Hypothetical high iso question

[gad-westy]

Something I've been pondering. Recently there has been a lot of talk about the d800 possibly having a 36mp sensor and the general consensus seems to he that this will result in poor noise control.

So what I'm wondering is why there are no low mp dslrs with incredible high iso capability? A D3s for example might be used at a wedding where print sizes mean that anything more than about 6mp is pointless. So if a d3s sensor was only 6mp say, would its noise handling be even better? And if so, would there be a market.?

Typed this on my phone. Apologies if apple have converted it to gibberish.

 

[woof woof]

I think it's because the makers marketing departments and large sections of the great unwashed want a nice easy number to think about. 20mp must be better than 10, marketing nailed, easy peasy.

 

[gad-westy]

I think it's because the makers marketing departments and large sections of the great unwashed want a nice easy number to think about. 20mp must be better than 10, marketing nailed, easy peasy.

I did wonder that but I would have thought that there may be a market at the higher end of the scale where results are everything. I doubt most pro usage actually requires high mp but I can envisage a lot of uses for clean high ISO. I do realise that the best full frame cameras aren't exactly lacking in this department to begin with but I do wonder 'what if'?

 

[woof woof]

Same here.

I can make an A3 print with my 8mp 20D but while I can accept that others want to photograph a starlings eyelashes at 600m and therefore need a high mp count camera so that they can throw away 75% of the image and still print A3 I do wish that there was the option of buying a state of the art low mp count camera and being able to shoot in the dark or even just shoot clean ISO 100 daylight shots without having to use noise reduction.

I think it's also a macho thing, "I need 30mp" being a real mans statement and "8mp is quite enough thank you" being something only a weedy little bloke who gets sand kicked in his face would say.

 

[HoppyUK]

Nikon D4, 16mp, ISO 204,800.
Nikon D3x, 24mp, ISO 6,400.

The principle at least is well enough established. The new Canon 1Dx has a similar high ISO capability to the D4, but less pixels than the 1DsMk3 and 5D2.

 

[woof woof]

The 5D was 12mb, I'd like to see a 12mb FF state of the art.

 

[gad-westy]

The 5D was 12mb, I'd like to see a 12mb FF state of the art.

Given the strides made with sensor technology, presumably even a cropper could perform exceptionally well if it weren't hobbled with such a high pixel density. The Nikon D7000 (and any other camera that uses that sensor) performs miracles with 16mp so I'd love to know what it could do with 6mp.

 

[woof woof]

Indeed.

 

[ausemmao]

Something I've been pondering. Recently there has been a lot of talk about the d800 possibly having a 36mp sensor and the general consensus seems to he that this will result in poor noise control.

So what I'm wondering is why there are no low mp dslrs with incredible high iso capability? A D3s for example might be used at a wedding where print sizes mean that anything more than about 6mp is pointless. So if a d3s sensor was only 6mp say, would its noise handling be even better? And if so, would there be a market.?

Typed this on my phone. Apologies if apple have converted it to gibberish.

Because unlike most on forums, the engineers at camera companies actually understand physics and signal theory and know that provided the steppers can make the features the right size, larger pixels provide no image level benefit at all.

 

[ausemmao]

Nikon D4, 16mp, ISO 204,800.
Nikon D3x, 24mp, ISO 6,400.

The principle at least is well enough established. The new Canon 1Dx has a similar high ISO capability to the D4, but less pixels than the 1DsMk3 and 5D2.

1D3 10 MP
1Ds3 21MP
Despite the similar tech and smaller pixels and same indicated ISO limits of the 1Ds3 it outperforms the 1D3 quite handily. That should be enough to show the lagest determinant of most factors related to IQ.

 

[woof woof]

Because unlike most on forums, the engineers at camera companies actually understand physics and signal theory and know that provided the steppers can make the features the right size, larger pixels provide no image level benefit at all.

 

[HoppyUK]

Because unlike most on forums, the engineers at camera companies actually understand physics and signal theory and know that provided the steppers can make the features the right size, larger pixels provide no image level benefit at all.

