[linux-audio-user] 44.1, 48 or 96?

[Paul Davis]

On Mon, 2005-09-12 at 13:30 +0100, james at dis-dot-dat.net wrote:
> On Mon, 12 Sep, 2005 at 07:59AM -0400, Greg Wilder spake thus:
> > On Sun, 2005-09-11 at 18:54 -0400, Joseph Dell'Orfano wrote:
> > > After a discussion
> > > with a friend, I´m not so sure anymore. He is recording his album at
> > > 44.1 kHz, arguing that this will avoid downsampling when finally
> > > pressing a CD. So, are there any opinions about this? 
> > 
> > The quality of the dithering algorithms can make very a noticeable
> > difference in your final product if you have to down-sample - but that
> > doesn't necessarily mean that you're relegated to life at 44.1.
> > 
> > IMO, the best solution is to work at 88.2khz whenever possible.  This
> > way, you can down-sample to CD quality without quantization errors.  I'm
> > no DSP guru - but as I understand it, sampling at a multiple of 44.1
> > (88.2, 176.4, etc.) is the cleanest way to go outside of DSD.

in my understanding, dithering has nothing to do with downsampling.
dithering is done to reduce artifacts from sample bit width reduction,
not from sample rate reduction. sample rate changes can be handled by a
variety of interpolation techniques, which exist in a plane defined by
two axes: quality and CPU cycles per sample. you can do cheap linear
interpolation and get a pretty unpleasant effect, or use more cycles for
a better result. dithering comes in its own varieties, but they are not
arranged on a simple 2D axis the way that interpolation is. AFAIK, so-
called "pow-f" dithering is spoken of the most highly at present, but
relies on a proprietary (patented, too, i think) algorithm.

consequently, yes there are good reasons to record at 88.2 if the final
SR will be CD quality at 44.1 because of the ability to almost entirely
avoid interpolation when downsampling. but this has nothing to do with
dithering, which handles the sample bit width change from (presumably)
24 down to 16. if done naively (e.g. 

	float twenty_four_bit_sample;
	short sixteen_bit_sample = (short) twenty_four_bit_sample;

various noisy artifacts end up in the signal. dithering is a way to
reduce their audible effect.

--p