Sony 12 mp, 11fps APSc CMOS ...

Has the Pentax shown any benefit from a 22bit ADC? None that I see.
Thing is, the decision to use a 22bit ADC could translate into flexibility/simplification of the K10D's internals. For example, you could get away without using analog ISO amps, doing all multiplications digitally. In fact, this may have been an engineering decision that simply got picked up by someone in marketing.

David
 
Unless the sensor has a DR > 72dB, the 2 extra bits aren't really
going to have much of an impact. If the DR is less than 72dB, then
all the extra bits are doing is specifying the noise more precisely.
For a single pixel, true. However, images are collections of pixels
and many gradual tone gradients span very many pixels. With each
pixel quantized to 14 bits, the average across a cluster of pixels
can settle to a value consistent with 14 bits. Plus, it's a bit
easier to manipulate the data during postprocessing (esp. steps that
stretch the tonal range) without adding gaps.
The same thing happens with 12 bits averaged over an area, the underlying average will have greater accuracy than the individual samples.
Exactly. It should be clearer with some real type numbers and
quantization noise.

Assume that the noise floor is at 11 electrons and a full well
capacity is 40,000 electrons (these are in the ball park, but I don't
have exact numbers). There are only 11.8 bits of possible dyanamic
range. But this would likely have the AD count 10 electrons for each
bit, and there are a number of tones missing.
Number look sort of in the ballpark. From some measurements of the 1DII (Clark), I estimate the FWC of the 40D (taking into account pixel size) to be about 25k e at base ISO. Read noise is a bit harder to guess, but charitably might be 8 e(?) giving 11.6 bits DR. A 14 bit sample with 11.6 bit DR would have the last 2 bits fluctuate quite randomly. If you took the 12 bit sample and added 2 bits with some dither, you could fill up the gaps. Since in the 14bit sample, the 2 smallest bits are noise, you achieve very much the same effect. The 12 bit sample has nearly all the information as the 14 bit sample.
Demosaicing, white
balance, and contrast processing contain multiplications and
divisions where quanitzation errors are compounded. At 14 bits you
can count every 3 electrons and reduce these errors. When displaying
or printing this extra precission disapears, but the extra tones can
yield better processing results.
I'm not talking about post-processing the RAW, of course minimizing arithmetic round-off error is important by doing calculations with high bit depth, but I'm talking about the information contained within the actual RAW file.

I propose an (simulated) experiment. Take an ideal image. You can get a "practically noiseless" image by heavily downsampling a clean image to give you a 16bit image. Recale the image values so that the max pixel value is 25,000. Add noise with an RMS value of 8 (e.g. could use a Gaussian with SD=8). Quantize using 12 or 14 bits. These represents the the RAW files.

Now we can see if the 14 bits in the latter file actually make a difference, or is it simply a case of dither. Take the 12 bit image, multiply by 4 (to get a 14 bit number) and then add a random number between -8 and 8 (to randomize the last two bits). Process both images exactly the same and see if there is any discernable difference. Actually, it'd be good to run this as a double blind experiment.

Cheers,
Daniel.
 
IMO, ADC with 14 bit carries real benefits to images. More dynamic range means more tonal levels, especially in dark tones, where the effects of posterization can become less visible, expecially when the image is edited by image processing programs.

I think that the number of bits of ADC, the pixels number and the noise control are all areas where the today technology has reached great result.
IMO, another important factor that can become important in the near future.

Actually, with "Bayer technology", each pixel captures only one channel (red, green or blue) of incident light. The other two channels are interpolated, using the value of the missing channels read by adjacent pixels. This means that the for each pixel of final image, the 66% of color information is "estimated". Then, the resulting color is only an approximation of the real color of light inciding on that pixel.

Viewed by another side, only the 33% of light that reach the sensor is really read; the other 66% is trashed! This is a big waste of "information", that has negative effects on image quality (less details are caught) and noise (less light is read).

