Auto Focus Step Size and Data Points


Hello again.

Finally had enough clear skies to try it out. First [reasonably] focused the system, then started A/F with step size of 22 and 9 data points, star size of 6 at 1x1 binning. I watched it go back and forth and acquire what it thinks is focus. I never got a perfect V shaped plot of data point metrics. But…

Contemplating later, I realized that I might want to use 3 data points for focusing after the initial focus is established using 9 points, and break the step size down to a lower value so the movement is more refined and basically right behind and in front of what is the current state of focus at the time.

Anyone has any thoughts on that and also why a perfect V (or near perfect) is sometimes not achieved?




I found my “V” by starting with some reasonable step size and then trying others until I got good Vs. I always use the recommended number of focal points. I always check my initial focus and make sure that the FOV contains enough quality stars to perform an automatic focus.


I don’t really know if there is a recommended number of data points, but if you have found your groove then that is great. I am hearing that your goal is to get a good V plot and that you keep using different step sizes until you get a good V plot.

I am thinking that the depth of field, which is called Critical Focus Zone is a very reasonable basis for initial auto focusing since SGP isn’t going to bring a totally put of focus OTA into focus. But after that initial focus then the step size shouldn’t have to be a function of the full CFZ, and in fact it should be a fraction of it (70%, 80%…) because if we are already in focus then it should take very little to keep in focus.


In my experience with SGP autofocus I’ve found it works best where there is a larger range of focuser movement and relatively steep sides on the V curve. The area near best focus is more like the bottom of a parabola for me and this means fitting regression lines at this part of the curve leads to quite flat lines and the intersection point can move around quite a lot.

I also use 11 points so there are more points in each regression as I was finding one bad point can throw out the regression line if 7 points are used. It takes a little bit longer to run the focus but the results are more repeatable.


I am still trying it, and last time, after getting the first focus, I kept varying the step size and evaluated the plot. I could never get a perfect V, but I did come close.


I think you mean “depth of focus”. Depth of field has no real meaning for objects at infinity. Just trying to be clear on terms.


the problem usually comes from step size. You must increase to 200 and in some cases to 500 or more. I had the same problem and it lasted until I received this information.