Auto Focus Woes


#1

So I’ve spent 5 nights working with auto focus, with it failing/giving erroneous results 9 times out of 10.

I rarely get a v curve, and when I do it is shallow and has “wings” often on one side only. Most of the time I get a jagged line that is nearly flat or slightly sloping.

I’ve tried a bunch of different settings; step size, cropping, threshold, nebulousness, exposure, center obstruction/autoscale etc. (and the various setting options keep changing with each beta).

For step size I used the reference calculator and my CFZ is 56 um. My focuser is giving me 2.2 um per step so my step size for autofocus is 24, but I’ve tried half that and double that and others as well.

I’m using the same exposure I use for platesolving, 10 seconds @ 12800 iso, but I’ve also tried 20 seconds. I image @ 480 seconds @ 1600 iso. All of this is through a CLS filter, and a clip-in IR/UV filter (The DSLR is full spectrum).

I don’t have working darks for autofocus yet, due to the file naming and format issues I’ve posted about in other threads, but I will try to make acceptable fits files next (is there a log that tells you when the dark files are found and used?).

Focusing with a bahtinov mask on a central bright star works fine and gives me a consistent starting point (in number of steps/focuser position).

I’m happy to make my focus packages available if that would help debug.

One thing I’m thinking might contribute to the problem (although cropping the focus runs doesn’t seem to help) is the curvature I have with my current optical train:

8" RC with a focal reducer AND a field flattener. According to CCDInspector this gives me the flattest field at 29% however there is kind of a double dip in the focus as you go out from the center. My guess is that this makes “whole field” focusing challenging, as different parts of the field come into focus with different focus positions. Except that, again, no amount of cropping seems to solve the problem.

Would FWHM with pinpoint be worth trying?

Any other ideas?

I do plan to experiment with removing the Field Flattener and/or the focal reducer just to eliminate those as the problem.


#2

If you’re getting mostly flat curves I would recommend upping the step size. The CFZ method is somewhat outdated and in the 2.4 docs we recommend that your max HFR be 2-3x of your min. With scopes with central obstructions this can be tricky. In those cases you want to minimize donuts as our current star detection algorithm does not like them and will report them as low HFRs.

So in short, for a scope with a central obstruction you want to:

  • Attempt to get a max HFR of 2-3X your min (in focus) HFR
  • Keep the focus range small enough such that stars don’t go to donuts.

Doing those two things and you should have good results. I would dismiss the information about CFZ. It’s good to know but I’m finding I get better results with the 2x-3x recommendation.

If you post some screenshots of the curves we can take a look and make some recommendations. If you have pinpoint you can try the FWHM method as it might behave better with donuts.

Thanks,
Jared


#3

It’s possible that they might.

You will also need to rule your focuser out as an issue. Find focus with a bahtinov mask, then move the focuser outside and then inside of focus by a number of steps roughly equivalent to the AF run (keep Jared’s comment above in mind), then return to the focus position indicated by the mask. If you are still in CFZ, you can probably eliminate the focuser as an issue.


#4






These were all from a few nights back. Since then I have not been able to get any “V” shaped curves at all. Just jagged lines, no matter the settings.


#5

Auto Focus Package That corresponds to the 2nd to the last image in my last post (the “best” looking V-Curve).


#6

The last two curves look similar - but were extended too far out and had flat ‘wings’ and it looks like 20680 is the value, some of the earlier graphs are off range and you have got random results, I’m suspecting that the others had brief exposure times.
I used to get similar results when my exposures were too brief. Coming from FocusMax, I was used to very short exposure times on a bright star. With SGP I had to re-learn my technique and I am normally using 8-10 seconds with a luminance filter on a f/6 scope with a CCD on a refractor. This has a focal length of about 600mm and I’m using a step size of about 100 microns. Your RC has a large central obstruction and as soon as you get donuts, the HFR calculation goes awry. Are you guiding during these exposures? Could the randomness be caused by tracking or PE related?


#7

Hmm, come to think of it, guiding might be the thing that gave me the somewhat V curves vs. what I’ve been getting since.

Another variable to revisit.

With the current optical train I’m at f4.92 and 0.89 arcsec/pix, but with the filters on the camera an above the light pollution exposure would be 60 seconds @ iso12800 vs. the 10 seconds I use for plate solving. I can try 60 second exposures (That’s a long time for a focus run!) and certainly anything longer thatn 10 seconds probably requires guiding.

