new DxOmark test results

[ejmartin]

 Maybe you could clarify the relationship between S/N and DR ?

Edmund

S/N is the ratio of the signal (illumination level) to the amount of noise at that signal level; DR is the range of illumination levels (S_min to S_max) for which the S/N exceeds whatever quality threshold you wish to impose.  Engineering DR takes that threshold to be (S/N)_min = 1; most photographers would ask for something substantially higher.  Thus S/N and DR are not directly related.  They are indirectly related in that light carries its own noise (photon shot noise) for which the noise level is the square root of the number of photons captured, and thus goes as sqrt(S), for a S/N~sqrt(S); this imposes a constraint on how low S can go before S/N drops below your threshold for acceptable quality.  Another constraint is electronic read noise, which increases the noise and therefore may make S/N drop below your threshold sooner than if the noise was photon noise alone.

Larger sensors gather more light for a given f-ratio (but also have less DoF), so for fixed exposure and similar read noise will have a higher DR, but not by a lot; I think rather that much of what people are seeing as a 'DR advantage' of MFDBs is actually a higher S/N over midtones caused by the greater light gathered rather than a greater range in shadows where the S/N, while poor, exceeds the quality threshold.

[eronald]

Bill,

 Thank you very much for the reference.
 
 As I am getting every day more lazy and no less dumb in my geriatric age, would you indulge me and tell me how one would choose a (photographically meaningful) noise floor?

Edmund

Edmumd,

It is all explained in the link to Emil Martinec that I provided earlier but perhaps you didn't see. He explains it more clearly and authoritatively than I could. Basically, the DR is the number of stops from saturation down to the chosen noise floor. The engineering definition uses a SNR of 1.0, which is too low for practical photograrphy.

Regards,

Bill

[Bart_van_der_Wolf]

Bill,

 Thank you very much for the reference.
 
 As I am getting every day more lazy and no less dumb in my geriatric age, would you indulge me and tell me how one would choose a (photographically meaningful) noise floor?

Hi Edmund,

The great thing about the most strict 'engineering definition' of DR is that it, as a rule of thumb, also applies to more relaxed definitions. Hence, better is best ...

Cheers,
Bart

[eronald]

Hi Edmund,

The great thing about the most strict 'engineering definition' of DR is that it, as a rule of thumb, also applies to more relaxed definitions. Hence, better is best ...

Cheers,
Bart

Hi Bart,

 I assume you would mean an SNR of 1 (0 db) but don't think that would necessarily be considered very strict ...
 Or, more probably I am missing something (a lot?).

Edmund

[Bart_van_der_Wolf]

Hi Bart,

 I assume you would mean an SNR of 1 (0 db) but don't think that would necessarily be considered very strict ...
 Or, more probably I am missing something (a lot?).

Hi Edmund,

How could it be any more strict than that? And, as long as that (proportionally) translates o less strict (is higher) noise thresholds, it would offer more playroom for more relaxed criteria.

Cheers,
Bart

[bclaff]

BCLAFF ... a forum contributor ... has done excellent work on measuring what he calls PDR - Photographic Dynamic Range.

http://home.comcast.net/~NikonD70/Charts/PDR.htm

Unfortunately, it doesn't have any MFDBs but has a nice selection of DSLRs.

I accept PDR files in any raw format that can be converted to DNG.
Anyone wanting to contribute should simply contact me by email for details.
Generally I want one shot per ISO of a particular target that I provide.

:-)
Bill

[bjanes]

Bill,

 Thank you very much for the reference.
 
 As I am getting every day more lazy and no less dumb in my geriatric age, would you indulge me and tell me how one would choose a (photographically meaningful) noise floor?

Edmund

Edmund,

That is pretty much a personal decision. Also, one should keep in mind that pattern noise such as banding (noted on some older Canon cameras) is more disturbing than white noise, so the SNR floor may not tell the whole story. Emil has a demonstration of S:N ranging from 0 to 8 (see his figure 13). Perhaps Bart can comment--he always provides a wealth of good information.

Regards,

Bill

[bclaff]

how one would choose a (photographically meaningful) noise floor?

For PDR I chose and SNR of 20 for the "standard" Circle of Confusion (CoC).
I chose 20 because ISO considers 10 "acceptable" and 40 "excellent".
The "standard" CoC is .033mm for FX, etc.

:-)
Bill

[eronald]

Bill (bjanes),
 Thank you.

Bill (bclaff)
 Thank you.

