MF CMOS -- now it gets interesting

[Theodoros]

……….if that boosts dynamic range further (it's already low in the 35mm case).


Cheers,
Bart

It's low, but its "friendly" to the extend were many think that 35mm Cmos has more usable DR than MF CCD… where in fact its no where near!

People usually think of DR with sensor's ability to "save" the HL from blowing when ETTR… they don't consider if saving the HLs leads to a dull or low-contrast unreasonable presentation, while the "right" approach should be, "protect" HLs with exposure as much as one wants and dig into the deep shadows while keeping contrast and picture drama at the same time…. From that POV, CCDs are even more "sensitive" than they claim in their ISO specification.

[ndevlin]

By Photokina, there will be many CMOS MF backs in the ether, how many will be in-the-metal is a more interesting question.

While aspects of CMOS are nice, I am less hyped about this than many.  I don't need an MF camera to shoot high ISO.  Cleaner 400 would be nice, but for true high ISO work, I'm not carrying the bulk of an MF camera, with its slow lenses and inferior low-light AF.  That's work for the Fuji X cameras, or the Df. 

MF, for me at least, is for landscape and studio (or studio-lit location) work.  In that context, ISO is mostly irrelevant. The architechtural guys might love higher ISO for their location work, but it means zippo to me.  I also can't see shooting at other than base ISO for portrait work.  My E640s are never near full power.

Personally, I just like how CCD images look.  This chip will not likely create a better 'look' than the 50 or 60MP CCD backs.  What it will do is (further) crush the prices on those backs, which suits me just fine. 

As for the much-vaunted Liveview....I bought the 800e thinking LV would be the panacea to my high-res focussing woes. No to be. Nikon's implementation was so dreadful I find LV only marginally useful. On MF it would be (i) such a massive power draw and (ii) such a massive heat producer than I can't see this working well other than tethered.  If I'm shooting tethered, I can already  get the focus right, thank you vey much.

So yes, an interesting and inevitable innovation.  But a game chnager? Not so sure about that, though they could prove me wrong.

What if Pentax makes a camera with this chip? Handles beuatifully already, but has focus problems and shutter vibration probems.  CMOS ain't going to fix that.  Just sayin'.

- N.

[Christoph C. Feldhaim]

Fuji X Pro 1 blown up to MF.
Just dreaming ...

[EricWHiss]

Wait, back the the Lunar for a sec…   Is there anyone here that bought one?  Or rephrased - anyone here that will admit they bought one?   I would have thought they didn't sell well at all.  I'm happy to read that the brain behind that move has been replaced.   

As far as only 50mp for the CMOS entries by both Phase and HB - I'm betting there will be more versions soon enough.

No doubt that the CMOS backs will be great for higher ISO shooing, but I'm still dubious that these backs will be as nice as CCD's at base ISO.  Only time will tell.

[Misirlou]

Fuji X Pro 1 blown up to MF.
Just dreaming ...

Nah, Foveon. That would really be something.

[jerome_m]

Can someone please explain me what the hype is about cmos and high isos? I've tried an old H3D-31 in parallel with a D800 at isos 100 to 1600 and the difference was tiny. I mean iso 1600 was less than ideal on the H3D, but not much worse than on the D800.

Obviously, backs without micro lenses see their sensitivity divided by 2, but that is a direct function of the micro lenses, not of cmos or ccd.

[TMARK]

Maybe your D800 is broken.

Regards,

TM

Can someone please explain me what the hype is about cmos and high isos? I've tried an old H3D-31 in parallel with a D800 at isos 100 to 1600 and the difference was tiny. I mean iso 1600 was less than ideal on the H3D, but not much worse than on the D800.

Obviously, backs without micro lenses see their sensitivity divided by 2, but that is a direct function of the micro lenses, not of cmos or ccd.

[hjulenissen]

It's low, but its "friendly" to the extend were many think that 35mm Cmos has more usable DR than MF CCD… where in fact its no where near!

People usually think of DR with sensor's ability to "save" the HL from blowing when ETTR… they don't consider if saving the HLs leads to a dull or low-contrast unreasonable presentation, while the "right" approach should be, "protect" HLs with exposure as much as one wants and dig into the deep shadows while keeping contrast and picture drama at the same time…. From that POV, CCDs are even more "sensitive" than they claim in their ISO specification.

My mental model of sensor DR is that of an (essentially) linear photon counter that clips at some point, adds some noise at another point, and then introduce signal-dependant noise (shot-noise). I try to set my exposure parameters and interpret my histogram accordingly.

-h

[Theodoros]

Maybe your D800 is broken.

