Re: [LAU] jack/oversampling

On Sunday 16 March 2014 15:11:20 Ralf Mardorf did opine:

> On Sun, 2014-03-16 at 14:39 -0400, Gene Heskett wrote:
> > On Sunday 16 March 2014 14:25:14 Ralf Mardorf did opine:
> > > On Sun, 2014-03-16 at 08:58 -0700, Len Ovens wrote:
> > > > I would mix the project at 48k or 96k
> > >
> > > Why 96 KHz? 48 KHz doesn't cause any issues, but already provides
> > > best sound quality.
> >
> > That I think is a personal call Ralf, primarily because at 48 Khz,
> > your anti-aliasing filters had better be very very good brick walls
> > by the time you get above 24Khz in input content just for the
> > aliasing control. And aliasing noise, once introduced, cannot be
> > removed by any known math function that does not have precise
> > knowledge of the phasing (aka group delay) of the original signal.
> >
> > And those very good brick wall filters _will_ have a considerable
> > group delay. IMO doing the sampling at 240K, then doing a weighted
> > sum shift (5 stage shift) to decimate the data down to 48K, should
> > result in dropping the alias caused noise floor by several db. Just
> > bring lots of expensive hardware to do that.
>
> Are we talking about reality or golden ears?
>
> I disagree with Fons regarding to the 44.1 KHz are as good as 48 KHz (in
> theory he might be right), but at least as long as there aren't software
> or hardware issues, _we_ are unable to hear a difference between 48 KHz
> and > 48 KHz.
>
Pretty much true, even for golden ears, PROVIDED the anti-aliasing
filtration in front of that digitizer is down at least 70 db by 1/2 the
sample frequency.

You would be surprised at the gear available that depends on the microphone
to do 80% of the filtering, and at the atrociously digitized sound you
could get when miking a set of snares or cymbals with an Altec M21b or a
PZM, both of which have pretty good response yet at 25Khz. Even some of
the better ribbon mikes will cover those frequencies although they would
tend to be very "peaky".

> Analog audio quality is something different Gene. Is you claim that > 48
> KHz you get that analog thingy, I'm missing for digital recordings?

I hesitate to answer this, my German is far worse than your English, and I
don't quite get the translation.

What I am saying is that when digitizing at 48Khz, any content in the input
signal that has components approaching 1/2 the digitization sampler
frequency will be aliased. Said another way, a 20Khz input will be
digitized as both the 20Khz fundamental, and an 8Khz beat frequency "alias"
component, and that this is one of the more distracting artifacts. The
only way to stop it is to filter that input to where the stuff above 16Khz
has been reduced to undetectable levels.
 
> My RME card can do 192 KHz, I never tested it, but I will compare
> recordings ASAP, likely after August this year I have got the time to do
> it.

To properly decimate to 48Khz that would take a 4 stage weighted summing
adder, but given that, the results might be startlingly better (at least
for the golden ears) than a straight 48Khz digitization.

Listening to that 192Khz recording should be a noticeable improvement
compared to a 48Khz recording, just from the reduction of aliasing
artifacts alone, given good enough mikes and sound sources suitable for
exploring/exploiting the alias that will happen in the real world.

These aliasing artifacts you don't hear during the quiet passages of the
music, but will be superimposed on the higher frequency portions of the src
audio signal, contaminating it as it gets louder. It will convert that
cymbal to fingernails on a blackboard, ditto for the brushed snare.

Cheers, Gene

-- 
"There are four boxes to be used in defense of liberty:
 soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author)
Genes Web page <http://geneslinuxbox.net:6309/gene>
_______________________________________________
Linux-audio-user mailing list
Linux-audio-user@email-addr-hidden
http://lists.linuxaudio.org/listinfo/linux-audio-user