Godfrey DiGiorgi wrote: > On Oct 19, 2007, at 3:09 PM, Toralf Lund wrote: > > >>> Possibly with 14bit RAW's >>> >> Do you think that would be possible? >> >> I know you (or was it someone else?) have been saying the light >> collecting capability of sensors has been improved lately, but in >> order >> to achieve a higher bit count, a better electric charge storage >> capacity >> would probably be needed as well. Or at least, some papers I read on >> this a year or so ago seemed to suggest that 14 bits won't be possible >> with the current element sizes even with 0 noise. >> > > My take on this: > > 12 bit vs 14 bit RAW files has more to do with quantization than with > charge capacity, which translates to DR. They could quantize the data > captured by the current sensor to an 8, 11, 12, or 14 bit > representation. The question is whether or not it is useful to do so. > Exactly! > 12 bits is 4096 discrete energy levels; 14 bits is 16,384. Is there > adequate resolution of energy levels in the sensor to make expressing > that energy in a larger quantization meaningful? That was essentially the question I was trying to ask.
I was saying that based on some papers I've read in the past (but don't have in front of me, or care to search for right now), the answer seems to be "no" unless *something* has been done to improve the sensor's capacity for storage of energy levels - or if you like, to make room for some extra electrons/electron holes. But actually, I didn't really do any maths when I said that. After doing some quick calculations, I think that there would actually be enough levels for 14 bits, or 15 even - but not 16. You have to consider the noise as well, though. The noise will essentially add random variation to the (theoretical) lower bits from the sensor so that it is not possible to extract useful info about the energy levels from them, and including them in the final output gives no extra value. The number of unusable bits depends on the amount of noise, obviously, but having just 1 bit to throw away may not seem sufficient. In other words, a sensor using the "old" technology would in fact have enough levels for 14 bits, but such a setup would leave virtually no room for noise. Oh, and if you wanted to know, the papers I mentioned suggested that you can simply assume a storage capacity of 1000 electron/electron holes/energy levels per square micron of sensor area. > The assumption is > that, given the present sensor, 12 bits is an accurate representation > and there is no added value in going to a larger representation. > > The implication is that a new sensor might have sufficient resolution > of energy levels to make a 14 bit quantization an advantage. Improved > accuracy is the result, exercising more of that 22bit A->D converter. > Maybe increased DR as well, if the imager actually does have > additional DR. > > Godfrey > "I'm no electrical engineer, I play one on the PDML." > - Toralf "Me too (I did study at the electrical engineering department, but chose the IT option.)" -- PDML Pentax-Discuss Mail List PDML@pdml.net http://pdml.net/mailman/listinfo/pdml_pdml.net to UNSUBSCRIBE from the PDML, please visit the link directly above and follow the directions.
