...Jesus, talk about splitting hairs ;)

This is an important topic (and certainly worthy of a sidebar entry) but I think this is a poor example. For one thing the original jpeg looks fine, the PP'd shot is really hardly much better, and this situation could be handled easily with a slight reduction of exposure (yet another reason not to shoot to the right, when shooting in bright sunlight always leave a good amount of room for highlights on the RHS). Illustration of the dynamic-range advantage of raw-shooting [to me it's just as much a matter of eliminating NR and in-camera color processing as HDR, especially in a small-sensor camera] requires a good selection of an example shot as much as a good selection of solution-methods and *possibly* even an HDR as well. Because you simply cannot see more than 8 bits of dynamic range in a jpeg in any case and even if you could you can't see it on most computer display hardware. And this is a website, right?

Pick a demo where the image under question suffers blown highlights in areas that aren't normally almost-white to start with. I would have picked an exposure where the sky itself, instead of being blue with white clouds, was so hot that the sky was entirely white, and then shown how the blue sky and white clouds can be rescued with an ER from a raw shot. At the least. But to really get the full DR on the web you have to use an HDR technique. 8 bits/channel is simply not enough to display a realistic DR in a well-lit high-contrast scene.

plus you can see here how the prototypical "ACR best" negative contrast curve washes out the shot, who would want that result just to save the highlights on the statue, half of which get blown-out again on the custom-parameters shot. This shot basically needs an exposure-reduction (and using the raw files as a source can save your bacon here, but even raw files have limited DR) and an HDR technique to get it all right.

for the record I categorically recommend against right-biased shooting because the loss of camera speed (and thus shot-stability and DOF) and loss of highlight dynamic range (both from cramming the shot to the right and the use of higher ISO which is often required) often outweighs the increase in color resolution from shooting to the right, especially given the reduction in overall dynamic range. Like most mantras it not only tends to be overused it's actually counterproductive in some situations. I want a hot-enough exposure to give me acceptable DR but not hot enough to ruin everything that is captured. In this case the choice was made to shoot it hot enough so that the statue was almost entirely blown-out and only the fact that it was shot raw allows for any recovery. But in my opinion the degree of IQ reduction due to the blown-out statue is minor even in the original camera jpeg, at least in this shot. To me this is only a good example of what one could do, but it is far from the best example of what could be done if the shot were taken raw. And half of the problem is this obsession with pushing shots to the right. When bright scenes are shot during the day in strong light half of the highlights will be lost due to that simple reason. Of course with so much light there isn't much chance of instability, but still in lower level of lights, it becomes a huge factor. Just about any shot can be pushed a stop without significant loss of color resolution and that can make a huge difference in terms of stability and noise. I routinely exploit this in my shooting. And to me the real benefit of highligh-recovery in raw shooting is the ability to recover entire regions that are blown-out in HDR scenes, or very bright-areas of shots taken at normal exposure that inevitably get blown out in bright light. Just having that ability means that I can knock the exposure down somewhere between a half-stop to a stop and get a great overall exposure, without having to shoot brackets all the time just in case.

...last but not least many cameras and even RCs will push the exposure even further, "automatically", when converting from raw to jpeg. Simply shooting raw means that if that "feature" is disabled a much-better overall exposure will result. Dcraw for example has default settings that will push the exposure so that at least 1% of the shot is blown-out, regardless of the overall exposure. Raw-shooting has many many advantages over shooting camera jpegs, this is just one of them.

" I'm sure that everyone who is not a fanboy or an HDR freak will see the point you're trying to make and appreciate the effort (thought they probably won't feel compelled to post a comment)."

..but if someone *does* post a comment, then they are compelled to...for some reason, probably bad?

at the bottom of the page here:

http://www.dpreview.com/reviews/canoneos50d/page19.asp

...you will see a much better example of why this is important (still not the one that I was looking for on dp but much better), but still this also highlights the issue of limited DR both in the jpeg format and in 24-bit display systems, that pretty-much forces the photographer to resort to HDR compression techniques. Depending on what is lost and gained, in terms of image-content, as a function of exposure. Still, no matter what, the brain is just not going to be able to mentally-process 12-14 stops of DR in a scene. 8-bit displays and the 8-bit jpeg format are popular for good reason. But I think that 10 stops would be a reasonable compromise. Anyway the 12-14 bits of dynamic range that are available in raw shots is really just "save your bacon" range, maybe 1 or 2 cameras (the D3X and the S5 as noted) can actually exceed the standards 12bits. What impresses me more is the fidelity of a 14-bit A/D system, and the fact that it can save a shot by providing a DR buffer. An esoteric topic but still very important.

