Doug Kerr
Well-known member
Notwithstanding the many useful observations presented here, I have still not developed a comforting insight into how, from theoretical considerations, we might expect differences in sensel digitization bit depth (e.g., the difference between 12-bit and 14-bit systems) to influence the dynamic range capabilities of our cameras.
I have just discovered a 2008 paper I think will turn out to be a treasure in this area. It is by Professor Emil J. Martinec of the Enrico Fermi Institute of the University of Chicago, who has taken time from his work on string theory and particle physics to prepare this paper, entitled "Noise, Dynamic Range and Bit Depth in Digital SLRs ", which is available here:
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html
The paper is large (it printed out at about 45 pages using my normal print typography). Although it is rigorous and of course involves a lot of mathematics, the level of the mathematics is not esoteric, and Professor Martinec provides nice summaries of the "bottom lines" of the various areas. There are numerous excellent illustrations.
Interestingly enough, Professor Martinec proposes a scheme seemingly equivalent to one recently proposed here by one of our members, in which the sensor is in effect simultaneously operated at two different ISO sensitivities (by having two parallel amplifier and ADC chains, with the amplifiers having different gains) and the two images are combined, in-camera, in an "HDR" scheme to provide an enhanced dynamic range from single exposures.
I have not yet in fact read the paper. After I have, I may have some observations to post here on what it has taught me.
I have just discovered a 2008 paper I think will turn out to be a treasure in this area. It is by Professor Emil J. Martinec of the Enrico Fermi Institute of the University of Chicago, who has taken time from his work on string theory and particle physics to prepare this paper, entitled "Noise, Dynamic Range and Bit Depth in Digital SLRs ", which is available here:
http://theory.uchicago.edu/~ejm/pix/20d/tests/noise/index.html
The paper is large (it printed out at about 45 pages using my normal print typography). Although it is rigorous and of course involves a lot of mathematics, the level of the mathematics is not esoteric, and Professor Martinec provides nice summaries of the "bottom lines" of the various areas. There are numerous excellent illustrations.
Interestingly enough, Professor Martinec proposes a scheme seemingly equivalent to one recently proposed here by one of our members, in which the sensor is in effect simultaneously operated at two different ISO sensitivities (by having two parallel amplifier and ADC chains, with the amplifiers having different gains) and the two images are combined, in-camera, in an "HDR" scheme to provide an enhanced dynamic range from single exposures.
I have not yet in fact read the paper. After I have, I may have some observations to post here on what it has taught me.
