On 2013-05-27, David Pickett wrote:
Correct me if I am wrong, but it seems to me that the reference in this documents to "12 bits" concerns an already packed signal, which needs to start off at a higher bit rate.
That's right. What they talk about is the final distribution format, where you can do willy-nilly in-band noise shaping. If you don't do that, but try to just go with a rectangular rate-depth window, then it's going to be something like 56kS/s times 19-20b/S. That is what you pretty much have to do in any case in intermediate formats where you expect that your signal will be processed again, because otherwise you'll *certainly* end up with recodig artifacts and noise accumulation even with this sort of plain, minimal, PCM.
(I've been coding a little something in the subtractive dither vein, which might help here. But even that doesn't fully take away the basic problem. Subtractive dither can take away all of a single quantized PCM channel's distortions, but it can't negate all of the noise accumulation over multiple requantizations.)
The BBC did a study years ago and I seem to recall that they decided that the worst case was the dynamic range of big band jazz which, they determined needed 20 bits of linear PCM.
The numbers I'm citing are for the absolute worst case. That is, so that the loudest sound you are able to reproduce breaks your eardrums, and the softest sounds are just below the hearing threshold.
Be that as it may, somethin likeg 24/96, as a simple rectangular window, is always and everywhere just perfect. The only reason we ever need over 16 bits is if we want to cater to those with anechoic rooms. The only reason we would ever want to sample at over 40kHz is because certain very young individuals at some point in their life apparently can just hear upto 25kHz, momentarily, and so we need 50kHz sampling plus a small relative margin of error so as to do the anti-aliasing filters right. 24/96 is already twice as much even as a non-shaped format, but perhaps has to be chosen evenso if we want to be sure it's transparent; as the next common format which includes both sufficient sampling rate and sufficiently low self-noise to truly cover even the most nastiest of circumstances. If nothing else, we can be fully sure nothing above that will *ever* be needed even if we just treat it as a naively, TPDF-dithered, somewhat frequency limited at the upper end channel.
As far as sample rate is concerned, I agree about 48kHz before sampling, but there are a lot of 15ips tapes that sound excellent but which hardly made it to 20kHz.
Yes. And I repeat: the above numbers are the absolute, awesomest, most special case which covers the best known cases of child prodigy put in a below-measurement-floor-quiet room as well. For every practical application something like 36kHz/20bits would already suffice, even in a rectangular channel, and in-band noise shaping could bring that back into 40/14 or so. The above numbers are the absolute worst case ones for a rectangular, non-shaped channels.
-- Sampo Syreeni, aka decoy - de...@iki.fi, http://decoy.iki.fi/front +358-50-5756111, 025E D175 ABE5 027C 9494 EEB0 E090 8BA9 0509 85C2 _______________________________________________ Sursound mailing list Sursound@music.vt.edu https://mail.music.vt.edu/mailman/listinfo/sursound
