Friday, December 01, 2006

The Secrets of the K10D (Part 1 of 3) - The Heart: CCD Imager

In my this new series of technical articles, I am going to write some of the inside "secrets" of the K10D. Well, I would call those inside things the "heart", the "brain", and also the "bridge" connecting the two, i.e., from the CCD, to the Analog to Digital Convert (ADC), and then to the Image Processor Unit (IPU).

As the first Part of this series of articles, I shall brief about the CCD used in the K10D. As you may know, it is a Sony CCD imager/sensor. First of all, here is the simple (marketing) datasheet, in Japanese:-

http://www.sony.co.jp/Products/SC-HP/cx_pal/vol69/pdf/icx493aqa.pdf

Well, the model name of the CCD used is called "ICX493AQA". The ICX493AQA is a newer generation of the 10 MegaPixel APS-C sized CCD imager from Sony, which its predecessor is the Sony ICX483AQA that used in the Nikon D200. Do note that the K10D is not the only nor the first DSLR model that uses the ICX493AQA, actually the Sony A100 is the first one used this new model, for Sony themselves.

Okay, let's go through a bit about the information provided in the above factual datasheet. Firstly, the new ICX493 has 2-channels in two pipelines which allow *analogue* pixel data to be "clocked out" by shifting the signal (voltage) levels in each recorded pixel and then propagated in the two analogue chains of electrical waveform, at a clock frequency of 25MHz.

For the (analogue) addressing and sequencing of the waveform chain, diagram 1 on the page two of the datasheet is referred (note the numbered pixels of R1, R2, R3 ... to R8, ditto for B and G). In short, Red and Blue pixels are "queued" and put together in the same channel (which is actually the same "chain", I will have it explained in more details later in the next paragraph) for output and then the Green pixels, denoted Gr (along the Red pixel lines, horizontally), and the Green pixels, denoted Gb (along the Blue pixel lines, horizontally) are grouped together and output at the other second channel.

In more details, The Red pixels are clocked horizontally in the sequence and pattern as shown in the diagram 1, for two adjacent horizontal lines of pixels, from left to right. The first cycle is for the Red lines of two and then the second cycle is repeated similarly to the first cycle, but for the Blue pixels this time (again of two adjacent horizontal lines). The same applies for the green channels for the two type of green pixels. Recursively, the whole frame is hence "clocked" out of the whole CCD arrays from bottom to top with repeating processes, then the data from the two channels is re-composed again, to form a complete composite picture data of the whole frame.

To compare against the older ICX483AQA, which has 4 individual channels of R, B, Gr and Gb respectively, the ICX493 undoubtedly will have a lower output data rate and hence a lower frame per second rate (3.3 frame/s Vs 5.57 frame/s) just because the ICX483 has double the channels but both models run at the same clock frequency of 25MHz. In this connection, it is interesting to note that the maximum number of continuous shooting frame rate of the K10D is by itself limited by the specifications of the Sony CCD sensor, but not the peripheral components within the K10D. But there is a small mystery on the K10D specifications: I actually had got a copy of the Pentax Press Release file well before the embargo released date of 14 September, the K10D specifications contained in it stated that the continous frame rate is 3.3 fps but now that all the published specifications have been updated to 3 fps only. Interestingly, the Sony A100 has also a 3 fps rating but not 3.3 fps, which is again lower than the theoretical maximum allowed rate for the Sony sensor.

Although more channels will lead to a faster frame rate, for the same clock speed, the more channels a CCD has, the more vulnerable the CCD would be for producing artifacts by itself, for the composite picture frame. Just say if the ICX493 which has two channels for the picture data, the whole picture will be re-built by using the two channel data streams. Similarly but in contrast, the ICX483 will require four data streams to build up the whole picture frame. Just in case one or any of the data streams does have some inconsistency than the others, owing to the fact that no two circuitries or (micro-)electronic components like amplifiers, etc. are built and could be built *totally* identical practically, some inconsistency within the frame might be seen for the final composite picture. A good example for this problem/technical limitation is the famous banding issue of the D200. As a remedy, the subsequent ADC and/or the IPU can take up the adjustment and preset calibration (at the factory) jobs, so as to eliminate the undesirable effect(s).

Regarding the "banding" issue, there are actually many causes for the generally called "banding" phenomemons (which are more than one type). I opt to write more about "banding" in a coming article, after I finish this series of the 3 "K10D inside" articles.

Afterall, a two-channel versus four-channel design is just a case of give and take. Whilst the continuous frame rate has been decreased (almost by half), a more consistent the picture quality would be better guaranteed, comparatively. Well, I would prefer the former as still picture quality should be of no doubt more important than continuous shooting, at least for me. Also, 3 fps is not a very slow frame rate already, it should be adequate for typical use.