Regards,

TM

No, Mr. Marot's D800 isn't broken… in fact micro lenses absence, makes more "high iso damage" than Jerome says… If one considers P30+ and P45+ and compensates for the size difference, it's nearly 2 stops.

[design_freak]

Wait, back the the Lunar for a sec…   Is there anyone here that bought one?  Or rephrased - anyone here that will admit they bought one?   I would have thought they didn't sell well at all.  I'm happy to read that the brain behind that move has been replaced.   

+1
I agree in 100%

[Theodoros]

+1
I agree in 100%

LOL… I met a Leica fanatic once (pretty rich fellow) that bought the Panasonic alternative instead of the V-lux he was claiming he would buy next to his "M" system… I asked him: "why did you buy this instead?", he said… "when you are in vacations with a Leica, all thieves are watching your moves"… Wise move from Hassy, …wise indeed, they'll sell lots!   

[BJL]

There is a persistent myth about cost differences between CCDs and CMOS sensors. The fact is that CMOS sensors of "SLR size" and larger have never been less expensive than CCDs, and in fact during the transition, CCDs stayed around longest in the lower-priced DSLR models while the more expensive, better performing models changed to CMOS sensors. CMOS wins on performance, not price, particularly lower read noise and thus better high ISO performance.

This myth probably started from the case of very low-cost "camera-on-a-chip" devices for mobile phones and such, which reduced costs by combining a (very small) CMOS sensor and the related processing circuits on a single CMOS chip. But those cost savings are irrelevant once the sensor gets up to SLR or MF sizes.


Also, this lower noise advantage of CMOS sensors over CCDs is not just about different CFAs or microlenses: many DSLRs used to have CCDs from Sony (and Kodak for early Olympus DLSRs) with micro-lenses and with similar CFA designs as the CMOS sensors that displaced them, and the CMOS sensors won clearly on low-light performance. If Sony and Nikon could have made their (micro-lensed) CCDs match Canon's CMOS sensors at high ISO simply by changing the CFA designs, they would have -- instead, Sony had to develop its Expeed CMOS sensors to compete more effectively against Canon's CMOS.


Finally, a wild guess: what we are about to see revealed is one or more cameras using a Sony CMOS sensor in a near 645 size like 54x40mm and so with pixel pitch slightly larger than 6 microns -- because the target for these is those MF users who want features like better low-light performance, good DR and more usable Live View, and minimal cropping of their prime lenses relative to 645 film, and for whom 50MP is more than enough. Look for DR to be hyped in the marketing!

[eronald]

Price is directly related to chip size. Which is why the MF makers will keep selling smallish chips while explaining how large they are compared to dSLRs. In other words what they are selling is mostly "crop frame medium format".

Edmund

There is a persistent myth about cost differences between CCDs and CMOS sensors. The fact is that CMOS sensors of "SLR size" and larger have never been less expensive than CCDs, and in fact during the transition, CCDs stayed around longest in the lower-priced DSLR models while the more expensive, better performing models changed to CMOS sensors. CMOS wins on performance, not price, particularly lower read noise and thus better high ISO performance.

This myth probably started from the case of very low-cost "camera-on-a-chip" devices for mobile phones and such, which reduced costs by combining a (very small) CMOS sensor and the related processing circuits on a single CMOS chip. But those cost savings are irrelevant once the sensor gets up to SLR or MF sizes.


Also, this lower noise advantage of CMOS sensors over CCDs is not just about different CFAs or microlenses: many DSLRs used to have CCDs from Sony (and Kodak for early Olympus DLSRs) with micro-lenses and with similar CFA designs as the CMOS sensors that displaced them, and the CMOS sensors won clearly on low-light performance. If Sony and Nikon could have made their (micro-lensed) CCDs match Canon's CMOS sensors at high ISO simply by changing the CFA designs, they would have -- instead, Sony had to develop its Expeed CMOS sensors to compete more effectively against Canon's CMOS.


Finally, a wild guess: what we are about to see revealed is one or more cameras using a Sony CMOS sensor in a near 645 size like 54x40mm and so with pixel pitch slightly larger than 6 microns -- because the target for these is those MF users who want features like better low-light performance, good DR and more usable Live View, and minimal cropping of their prime lenses relative to 645 film, and for whom 50MP is more than enough. Look for DR to be hyped in the marketing!

[jerome_m]

Maybe your D800 is broken.

Maybe you have never seen how poor the D800 is before noise reduction or how good the H3D-31 is after noise reduction?

