Hi!

I've been using autofocus for some months now, and here are a few thouhgts that I'd like to share with the rest of you. I find the module and the function not quite as easy to use as I expected. Often, I get more of a "ramdom scattergram" in the graph, than a neat V-shaped curve. I guess this has to do with atmosheric turbulence and random variation in the HFR. Despite this, the module mostly finds a focus point. So it's working reasonably well.

However, this leaves me with at least one question and one idea:
My question concerns what settings are best for focusing, to handle the randomness in the HFR. I've been playing with shorter (5 secs) and longer (20 secs) exposure times, and with between 1 and 3 frames. However, I'm not sure what strategy is better. Sometimes focus fails, and I increase or decrease exposire time, and it succeeds, but that might as well be just randomness. So are there any general ideas out there on strategies for this?

My idea is that when using the automatic focus as part of the capture module, that is, "focus if HFR > xx", the value for this is taken from the successful autofoucs, as I understand it. However, my focus procedure uses short exposures of 5 - 20 secs, while my captures typically are 180 - 300 secs. The longer exposure naturally causes the HFR to tend to be higher - due to seeing and gudiing residuals, right? So I end up refocusing after almost every exposure. Is this a good idea, or should there be some allowance for HFR to be slightly higher in captures than in focus subs, to avoid unnecessary focusing?

Magnus

There is a gradient-based HFR algorithm that is in the works but it is going to be a few weeks before I have time to do it. Meanwhile, the HFR that is checked in the in-sequence-focusing is the one specified in the capture module, not the focus module. On first autofocus, and if the capture module HFR value was not already set, then the HFR value of the first autofocus operation is set in the capture module + 2.5% of the HFR value, so that autofocus doesn't have to fully run for minute deviation in HFR. However, you can set this value yourself in the capture module.

Finally, I think the "V-shape" is perhaps a bit misleading since the focus module does not take traverse the full length of the curve, it only goes as far as necessary, so usually you end up with half a V-curve, plus a point or two on the "other side" of the curve.

Perhaps there should be an option to do the full curve and then do a polynomial fit to the points? I know focusers sometimes slip a little and have backlash, but if you only do one side of the curve, the exact position of best focus may be hard to determine, particularly if there is some variation in the HFR due to noise.

If you were to plot the full v curve twice, both focusing in and focusing out, you could probably determine the best focus position number for each direction, either moving in or moving out. Or, you could even determine a value for the amount of backlash based on the variance between the vertices. You could use that information to make it even more precise.

Hi!

Just to explain what I mean with randomness - attached is a screenshot of the focus process just now.

Magnus

Yes, mine have often looked like that too.

Hi again!

As I am continuing my focus experiements, I've got another idea:

I'm using a DSLR, with a raspberry pi at the telescope and a laptop running Kstars and Ekos indoors. Now, I set the camera to use Raw format, for my images, but I realized that that means that also the focus frames are in raw format. Would it not speed up the focus process of they could be in Jpeg instead? I already use subframes, but there is still a considerable latency between each capture.

Magnus