Wednesday, December 06, 2006

The Secrets of the K10D (Part 2 of 3) - The Bridge: A to D Convertor

One of the most impressive things of the Pentax K10D, as marketed by Pentax, is the 22-bit Analog to Digital Convertor (ADC). My own first impression was even that it was *impossible*, by considering that the competition only offers 12-bit or 14-bit at most, even for the RAW format files.

Right before the official announcement of the K10D in September, some people around the net had already started to guess about the make/model of this special ADC (of course these people already got known to the K10D specifications, by some "insider" channels), a few of them had been able to spot out that it could be a device designed and made by NuCore. Here is one of the very first posts made:-

http://forums.dpreview.com/forums/read.asp?forum=1036&message=19822136

Of course, the use of the NuCore device has never been verified by any official means, nor there is any clue in any official documents. However, the rumours continued and still continue on and off. From time to time, the only device, which contains a 22-bit ADC, which people can find and talk about up till now is only the NuCore NDX-2240. Here is a brief technical catalogue:-

http://www.nucoretech.com/nu3/images/80_downloads/pb_ndx2240.us.pdf

Well, if NuCore is compared with Sony, the maker of the CCD used in the K10D, NuCore is indeed a much smaller company which is not quite well known by the public. A quick glance on the company's background will find that it is actually a very high-tech one with joint venture of different vendors, with strong R&D in specific signal processing devices. You can find a list of the joint venture companies in NuCore's company background information page, as follows:-

http://www.nucoretech.com/nu3/10_company/corp_backgrounder.html

Despite that Pentax is not one of the listed joint venture vendors, probably engineers of Pentax were wise enough to source and purchase the device "off-the-shelf" as an OEM product from NuCore and use it in the K10D.

Now, let's go into more depth about the NDX-2240 and you can look at the above technical catalogue again. The fact is that it is not just an ADC, but actually it is an "Analog Image Processor", which means that the NDX-2240 does have the ability to handle analog image signals as well as to carry out certain image processing jobs. As you will see following this article, the NDX-2240 is actually an analog *and* digital *hybrid* device.

Referring to the device architecture block diagram on page two of the catalogue, the 22-bit ADC is only part of the NDX-2240 and it is located at the AFE (Analog Front End) of the device (the light pink purple block in the diagram). With the ADC are some built-in analog correction/adjustment circuits/function units, including a primitive "Analog White Balance" which adjust the signal level gains of the R, G and B input analog signals. And then, the preliminarily re-conditioned signals are passed into the digital processing block (the yellow one) for "prevention" of various artifacts (and again, those are actually corrections and adjustments) and then a *low-level* digital "gamma" correction is performed.

The "gamma" correction is actually not a gamma function (which is an exponential function). I think NuCore call this to be the "gamma" just by convention as the term has been widely adopted and is easily understood. Instead of a true gamma function, it is a simple "knee points" correction with linked linear lines. To make it more clearly about how comes the term of "knee", which is actually self-explanatory, just look at the knees of our legs, where a change in angle and direction occurs and the knee joints two linear fixed parts (bones) of the leg.

Afterall, the knee adjustment curve or the so-called "gamma" correction is just a tone curve which relates the input and output signals and define the gain at different levels of the input for the output. According to the information provided by NuCore, there are a maximum of 16 numbers of knee points that can be inserted to define the curve, when all colors are processed "altogether". If individual colors are processed (which practically should be the case, as each individual R, G and B waveforms are being passed out from the CCD and then input into the ADC of the AFE of the NDX-2240), only 4 knee points can be inserted for each individual color channel. (But I have no idea on why it is 4 instead of 5. Where have the remaining 3 points gone?)

Well, for the remaining 3 blocks in the diagram are those supporting functions like timing and sychronisation (the light brown block) or just the bus(es) for internal data transfer and control (the light green block) and the Input/Output (I/O) interface of the Digital Back End (DBE) of the device (the last block), of which data will be the finally output from the serial port of the device, in a serial manner.

Indeed, there have been a lot of "massage" to be applied to the original CCD signals received at the AFE until finally the data are output at the port of the DBE of the device, of which the key functions I have briefed in the above. Actually, there are even more to do by this *processor*.

To name a few about those additional signal processing, which are mentioned in the catalogue, include also smart/adaptive "true black level" detection, "automatic smear detection and suppression" (for charge overthrown case across line(s) of pixels for very high contrast scenes) and so on.

Finally, it would be interesting to note that the NDX-2240 is available in 12, 14 or 16-bit versions, despite that all variants will have a 22-bit ADC, which is used for *internal processing* and adjustment ONLY. Just say if the K10D is using the 12-bit version of the NDX-2240, the RAW file cannot store more than 12-bit as all the extra bits of data have already been truncated once the data are output from the port of the DBE of the device.

