IQ250 14 stops DR

[ErikKaffehr]

Hi,

Just fine! Didn't know Hasselblad was also using Dalsa.

About Leica, we will see. A while ago I speculated they would use their own design, but now I guess they will go for the Sony sensor.

By the way, Leica is playing smart. They make a body that can take Contax and Hasselblad lenses, makes for a large market.

Best regards
Erik

Erik, two slight modifications to your list, an some comets in italics:

[synn]

Update: The IQ 250 profile makes the awful skintones from a Canon 5D Mk III look better too.
The Nikon D800 files look nicer after this, but the Canon ones are definitely losing their waxy, global look after this.

[Theodoros]

Update: The IQ 250 profile makes the awful skintones from a Canon 5D Mk III look better too.
The Nikon D800 files look nicer after this, but the Canon ones are definitely losing their waxy, global look after this.

it's because they've "worked" (due to the IQ250) on the over saturation that Cmos sensors have…. The blue cast that is developing with your D-800 when you "dig" in the shadows, is most probably because there is excessive colour information intentionally "packed" into the shadows to look like there is more DR than there should be… That's why I say that CCDs have more usable DR.

[Bart_van_der_Wolf]

it's because they've "worked" (due to the IQ250) on the over saturation that Cmos sensors have….

Hi Theodoros,

CMOS sensors do not oversaturate, they just capture photons. Could you explain what you are referring to (preferably with some authoritative references, if any), maybe something got lost in translation?

The technology differences between CCD and CMOS have nothing to do with color rendering, only with storage of photon affected electron charge, and how it's read out. Any color rendering related differences are only due to Bayer CFA design, Raw conversion, and profiling.

Cheers,
Bart

[synn]

Hi Theodoros,

CMOS sensors do not oversaturate, they just capture photons. Could you explain what you are referring to (preferably with some authoritative references, if any), maybe something got lost in translation?

The technology differences between CCD and CMOS have nothing to do with color rendering, only with storage of photon affected electron charge, and how it's read out. Any color rendering related differences are only due to Bayer CFA design, Raw conversion, and profiling.

Cheers,
Bart

 

Are we doing this AGAIN...?

[esox]

Hi Theodoros,

CMOS sensors do not oversaturate, they just capture photons. Could you explain what you are referring to (preferably with some authoritative references, if any), maybe something got lost in translation?

The technology differences between CCD and CMOS have nothing to do with color rendering, only with storage of photon affected electron charge, and how it's read out. Any color rendering related differences are only due to Bayer CFA design, Raw conversion, and profiling.

Cheers,
Bart

Bit resolution as no importance ? No matter if it is 8 bits, 10 bits, 12 bits 14 bits or 16 bits ? Surprising.

[torger]

Bit resolution as no importance ? No matter if it is 8 bits, 10 bits, 12 bits 14 bits or 16 bits ? Surprising.

Today sampling is either 14 bit or 16 bit. Noise floor has not yet passed 14 bit. MFDBs had 16 bit sampling when noise floor was still at 12 bit, and marketing claimed it was an advantage, ignoring read noise levels. Most (not all) has left this fraudulent marketing though and now just talk about the dynamic range, which is based on actual measurements of signal to noise ratio.

Still the quality and shape of noise is one of those hard-to-measure aspects that can affect subjective image quality. Random smooth noise is good, blotchy patterned noise is bad, but in a signal-to-noise ratio measurements these qualities will not be measured. A camera with higher engineering DR can have less useful DR due to this.

I haven't seen any recent comparisons of noise quality out of a recent CCD back and a recent sony CMOS sensor so I don't know if there are any differences there or not. It would be interesting to see. I'm not excluding the possibility that noise quality is visibly different in a typical CCD architecture compared to a typical CMOS architecture.

But I still find it much more likely that these days when comparing a Dalsa CCD with a Sony Exmor CMOS it's all about the CFA and profiling. In other words I would not dare to claim that CCD has some special quality just because it's CCD, I believe that's like the "16 bit myth" it's a work of MF marketing to differentiate from smaller formats, and CCD vs CMOS was simpler to talk about than differences in CFA design and profiling. That language will probably change now when we get the first CMOS MFDBs.

[ErikKaffehr]

Hi,

This is a comparison I made a couple of months ago between my P45+ and my SLT 99. The P45+ is an old design, but I still payed 10000$US for it this summer, 4 times the cost of the SLT 99.

