Phase One IQ3 100mp. Shipping now.

[ErikKaffehr]

Hi,

The Sony A7rII has a special trick called the Aptina Patent. They reduce full well capacity at 640 ISO by disconnecting an internal capacitor. That reduces readout noise a bit. At real high ISOs they probably just kick in noise reduction.

Quite smart actually.

Best regards
Erik

Yes, it seems the D810 gets the most DR out of the sony 35mm sensor.

Low iso performance is definitely a priority on a back like the IQ3 100MP while the A7RII being a much more consumer oriented product is tuned more for high performance across the whole iso range with obvious bumps (compared to the D810) in performance at several iso settings.

I have no clue what each company does on top of what the sensor can deliver from the factory or if some changes are made inside the sensor itself or if its a combination of hardware and programming (on sensor).

I don't think anyone without a signed NDA really knows.

I though Sony improved the depth and DR of the A7RII with the new firmware? and got it to 14bit.

[Bart_van_der_Wolf]

I've done a bit more Read-noise and Dynamic Range number crunching based on some more files, and at different ISOs. The conclusion is that the 16-bit lossless mode does add DR compared to the 14-bit 'lossless' mode, but it remains a bit below the claimed 15 stops. Also, there is no real difference in readnoise between nominal ISO 50 and ISO 100. Read noise for ISO 3200 and above remains unchanged, so becomes basically ISO-less.

Attached are the summarized 14-bit mode shot results of the 'Roundtower' sequence. The numerical differences between the R/G/B/G2 color planes are marginal, they all show very similar values.

I've used RawDigger Version 1.2.4 Build 437 (which is supposed to read the 16-bit mode correctly) to read the files. The DR of 13.65 stops at nominal ISO 50 in 14-bit mode, change to a DR of 14.39 stops in 16-bit mode (not quite the claimed 15-bits but still very high).

Read noise drops from a standard deviation of 5.03 DN to 3 DN, at ISO 50 (and I assume at ISO 100, but I have no file to verify yet). Base ISO therefore seems to be nominal ISO 100. We'll see when DxOlabs publish their analysis how they rate the real ISO sensitivity.

At ISO 800 the DR changes from 11.66 stops in 14-bit mode, to 12.06 stops in 16-bit mode.

The observed 4-DN combing of the histograms in 14-bit mode, indeed disappears in 16-bit mode, so the electronics do seem to produce 16-bit native signal levels, with a noise floor in the 1.61 LSBs (least significant bits).

Low ISO read-noise is very low, and DR is impressive, assuming these files are representative for a larger population of sensors.

Cheers,
Bart

[ErikKaffehr]

Hi,

I guess that Sony/Phase One really pulled that rabbit out of that hat!

As far as I recall, Nikon D810 was the camera with the highest DR (at the pixel level), and DxO measured that camera to 13.67EV, so this reaches 14.39 (according to your calculations) and that is quite a bit more.

On the other hand, I have noticed that D810 was close to 14 bit limit and DR. Going to 16 bit may have been needed to fully utilising those sensor pixels.

Seems like Phase One put together a really impressive back!

Just to add, thanks for sharing your findings!

Best regards
Erik

I've done a bit more Read-noise and Dynamic Range number crunching based on some more files, and at different ISOs. The conclusion is that the 16-bit lossless mode does add DR compared to the 14-bit 'lossless' mode, but it remains a bit below the claimed 15 stops. Also, there is no real difference in readnoise between nominal ISO 50 and ISO 100. Read noise for ISO 3200 and above remains unchanged, so becomes basically ISO-less.

Attached are the summarized 14-bit mode shot results of the 'Roundtower' sequence. The numerical differences between the R/G/B/G2 color planes are marginal, they all show very similar values.

I've used RawDigger Version 1.2.4 Build 437 (which is supposed to read the 16-bit mode correctly) to read the files. The DR of 13.65 stops at nominal ISO 50 in 14-bit mode, change to a DR of 14.39 stops in 16-bit mode (not quite the claimed 15-bits but still very high).

Read noise drops from a standard deviation of 5.03 DN to 3 DN, at ISO 50 (and I assume at ISO 100, but I have no file to verify yet). Base ISO therefore seems to be nominal ISO 100. We'll see when DxOlabs publish their analysis how they rate the real ISO sensitivity.

At ISO 800 the DR changes from 11.66 stops in 14-bit mode, to 12.06 stops in 16-bit mode.

The observed 4-DN combing of the histograms in 14-bit mode, indeed disappears in 16-bit mode, so the electronics do seem to produce 16-bit native signal levels, with a noise floor in the 1.61 LSBs (least significant bits).

Low ISO read-noise is very low, and DR is impressive, assuming these files are representative for a larger population of sensors.

Cheers,
Bart

[Bart_van_der_Wolf]

Hi,

I guess that Sony/Phase One really pulled that rabbit out of that hat!

As far as I recall, Nikon D810 was the camera with the highest DR (at the pixel level), and DxO measured that camera to 13.67EV, so this reaches 14.39 (according to your calculations) and that is quite a bit more.

