Цифровое фото и аппараты для оного

Think of a set of stairs. The height of the stairs between floors is the Dynamic Range. The number of steps in the stairs between floors is the number of ADC bits.

Any given sensor technology has a fixed dynamic range (stair height). You can cover that dynamic range (stair height) with as many bits (steps) as you'd care to. A 16-step stair allows you more discrete positions between the floors, but it doesn't increase the height you travel. In essence, the more steps (bits) you use, the smaller the discernible increment is. But since sensors output an analog integer value (electrons can't be split), and we can only count to 60,000 or so, we'd never need more than 16 bits to accurately record the data.

Where people get confused, I think, is in making the assumption that sensors are perfect. If I had a sensor and ADC that were perfect, collecting 0 photons would be black, collecting 1 photon and converting it to an electron would be a stop, collecting 2 photons would be the next stop, collecting 4 photons would be another stop, collecting 8 photons would be another stop, and so on until we got to 65,536 (16 stops up the scale). But in reality, what happens if you have a base noise level of 128 electrons laying around in the sensor at all times? Then we can only distinguish 256, 512, 1024, etc., or about nine stops of light differential.

Other factors come into play, too, but I think that's enough to give you the idea. Bits do not automatically equate to stops. In the current sensor designs, 12 bits is about the right value to get an accurate count from the noise floor to the maximum filled well.