×

INDI Library v2.0.6 is Released (02 Feb 2024)

Bi-monthly release with minor bug fixes and improvements

Temperature and Altitude (residual) focus compensation

  • Posts: 33
  • Thank you received: 15
Let me simplify then. For the telescope you used to generate this graph, calculate the width of the CFZ , and the focal length change of a single step (in microns). Now, in your graphic, how many CFZ widths error are contained in the extreme spread (roughly 1000 steps) of the values you generated?

Thus, estimating a step position based on that graphic, is statistically more likely to produce an out-of-focus image, than one in focus.
3 years 3 months ago #63739

Please Log in or Create an account to join the conversation.

  • Posts: 33
  • Thank you received: 15
Giving this a little more thought, I can visualize a "predictive focus tool" that could set the correct initial position and step size based on local conditions and the parameters of the rig being measured, by using a feedback loop created by logging the values of the independent variables (that have a significant enough effect to bother with) of each focus attempt, including the available weather data, into a text file database. This would need to be performed against the same starfield for some base number of repetitions at different weather conditions. You'd need to input the measured step distance of your focuser assembly, and the tool would use the telescope parameters configured elsewhere to calculate things like step-width of the CFZ and use the curves generated by the data set at each variable point to build a mesh of actual resulting values, that could be used as a predictor for focus setting(s) that should be pretty accurate once it has enough training runs in it. So as a user, you'd do the measurement, configure the tool with it, and select a reliable and non-variable star (like polaris, for example), run "Train Focuser" (autofocus that saves values to training file). Then, every time you set up, point to Polaris, and run "Train Focuser" again. Do it enough times, and it should get pretty good at predicting focus position for a given set of conditions. The mesh generated, likely might have value elsewhere in the software as well, since it effectively models the performance of your rig at a given location over time.
Last edit: 3 years 3 months ago by Brian Davis.
3 years 3 months ago #63746

Please Log in or Create an account to join the conversation.

  • Posts: 398
  • Thank you received: 117
Hi bdavis. I think you're driving toward a much more complicated automation regime than what I've suggested (i.e. elimination of existing autofocus loop). I fear that would be very unlikely to ever get implemented (unless you write and test it yourself). What I have suggested in this thread HAS been implemented at multiple large observatories. I know this because I personally worked on the temperature and elevation compensation model at the Large Binocular Telescope, and a similar strategy (lookup tables) was employed at Keck where I worked before LBT. I have used the data I posted for my own f/2.2 RASA 11 (highly sensitive to focus) and it works great. Whether I can convince you of the suggested approach is moot. The proof is in the results and is available to everyone. For the unbelieving, a better question might be: "how and why could this improvement approach possibly work?". So let me try and address that just a bit more.

The temperature component works because the dominant error we're beating down is structural (shift of optics due to changing temp of the OTA). The next component (residual of the temp function) has complex / multiple underlying reasons, but airmass (target el) empirically correlates with the residual to help beat it down. Together, these two available variables get us within the ball park to improve focus management. The key to why this simplicity works is that we're NOT doing a one-time blind position update. We're not replacing the closed loop autofocus routine. We're only "seeding" the autofocus loop's start position. The autofocus loop is still responsible for resolving best focus position. Therefore, all the discussion about spread and CFZ size is misdirected. Ok, now to the next suggested higher automation level (updates between exposures). In that scenario, any position update would use the function outputs as integration OFFSETS from the base position determined by prior autofocus. Those offsets are small and driven by trendline direction. Here, the spread isn't as important as having the trend SWAMP the spread uncertainty so the direction and magnitude keep us within the CFZ and REDUCE our need for another autofocus run. I mentioned to Bart that I'm not sure I would trust a blind update for a large elevation difference slew (and I still believe that). Bottom line: The work needed to do a one-time blind offset and ELIMINATE autofocus runs is well beyond the scope of this thread. So, whether or not I've convinced you, I hear you. I suggest if you want to discuss a "predictive focus tool" that eliminates the autofocus loop, please open a new thread for that topic. I'd like to keep this thread within the bounds of improving and reducing the existing autofocus loop interface (as opposed to eliminating it). Cheers, Doug
The following user(s) said Thank You: Jasem Mutlaq
Last edit: 3 years 3 months ago by Doug S.
3 years 3 months ago #63766

Please Log in or Create an account to join the conversation.

  • Posts: 398
  • Thank you received: 117
I plan to make progress on this little project in the new year. Adjusting the autofocus "seed" position should be simple enough; Everyone should be able to benefit. More complicated, but also more rewarding, is adjusting focus position between exposures to maintain more time on target (i.e. reduce autofocus runs). Currently, I've got a functional test stub that adjusts an autofocus position using an integrator offset defined from temperature and elevation trendline slopes/curves. l've already shown that for f/2 (RASA, Hyperstar, etc.), adjustments on the order of a full CFZ per degree(F) temperature must be done. Also needed is ~1/3 CFZ adjustment per degree of elevation change (below 45 degrees). This is a LOT of needed focus adjustment for beginning of night and low elevation scenarios! The need for adjustments in f/4 through f/8 systems is TBD, but I would expect f/4 and f/6 to benefit (39um and 87um CFZ size). At f/8 and f/10 (155um to 244um CFZ size), it's likely not important (beyond the seed update).

