are ISO settings a nonsense on digital?

So - I think a lot about ISO and noise and camera design relating to those things because I take shots in low light.

Now as I understand it, the iso settings simply change the gain of the amplifier on the output of the CCD before it goes to the A/D converter.

that means that at all settings other than the slowest, the image will clip BEFORE the CCD saturates (for any given pixel), so we use progressively less of the dynamic range of the sensor as we increase ISO. no wonder the noise levels go up!

now the K10D has a 22bit a/d converter, thats significantly more steps of dynamic range than photons per pixel at the slowest ISO, in other words its massive overkill. (which is good)

so seems to me that if we used 16bit or better raw files we would be capturing the entire dynamic range of the CCD anyway, brightness would get adjusted later, we'd never clip the image at low light, there would be no need to ever set the ISO, it would become another of those many camera settings that only affects the jpg. (or gets adjusted automatically to create the jpg)

so why isn't this happening? seems like it might even be possible with a firmware update to the K10D?

or have I misunderstood something?

I think the processing on the K10 is only 12-bit so you'd be hard pushed to get any 16-bit files out of it...

so what's the point of the much touted 22bit A/D then?

or are they just "marketing bits" like we so often see in audio....

http://www.dailytech.com/Major+Lowlight+Digital+Photography+Breakthrough+Inbound+from+Korea/article7401.htm

This could solve all our low light problems...

I'd like to see some actual real info on what it does different.

theres a fundamental limitation in terms of the number of photons hitting each pixel, and higher pixel counts and thus smaller pixels make that worse.

also using bayer filter arrangements to do colour essentially throws away 2/3 of the light so we need to get away from that too. 3CCD and a prism would be good but bulky and expensive for a DSLR.

its easier to make a low res sensor more sensitive, because you can have bigger pixels, but i'm not clear that this new technology (whatever it is) will apply to hi res sensors.

the ultimate sensor would count individual photons digitally, with no saturation point, and recording the wavelength (colour) of each photon. (although given the trichromatic nature of the eye the latter might be overkill for human viewing)

although we are still a fair way away from that, I think that real world photon counts will always put a limit on what can be done with low light.

the eye can detect a single photon (although not in colour!) but our vision still gets fuzzy/grainy in low light.

although, i guess, that to do images of typical current dynamic range ie 8bit, you could get away with only 256 photons per pixel per channel if you could count them accurately. or 768 photons per pixel total for full colour.

that certainly is low light, and i would assume that this new device is getting closer to that theoretical limitation, so it could be good!

the one they are selling is only 640x480 pixels I note!

I can't interpret the tech specs given here into bit depth dynamic range or number of photons required to make a difference in the output, can anyone else?

http://www.planet82.com/english/product/smpd_02.asp

they claim 120dB dynamic range, thats well in excess of 16bit and implies a very high photon per pixel saturation point at least compared with the 80,000 photons I've seen quoted as normal.

I was thinking about this a while ago, basically triggered by the equipment we sell at work.
We basically see CCD and CMOS sensors. If you look at spectrophotometers you see a lot of diode array sensors and NMOS sensors but I'm not sure if it is possible at all to use these for imaging purposes.

I had to look up to be sure what a spectrophotometer was , but i see that (as I thought from the name) it measures actual wavelength rather than just human perceived colour.

is that the bit of what I said (about recording the wavelength of each photon) that you were thinking about?

what is your work?

OK so I'm talking to myself here, but i realised that they are claiming smaller pixel size as well as higher sensitivity, so I guess it will apply to hi res sensors in due course.

Hi Thomas,

photons don't have a wavelength. Light at the quantum state is measured either as a particle (photon) or by it's wave like properties. You may have heard the term wave/particle duality. Generally we'd be concerned with the section of the electro magnetic field we perceive as light. These are our our wavelengths of light..

Err.. now it's time to wait for Mongoose.. he's the physics boffin.

Cheers

Liam

This allows more accurate conversion of the data, it is processed as 22-bit from the analogue signal and then compressed down to 12-bit to save a lot of space. The 22-bit to 12-bit conversion still gives a big tonal improvement from a straight 12-bit ADC.

individual photons have an energy, which corresponds to wavelength i think.

certainly its difficult to observe wavelike behaviour in a single quanta, but I think the concept of a wavelength of an individual photon has some validity.

if not then colour and wavelength must be built up through interaction between quanta. I'll look it up if i can find out what exactly to look up!

I'm not a physicist though...

a quick google found this:

http://focus.aps.org/story/v3/st21

which tends to support my understanding about it, even though its actually about soemthing more esoteric again

EDIT & addition: the famous two slit experiment, in which photons fired individually display interference effects (ie the single photon "interferes with itself" [!!]) would seem to indicate that single photons have wave like properties.

however i'm sure there would be quantum effects in a photon counting sensor as I was advocating, and it would be great if someone who knows could comment on what they might be.

exactly, but the resolution of the full 22bits is greater than the resolution of the sensor. so why do we need to change the pre ADC amplifier gain with the ISO setting?

if the full 22bit data were scanned for peaks before/during resampling down to 12bit, then we'd effectively have after the fact automatic iso and never have a clipped image unless the CCD was saturated (ie never clipped in light levels that we'd normally use faster iso speeds for.)

OK my optics is good, my quantum mechanics less so but if I recall correctly this is the deal:

light is really neither particle (photon) nor wave, these are just two ways of thinking about it which fit certain properties. No one has yet come up with a satisfactory way of thinking about radiation which explains all it's properties, which is why we are stuck with the messy "wave-particle duality" description.


Thomas (who I note is still not rhyming, although rhyming in quantum mechanics is not easy ) is quite right, the energy of a photon corresponds to the wavelength of the light via

wavelength = (plancs constant x the speed of light)/Energy in electron volts

so even if you only catch one photon, if you measure it's energy you can get the wavelength of the light and thus the colour. Of course catching single photons presents issues with measuring light intensity, but if it's that dim TBH it's probably time to put the camera away anyway.

Hi Mongoose,

I'm rapidly getting away from my comfort zone now.. not that that'll stop me makiong a fool of myself

This would mean then, that the speed of light must be a variable? Planks constant is just that, and with energy/mass being interchangeable, and photons having only a rest mass, that mass should be constant.. therefore the only variable on that side of the equation is C.. and it's also a constant..

Now we get into the Heisenberg Uncertainty principle, where if you measure something at the quantum level, you change it's properties by that act.. so that makes the measurement of the energy level of a single photon impossible, because when you do it err.. changes!?

I wish I'd listened more..

Cheers

Liam