Oversampling vs upsampling

By the way, the number 24 in 24/96 has nothing to do with
upsampling or oversampling. It simplies indicate the digital filter (or DAC) would convert the original 16 bits into 24 bits. Number of bit and sampling rate are mutually exclusive. You can certainly have the original 16bit/44k signal converted (or upsampled) to 18bit/96khz or 18bit/192khz.
But nowaday, most DAC companies such as Analog Devices and so on only manufacture 24 bit DAC hence their popularity in cdp.

I think a lot of confusion are due to the way companies marketing their products but it really boils down some basic technology.

Andy, Thank you very much!!

Thanks for a great explanation, Andy2! Now does the number of DAC chips used in the upsampling/anolog conversion have anything to do with sonic improvement. I know that Accuphase uses more I/C's in their higher pried models?

Statistically, I would think that it "smooths" out the signal at the output.
Having two dac is equivalent to summing the output level of the DAc, average them (like (a + b)/2), then send them out to the output. It's like taking the average of two signals.

Whenever you take the average of anything, you reduce the peak and dip of that things that you try to average.
It's like the average income of people who live in California and people who live in Nevada is more or less the same. But if you look at the income of each individual in each state, you would see a lot of "peak and dip".

The same applies for electrical signal when you try to average them.

I have a Cary 303/200 and it also uses 2 dac per channel.
The Cary 306 uses up to 4 dac per channel and I think that's why it sounds smoother than the 303.

Close. What happens when you sum the outputs of two devices is that the signals they have in common (the real data or music) is summed (x2) while the random signals (noise) is cancelled. This is because the real signals occur at the same time in the same phase while the noise on one DAC is likely to be slightly different from the noise on the other. When you sum, the noises cancel and the real data sums.

Result is improved SNR.