How are dynamic range and sensor size related?

[V8Beast]

First off, let's not turn this into a D800 vs. 5D3, or Canon vs. Nikon debate Instead, the I have a very simple question regarding dynamic range, since it's such a hot topic lately.

More specifically, lots of people say they prefer shooting full-frame because of the improved dynamic range over crop bodies. The common argument is that the larger photo sites collect more light, thus translating to improved DR. According to DxO sensor tests, however, this simply isn't true:

5D3 = 11.7 stops of DR
7D = 11.7 stops of DR
G12 = 11.2 stops of DR

So, a 1.6:1 sensor 7D matches a full-frame 5D3 in DR, and the tiny sensor of the G12 point-and-shoot is only half a stop behind either of the SLRs.

I'm not the most tech savvy person out there, so I'm just curious how DR and sensor size are related, if at all, and why. Thanks

 

[well_dunno]

thanks for starting this thread - I asked the same question on 5d Mark III DR thread. Most of the available info I found online points out a correlation between photosite size and DR but it just does not seem to be true considering the measured DR for several sensors as you pointed out...

 

[Loswr]

How are dynamic range and sensor size related?

Ambiguously.

 

[jrista]

Dynamic Range is related to a couple of things. Pixel well size, which is classically thought of as the most important factor, and electronic noise. The size of a pixel really affects the maximum saturation limit in electrons, which can theoretically improve your signal to noise ratio (SNR). Just having a higher SNR is only part of the story, though...the level of electronic noise is what the ratio is relative to, and higher electronic noise can still diminish your maximum saturation SNR even if you have a larger pixel.

The reason the D800 with its Sony Exmor sensor, to use the most ideal example, has so much more DR (13.2 native stops) is because it has extremely low read noise. The 5D III, on the other hand, has particularly high read noise. In exact terms, the D800 has about 2-3 electrons (e-) worth of read noise at ISO 100, where as the 5D III has over 33e- of read noise (over 11 times more than the D800, and ironically, even more than the 5D II had at around 27e-.)

A larger pixel is definitely a factor in improving dynamic range, but if you concurrently increase electronic noise, you won't gain much, and may even lose dynamic range if you increase electronic noise enough. A smaller pixel (i.e. a 7D or G12) can theoretically attain similar or even better dynamic range if you keep electronic noise low. The story is certainly more complex than this when you factor in all the details that affect sensor efficiency and IQ, such as quantum efficiency, ADC gain, etc...but in rougher terms: pixel size and electronic noise both significantly affect the dynamic range you can get out of a sensor.

 

[V8Beast]

A smaller pixel (i.e. a 7D or G12) can theoretically attain similar or even better dynamic range if you keep electronic noise low. The story is certainly more complex than this when you factor in all the details that affect sensor efficiency and IQ, such as quantum efficiency, ADC gain, etc...but in rougher terms: pixel size and electronic noise both significantly affect the dynamic range you can get out of a sensor.

Thanks for your response. I guess no one else knows the answer

What I gather, then, is that it's all about read noise. Even without doing the math, it doesn't take a rocket scientist to figure out that when comparing an APS-C 7D with 18 megapixels to a full-frame 5DIII with 22 megapixels, each pixel on the 7D's sensor is much smaller. If I understand correctly, the reason it can match the 5DIII's DR is because it's read noise is lower? I'd imagine that the 7D's read noise must be substantially lower than the 5DIII's to make up for the difference in pixel size.

If so, the obvious question is why the read noise is so high on the 5DIII. In comparison, the read noise of a G12 must be insanely low to come so close to the 7D and 5DIII in DR. How's this possible?

 

[jrista]

A smaller pixel (i.e. a 7D or G12) can theoretically attain similar or even better dynamic range if you keep electronic noise low. The story is certainly more complex than this when you factor in all the details that affect sensor efficiency and IQ, such as quantum efficiency, ADC gain, etc...but in rougher terms: pixel size and electronic noise both significantly affect the dynamic range you can get out of a sensor.

