M9: CCD or CMOS

[brianv]

Advertisement (gone after registration)

I also have an intuitive preference of the CCDs over Cmos. However, so far, the best prints I ever got were actually from a Cmos sensor,that is, the one in the Kodak slr/C.

 

Oddly enough- the sensor used in the SLR/C and slr/N were not made by Kodak. I forget the company, it was acquired by someone a few years back. I downloaded the datasheet several years ago.

[Advertisement]

The Leica Q3 is available for pre-order at B&H Photo Video and Adorama

[adan]

FillFactory made the 24 x 36 CMOS sensors for the Kodak SLRs.

 

Now owned by Cypress Semiconductor: Cypress Completes Acquisition Of IMEC Spinoff FillFactory NV

[mjh]

Okay, sounds like you know a lot more of the engineering then I ever will, so then how does the Nikon D3 CMOS sensor perform do much better at high ISOs then the CCD cells in former Nikon cameras, and the M8 series?

Does it really? The sensor in the D3 and D700 has just 12 million pixels so the pixel pitch is a generous 8.45 µm. That translates to 2.4 times the pixel area of the D2X, for example, or 1.7 times the pixel area of the EOS-1Ds Mark III, or 1.5 times the pixel area of the M8/M8.2. Everything else being equal, this advantage is good for the same low noise at 2.4, 1.7, or 1.5 times the ISO value. Most everything else you see is the result of the noise reduction, be it in-camera or by the raw converter. Nikon has been almost brutally keeping chroma noise in check so it is nearly constant across the whole ISO range. This requires the noise reduction to get more aggressive with increasing ISO numbers. Canon used to apply about the same noise reduction at all ISO settings so the chroma noise did increase with the ISO setting, but recently with the EOS 5D Mark II they chose to do it the Nikon-way, maybe because the unnaturally noise-free images turned out to be popular. Leica’s noise reduction in the M8 clearly isn’t the most sophisticated (and neither is Olympus’ or Sony’s).

 

In my experience, the noise in the raw image data is nearly linear with the pixel area these days, regardless of the technology or vendor of the sensor used. CMOS sensors once used to be associated with cheap webcams and such, which is why nine years ago, Canon’s introduction of the D30 had everyone baffled – why was Canon using a sensor technology that everyone knew was only good for toys? CMOS sensors have individual amplifier circuits for each pixel, and as each amplifier is slightly different, there is a lot of fixed-pattern noise in the signal. Fixed-pattern noise can be effectively suppressed, though, and Canon (and, more recently, other vendors as well) managed to get a reasonably low-noise image from a CMOS sensor, making it a viable alternative to CCD sensors. If you compare fill-factors, though, full-frame transfer CCDs still have a slight edge.

 

The main factor contributing to noise isn’t the dark current that partly depends on the chip’s temperature, but shot noise which depends on the number of photons detected by each sensor pixel, completely independent from the technology used.

[sandymc]

Another major source of sensor noise is signal contamination. Historically, that was one of the major reasons (together with fill factor) as to why CMOS sensors were noisy; the digital circuits you need on chip inject noise into the image as the digital signals switch. Canon's big break though was to design a CMOS sensor that was very close to a CCD sensor in terms of injected noise - partially a better CMOS process, partially a better design. Till then, nobody believed you could get a CMOS sensor to near CCD performance levels. Now most of the new generation sensors are CMOS, but at the time it was a big competitive advantage for Canon.

 

Sandy

[jaapv]

You are number four or five to tell that story. There is also a shop in Jakarta taking preorders, but none of them can actually confirm anything.

Well, not just Jakarta

[Mauribix]

Well, not just Jakarta

 

Ohhh yesss!

[adan]

Advertisement (gone after registration)

Canon used to apply about the same noise reduction at all ISO settings so the chroma noise did increase with the ISO setting, but recently with the EOS 5D Mark II they chose to do it the Nikon-way, maybe because the unnaturally noise-free images turned out to be popular.

 

Agreed - I noticed in my brief stint with a 5DII that at ISO 100 it was noisier than the M8 at 160* - but that ISO 800 looked almost the same as 100. A flattening of the noise curve through processing.

 

*In 100% view - in a print the smaller pixels meant the noise was not significantly different, visually.

[lars_bergquist]

The 'noiselessness' of some Nikon high-ISO images is of course largely a product of very aggressive noise suppression in-camera. Leica could have chosen to raise the level of noise suppression to Nikon levels, but they did not do so. I think that choice was deliberate, a consequence of the same thinking that led them to abolish the over-sensor soft filter ("AA filter"). They wanted maximum detail. Those who prefer noislessness to definition can get what they want in PP.

 

No noise-reduction software can know the difference between noise and fine image detail. So it must suppress one with the other. So Canikon don't get their pixels' worth in definition. But because of that soft filter, the detail may not be there in any case.

 

The old man from the Age of Panatomic-X

[lars_bergquist]

Well, not just Jakarta

I have pre-ordered mine. My dealer told me that he has a foot-long list of pre-orders for not yet existing cameras, including Canon ones.

 

The old man from the Age of Roll Film

[nugat]

From 1600ISO up the better way is to shoot raw with all noise supression turned off. The detail remains and noise is better dealt with software in post.

[Stefflaurent]

I have pre-ordered mine too...

[adli]

My dealer told me that he has a foot-long list of pre-orders for not yet existing cameras, including Canon ones.

 

 

I guess that includes a discontinued pre-order list for the R10

[lars_bergquist]

I guess that includes a discontinued pre-order list for the R10

No.

 

The old man from the Age of Adox

[t024484]

The sensor in the D3 and D700 has just 12 million pixels so the pixel pitch is a generous 8.45 µm. That translates to 1.5 times the pixel area of the M8/M8.2. In my experience, the noise in the raw image data is nearly linear with the pixel area these days, regardless of the technology or vendor of the sensor used.

