Questions about Dynamic Range on DXOMark and 14-bit / 16-bit ADC

SixteenBit said:

I have a few questions about Dynamic Range testing on DXOMark and about 14-bit vs 16-bit ADC.

Dynamic Range Ratings by DXOMark (sorted by Print, normalized to same print size)
1) Nikon D800 - Screen 13.23 Ev, Print 14.33 Ev
2) Pentax K-5 - Screen 13.61 Ev , Print 14.12 Ev
3) Nikon D7000 - Screen 13.35 Ev, Print 13.87 Ev
4) Nikon D3x - Screen 12.84 Ev , Print 13.65 Ev
5) Nikon D5100 - Screen 13.04 Ev, Print 13.56 Ev
6) Phase One IQ180 - Screen 11.89 Ev, Print 13.56 Ev
7) Fujifilm S5 Pro (6mp) - Screen 13.71 Ev, Print 13.51 Ev
8) Fujifilm S3 Pro (6mp) - Screen 13.65 Ev, Print 13.45 Ev
9) Sony NEX-7 - Screen 12.59 Ev, Print 13.39 Ev
10) Sony Alpha 580 - Screen 12.76 Ev, Print 13.27 Ev

Dynamic Range Ratings, sorted by Screen (per-pixel performance)
1) Fujifilm S5 Pro (6mp) - Screen 13.71 Ev, Print 13.51 Ev
2) Fujifilm S3 Pro (6mp) - Screen 13.65 Ev, Print 13.45 Ev
3) Pentax K-5 - Screen 13.61 Ev , Print 14.12 Ev
4) Nikon D7000 - Screen 13.35 Ev, Print 13.87 Ev
5) Nikon D800 - Screen 13.23 Ev, Print 14.33 Ev

From

http://www.dxomark.com/index.php/About/In-depth-measurements/DxOMark-testing-protocols/Noise-dynamic-range

"We use filters having different light absorption levels ranging from 0% to 99.99% in order to test across a dynamic range of 4 density steps (= 13.3 f-stops — a dynamic range much greater than today’s digital cameras). When shooting such a chart, the sensor of the camera being tested sees a wide range of light levels, with a 1/10,000 ratio from minimum to maximum."

Question 1) due to 13.3-stop limitation, are test results from 13+ DR sensors are still accurate and reliable ?

Question 2) Even Nikon and Pentax use Sony CMOS. The main difference is that Nikon and Pentax uses 14-bit ADC but Sony uses 12-bit ADC. This seems to limit DR of sensor. Will 14+ stop DR get a little more improvement from 16-bit ADC (linear) or still no changes ?

Question 3) assume sensor has 14-stop DR and 12-stop usable DR. In case of 4-6 stop shadow lifting, will 16-bit ADC yield better overall image quality than 14-bit ADC due to more raw levels in shadows ? or still no benefits ?

Linear Raw Levels
Ev 16 bit 14 bit 12 bit
+3 32768 8192 2048
+2 16384 4096 1024
+1 8192 2048 512
0 4096 1024 256
-1 2048 512 128
-2 1024 256 64
-3 512 128 32
-4 256 64 16
-5 128 32 8
-6 64 16 4
-7 32 8 2
-8 16 4 1
-9 8 2
-10 4 1
-11 2
-12 1

Click to expand...

fvdbergh2501 said:

The dynamic range of a sensor is typically limited by the signal to noise ratio (SNR), not the bit depth of the analogue to digital converter (ADC). Of course, I am assuming that we have sufficient bit depth to begin with, as in your example of going from 14 bits to 16 bits.

Take for example the D7000, which has exceptionally low read noise. IIRC, Marianne's measurements put the read noise at about 3.7e-, or 1.4DN. In other words, the read noise's magnitude is greater than the step between adjacent 14-bit numbers.

This implies that noise will prevent any details from being captured in the lower two bits of a 16-bit ADC.

You cannot use the extra bits at the bright end either, because the standard deviation of shot noise will be more than 50 DN near full well capacities.