It's an interesting subject, but by and large the real world experience is that fewer pixels over a given area is less noisy.

This is only a hunch, but one thing we don't hear much about is how much light 'falls between the cracks' of the pixels. Marketing folks talk about gapless microlenses and stuff, but I just don't believe that every photon is collected, or anything like it TBH.

It stands to reason that even if losses in the gaps are minimised, the more gaps you have, the more you'll lose. Until we have truly gapless pixels and 100% efficient microlenses, lower pixel density will always mean more light collected, and less noise.

Having said that, we now seem to have reached a level with the 1Dx and D4, where extra potential image quality is being traded for mega-ISO. Fine if that's what you want, but there are other ways of balancing the benefits.

Such as much higher pixel density, eg rumoured 36mp Sony sensor possibly heading for a new FF Nikon. Another way of doing it, based on the 'how much image quality do you need' question, is to reduce sensor size and gain the inherant benefits available down that road.

I think the future of crop formats and 4/3rds is bright, but it will need better lenses to also step forward - more sharpness, and lower f/numbers.

 

[ausemmao]

It's an interesting subject, but by and large the real world experience is that fewer pixels over a given area is less noisy.

Also untrue.

Every DSLR on the market is beaten for quantum efficiency (that's the percentage of light hitting the sensor that becomes electrons) and read noise (the noise added by all the electronics between hitting the photosite and being written to a file) by compact camera sensors. The sensors that come closest are the G3's sensor and the D3s. (fun fact: if Canon did nothing but make the 1D4's sensor FF, that would make it near indistinguishable or maybe slightly better than the D3s for high ISO noise but with better DR and 26MP. Would be kind of expensive though (IIRC they would need to retool a factory to be able to do multiple passes on the same wafer.).

In other words, right now it is possible to make a 100MP sensor with better than D3s noise performance at the image level.

There are many reasons this isn't done (for one, we don't have the read electronics to get a decent framerate out of it given the frequency that they'd need to run at), but the biggest is cost. DSLR sensors tend to get made on junk steppers. For reference, i think Canon still uses 130 if not 180nm equipment for its sensors where CMOS cutting edge is 22nm (but that won't get used for image sensors for a long time). It's not cost effective to use newer tech for DSLRs when the rest of the imaging chain hasn't caught up, and the cost required would be astronomical. You can get 20 1/1.7 (s95 size sensors) in the same wafer area as a FF sensor, and you'll get better yields from them to boot given a defect will only affect 1 of those 20 where it would wreck that whole FF sensor. So even if the electronics could keep up, I don't think there's much of a market for a £30,000+ FF DSLR

This is only a hunch, but one thing we don't hear much about is how much light 'falls between the cracks' of the pixels. Marketing folks talk about gapless microlenses and stuff, but I just don't believe that every photon is collected, or anything like it TBH.

You're right here. The best CMOS sensors collect about 60% of the light after the CFA (again, looking at sensors pixel size simply doesn't come into it in DSLR or even compact camera ranges). There's light lost to reflection off the microlenses, off the photosite itself, off the wiring, then light that penetrates too far...
That's not counting the light absorbed by the CFA itself. We could try using 3 sensor cameras with dichroic prisms, but if you thought focus microadjust was an issue now.....

It stands to reason that even if losses in the gaps are minimised, the more gaps you have, the more you'll lose. Until we have truly gapless pixels and 100% efficient microlenses, lower pixel density will always mean more light collected, and less noise.

For DSLR pixels, the limiting factor isn't pixel size.

Having said that, we now seem to have reached a level with the 1Dx and D4, where extra potential image quality is being traded for mega-ISO. Fine if that's what you want, but there are other ways of balancing the benefits.

Such as much higher pixel density, eg rumoured 36mp Sony sensor possibly heading for a new FF Nikon. Another way of doing it, based on the 'how much image quality do you need' question, is to reduce sensor size and gain the inherant benefits available down that road.

I think the future of crop formats and 4/3rds is bright, but it will need better lenses to also step forward - more sharpness, and lower f/numbers.

