Hi everyone,

After a weekend of shooting nightscapes with a canon 5D and nikon D80, I am very happy with the canon's cmos sensor. The cmos is much better with regards to noise at long exposures as well as high iso.

In medium format there are no cmos sensors (to my knowlege). Why is this? Is there a certain technological barrier? Certainly a camera like the H3D31 would benefit from having a cmos sensor.

regards

Mike

There isn't a CMOS sensor out there with low ISO image quality which can compete with the CCD-based MFDBs. So CMOS makes no sense for MF.

Mike:

I believe that CANON, who has their own fab plants makes their own sensor chips, and further that no one else produces CMOS sensors, at least in sizes that would be useful in DSLRs. Should CANON manufacture a MF DB, then their head start with CMOS technology might be an competitive advantage.

Jerry

I believe Canon chose CMOS because it is cheaper to produce, not for reasons of quality.

that's right, much cheaper to produce.

Thierry

Canon's worst current performer at high ISO (400D), still has less than half the read noise at ISO 1600 than the best Sony CCD. Canon's best, the 1Dmk3, literally has only about 5.5% the read noise of the worst CCD (worst in read noise, that is; not in general), in the D2X.

At the low ISO end, Canon also has the lead, by 1/2 stop, typically, at ISO 100.

Now, CMOS is supposed to mean a higher quantum efficiency, but in practice, the Sony DSLR CCD sensors to date have not had very high QE. The P&S CCD sensors have higher QE and and lower read noise than the DSLR sensors. The only reason they give inferior images is because of sensor size.

For those who like to read up on the subject, here is an interesting article (PDF) from Dalsa:

http://www.dalsa.com/shared/content/pdfs/C...willer_2005.pdf

A somewhat older, but also very interesting scientific article on CCD vs CMOS imagers:

http://www.ifp.uni-stuttgart.de/publicatio...owo01/blanc.pdf

While Canon was the first ot embrace cmos in mass in dslr's, Sony also produces cmos sensors for dslr's.

And I think that's the single biggest reason you don't see a CMOS DB. It's because Sony and Canon aren't producing for the larger format. Dalsa and Kodak have invested considerable amount in CCD--I imagine that it would be like starting from scratch to R&D CMOS chips--thus being cost prohibitive and (very) unlikely to yield DSLR/ compact profits in the long run.

Dalsa produces both CCD and CMOS chips. But if you read the articles I linked to, you would understand why CCD's are still preferred for MF backs.

What I've always wondered is how can the CMOS chips produce lower noise than the CCD but still have less total dynamic range? I thought DR was a function of the signal to noise ratio? Seems like even canon's latest chips for the 1D III bodies can't match even older MFDB chips in DR.

There were two Leaf backs that are no longer in production that used a CMOS sensor: The Leaf C-Most and the Leaf Valeo 6. They were actually both using the first "full frame" (24X36) chip, at 6MP, that was rated to 125/250 iso.

For the time (2000) and for the price ($8,000 IIRC) they offered good value. However the cost or R&D (the chip was a joined venture between Leaf and two other companies) to develop it further into a larger sensor, was higher than using the higher MP CCDs that came shortly after from Kodak and Dalsa.

Oddly enough, I tested a Valeo 6 on a Contax last week and tethered to a fast Xeon Mac, it gave us a long burst of capture at 2 frames/ sec....

Sorry for the hijacking...please continue:-)

Yair

Both CMOS and CCD can produce great results. And it's not just Canon who make their own CMOS - we make our own high speed CMOS at RED Digital Cinema, but it's designed for movie use, so the size is about that of a 1.6 crop sensor from the 35mm stills world though.

Basically, dynamic range = signal to noise ratio. Lower noise is equal to better dynamic range. I don't think there's any real reason why CMOS can't be used for medium format. Because of the on chip a to d, it might even help get those large megapixel counts off the chip faster and more easiliy. The "Mysterium" CMOS chip we use can generate about 1GB of data per second (that's the full 12mp sensor running at 60 frames per second), so CMOS really is an advantage for us.

Graeme

It sounds to me like a price issue which Yaya confirmed. Which was my initial thought. The R&D needed to develop a competing CMOS sensor isn't going to bring in the money to cover both costs and make a profit.

I liken it to the LBCAST sensor by Nikon. They attempted to develop and fabricate their own sensor and had they continued to pour money into developing it further it would have flourshed. But in the mean time it would have put them further behind Canon in image quality and been leaking money in process. They weighed the costs and benefits and decided it was better if they had Sony implement their designs instead of fabricating the chips themselves.

Yhe imagery you have posted is beautiful. I'm impatiently waiting for a more miniaturised model of your camera.

Edmund

That might depend on your standard of DR.

