Hi Buzz and RockstarBill
Thanks for all the feedback and tips. Weather has been bad and I have not done much, but did manage to sneak in some additional tests last night.
I think by now it is pretty clear that my babyQ has field curvature which is at the upper limit of the specifications. The integral flattener will only produce flat field if focus is carefully achieved. Your baby-Q is likely better than mine so you have been able to get good results with standard AF parameters. I know it is not tilt as round stars can be achieved - it is just fiddly to do so.
At the beginning, I have been assuming that bad field flattening was due to an apparent difference between the best focus for HFR and the best focus for round stars. I have been trying to determine what this difference was so that I could focus for best HFR and then apply an offset to get round stars. However, I have not had any luck with it. Firstly, the focus point for best HFR was often different for different runs; and secondly, the offset that was needed to get round stars was also different for different runs.
Then I thought, may be I could minimise this difference by including as many corner stars as possible in the AF routine for HFR calculation. My hypothesis was that when the corner stars were round, they would also have the lowest HFR and by including as many corner stars as possible in the calculation, I would also optimise the corner stars. To this end, I have changed to bin 1x1 and reduced the min star size to 2. I thought I hit the jackpot as this strategy appeared to produce good results with L-filter, albeit very slowly. Unfortunately I did not have much luck with the other color filters.
But then I started thinking, why was the AF focus so inconsistent between runs? Clearly the inconsistency was the reason why simply applying a constant offset to the AF result did not work. Perhaps there was something about the autofocus process that could be improved?
So two nights ago I set up to test whether I could get more consistent AF results by just using the centre of the FOV for focusing. I activated the "crop" function so that the elongated corner stars were excluded from the analysis. That's when I found something quite interesting:
Without changing the step size, I found that the previous V curve became a very very shallow U. When the corner stars were excluded and using the old step size, the HFR at max outside focus was like 1.5, it went through focus at 0.8-1 and back to about 1.5 at max inside focus. The centre 3-5 data points all had the HFR around 0.8-1.0. It became very clear that the AF focus was very susceptible to error (a slight change in the HFR results in a big change in the calculated focus point). To make a better V-curve with well defined slopes, I had to very substantially increase the step size from 18 to 55 and the min star size from 2 to 10!!
It would appear that previously the majority of the HFR change through the focus range was in fact due to the corner stars, and the centre stars did not change much through the focus range at all. It also means that the AF focus result previously was likely not very accurate.
I have since changed back to bin2x2, with 40% cropping and increasing the step size from 18 to 55 and min size to 10. By doing this, it appears that I am able to get good corners with my LRGB filters although more tests are required.