Path: utzoo!attcan!uunet!mcrware!droid
From: droid@mcrware.UUCP (Andy Nicholson)
Newsgroups: comp.sys.next
Subject: Re: 16-bit Digital Sound: FP or int?
Summary: I think CD is linear encoded
Message-ID: <842@mcrware.UUCP>
Date: 14 Nov 88 01:12:25 GMT
References: <44@softart.UUCP>
Organization: Microware Systems Corp., Des Moines, Ia.
Lines: 48

In article <44@softart.UUCP>, riml@softart.UUCP (Research in Motion Limited) writes:
Stuff about cd quality sound deleted
> The reason I ask these questions is simple.  I have been convinced that
> the "CD" digital encoding uses floating point.  Why?  If it didn't, low
> level signals would be hopelessly distorted due to the lack of information
> content in a LINEAR encoding at low levels.  
> 
> Thus, LINEAR 16-bit output D/A systems will NOT produce "CD" quality sound.
> 
> If my assumptions are incorrect, I would be happy to be told so by an
> INFORMED source.
> 
> Michael A. Barnstijn,         USENET:  {uunet|sun|decvax}!watmath!softart!riml

For those who are interested, Philips publishes a magazine called "Philips
Technical Review".  I can't recall the exact date, sometime in '82 I believe,
they dedicated an issue to CD technology (Philips invented most of it, you
know).  I have read it, but I have it at home and not here, so I can't quote
from it.  Also, my employer, Microware, is involved with software development
for CDI, so I have heard a few things about this as well.

I'm not sure if all this qualifies me as in informed source, but I near as
I can remember, I don't recall ever reading or hearing anything about the
use of non-linear encoding for CD.  In fact, Philips pioneered the use
of oversampling in CD players so they could use (cheaper) 14-bit DAC's
(linear) in early CD players.  The only commercial product I know of that uses
exponential (or floating-point) encoding are Kurzwiel synths which are
18-bit (12 signal, 6 exponent).

As far as the "hopeless" distortion of low-level signals, I'm not sure I
follow.  Techniques like dithering can be used to make low-level signals
sound ok down near the noise floor, and your absolute signal-to-noise ratio
is limited by the precision (# of bits) in your signal.  Use of exponential
DAC's only gives you a greater dynamic range without increasing S/N.
The lowest level signals (0 exponent and small mantissa) are still going
to be buried in the noise.  Quoted S/N ratios are (of course) _maximum_
S/N.  The noise is always going to be at about the same place, it's just
a matter of how far your signal is above it.

Finally, exponential DAC's aren't all that expensive.  Just get a DAC of your
desired bit resolution (8, 10, 12, 14, 16, etc.) and hang a little digitally
controlled op-amp on the output.  Voila!  Exponential DAC.  The exponenent
varies the gain of the op-amp from 1 on up (1, 2, 4, etc.).  You can
build the resistor network on the op-amp to do the conversion.  The Amiga
computer uses this technique (Look at the schematics!).
A really good look at how A/D and D/A works can be found in "Musical
Applications of Microprocessors", by Hal Chamberlain, published by Hayden
Books.  It's in its second edition.