Thanks for your response. I guess no one else knows the answer

What I gather, then, is that it's all about read noise. Even without doing the math, it doesn't take a rocket scientist to figure out that when comparing an APS-C 7D with 18 megapixels to a full-frame 5DIII with 22 megapixels, each pixel on the 7D's sensor is much smaller. If I understand correctly, the reason it can match the 5DIII's DR is because it's read noise is lower? I'd imagine that the 7D's read noise must be substantially lower than the 5DIII's to make up for the difference in pixel size.

Aye, the 7D's read noise is quite a bit lower (although not nearly as low as an Exmor sensor, which ranks in at 2-4e-). I think the 7D has around 17-18e- read noise at ISO 100, which is roughly HALF that of the 5D III. I am honestly rather baffled by the read noise of the 5D III sensor. For all the progress that is apparently possible in the area of read noise, and how important it is to maximum DR, it really surprised me to see it JUMP from the 5D II's 27-28e- to over 33e-. Truly has me baffled. Either Canon is just simply not paying attention to the desire of so many of their brand-locked photographers for better DR, or they just don't care...either way, its a pretty major step back.

If so, the obvious question is why the read noise is so high on the 5DIII. In comparison, the read noise of a G12 must be insanely low to come so close to the 7D and 5DIII in DR. How's this possible?

AYE! That would be the question indeed! I don't have any details about the G12. Either Canon has found a way to greatly improve the SNR of the G12, or it must have pretty low read noise. It is a substantially different sensor than those found in Canon DSLR's, though...it may be backlit (which, much like microlensing, can greatly increase quantum efficiency), and if so, that would indeed help improve SNR. Unless Canon started work on the G12 sensor so late in the game that they couldn't migrate any of the read noise/QE improvements to the 5D III sensor, the ridiculously high read noise of the 5D III (which is the highest of any camera over the last several years, as far as I am aware) is completely unwarranted in my opinion. As such...the only logical conclusion I can really come to is that Canon simply does not care about low-ISO anymore. Their push for "improvements" would then seem to be driven by three large but very specific markets:

* Video (which is rather annoying for stills photographers, who used to be the primary drivers of DSLR use)
* Wedding photographers
* Sports/Photojournalist photographers

Canon seems to have entirely lost site of other forms of photography, such as:

* Landscapes (resolution and DR, DR, DR!! You can never have enough DR for landscapes.)
* Studio (resolution)
* Portraiture (DR)
* Wildlife (resolution, sometimes DR)
* Astrophotography (Noise floor!! Noise is a massive killer for astrophotography.)
* Macro (resolution and DR, but NOISE can be a big one.)

The really sad thing is, when I check local stores, and ask friends about local stores in their areas...the D800 is in HUGE demand (more requests than for the 5D III)...but the 5D III is what sells. The primary reason I get is that they have 5D III's...but they don't have D800's. If Nikon could get their manufacturing and supply chain in order, they might actually sell enough D800's to really hurt Canon, which might force them to think about all the photographers they are leaving behind in their relentless quest to fill the HDSLR void and satisfy the evil hordes of wedding photogs, sports photogs and photojournalists.



Personally, I feel like I'm in a bit of a bind. Few Canon photogs I know have ever said anything bad about Canon support. I've had a few experiences with them myself, once for a body and twice for lenses. Their turnaround is pretty quick, their service is pretty excellent, and the quality of their work is stellar. I have not heard the same from Nikon users, even die hard Nikon fans...from the way it sounds, Nikon support is anywhere from lackluster to terrible. Canon also has support centers pretty much everywhere in the world, and turnaround is pretty quick wherever you are, but I've not heard the same about Nikon. There are also what seem to be ever-present supply problems for Nikon gear, particularly professional-grade gear, in general. I've never once had problems finding Canon gear, online or locally, for anything from a small replacement part to expensive L-series glass. I might even be a Nikon user right now if it wasn't for the fact that the local stores didn't have the entry-level DSLR and lenses I wanted way back years ago when I first got into photography...all they had was a display model with a different lens (well, and the Nikon menu system...never cared for it, nor their body ergonomics. I clicked pretty quickly with Canon's scroll-less and color-coded menu system, and I loved their pro-body ergonomics...which are some of the things that ultimately swayed me towards Canon, that and the fact that the local stores actually had the Canon gear I wanted.)

