In article , Chris Hornbeck
wrote:

On Mon, 17 Nov 2003 22:44:52 GMT, Jay - atldigi
wrote:

Something to ponder: Why does DSD (SACD) work?

Because it's a different data stream and it's not PCM.


But how do they make 1 bit work if it can only be off or full 6dB?
The point is that digital doesn't actually work that way.

But it *is* PCM. Sampling rate and word depth can be traded
pretty liberally, I think is your point. Greater word depth
reduces the ambiguity of quantization, and ambiguity is
"like" noise.


DSD can essentially be thought of as PCM that is 1 bit with a very high
sample rate, severely noise shaped, and greatly reduced filter worries
(I'd say no filters, but many manufacturers suggest using some filtering
to prevent the high level ultrasonics due to the noise shaping from
damaging equipment). The noise shaping is one reason the high sample
rate helps so much; you can shape all that noise far away from your
desired passband (20-20k) and get the equivalent of around a 120 dB
signal to noise ratio within that limited bandwidth. However, the noise
in the ultrasonic range is ridiculous.

It's like recording right off the oversampling ADC without taking a trip
through quantization. It's 1 bit. That's all there is. But it seems to
work, doesn't it? It doesn't just output 6dB square waves. It plays
music. This couldn't be possible under the criteria that some are trying
to impose. But it works. It works due to the trade off Chris alludes to.
But there's no free lunch. All that noise has to go somewhere, but after
it does, you can hear music in the 20-20k range, even though there's
only 1 bit. And it's not because of the "rising or falling" illustration
used with 1 bit recording or you'd never be able to start a song in the
middle.

I know, it's hard to wrap your head around, but viewing a digital audio
system as a whole instead of in pieces, and taking into account proper
design and implimentation, digital audio works, even though some of the
concepts seem pretty wierd in practice. Some things when illustrated on
paper are good for learning and visualization, but you eventually have
to move away from the illustrations and into the wierd world where there
are no stairsteps and you only gain more dynamic range with more bits.

Add filtering, use crappy filters, forget to dither, or use the wrong
dither, and it can all fall apart, and all of these isolated evils that
people worry about can actually come into being. Do everything right and
this stuff isn't a problem within the limits of the system, which
admittedly do exist. The filter issues (ripple, group delay, ringing,
noise or poor performance from the analog stages) that certainly affect
the audible band in the CD standard, poor filters which may not entirely
prevent aliases or images (that would be **** poor design with no excuse
these days) or amplify the normal filter issues, the noise floor which
also can be heard in the CD standard and prevent the lowest level
details from being captured (assuming the source didn't have a bunch
more noise to begin with), and poor practice in preparing the masters,
whether it be dither problems or poor processing due to bad algorithms
or insufficient resolution (we won't even mention clocking and jitter),
and you have plenty of land mines to screw you up.

24/96 solves or lessens some of those problems, gives us some margin of
safety, and offers possibilities for currently unused, simpler filter
techniques that could really help. I'm not saying 24/96 isn't better
than 16/44.1. I'm just saying that many reasons we see given aren't
always correct, and the extent of the problems are sometimes overstated.
What do the details mean to the guy just recording some music? Not much
usually. But from the standpoint of tecnical learning, the fine
distinctions are worth making.

--
Jay Frigoletto
Mastersuite
Los Angeles
promastering.com