FYI (in case it wasn't obvious): if you have low elevation targets at the beginning of the night to image, it pays to image these EAST if possible! Needed adjustments for falling temperatures partially negate needed adjustments for rising elevation. On the WEST side, these factors are additive AGAINST you, possibly preventing good hold on focus. Something to think about in your planning until we have compensation in place!

Finally, a few requests. I need some 3.5.x focus log data for f/4 through f/8 systems. If someone can send me a debug focus log snippet (see OP) or better...trendline data, it would be helpful for sensitivity analysis. If you might be interested in discussions of the GUI requirements/design for this project, I am toying with the idea of hosting a Zoom call before implementation goes too far. If interested, send me a separate note. Finally, I am still looking to partner with a GUI developer. This isn't my primary skill set, so I could use some help. Hopefully someone who is in the f/2 to f/7 camp will see their own need and want to partner! We can get this done.....Cheers and thanks. Doug
3 years 2 months ago #65251

Please Log in or Create an account to join the conversation.

  • Posts: 398
  • Thank you received: 117
I've updated the charts in the OP (now ~1 year, 350 autofocus runs at night). The temperature data now slightly favors a weak quadratic fit (but I'm still showing a linear). The temp ranges from 20F to 70F. I've also examined focus sensitivity for different systems against seeing (1.5 to 3 arcsecs, f/2.2 to f/10 setups). For ease of understanding, I have attached two graphs, one showing a CFZ perspective (microns), and the other using an EAF quality focuser (5760 counts per revolution) combined with a 750um focuser thread pitch (resulting in focus counts). In case it's not obvious, f/10 or f/7 setups won't need Adaptive Focus Control (AFC for you football fans). In good/best seeing, f/4 might benefit. In all cases, f/2 and f/3 need AFC. Those of us with f/2 or f/3 systems know intrinsically what the charts are saying; the current focus adjustment controls (think scheduler, time/temp deltas) do not suffice. I've also attached a temperature chart from this month to support that assertion (my site/gear).

This month, I used a stand-alone program to manage an autofocus "integrator", and manually offset focus between exposures as temperature and elevation changed. Manually intensive, but the results were very satisfactory! There appears to be no Ekos "gotchas" for updating focus between exposures. Adjustment can coexistent with download, dither, or exposure delay timing. I was easily able to maintain focus in my f/2.2 rig as temperatures rapidly changed. One good seeded autofocus and then occasional updates between exposures will keep focus well managed. Only 1 autofocus needed per target!

So a couple of thoughts. My 7 night January run had some bad seeing in it. Autofocus has difficulty when HFR readings are bouncing all over the place. This adds to spread in the logged data. Linear autofocus tends to pull up short of the bottom of the V curve in bad seeing, especially if the last HFR reading (averaged or not) is marginally higher than the prior HFR reading. Getting one good autofocus result (even if manually adjusted), followed by AFC updates, seems like a better approach to managing focus than risking N funky autofocus results. Finally, while I started this idea thinking that a seeded autofocus was the goal, it now seems clear that AFC between exposures IS the goal. A seeded autofocus start is desirable, but realized AFC between exposures is a "can't live without" feature for f/2 & f/3 systems! So work will continue.... cheers, Doug


The following user(s) said Thank You: Peter Sütterlin
Last edit: 3 years 2 months ago by Doug S.
3 years 2 months ago #66254
Attachments:

Please Log in or Create an account to join the conversation.

  • Posts: 1009
  • Thank you received: 133
Very interesting! Thanks for sharing.
One thing that keeps me from 'blindly' adjusting focus based on temperature is focuser backlash. How do you handle this, especially how did you measure it (assuming you do have some)? One of the reasons I do like Hys focus routine that is insensitive to (at least small) BL.
An EKOS module to automatically compute focus backlash would be a nice thing :)
3 years 2 months ago #66267

Please Log in or Create an account to join the conversation.

  • Posts: 398
  • Thank you received: 117
I haven't yet measured my EAF focuser backlash, but it's on the list of things to confirm this month. I'll use a caliper and manually drive focus to check it. I'm a bit uneasy about the flexible coupler ZWO uses (between motor shaft and focuser shaft), but I think it likely is ok and shouldn't add significant BL. I suspect EAF, Pegasus, and other similar focuser offerings are going to be well enough behaved to ignore BL. Just thinking out loud, I would guess that BL related skew would bias measurements without altering overall trendline slope. The integrator already addresses prediction bias (back to autofocus position); it should cover BL bias too. It's a beautiful theory that could be wrong, but early testing seems to bear this idea out.

