If I'm not mistaken, I believe that the multiple concurrent sampling technique that Marakanetz and Jeff refer to above is actually descriptive of *oversampling*, rather than "upsampling", which has more to do with Jeff's reference to the use of alogrithmic interpolation to synthesize "extra" samples in between the data points to "smooth" the sampling frequency. Both of these are distinct from the adding of extra bits of random dither to "extend" word length from the 16-bit CD data to 24-bits before it is fed to the DAC chips, which is done in an attempt to better linearize the chips' low-level performance. Again, to the best of my (admittedly limited - thanks, Sean!) knowledge, neither of these techniques can actually increase the resolution of the recreated analog signal, but are efforts to extract more subjectively pleasing results out of the digital-to-analog conversion proccess. (I do promise to read that link, Sean - I hope it doesn't show me to be too far off base here!)
I will add that I recently bought a newer generation, popular "upsampling" DAC to audition against my well-regarded older unit, which does only tradtitional oversampling. While I found that upsampling did indeed change the sound, I contucted a controlled test which informed me that - at least for this particular DAC - the results were actually less faithful to the digital input signal it was being fed when upsampling was engaged than when it was defeated. I also wound up preferring, and finding to be most accurate, the performance of my reference model. I intend to write a full review of these sessions and post them separately here on the forum soon, but this experience did once again drive home the fact that it's not the type of technology used, but its implementation, that will often make the biggest difference in performance.