The "old" film technology, at the opposite, captures all the incident light, without any lost! I think that big improvements could became considering also this "lost" light, like the way followed by Foveon sensors.

\Pino
 
IMO, ADC with 14 bit carries real benefits to images. More dynamic
range means more tonal levels, especially in dark tones, where the
Yeah, if the DR of your system is 14bit. If your sensor is limited to around a 70dB ratio then there is no point in 14bit per se (assuming you're AFE amps are doing a good job)
'
Viewed by another side, only the 33% of light that reach the sensor
is really read; the other 66% is trashed! This is a big waste of
"information", that has negative effects on image quality (less
details are caught) and noise (less light is read).
Yes, theoretically a bit more than 3dB loss compared to optimal.
The "old" film technology, at the opposite, captures all the incident
light, without any lost! I think that big improvements could became
considering also this "lost" light, like the way followed by Foveon
sensors.
Except that you don't know how efficient the chemicals are at capturing photons... My understanding is that CCDs are already much better...

Laurens
 
So if there are some substantial improvements in Sony's design, it
could make a big difference in their competitive position with Canon.
Although what you said is basically true about bodies, the only way Sony is going to make a "big difference" in its competitive position in this marketplace is with a substantial improvement in, and enlargement of, its current (non-)system. Having a camera body which is technically more advanced (for a few months) is not going to sway a pro who needs lenses or a FF camera that can be rented anywhere (just one example) or even a consumer with some investment in lenses and flashes of another system.

I have no doubt Sony can and will increase their marketshare and, over the next decade, attract some professionals and consumers who are willing to buy the higher end equipment which has the margins that enable further R&D, but it is not going happen with a couple of excellent bodies and/or new chips.
 
There seems to be a lot of interest in the use of the IMX021 in dSLR. Fair enough, but my own interest would be to see this in a camera such as an upgraded DSC-R1.

One of the other interesting (but unanswered) things about this sensor is how many frames per second it could do at 4> 1 and 9> 1. Nominally, 12.4Mpx/s translates to 1920x1080p60.

One disapointment of the DSC-R1 was no movie at all. I'm not suggesting that putting 1080p60 into a revised DSC-R1 is sensible, but say 1280x720p24/25/30 would be a welcome addition.
 
The Fuji S5 (and S3 for that matter) have double the photosites. They have the "normal" size photosites, and a set of smaller photosites. These smaller sites are less sensitive to light. Which means they can capture some detail in areas that would normally blow out on normal sensors. I'm probably oversimplifying things....but I think the general idea is right.

-Tim
 
I think your "right on" with your evaluation of Sony. They dominate the high end Video Camera business over Canon and they have the necessary financial strength to throw some serious "left hooks" to Canon's midsection within a couple years. No manufacture in the world does better electronics than Sony and if they are serious about entering the Pro DSLR business, they will be heard from.
--

' You don't have to have the best of everything to get the best out of what you do have'.

 
Has the Pentax shown any benefit from a 22bit ADC? None that I see.
Thing is, the decision to use a 22bit ADC could translate into
flexibility/simplification of the K10D's internals. For example, you
could get away without using analog ISO amps, doing all
multiplications digitally. In fact, this may have been an engineering
decision that simply got picked up by someone in marketing.
I strongly agree, and things I've seen floating around written by people who know people who should know (hehe) say that that's the exact truth of it. The K10D does a bunch of things on the digital side that on many other cameras are in the analog circuitry.

-Matt
 
For one thing, Sony needs to stop thinking it can charge 30% more than Canon for equivalent lenses and accessories. If they're trying to get market share. It's quite probable, of course, that on many lenses it costs Sony 30% more to make them than Canon, because of the low Sony production levels right now, but frankly Sony isn't going to get market share unless it gets competitive on price.
 
Daniel,

I was thrown off by the chip size, it seemed like it was neither DX nor full frame. I went back and re-read the note and it certainly is DX compatible.

Sorry 'bout that!

--
Ignacio Féito
México
 

Keyboard shortcuts

Back
Top