"But measuring off a pseudo-scientific process and presenting them as quantified numerical data is a bit irresponsible."

...so is raising a big alarm over a minor issue.

The A/D circuitry has a resolution and dynamic range which exceeds that of the sensor. They do cover this in detail in their SLR reviews...this is a fairly high-tech area but it's all behind closed-doors if you don't shoot raw and in any case the output is limited by either 24-bit display resolution or by printer resolution. The real problem is the relatively low resolution and DR of standard output devices, compared to even a P&S shot raw.

...great then since we're both professional EEs and we both have experience with mixed-mode design, I hope, then you can just straight-out explain in a paragraph or so how you can get more DR out of a sensor than the A/D system can support, even if it's a 14-bit A/D system. If it'll help to make this more spicy for you I can point to several camera reviews on multiple sites including dp that demonstrate that there's not a single camera short of possibly a high-end MF digital back that I've yet to see that has much more than 12 bits of DR and most have around 10 bits. Feel free to link to any site that you like. But enough of the smoke & mirrors, chum.

plus, if I can say this without getting spammed, why would a MFG design, build and sell hundreds of millions of cameras with sensors that have more DR than the RO electronics can support? Not to mention the output data format?

Ignoring the fact that I find 10bit DR 12MP images in 24-bit jpeg to be quite enough, really. Sure I would appreciate more spatial resolution but the last thing I really need is more fidelity or DR. I'd *love* to have the optical gain without the noise but of course that would require me to use much-bigger sensors and/or bigger & faster lenses. But ok maybe I've made the mistake of putting actual *numbers* to the concepts that you spoke about in your post, so I'll be quiet now and see what you have to say. Sure: maybe Nikon has used a sensor with 16bits of DR in the D3X but saddled it with an A/D system with only 10bits of DR yet used back-end electronics that produce 14-bit raw files. I'm sure that the explanation is "in the texts" and you can point out specifically where since many top engineering schools now have their entire curriculum posted on the 'net plus there are plenty of "how does this work" sites out there with similar info. Shouldn't be a problem.

...remember that in digital photography (especially digital cameras with CMOS sensors with digital read-outs), DR is measured in bits, not in dB like an analog signal. ISO and exposure-time provide the conversion from integration to output-level and one is left with the DR measured relative to the noise floor at each ISO. The true DR, in dB, depends on the display circuitry. Since that is device-dependent digital photographers don't use that when they speak of DR. In general is of course possible for a sensor to have a DR measured in dB that exceeds that of its associated digital RO circuitry but now you are mixing metaphors. The DR of a digital camera is measured in bits, not dB, and dpreview and also dxo routinely provide test data that demonstrates that the 35mm digital cameras and below, at least, have dynamic-ranges of no more than 12 bits. Do the sensors have dynamic-ranges of 14 bits, 16-bits? Do they have a DR higher than the RO electronics if the DR is measured in dB? Possibly, but I doubt it if they are using the same power-source. Is it possible to determine the sensor DR in dB independently of the RO electronics, in an optimal configuration? Sure, rip the sensor out and see for yourself. Would that mean anything? No. You have to put real electronics behind the sensor to make a real camera and I see no rational reason why a 14-bit file format would not be able to adequately-support a sensor with 14 or 16-bits of DR. These cameras are barely hitting 12 bits of DR, unless there's some major flaw in the testing methodology.

Even so given an 8-bit/channel computer you would only get DR headroom out of this. Which is the whole point of this article.

...dxomark compares *cameras*, not sensors. Unless you operate under the assumption that the sensor is the limiting factor, not the back-end electronics. Even so never forget that just about any camera shot raw will provide more DR than you can print or display on commonly-available computer hardware. The limitation is your computer, not the camera.

...except where the camera ISO is high enough so that the DR at that ISO is lower than 8bits/channel, the factor that limits DR for common display hardware. As the ISO is increased the camera sacrifices color-resolution, image DR and SNR for better overall image exposure, likewise the photographer sacrifices all of these plus exposure for speed, DOF and reduced lens-blur across the image.

...to be technically-accurate the display DR is set by contrast-ratio, the bit-depth determines resolution. You trade-away resolution, DR and noise-immunity for exposure, speed, DOF and lens-sharpness with increasing ISO. That's in the camera. The display performance is set by the display hardware. Sorry I would have said this above but I can't edit these things.

by the way there is a lot of talk now on the 'net about this topic with regard to the 5dMk2:

http://forums.dpreview.com/forums/read.asp?forum=1032&message=30799570

...seems a lot of people are not happy with having to shoot it raw to get good IQ out of it :) But the people who normally shoot raw are quite happy with it. The same goes for a lot of cameras due to poor in-camera image-processing with regards to metering, white-balance, NR, tone-curves sharpening and the like, but similarly a lot of cameras have great in-camera image-processing and this is not nearly as much of an issue.

