The GetDPI Photography Forum

Great to see you here. Join our insightful photographic forum today and start tapping into a huge wealth of photographic knowledge. Completing our simple registration process will allow you to gain access to exclusive content, add your own topics and posts, share your work and connect with other members through your own private inbox! And don’t forget to say hi!

Noise of 6,0µm Kodak CCD (H3DII-50 & S2)

georgl

New member
The new Hasselblad H3DII-50 uses the first entirely new Kodak full-frame-CCD-architecture since 4 years, has anybody compared noise & DR with it's predecessors (6,8µm -> H3DII-39)?

It would be quite interesting to see how much has changed, especially since the Leica S2 will use the same sensor-architecture?

What about Dalsa's new 6,0µm-CCDs (P65+ for example) - nobody compared yet?

Thanks!
 

Professional

Active member
We need a lot of money to get those bodies to compare it, otherwise wait some pros and experts to get their hands on them all to compare.
 

dougpeterson

Workshop Member
As anyone who has ever tried to do direct head-to-head DR analysis knows it can be much more difficult then setting up a shot, shooting the first back, swapping backs, and shooting again. Variation in color spaces, processing choices, and different true native ISOs can all impact the head to head example. When you're looking for what you already know will be relatively minor differences (0.5 stop of DR at the most either direction) these otherwise trivial differences can really obscure your results.

Jack and Guy did some tests in Moab with the P65+ I brought and Guy's P25+ and Dave, our owner, did some tests in ATL. As best as I can tell the DR is a bit higher than either the P25+ or the P45+. Given that the size of the pixel has decreased I think this is an achievement in of itself.

There are actually a lot of fun to learn about (if you're a big nerd) factors in the architecture of a digital back which result in the system's DR. The quoted pixel size on tech sites is almost always the largest possible way of measuring each pixel. The part of each pixel which is actually light sensitive was, especially in the 9 micron sensors, much smaller. So when Kodak went from the P25+ to the P45+ the size of the light-sensitive area of each pixel barely changed, while the wasted (non-sensitive) area within each pixel (the electronics package and border of each pixel) was drastically reduced. Likewise with the new Dalsa 6 micron design the wasted space has been effectively eliminated so the loss of light sensitive is a little less than the 6.8 --> 6.0 micron change would indicate. Then after the sensor has pulled in all the photons it can the resulting electrical charge has to be read off of each pixel (generally down a row like "pass the bucket") and to a AD convertor where the charge is translated to a numerical value. The accuracy of this conversion is governed not only by the bit depth (e.g. 12, 14, 16) but also the quality of AD convertor (i.e. not all 16 bit convertors perform identically well). Depending on the architecture there is some (usually patented way of reading a black-calibration off of the unexposed chip and negating much of the noise of the chip). Once the signal is converted to digital you are left with a very unusable array of bayer-patterned pixels. It's then up to the math-gods of the raw conversion software to make a pretty picture from an otherwise random-looking group of pixels. The math here is so complicated that there are literally only a handful of individuals in the entire world who could truly be called experts on it. Finally you have a picture you can print or compare (and I've surely omitted several steps).

Don't know where this ramble came from now that I've arrived here. Other than to say a LOT of work goes into trying to squeeze out the last bit of quality from these sensors. And while the theoretical ceiling of DR may decline with the declining size of the pixel it is often the case that the actual performance stays the same or increases as the other mitigating factors that prevent you from reaching that ceiling are alleviated.

Sometimes I wonder if anybody but Bob makes it to the end of my poorly structured and inadequately explained technical rants. Hi Bob!

Doug Peterson, Head of Technical Services
Capture Integration, Phase One & Canon Dealer | Personal Portfolio
 

Guy Mancuso

Administrator, Instructor
We read them Doug. Just to add when we did the p25,P45 and P65 Plus backs the DR went up as the microns went down. Interestingly the P65 plus was about 2/3 a stop better than the P25 plus and about a 1/3 better than the P45 Plus. Now i will add these are are visual findings and not some dumb DXO numbers that tells you basically nothing. I'm a real shooter that shoots real equipment in the real world and not test machines. :ROTFL::ROTFL::ROTFL:
 

georgl

New member
Thanks for your extensive post, Doug!

I understand that various aspects besides the sensor itself influence the RAW-IQ. But when these aspects are handled carefully (thermal management, converters...) it all comes back to the sensor.

I suspect that all these aspects are handled similary between the mentioned Hasselblad/Imacon-models (39 vs. 50MP) and whenever two MFDBs used the same sensor, their perfomance wasn't to different, so I don't think the S2 will be too different from any other cropped to (37,5MP) 6,0µm-Kodak-based MFDBs regarding RAW-IQ.

Some people here already have the H3DII-50 (and had the 39MP-version), or am I wrong?
 

Professional

Active member
Thanks for your extensive post, Doug!

I understand that various aspects besides the sensor itself influence the RAW-IQ. But when these aspects are handled carefully (thermal management, converters...) it all comes back to the sensor.