Anyway, back to the bind. My landscape, macro, and astrophotography could really benefit from lower noise and better shadow recovery. Particularly my landscape work, which I haven't done much of over the last 8 months, and none of whatsoever this year so far. I always find myself needing to push shadows around, and they have never really looked all that good, on any of the Canon cameras I've used (450D, 7D, 5D II). My options are either to add Nikon to my overall kit, deal with a menu system which irks me and body ergonomics that don't fit my hands right...deal with what I've got...or deal with what Canon has to offer. None of those options really solve my problem well, and the only one that solves it at all will also cost thousands of dollars for a new body (D800) and a couple lenses to cover the bases...such as a 14-24 and maybe a 24-70. Thats about $7000 worth of gear...something I can't afford at all right now (I can't even afford a 5D III, which I expected to list for around $3000, some $200 over budget already.)

All I can say is its rather frustrating when your chosen brand just forgets you and shows no interest in even demonstrating they actually have not forgotten you, or that they are working on the problem, or that they have even acknowledged a problem (or gap, or discrepancy) even exists between them and their customers. Quite frustrating. So yes...it blows my mind that the 5D III has such atrocious read noise, when Canon has certainly done better in the past.

 

[V8Beast]

Aye, the 7D's read noise is quite a bit lower (although not nearly as low as an Exmor sensor, which ranks in at 2-4e-). I think the 7D has around 17-18e- read noise at ISO 100, which is roughly HALF that of the 5D III. I am honestly rather baffled by the read noise of the 5D III sensor. For all the progress that is apparently possible in the area of read noise, and how important it is to maximum DR, it really surprised me to see it JUMP from the 5D II's 27-28e- to over 33e-.

Another elementary question: Can there possibly be any benefit of increased read noise? I assume not, but it is very odd that it would increase on the 5DIII over the 5DII. Also, is read noise the same as visible noise in an image? If so, how does the 5DIII manage to do well at high ISO in terms of noise visible in the image when read noise is so high?

If Nikon could get their manufacturing and supply chain in order, they might actually sell enough D800's to really hurt Canon, which might force them to think about all the photographers they are leaving behind in their relentless quest to fill the HDSLR void and satisfy the evil hordes of wedding photogs, sports photogs and photojournalists.

I think that's the catch right there. As much as Canon is lagging behind in DR, the market for wedding photogs, sports photogs, and photojournalists is huge. Interestingly, the stuff I shoot is a cross between what's needed for a good wedding, sports, of photojournalism camera, which is probably why the 5DIII suits my needs very well, despite the read noise

Anyway, back to the bind. My landscape, macro, and astrophotography could really benefit from lower noise and better shadow recovery. Particularly my landscape work, which I haven't done much of over the last 8 months, and none of whatsoever this year so far. I always find myself needing to push shadows around, and they have never really looked all that good, on any of the Canon cameras I've used (450D, 7D, 5D II). My options are either to add Nikon to my overall kit, deal with a menu system which irks me and body ergonomics that don't fit my hands right...deal with what I've got...or deal with what Canon has to offer. None of those options really solve my problem well, and the only one that solves it at all will also cost thousands of dollars for a new body (D800) and a couple lenses to cover the bases...such as a 14-24 and maybe a 24-70. Thats about $7000 worth of gear...something I can't afford at all right now (I can't even afford a 5D III, which I expected to list for around $3000, some $200 over budget already.)

Have you considered a D5100 or D7000? They'll get almost as much DR as a D800 for a fraction of the cost.