It is actually difficult to see the raw data of modern cameras without noise reduction: adobe camera raw (all adobe products) or apple camera raw (all apple products, including the preview function of os x and most non adobe software running on apple) include noise reduction even if you don't want it. It is automatic and very good at hiding the noise of the base sensor. If you want to see how poor the noise floor really is, you have to use free software based on dcraw ands even then the raw data itself is not 100% clean. Have you done that?

OTOH, images from old cameras posted on web sites devoted to reviews never gets updated with more recent processing. Images from an H3D taken in 2007 looked horrible in 2007. But if I use Phocus on that old camera today, I use a recent version of Phocus and that gives good results.

Does it mean there have been no progresses? No, but the progresses have mainly been in software and progresses as to the sensor noise level have been much smaller than most people believe. Which is not surprising: CCDs used in 2007 already achieved about 80% quantum efficiency (Kodak data sheets, not dxo mumbo-jumbo). There is not that much to be gained or we would start to split photons and that is not possible.

So, I am sorry, but I fail to share the excitement about cmos versus ccds, simply because ccds are not so bad that major progress is possible (remember that the same ccds are still used in professional astronomy, where noise levels really matter). cmos may allow us to select iso 25000 on our next MF back, as it does on recent 24x36 cameras, but the results would be as bad as it is on 24x36 cameras and I don't feel that this is a market for MF. Especially when one does not want micro-lenses, which is a net light loss of a full stop and one one consider that the average MF lens is 2 to 3 stops slower than what is available for 24x36 cameras. What would be the point?

Obviously, cmos should also allow the implementation of a decent live view function and that will be useful to many. It may even allow video and I am sure that filmmakers will be tempted by the elusive "MF look". But high isos? Not in an acceptable quality.

[jerome_m]

Also, this lower noise advantage of CMOS sensors over CCDs is not just about different CFAs or microlenses: many DSLRs used to have CCDs from Sony (and Kodak for early Olympus DLSRs) with micro-lenses and with similar CFA designs as the CMOS sensors that displaced them, and the CMOS sensors won clearly on low-light performance.

Not really, no. If I remember what the experienced astronomers measured at the time, Canon sensors at the time (e.g. in the EOS 20D) had lower noise in the readout circuitry. The reason is that the readout happens directly on the sensor with cmos, so it is easier to design a path better protected from outside noise. But this is certainly possible if one is willing to invest enough money in protecting the external readout circuits. MF manufacturers certainly had the money to do that.

[BJL]

Price is directly related to chip size.

Indeed: for SLR sizes and up, size is the dominant factor in the unit cost of sensors, not CCD vs CMOS, which is why I would not expect the addition of CMOS options to have much short-term effect on prices. Optimistically, it could help to bring prices down a bit in the longer term, if it expands the range of "use cases" for MF and so increases the pool of potential customers, and if the makers see more opportunity to increase profits by shifting a bit in the direct of "lower price, higher sales volume".

This is also why I expect that these MF-sized CMOS sensors will be available to other makers, even if Hasselblad has some initial exclusive, and even if Hasselblad has tuned this sensor's envelopment in exchange for a cut of revenue on sales to competitors: the cost savings through higher sensor production volume should in the end be attractive. More so if the MF makers see the market share battle between DMF and 35mm format as worth collaborating on.

[eronald]

Not really, no. If I remember what the experienced astronomers measured at the time, Canon sensors at the time (e.g. in the EOS 20D) had lower noise in the readout circuitry. The reason is that the readout happens directly on the sensor with cmos, so it is easier to design a path better protected from outside noise. But this is certainly possible if one is willing to invest enough money in protecting the external readout circuits. MF manufacturers certainly had the money to do that.

Actually, what you wrote above has good points, this last I regretfully feel is not correct. Getting a signal off a chip at high readout frequencies costs power, which means strong amplification, which means introducing noise as well as all the assymetries and banding effects inherent in multiple readouts through dis-assorted discrete components - if you can convert on-chip you should win something.

But I would agree that the base technology matters less in a way than the implementation, and there is little reason to think that V1.0 of the MF CMOS sensors will have much to offer over the existing products. If anyone here is throwing their H5D-50 or IQ-280 in the bin, I will be very pleased to come and collect and do my bit to recycle used electronics

Edmund

[jerome_m]

Actually, what you wrote above has good points, this last I regretfully feel is not correct. Getting a signal off a chip at high readout frequencies costs power, which means strong amplification, which means introducing noise as well as all the assymetries and banding effects inherent in multiple readouts through dis-assorted discrete components - if you can convert on-chip you should win something.

