Home Vinyl Asylum

Welcome Licorice Pizza (LP) lovers! Setup guides and Vinyl FAQ.

Re: er...

"Yes, digitization throws away a lot of information. BUT, as Nyquist showed, it can be recovered surprisingly accurately"

It is actually more subtle than that.

What Nyquist-etc show is that sampling a band-limited signal and reconstructing it with a very specific filter at replay is a 100% lossless operation. This is still without quantisation, i.e. they are talking about an analogue sampled system.

Quantising ('digitising') the individual samples reduces the information due to the granularity of the quantizer. But this granularity can be pushed down to good-enough levels. In practice 20-22 bit is possible, giving signal-to-quantisation distortion levels of over 100dB. In short: the quantisation loss can be made arbitrarily small (but requires some competence).

The required reconstruction filter is the Sinc function. Sadly this cannot be realised in this universe, but it can be approximated to levels that are quite decent. Even so the human ear constitutes an additional reconstruction filter, making it debatable whether, for audio, one must strictly adhere to the Sinc reconstructor. After all, many people listen quite happily to non-oversampling non-filtered DACs.

So what's left of digital's harmful processes?

Ah yes, the 'sampling a band-limited signal' thingy. Nyquist-... don't teach you how to band-limit an arbitrary signal. And that's where the rub is.

You cannot have harmless band-limiting AND a totally flat frequency response to 20kHz AND a sampling rate of only 44.1kHz.

That's where CD seems to fail.

It is the CD manufacturers insistence on reaching out to 20kHz (do we really need this? Tell me which analogue tape machine or LP does so at all levels?) that causes a lot of the problems. Why? Because of the band-limiting anti-aliasing filter being flat out to 20kHz and then -100dB or less at 22kHz. Too steep. Analogue filters needn't apply here, and regular digital filters cause pre/post-ringing(*).

Solutions?

1)
You can reach out pretty harmlessly with digital to 20kHz, but then you'll need a sampling rate of more than 44.1kHz. Probably around 60kHz, earlier studies indicate 50kHz bu that seems to be a bit right.

2)
Forget the flat-to-20kHz requirement. Allow the system's frequency response to droop monotonously above, say 15kHz, to reach -3dB at 18kHz and less than -60dB at 22kHz. That way you can make pretty decent band-limiters.

Almost no-one is doing so. That's bad. But there is hope. Some manufacturers of ADCs and some recording engineers are picking up on this.

---

Much longer than I intended to reply, and no analogue contents. Sorry for that.

(* That's at the ADC side. At the DAC side we need those filters
to ring. This often not-understood dichotomy is one of the reasons of many people in the industry doing the wrong thing half of the time: they apply DAC filter lore to ADCs!)



This post is made possible by the generous support of people like you and our sponsors:
  VH Audio  


Follow Ups Full Thread
Follow Ups


You can not post to an archived thread.