The DxOMark normalized (print) scores are derived from 8MP images, which means the averaging of neighbouring pixels will help to reduce read noise, thus increasing SNR. The D800 is a good example: if you bin 2x2 groups of pixels, you will obtain (roughly) an 8MP image with read noise reduced by a factor of 2, i.e., then you effectively end up with 15-bit DNs.

Lastly, measuring engineering DR using Marianne's method (or Emil's method, which is the same thing) does not require the full range to be measured to compute accurate DR values. A single point on the PTC will give you the required SNR value, and hence the DR. Not sure if this is how they calculate DR at DxO, but is is certainly possible using their stepped exposure data.

Click to expand...

SixteenBit said:

16-bit ADC could be overkill for digital backs, right ?

Click to expand...

falconeyes said:

SixteenBit said:

16-bit ADC could be overkill for digital backs, right ?

Click to expand...

You already got the answer, yes.

There is another take at this: Even at zero readout noise, you don't need higher bit depth than you have electrons in your pixel well. Sony sensors are all below 50,000 e-. The 14 bit limit is 65,000. This may be easier to understand.

The D4 has larger pixels and a larger well, with more than 100,000 e- capacity. But like digital backs, it doesn't need a 15 or 16 bit AD because read noise still is much larger than quantization noise.

Click to expand...

SixteenBit said:

I wonder if 16-bit ADC could be adding a little more quality or flexibility to D800 files like wider range of shadow/highlight recovery or better shadow details due to more raw levels. If not, then 16-bit ADC could be overkill for digital backs, right ?

Click to expand...

fvdbergh2501 said:

SixteenBit said:

I wonder if 16-bit ADC could be adding a little more quality or flexibility to D800 files like wider range of shadow/highlight recovery or better shadow details due to more raw levels. If not, then 16-bit ADC could be overkill for digital backs, right ?

Click to expand...

I'll rephrase my previous response: read noise on the D800 is about 1.4 DN (see Bill Claff's charts) . In a dark patch of uniform colour (deep shadow), two nearby photosites (say green) will have values that randomly differ by more than 1 DN (gray level) about 48% of the time, meaning you only have 13 bits of non-random data. For example, your 14-bit number for one photosite might be 112 (out of 16486), and its neighbour will be 113, even though the light levels were identical at both photosites.

This is still only read noise. Photon shot noise will make this situation worse.

Adding a 16-bit ADC will simply mean that your read noise will now be 1.4 DN * 4, i.e., 5.6 DN in your new 16-bit range. In your uniform dark patch, nearby photosites will now differ by 1 DN about 86% of the time, and they will differ by 4 DN about 48% of the time.

In short, the 16-bit ADC will allow you to measure the noise more accurately, but it will not improve DR. You will be able to "lift the shadows" more, but you will not see any real image detail, just more noise.

As falconeyes pointed out, the only real reason for a 16-bit ADC would be if your full well capacity was quite large, but this will typically only happen if you have photosite much larger than 8 microns.

Click to expand...

Zlik said:

fvdbergh2501 said:

SixteenBit said:

I wonder if 16-bit ADC could be adding a little more quality or flexibility to D800 files like wider range of shadow/highlight recovery or better shadow details due to more raw levels. If not, then 16-bit ADC could be overkill for digital backs, right ?

Click to expand...

I'll rephrase my previous response: read noise on the D800 is about 1.4 DN (see Bill Claff's charts) . In a dark patch of uniform colour (deep shadow), two nearby photosites (say green) will have values that randomly differ by more than 1 DN (gray level) about 48% of the time, meaning you only have 13 bits of non-random data. For example, your 14-bit number for one photosite might be 112 (out of 16486), and its neighbour will be 113, even though the light levels were identical at both photosites.

This is still only read noise. Photon shot noise will make this situation worse.

Adding a 16-bit ADC will simply mean that your read noise will now be 1.4 DN * 4, i.e., 5.6 DN in your new 16-bit range. In your uniform dark patch, nearby photosites will now differ by 1 DN about 86% of the time, and they will differ by 4 DN about 48% of the time.

