If there is any digital preprocessing before outputting The RAW file to the user, its always better to do The processing at higher bit depth because there will Be less interpolation errors at each procession step.
For example, when I make CDs from LPs, I record and do All my editing and processing at 24 bit and only down Convert to 16 bits at he very last step to make the CD And it sounds better than doing everything at 16bit. jco -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] On Behalf Of John Francis Sent: Sunday, October 15, 2006 3:13 PM To: Pentax-Discuss Mail List Subject: Re: K10D 22 bit A/D conversion On Sun, Oct 15, 2006 at 10:02:36AM -0400, graywolf wrote: > Well there is the fact that your analog signal from the sensor is for > one color pixel from the sensor. Then several pixels are analysed, > combined according to that analysis, and written out as 4 3-color > pixels. And that is just for a RAW image. It is not a simple convert one > pixel to x bits operation. Is 22 bits over kill for that? I do not know. > > I suspect that they are using a 22 bit ADC simply because that is what > was available. But if they are eventually planing 16 bits per pixel > output, a 22 bit ADC is not over kill at all. A point to consider is > that with modern IC's that ADC is not just an ADC it is most likely a > computer chip as well. > > The point I am trying to make here is I think the analysis of this that > folks are making here on the list are way over simplified. They're over-simplified because I can only think of one or two people on the list likely to be both capable of, and interested in following, in mind-numbing detail, the full argument. But, briefly, for all the sorts of operations we are talking about here (addition, multiplication, and division) all you need is one or two guard bits of additional precision to be sure that your total possible error, at the end of the whole string of calculations, is less than one low-order bit. Using a 22-bit data path for data which only starts off with (at best) 15 bits of significance is totally unnecessary. As you suggest, I suspect they are using a 22-bit DSP mainly because that is what is available. Furthermore, unless they intend to replace the sensor, providing 16-bit RAW output still wouldn't require a 22-bit processor; the limiting factor in the chain of calculations would be the signal-to-noise ratio of the original signal. To use a simple analogy: You can calculate the great circle distance between New York and San Francisco to any precision you choose - down to the nearest mm, if that suits you. But if your value for the radius of the earth is only measured to the closest metre or so, then calculating distances to a precision of mm is wasted effort, with meaningless results. -- 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