I suspect that all these aspects are handled similary between the mentioned Hasselblad/Imacon-models (39 vs. 50MP) and whenever two MFDBs used the same sensor, their perfomance wasn't to different, so I don't think the S2 will be too different from any other cropped to (37,5MP) 6,0µm-Kodak-based MFDBs regarding RAW-IQ.

Some people here already have the H3DII-50 (and had the 39MP-version), or am I wrong?
Too bad, i have only HassyII 39mp, got it about 1 month ago and so, still new in MF world.
 

dougpeterson

Workshop Member
Thanks for your extensive post, Doug!

I understand that various aspects besides the sensor itself influence the RAW-IQ. But when these aspects are handled carefully (thermal management, converters...) it all comes back to the sensor.

I suspect that all these aspects are handled similary between the mentioned Hasselblad/Imacon-models (39 vs. 50MP) and whenever two MFDBs used the same sensor, their perfomance wasn't to different, so I don't think the S2 will be too different from any other cropped to (37,5MP) 6,0µm-Kodak-based MFDBs regarding RAW-IQ.

Some people here already have the H3DII-50 (and had the 39MP-version), or am I wrong?
Honestly I'm not sure.

On the one side: As an example of how important those "other" things are it's taken Hasselblad a long time to (just this week) deliver ISO 800 on their 39 megapixel model which uses the same chip as the P45+ which had ISO800 from the start. Same thing with long exposure where the P45+ can go to one hour (or longer in cold weather) using the same sensor that Hasselblad can only do 30 seconds on. I'm not picking on Hassy here; they just happen to be the only ones to use the same sensors as us.

On the other hand: when it comes to normal shutter speeds at base ISO with near proper exposure (which is most of captured images) performance the difference in IQ is negligible. So you're probably right that the RAW-IQ of the S2 is likely to be stellar and on par with other 6 micron sensors.

Assuming Capture One supports the file-type (seems likely, but nothing is for sure until it's already happened) then you can be sure to squeeze the most out of that raw.

I want to SEE it though rather than speculate.

Doug Peterson, Head of Technical Services
Capture Integration, Phase One & Canon Dealer | Personal Portfolio
 

georgl

New member
"black-calibration off of the unexposed chip and negating much of the noise of the chip"

You've mentioned that, I think that's one of the tricks (besides thermal management) - but only used on long exposures, right?

You've also mentioned 16bit-converters, I've looked on the Analog-Devices-page and they only offer converters for full-frame CCDs with up to 14bit (sufficient for today's dynamic range?) and as far as I know, the S2 also uses 14bit-converters and then generates 16bit-RAW-files - isn't that the case with all "16bit"-MFDBs?

@ Professional
So now you have a reason to buy the 50MP-version ;-)
 

Professional

Active member
"black-calibration off of the unexposed chip and negating much of the noise of the chip"

You've mentioned that, I think that's one of the tricks (besides thermal management) - but only used on long exposures, right?

You've also mentioned 16bit-converters, I've looked on the Analog-Devices-page and they only offer converters for full-frame CCDs with up to 14bit (sufficient for today's dynamic range?) and as far as I know, the S2 also uses 14bit-converters and then generates 16bit-RAW-files - isn't that the case with all "16bit"-MFDBs?

@ Professional
So now you have a reason to buy the 50MP-version ;-)
If i have enough money i will buy :grin:
 

dougpeterson

Workshop Member
You've[...] mentioned 16bit-converters, I've looked on the Analog-Devices-page and they only offer converters for full-frame CCDs with up to 14bit (sufficient for today's dynamic range?) and as far as I know, the S2 also uses 14bit-converters and then generates 16bit-RAW-files - isn't that the case with all "16bit"-MFDBs?
I am not privvy to the manufacturing origin of the 16-bit AD converter but it is in fact a true 16-bit AD converter producing a 16-bit raw file*. This is the case of all major digital backs except for the Mamiya ZD (which uses a dSLR class 14-bit AD converter).

I was also not aware that the S2 would use a 14-bit AD converter. I've been waiting a bit on the sidelines to see if the S2 will come out within a reasonable time of it's estimates and so haven't put as much time in as I probably should to know every spec.

Then again I don't really care (even as a huge nerd) HOW the quality is achieved so long as it is. All I care about in this case is that if I have an image with important detail deep in the shadows that I be able to pull as much clean, color accurate, and artifact-free detail from those shadows. It's a real simple test (because the absolute difference allows a lot of slop-room) to shoot your P+S, your dSLR, and your digital back and start pulling detail from the shadows. It's one of the most obvious differences between mediocre, good, and great digital.

So when someone can put a final-production S2 in my hands I guess we will see just how well it can pull and push shadow/highlight detail around.

Doug Peterson, Head of Technical Services
Capture Integration, Phase One & Canon Dealer | Personal Portfolio
 
I can't speak for other manufactuers, but the AD converter in an H3DII is a 16Bit converter.

In the case of the scanners they use three AD converters, one for each channel and again at 16bit.