 

[V8Beast]

Either Canon has found a way to greatly improve the SNR of the G12, or it must have pretty low read noise. It is a substantially different sensor than those found in Canon DSLR's, though...it may be backlit (which, much like microlensing, can greatly increase quantum efficiency), and if so, that would indeed help improve SNR. Unless Canon started work on the G12 sensor so late in the game that they couldn't migrate any of the read noise/QE improvements to the 5D III sensor, the ridiculously high read noise of the 5D III (which is the highest of any camera over the last several years, as far as I am aware) is completely unwarranted in my opinion. As such...the only logical conclusion I can really come to is that Canon simply does not care about low-ISO anymore. Their push for "improvements" would then seem to be driven by three large but very specific markets:

I brought up the G12 merely as an extreme example of the sensor size vs. DR correlation, but looking back as several generations of Canon DSLRs, the larger sensor bodies have never really outperformed crop sensors in the DR department, at least according to DxO. Let's see:

1DsII = 11.3 stops of DR
5DC = 11.1 stops of DR
20D = 11 stops of DR

So, the "bigger photosites collecting more light" argument never seems to have held up as far as DR is concerned. I'm just curious as to why

 

[ChodTheWacko]

So, the "bigger photosites collecting more light" argument never seems to have held up as far as DR is concerned. I'm just curious as to why

One of your problems is that your question is no good, because it is oversimplified.
It's kinda like asking something like 'How is gas mileage and horsepower related?'

I can almost guarantee that if you had two sensors, with identical technology and megapixels, a later sensor will have higher dynamic range. It just has larger photosites and smaller signal to noise ratio.

The problem, once you compare different chips/systems, there's tons of other factors.
Different circuitry. Different microlenses. Different noise reduction.

And what does a score of X for dynamic range mean? How does it take into account different types of noise, some of which is more distracting and less usable? Does a camera with less, but more offensive noise have the same score as one with more noise, but kinda ignorable grey noise?

- Frank

 

[Aglet]

I'll try put this answer this a shorter way that covers most of the bases; everything photographers need to know about Dynamic Range (and Signal to Noise Ratios) without getting bogged down in technical details.

DR is directly linked to Signal to Noise Ratio. That's because, at some level of darkness, the camera's overall noise signal means you can no longer measure meaningful levels of tonal variance, they're obscured by noise.

Bigger pixels can produce a larger SNR because, all other things assumed equal, they should be able to catch a larger signal (more electrons) than a smaller pixel when they're both "full." (at the highlight limit)

If the read and other noise part of the signal is the same (constant) for a large and small pixel, then the SNR is greater for the larger pixel. If you choose the same level of SNR as a cutoff level of usability, then the larger pixel is going to be able to provide more tonal information at the darker end than the smaller one, hence a greater overall DR.

The DR should decrease pretty much linearly with increasing ISO sensitivity because that's just the basic math of the above description. 1 stop more ISO = 1 stop less DR. This pretty much holds at high ISO for all cameras.

The difference occurs between Canon and SoNikon at the low ISO levels on their more recent sensors.

SoNikon pixels have much lower, but consistent read noise across their ISO range. Their resulting DR vs ISO is a nice linear plot, for the most part, which is close to theoretical ideal.

Canon seems to be using whatever original tech they developed way back, which is likely a variable analog gain stage between the pixel and the (ADC) circuit that converts the signal level to a digital number. This may have the effect of covering up some of the noise of the ADC system at the lower ISO levels which then results in that flattened DR vs ISO at the lower ISO ranges. Someone who really knows what's going on with Canon's sensors, readout circuitry and ADC system please correct me if I'm wrong on this aspect because it's a pretty strange DR vs ISO curve and this is the only explanation I came up with after thinking about it briefly.

So, a small pixel sensor, like in the G12, is capable of having the same DR at base ISO (it does) as a large pixel sensor like the 5D2/3 at the same base ISO if the G12 has a very low read noise (it does) compared to the larger pixel sensor. However, because the pixels are so small in the G12, only going up 2 stops from base ISO mean it now loses the DR comparison against the larger pixel'ed, but underperforming system of the 5D3.