Hi Erik, yes that is indeed impressive (and so are the filesizes of course).

It may be that with a bit of tweaking of the sensor calibration/stitching/LCCs/NoiseReduction one also achieves a superior 'photographic DR' and not only an engineering DR. Something else I noticed earlier with a DCamProf profile for the EOS 5Ds that 'AlterEgo' made available in that application's main thread, is that the profile may have a significant (in that case positive) impact on the demosaiced and rendered noise. So the Raw conversion quality also makes a difference.

On the other hand, I have noticed that D810 was close to 14 bit limit and DR. Going to 16 bit may have been needed to fully utilising those sensor pixels.

Absolutely. Outputting more than 14-bit data from the ADC (by whatever means) is necessary to exceed the 14 stop boundary.

Seems like Phase One put together a really impressive back!

They sure did. I think it's not just the Sony sensor that made it possible, but we can only be sure if others start using the sensor as well.

Just to add, thanks for sharing your findings!

You're welcome.

I wanted to satisfy my own curiosity, and I don't mind sharing what I've found so far. Much more can be done to unravel the under-the-hood state of affairs, but that would require constructing Photon Transfer function curves for the different color planes at different ISOs. That would be a lot of work, based on image pairs at many exposure levels that need to be subtracted to isolate pattern noise, Read noise, reset noise and photon noise, and determine ADC gain and other things that one might stumble on. So far there seems to be no White balance preconditioning, but maybe something like that will show up at specific settings.

Cheers,
Bart

[torger]

Something else I noticed earlier with a DCamProf profile for the EOS 5Ds that 'AlterEgo' made available in that application's main thread, is that the profile may have a significant (in that case positive) impact on the demosaiced and rendered noise. So the Raw conversion quality also makes a difference.

That was an "aha"-experience during the development of DCamProf. It's not a good idea for the profile to push saturation of darkest shadows a lot, as it just makes the chroma noise more visible....

[Bart_van_der_Wolf]

That was an "aha"-experience during the development of DCamProf. It's not a good idea for the profile to push saturation of darkest shadows a lot, as it just makes the chroma noise more visible....

Hi Anders,

Indeed, and it's very effective in restraining the not very noise free 5DS Raw data conversions. The IQ3 100 sensor needs a lot less help with its 3 stops better DR, but it won't hurt either for less well controlled lighting conditions, e.g. landscape and natural light architecture, or at higher ISOs.

Cheers,
Bart

[Jim Kasson]

Yes, it seems the D810 gets the most DR out of the sony 35mm sensor.

Low iso performance is definitely a priority on a back like the IQ3 100MP while the A7RII being a much more consumer oriented product is tuned more for high performance across the whole iso range with obvious bumps (compared to the D810) in performance at several iso settings.

I have no clue what each company does on top of what the sensor can deliver from the factory or if some changes are made inside the sensor itself or if its a combination of hardware and programming (on sensor).

I don't think anyone without a signed NDA really knows.

I though Sony improved the depth and DR of the A7RII with the new firmware? and got it to 14bit.

The new uncompressed raw mode does not improve the EDR of the a7RII.

http://blog.kasson.com/?p=12287

Even at ISO 100, the camera has sufficient read noise to adequately dither the ADCs at 13 bits precision.

Jim

[Miyata610]

14.39 is very impressive indeed.

In fact, it needs only 0.11 more to be 15 stops (as opposed to 15.0 stops).

This is probably within margin of error.

[ErikKaffehr]

Hi Jim,

In my simple tests I did not see an improvement in dark noise. But I have noticed that dark tones that were pushed 4-5 stops remained neutral grey with the new uncompressed formate but went like brownish with the old formate.

So I felt the new file format yielded more neutral colours in extreme pushing. I did not come up with an explanation for the effect.

I have some samples, if needed, but it would take some time to find.

Best regards
Erik

The new uncompressed raw mode does not improve the EDR of the a7RII.

http://blog.kasson.com/?p=12287

Even at ISO 100, the camera has sufficient read noise to adequately dither the ADCs at 13 bits precision.

Jim

[alatreille]

That was an "aha"-experience during the development of DCamProf. It's not a good idea for the profile to push saturation of darkest shadows a lot, as it just makes the chroma noise more visible....

Could some please provide a link to this thread?  I had a look but couldn't seem to find it. 

Thank you


Andrew

Sent from my LG-H812 using Tapatalk

[voidshatter]

I've done a bit more Read-noise and Dynamic Range number crunching based on some more files, and at different ISOs. The conclusion is that the 16-bit lossless mode does add DR compared to the 14-bit 'lossless' mode, but it remains a bit below the claimed 15 stops. Also, there is no real difference in readnoise between nominal ISO 50 and ISO 100. Read noise for ISO 3200 and above remains unchanged, so becomes basically ISO-less.

Attached are the summarized 14-bit mode shot results of the 'Roundtower' sequence. The numerical differences between the R/G/B/G2 color planes are marginal, they all show very similar values.

