Autofocus difficulties


#1

I’ve recently purchased a Meade LXD55 SN10 that I’m hoping will be good for astrophotography. I’ve been having some difficulties getting consistent autofocus results with this scope and need some advice. I’ve tried focusing with the Bahtinov mask and then determining the point where stars become donuts and using the (f’ - f) * 2 / steps approach but can’t seem to get SGP to give me a decent v curve. Here’s an example

  1. I focus with the mask - focuser position is 8690 steps:
  2. I run autofocus which struggles and comes up with 8702 as the focus point:
  3. I check this point with the mask and it is clearly not as good as when the focuser was at 8690

I’ve tried increasing and decreasing the step size and 7 as well as 9 steps. I think I’ve got logs enabled so could post these (if I can find them).

Any advice on where to go from here?

Thanks,

Peter


#2

The good news is that you clearly have a trend. This means that you are pretty close. Looks like you were using 10 steps and didn’t start from the bahtinov focus you noted. I don’t think that’s of any real matter though. The question is: why is there almost no movement in HFR from 8690 to 8726? This makes me think that your auto focuser may have a bit of “slop” in the gear meshing and the first outward movement didn’t really do much in terms of moving the focuser. It Just moved a bit, then unwound it self with the same issue and finally engaged at 8678. Have you tried using backlash compensation?


#3

Thanks for the feedback Ken. The Autofocus run started at 8690 and moved out 3 lots of 12 steps however the 4 inward points didn’t yield much movement in HFR. I’ve never found issues with this Moonlite before (I had it on my 10" Newt) but I am finding the SN10 quite a challenge. I think the CFZ is quite small. I’ll get the dial gauge out and give the backlash idea a test and report back.

Regards,

Peter


#4

Also… just as a note, it looks like there might be a bayer matrix present here? Are you using an OSC?


#5

yes - it’s a QHY8 - does that introduce other issues? I’m using 2x2 binning for focus runs.

Also checked the backlash in the Moonlite with the dial gauge - it was very reproducible (within 0.005mm on the dial gauge stepping in and out over 120 steps in 10 step increments over 9 runs. I made a small tweak to the stepper backlash adjustment from 8 to 7 steps.


#6

Well, that’s a good question. My initial response for you is that you are fine where you are because of the distinct trend you exhibit (at least on the inside of the “V”). But… binning at 2x2 does not remove all of the aberrations from the image that can cause errant centroid detection. You may find that it helps at 3x3 (in terms of proper centroid detection, but I think your primary issue lies elsewhere for the time being). This would be empirical, but may be worth a shot.


#7

Yeah - I should have bought a TAK.


#8

Could the small difference between 8690 and 8702 be in the noise? The HFR value is hardly different. If so the difference may not matter, even if the Bahtinov mask can see a difference. After all it is the star size that matters in the image.

I would not expect backlash to matter much because all the movements when doing auto focus are in the same direction. It’s only the move back to the best focus position that’s different.

Why 3x3 binning Ken? I can see that 2x2 binning should remove the effects of the Bayer array but 3x3 will reintroduce it.

Chris


#9

Hi i am having similar issues with my reflector, like you I have the bahtinov mask showing good focus but the the auto focus fail to repeat it…i also use a moolite on the reflector a cr1…no real back last issue and the positioning is repeatable…
The cfz is quite small 48 microns.
I am still in the process of testing with alternative settings but as of yet I cannot repeat…
The other night I had what I would describe as a perfect V curve but the focus was clearly wrong…i do have an fsq106 and this works fine but it does cconcern me and I will now double check that.

Ray


#10

Just as an additional point, my camera is a qsi 583 mono…
The focus difference shows up quiite obviously in the diffraction spikes from the secondary i considered collimation as a potential reason but CCD inspector confirms that its within reasoñable boundaries!

Ray


#11

I know this is frustrating for long focal length and secondary obstruction users. The current AF works really well with refractors and we are going to do our best to create a more robust method in 2.5 that can handle donuts, bayer matrices, diffraction spikes and star poor areas of the sky.

Ignore the part about SCT…


#12

The screen shot above at the 8702 final position looks like it was happy with star selection and there were plenty of stars that were not selected but clearly distinguishable from the noise.


#13

I meant noise - or variation - in the focus position if multiple auto focus runs are made. I would expect that the position would change slightly each time. This isn’t the same as noise in the image.

