ASI1600 Bias Stacking Effectiveness
I did some testing with Low Gain (Gain 0) data. I acquired 256 bias frames at -20C with SGP using the latest ASCOM driver. Stacked them in PI, with stack sizes spanning the range from 4x through 256x frames. I performed statistical
analysis on the whole lot to determine how many bias frames need to be stacked for the most effective and efficient results.
Based on this analysis, and the visual observation of noise reduction below, I do not believe there is a need to stack more than 64 bias frames in most cases, maybe 81.
Full Sequence GIF: http://i.imgur.com/QUIDrOm.gif
Based on the convergence of the histogram in the animation above (click the link below the short GIF here to see the full series progression GIF), it does indeed appear that to get the full 16-bits of precision out of ASI1600 data, you do need to stack 256 subs. However, by 81 subs you have gained another full three bits of precision, which I think will probably be enough for most people. I think some highlight precision testing is still needed, as when I processed some of the M57 data previously, there did appear to be some posterization in the highlights, and that was even at 76 frames. If posterization in the highlights is an issue, then I believe stacking a full 256 sub (or larger) stack might indeed be warranted.
Edited by Jon Rista, 11 June 2016 - 01:21 PM.