"As to smoke and mirrors how can suggesting you read and be factual as opposed to misleading be smoke and mirrors? "

Simple. Because you're not being factual, you're just being suggestive. Which is the basis of all magic. Likewise the reference that you link to is the same way, the author doesn't actually test sensors, he estimates sensor performance based on a model and relies on mfg datasheets for support. The trend is clear. It is one thing to make a claim, another to state facts. The mistake that you make (and that he makes) is that you substitute the former for the latter. Again, digital photographers do not measure DR in terms of electrons or even in terms of dB. Certainly it is *possible* for a camera sensor to have a dynamic range that exceeds that of the RO electronics, but that would depend on how you measure DR for the sensor and for the electronics. If you buy a camera, is that the case for that specific camera? Maybe. Is this issue really all that significant since you have to use the camera short of ripping it apart and putting-in your own RO electronics? No. This whole discussion borders on complete irrelevance for that reason. Not to say that you are wrong. It just doesn't matter. What matters is what is in the data-file. The theoretical estimates that you see here

http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary/dynamic_range_a.gif

are not in the data files. The MFG data is.

Likewise here, at least in the general sense...

http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary/dynamic_range_4cameras_2.gif

and here:

http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary/dynamic_range_3cameras+models_3.gif

...with the exception of the newer Sonys which don't show that rolloff at the lower ISOs.

You can check this at dxo if you wish.

So the question of do the sensors have a DR which exceeds that of the RO electronics seems to yes for some and no for others, depending entirely on the sensor, the RO electronics, and exactly how you measure DR. No "misleading" at all. Even if you wish to be sidetracked by that question. But if you wish to be more specific than this, be my guest, perhaps that will make for interesting reading as well. I caution you that this is not my site, though.

...the argument that the guy at Clarkvision is making is that at low ISOs you should see well above 12 stops of DR out of these cameras except that the RO electronics are limiting the performance (neglecting the fact that some are using a 12 bit data format, some 14 bits). Not one of the cameras that they test actually exceeds about 12.5stops of DR. However they estimate that this is well below what their sensors are capable of and that the general trend at higher ISOs supports their model. But this all flies in the fact that the mfg data does NOT support their model. Logically one must conclude that they believe that the MFG data is incorrect or else specified at high ISOs. Either way the MFG data does not support their theoretical estimates.

And in any case if you were to go look at the dxo DR plots vs ISO for the newer Sonys you wouldn't see this rolloff at low ISO. And their DR plots match up pretty well with the last two plots that I linked to above. They are also supported by the tests done here at DP. So the question is are the sensors accurately represented by the model or not, well, they would have to replace the RO electronics with higher-performance electronics to know for sure...but they don't do that, they just rely on the models that they have come up with. You, likewise, in relying on this link, are relying on their models. I am just saying from a practical point of view that it would not make sense to put a sensor in a mass-produced camera that has a DR that far-exceeds the DR of the camera electronics and output file format. Even if we were talking about low-volume cameras for scientific applications, not to mention cameras with production numbers in the millions of units, from mfgs that produce dozens of cameras. You can't look at ideal photoconversion in an ideal silicon wafer in an ideal sensor and think that's going to carry over to the sensor in a mass-produced digicam. That doesn't make sense.

...the D300, D700, D3, D3X, A700 and A900 are notably-absent from those charts, not to mention any and all MF cameras. Why include the 5DMk2 and not the D700?

I would not have been satisfied until I could not find even one camera, not even a $15k MF digital back, that shows strong correlation with my predicted results all the way through its ISO range. Only then would I conclude that this rolloff is due to a lower DR in the RO electronics. It could just be due to a "no better than necessary" sensor. No engineer worth his salt would pitch a sensor as having a FOM better than the RO electronics, and then use that sensor/electronics combination to support a model that he came up with when he didn't have experimental data to actually prove that.

Of course, he might be right, but he hasn't proved it. It's just speculation at this point.

...interesting how dxomark just posted results for 4 MF digital backs, and the D3X beat all of them for dynamic range, even with a significantly-lower pitch, across the ISO range. And its DR rolled-off at a higher value and a higher ISO, while most of the digital backs didn't roll off at low ISO. So what's happening, at high ISO the response is defined by electronics? The measured data seems to fit their theoretical model

http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary/dynamic_range_3cameras+models_3.gif

...fairly well at *high* ISO, its the lower ISOs that are giving the model trouble. But then again you don't see many digital cameras with ISOs below 100. But their performance model is failing even at ISO1000, for the 5DMk2 and 50D. And why would the measured system DR be limited by the electronics at low ISO but limited by the sensor at high ISO. Their model is just not general enough.