Chris


#14

I’ve been doing more work on this over the past few weeks. I’m now imaging with my QSI683, having reinforced the OTA where the focuser attaches so the bayer matrix is no longer a factor. I’m still seeing the same issue with the SGP focus position not matching up with the Bahtinov mask position. I’ve tried different binning settings and varying the exposure length and can appreciate the importance of getting good SNR to get reliable HFR results. I’ll be sticking with 2x2 binned autofocus images going forward to ensure I gather enough signal as this seems to be more important than the resolution.

I’ve also read a number of papers written about Focusmax and I’ve come to the conclusion the HFR results near focus are not going to respond linearly to focuser movement and this is going to create issues using a linear fit of the v-curve if the points close to focus dominate the regression. I’m wondering if it would be better to exclude the points at the bottom of the curve from the calculation?

Likewise if you get too far from focus the donuts cause havoc and the HFR line tends to flatten out. It looks like there is a sweet spot where the line responds linearly.

I’ve found the results can be quite erratic if you try to create a curve with 7 points - ie three either side of the focus point. This is because one dodgy measurement throws out the line fit. Although it takes longer to run autofocus, the results are much more repeatable if there are 5 or more points either side of the focus to fit a line to.

I’ve also done some analysis of the guide logs and believe there needs to be a criteria where the slope of the regression line is similar inside and outside focus - especially if the focus point is determined from the intersection of the regression lines. Wildly differing slopes seems to be a symptom of bad HFR data, influenced by non-linear HFR response near focus or donut stars too far from focus.

Is there any documentation on the method SGP is using to determine the focus point?

Finally I’ve written some very dodgy Visual Basic to extract the autofocus run info from the SGP logs and dump this in Excel - it would be great to see a tool in a future release to allow the logged autofocus data to be viewed more conveniently.

Cheers,

Peter


#15

I’m not sure how this influences the bayer matrix one way or the other.

This is almost always a good decision. It also helps with the time required to run AF. If your QSI is of the OSC variety AF at 1x1 is not even really an option… the bayer matrix will almost always muck with centroid detection.

This is true. Smart focus, which uses regression for the ideal focus position, does not consider bottom points within a certain range of each other… indicated by a red X (i.e usually the flat part of of the “curve”… non-linear). When not using smart focus we use a simple weighed average of the bottom 3 points (so long as they are grouped).

As with most things… more data is more better. The key is finding the minimum sufficient.

This is true. We account for this by defaulting to weighted average method when detected.
.
SGPro 2.5 will have a fairly large focus on hardening AF. We have code that can deal with donut stars, but it is not yet efficient enough to be used at the “whole image” level as of yet… especially for the average astro “field-machine”. In addition to this, it is also far more susceptible to false positives introduced by nebulosity.

The current method is far more tolerant of setups < 1000 mm of FL with refracting lenses. We want to expand AF’s tolerance in 2.5 so people don’t need to write scientific papers on how to use SGPro with fast (small CFZ), long FL scopes using a secondary obstruction.

We are still working on fulfilling promises we made for 2.4 so we have not yet begun 2.5 in earnest. That said, we have at least a dozen posts (like this) from smart people full of good ideas. All from folks with the same frustrations. We will get something together… not sure exactly when.


#16

The significance of reinforcing the ota is I can now use the mono QSI rather than the OSC QHY8.

Is there a way to get the sequence to automatically slew to a focus field and then slew back to the target when focus is complete?


#17

There used to be but it has long since been removed in order to promote the “on target” focus methods for which we have been striving. Maybe there is a place for this in the future (2.5?)… Depends on how well things work out. The only reasonable excuse we would like to have for violating on-target focus is when imaging truly star-poor regions.


#18

Sounds like I might be manually slewing and focusing with a mask for a while if I can’t find some settings that work with this OTA.


#19

Ken - can you elaborate on this? How is the weighting done?

Peter


#20

Hi Ken. I’d like to add an expression of interest for re-introducing the slew to a focusing target. Recently I’ve been imaging NGC104 47 Tuc globular cluster - hardly a star-poor region. Normally my V curves are distinct and AF runs fine, but for some reason this particular target is causing some issues - only think I can think of is that the out of focus glob confuses the HFR estimations?


www.mainsequencesoftware.com