[linux-audio-user] 192kHz

[David Cournapeau]

On 1/27/06, Folderol <folderol at ukfsn.org> wrote:
> On Thu, 26 Jan 2006 16:11:48 -0500
> Gene Heskett <gene.heskett at verizon.net> wrote:
>
> > On Thursday 26 January 2006 15:36, Wolfgang Woehl wrote:
> > >Ismael Valladolid Torres <ivalladt at punkass.com>:
> > >> I don't see any reason to work at 192KHz. Apart from huge
> > >> files, Nyquist is on my side.
> > >
> > >Wouldn't interference of 2 or more signals from above the
> > >audible band have the potential to produce energy within the
> > >audible band?
> > >
> > If there is something non-linear in the mixing process, yes.
> >
> > >Wolfgang
>
> A fact often forgotten is that the human ear is itself non-linear in
> response. It can be reasonably argued that (assuming no other
> distorting factors exist) well out of band signals could interfere
> 'mechanically' inside the ear and produce audible difference signals.
> What is more, these signals will vary from person to person.
>
The ear systen is obviously non-linear: otherwise, you wouldn't have
any masking systems. One of the explanation I am aware for such
effects are interferences in the  membrane basilar in internal ear.

Concerning 192 khz, 24, 32 bits, etc... I think it is important to
make the distinction between storage and reproduction. 32 or 64 bits
(!) is stupid if you feed your convertor with it directly. Now, if you
intend to import this file into some other gear, it may be useful
(note the may, I don't have experience with the use of high end audio
gear, and at  the end, only the ear can tell if something matters).

Also, both frequency range and bit width have influence on maximum
SNR: bitwidth change is obvious; in the case of frequency range, you
can think as a way to "spread" the "floor noise" on higher bandwith,
effectively decreasing the level on the hearing range. In ideal
conditions which never happen in the reality, on bit more is 6 dB, and
a doubling of frequency bandwith is 3 dB.

Nyquist is an *exact* theorem. Mathematically, any function without
any frequency above fr/2 can be *exactly* characterized by its sampled
coefficient. Now, it assumes a perfect filter (infinite slope, no
modification below the frequency cut) which does not exist.

David