If the 5D3's DR curve continued to the upper left on the same slope it has at higher ISO, then it'd be topping the base ISO chart near 14 stops too, much like the D800. But something about Canon's sensor system limitations prevents that.

AFAIK, Canon uses a lot of Sony-made (CCD) sensors in their compact cameras. I think the G11 and G12 use Sony-made sensors.

The above completely ignores the quality of the noise produced by the overall sensor and how it affects the final image produced. Fine-grained random noise and patterned noise may cause the same overall SNR and therefore DR measurement if the noise for both is used as an average value for the calculation. But we certainly notice the patterned noise a lot more than the random noise when we look at an image and random noise is also dealt with better than noise that has structure to it.
If noise structure was constant, always in the same place in the same way in the image, it would be possible to mathematically reduce it by averaging and subtraction. The problem with some of the pattern (banding) noise is that altho it has a pattern or structure, it isn't consistent from shot to shot; it shifts around. This likely occurs from frequency aliasing between noise on the camera's internal power supplies compared to the sensor readout frequency.
Canon, fix your internal power supplies and at least you'll reduce the darn random-banding issues.

so, cameras with the same DR and SNR ratio can produce images with different usability depending on the type of included noise, random or patterned. The camera producing more uniform random noise may be able to provide an extra stop or more of usable image quality.

 

[weixing]

Hi,
I believe the sensor technology gap between Sony/Nikon and Canon is not that big. IMHO,
the Canon sensor are just as good as Sony/Nikon sensor... the different is how they handle the signal and noise.

Here is what I think:
Canon RAW file include the bias signal which mean all the noise (positive and negative) are include in the
data, but Nikon RAW files somehow do not record the bias signal... I think may be Nikon filter out all those data, so only those information (signal and noise) that are above bias level are recorded in the image.

As a result, Nikon sensor perform better at lower ISO as noise below the bias level are not recorded in the image. But as ISO increase, those noise lower than the bias level are now above the bias level, so they are recorded in the image. So at higher ISO, performance of Nikon sensor are now on par with Canon sensor.

Now if you take the above and look at all the review on Nikon D800 and Canon 5D3, it does seem to be the case: at low ISO Nikon D800 perform better than Canon 5D3, but at high ISO, Canon 5D3 perform better due to a larger pixel size (higher signal) compare to smaller Nikon D800 pixel size (lower signal).

Just my $0.02.

Have a nice day.

 

[birtembuk]

Their push for "improvements" would then seem to be driven by three large but very specific markets:

* Video (which is rather annoying for stills photographers, who used to be the primary drivers of DSLR use)
* .......

" They’re in mood for Hollywood n’ Bollywood ... "

Fine with your EOS video Canon, but please, don’t leave us, still-phogs, in the OOF

 

[Astro]

i wrote it in another thread.

photographic dynamic range = full well capacity / read noise.

that´s it in short.
i also wrote that canon imo has to work on it´s analog digital conversion to get better DR.
the cleaner D800 signal and the higher DR seem to affirm that.

increasing the full well capacity while having smaller pixel is a hard task so i guess sony has managed to reduce the read noise siginficantly and increased the quantum efficiency.

i have not read the whole amout of writing above so i guess it was said already.


so... that said i read from various sources that most of the noise today (in low iso images) from digital SLR cameras is not read noise but photon statistic noise. you can influence photon statistic noise only one way... collecting more photons.
and that would mean higher full well capacity and better quantum efficiency.

but im not sure if its correct that read noise can´t be much improved these days.

 

[PeterJ]

Another elementary question: Can there possibly be any benefit of increased read noise? I assume not, but it is very odd that it would increase on the 5DIII over the 5DII. Also, is read noise the same as visible noise in an image? If so, how does the 5DIII manage to do well at high ISO in terms of noise visible in the image when read noise is so high?

