I think 1 electron read noise would be a great improvement. 0.1, even greater.
OK, I now see you had a typo in the previous sentence.
Read noise is the issue, and it is not proportional to absolute gain. Read noise is *NOT* the amplification of existing noise in the sensor wells; it is noise *GENERATED* in the reading of the sensor. That's why it is different as enumerated in electrons, at different ISOs in Roger's experiments, and in mine.
From some of the statements you have made, you seem to think that read noise is a quantum event, like the captured electrons. When someone says"the read noise at ISO 100 is 30.1 electrons", this doesn't mean each pixel is off by some integer number of electrons, the standard deviation of which is 30.1 electrons for the entire image. It means that the read noise was measured in ADUs, and then on assumed information about the relationship between ADUs and electrons for that ISO, the ADUs are translated into units of "electrons". This figure has nothing at all to really do with sensor electrons and one can calculate the gain in electrons per ADU.
[{POST_SNAPBACK}][/a]
Read noise, like other forms of noise, can be expressed in terms of ADUs or electrons and it is expressed in terms per pixel, not for the entire picture as you seem to imply. Interested readers can refer to this [a href=\"http://www.qsimaging.com/ccd_noise.html]reference[/url], a portion of which is quoted below:
"CCD manufacturers measure and report CCD noise as a number of electrons RMS (Root Mean Square). You’ll typically see it presented like this, 15eˉ RMS, meaning that with this CCD, you should expect to see about 15 electrons of noise per pixel. More precisely, 15eˉ RMS is the standard deviation around the mean pixel value."
Nowhere did I state that read noise was some type of constant or offset and not related to ISO. The 3-4 electron value to which I referred is for the ISO at unity gain. At base ISO under normal photographic conditions, noise is shot limited, not read limited. The read noise of 30 electrons you quoted at ISO 100 might be related to a full well value of 80,000 electrons, and is not significant except in the deepest shadows. With a full well of 80,000 electrons, the shadows 10 f/stops below clipping would still have 78 electrons.
Read noise is normally determined by subtracting two bias frames as described in the above reference or in Roger's essay. The bias frames could include the whole picture, but one normally measures a representative cropped area from the center, perhaps an area of 256 by 256 pixels. The results can be shown as a histogram or expressed as a SD. If the sensor is completely uniform, one could measure a single pixel 65,536 times. One can easily convert from ADUs to electrons, since the number of electrons is the square of the signal to noise ratio as Roger explains.
The main source of
Read Noise is from the on chip pre-amplifier.
Contrary to what you imply, the number of electrons collected is not constant for a given luminance, but follows a Poisson distribution.
Bill