I don't believe theoretical sensor IQ is the limit here. The surrounding systems (cost of making the sensor, possible framerates and heating issues, AF accuracy, ability to manufacture to sub micron accuracy with small components to a reasonable cost, and not least consumer misinformation (businesses that tell their customers that they're plain wrong don't tend to do well unless very, very specialised)) are much more of a problem.

 

[Canon Bob]

.....Until we have truly gapless pixels and 100% efficient microlenses....

If we had "truly gapless pixels" then why would we need microlenses?

Bob

 

[HoppyUK]

Thanks for your reply, but if you say this:

Also untrue.

Then you cannot also say this:

You're right here.

It's a complicated question because there are theories on one hand, and practise on the other that often appears to contradict it. And I accept that there are other manufacturing and cost issues that could well be overriding.

The point I was trying to make was simply that if you slice and dice a given area into ever smaller pieces, the the gaps between the pieces will take up an ever increasing percentage. Therefore, more pixels will inevitably mean less raw data for the rest of the imaging chain to work with (and a whole bunch more theories and practise to go with that).

However, the thing I was saying is that there comes a point when the law of diminishing returns means that whatever we've got, it's more than enough for most purposes. And that opens up an opportunity to cut the cake differently.

For example, I don't need or even want 204,800ISO. It's useless to me, gives no control of shutter speeds and depth of field in normal light. I'll trade that for a 15mp crop sensor, still with more ISO capability than I'll probably use. I'll get more long lens reach and smaller formats make low f/numbers possible, super-sharp f/2 zooms and f/1.0 primes. That's not going to happen next week, but just checking my crystal balls, I think it might be a (continuing) trend over the next five years.

 

[HoppyUK]

If we had "truly gapless pixels" then why would we need microlenses?

Bob

Haha I don't think we'll ever get gapless pixels the way things work at the moment.

The actual light sensitive bit is a much small area sitting in the bottom of the pixel well. The microlens sits over the top to collect and channel the light from the whole pixel area down to the bottom.

This all sounds great in theory, but we are talking about such microscopically small components that the word 'lens' hardly applies, and the term 'gap-less' is marketing speak for a bit closer than they were before.

Either way, a lot of the light falling on the pixel area is lost before it even gets to enter the imaging chain, and my point above is that the more pixels you have, the greater proportion of the total area is wasted.

 

[ausemmao]

Thanks for your reply, but if you say this:



Then you cannot also say this:

I can, because they don't contradict one another.

Statement 1:

Sensor area efficiency is not negatively affected by pixel size at DSLR scales. Evidence: Comparing best DSLRs to best compacts per unit area.

Statement 2:

Not every incident photon is collected by camera sensors due to CFA, wiring etc. This is unrelated to pixel size at DSLR scales, and affects every CMOS sensor in existence.
Evidence: Comparing quantum efficiency between sensors and knowing the CFA exists.

In other words, there is light lost to all sorts of things on CMOS sensors. This quantity of light lost is not significantly affected by pixel size in DSLR scale sensors.

'Gapless microlenses' is a marketing term. Even if they were truly gapless, it's far from sufficient to improve IQ.

It's a complicated question because there are theories on one hand, and practise on the other that often appears to contradict it. And I accept that there are other manufacturing and cost issues that could well be overriding.

The point I was trying to make was simply that if you slice and dice a given area into ever smaller pieces, the the gaps between the pieces will take up an ever increasing percentage. Therefore, more pixels will inevitably mean less raw data for the rest of the imaging chain to work with (and a whole bunch more theories and practise to go with that).

The thing is, the actual light sensitive part of the sensor is a lot less than I think you think it is (clearly untrue based on your post above )(it's why microlenses exist. The other main reason is that the sensors respond horribly to off axis incident light - creates crosstalk issues, isn't absorbed properly, can be reflected. It's an even bigger issue for Foveon sensors). The actual light sensitive area alone doesn't affect how good the sensor will be, except that making the gap between the wells too small increases crosstalk, and deeper wells increases full well capacity, which can increase dynamic range.