With an ADC (analog to digital converter) there's a compromise between read speed and noise. While the resolution hasn't increased a lot the frame rate is 50% higher to it must be sampling over 50% faster than the 5D2. ADC converters get better all the time, but given the above I'd say they've made a compromise to favour the faster frame rate a little over read noise. I guess you could say that's the only 'advantage', getting more frames per second, which is best would depend on the use.

ISO is different because that largely comes into the analog sensor / amplifier domain. While lower read noise would potentially allow you to boost the signal digitally with less noise if the analog sections get the voltages up to acceptable levels in low light then for the ADC it can be doing the same job at higher ISO settings.

 

[well_dunno]

Thanks all for your input - that is helpful! I recall reading the exmor sensors had the processor on them decreasing the noise that is picked up during the transfer over the circuits whereas Canon's processors are not on the sensors due to which extra noise is picked up. So that is a part of the lower DR story?

Cheers!

 

[Canon-F1]

DR has more to do with the design of the sensor circuitry away from the pixel. The analogue to digital converters, for example. Interference in the circuit from other circuits or the camera. DR is all about quality.

nice platitude ... ...

if everything at the sensor/photosite level works optimal then the ADC is the factor that determines the achievable DR.

but i guess there is room for improvement on the sensor level.
for example the filters can be improved, the substrate can be improved, coatings can be improved, the fill factor can be improved, gate structure can be improved.

all things that affect quantum efficiency, well capacity.

 

[KarstenReis]

Roger from lensrentals wrote an article explaining some of this stuff.

http://www.lensrentals.com/blog/2012/02/sensor-size-matters-part-2

 

[jrista]

Aye, the 7D's read noise is quite a bit lower (although not nearly as low as an Exmor sensor, which ranks in at 2-4e-). I think the 7D has around 17-18e- read noise at ISO 100, which is roughly HALF that of the 5D III. I am honestly rather baffled by the read noise of the 5D III sensor. For all the progress that is apparently possible in the area of read noise, and how important it is to maximum DR, it really surprised me to see it JUMP from the 5D II's 27-28e- to over 33e-.

Another elementary question: Can there possibly be any benefit of increased read noise? I assume not, but it is very odd that it would increase on the 5DIII over the 5DII. Also, is read noise the same as visible noise in an image? If so, how does the 5DIII manage to do well at high ISO in terms of noise visible in the image when read noise is so high?

Its more than just "read" noise...that is why I tried to use the term "electronic noise". There are a whole variety of forms of noise in an electronic system, including read noise, but also dark current noise, thermal noise, differential noise (due to inconsistencies in the efficiencies of individual transistors for each photosite), etc. There are ways of mitigating each of these forms of noise. Canon sensors only seem to do a basic form of CDS, or correlated double-sampling, which mitigates a certain amount of dark current noise. Sony sensors, on the other hand, have a variety of ways of mitigating most types of electronic noise, which is why their sensors perform better.

As for high ISO, I am not sure of the exact mechanics, but I believe the black point rises as you increase ISO (thanks to Canon's use of a bias offset of 2048, signal actually starts at -2048 and rises from there). You effectively lose DR on both ends, shadows and highlights, as ISO is increased. By ISO 400 or so, it seems the black point is high enough that it starts above the electronic noise floor, meaning the only really significant form of noise left is photon shot noise (a matter of physics and therefor beyond our control.) Sony sensors, it seems, don't have a bias offset, so they are pretty much always starting from the same point (zero, effectively), and they lose DR on the highlight end of the scale as you increase ISO. Their shadow DR seems to be pretty static.

