AF HFR out of tolerance warning: real issue or false alarm?


#1

I frequently get the HFR out of tolerance popup message and alerts through the notification system when AF runs. It is not clear to me whether there is a real issue or if it is a false alarm. My AF curves are asymmetrical, like the picture below, and I wonder if that is what is causing the warnings. The scope is an astrograph with a large central obstruction.

The lowest points at the bottom of the V can sometimes be wider and flatter than I indicated in the sketch when seeing is bad.

My refractor’s V-curves are nice and symmetrical and I rarely if ever get the tolerance warning when using that scope.

I can get you AF packs and logs if you need them.

Andy


[Smart] Auto focus help
#2

I also frequently get the warning and frankly ignore it. I haven’t taken
the time to understand what the warning means, but when I check focus it is
spot on regardless of the warning or not. This is with a refractor that
usually gives me a nice V curve.


#3

I’m getting these warnings too, but my curves don’t look any different than before. I’ve just ignored them to this point but I too have wondered if it’s a real issue.


#4

I have a 50% CO. The measurement are not accurate outside of HFR of about 3.0. You curves look better than mine. I get a M and a W curve half the time. I happy when get a J. Rarely, get a V.


#5

Neither is it to us… that’s why it’s warning and not an error. Right now this is just a “finger in the wind” check that lets you know something may be a little wonky (or not… warnings are for you to decide). Right now, in terms of minimizing HFR, it means the verification check is > 10% above the lowest HFR seen here:

I realize this is a bit lazy and doesn’t really fall in line with the regression we use to calculate the actual focus position. Until we get around to making this more useful, just take it for what it is… a warning. If it means nothing to you because you always see it and your frames are just as good, then you can feel free to ignore it. If you don’t usually see it (I personally don’t), it may mean you need to check something.

Moving to requests.


#6

Ken,

Thanks for clarifying the meaning of the warning.

That would seem to reinforce the idea that SGP is being fooled by the asymmetry. If SGP is choosing the point where the green regression lines cross, it may be choosing a less than optimal focus point (too far to the right). So in may case the warning is flagging a real issue.

Can you suggest any workaround me for getting focus closer to the real minimum (close to the point where you drew the blue arrow)? The weighted minimum point would appear to be a better choice than the intersection of the regression lines… any way to configure SGP to use that method? Perhaps SGP could automatically use the weighted minimum method when the regression line slopes differ significantly?

Andy


#7

Yes, disable “smart focus”.

It already does this… I don’t think your slopes would trigger this behavior.


#8

ok, thanks, I will give that a try. Does that mean that it will not expand the focus range any more though if I choose that option?


#9

It does. Disabling smart focus disables range extension, range reset, and calculation of focus pos through linear regression. It essentially honors your settings. It will take N samples M steps apart (and no more) and then take the weighted average of the lowest 3 consecutive points.


#10

Ok. I will give it a try.

Andy


#11

Just to add to this conversation, I saw the warning and found that it was indeed genuine… However I found that the cause was essentially my backlash setting.
Once I increased it, I found the warning appeared less often.

Ray


#12

That’s what I think is happening with my system. I am planning today to make sure the hardware has as little flex as possible and look at adjusting the backlash in the driver.
…Keith


#13

Just FYI, If it fails based on slope then the lines will be red rather than green.

Jared


#14

Thanks Jared. My slope lines are always green.

Just to make sure I understand: is SGP doing a slope asymmetry check, or just some kind of goodness-of-fit check? If it is doing an asymmetry check, perhaps the tolerance there is too loose. My AF curves are probably a good example of something that should trigger an asymmetry check rejection and fall back to the weighted lowest points method.

I hate the idea of losing range extension and range reset by disabling smart focus altogether. I know you guys prefer to avoid configuration settings, but maybe the asymmetry check tolerance needs to be configurable, at least in the beta. Once you get some feedback from your beta testers you may be able to find a reasonable threshold for the asymmetry check and remove the setting.

Andy


#15

If I recall it’s checking to make sure the slopes are within 3X of each other which is likely a little on the high side. However since the data is actually quite flat (moved 100 steps, hfr changes by 0.2) this might be warranted. since a change from a slope of 0.002 to 0.006 isn’t really that massive. Or maybe we just consider the focus run a single unit of “1” for the X.

This asymmetry check isn’t new. This has been in since we started the “best fit” focus a couple of years ago.

Jared


#16

Hmm… maybe slope ratio is not quite the right metric. I think a better metric would be to compare the angles of the regression lines, i.e. theta_right = arctan(right_slope) vs theta_left = arctan(**-**left_slope), if the abs difference between theta1 and theta2 exceeds more than, say, 10-20 degrees, reject and fall back to the weighted minimum method.

yeah, I’ve had this problem all along since I started using this scope (asymmetry leading to less than optimal focus point). The warning message and the AF validation frame is relatively new.


#17

The problem is that we’re not dealing with differences of 10-20 degrees (when we consider step size for X). It’s rather miniscule as the 700 step window “beats down” the HFR delta of maybe 3. The view that you see is very compressed in X which makes the slopes (and angles) look more extreme.

So I think we’d have to do something a little different like take the focus window and consider it a “unit” from min to max step and reduce this down considerably. Maybe even consider it a square with the HFR. Otherwise you’re talking about the difference from 1.15 degrees to 3.43 degrees (numbers coming from my above example of a slope of 0.002 and 0.006 based on a 100 step sample and 0.2 HFR change)

I’ve likely done a poor job explaining my reasoning here…but I believe this boils down to 2 things:

  • To be able to determine if the slopes are asymmetrical we a way to remove the step size from the equation (I think pivoting the HFR across the X would be a decent start. ie focus went from 0.8 to 2.4 HFR and steps went from 1000 - 2000, set low steps at 0 (maybe 0.8?) and high steps at 2.4 and extrapolate in between to effectively “square” the graph). Having the determination tied to step size is likely misguided and should be refactored.
  • Based on the above we likely need to be more aggressive about when things are asymmetrical.

Jared


#18

Hello,

I posted a different post on the same subject. I get the same warning. This post helps a lot in allowing me to understand what the issue is. Does this error actually stop the sequence. For me it seems that I have to answer the popup in order to progress with the sequence. Is this true, or am I impatient and if I wait enough everything goes away and the sequence finishes with no intervention.


#19

The popup goes away on its own after 10-15 seconds I think … you can see the countdown timer ticking.

Ah, good point about the relative scales of the x/y axes and the need to be scale-agnostic. It sounds like you have a good approach to fix it. Any chance we might see a fix in the next beta? Would it help if I sent you my AF packs and/or log files?

Andy


#20

Thanks,

I will wait. Thank goodness I am not the President and have access to the nukem button.


www.mainsequencesoftware.com