are ISO settings a nonsense on digital?

For some more physics, I think i'm right in saying that this isn't possible. At a quantum level in the photo-electric effect -the basis of a CCD-, every conducting material has a saturation current - a point at which no matter how many photons hit the target, the photocurrent can't increase further.

i think not. my understanding is that photons have no mass at rest, since anything with at rest mass must have infinite mass at light speed and thus photons mass at lightspeed (which is the only speed you ever see them at!) is zero x infinity which is indeterminate. in fact i guess their mass at speed should be calculable from their energy but i don't know. certainly energy varies with frequency just like any other wave. (even though they are not really a wave)

I think it is possible to determine precisely the details of a quantum particle's state in the past, if you don't mind precluding knowledge of where it went next. is that right?

we don't care what happens to the photon after it hits the sensor, so i suspect that allows us to determine both its energy and its position when it does hit the sensor.

also we don't actually need to know its position very precisely, even the mini photosites on that new chip are 5-10 wavelengths across.

this is getting a bit speculative though.

I'd really like to know if the variable gain amplifier ISO stage can be dispensed with when using high bit depth A/D and existing sensors, and why it hasnt been done already, which is a question a pentax engineer could certainly answer if they wanted to...

of course its likely that i am misunderstanding something about how current cameras work...

I'd also like to know what the obstacles to an inherently noise free 'photon counter' type sensor are, where the only "noise" is not technology noise but random fluctuations in the photon stream, plus photon quantisation issues. this is fascinating because its one of the ways in which the real world is actually digital rather than analog and we could concievably capture it bit for bit.

I think you are right as far as CCD type tech goes.

but the sensor could be reset and keep on counting from 1 and add that to the total, like the next frame in a vidcam, or a sort of automatic HDR, or it could use a completely different sensing technology.

actually talking about the photon's rest mass is misleading since a photon can never come to rest, they have to be doing the speed of light all the time and in all frames of reference (and no, c is not a variable, but PLEASE don't make me go into relativity, that I really will have to go look up)

Heisenbergs uncertainty principle puts a limit on how accurately certain pairs of things can be measured, but I don't think it will ever come into play in camera design. As was mentioned earlier, we don't need to know either energy or position particularly accurately by quantum standards. You'd have issues with diffraction limits in the lens long before you hit Heisenberg limits in the sensor.

Hi Mongoose,

I thought I'd check on a science forum I'm a member of.. and I got it wrong.. as I thought I probably had. Here's the thread if you want a look.. don't worry, it has you as right and me as wrong..

http://www.sciencefile.org/cgi-bin/yabb2/YaBB.pl?num=1179941619/0

I don't know if you ever went to the BBC science boards when they were open.. this forum is a spin off from it.

Apologies for off topic-ness Matt/John.

Cheers

Liam

hi Liam, interesting thread that, populated by people with a far deeper understanding of relativity and QM than I have! Glad to see I was at least close to right.

except for the last post they don't seem to have understood that ISO setting in a digital camera is achieved by changing the amplifier gain on the CCD output.

but I don't have time to sign up to another forum to discuss it with them!

my original point was that by reading the CCD at lowest amplification and high bit depth to digitise the entire dynamic range of the sensor in one go we both reduce amplifier noise and eliminate the need for ISO setting entirely. assuming the ADC is capable of accurate conversion at very low input levels of course, but then if its isnt, then theres no point to the high bit depth anyway.

Thomas,
I have a feeling we discussed this before... (at least I have )
From what I understand, Pentax opted for removing the pre-amp and associated electronics and went for the 22-bit ADC. It was one of the head guys in Japan that mentioned this - and they did it to gain a better image "quality" (more film-like).

Like you, I reckon they have just slapped the CCD output (which is analogue) straight into the 22-bit ADC with no ISO-setting-preamp. And, like you, I came to the same conclusion.

What's more, I've mentioned it on more than one occassion and I hope the message has got back to Pentax Japan. At the very least, a 16-bit RAW format (or better a 22-bit) that works in conjunction with Silkypix would be really cool - it'd save me fiddling about with tons of exposures for HDR images, and allow all kinds of possibilities.

But... allowing that option does open rather a lot of other technological issues - such as the RAW convertor. There's no point Pentax releasing such a feature with nothing to support it!