It is a trade-off. On chip conversion can only have so much space, power and complexity. Getting the analog signals out of the chip indeed adds its share of problems but, given enough care (that is: $$$ for the design…) it can be quite good indeed. There must be a reason why the big telescopes which costs immense sums of money still use CCDs.

There has been a camera with a cmos sensor and an external converter, BTW: the Nikon D3x. It uses the same sensor than the Sony A900 but rumour is that Nikon replaced the converters by external ones (and uses 14 bits instead of 12).

[TMARK]

Edited to add:  I apologize for the profanity and tersness of my post.  I'm sick and cranky.

I don't care about any of this shit.  Quantum efficiancy blah blah fucking blah.  Where the rubber meets the road is what is important.  If I could have had a back that produced a clean 1600 - 2000iso in 2008, when we started in motion and I lit everything with ARRI Fresnel's and HMIs, I would have kept using the backs.  At the time the only option was the P65+ using sensor plus.  The fuck if I was going to spend $40k for a 15mpx file for a still image, which in terms of ad production work, is an after thought.  I had already sold my P30+ (same chip as the H31?) to spend copious amounts of money on Red stuff, which actually returned the investment quickly, even in the depths of the recession.

And no, I'm not excited about CMOS.  I think it will be different.  If it does make a clean file at 1600 - 2000iso, it re-opens the door to MF for many professional ad shooters.  I think that's great.

Maybe you have never seen how poor the D800 is before noise reduction or how good the H3D-31 is after noise reduction?

It is actually difficult to see the raw data of modern cameras without noise reduction: adobe camera raw (all adobe products) or apple camera raw (all apple products, including the preview function of os x and most non adobe software running on apple) include noise reduction even if you don't want it. It is automatic and very good at hiding the noise of the base sensor. If you want to see how poor the noise floor really is, you have to use free software based on dcraw ands even then the raw data itself is not 100% clean. Have you done that?

OTOH, images from old cameras posted on web sites devoted to reviews never gets updated with more recent processing. Images from an H3D taken in 2007 looked horrible in 2007. But if I use Phocus on that old camera today, I use a recent version of Phocus and that gives good results.

Does it mean there have been no progresses? No, but the progresses have mainly been in software and progresses as to the sensor noise level have been much smaller than most people believe. Which is not surprising: CCDs used in 2007 already achieved about 80% quantum efficiency (Kodak data sheets, not dxo mumbo-jumbo). There is not that much to be gained or we would start to split photons and that is not possible.

So, I am sorry, but I fail to share the excitement about cmos versus ccds, simply because ccds are not so bad that major progress is possible (remember that the same ccds are still used in professional astronomy, where noise levels really matter). cmos may allow us to select iso 25000 on our next MF back, as it does on recent 24x36 cameras, but the results would be as bad as it is on 24x36 cameras and I don't feel that this is a market for MF. Especially when one does not want micro-lenses, which is a net light loss of a full stop and one one consider that the average MF lens is 2 to 3 stops slower than what is available for 24x36 cameras. What would be the point?

Obviously, cmos should also allow the implementation of a decent live view function and that will be useful to many. It may even allow video and I am sure that filmmakers will be tempted by the elusive "MF look". But high isos? Not in an acceptable quality.

[BJL]

Canon sensors at the time (e.g. in the EOS 20D) had lower noise in the readout circuitry. The reason is that the readout happens directly on the sensor with cmos, so it is easier to design a path better protected from outside noise.

That is a fair explanation of the inherent noise advantage of active pixel CMOS sensors over CCD's; and I emphasize the full name including "active pixel", because that is a fundamental difference from CCD's.

CMOS sensors amplify the signal early (during direct transfer from photosites to sense capacitors at the edge of the sensor), reducing the effects of subsequent noise. Further, the newer column-parallel ADC approach (now used by everyone except Canon) also eliminates any further transport of the analog signal by converting to digital directly at each sense capacitor.

But this is certainly possible if one is willing to invest enough money in protecting the external readout circuits. MF manufacturers certainly had the money to do that.

No it is not possible with a CCD. A fundamental feature of a CCD sensor is that it is a passive device, in which the charge produced at each photosite by the photo-electric effect is transferred in thousands of hops from one photosite to the next: first hopping down to the sensor's edge, and then along the sensor's edge to one corner. Only at the corners can any amplification or charge-to-voltage conversion be done. This limit to can be overcome in only one way: by changing to an active approach with earlier amplification --- and that is exact the design change that lead to active pixel CMOS sensors and differentiates them from CCDs.


Trivia: most or all current CCDs are made using CMOS process; the short-hand "CMOS" for "active pixel CMOS" is this misleading, since the fundamental difference from CCDs is the "active" not the "CMOS".