neuroanatomist said:
jrista said:
neuroanatomist said:
Yes, it's quite a good trick to have DR "in mid 14's" when your camera has a 14-bit ADC. Hail to the almighty DxO Biased Scores, and kudos to those that revel in that BS. :
It's actually the compression algorithm. Also technically speaking, it isn't actually RAW. However, cRAW applies a tone curve to the data coming off the sensor before it black clips and compresses. It's actually the same thing they do with the A7s, and will probably do with future cameras. They take a greater dynamic range than their bit depth and use a mathematical curve to compress it into a smaller space. Same thing we do with our RAW editors.
The thing with Sony cameras is they have the data fidelity to actually do that. The data precision in their RAW files may only be 12 bit, however in terms of usable information they still deliver stops more than any Canon camera. You can hate DXO all you want (I don't like most of what they do either), but the real world results are all you should need to understand that bit depth and usable information are not synonymous.
The thing I don't get is why Sony doesn't just go strait to 16-bit RAW. In digging around with their SDK lately, I think Sony has some strange data bottlenecks somewhere in their readout pipeline. They have a number of cases where they restrict the sensor readout to 12 bits even, and it always seems to be throughput related. If at some point they resolve those issues, I'd be willing to bet Sony puts the first 16-bit RAW consumer camera on the market.
Regarding the rumored A7000, yes it's certainly possible to map higher DR from on-sensor HDR down into a lower bit depth file (assuming the rumors are true). As you state, it won't be RAW data at that point, but Sony doesn't seem to care much about that anyway. It certainly breaks the long-standing common practice of using an ADC that can fully encompass the DR from the sensor. Maybe it's common to compress/map wider analog data down prior to digitizing it in other fields e.g. audio, I'm not sure.
Regarding the more recent discussion in this thread regarding current cameras, it has nothing to do with in-camera mapping/compression, and everything to do with DxO's downsampling the files to 8 MP. AFAIK, no current dSLR/MILC can record >14-stops of DR
at capture (but for example the Red Epic Dragon can, and it uses a 16-bit ADC).
You know I don't like DXO any more than anyone else. Their persistent attempts to box all of the complexity of sensor IQ into a single scalar number is as annoying as ever. That said, all they are doing is normalizing. Normalization should be a well understood concept, especially by someone such as yourself, given how many times it's been explained on these forums.
Personally, outside of a pure comparison context, I don't believe the numbers spit out by 8mp normalized results tell us much about what we'll experience when actually editing a RAW file in a program like lightroom. I believe that Screen DR (non-normalized DR) tell us that, since that is the per-pixel DR of the RAW data that we are literally working with. Another issue I have with DXO's Print DR is that,while they call it a "measurement", it is absolutely nothing of the sort. It is a purely extrapolated number, acquired by running another DR value through a simple mathematical formula. It's the purely theoretical maximum DR that the camera might achieve if it had perfect noise characteristics. It does not account for actual noise characteristics, because it is not an actual measurement...it's an extrapolation from an actual measurement (which so happens to be Screen DR).
The fact that DXO does not make that clear, and worse the fact that they utilize their Print DR numbers as THE DR numbers, has lead a significant percentage of the photography community regurgitating numbers like 14.8 stops of DR as actual real-world DR. Assuming all of those people are downsampling all of their photos to 8 megapixels, then they may well indeed be getting 14.8 stops, or 14.5 stops, or 14.2 stops or whatever it is for the camera they use. If they are keeping their data RAW, or worse, printing at even larger sizes than native, then they are decidedly NOT getting the theoretical maximum potential DR.
Sadly, I do not believe that actually downsampling images and actually measuring them would paint Canon in any better light. Canon has worse noise characteristics than the competition, so actual Print DR measures for Canon would probably end up worse than the extrapolated Print DR "measures" that DXO uses now. I spend a lot of time working at the noise floor with astrophotography. I run FFT's on individual subs and integrations every so often. Canon data doesn't even come close to exhibiting a gaussian distribution of noise. Sony cameras are closer, but their compression limits how close they can get. Nikon D800 data with the black point hack actually has the closest to gaussian representation, but even they aren't purely gaussian. A clean, pure gaussian noise profile would result in ideal downsampling results, while the non-gaussian noise profile of Canon data is going to result in less than ideal downsampling results. I don't believe any ILC currently on the market could actually achieve exactly the Print DR that DXO lists, although I think the D800 & D810 probably get closer than anything else.
As for recording more than 14 stops of DR "at capture". Depending on exactly what you mean there, it's possible. Assuming "at capture" means in the analog signal on the sensor, then if you had a sensor with, say, an 80,000e- FWC and 3e- read noise. The dynamic range of the analog signal is 14.75 stops of DR. With a strait ADC conversion using a 14-bit ADC, you would have to clip that to 14 stops in some way. Alternatively, you could compress the dynamic range, preserving the original bounds of the information it represented by combining some of the information with less precision. Lower precision, same range of information. That's what cRAW does...applies a curve that compresses the sensor DR into a tighter range, then digitizes it.
Conversely, if you have a sensor with 80,000e- and 25e- RN, then your dynamic range is 11.68 stops. Your analog signal doesn't have enough tonality to even use 12 bits, so storing the signal in 14 bit data is really just wasteful. It could affect camera design in other ways...larger numbers use more bandwidth, thus potentially limiting your maximum throughput, putting a cap on frame rate, etc. Canon could easily be using 12-bit data, and we wouldn't be losing a thing. We wouldn't be able to represents steps of noise as accurately, but we certainly wouldn't be losing tonality. Because of the higher read noise, the sensor isn't capable of delivering more than 14 stops of DR, so compressing the signal makes no sense.
If Canon delivers a 5D IV with an 80ke- FWC and 5e- RN, on the other hand, they would have 14.01 stops of DR, and at that point, they would be able to fully utilize the bit depth of their ADCs. They still wouldn't need to compress the original sensor information to fit the bit depth of the ADC, but they would at least be able to use all of it.
Bandwidth is, IMO, the primary reason Sony hasn't gone to 16 bit ADC yet. They seem to be borderline on 14 bit as it is, but since their sensors do deliver at the very least more than 13 stops of DR, they are able to utilize a 14 bit ADC unit. If it wasn't for any kind of throughput bottlenecks, then I suspect Sony would already be using 16 bit ADCs. One of the reasons Exmor has lower noise is because their ADC units are column parallel, which allows them to be run at a lower frequency. It is entirely possible that running them at a higher frequency to handle 16-bit conversion introduces more noise, which would diminish dynamic range. I'm not sure. Either way, what Sony is doing, compressing the original signal before conversion, is the best way to preserve the information their sensors are capable of delivering, even if it costs some precision. In practical use, the loss of precision doesn't seem to be a huge issue. It might result in some posterization of smooth gradients at the low and high ranges of the signal, where linearity may drop. Again, in practice, it seems that photon shot noise is generally high enough that posterization isn't a big problem outside of extreme circumstances.
Camera companies are employing different and alternative ideas in the way they process the data on the camera hardware and firmware. Some of them are clearly preserving more information, despite the fact that they are using less precision to store it. I wish it really was just as simple as 14 bit ADC, 14 bit file, max 14 stops of DR. I wish it was just as simple as Sony moving to a 16 bit ADC in the A7000 and storing their 15.5 stops of DR without compression or anything else as-is in a 16-bit data file. I would so much prefer that. But...it's just not that simple anymore.
