Hi Ronald,
If I switch to using centroid then I need to set up some other parameters, as I recall. I will have to dig up my first posts to see what you said about that.
Centroid doesn't really have tuneable parameters.
I am using a step size of 50 and an Over scan of 1000 (I have a feather touch focuser on my SCT and it takes 5270 steps for a full turn ) with the backlash set to 0. 1000 steps over scan may be too high but I know when I reverse direction on my focus knob there is a slight offset in the image due to mirror shift. It does not seem to cause a problem with the autofocus routine other than the second point is almost always worse than the first point in the curve. I kept increasing the over scan to see if I can keep that from happening, but it still does.
The feather touch is supposed to be low backlash so 1000 sounds too high for Overscan. Sounds like something else is going on. There is a settle parameter that can be set after each movement before the focus frame is taken. It might be worth increasing this and seeing if that improves things.
If I am using 'all stars' to focus when does it select the stars? I assume it selects them after doing the over scan and has returned to the first data point.
Yes, correct.
By then the focus should be stable going CCW on the focus knob, so I don't understand why the second data point is almost always larger than the first data point.
So there are a few things that come to mind. If the first datapoint is in the "donut zone" then that could be a reason. There is some functionality coming that will substantially improve this situation. Its merged into the code base and will be generally available in the next release. There is a video on my youtube channel about this - its called Donut Buster.
It could be that the first point hasn't "settled" properly although I would have thought that would make the point bigger, not smaller.
Often I see people using too short an exposure in order to make Autofocus run faster. You could try increasing the exposure.
Are you sure you have the "sense" of the focuser correct and that the last movement is CCW (pushing the primary uphill)?
With the step size of 50 I seem to take a lot of data points to establish the curve. How much accuracy do you think I will loose if I increase the step size to 100?
I would start with 11 points and reduce them when you gain confidence. The thing to remember is that with smaller datasets each point makes a bigger contribution to the curve so 1 bad datapoint will have a bigger effect. Set useWeights=on and RefineCurveFit=on and the software will attempt to compensate for this.
Is there a way to take a first pass at a larger step size and then take a second pass with small step sizes just right around focus?
Yes, set walk=CFZ Shuffle.
I can play with it under the stars, but then I am taking up 'imaging' time. The routine seems to be working for me right now as it is, but I would like to improve where I can and make it more time efficient.
What I do is to collect Focus data whilst imaging and then "play" with it in daylight by running it through the simulator. This allows to run "what ifs" where you can change parameters and see what happens in Autofocus. You will need some technical skill but you don't need to be a programmer. This video explains how to do this...
With the filters, if I recall from the video, you can either reference off of a particular filter (like referencing Green off of the position for Lum) or you can reference off of the previous filter in the sequence. So I could have two reference points, say referencing RGB off of L and referencing Sii and Oiii off of Ha, as I would normally autofocus on L to start with for broad band. Then for the NB filters I would start with an autofocus on Ha and reference the other two from it. Does that sound right?
You could do it that way but I'm not sure there's any benefit to it. What I would do if pick a filter that focuses well, e.g. Lum or a light pollution filter. This will be faster than a NB filter as you can use shorter exposures. Set offsets for everything relative to this. When you change filters, Focus moves the focuser from the old filter to the new, so...
Filter Offset
Lum. 0
Red. 100
Ha. 75
Oiii. -50
Moving from Lum to Red: focuser moves 100 - 0 = 100 (out by 100)
Moving from Lum to Ha: focuser moves 75 - 0 = 75 (out by 75)
Moving from Ha to Oiii: focuser moves -50 - 75 = -125 (in by 125)
Hope that makes sense.