Originally posted by MJKoski It is limitation of camera hardware and/or software when doing the calculations
Interesting. I ran on test comparison between composite average at base iso underexposed, composite average at high iso to compensate for underexposure, composite adding underexposed, composite adding at high iso. And I think I (kind of) understood what the camera does. When averaging, the camera does a bit wise average before storing the result in DNG without rescaling to 14bits, this leads to banding because the small gradations of tones are discarded by the average. While for the adding mode, the camera performs sequential 14bits wide word additions, then store the results in DNG. In the case of average, shadows that code barely over 5bits remain 5bits deep, and as the averaging increases the value can even become corrupted if one bit on a word flips from 0 to 1 but least significant bits don't change, so this is the banding I observed. Now, the adding mode is much more interesting, as it is effectively pushing up least significant bits to higher positions within 14bits words as the value of the stack increases, there is no loss of information, on the contrary as the stack grows the values get more refined until reaching the 14bits range, as which point the image quality is at its max (better than ISO100 quality since the DR is 14bits and the read noise has been cancelled out along the way). So... the best is to lower to ISO100 and count the number of stops of underexposure, then select the number of frames to be added based as 2^(# of stops of underexposure), and this maxes out image quality without any banding or artifacts.