And, besides, there is already an auto-ISO feature. Anything more and you're getting into HDR tone-mapping territory (which is a minefield, and certainly not simple).

I'm still holding out for an extended RAW format....

Matt

Matt, yes we have discussed at least part of this before. i was taking a new (for me anyway) angle on it

so you are saying that they've already done away with the variable amplifier!

which means that the ISO setting on the K10D is ALREADY done in software!

which is 99% of what i am advocating hardware wise. but it fails to really take advantage of doing it this way.

I'm assuming the ISO setting is still set pre exposure, but if the whole dynamic range of the sensor is already being captured then that makes no sense, the 22bit data could be scanned for peaks then normalised and dithered to fit the 12bit raw file, the files would be the same raw we have now, but theyd never be clipped except in high light, the brightest level would be preset to the brightest level in the file, and you'd never have to set the ISO.

If this is already what the K10 auto iso setting does i'll be impressed! but i think not! (i don't have one to test)

Thomas,
I think you've greatly oversimplified the issues

If you've ever dealt with wide dynamic range images you'll realise that it would be highly annoying to just "not clip". Sure, for flat and contrast-less images this would be fine. But for real images it'd be a right pain.

You'd find that you'd have detail in the highlights - sure - but you could never work out where the midtones are going to be

As I said, handling wide dynamic range images requires complex tone-mapping. I'd be happy to do that, but only at the PC.

Matt

I take your point, but i wasnt really meaning to do it that way as a perfect solution just as an improvement on what we have now that still allows the same raw format.

normalising to full scale will increase shadow detail too since any level steps available above the brightest recorded in the raw file are wasted. The system i describe ensures maximum DR possible for the 12bit file. of course a 16bit file would be better though!

in a raw file you don't really need to know where the midtones will be because its not attempting to be the final image, that can be worked out later.

the system i describe would work best if the brightest point in the pic is an important part of the image, if not it will throw away a little shadow detail in order to preserve the unecessary highlight detail, but it won't throw away as much as we currently have to to ensure that we don't clip using preexposure ISO setting, so its still an improvement unless you are in the habit of always clipping unecessary highlights by the exact perfect amount when shooting.

since clipping is always the absolute end of that part of the image i often try to expose to avoid it completely, this involves me in wasting DR to ensure that. i have to leave headroom. we shoot with raw because it gives us more headroom than 8bit, but a 12bit raw that doesnt need any headroom is better again, although still not as good as a 16bit raw.

so i wasnt advocating any more complex tonemapping than is already done by the ISO setting.

I do see what you mean though...

The point is that you don't always expose to get everything in. That's part of the creative process

To do what you want, you can just use spot metering and fully manual. Point at the highlights, get an exposure then add around 2Ev (because it metered as a mid grey) You can do that already.

Personally, I'd rather know where my midtones are.... since CCDs are linear, the detail is in the upper part of the histogram. In fact, as I've often said, the first 1-stop is in the upper 2048 values (of a 12-bit RAW), the next 1-stop occupies 1024 values, the next 1-stop occupies 512 values, and so on. If you are exposing so your detail is 3 or 4 stops underexposed you can guess what's happening to your detail


Frankly, I reckon the only "real" way is a full 22-bit RAW file with the JPG thumbnail processed exactly as it is now. But it'll mess up the histogram display on the camera too ('cause I reckon they work that out from the JPG to save time), as well as requiring a complex piece of RAW conversion software as well.

And then what about the K10D's ability to do in-camera RAW processing? How would they handle that with a 22-bit RAW. I can see the user interface getting unwieldy

But, I'm still hoping
Matt

does the auto ISO on the K10D work in smaller steps than the usual 200 400 etc? if so then i guess its going to be close with multi segnment metering to doing it quite well anyway...


that detail being in the upper part of the histogram is just bad design seems to me, surely that should be mapped on a standard curve anyway?

otherwise you are constantly on a knife edge between clipping and loss of detail...

it's not bad design, its a matter of physics

our eyes and photographic film respond logarithmically to light, CCDs respond linearly. Unless someone manages to produce a digital sensor with a logarithmic light response (and as I understand it that would need to be drastically different from any current system) it's something we are just stuck with.

well physics vs physiology if you like.

so when reducing resolution to a given file size it should be mapped in a way that is useful to our physiology.

light isn't made up of just red green and blue either, but we represent it like that because our eyes sense it like that.

see what I mean?