I will look through that site of emil's.
At the moment I quite like my D4; however, I am idly wondering whether as it uses amplification for low ISOs one could not extract much more DR from it by some trick such as converting both the unamplified (high bits) and amplified (low bit) data.

Edmund

For PDR I chose and SNR of 20 for the "standard" Circle of Confusion (CoC).
I chose 20 because ISO considers 10 "acceptable" and 40 "excellent".
The "standard" CoC is .033mm for FX, etc.

:-)
Bill

[torger]

2012 vs 2004  
http://dl.dropbox.com/u/17452116/DynamicRange_5DMIIIvsAptus22.jpg

It should be noted that Canon is not near Sony Exmor in CMOS noise performance at base ISO. Additionally, to make a fair test you cannot blow highlights more in one shot. There seems to be more highlight detail in the 5D shot, although it is a bit hard to judge since the Aptus 22 is more out of focus in the bright background. But anyway I do expect the Aptus 22 to win over 5D mark III. Canon doesn't have Sony Exmor technology, only Sony has. It can also be noted that Aptus 22 is still quite competitive with 2012 MF backs in terms of DR. The development of CCD tech has not been as dramatic as with CMOS, but on the other hand noone can at base ISO do as good as Sony, so the best CMOS tech is not found in all cameras, in no Canon camera for example. With MF CCDs as far as I know each one since 2003 has had great DR, so you can't really get a bad one there, at least if you will shoot at base ISO and have short exposure times.

If one wants to be pleased with an old 22 megapixel back the best way is to compare with a Canon, I know I will :-) (I plan to get an Aptus 22 or P25 myself...)

[torger]

Perhaps you should recheck your math.

Uhh... you're probably right, but the comparison was with the D5100 in that case though, but I thought 3 dB = 1 stop, 48.7 vs 41.2 that is 7.5 dB, > 6 dB, larger than 2 stops. I guess the large difference in midtones is due to photon shot noise...

[torger]

       I wonder why no MFDB has gone CMOS? A double-size D800E sensor might be much cheaper to produce than the IQ180 sensor, and perhaps as good?

I think it is because that it is not so easy. Today only Sony can make CMOS sensors that is up there competing with MF CCDs in performance. Canon can't, Nikon can't (but use Sony sensor in D800), Dalsa can't. It is also a bit tricky to make large chips, so it could be the case that noone has the technology to make a CMOS chip of that size (which are more complicated than CCD). Concerning cost my guess is that it would be more expensive to make due to being more complicated, perhaps leading to lower yields.

Fast refresh liveview directly on the back, great ISO performance, long exposures, comparable (or perhaps better) low ISO performance. It would surely be a market for that thing, but someone must have the capability to manufacture it. Sony may be able to do it, but MF would be a whole new market for them, and much less profitable than the smaller sensors. We will see what happens.

[eronald]

I read http://www.sony.net/SonyInfo/technology/technology/theme/cmos_01.html

It seems Exmor sensors are made with a dedicated CU wiring process, and something in the article talks about low silicon defects which makes me think they may have an additional wafer processing step, or maybe qualify the wafers in some way. But I see no reason why other companies cannot at least copy the architecture, invent something similar, or just license/subcontract MF sensors from Sony. After all, all their competitors are also their customers for sensors.

Another question is when sensors on dSLRs are going to be back-illuminated? The Sony site states there are 8db to be gained there ...

Edmund

I think it is because that it is not so easy. Today only Sony can make CMOS sensors that is up there competing with MF CCDs in performance. Canon can't, Nikon can't (but use Sony sensor in D800), Dalsa can't. It is also a bit tricky to make large chips, so it could be the case that noone has the technology to make a CMOS chip of that size (which are more complicated than CCD). Concerning cost my guess is that it would be more expensive to make due to being more complicated, perhaps leading to lower yields.

Fast refresh liveview directly on the back, great ISO performance, long exposures, comparable (or perhaps better) low ISO performance. It would surely be a market for that thing, but someone must have the capability to manufacture it. Sony may be able to do it, but MF would be a whole new market for them, and much less profitable than the smaller sensors. We will see what happens.

[bjanes]

Uhh... you're probably right, but the comparison was with the D5100 in that case though, but I thought 3 dB = 1 stop, 48.7 vs 41.2 that is 7.5 dB, > 6 dB, larger than 2 stops. I guess the large difference in midtones is due to photon shot noise...

