spaceman How far to the left is too far?
The optimal value (minimizing error while not losing too much dynamic range) depends on your copy of the camera, and what gain you have set it for. You need to measure your own camera.
My 585MC for example with the default bias (ADC offset), at gain of 0 dB and 1 millisecond exposure, the standard deviation is about 11 ADU, with min ADU of 112. So, it is not likely that the analog noise will drive the min ADU below 0 with gain of 0 (it is a whopping 10 sigma).
However, at a gain of 6 dB, the same exposure gives me a standard deviation of 17 ADU, while the min ADU is 15. This is way too close -- about one sigma; so I would expect the sensor signal to fall below 0 about 32% of the time!
For my camera with ZWO's default bias (offset), any gain of 6 dB or over (gain of "60" using ZWO's scale) renders the camera useless, since stacking will not improve the signal to noise ratio by the amount that you expect (3 dB per double integration time).
If you adjust the bias (ADC offset) so that 3 times the standard deviation is equal to the min ADU, then the sensor will only fall below 0 about 0.3% of the time, and is probably acceptable for SeeStar type users, but not for real DSO people.
Larger margin is better, but you lose dynamic range. But you don't really have to be that extreme. Just 4 sigma would reduce the error to 0.006% for example, from the 0.3% from 3 sigma. And probably good enough for any hobby type user.
So, in answer to your question of how far to the right? Keep increasing the offset until the min ADU of a bias frame is more than 4x of the standard deviation, and you will be quite safe, and not lose much dynamic range than neccessary. Make sure you remeasure when you change gain.
(Unfortunately, this is not possible if you use ASIAIR. So, the camera is a door stop for ASIAIR users.)
I am just using the percentages for standard deviation of a Gaussian, but it gives you an idea of the magnitude needed. The camera noise statistics in not Gaussian (closer to Poisson distribution, which you can check here https://en.wikipedia.org/wiki/Poisson_distribution to get more precise percentages).
Chen