I've used RawDigger Version 1.2.4 Build 437 (which is supposed to read the 16-bit mode correctly) to read the files. The DR of 13.65 stops at nominal ISO 50 in 14-bit mode, change to a DR of 14.39 stops in 16-bit mode (not quite the claimed 15-bits but still very high).

Read noise drops from a standard deviation of 5.03 DN to 3 DN, at ISO 50 (and I assume at ISO 100, but I have no file to verify yet). Base ISO therefore seems to be nominal ISO 100. We'll see when DxOlabs publish their analysis how they rate the real ISO sensitivity.

At ISO 800 the DR changes from 11.66 stops in 14-bit mode, to 12.06 stops in 16-bit mode.

The observed 4-DN combing of the histograms in 14-bit mode, indeed disappears in 16-bit mode, so the electronics do seem to produce 16-bit native signal levels, with a noise floor in the 1.61 LSBs (least significant bits).

Low ISO read-noise is very low, and DR is impressive, assuming these files are representative for a larger population of sensors.

Cheers,
Bart

Hi Bart,

Thanks for your input. I have analyzed the darkframe shots I made. For the IQ3100 case it was shot in 16-bit mode. I used RawDigger 1.2.4 (black level=0) to manually select a sub-region in each darkframe to avoid tiling of readout noise. Then I used the following to calculate the dynamic range:

DR = ln( (65535 - mean(darkframe)) / std(darkframe) ) / ln(2)

It appears that the DR of this new sensor is really great!

See attached plot:

[AlterEgo]

Could some please provide a link to this thread? 

the main DCamProf topic = http://forum.luminous-landscape.com/index.php?topic=100015.0

[Jim Kasson]

Hi Jim,

In my simple tests I did not see an improvement in dark noise. But I have noticed that dark tones that were pushed 4-5 stops remained neutral grey with the new uncompressed formate but went like brownish with the old formate.

So I felt the new file format yielded more neutral colours in extreme pushing. I did not come up with an explanation for the effect.

I have some samples, if needed, but it would take some time to find.

I've noticed that, too.  Take a look at these asymmetrical dark-field histos in compressed mode:

http://blog.kasson.com/?p=12299

I'll look around and see what else I can find.

Jim

[AreBee]

Does colour reflected by a sensor's physical surface visibly change as pixel density increases, similar to the way in which colour reflected by CD/DVD/Blu-Ray visibly changes as pit density increases?

[Jim Kasson]

In my simple tests I did not see an improvement in dark noise. But I have noticed that dark tones that were pushed 4-5 stops remained neutral grey with the new uncompressed formate but went like brownish with the old formate.

So I felt the new file format yielded more neutral colours in extreme pushing. I did not come up with an explanation for the effect.

Eric, I found the testing that I did that bears on that point:

http://blog.kasson.com/?p=12372

It quantitatively validates your findings. The bias is low, and will not be material for most photographers, I think.

Jim

[AreBee]

Bart,

A 4.6 micron pitch will mean that at f/5.0 or narrower, diffraction will start to function as an OLPF of sorts, to reduce aliasing effects even further than the dense sampling already does, and apertures of f/16 or narrower will lose effective resolution due to diffraction...

Setting aside advice not to shoot at an aperture of f/16 or narrower for exactly that reason, what can one say if one does? Does deconvolution sharpening return zero benefit?

[ErikKaffehr]

Hi,

When MTF at Nyquist reaches zero, no more detail can be recovered regardless of sharpening method.

But, visual perception is dominated by low frequencies and they are still improved by sharpening.

Another way to see it is that system MTF is like:

MTF(system) = MTF(lens design) * MTF(diffraction) * MTF(sensor)

Increasing MTF for any part still improves MTF(system) unless any part goes to zero.

I would suggest that diffraction needs to be taken into account when stopping down for DoF. The CoC in the calculation should be larger than the diameter of the diffraction disc. Bart can explain the fine details, he has both the math and the language skills :-)

Best regards
Erik

Bart,

Setting aside advice not to shoot at an aperture of f/16 or narrower for exactly that reason, what can one say if one does? Does deconvolution sharpening return zero benefit?

[ErikKaffehr]

Hi,

That is angle dependent. So different spectral line will be have maxima at different angles.

Best regards
Erik

Does colour reflected by a sensor's physical surface visibly change as pixel density increases, similar to the way in which colour reflected by CD/DVD/Blu-Ray visibly changes as pit density increases?

[AreBee]

Erik,

When MTF at Nyquist reaches zero, no more detail can be recovered regardless of sharpening method. But, visual perception is dominated by low frequencies and they are still improved by sharpening.

I did wonder. Thank you.

That is angle dependent. So different spectral line will be have maxima at different angles.

In layman's terms, is that a "Yes"?

[ErikKaffehr]

Hi,

If you look illuminate and look from the same angle you would see a different colour.

Best regards
Erik

Erik,

I did wonder. Thank you.

In layman's terms, is that a "Yes"?