In order to verify if the K10D RAW file can contain more bits of data and to sensibly guess which version of the NDX-2240 chip it uses, I have checked the latest version (dated 2006-10) of downloadable K10D manual at the Japanese digital site of Pentax, as follows:-

http://www.pentax.co.jp/english/support/man-pdf/k10d.pdf

What I can find relevant is on page 227 of the manual, it states: "RAW data is 12-bit data" and also that on page 51 of the manual it also shows that the K10D RAW file, no matter they are in PEF or DNG, are both same in size, which simply implies both format are in same bits.

Furthermore, it states that a 512MB SD card can host 29 RAW images of both types. So, each image is approximately 17.66MB. A simple calculation can verify if it really contains 12-bit data, the size of the RAW data alone will be: 12 bit x 3872 x 2592 / 8 bit = 14.36 MBytes. And, the extra 3.3 MB room is for the embedded full-size jpeg image and other data like the EXIF. Therefore, unless the RAW file is compressed (which has never been mentioned and confirmed by Pentax in any of their documentation. So, at this moment it should be the case that the K10D RAW files are uncompressed (and there is no option of compression of RAW data neither), the chance for K10D RAW file contains more than 12 bits of data is very unlikely (do note again that the manual have already said that it is "12-bit").

At this point, I hope that I have explained one of the FAQs about why whilst the K10D has a 22-bit ADC, as marketed widely, it cannot have a higher than 12-bit RAW. In short, it is just because the hardware device of the K10D uses probably does NOT support this.

Okay, in the coming last part of this series of articles, I shall brief about the digital Image Processing Unit (IPU) of the K10D. And that I shall reveal what actually the Pentax "PRIME" image engine is.

Update on Dec. 11: To discuss about the significance of the number of bits but as well as the insignificance of the number of bits of a DSLR and its RAW format, I opt to write another article on this topic, in more details.

>> See also: The maker of the ADC and IPU of the K10D

>> Back to Part 1

11 Comments:

Priyantha Bleeker said...

I saw at the dpreview forums posts from users wich reported that the PEF RAW format is compressed and the DNG is not compressed.

And there are also posts that suggest that Pentax doesn't use the NuCore ADC, this Roland Mambo said this and others.

Can you explain this better ? And come with some evidence ? I am very intreseted !

PS. Sorry for the bad English ;)

BasK said...

Very interesting article, and very interesting questions of Priyantha. I really would like to see the answers on this questions too.
Why would Pentax implement a 22 bit ADC without using it?

Anonymous said...

A lot of conjecture, but little proof of anything here!! Nice fantasy story though ;-)

RiceHigh said...

Thank you for all your replies!

-> In response to Priyantha:

Besides the latest K10D user manual which I can download at the Japanese Pentax digital support site, I have no further information on whether the PEF or DNG is compressed or not. But I suppose that manual is updated. At least at this moment, according to the information in the mentioned manual, both PEF and DNG are in the same file size and are NOT compressed (actually, I've already explained the reasons for drawing up the conclusion in my article, with some simple calculations).

Regarding Roland denied the use of NuCore device in the K10D, have he mentioned clearly about what device has actually been put in the K10D body? Pls tell us if you have further information from him. TIA! (but pls don't tell us it is a "PRIME" :-))

-> In response to Bask:

Pentax is not "implementing a 22-bit ADC without using it". As explained in my article, the ADC is only ONE part/component of the device NDX-2240 and within the device. The 2240 receives raw analog signals from the CCD and input them into the ADC and then finally *output* as either 12-bit, 14-bit or 16-bit digital data, depending on the actual version of the chip.

-> To anonymous:

As I have already said, except that it has not yet any *official* proof on the use of the NuCore in the K10D, everything matches very well with those Pentax published specs and marketing press releases for the key features and functions.

Regarding all the technical talks about the NDX-2240, it is all based on the technical information contained in the technical catalogue, along with some other technical information published in NuCore's website.

Whilst you can still say it is a rumour for the use of the NuCore and you can still doubt that if the K10D adopts the NuCore, I cannot agree anymore with you about your words of "conjecture" and "story"!

Instead, all these are factual information. I just try to explain those information with my words.

Priyantha Bleeker said...

RiceHigh, are you planning to buy one or more K10D's ?
So yes, maybe you can open it ? :P

I don't know if Roland knows more, I didn't shaw any more, but maybe he will explain it later ;)

I've some official PEF and DNG files from Pentax and the size of an PEF differs much from the DNG, a PEF is around 9 a 11 MB and an DNG is around 16 a 18 MB.

