OM PEN E-P2

[Wayne Fox]

Here you go- in a great explanation of the Digic processor, Canon does talk a bit about what is happening, and where.  A snappy illustration of bit depth, too.  Gonna have to bookmark this one.  
http://www.usa.canon.com/dlc/controller?ac...;articleID=2748

Considering that cmos sensors do the A/D conversion on the sensor at the pixel level, the ability of Digic4 to do anything will be quite dependent on chip design.  sounds like this goes hand in hand... you have to improve the sensor at which point you can improve the firmware to take advantage of the new sensor.

Of course, the EF-1(2) and the GF1 don't use a cmos sensor, or a CCD for that matter.  I don't know what "Live mos" brings to the table - it appears to be some hybrid mix of CCD and cmos technologies. It apparently improved the noise of the original 4/3rds sensors which were ccd based, but how that relates to chip design vs firmware ... maybe completely different than other cameras.

But I'm not engineer, and most of this getting over my pay grade.

I'm actually becoming a fan of the micro 4/3rds format, and I was wondering if the 4/3rds cameras lag behind in noise?  To me it seems they could do better, but then again my own test seems to show they are only about a stop behind the recent Canon 7D, and about a stop better than the recent G11/s90.  considering the  size of the sensors perhaps that's not bad.

I really like the Lumix, but the in camera IS of the Pen is a nice feature.

[teddillard]

Considering that cmos sensors do the A/D conversion on the sensor at the pixel level,

Sorry, where do you get that?  I may be completely off-base here, but the CMOS just gives you a readout of each pixel, not A/D at the pixel...  in any case, the processing is the real "secret sauce" here.  As I mentioned, one sensor design had noise filtering and some limited color processing before the A/D conversion- working the analog data.  There was a great deal of promise with that, but I believe it fell short in actual practice, and I haven't heard of anyone else doing it.

I totally agree with you on the format, Wayne, it's really exciting but has, so far, fallen short on image quality across the board.  And DarkPenguin, the firmware is what is in the processor- what you're calling the chip.  Upgrade the firmware, upgrade the processor...  so you could keep the same processor and change the processing by reprogramming the "chip".

[DarkPenguin]

And DarkPenguin, the firmware is what is in the processor- what you're calling the chip.  Upgrade the firmware, upgrade the processor...  so you could keep the same processor and change the processing by reprogramming the "chip".

You could if that's how they designed it.  I would be stunned if it was doing any significant image processing in firmware.

[teddillard]

You could if that's how they designed it.  I would be stunned if it was doing any significant image processing in firmware.

Uh, not sure where we're disconnecting, but take a look at that Canon link.  That's exactly where they're doing all the processing.  Every camera I've ever worked with does the image processing in the camera processor, the firmware.

[Wayne Fox]

Sorry, where do you get that?  I may be completely off-base here, but the CMOS just gives you a readout of each pixel, not A/D at the pixel...  in any case, the processing is the real "secret sauce" here.  As I mentioned, one sensor design had noise filtering and some limited color processing before the A/D conversion- working the analog data.  There was a great deal of promise with that, but I believe it fell short in actual practice, and I haven't heard of anyone else doing it.

I totally agree with you on the format, Wayne, it's really exciting but has, so far, fallen short on image quality across the board.  And DarkPenguin, the firmware is what is in the processor- what you're calling the chip.  Upgrade the firmware, upgrade the processor...  so you could keep the same processor and change the processing by reprogramming the "chip".

I thought I'd heard this before, so I did a little research.  Turns out CMOS does the electron to voltage conversion at the pixel level which I believe has been misinterpreted as the A/D conversion by some in various forums. This is one of the major difference between CMOS and CCD, and is one reason the actual light sensitive area of the CCD sensor is a greater % of the chip.  (Surprisingly enough even on a CCD that light sensitive area is only 30% of the surface of the chip).

However, from everything I can find, the CMOS sensor does do much of the image processing in the chip itself, including the A/D conversion.  The Camera firmware receives digital data from a CMOS sensor, not Analog data.  Even the article you linked indicates this to be true ..."In the case of the EOS 5D Mark II and 50D cameras, the raw digital data produced by the A/D converter is fed directly to the DIGIC 4 image processing circuit".  So the A/D conversion occurs before the camera firmware/processor sees it.

There is a couple of diagrams in this article by Dalsa which also shows the A/D convertor is part of the CMOS circuitry ... http://www.dalsa.com/public/corp/Photonics...S_Litwiller.pdf.

To me it appears CMOS IQ is very dependent on in chip circuitry.

One apparent advantage of LiveMOS sensors is they allow a CMOS type of sensor to obtain the same light sensitve surface as a CCD ... about 30% of each pixel.  According this this source, a CMOS sensor is more in the neighborhood of 10%  http://www.dcviews.com/press/Olympus-Panasonic-MOS.htm.  This also shows one of the purported advantages of Live MOS.

It's easy to see how even minor efforts to shrink the on chip electronics of CMOS sensors can lead to substantial gains in the light sensitive area of the sensor, yielding improved signal to noise ratios.

[teddillard]

I thought I'd heard this before, so I did a little research.  Turns out CMOS does the electron to voltage conversion at the pixel level which I believe has been misinterpreted as the A/D conversion by some in various forums. This is one of the major difference between CMOS and CCD, and is one reason the actual light sensitive area of the CCD sensor is a greater % of the chip.  (Surprisingly enough even on a CCD that light sensitive area is only 30% of the surface of the chip).

However, from everything I can find, the CMOS sensor does do much of the image processing in the chip itself, including the A/D conversion.  The Camera firmware receives digital data from a CMOS sensor, not Analog data.  Even the article you linked indicates this to be true ..."In the case of the EOS 5D Mark II and 50D cameras, the raw digital data produced by the A/D converter is fed directly to the DIGIC 4 image processing circuit".  So the A/D conversion occurs before the camera firmware/processor sees it.

