[sdiy] VC variable bit-rate audio digitizer? help!

[Magnus Danielson]

From: Grant Richter <grichter at asapnet.net>
Subject: Re: [sdiy] VC variable bit-rate audio digitizer? help!
Date: Thu, 18 Jul 2002 21:05:49 -0500

> >> So obviously there are various known ways to do plain bit rate
> >> conversion. A much more interesting module, in my opinion, would be
> >> one that could vary the amplitude quantization _linearly_ from 2
> >> states to some hi-fi number of states. Is THAT feasible?
> 
> The process of limiting a continuous function to certain levels is called
> quantization. The process produces quantization noise, which is reduced by
> 6 dB for each additional bit. So 8 bit digitization has a signal to noise
> ratio of 48 dB (8 x 6dB) and 16 bit audio has a signal to noise ratio of 96
> dB (16 x 6 dB).
> 
> A sine wave recorded on a 16 bit system at -48 dB is only going to be
> digitized to 8 bits.

At this point I realized what one could do, but only reading on further
you suggested exactly what I was thinking of. I knew there where a better
solution around, but damping the analog signal does all the difference!
Also, this would give true CV control over the phenomena!

> This is where the quantization noise is most apparent,
> at the tail end of decays and the like. People have become very sensitized
> to quantization noise (sizzle) and complain about being able to hear it in
> 12 bit systems like the early DX7s.
> 
> You can control the quantization by recording at various low levels, and
> then do digital gain increase back to 0 dB, which is a multiplication
> without removing the quantization noise introduce by the low level
> recording.

That would do it, it is simple enought to do. Basically put an A/D back-to-back
with an D/A and then have a VCA in front and a VCA after. The VCAs then
driven so that the gain reduction in the first is balanced with the gain in
the second. Could be done.

However, this assumes true linear A/D and D/As. If one uses sigma/delta stuff
the result may be quite different, since the sigma/delta oversampling shapes
the noise. I am not judging on what is "best", just hinting that it may give
a different effect than what one expected. I guess only shootouts will tell
which is the preferred solution. Maybe some hybrid solution is best.

If one is up for nasty tricks, let's vary the sampling rate with another
CV. Hm... now we have a really nasty thing here.

Cheers,
Magnus