These images were compared with sharpening and with noise reduction set to zero, LR 5.3 was used. I did test with Capture One, too, and the result was not better.
P45+:


SLT 99:


Raw images:
http://echophoto.dnsalias.net/ekr/index.php/photoarticles/20131117-CF044323_LR5.iiq

http://echophoto.dnsalias.net/ekr/index.php/photoarticles/20131117-_DSC3262_LR5.dng

My article is here: http://echophoto.dnsalias.net/ekr/index.php/photoarticles/80-my-mfd-journey-summing-up?start=1

Chris Barret has posted a couple of images from his IQ 260 compared with A7r. Regardless of LR5 or Capture One 7.1.3 the images from the Alpha 7r were less noisy, but those shots were made at 100 ISO.


Best regards
Erik

Today sampling is either 14 bit or 16 bit. Noise floor has not yet passed 14 bit. MFDBs had 16 bit sampling when noise floor was still at 12 bit, and marketing claimed it was an advantage, ignoring read noise levels. Most (not all) has left this fraudulent marketing though and now just talk about the dynamic range, which is based on actual measurements of signal to noise ratio.

Still the quality and shape of noise is one of those hard-to-measure aspects that can affect subjective image quality. Random smooth noise is good, blotchy patterned noise is bad, but in a signal-to-noise ratio measurements these qualities will not be measured. A camera with higher engineering DR can have less useful DR due to this.

I haven't seen any recent comparisons of noise quality out of a recent CCD back and a recent sony CMOS sensor so I don't know if there are any differences there or not. It would be interesting to see. I'm not excluding the possibility that noise quality is visibly different in a typical CCD architecture compared to a typical CMOS architecture.

But I still find it much more likely that these days when comparing a Dalsa CCD with a Sony Exmor CMOS it's all about the CFA and profiling. In other words I would not dare to claim that CCD has some special quality just because it's CCD, I believe that's like the "16 bit myth" it's a work of MF marketing to differentiate from smaller formats, and CCD vs CMOS was simpler to talk about than differences in CFA design and profiling. That language will probably change now when we get the first CMOS MFDBs.

[Paul2660]

Eric,

If you had shot the same series and bracketed the exposure on the P45+ allowing more light in, don't you feel the result would have been better.  Your P45+ shows the classic noise/shadow issues I had for 3 years shooting that back.  The A99 has more room and that to me is not a CCD vs CMOS issue, it's just a better overall process.  Sony has produced some very good chips in the last few years and the P45+ as you mention was first sold in 2007, and shipped in 1st half of 2008. 

When Phase One released the later firmware for the P45+ that allowed it to get to a full hour exposure, they did improve the quality of the images at base iso (at least I felt so), and iso 400 became possible to use.  I was curious if you were running the latest firmware on the P45+. 

I feel the 260 allows more reach into the shadows for sure I have seen Chris's comparisons and in my work, I will most times be able to push the shadows 1 to  1.75 of a stop before the noise becomes destructive, again at base iso of 50.    Not sure if I have ever seen the finer color gains or losses between the CCD and CMOS chips I have used, but the noise issues like you point out are there for sure with the older Phase backs unless the exposure was bracketed.

Paul C.

[ErikKaffehr]

Hi,

That image was from a series that was bracketed. Both original images were exposed to the right, with the P45+ having a bit more exposure. See raw histograms from "RawDigger" below.

This is no test shot, I happened to have two exposures from both cameras on the same shoot. These cameras I own and use, so I am familiar with them.

I went back to Chris Barrets tests. Here I pushed exposure 4 stops (like in my own test). I developed the IQ images using standard setting but without noise reduction in both LR 5.3 and Capture One.

I also downsized the images to A7r vertical pixels, so the size advantage of the IQ 260 is taken into account. See below.

Two small comments:

1) In the histogram you can see that the Sony histogram drops sharply at -7 EV, while the P45+ goes down to -12. I am pretty sure that the signal we see on the P45+ below -8EV is just noise.

2) The P45+ has some more exposure but it seems that the SLT 99 image is a bit brighter. One explanation could be more lens flare on the SLT, as a fairly complex zoom lens was used. That said, I could see some advantage to the Alpha 99 in all tests I did. Or it could be said that I am essentially confident to be able to pull shadows on the ALpha 99 but less so on the P45+.