Three interesting quotes in closing:
..."Figure 4 shows the maximum dynamic range possible from each sensor, based on full-well capacity / best read noise [i.e. their theoretical estimate], *assuming no limitation from A/D converters*. Figure 5 shows the *measured dynamic range* from 3 cameras with significantly different pixel sizes as a function of ISO...One sees that actual dynamic range of a digital camera decreases with increasing ISO as long as the range is not limited by the A/D converter... If 14-bit or higher analog-to-digital converters were used, with correspondingly lower noise amplifiers, *the dynamic range could increase by about 2 stops on the larger pixel cameras* (so it is limited by the A/D resolution, not "A/D dynamic range" as some hypothetical "noise-based DR limiter" concept separate from the resolution).

Ok so throw the 5Dmk2 and 1DsMk3 into the mix which do have 14-bit A/D and you see that the upper DR limit hardly improves, proving them wrong:

http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary/dynamic_range_4cameras_2.gif

http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary/dynamic_range_3cameras+models_3.gif

(and also demonstrating that their prediction that future cameras will have 16-bit A/D is nonsense, especially with decreasing pitch, unless some MFG wants to provide more DR headroom and lower q-noise at the cost of lower packing-efficiency and higher data bandwidth requirements)

...but note the D200 performance...which *is* a 12-bit camera, is nice and linear all the way down to ISO100. Better read-out electronics...on a subframe, compared to two fullframes? Or just a higher-quality sensor? My money is on the latter. If only because I see nothing here which demonstrates that the A/D system affects sensor DR in any way other than by resolution.

Last but not least the part about CFOM=MP*pitch is sheer nonsense.
They just proved experimentally that camera performance doesn't increase linearly with pitch especially at low ISOs. Anyway this site does more to disprove your opinion than I think that you realized when you posted the link.

...not only that but the high-ISO DR performance of the 1DsMk3 is better than the high-ISO performance of the 1DsMk2 even though the Mk2 has about 15% higher pixel-pitch. Given that the low ISO performance is almost the same while one is a 12-bit camera and the other is a 14-bit camera, this quite-thoroughly blows a hole in your theory. The A/D circuitry, the RO electronics in general and the output file-formats are providing more than enough DR for the sensors in commercial cameras. DR headroom is another issue entirely. Simply switching to a 16-bit electronics and data-format would give you up to 4 more stops of DR headroom. But a commercial sensor would never match that, you'd have to convert a 20MP+ CMOS sensor from 12 to 16 bits that can't put-out more than 12 bits of DR at any reasonable ISO. And just think, if it were really that easy why not just go to 16-bits now instead of goofing around with 14-bit A/D and 14-bit/channel output file formats. That would drive the sensor cost through the roof. These sensors have to be phenomenal just to make good use of a 14-bit system, and even then, as you just saw, they can only make full use of it at low ISO. Most people do not drop $2500 on a FF or even $6k on a low-res MF digital back for dynamic-range. Those cameras are purchased mainly for low noise, second for high ISO, last but not least for high resolution. Never for DR. This whole idea of exploiting that high stop of DR headroom is just a trick like polishing CDs to make them play better. An act of desperation, made necessary by shooting to the right in the first place.

...but sure, yeah, if it were up to me I'd take an 8MP or even a 6MP 5D with 14-bit or even 16-bit A/D and a clean ISO12,800, no problem. Even if it were $4k I would happily buy that over a 24MP D3X at $8k for the body. When I can get 14MP raw shots out of a G10, with a 28mm IS wide-angle that fits in my front pocket, why do I need a 24MP FF? I don't need the *pixels*, I need the SNR!

...WJ, while that may be true, if you don't shoot raw you will not have the information required to repair blownout highlights (unless you can get it from another shot where they are not blown-out). Which is the point of this article.

Of course even shooting raw you get only so much extra DR. And it still has to be managed intelligently in PP which in my opinion is 80% of the challenge and often makes it not worth worrying about. To me the main benefit of shooting raw is in getting the raw sensor-data. The extra DR is just icing on the cake. If those two things are not important to you :) then by all means shoot jpeg but you can forget about recovering blown-highlights.

...this is a better example of the pluses and minuses of shooting jpeg vs raw...just to show that it isn't all about high DR & recovering blown-highlights vs huge files & the need for complicated post-processing.

Sometimes it's just plain and simply about getting a great shot instead of just a good one.

http://www.flickr.com/photos/26432908@N00/345788068/in/set-72157594542175419/