Originally Posted by majkel
Upsampling and oversampling is mostly dedicated for high fidelity. No oversampling means obvious sin(x)/x distortions when low fs signal is applied to the DAC but some people care for those special effects.
Exactly. Let me explain.
The oversampling part of a DAC is there in order to correct the output, that is natively wrong.
A bare DAC outputs a signal made of rectangular steps, while the sampling theory assumes that an ideal DAC outputs instant pulses, and returns to zero between every sample. The fact that real DACs outpout rectangular steps has an effect on the frequency response. The relation between the steps and the frequency response is not trivial, but here is the final result, taken from Gbo's french paper in the homecinema-fr.com
The frequency response of a bare, NOS, DAC playing a CD is shown on the left side, between 0 and 0.5fc. It drops from 0 dB in low frequencies, to -3.92 dB at 22050 Hz.
The effect of the oversampling part is to expand this response into the ultrasonic frequencies, so that audible frequencies lie entierly in the flattest part of the curve.
Here is a paper that explains why NOS DACs are made : kusunoki
It shows a complete lack of understanding of the way a filter works. The designer of those DACs have it all wrong. He relies on anthropomorphic, nearly superstitious interpretations of the way the electronic works.
In homecinema-fr.com, gbo have simulated the effect of not oversampling a CD on a musical sample. Here are the spectrums of the two samples that he made :
In blue, the frequency response of the correct file, converted to 96 kHz with a proper antialias filter, like in oversampling DACs.
In red, the frequency response of the simulation, converted from 44.1 kHz to 96 kHz with Mathlab, simulating a NOS DAC (limited to 48 kHz). We can see the ultrasonic noise caused by the lack of filtering, and the slight treble attenuation caused by the rectangular steps.
I have compared to the two versions in a double blind ABX listening test and scored 7/8, relying on the attenuation of nearly 1 dB at 13 kHz.
Note that I am unable to distinguish a full band musical sample from a 14 kHz lowpassed version (with an appropriate cosine filter or even a brickwall filter ringing at 14 kHz). People hearing up to 16 or 18 kHz should hear the difference more easily.
Note also that the simulation is imperfect, and underestimates the real attenuation.
So NOS DACs are definitely a bad idea. If they work for you, try to set the treble setting of your amplifier at -1 dB on a normal DAC. The result should be the same.