[Scott Dorsey]

William Sommerwerck wrote:
"Scott Dorsey" wrote in message ...

But, there _is_ absolutely no reason to save anything above 20kHz.
Try it for yourself... take a recording, bandlimit it to 20kHz, and listen.
Do you hear a difference? Can you say that benefit is an improvement?

I rarely disagree with Scott on anything. This is an exception.

If you cascade enough stages with a 100kHz bandwidth, you'll wind up with less
than 20kHz bandwidth, and a lot of phase shift.

IF you do that in the analogue world. In the analogue world, as I pointed
out earlier in this thread, you need to have extended frequency response in
order to get accurate response across the audible range.

But... the digital world is not like that, and that's a thing that we can
all use to our advantage.

The real worry is that, because there IS content above 20 KHz, the
additional recording bandwidth will be accurately recording it, but
the reproduction system will not accurately reproduce it and will
produce audible beat products from the inausible ultrasonics.
In this case, the additional bandwidth is /degrading/ the sound
and not improving it.

This assumes amplifiers and/or speakers have sufficient IM in the ultrasonic
region to produce audible beats. This is easy to test. Has anyone done so? Of
course not, because it costs money to run good tests, but nothing to
speculate. (Cary Grant once said something insightful about this.)

Speakers sure have sufficient IM in the ultrasonic range to produce audible
beats, because they have sufficient IM in the audible range to produce audible
beats. Speakers are the real problem here, amplifiers are more or less a
non-issue.

This argument has been applied in reverse to human hearing -- that
non-linearities in the ear generate IM products that we hear "live", but not
from band-limited recordings. Again...

That certainly is the case at very high levels and it's part of the reason
why some of the rock folks are so fond of mild clipping distortion; it mimics
the sound of the ear overloading and makes things sound louder than they
really are.

I'm not sure that this is a good thing that we want to model but it would
be a really interesting thing to try and measure accurately. Let me look
and see if anyone has done this.

The problem is that you cannot support higher frequencies without
/also/ degrading the signal in other ways, so you have to pick and
choose what converter attributes are going to give you the best gain.

What is your evidence for these degradations actually occurring?

So far just subjective listening tests on converters. There are a bunch
of converters out there that sound better at 44.1 than at 96 ksamp/sec
and there are lots of them that have more measurable clocking errors
at the higher rate. I'm not saying that there aren't ALSO converters
that sound better at 96 ksamp/sec than 44.1 but I have not encountered one.

It's the job of the production engineer to take the equipment that design
engineers produce and figure out how to use that equipment to make good
recordings. (What is a good recording? That's the producer's job to figure
out.) If a given piece of equipment sounds better configured one way than
another, by all means the production engineer should be using it in that
way.

I would be in favor of extended HF response if we could just get extended
HF response without any downside and without any side effects, because
in that case it may or may not have any benefit but it certainly could do no
harm. The problem is that the extended bandwidth is apt to do harm on
playback, and it requires sacrifices in converter design that may do harm
in recording.

Then simply have a switchable filter in the playback system.

If I had control over the customer's playback systems, by all means I would
do that, because it would solve all of these problems. But then, with that
in place, there would be no reason to use high sample rate systems at all.

One final note about the Sampling Theory For Digital Audio article: I
have nothing against the Nyquist Theorem, pure mathematics, or how the
author explained it. My problem is with the limited thinking in the audio
industry, and too many times, I've seen that article used to support and
maintain that dogma. I think that is inappropriate; that's all. I think
my negative statements earlier were more about that than anything in the
article itself.

The Nyquist theorem tells us the sampling rate needed to avoid losing
information. It tells us nothing about the audible effects of the surround
circuitry needed to make a sampling system work properly.

This is true. It's the job of the design engineer to take the stuff that
information theorists have done and make a device that the production
engineer enjoys using.
--scott


--
"C'est un Nagra. C'est suisse, et tres, tres precis."