One camera can have more range between 1:1 S/N and saturation that another, while the latter has a greater range between 3:1 S/N and saturation, if the latter has more read noise but less shot noise due to a higher maximum photon count.

Another possibility is that the testing methods are flawed.

Most likely you have more depth to the info you post. But surely the real ISO definition of dynamic range is not so much use to working photographers nor much of the science behind it. Same way pixel count doesn't mean much - we all know you can't compare the digi cam pixels to high end camera pixels. I'm saying that while quite a few dslr's test out at nearly 12 stops DR same as what's quoted for digital backs, there is not much the same when you work with the files. I also agree with you that the testing may be flawed. I mean when you average a region of say 100,000 pixels in calculating DR, you could have a good technical value however all 5 of the splotches in that region are going to show up in an image if the shadows are pushed up even though they didn't change the average value a whole lot.

I read the articles linked and it looks like there are three reasons CMOS can't get the DR of CCD....the circuits in CMOS take up well space that could be used to capture photons (thereby increasing SNR), as the cicuit complexity goes up along with density of the sensor, the coupling noise goes up and the supply voltages go down which limits the dynamic range on CMOS.

Of course the Dalsa article was written in 2005 so I would guess there may have been advances in both CCD and CMOS tech since then. 2 years is like a whole generation in silicon time.

And in CCD, there's the trickyness of transporting that analogue voltage off the chip to your external A-to-D without adding any noise to it.

More often than not, it's not the chip itself, or the design be it CMOS or CCD, but how it's implemented and engineered that is the deciding factor.

Graeme

Hehe, I took some images from the empire state with a 1Ds, every image striped, the cellular antenna can be sen clearly in the pictures

Edmund

Long. long ago, Kodak made the CCD for their DCS 600 and 700 series DSLR and they was on top of everyone, Canon came along and did not want to paid for the copywrite of using the CCD so they develop the CMOS. Everyone said it won't work, but I guess Canon made it work and I wonder where Kodak is now, with their DSLR. The last DSLR Kodak made was the SLRn and SLRc and it was made with the CMOS but it was crap.

Let me try to point the discussion in another direction:

As I understand from some posts higher dynamic range means less noise.

In 35mm(!) the cmos sensor works better than a ccd (IMO) at exposures up to 30sec and longer.

DMF has higher dynamic range, so technically (if higher dynamic range means less noise) it should be possible to get more quality from a medium format DB when doing nightshots even though its a CCD sensor.

With a canon and 24mm L lens I get the best results at f5.6 and f8 @100iso at a maximum shutterspeed of 30 seconds.

Anyone ever do the same or longer exposures with a DB?

regards

Mike

The was no copyright on CCD's; many companies made them. Canon saw other advantages to CMOS, as Sony and Panasonic [NMOS] now also do, and also Kodak and Dalsa for some of their smaller format sensors.

And Kodak does still produce CCD sensors for at least one "DSLR": the Hasselblad H3D.

My hunch is that the advantages of changing to CMOS/NMOS pay off most in higher volume and or lower cost situations.

Phaseone recent backs seem to perform well with long exposures.

Regards,
Bernard

Indeed, but the +line is made for long exposures. I'm looking to find out how other digital backs without this function are holding up.

I think most DB makers list 30 seconds as the longest possible exposure (without using bulb of course)

Huh????

Even with bulb none of my backs got further than around 30 seconds. My Leafs only came to about 25 seconds before I got an error (without image).

The noise was really ugly. The best results were with exposures not longer than about 20 seconds.

I have not yet tested my Hasselblad backs (yet).

Why don't people just use something like a Canon or Nikon for long exposures? Much better equipped to do so.

Hmm too bad. I was looking into medium format to be able to make larger prints.

As the saying goes; horses for courses....untill digital back makers get it together, I'll be shooting digital 35mm then.

Or film

Mike

PhaseOne + backs are the clear leader as far as long exposures go.
I have gone 4 minutes with very acceptable results.

As mentioned above Dalsa chipped backs seem to run into trouble once exposures
pass 15-20 seconds.

Leaf has made very visible improvements with regards to long exposure processing with
the introduction of v11 and previously unuseable 30 second exposures reprocessed in
v11 were significantly improved.

Mamiya seems to hit the wall around 10-15 seconds based on secondhand reporting by
users on this site.

With the release of their new processing software,Phocus,later this year Hasselblad
is saying that exposures will increase to 60 seconds
Currently 30 second exposures using an H3D39 are very clean.

You referenced 30 seconds @ f8 as being a requirement for you.
From what I have seen,firsthand, that need is currently met by at least 3 systems
and I would suspect a fourth,Sinar,would not disappoint.

Mark