Can someone explain why DSLRs with full format sensor (like the Nikon D3 and d700) have excellent high ISO performance/ low noise but yet struggle with dynamic range?

I have shot the d700 the other day, also the D3 which seems identical in image quality, and while they were quite good overall, it was obvious that the highlights were prone to blowing out.

My Fuji S3 and S5 are definately better, but they use the dual pixel sensor with disadvantages in file size and speed. It is actually "2 sensors in one".

MF backs however advertize a DR of 12-12,5 stops of light. Pixel size in the 39MP backs is not bigger than in the FF DSLRs, so how can they achieve such a great dynamic range? Do they use higher quality AD converters, or other electronics?

I don't think that file depth is directly linked to dynamic range, and I have read that several times.

So if someone can contribute an explanation.....?

Thanks
Bernie

First to understand is, that the dynamic range expressed as a number depends on the way it is measured/judged to a higher degree than on the sensor. When adopting the same measurement to MFDBs as for example DPReview measures the DSLRs, then the 12-12.5 stops shrinks to 10-10.5 stops.

The second difference is in the adaptation of ISO sensitivity. All DSLRs I know see the data under "different filters", i.e. interpreted with different ISO gains. Thus the potential dynamic range of the D3 can be seen as the one starting with the acceptable shadows at the highest real ISO (6400), up to the saturation level with ISO 200 (I think that's the base ISO). However, you can't achieve this DR in a single shot.

Most MFDBs work differently: they don't have different ISO settings; they offer the entire dynamic range in a single shot (therefor they do need a greater bit depth than DSLRs, though the 16bit is too much). The Phase One P45+ is an exception I know of, in that it has different ISO gains; there may be other exceptions.

OK, you are asking several different questions here, on slightly different aspects of the same subject, and so the answer depends on the question.

First, about DSLR dynamic range. It is limited at the high end by the saturation of the ADC, which takes the ouput of the ISO amplifier and digitizes it; there is a maximum value it can output, and therefore a maximum input it can cope with. When the ISO is doubled, one more stop or EV of input is pushed past the saturation point of the ADC and is lost.

At the low end is the noise "floor", really the point at which one chooses to quit because the signal/noise ratio becomes too low. This is governed by the electronic noise in the circuits that read the sensor, amplify the signal (ISO amplifier), and quantize it (ADC). This noise varies with ISO, and thus how much room on the noise floor is available depends on ISO.

The result is that, in terms of absolute exposure, one has a "window" between noise floor (the minimum S/N ratio deemed acceptable) and saturation, which is the DR for that ISO. For the Canon 1D3, it looks like this for various ISO:



Note that each subsequent stop of ISO amplification lops off one stop from the upper end, while the room on the lower end improves because read noise on CMOS sensors drops in absolute exposure terms with increasing ISO, until it saturates about ISO 1600.

Note that the expansion on the bottom end yields less and less -- going from ISO 800 to 1600 doesn't make much difference in shadow S/N, but one loses an entire stop of raw headroom. Above ISO 1600 there is no expansion of the shadow range whatsoever, just more and more lost from the top end. This is why it makes no sense to use absurdly high ISO's like 6400 if you shoot raw -- you're just throwing away highlight headroom and not getting anything back at the shadow end; it's better to underexpose by a stop or two at ISO 1600 if you need the shutter speed, than to use higher ISO.

All this "window" behavior is occurring because the camera doesn't deliver the full DR that the sensor is capable of at base ISO; rather it is limited by the rest of the electronics, the ISO amplifier and ADC. The sensor has a DR of about 14 stops, while the amp/ADC have a bit less than 12 stops DR. There are about two stops of extra DR to be had in DSLR's that is currently being thrown away by limitations of components other than the sensor. This is true of both Canons' and Nikons' CMOS DSLR's.



That sounds more like an exposure issue than a DR issue, that the metering wasn't what you expected it to be; unless you are saying you chose the exposure to hold some chosen level of detail in the shadows, and that resulted in the highlights overexposing.




Well, don't believe everything you read The 39MP backs use a Kodak CCD sensor that has slightly less than 12 stops of engineering DR at the pixel level, IIRC. The MFDB pixel DR is comparable to that of pro level CMOS DSLR's (Canon 1 series, Nikon D3). But here one should distinguish pixel level DR and image DR. With twice the area of a FF sensor, a MFDB captures more light over the frame than the FF camera, so more signal and better S/N ratio and DR at the image level. The whole is more than a collection of isolated parts -- the individual pixels are similarly spec'd, but the MFDB has more of them collecting more light and when you do the math as to how DR scales when you scale up the sensor the bigger sensor wins in DR by about the change in linear size of the collecting area, assuming that the collection efficiency per unit area is similar (which it is, IIRC).