Honestly, I'm much less worried about BL than I am with position errors introduced when linear autofocus pulls up short. That kind of error needs to be addressed. Beyond the obvious logged position error, introducing AFC would have the effect of "freezing" that bad focus (relative position offset), resulting in AFC being less ideal than if multiple sloppy AF runs were done (assuming that some of those AF solutions would be good). I need to talk to Hy offline about what these charts suggest regarding sufficient precision of linear AF solutions, and how that might lead to some improvements. More later.... Cheers, Doug

Edit: One important thing to remember about AFC is that it's operating a trendline. Generally speaking, the temperature trend will be downward (bumps excepted). This means that BL isn't a factor because the adjustments are all unidirectional. If you were really worried about BL, you could avoid adjustments during reverse bumps (they typically don't last long). Guess: the magnitude of the BL probably needs to be less than 1/2 of the CFZ to be safely ignored otherwise.
Last edit: 3 years 2 months ago by Doug S.
3 years 2 months ago #66273

Please Log in or Create an account to join the conversation.

  • Posts: 1009
  • Thank you received: 133
I'm afraid you are over optimistic there. The BL of my EAF is 84 units. An error of 20 already produces a slightly visible loss of sharpness (I do have good seeing though). But 80 produces doughnuts (well, not really as I don't have central obstruction...). So a correction that reverses direction would completely set off my focus without proper BL compensation. Well, rather it wouldn't do anything before BL is eaten up by corrections.
As for the value - I measured that with a piston micrometer (no idea if that is the proper word). The value should be fine, although I did that when the telescope was horizontal. Usually the scope will point up. I measured it from the slope of the up and down scans, so that shouldn't matter. But I have no real feeling how well that compensation works for corrections that are smaller than the BL....

As for the flex coupler: I was woried about them too, and (with my older Pegasus DMFC) did a test replacing it with a torsion-free version. It didn't affect BL. That of course depends on actual load and stick slip in the focuser gear which might be stronger at low temperatures. But for me it's indeed no issue.

Uh, you lost me there. I might have to read through this thread again...
It has improved substantially IMO. And it is insensitive to BL. Without compensation, at least for me the polynomial AF routine is close to unusable, and leads to errors larger than the position errors I get from linear. That one quite sometimes is 10 units short, very rarely 20.

Looking forward to it :D
The following user(s) said Thank You: Doug S
3 years 2 months ago #66274

Please Log in or Create an account to join the conversation.

  • Posts: 398
  • Thank you received: 117
Thanks for the BL report; I'll definitely check mine for comparison. What focal ratio are you operating at? It would seem that 84 counts on an EAF shouldn't produce that visible of an impact unless you're operating at low f ratio....but it's an interesting data point....so thanks for commenting!

About poly focus alg, we totally agree. I can't use it at all.... In that sense, the linear alg is head & shoulders above the others. I do think there's still room for improvement however, even given how much better it already is.
Last edit: 3 years 2 months ago by Doug S.
3 years 2 months ago #66276

Please Log in or Create an account to join the conversation.

  • Posts: 1009
  • Thank you received: 133

That's on a Sharpstar 140, F/6.5. So not fast at all. 1 step corresponds to 2.8μ movement, so 84 steps is 0.235mm. That is a R&P focuser, maybe it just has a coarse gear ratio.
IIRC, around 30-40 steps of that BL are internal of the EAF, based on various forum posts here and there.
3 years 2 months ago #66277

Please Log in or Create an account to join the conversation.

  • Posts: 398
  • Thank you received: 117
I'm going to need to think about what you've said. Using the CFZ formula from here (www.goldastro.com/goldfocus/ncfz.php), it seems you should not be able to tell any difference in 20um of change if already well positioned in the CFZ. Your CFZ size for 1, 1.5, 2, and 3 arcsecs seeing (10% tolerance) is 42um, 63um, 84um, and 126um respectively. I guess the real question is how did you arrive at 2.8um / step for your system? We're both using EAF. My Celestron focuser thread pitch is 750um/rev (0.13um/step). You've calculated 21x coarser threads (seems hard to believe, but I couldn't find the thread pitch for your focuser to confirm). Were you forced to not use the 10/1 fine focus on your setup, or have I screwed up somewhere in the calc?
3 years 2 months ago #66278

Please Log in or Create an account to join the conversation.

  • Posts: 1009
  • Thank you received: 133
Doug,

I'm indeed not using the 10:1 fine focus. The reason is that those are friction-driven(*), and, like crayfords, are sensitive to slip. One can usually hold/fix the coarse knob on the opposite side and then turn the fine knob, and will be able to move it. I think I even read somewhere in a manual (maybe from the Pegasus? don't remember) to explicitly not use the fine gear for motor focusers.
And if the CFZ is around 50μ, a step width of 0.13μ sound like clear oversampling, isn't it? :D

(*) at least those that I know; they use an outer cup, an inner axis, and three steel balls as "gear". Don't know if there are others that have a real planetary gear for this.
3 years 2 months ago #66280

Please Log in or Create an account to join the conversation.

Time to create page: 1.643 seconds