Secondly, another major advantage/improvement of the ICX493 over the ICX483 is about the power consumption. Now that the 493 requires only 6 Volts to drive and the previous 483 requires 7V. As a result of a lower voltage and an overall simplified circuit(ry), the power consumption has been dramatically reduced by 37%, as claimed by Sony. So, what's the implication of this? Beside it consumes less power which means a longer battery life, more importantly is that there is less heat generated which simply means less electrical noise (and hence picture noise in the final image) will be introduced as a result of a higher temperature within the CCD during operation. As a final result, the Signal-to-Noise ratio of the 493 CCD could be improved.

Thirdly, another advantage of a simplified circuit design is the package size of the CCD component is minimised. The new QIP (Quad In-line Package) has only 64 pins, versus the 80 pins QIP of the 483, which means fewer connection points to save manufacturing cost and increase the reliability (very marginally), too. Thus, the outer flat area size of the 493 package boundary is only 81.8% of the 483 (there is a typo in the page 2 of the datasheet which mentions "91%", as I calculate it to be only 81.8%, according to the actual dimensions provided in the datasheet) and that the weight of the new CCD is only 75% of the old model.

But do note that the gross total light sensitive image area of the 493 is actually smaller for the vertical height (18.42mm Vs 20.3mm) whilst the horizontal length is the same (26.1mm). However, since not all the area of the CCD light sensitive area is used for imaging, as some areas must be masked out for calculating the dark current (to rule out the background noise for the true image signal), the actually "sensible" area and number of pixels remain the same (3872h x 2592v for K10D, ditto D200), which results in a recommended diagonal length of 28.328mm by Sony and thus the multiplying factor is 43.2666mm / 28.328 mm = 1.5273. Just for your further information, the 43.2666mm is the diagional length of 135 film frame which is in 36 x 24mm.

Okay, in the next Part of this series of articles, I shall talk in-depth about the "Bridge", that is, the AD Converter used in the K10D! Please come back later! (Of course, I have done my research already for something to be shared with you folks! :-))

>> Go to Part 2

9 Comments:

Elliot said...

Thank you. Informative read, look forward to parts 2 & 3.

Anonymous said...

Hallo HighRice,
nice artikel.
People who used the K10D stated, that the frame rate is 3.2. Therefore it is faster as the stated 3 fps and close to the theoretical 3.3.
regards
Rüdiger

RiceHigh said...

Rüdiger,

Hallo, very long time no see!

Thanks for your kind comment and I shall try to finish the remaining parts in details when I have the time.

Regarding the actual continous frame rate, I just wonder why Pentax don't publish it as 3.2 fps if it actually reaches the number.. Would it just be the peak but not the average performance of the K10D, so that Pentax needed to be conservative?

RiceHigh (well, not "HighRice"! ;-) Of course, as one of my old net friends whom we have got known for years, pls feel free to call me Michael if you like :-))

p.s. Have you seen Roland Mabo recently? It seems that he has disappeared recently.

Anonymous said...

any indication of the dynamic range of the sensor?

RiceHigh said...

Yes, the Dynamic Range (DR) can be derived.

The Sony spec states that the Signal-to-Noise (S/N) ratio is 100dB. Thus, first divide 100dB by 20 (explained more below), which is equal to 5. That means that the ratio between the highest possible signal level (the brightest one) and the lowest noise signal level is 10 to the power 5. In log base-10 scale, the (ratio) unit is simply 5. Do note that no "dark" signal could be darker than the noise signal in the background.

Now that when we talk about the DR, it is simply a ratio in log base-2 scale. To convert the figure is simple, just multiply 5 by the scale ratio difference between log-10 and log-2, thus: 5 x log10/log2 = 16.6 EVs. But then this figure is impractically high, IMHO. It is just a theoretical figure of the CCD itself, or a figure which is obtainable under a specific (test) condition. Indeed, the final DR obtained is to be determined by a bundle of other factors within the camera, e.g. the contrast level of a lens, exposure level, and etc.

In fact, the dB figures quoted by CCD makers are usually misleding, they should multiply by 10 only instead of by 20 but then they used to do so, just for a larger figure! It is just because the output signal from the CCD is current or voltage, but not (electrical) power, despite the CCD detects light energy power.

Since CCDs detect light energy power and converts into linear current/voltage signals, they are also called the "square law detectors".

As I have noted that the 100dB figure quoted is somehow unrealistic, owing to the large figure shown. To compare, a similar Kodak 10MP CCD clearly states about the DR in dB is only 71.5 dB, which I think is more reasonable, see:-
http://www.kodak.com/US/en/dpq/site/SENSORS/name/KAF-10500_product/show/KAF-10500_productSpecifications

chocoberry said...

the number "-100dB" in the Sony PDF file indicates smear(スミア), not dynamic range. I think that's why you thought the 100dB figure is unrealistic.

Anyway, very nice article.

RiceHigh said...

Thanks for the supplement, chocoberry.

Todor said...

Is it full frame or interline CCD?
Does it have microlenses?

RiceHigh said...

I think it is just a typical full-frame CCD architecture without any buffer array nor features any electronic shutter and it is slow in speed.

As for microlenses, typically products of the same generation in APS-C size do not have them.

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.