quite gusty right now so there are spurious points here and there. Even so, there are some trends.
2.4 seems to find fewer stars and, looking at the AF graph only, the HFR is much higher. My best HFR on v2.4 was between 3.5 and 4, and on v2.3 was between 2.5 and 3.
For example, v2.3 screenshot, showing a typical curve I’d get trying to keep the step size a bit on the narrow side – not perfect but a good selection of stars:
Now, a v2.4 screenshot – this was a successful run! – the curve actually looks not too bad, but it just doesn’t seem to find as many stars in general:
Or this one (v2.4), found more stars by the graph is quite flat:
At any rate, I did have some success with 2.4 tonight. I’m not sure why the HFRs are so different, and the number of stars detected seems to vary so much.
Here are logs, full sized screenshots, and AF images for a few runs:
Thanks. That is certainly helpful and exactly the kind of data we need to be of assistance. In the 2.4 runs, what was your sample size set to? Did you actually move the slider to another value or just leave it at the default 100?
The HFRs are different because we have decided to prioritize larger stars for the HFR mean as opposed to 2.3 which prioritizes higher intensity stars, regardless of size.
Sample size was set to 100 across the board – I even tried bringing it down and then back up, in case it just needed a kick (if that makes sense).
That’s good to know about the HFR differences. The exposures were all 25s I believe – though I’ve tried a bit less at some point. Maybe they are oversaturating?
)