Toralf, I think you got it all just right and I believe we were both talking about the same thing in the first place :-). Sometimes it just takes a lot of writing when trying to describe not so easy things.
I have to admit I wasn't looking too deep into the DSLR ccd data sheets when trying to figure out why they chose 22 bits. Plus, it is not very easy to get the bare foot data from the current sheets, since the ccd manufacturers seem to be hiding the basic facts to get more of their sensors sold (?). Astronomical sensors generally have quite big pixel sizes (but there are quite small as well, comparable to DSLR). The smaller the wells, the higher the noise (because of the smaller area to capture photons and smaller space to store electrons. Making better signal processing stages is one way (maybe the only one at this time) to improve the results. Pentax seems to be ahead of the competition with the new (to DSLR world) approach when using better designed signal processing. Let us hope there will also be better sensors in the future :-). Actually the only negative thing I have with the K10D is I hoped they would incorporate a full frame sensor to get more sensitivity and lower noise (more exposure latitude). But maybe it would have competed with their coming 645D too much. BTW I did preorder my K10D today! It will be exciting to see the sample photos of production model K10D. I think what we will see is a little different digital pictures (closer to film kind of images). But I have been proven wrong before... Antti-Pekka ________________________________________ Antti-Pekka Virjonen Computec Oy R&D Turku www.computec.fi > -----Original Message----- > From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf > Of Toralf Lund > Sent: Monday, September 18, 2006 3:41 PM > To: Pentax-Discuss Mail List > Subject: Re: It is ingenious! Was: Does this mean what I think it > means? > > Antti-Pekka Virjonen wrote: > >> I think the actual sensors do have the 14-16 bits you mention - [ > ... ] > >> > >> > > > > Hi, > > > > Yes, actually I was talking about usable bits, taking the noise into > > account, thus giving 14-16bits of usable data. > > > > This is so at least with the good quality astronomical ccd sensors > but I > > > > am sure the DSLR camera sensors are coming close nowadays (or am I > just > > hoping?). > > > As has also been mentioned many times, the max. charge or "full-well" > capacity of the sensor depends directly on the pixel area. Apparently > it's approximately 1000 electrons per square micrometer in a normal > CCD > > I'm not sure about astronomical sensors, but in the linear CCDs I've > used at work, the pixels are 12x12 or 14x14 micrometers, which should > give you 2-3 times the data required for 16 bits. That may not be > quite > enough to leave you with 16 "good" bits after you've thrown away the > noise, but 14 should be no problem. > > On a DSLR, on the other hand, the pixel size has been something like > 8x8 > microns for a while now, but on the latest (10MP) cameras, it's down > to > 6x6 or so. In other words, the DSLRs cameras are not "coming close"; > they are actually moving further away in that the pixels keep getting > smaller! > > A different side of this is obviously the noise situation. You can > also > improve the number of usable bits by reducing the noise, and this is > an > area where the DLSRs probably are getting better - but they also have > to, in order to compensate for the smaller pixels. And it wouldn't > surprise me if the astronomical sensors have very low noise, too. > > In any case, when designing measuring systems (using amplifiers and > a/d > > converters) it is a good thing to have plenty of more bits and > precision > > compared to the original signal. This way you will minimize the > added > > noise. When having a 22 bits a/d converter on the K10D you can also > > measure and analyze the "noise" of the low expusure pixels. You will > get > > all the available information out of the sensor. > > > Yes. It definitely seems like I good thing to keep *all* the info for > the early image processing stages. Also, having some extra bits around > will minimise round-off errors that might accumulate across multiple > pixel value adjustment steps. You may think of extra bits as more > decimals in intermediate results. I think this may be the most > important > consideration behind the choice of 22 bits. > > - Toralf > > > > > -- > PDML Pentax-Discuss Mail List > [email protected] > http://pdml.net/mailman/listinfo/pdml_pdml.net -- PDML Pentax-Discuss Mail List [email protected] http://pdml.net/mailman/listinfo/pdml_pdml.net