However, the thing I was saying is that there comes a point when the law of diminishing returns means that whatever we've got, it's more than enough for most purposes. And that opens up an opportunity to cut the cake differently.

For example, I don't need or even want 204,800ISO. It's useless to me, gives no control of shutter speeds and depth of field in normal light. I'll trade that for a 15mp crop sensor, still with more ISO capability than I'll probably use. I'll get more long lens reach and smaller formats make low f/numbers possible, super-sharp f/2 zooms and f/1.0 primes. That's not going to happen next week, but just checking my crystal balls, I think it might be a (continuing) trend over the next five years.

This is true. It's a dilemma (and an opportunity) for the camera companies. They will not be able to sell cameras solely on the 'the image is better' basis for much longer because we are almost certainly at the 80/20 point for FF and APS-C (we will definitely be once fabrication costs fall further and segmentation based on sensor size goes away). Usability will become far more important.

That said, I'm the other way round. I really want better high ISO (I can end up at ISO 1600-3200 on an overcast day shooting sports, as below 1/1000 is too slow for wakeboarding, and I like night shooting) though I accept I'm in a niche. Luckily, most things done to improve IQ tend to improve it across sensitivity levels.

 

[ausemmao]

If we had "truly gapless pixels" then why would we need microlenses?

Bob

Even if pixels were gapless (they won't ever be because of crosstalk, wiring and because it would prevent microlenses doing their job among other things) microlenses would still be needed to make the incident light as close to on axis as possible.

 

[Flash In The Pan]

My eyes are bleeding.

 

[Canon Bob]

Even if pixels were gapless (they won't ever be because of crosstalk, wiring and because it would prevent microlenses doing their job among other things) microlenses would still be needed to make the incident light as close to on axis as possible.

I appreciate that gapless pixels (in the current sense of things) aren't feasible (hence the quotation marks).
The microlenses are, by neccessity, a compromise at the moment and this compromise is making larger aperture lenses a little less efficient than we would like. An f/1.2 lens feeding a 1Ds3 or 5D2 sensor is losing about 1/3 stop of light as the efficiency of the microlenses is aimed at slightly smaller apertures to cater for the more mainstream lenses. At f/1.4 then the difference is much less and by f/2 it can't be measured (by me at least).
Faster lenses aren't the answer until the next part of the system is redisgned to cope with them.

Bob

 

[ausemmao]

I appreciate that gapless pixels (in the current sense of things) aren't feasible (hence the quotation marks).
The microlenses are, by neccessity, a compromise at the moment and this compromise is making larger aperture lenses a little less efficient than we would like. An f/1.2 lens feeding a 1Ds3 or 5D2 sensor is losing about 1/3 stop of light as the efficiency of the microlenses is aimed at slightly smaller apertures to cater for the more mainstream lenses. At f/1.4 then the difference is much less and by f/2 it can't be measured (by me at least).
Faster lenses aren't the answer until the next part of the system is redisgned to cope with them.

Bob

Yep - but until someone finds a way to make the silicon itself respond properly to oblique light, that won't happen

 

[HoppyUK]

I appreciate that gapless pixels (in the current sense of things) aren't feasible (hence the quotation marks).
The microlenses are, by neccessity, a compromise at the moment and this compromise is making larger aperture lenses a little less efficient than we would like. An f/1.2 lens feeding a 1Ds3 or 5D2 sensor is losing about 1/3 stop of light as the efficiency of the microlenses is aimed at slightly smaller apertures to cater for the more mainstream lenses. At f/1.4 then the difference is much less and by f/2 it can't be measured (by me at least).
Faster lenses aren't the answer until the next part of the system is redisgned to cope with them.

Bob

Yes, it's all a balancing act - the whole imaging train has got to perform at the same level or we'll never see the benefits of the good bits.

Couple of examples - the new 1Dx and D4 have lower pixel counts than perhaps they might, and one reason for that is the processing engine that can't cope with any more and still run at 10-12fps.