David
 

georgl

New member
Do you know the manufacturer? I only know that analog-devices is by far the market-leader in this segment and I'm nearly 100% sure that they not only delivered the DSPs but also the converters for the Leica DMR which was always claimed as "16bit"-system. The Leica DMR was developed and manufactured by Imacon and is basically a "predecessor" of today's Hasselblad-backs.
There's a very popular high-end scanner in the film-industry which uses a second exposure to deliver real 16bit files, they claimed that this is the only way to achieve the necessary DR and they also use 14bit-converters...

I'm not sure if this is important at all, but there was a lot of discussion about these bits in the past and there seems confusion and misunderstanding about it...
 
I am afraid if I did know the manufacturer, I would have to keep it to myself. ;-)

Also the Leica DMR is certainly not to be seen as a predecessor to the H3DII. The hardware between both systems is completely different. Unless you know something I don't?

Best,


David
 

dougpeterson

Workshop Member
The Head of Technical Services at P1 confirmed for me. I too cannot comment on the component provider. However, it is really, truly, absolutely a 16 bit AD converter which is inside every Phase One single shot back (H5, H10, H20, H25, P20, P21, P25, P30, P45, P65+ and the other plus models). A long time ago early versions of the scan backs were 12 bit or 14 bit converters. But even the PowerPhase FX, the last scan back we made (shameless self-promotion: we have a great deal on one here in used inventory), was a 16bit AD converter.

When I ask about accuracy what I am told is that in an ideal mathematical sense you don't need a 16bit AD converter to get the dynamic range of a high-end back but that having more accurate information of the noise floor helps with the black-calibration and that having only a 14bit AD converter would have a meaningful detrimental impact on the image quality. To be frank, that part is over-my-head, but I trust them when they say it's a needed component.

Doug Peterson, Head of Technical Services
Capture Integration, Phase One & Canon Dealer | Personal Portfolio
 

tashley

Subscriber Member
As anyone who has ever tried to do direct head-to-head DR analysis knows it can be much more difficult then setting up a shot, shooting the first back, swapping backs, and shooting again. Variation in color spaces, processing choices, and different true native ISOs can all impact the head to head example. When you're looking for what you already know will be relatively minor differences (0.5 stop of DR at the most either direction) these otherwise trivial differences can really obscure your results.

Jack and Guy did some tests in Moab with the P65+ I brought and Guy's P25+ and Dave, our owner, did some tests in ATL. As best as I can tell the DR is a bit higher than either the P25+ or the P45+. Given that the size of the pixel has decreased I think this is an achievement in of itself.

There are actually a lot of fun to learn about (if you're a big nerd) factors in the architecture of a digital back which result in the system's DR. The quoted pixel size on tech sites is almost always the largest possible way of measuring each pixel. The part of each pixel which is actually light sensitive was, especially in the 9 micron sensors, much smaller. So when Kodak went from the P25+ to the P45+ the size of the light-sensitive area of each pixel barely changed, while the wasted (non-sensitive) area within each pixel (the electronics package and border of each pixel) was drastically reduced. Likewise with the new Dalsa 6 micron design the wasted space has been effectively eliminated so the loss of light sensitive is a little less than the 6.8 --> 6.0 micron change would indicate. Then after the sensor has pulled in all the photons it can the resulting electrical charge has to be read off of each pixel (generally down a row like "pass the bucket") and to a AD convertor where the charge is translated to a numerical value. The accuracy of this conversion is governed not only by the bit depth (e.g. 12, 14, 16) but also the quality of AD convertor (i.e. not all 16 bit convertors perform identically well). Depending on the architecture there is some (usually patented way of reading a black-calibration off of the unexposed chip and negating much of the noise of the chip). Once the signal is converted to digital you are left with a very unusable array of bayer-patterned pixels. It's then up to the math-gods of the raw conversion software to make a pretty picture from an otherwise random-looking group of pixels. The math here is so complicated that there are literally only a handful of individuals in the entire world who could truly be called experts on it. Finally you have a picture you can print or compare (and I've surely omitted several steps).

Don't know where this ramble came from now that I've arrived here. Other than to say a LOT of work goes into trying to squeeze out the last bit of quality from these sensors. And while the theoretical ceiling of DR may decline with the declining size of the pixel it is often the case that the actual performance stays the same or increases as the other mitigating factors that prevent you from reaching that ceiling are alleviated.

Sometimes I wonder if anybody but Bob makes it to the end of my poorly structured and inadequately explained technical rants. Hi Bob!

Doug Peterson, Head of Technical Services
Capture Integration, Phase One & Canon Dealer | Personal Portfolio

I got it, I think (?) and very interesting it was too... now if I could just work out how those dudes at DXO do their analyses, and what the results mean, I'd be in pixel heaven!
 
I got it, I think (?) and very interesting it was too... now if I could just work out how those dudes at DXO do their analyses, and what the results mean, I'd be in pixel heaven!
That is exactly the issue Tim and something we have failed to figure out as well.

I don't see how they can do an 'honest' test.

David
 
Top