Anyway, back to the bind. My landscape, macro, and astrophotography could really benefit from lower noise and better shadow recovery. Particularly my landscape work, which I haven't done much of over the last 8 months, and none of whatsoever this year so far. I always find myself needing to push shadows around, and they have never really looked all that good, on any of the Canon cameras I've used (450D, 7D, 5D II). My options are either to add Nikon to my overall kit, deal with a menu system which irks me and body ergonomics that don't fit my hands right...deal with what I've got...or deal with what Canon has to offer. None of those options really solve my problem well, and the only one that solves it at all will also cost thousands of dollars for a new body (D800) and a couple lenses to cover the bases...such as a 14-24 and maybe a 24-70. Thats about $7000 worth of gear...something I can't afford at all right now (I can't even afford a 5D III, which I expected to list for around $3000, some $200 over budget already.)

Have you considered a D5100 or D7000? They'll get almost as much DR as a D800 for a fraction of the cost.

Well, for landscapes, I really want that extra resolution. A low resolution of 12mp really doesn't interest me much, and I was hoping the 5D III would land with 28-32mp. The D800E with 36.3mp and its incredible DR is about as close to the holy grail of landscape photography as I think it can get these days. I'd much prefer if Canon could reciprocate with their own megapixel/DR monster, though.

 

[jrista]

Hi,
I believe the sensor technology gap between Sony/Nikon and Canon is not that big. IMHO,
the Canon sensor are just as good as Sony/Nikon sensor... the different is how they handle the signal and noise.

Here is what I think:
Canon RAW file include the bias signal which mean all the noise (positive and negative) are include in the
data, but Nikon RAW files somehow do not record the bias signal... I think may be Nikon filter out all those data, so only those information (signal and noise) that are above bias level are recorded in the image.

As a result, Nikon sensor perform better at lower ISO as noise below the bias level are not recorded in the image. But as ISO increase, those noise lower than the bias level are now above the bias level, so they are recorded in the image. So at higher ISO, performance of Nikon sensor are now on par with Canon sensor.

Now if you take the above and look at all the review on Nikon D800 and Canon 5D3, it does seem to be the case: at low ISO Nikon D800 perform better than Canon 5D3, but at high ISO, Canon 5D3 perform better due to a larger pixel size (higher signal) compare to smaller Nikon D800 pixel size (lower signal).

Just my $0.02.

Have a nice day.

Nikon cameras do not include a bias offset for the simple reason that the sensors they use now do not have enough noise to warrant it. Sony sensors produce excessively low noise around the 3e- level. Its pretty difficult to get much lower than that. SoNikon simply doesn't need a bias offset because there effectively isn't any noise to interfere with the image signal. The reason noise at higher ISO is on par with Canon is because the very vast majority of that noise is photon shot noise, resulting from the physical nature of photons, something we have no control over in the first place. Once shot noise dominates, there really isn't any difference between one brand or another...its all just physics.

 

[jrista]

i wrote it in another thread.

photographic dynamic range = full well capacity / read noise.

that´s it in short.
i also wrote that canon imo has to work on it´s analog digital conversion to get better DR.
the cleaner D800 signal and the higher DR seem to affirm that.

increasing the full well capacity while having smaller pixel is a hard task so i guess sony has managed to reduce the read noise siginficantly and increased the quantum efficiency.

i have not read the whole amout of writing above so i guess it was said already.


so... that said i read from various sources that most of the noise today (in low iso images) from digital SLR cameras is not read noise but photon statistic noise. you can influence photon statistic noise only one way... collecting more photons.
and that would mean higher full well capacity and better quantum efficiency.

but im not sure if its correct that read noise can´t be much improved these days.

Sony definitely has better ADC. The difference between Sony ADC and Canon ADC is that Sony's is on the sensor die in Exmor sensors, using CP-ADC or column-parallel ADC. This gives them a lot of power to control the read noise on a per-column basis, helping eliminate forms of fixed pattern noise. Canon's ADC is off-die in the image processor (Digic). It is also parallelized, I believe there are 8 read channels per DIGIC in the 1D X and 5D III, and 1 ADC per channel. Canon's ADC's have to work at a higher frequency since each one is processing more pixels than Sony's on-die CP-ADC, which introduces more noise. They also don't have the ability to filter out column-level noise like Sony does, so their ADC's convert FPN into each ADU as well.