There is a couple of diagrams in this article by Dalsa which also shows the A/D convertor is part of the CMOS circuitry ... http://www.dalsa.com/public/corp/Photonics...S_Litwiller.pdf.

To me it appears CMOS IQ is very dependent on in chip circuitry.

One apparent advantage of LiveMOS sensors is they allow a CMOS type of sensor to obtain the same light sensitve surface as a CCD ... about 30% of each pixel.  According this this source, a CMOS sensor is more in the neighborhood of 10%  http://www.dcviews.com/press/Olympus-Panasonic-MOS.htm.  This also shows one of the purported advantages of Live MOS.

It's easy to see how even minor efforts to shrink the on chip electronics of CMOS sensors can lead to substantial gains in the light sensitive area of the sensor, yielding improved signal to noise ratios.

Heading out the door, but thanks for that great info!  Yes, CMOS images are very dependent on the sensor circuitry.  They are inherently noisier, so anything that can keep the S/N ratio better is key.  The biggest contributor to noise, actually, is called "signal leakage" on a CMOS, and the best fix for that is to use a higher grade of silicon apparently.  This, ironically, raises the cost of the CMOS up to nearly the level of a CCD, from what I've been told.  

More later, got honey-doo stuff to take care of!  

[teddillard]

Absolutely fascinating bunch of information, Wayne...  thanks again.  There's a whole lot more to take in than I've had a chance to, but that .pdf from Dalsa is, in the very least, one of the best explanations of how the sensor works as anything I've read or been told.  

All that said and read, though.  Let me go back to the original statement, sensors is sensors...  I think what we're discussing here are generational differences in sensor technology, not a simple difference in brand or product.  There certainly is a huge difference in sensors between what we're seeing now, and what was there only a few years ago, whether you're talking CMOS or CCD, and I'd never mean to say that there's no significance between the quality of the files from one generation to another.  My only point is that in the comparison of two models of similar-priced cameras, or even, in most cases (and this one in particular, apparently) where you have a newer model of a camera released, the sensor model doesn't have much to do with the image quality.  The firmware upgrades, or processor (chip) updates are far more important.  

Sensor technology breakthroughs aside, sensors, themselves, are not analogous to film- they do not have a characteristic look.  That is a result of the processor.  They certainly have performance characteristics- noise, bit-depth (if the A/D is on the sensor), filtration capabilities, but those are what the processor is going to use to make that look.  

I feel like I'm mincing words, and I also feel like we're in agreement.  I only try to clarify this because I've read so many times how the expectation from a certain sensor is that it will perform in a certain way, and it just doesn't work like that.  ("Oh, that's a Sony sensor... we know that's going to do such and such...")

Bottom line- take two cameras that are similar products and the image quality you see from the files are about the tradeoffs that get made in the processor.  Show me the files.

[BJL]

I thought I'd heard this before, so I did a little research.  Turns out CMOS does the electron to voltage conversion at the pixel level which I believe has been misinterpreted as the A/D conversion by some in various forums.

To clarify: Panasonic's MOS-based 4/3 sensors are in most respects the same as CMOS sensors, in being active pixel devices: the charge from each photosite is amplified to a larger charge in the process of transferring it to the edge of the sensor. In at least some active pixel (CMOS/MOS) sensors, this amplification is variable, so ISO adjustment is done there. And I believe that charge to voltage conversion is also done on-chip. Canon and Panasonic sensors then output this voltage signal (though maybe the GH-1 sensor is different, outputting a digitized signal.) Canon DSLR's do A/D conversion off-chip, but before reaching the DIGIC chip, which is purely DSP as far as I know. But Sony EXMOR CMOS sensors to A/D conversion on-chip at the bottom of each column and output a digital signal (so ISO amplification and charge to voltage amplification is clearly done on-chip too).

What about CCD's? Dalsa at least incorporates charge to voltage conversion and amplification onto its FF CCD sensor modules, but this is with separate amplifiers mounted next to the corners of the sensor on the same package. Kodak might too, but I am not sure. CCD chips themselves are passive devices, only transporting charges, with no amplification or charge to voltage conversion.

One final note: the amplification stages, including charge to voltage conversion, are a major noise source, and refinements in that stage are often cited by makers of both CCD and active pixel (CMOS, MOS) sensors as contributing to lower noise levels. This can be improved incrementally in what is superficially the same design. For example the 10MP "High Speed LiveMOS" sensor of the Olympus E-3 was improved in this respect over the 10MP sensor in the E-410 and E-510 from earlier the same year, through wiring and amplification refinements.

Thus it is possible to reduce the noise levels of a new camera using the same or a superficially very similar sensor, through amplification refinements. Olympus and Panasonic have indicated that the next sensor improvements might well not involve increases in pixel counts, but improvements in other dimensions of performance.

[teddillard]

OK, sorry for the delay, I've just been off in the corner eating a little crow...  (pass the salt, please?    )

Seriously, this was really a lot of great info to digest- especially the last post.  It got me going on putting together a quick flow-chart (kinda) on what happens where... which I posted on the h2h site.  Have a look, and let me know if I've missed anything.  I still have my crow-eating bib on.  Of course, I haven't changed my mind at all, but you know...  that's just how I am.    

Thanks again guys...

[BJL]

Panasonic's MOS-based 4/3 sensors are in most respects the same as CMOS sensors, in being active pixel devices

I have an update on this: Panasonic's LiveMOS sensors are active pixel CMOS sensors pure and simple, according to the inventor of active pixel CMOS sensors, Eric Fossum, in a thread at DPReview. He suggests that emphasizing the NMOS parts of the sensor design, is simply a marketing thing.