The AF system is another limiting factor, in that it needs to get a good look at the subject to lock on and track. The key thing there appears to be the mirror, and both Nikon and Canon have spent a lot of effort speeding that up and reducing bounce.

Here's an idea - a concave sensor! Preferably with a variable curve!!!

 

[modchild]

I think the producers have been thinking the same, which is why they have dropped the MP count on their latest models and managed to get better photos with less noise (or so the claim goes) and brilliant resolution at the same time. I love the way my 5D2 handles noise already so a big improvement on that even with a few less MP would be incredible.

I'd be more than happy with 5fps if they were all perfect photos than 10fps where half are flawed. I really can't see birds flying faster or people playing football or baseball faster so why we need faster fps all the time is beyond me.

 

[Canon Bob]

..... I really can't see birds flying faster or people playing football or baseball faster so why we need faster fps all the time is beyond me.

Professional sports togs would probably disagree Stuart. Put some numbers between the shots and it starts to make a little sense. A cricket ball will typically travel about 4.5m between shots at 10fps, a tennis ball over 5m and a football around 3m. Double those distances at 5fps and the ball is out of the frame before you can react.
10fps still leaves a lot of anticipation required to get the money shot at times.

Bob

 

[Splog]

Also untrue.

Every DSLR on the market is beaten for quantum efficiency (that's the percentage of light hitting the sensor that becomes electrons) and read noise (the noise added by all the electronics between hitting the photosite and being written to a file) by compact camera sensors. The sensors that come closest are the G3's sensor and the D3s. (fun fact: if Canon did nothing but make the 1D4's sensor FF, that would make it near indistinguishable or maybe slightly better than the D3s for high ISO noise but with better DR and 26MP. Would be kind of expensive though (IIRC they would need to retool a factory to be able to do multiple passes on the same wafer.).

In other words, right now it is possible to make a 100MP sensor with better than D3s noise performance at the image level.

There are many reasons this isn't done (for one, we don't have the read electronics to get a decent framerate out of it given the frequency that they'd need to run at), but the biggest is cost. DSLR sensors tend to get made on junk steppers. For reference, i think Canon still uses 130 if not 180nm equipment for its sensors where CMOS cutting edge is 22nm (but that won't get used for image sensors for a long time). It's not cost effective to use newer tech for DSLRs when the rest of the imaging chain hasn't caught up, and the cost required would be astronomical. You can get 20 1/1.7 (s95 size sensors) in the same wafer area as a FF sensor, and you'll get better yields from them to boot given a defect will only affect 1 of those 20 where it would wreck that whole FF sensor. So even if the electronics could keep up, I don't think there's much of a market for a £30,000+ FF DSLR



You're right here. The best CMOS sensors collect about 60% of the light after the CFA (again, looking at sensors pixel size simply doesn't come into it in DSLR or even compact camera ranges). There's light lost to reflection off the microlenses, off the photosite itself, off the wiring, then light that penetrates too far...
That's not counting the light absorbed by the CFA itself. We could try using 3 sensor cameras with dichroic prisms, but if you thought focus microadjust was an issue now.....



For DSLR pixels, the limiting factor isn't pixel size.




I don't believe theoretical sensor IQ is the limit here. The surrounding systems (cost of making the sensor, possible framerates and heating issues, AF accuracy, ability to manufacture to sub micron accuracy with small components to a reasonable cost, and not least consumer misinformation (businesses that tell their customers that they're plain wrong don't tend to do well unless very, very specialised)) are much more of a problem.

Pixel size is a limiting factor and always will be! A QE of 100% is impossible, therefore we have to take the best we can, different sensors behave differently and the whole point of sensors is collect photons, count them and then pass that signal for processing. If the photosite is small it will collect fewer photons and have to process that weak signal either more often or guesstimate.

ISO is a bit of a falsity in that the sensitivity is quite often an algorithm within the processing, yes the sensor can have the gain increased but this is at the expense of electrical signal creating noise. If upping the gain had no adverse effect on SNR then all cameras would be built that way.