In short, the 16-bit ADC will allow you to measure the noise more accurately, but it will not improve DR. You will be able to "lift the shadows" more, but you will not see any real image detail, just more noise.

As falconeyes pointed out, the only real reason for a 16-bit ADC would be if your full well capacity was quite large, but this will typically only happen if you have photosite much larger than 8 microns.

Click to expand...

So in other words "tell the same story with more words"

Thanks for your technical input.

Click to expand...

SixteenBit said:

Zlik said:

fvdbergh2501 said:

SixteenBit said:

I wonder if 16-bit ADC could be adding a little more quality or flexibility to D800 files like wider range of shadow/highlight recovery or better shadow details due to more raw levels. If not, then 16-bit ADC could be overkill for digital backs, right ?

Click to expand...

I'll rephrase my previous response: read noise on the D800 is about 1.4 DN (see Bill Claff's charts) . In a dark patch of uniform colour (deep shadow), two nearby photosites (say green) will have values that randomly differ by more than 1 DN (gray level) about 48% of the time, meaning you only have 13 bits of non-random data. For example, your 14-bit number for one photosite might be 112 (out of 16486), and its neighbour will be 113, even though the light levels were identical at both photosites.

This is still only read noise. Photon shot noise will make this situation worse.

Adding a 16-bit ADC will simply mean that your read noise will now be 1.4 DN * 4, i.e., 5.6 DN in your new 16-bit range. In your uniform dark patch, nearby photosites will now differ by 1 DN about 86% of the time, and they will differ by 4 DN about 48% of the time.

In short, the 16-bit ADC will allow you to measure the noise more accurately, but it will not improve DR. You will be able to "lift the shadows" more, but you will not see any real image detail, just more noise.

As falconeyes pointed out, the only real reason for a 16-bit ADC would be if your full well capacity was quite large, but this will typically only happen if you have photosite much larger than 8 microns.

Click to expand...

So in other words "tell the same story with more words"

Thanks for your technical input.

Click to expand...

Thanks for info. Actually, I am thinking about playing with raw files from 14bit DSLR and 16bit digital back , maybe between D800 and IQ180 at ISO100. And try to lift shadows 4-5 stops and compare the results.

Click to expand...

Zlik said:

SixteenBit said:

Zlik said:

fvdbergh2501 said:

SixteenBit said:

I wonder if 16-bit ADC could be adding a little more quality or flexibility to D800 files like wider range of shadow/highlight recovery or better shadow details due to more raw levels. If not, then 16-bit ADC could be overkill for digital backs, right ?

Click to expand...

I'll rephrase my previous response: read noise on the D800 is about 1.4 DN (see Bill Claff's charts) . In a dark patch of uniform colour (deep shadow), two nearby photosites (say green) will have values that randomly differ by more than 1 DN (gray level) about 48% of the time, meaning you only have 13 bits of non-random data. For example, your 14-bit number for one photosite might be 112 (out of 16486), and its neighbour will be 113, even though the light levels were identical at both photosites.

This is still only read noise. Photon shot noise will make this situation worse.

Adding a 16-bit ADC will simply mean that your read noise will now be 1.4 DN * 4, i.e., 5.6 DN in your new 16-bit range. In your uniform dark patch, nearby photosites will now differ by 1 DN about 86% of the time, and they will differ by 4 DN about 48% of the time.

In short, the 16-bit ADC will allow you to measure the noise more accurately, but it will not improve DR. You will be able to "lift the shadows" more, but you will not see any real image detail, just more noise.

As falconeyes pointed out, the only real reason for a 16-bit ADC would be if your full well capacity was quite large, but this will typically only happen if you have photosite much larger than 8 microns.

Click to expand...

So in other words "tell the same story with more words"

Thanks for your technical input.

Click to expand...

Thanks for info. Actually, I am thinking about playing with raw files from 14bit DSLR and 16bit digital back , maybe between D800 and IQ180 at ISO100. And try to lift shadows 4-5 stops and compare the results.

Click to expand...

You would be comparing sensor technologies and not 14 vs 16 bit.

Click to expand...