For sure, a larger pixel has a better S/N when looking at shot noise, because the photon count goes up linear with area, and the shot noise with the square root.

So S/N doen not increase linear with pixel area but linear with size, and is indeed independant of technology.

So a D3 pixel produces 1.24 less Shot Noise (8.45/6.8) as a M8 pixel.

 

The main factor contributing to noise isn’t the dark current that partly depends on the chip’s temperature, but shot noise which depends on the number of photons detected by each sensor pixel, completely independent from the technology used.

Shot noise is directly proportional to the square root of the Signal, but there is more to noise than just shot noise.

PRNU, or pixel response non uniformity, is noise that is linear proportinal to the Sensor Signal.

 

At low ISO PRNU is the the dominant noise source at high ilumination.

Theoretically PRNU could be removed in the RAW signal, but I do not know if Leica or anybody else takes the trouble of doing this,

In the middle area of illumination it is the Shot Noise that dominates, and at low illumination the dominant noise source is the Read Noise from the Sensor added to the ADC noise.

 

At high ISO the ADC noise becomes insignificant, and the Read Noise remains as the dominant Noise Source at low Illumination.

PRNU also becomes insignificant and only Shot noise remains as the dominant Noise Source at middle and high illumination.

 

So if at higher ISO, noise in the dark parts is what annoys most, it is likely to be Read Noise, and here it is where the Leica M8 could be improved.

If on the other hand at all illumination levels noise is getting visible, it has to be accounted to shot noise, and can be reduced in the digital domain at the cost of having less resolution.

 

So when wanting to achieve as less noise as possible at higher ISO values, without having the digital manipulation, two factors are playing an important role which are: 1) Have large pixels, giving large sensors for a given resolution, and 2) Have read noise as low as possible by using very low noise amplifiers, plus a very sophisticated analog/digital layout all the way to the ADC, isolating injection of digital noise.

 

hans

[mjh]

So S/N doen not increase linear with pixel area but linear with size, and is indeed independant of technology.

Admittedly I didn’t make it clear how I was quantifying noise here. What I had meant – and should have expressed more clearly – was that doubling the pixel area gives you the same amount of noise at twice the ISO number.

 

Shot noise is directly proportional to the square root of the Signal, but there is more to noise than just shot noise.

Sure, but in my experience it is the dominant source of noise here. A current generation Kodak CCD captures like 40,000 photons per pixel at its native sensitivity so the signal to shot noise ratio is 200:1 for the highlights. Increase the ISO from 100 to 1600 and it drops to 50:1. Shadows are much worse, obviously; at the lower end of a hypothetical dynamic range of 12 EV we’d get 3:1 for the native sensitivity and almost pure noise at ISO 1600. Apart from increasing pixel size or applying digital noise reduction, there is nothing one can do about this.

 

PRNU can be an issue, but for those (mostly CMOS) sensors where it is, it usually gets removed from even the raw data.

[markowich]

The 'noiselessness' of some Nikon high-ISO images is of course largely a product of very aggressive noise suppression in-camera. Leica could have chosen to raise the level of noise suppression to Nikon levels, but they did not do so. I think that choice was deliberate, a consequence of the same thinking that led them to abolish the over-sensor soft filter ("AA filter"). They wanted maximum detail. Those who prefer noislessness to definition can get what they want in PP.

 

No noise-reduction software can know the difference between noise and fine image detail. So it must suppress one with the other. So Canikon don't get their pixels' worth in definition. But because of that soft filter, the detail may not be there in any case.

 

The old man from the Age of Panatomic-X

 

i'd bet quite a lot of money that neither leica nor phase has -at the moment-the in-camera noise reduction technology of nikon and canon.

peter

[lars_bergquist]

i'd bet quite a lot of money that neither leica nor phase has -at the moment-the in-camera noise reduction technology of nikon and canon.

peter

All right, enlighten me, someone: Is noise reduction an arcane secret known only to Canon and Nikon? Like the mythical "secret of the atom bomb"? Or is it partly a question of available processing capacity?

 

Nothing of this changes my argument, however. No software can discriminate between noise and fine detail on the pixel level. Both have to go, or none.

 

The same old man from the Age B.C. (Before Computers)

[carstenw]

With results which are already soft from AA filters, Canon and Nikon can accept more smudging from noise reduction than other companies are willing to... I don't find the Canon/Nikon high ISO results attractive at all.

[Advertisement]

Hallo,

Look here: Leica M

[jonoslack]

With results which are already soft from AA filters, Canon and Nikon can accept more smudging from noise reduction than other companies are willing to... I don't find the Canon/Nikon high ISO results attractive at all.

 

Hi Carsten

Nor do I, and I also feel that all that pre-processing has a bad effect on colour in unusual circumstances.

 

I don't often do anything about noise with M8 shots . . . . I'd rather the camera didn't make decisions about it either.

[rosuna]

I don't know exactly how Canon or Nikon are getting those results, but even in high resolution cameras (D3X, 5D Mark II, 1Ds Mark III), the files at high ISO values are just incredible. It may be some kind of mix of in sensor electronics, motherboard and processor design and software, but it works. Detail, color and tonal variability is preserved across a wide range of ISO values. I cannot get the same with the M8 sensor, even applying speciallized software for noise supression. Color and detail is severely affected.

 

As a pure photon collector the M8's CCD may be better than a CMOS (although the difference may be smaller than many think), but the whole M8 architecture introduces at subsequent stages a lot of noise to the signal finally processed. I am interested in the S2 as a new platform for Leica products, and I want to see how a CCD performs with improved electronics (motherboard, processor, A/D converter, sensor itself).