Originally posted by TDvN57 Apologies for being somewhat off topic.... I stumbled upon a video yesterday which explains in simple layman's style how gain and noise works from the sensor side and how a higher gain (iso) setting affects the end result. Helped me to better understand some comments from some tech guys on the forum who have spent their lives designing and building some of the stuff we take for granted.
Here is the link:
CMOS Gain Settings Explained - Atik Cameras What the video seemed to miss, which is essential to understanding the actual signal to noise in the pixel, is that the Gaussian noise in a collection of randomly generated electrons (from photons) is the electron count from the square root of the number of photons (2000ph => 45ph, rms => 22e, rms). So the signal to noise without counting the readout noise is also square root of the number of photons (2000ph/45ph = 45). For the 1000 generated electrons, 22e rms is 7 times the readout noise at low gain, and almost 22X the readout noise at high gain (if I recall the video reported details correctly).
Total noise is root-sum-square (rss) of the two noise sources; and hence, for photoelectron counts higher than, say, 100e, the noise in the signal dominates the noise in the readout, as one wants. So the readout is great, but nature is not so great. At 100e, the total noise is rss{7e, 3e} = 7.6e, rms, and the SNR (voltage) is 13. This would be a tad grainy, but once past the 2500 photon count region, the details should be pretty good, even pixel peeping, assuming the eye can detect contrast down to 2%. (That is, the apparent contrast between two pixels with nominally 2500 average photon counts each should look identical. This may not be true of actual video with the same parameters because the eye can sense small changes and the noise will have random values such that the contrast will intermittently be greater than 2%.)
Note: the quantum mechanical process of photons generating electrons (50% on the average) is itself probabilistic, and the above is likely optimistic.