This is because, when the ADC used has enough DR not to cripple the sensor DR, it outputs all the information that the sensor has to offer. There would be no purpose to offering hardware ISO gain, all it would do is reduce the highlight DR without offering more shadow range, just as the DSLR's behavior at ISO 1600 and above. The difference is that, when the ADC's DR exceeds the sensor DR, that property starts immediately above base ISO rather than at some higher threshold, and thus there is no reason to offer hardware ISO -- it would just remove highlight headroom. The only advantage might arise if MFDB's offered jpeg output

FWIW, my Leaf Valeo 11 was an 11MP back with a sensor the same size as a fullframe 35mm sensor, and it had 12 stops of DR.

I have the D3, and with Adobe Camera Raw there is a +0.5 EV baseline exposure offset (see the Adobe DNG spec for details) and most shots appear overexposed with ACR, but less so with in camera JPEGs and with Nikon Capture conversions. You really have to look at raw files to judge any clipping or how much headroom is present. Very few of my D3 images where I had set the exposure reasonably correctly have blown highlights in the raw file. Rawanalyze is an excellent tool for this purpose.

Bill

Double the sensor area, as in the P45+, and you get a theoretical increase of 1/2 a stop DR, ie. 12.5 stops.

However, in the absence of careful DR comparisons between, say, a 1Ds3 and P25 or 45+, we can't be sure what the differences are. Off the top of my head, I would have thought the increase in DR would be greater than 1/2 a stop, but not just because of increased sensor area, but also possibly a greater photodiode size within the photosite, which allows for a base ISO of 50 despite equal pixel count and similar quantum efficiency. But I'm just speculating.

Your statement that the camera has a dynamic range of 12 stops is incomplete, since you did not state the criteria by which the dynamic range was defined. Emil is talking about the engineering DR, which is the full well capacity/read noise, both in terms of electrons. This is for the base ISO of the sensor. For each doubling of ISO above base, the numerator in the equation is halved. The denominator of the equation also changes with ISO, since read noise decreases with increasing ISO up to a certain limit as Emil described.

Photographic dynamic range depends on what S:N you are willing to accept in the shadows, and is less than the engineering DR. For example, Imatest reports multiple DRs depending on the noise criterion used.

There is an ISO standard for dynamic range, but it is rather confusing and is not often used by photographers.

Bill

It's also important to take into account pixel size, since when comparing DR of different sized pixels one is comparing the DR at different scales, and S/N ratio and DR are scale dependent. The typical rule of thumb is that DR expands linearly with scale; that is, if one combined the data from a 2x2 block of pixels the scale would be twice as large for this superpixel, and its DR would also be twice larger.

I don't think that they do. Shadows are extremely clean, and exposure should be chosen taken that into account.

Besides, the highlight roll out is very soft and natural.

I have not done any rigorous measurement, but I feel that my D3 is not very far from the Mamiya ZD in terms of usable DR. The ZD appears to have more highlight headroom, but the shadows are in fact less clean.

Cheers,
Bernard

I am surprized that in-camera JPEG is used as criterion for image quality.

Hi I would put my two cents in:

1. user error is often a cause of issues you mentioned
2. too much contrast in the scene ie zone system not being used
3. 16 bit vs 12 bit
4. perhaps the detail is really there but your raw converter is set so it clips.

I own a digital tech business in LA and I'm asked and presented with challenges like this every day,
it's usually an error from one of the above...

Good points, although I would add that the newer DSLRs are now using 14 bits. I'm not sure I get much more (or any more) DR from my D300 using 14 bits than I get using 12 bits, but unless I'm shooting fast action I use it all the time anyway.

I'd like to thank everyone who has answered so far to my question. I am technically not knowledgeable enough to make a real judgement, so I read those answers and try to make myself a better idea...

regards
Bernie

Hi,

"The ZD appears to have more highlight headroom, but the shadows are in fact less clean."

It sounds to me that you don't expose to the right. The expose to the right philosophy essentially tries to utilize the highlight headroom to optimize shadows. My feeling may be that there is a good foundation for "expose to the right" but it can also cause quite a lot of trouble, if overdone.

Best regards
Erik

Ps. BTW, I enjoy both your pictures and your writing :-)

Bernard,

I too am an admirer of your photography and writing, but I must take exception to your statements about highlight headroom and roll off.

Like most digital sensors, the D3 response is linear up to clipping, at which point there is an abrupt flattening of the curve. Shown below is an Imatest analysis of a Stouffer wedge photographed with the D3. The raw file was 14 bits, but Imatest reports a normalized pixel value in 8 bit notation.

Click to view attachment

The highlight roll off is a function of the raw converter. Here is a plot of a normally exposed wedge converted with Nikon Capture. The camera was set to the Standard Picture Control.

Click to view attachment

Since the raw file has a linear tone curve, I don't think it makes much sense to talk about highlight head room, since there is really no midpoint separating highlights from shadows. DPReview uses 18% gray, but the location of this point is dependent on the exposure (normal or ETTR).

Regards,

Bill