The NuCore ADC...it is an intresting story ;)

_Mike_D said...

Hey Ricehigh, thanks for the great article.

I saw your comment about the lack of compression (acording to the manual) on k10d. It looks like on my k10d the PEF files vary wildly in size. Typically they weigh in around 8-12meg although I have seen them as small as 6.8meg and as large as 16.2. The DNG's on the other hand are almost always right around 16meg

This obviously suggests that the PEF's are compressed, and the DNG's are not. Weather they use lossy compression on the PEF's is a different question.

I've found your articles very helpful in the past, keep up the good work!

RiceHigh said...

Hi, _mike_d,

Thank you for your kind comments!

BTW, have Pentax included any amendment to the K10D printed manual for mentioning something about the PEF compression, estimated file size as well as if the compression is loseless or lossy?

Any further information provided will be greatly appreciated!

Basil said...

Just look to samsung GX-10 site:
http://www.samsungcamera.com/product/pro_view.asp?prol_uid=1923&cat_uid=62
On picture http://manage.samsungcamera.com/upload/editor/200612/15_2_2006121218131953.gif
we can clearly see NuCore 2240 chip.

RiceHigh said...

> Basil said...
> Just look to samsung GX-10 site:
> http://www.samsungcamera.com/product/pro_view.asp?prol_uid=1923&cat_uid=62
> On picture
> http://manage.samsungcamera.com/upload/editor/200612/15_2_2006121218131953.gif
> we can clearly see NuCore 2240 chip.

Yes, Samsung got the "secret" revealed and exposed despite that Pentax's own photo of "their" chip shows no marking on it at all! Yes, nothing. Well, what a good partner Samsung is! ;-) Nonetheless, it's good for *us* at least as we can know the truth and get something confirmed, finally and officially! :-D

Anonymous said...

As far as I know, in common DSLRs, there is no such thing as a 22-bit A/D converter, this may just be marketing hype describing the post-processed interpolated color reconstruction. The Pentax K200D has been said (in some reviews) to have such a device, but it is more likely to have an actual 10-bit or 12-bit A/D converter as is often used.

I believe that JPEG images use only an 8-bit conversion. RAW images saved as either PEF or DNG format are lossless except for some possible pre-processing that cannot be turned off. When downloaded into a computer, PEF files of the RAW data are more compact at 11.9 MB/image. DNG images are about 1/3 larger at 16.2 MB.(Try this to see for yourself.) I don't understand why the tables in the Specifications section of the User's Operating Manual show a given size of memory storing the same number of images in either PEF or DNG format.

In the 10MP PEF format, each pixel has 11.9 MB/10MP = 1.19 bytes of memory or up to 9.5 bits of brightness information.

Colors are usually interpolated from an RGB or similar Bayer color filter pattern on the image sensor to reconstruct what is often 8-bits of resolution for each of the 3 neighbouring primary colors of nearby pixels. The resulting 24 bits would give us "truecolor" of 16 million colors, but Pentax's "22-bits" implies a lessor 4 million colors which may be just slightly less than what the human eye may be able to differentiate.

RiceHigh said...

Not exactly, it is really a 22-bit convertor, but the 22-bit sampling is just made at the "Analogue Front End" (AFE), which is the first stage where the analogue signal is received. The output data is shortly truncated to 12-bit as output at the digital output of the ADC (called the Digital Back End (DBE) in the NuCore's Catalogue). There is a version of the NuCore ADC which has a 14-bit output but unfortunately Pentax hadn't used it.

Theoretically, it is always more beneficial to oversample (not in the frequency sense, but just talking about quantization level and resolution, e.g., for both audio and video data, and then downsample/downsize the original data into a more compact content. However, in the case of K10D, I really doubt the meaningfulness of doing so as the CCD imager (background) noise would be dominating factor here, i.e., most of the Least Significant Bits of the 22-bit ADC would be containing only noise, just for the first time when it gets the signals. The 22-bit oversampling ADC will only be useful only if the sensor's S/N ratio is higher, not so for the Sony CCD used in the K10D. The 14-bit output is useful, though, but anyway there is no Pentax/Samsung DSLR has RAW output higher than 12-bit so far.

Last but not least, the Samsung GX-20 has a 22-bit ADC (the same one in the K10D, I bet) but not so for the K20D, which has only a pure 12-bit ADC, from what it is told in the official information disclosed by Samsung and Pentax, respectively.

Post a Comment

Related Posts

 
Creative Commons License
RiceHigh's Pentax Blog by RiceHigh is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.