Yeah, and just to say, I am quite happy with my P45 and I actually enjoy shooting with it! I am not going to get rid of it. But I did not expect miracles from it and I don't see any miracles. It actually works as I have expected.

Best regards
Erik

[Theodoros]

Hi Theodoros,

CMOS sensors do not oversaturate, they just capture photons. Could you explain what you are referring to (preferably with some authoritative references, if any), maybe something got lost in translation?

The technology differences between CCD and CMOS have nothing to do with color rendering, only with storage of photon affected electron charge, and how it's read out. Any color rendering related differences are only due to Bayer CFA design, Raw conversion, and profiling.

Cheers,
Bart

Hi Bart… it's just that I have that opinion (which I can't prove - none can), that A/D converters are "tweaked" so that they don't provide a linear output of the (linear) analog input of photons… I believe that the (raw) output is a "linearised S-curve" other wise measurable DR would have never improved to todays values… I believe that Cmos sensors of DSLRs amplify the LLs and restrict the HLs (possibly in the A/D conversion process that is build in to Cmos sensors) to look like they have extended DR than CCDs… My suspicion is that this the reason why with CCDs you have usable DR (on a realistic print) that is very close to the maximum of the recorded DR, while if one tries to print as close as possible to the full extension of the recorded DR on a Cmos sensor, he ends up with a dull (unrealistic) print and with false (cool) colour casts in the darks and (warm) in the HLs...

[tho_mas]

3 lengthy pages about the 14 stops DR of the IQ250... although there is not a single RAW file of the IQ250 available yet.
Something's really wrong in here....

[torger]

3 lengthy pages about the 14 stops DR of the IQ250... although there is not a single RAW file of the IQ250 available yet.

No thread stays on topic for more than one page or so anyway 

[BJL]

.. it's just that I have that opinion (which I can't prove - none can) that A/D converters are "tweaked" so that they don't provide a linear output of the (linear) analog input of photons… I believe that the (raw) output is a "linearised S-curve" other wise measurable DR would have never improved to todays values… I believe that … My suspicion is that  ...

In summary, you state a list of beliefs based on nothing and contradicted by well-known properties of ADCs, simply because you are unaware of (or in denial about) the well-know reason why new CMOS sensors have significantly better "engineering dynamic range" than CCDs. Which is that modern active pixel CMOS sensors have far lower dark noise levels than the best modern CCDs: about 2 electrons (RMS) vs about 10-12 for those CCDs.  You can read this in spec sheets at Truesense, Dalsa and so on.

This advantage in dark noise is in turn because modern active pixel CMOS sensors apply amplification to the signal very early (while transferring directly from the photosite to the edge of the sensor) and then (in column-parallel designs like Sony's) the sensor does analog to digital conversion immediately there with no further transportation of the charge signal, whereas CCD's cannot do any amplification or ADC until the signal has been transported off the sensor, which has to be done by moving it in thousands of times, each move a hop from one photosite to a neighbor.  Those hops introduce some "read-out" noise beyond that arising in the photosites.

[Theodoros]

….a list of beliefs based on nothing….

Nothing? ….I (and many others) find CCDs to have much more usable DR on prints than Cmos (as I stated on my reply to Bart)… But again, I only print about 8 square meters a day on year average… MO is simply based on (photographic) findings… which are the only ones that matter ...to photographers.

[tho_mas]

No thread stays on topic for more than one page or so anyway 

[BJL]

Nothing? ….I (and many others) find CCDs to have much more usable DR on prints than Cmos (as I stated on my reply to Bart)… But again, I only print about 8 square meters a day on year average… MO is simply based on (photographic) findings… which are the only ones that matter ...to photographers.

I was referring to your stated beliefs about ADC's in CMOS sensors being tweaked to producing non-linear responses,and declare that this is the only possible explanation of the improvement in total DR given by CMOS sensors.

I wil say it yet again: these "CCD vs CMOS" DR comparisons are mostly sensor size comparisons: they reveal the genuine advantages of a _larger_ sensor that can gather more light (count more photons total across the image) when fully exposed (base ISO speed, ETTR etc.), which gives a better ratio of signal to photon shot noise, and thus a better SNR in the relevant parts of most images, and so more "usable" dynamic range, even when compared to a smaller sensor with higher engineering dynamic range due to a far lower noise floor. Engineering DR is only of much relevance to photosites that get far, far less than full-well illumination, more than ten stops below saturation. Such low light levels are mostly relevant to (a) high ISO, low light photography, and (b) gear-heads who seek out extreme cases as reasons for wanting to pull details out of shadows way, way darker than are usually relevant.

