good post.

its hard to A/B this stuff since they wont send me one to compare, so trusting your ears is easier said than done. i would like to see something on paper(or screen) that proves the accuracy is there. 24bit processing has been refined and proven, while this is almost unknown and unproven. i need more.

i never buy a car the first production yr of the model either.

Totally agree Jimna.

My gut is telling me that the analog noise floor of the output stage won't be low enough to make use of the extra DAC bits, but I can't prove that.

That and the human ear only has a dynamic range of about 130 decibels, well at least until the pain really sets in.

I've been doing a little research:

Pretty much all the digital processing chips out there are already using 32-bit data lengths. In the case where 24-bit DACs are used some of the data is just not sent to the DACs.

Cirrus Logic has an 32-bit Audyssey processor chip (CS48AU2) with built in algorithms. The interesting note on the data sheet was that Audyssey algorithms only support 48, 44.1 and 32 KHz data rates.

AKM Semiconductor offers the AK4390 32-bit DAC which is used on the Esoteric SACD SA-50. For $5,000 you get a 130 dB SNR.

Wolfson has the WM8741. (See ARCAM products)

TI/Burr Brown has their PCM1795. On paper it is not as good as the AKM or Wolfson, but there's a lot of other factors to be considered.

These all seem to be in production, mainly in high end SACD players.

32 bit DACs or 32 bit ADCs would just add to the expense and provide no real world benefit. Even with 24 bit DACs or ADCs, the least significant bit (LSB) is below the noise floor. So is the next LSB and more. While properly implimented 24 bit DAC and ADCs can give a very worthwhile improvement over 16 bit, it is far short of the 8 bit difference.

32 bit digital processing, on the otherhand, is worthwhile compared to 24 bit processing. 32 bits gives you a lot of leeway for digital junk (truncation, roundoff errors and such) that accumulates due to every digital operation, the total error part of the digital word working it's way from the LSB end of the digital work towards the MSB of the digital word as the digital processing continues. 32 bit processing allows more bits to be "wasted" at the LSB end (for a given amount of proccessing) and still have enough "untained" bits to feed the 24 bit DAC.