When one uses dB to measure power (as with sound) the equation is dB = 10 * log(p1/p0), while when one is using amplitude the equation is dB = 20 * log(a1/a2), since power is the square of the amplitude. The output of the sensor is amplitude, so 1 stop = 6 dB. See here. It is a bit confusing. 

Regards,

Bill

[torger]

When one uses dB to measure power (as with sound) the equation is dB = 10 * log(p1/p0), while when one is using amplitude the equation is dB = 20 * log(a1/a2), since power is the square of the amplitude. The output of the sensor is amplitude, so 1 stop = 6 dB. See here. It is a bit confusing. 

Thanks for clarifying, I always mix those up... it's not the first time :-).

[bjanes]

Bill,

 Thank you very much for the reference.
 
 As I am getting every day more lazy and no less dumb in my geriatric age, would you indulge me and tell me how one would choose a (photographically meaningful) noise floor?

Edmund

Edmund,

I don't know about you, but my geriatric brain is a bit muddled. For the DR at 18%, I was using the number of stops that the SNR at 18% saturation is above a SNR of 0 dB (the engineering noise floor), and the floor was not at 18%. 18% is about 2.5 stops under 100%.

Regards,

Bill

[torger]

I read http://www.sony.net/SonyInfo/technology/technology/theme/cmos_01.html

It seems Exmor sensors are made with a dedicated CU wiring process, and something in the article talks about low silicon defects which makes me think they may have an additional wafer processing step, or maybe qualify the wafers in some way. But I see no reason why other companies cannot at least copy the architecture, invent something similar, or just license/subcontract MF sensors from Sony. After all, all their competitors are also their customers for sensors.

Another question is when sensors on dSLRs are going to be back-illuminated? The Sony site states there are 8db to be gained there ...

I hope there is someone that knows the details about chip manufacturing here, I don't. I've just noted that it seems to be quite hard to do "obvious" things... Back-illumination, that is putting the wiring behind the photodiode instead of in front of it (blocking light) is an obvious thing to do, so why it is not done is not because they could not figure out that placing it behind would be a better construction, but because of that it is harder to manufacture the chips that way (why it is so I'd be glad to hear, I don't know). Back illumination is more important with smaller pixels since the wiring will block a larger part of the pixel then, so the 8 dB was probably for mobile phone type of chips I guess (which has extremely small pixels).

I don't know why Canon for example hasn't licensed from Sony, but I guess it is a bit of prestige and competition ongoing (and better profit using own tech), and Canon provides something that is "good enough" for nearly all types of use, so I guess they don't care too much. This very strong interest in DR as we have in MF and landscape photography is probably a niche interest. I've shot quite some landscapes with a 5Dmk2 and even 7D and frankly I rarely have had problems in practice with their in comparison rather poor DR. But I still would like more... always want more DR and more resolution :-)

[yaya]

The development of CCD tech has not been as dramatic as with CMOS, but on the other hand noone can at base ISO do as good as Sony, so the best CMOS tech is not found in all cameras

I have yet to see an image (as in a photograph...with light and a real subject, you know what I mean...) that demonstrates this allegedly superior base iso DR of the Sony...

[torger]

I have yet to see an image (as in a photograph...with light and a real subject, you know what I mean...) that demonstrates this allegedly superior base iso DR of the Sony...

Superior to Canon is quite easy... superior to an MF is harder. My point was that demonstrating MF dynamic range superiority by putting it head to head to a Canon sensor (even the recent ones) is not giving it the toughest challenge, but rather the opposite.

Also, the MF supposed DR advantage is perhaps more about midtones than shadow noise. Or perhaps it holds color information better in shadows, or perhaps less color-blotchy noise due to larger area. It must be something else than pure signal-to-noise ratio 10+ stops down. It's a mystery which would be most interesting to get an answer to.

A side-by-side shoot on a high DR landscape scene with perfectly ETTR-exposed RAW files a D800 vs an IQ180 would provide quite much information about "state of the art". But I would not hold my breath. Testers almost never provide raw files and almost always expose the cameras differently (trust autoexposure or jpeg histograms or whatever).

[yaya]

Testers almost never provide raw files and almost always expose the cameras differently (trust autoexposure or jpeg histograms or whatever).

Which is why we always encourage potential buyers to do their own testing...if you work with a dealer who sells both formats I am sure that they will be happy to do this with you, especially if they know that there's a good chance that you will buy from them

many dealers offer a "try before you buy" where if you rent the kit they will put the rental fee (or some of it) towards a purchase

Yair