Short version: photons shot noise is becoming progressively more important than dark noise or read noise, and so engineering DR is becoming less relevant as a measure of sensor performance, except in low light situations.

[Theodoros]

I was referring to your stated beliefs about ADC's in CMOS sensors being tweaked to producing non-linear responses,and declare that this is the only possible explanation of the improvement in total DR given by CMOS sensors.

I wil say it yet again: these "CCD vs CMOS" DR comparisons are mostly sensor size comparisons: they reveal the genuine advantages of a _larger_ sensor that can gather more light (count more photons total across the image) when fully exposed (base ISO speed, ETTR etc.), which gives a better ratio of signal to photon shot noise, and thus a better SNR in the relevant parts of most images, and so more "usable" dynamic range, even when compared to a smaller sensor with higher engineering dynamic range due to a far lower noise floor. Engineering DR is only of much relevance to photosites that get far, far less than full-well illumination, more than ten stops below saturation. Such low light levels are mostly relevant to (a) high ISO, low light photography, and (b) gear-heads who seek out extreme cases as reasons for wanting to pull details out of shadows way, way darker than are usually relevant.

Short version: photons shot noise is becoming progressively more important than dark noise or read noise, and so engineering DR is becoming less relevant as a measure of sensor performance, except in low light situations.

I was referring to results…   ...that don't coincide with scientific theories.  and thus, expressing a suspicion that my conspicuous mind finds to be an explanation…  …if I have your permission of course. 

[BJL]

I was referring to results ... that don't coincide with scientific theories.

The results you see conform perfectly with scientific theory, as I explained in my last post: larger sensors that can count more photons have a better ratio of signal to photon shot noise and thus better signal-to-noise ratio when given full exposure (base ISO speed), whereas total dynamic range only tells you about noise levels in the deepest shadow regions, about ten stops or more below saturation level, which is the only place that dark noise is significant compared to photon shot noise in modern sensors.

But apparently you prefer to ignore that explanation, just as others want to ignore differences in CFA design and other aspects of color handling, and instead attribute the differences seen to the difference between CMOS and CCD technology.

P. S. This is nicely explored by torger at http://www.ludd.ltu.se/~torger/photography/noise-test.html as discussed in another thread starting at
http://www.luminous-landscape.com/forum/index.php?topic=86989.msg708619#msg708619

In particular, I quote one of torger's conclusions:

The Aptus is very slightly better than the D7000 down to about 7 stops from saturation (max 1/3 stop better if even that), then read/dark noise becomes a factor and the D7000 wins with about 2/3 stops towards the noise floor.

The MF camera there is from 2005, so more recent MF models should perform a bit better, with the far bigger CCDs staying ahead of the smaller Sony EXMOR CMOS sensors for more than the top 7 stops, but the overall pattern is probably the same.

[Theodoros]

The results you see conform perfectly with scientific theory, as I explained in my last post: larger sensors that can count more photons have a better ratio of signal to photon shot noise and thus better signal-to-noise ratio when given full exposure (base ISO speed), whereas total dynamic range only tells you about noise levels in the deepest shadow regions, about ten stops or more below saturation level, which is the only place that dark noise is significant compared to photon shot noise in modern sensors.

But apparently you prefer to ignore that explanation, just as others want to ignore differences in CFA design and other aspects of color handling, and instead attribute the differences seen to the difference between CMOS and CCD technology.

P. S. This is nicely explored by torger at http://www.ludd.ltu.se/~torger/photography/noise-test.html as discussed in another thread starting at
http://www.luminous-landscape.com/forum/index.php?topic=86989.msg708619#msg708619

In particular, I quote one of torger's conclusions:The MF camera there is from 2005, so more recent MF models should perform a bit better, with the far bigger CCDs staying ahead of the smaller Sony EXMOR CMOS sensors for more than the top 7 stops, but the overall pattern is probably the same.

Have you ever tried an old MFDB with 6 or 11mp 24x36mm sensor? ….Obviously not!