Non Over Sampling DAC

[jaimilson]

1. The signal out of a DA is made of "steps". It is in fact the desired analog signal but with some added high frequency energy (the "steps") which is all above Nyquist, but needs to be removed. True, we do not hear such high frequencies, but sending such signals through electronic gear can cause problems. Say a power amplifier or headphone, or speakers are not designed to processes such high frequencies. The "step energy" covers a wide frequency range, from Nyquist to many MHz or even higher. So we want to remove (filter) that energy.
Now, say you have a 44.1KHz material ,ƒÏ Nyquist is at 22.0KHz. That calls for removing the energy above 22.05KHz, but keeping the audio (say up to 20KHz). It is very difficult to have a practical filter that does such a job. We have 2KHz of transition between all pass (without distortions and phase alteration) and full rejection.

La contrainte est effectivement maximum sur le filtre, comparé à l'usage d'un sur-échantillonnage associé à une interpolation nécessairement arbitraire et l'usage d'une circuiterie d'horloge plus complexe (cf. gigue etc).

2. Theoretical sampling is not a "stepped wave,ƒ˘, it is a very narrow pulse at sample time, but zero most of the time (zero between samples). But such signals have tiny energy so we resort to a practical wave ,ƒÏ the step wave (keep the value of a sample until the next sample). "The amplitude vs. frequency response of a "stepped wave,ƒ˘ has some loss as you approach Nyquist. That is undesirable, and difficult to compensate for.

C'est ce que fait le TOTALDAC. Je n'ai lu aucun commentaire concernant un effet audible indésiré de son filtre, qui, pourtant, corrige quasiment les 3 dB à la fréquence de Nyquist. On pourrait ne pas corriger autant à 22,05 kHz, d'autant plus que l'on s'intéresse à 20 kHz.

Therefore, increasing Nyquist is a good thing. It makes a good filter design practical, and the flatness response to say 20KHz improves dramatically.[...]

Si on se donne e.g. des spécifications d'horloge très serrées, cela peut changer la vision de l'ensemble du design...

As a rule, DA's do it (upsample) internaly. But say your DA can recieve 44.1KHz and 88.2KHz by design, then the designer can take a 44.1KHz signal and do a X2 upsampling before the DA (thus feeding the DA 88.2KHz). There is no reason to do it if the DA has a good internal upsampler. You can do it if the DA internal upsampler is poor. That is an implementation issue. [...]"

Le choix du circuit intégré et son usage sont liés à l'ensemble du design.