My guiding is mostly within 2 arcseconds, with higher altitudes being mostly under 1 arcsecond. Still dialing in a new mount, no PEC applied yet. Gearing up to experiment with guiding using IR only (hopefully will help at lower altitudes).


#8

You will also need to rule your focuser out as an issue. Find focus with a bahtinov mask, then move the focuser outside and then inside of focus by a number of steps roughly equivalent to the AF run (keep Jared’s comment above in mind), then return to the focus position indicated by the mask. If you are still in CFZ, you can probably eliminate the focuser as an issue.

Focuser ruled out as an issue. Ran 1000 steps in either direction and returning back 1000 steps mask pattern is dead on focus.

Guiding Helps.

Reducing the AF sample size helps.

Increasing the exposure does not help.

Increasing the step size to 4x the CFZ amount seems to help, but I still don’t have 3 to 4x HFR values. Going further only increases the size of the wings on the V.

This is my best result to date:


#9

Hi I had similar problems using celestron C11 sct (at F6) and starizonas microtouch focuser - both good bits of equipment.

I started out working out step sizes using CFZ calculations, but this was not helpful. I experimented with step size and number of points and have now settled on 7 points and a step size of 85 which for me, fits with what Jared said:

'Attempt to get a max HFR of 2-3X your min (in focus) HFR Keep the focus range small enough such that stars don’t go to donuts. ’

My curves are variable and my best looks like your best; i sometimes get one like your 20681 focus point.

However, I think the main point which made me let it go in the end is that there is no discernible difference in focus that I can perceive in the images I take. I think it’s good to strive for perfection, but quite often its just that.

I think you’ve done the right thing to experiment. Your equipment is pretty fast at F4 ish so I don’t think that exposures over 5 sec will help. I use 2 sec exposures (@F6.3) and bin 2X2 which helps the donuts problem too.
Because my star shapes are a bit off in the corners (celestron F6.3 reducer), I crop by 10%. Another good thing which you may already be doing is to have the HFR calculation turned on for the focus frames - you can see what’s happening a bit from that.
Good luck with it.
Paul


#10

With an HFR in the 5-6 range you will essentially be trying to focus a fuzzy blob which, at around critcal focus point, will result in a flat-ish line for some considerable distance either side before the donut enlarges again. Basically, I would say optical performance is poor resulting in a ‘U’-shaped curve with a flat bottom. You might try collimating your optics carefully to see if you can improve things. I know it is unfair to compare with a refractor but as a reference point I get a minimum HFR around 0.5-0.6 and at the extreme ends of the focus run (15 point curve) I get HFR around 3, and this results in a nice sharp ‘V’ curve. Bottom line: the issue is not so much with the software as the data it has to work with.


#11

OK, except that CCD inspector is reporting FWHM under 2. So I if I understand that right that’s about as good as you can get with typical non observatory location seeing conditions?

As to Collimation, it looks good by various tests and the experts at my club’s fixit sessions have also tested and found the collimation good on two recent occasions.

I’ll probably run some autofoucs tests without the FR and/or FF to see what combinations work best.


#12

So far mechanical issues have not been mentioned. Have you checked that the linkages between the motor and focuser are tight with no play. Also is the backlash setting working as it should. Either of these will give graphs like you show. I know, I had a loose coupling that I thought was tight. I fixed that and I have no problems with getting great V curves.

Mike


#13

Mike,

I don’t have an SCT so backlash shouldn’t be an issue. I have an RC with a Crayford style focuser.

I’ve run the focuser out 1000 steps and back in either direction and it goes right back to focus (Bahtinov mask), however I do see some changes over repeated focus runs in SGP, (absolute focus position, in steps, changing over the night) and need to determine if that is temperature or some mechanical problem.


#14

I concur that mechanical issues like @MikeOates has had will defeat autofocus. I have had many: loose gear coupling, loose attachment of focuser to scope body, poor quality focuser, wobbly camera connection. The focuser that came on my RC8 was so bad, I had to replace it with a good quality one. Also had bad collimation on both the RC8 and my 90mm refractor which raises havoc with the FWHM values. Also poor guiding during the focus exposure is bad.


www.mainsequencesoftware.com