Hmmm. I want to pick up on the discussion a few comments back, about bit depth and noise and colour gamut, before it got on to lenses.
I would guess that many of the topics being discussed here are independent variables, and are being mixed together into a bit of a messy soup. Noise, for example, is generated by each photosite, along with the signal. Every electronic device creates some noise. If the signal level is very low (dim light) then noise becomes a bigger issue, starting to overpower the signal itself. Hotter sites also generate more noise and there has been discussion about the heating of various sensors and how that affects the image. Amplification of the signal also amplifies noise, so higher ISO (higher amplification) means more noise. If a photosite could be built that generates no noise (impossible) then there would be no noise, irrespective of bit depth, 2 or 22. Noise and bit depth are independent variables.
Bit depth is a design feature of AtoD converters. Converters are electronic devices that are built with a specific bit depth. That is just a physical feature of the converter. An analogue signal being converted to digital will have more detail saved if converted with an AtoD converter that has more bits. Audio fanatics talked about this a lot when CDs were replacing LPs.
So, not having looked into the details, I am guessing that the analogue signal received through the lens (this device does not count photons per se) is converted to a digital signal after each photosite. I’m sure there are not 16 million A to D converters on the chip(s), so it must be somehow streamed sequentially to a number of parallel converters. Whoever talked about detail between limits is right. The A to D converters can only use the dynamic range fed to them by the photosites. Again, these are two independent operations. The bit depth does not extend or reduce the dynamic range (unless the Ato D converter was very badly designed, which I doubt). The AtoD converter simply stores more or less detail within the photosite’s dynamic range, and, I’m sure, is designed to cover the total range.
Then there is colour. Each photosite is sensitive to a range of the electromagnetic spectrum and produces a signal when that part of the spectrum ‘excites’ it. This is nonlinear, and I have no idea if the camera manufacturers try to do something to account for that. I could get into colour theory ( but would have to look it up again) so I will just remind that there are 3 sensors per site, one for each of the primary colours. So a measure of the real colour is based on the relative values from each of the three sites. Bit depth should play into this I guess, because the greater the detail about each of the primary colours ( more bits) the greater the accuracy of the final mix.
The last bit is the construction of the sensor. There are lenses at each site. The bigger the lens the higher the signal so the lower the noise (relatively speaking). The lenses, I am again guessing, also have colour filters on them ?? I haven’t really looked into that but assume it is the case, so they can differentiate between the primary colours. Then there are UV filters built in because the photosites do not have the same colour sensitivity as our eyes and we do want the pictures to look the same as when we see the real thing. I haven’t explored the details of sensor spectral sensitivity at all, but the point I want to make is that the colour gamut is a function of the sensor and the filters that are used, not a function of bit depth. It is the fine gradations within the gamut that are a function of bit depth.
If there are any Electronics Engineers/camera/chip designers out there that find fault with my reasoning, please let me know.