NetNews Usenet Archive 1992 #31/NN_1992_31.iso/spool/rec/audio/17511

Text File | 1992-12-30 | 5.6 KB | 126 lines

Path: sparky!uunet!cs.utexas.edu!uwm.edu!linac!att!att!allegra!alice!andrew
From: [e-mail] (Andrew Hume)
Newsgroups: rec.audio
Subject: TAS article; comments please?
Message-ID: <[e-mail]>
Date: 30 Dec 92 06:39:26 GMT
Article-I.D.: alice.24510
Organization: AT&T Bell Laboratories, Murray Hill NJ
Lines: 115


    the following is the text of a letter i am planning to
send to the absolute sound (TAS). i would be especially pleased
if someone would point out any errors i have made. if you
haven't seen the article, you will have to wing it.

    thanks,
            andrew hume


Dear Sir,
.PP
I wish to comment on three matters raised in ``The Sound of Digital'' in issue 81.
I apologise for the tardiness of this note;
I have only recently started a subscription to
.I TAS
and was unsure of the publication lag for letters
(I was sure someone else would have written in!).
.PP
The first matter is a blatant mistake; the samples on a CD are 16 bits.
Not 15 bits.
This is an overt error; Rejskind mentions it twice and even uses a footnote
to assure the reader he knows what he is saying.
I am at a loss to understand how he could think that the top bit is a parity bit.
The most likely explanation of this mistake is that Rejskind confused
the 16 bit samples on the CD with the AES/EBU subframe data format.
This latter is a 32 bit format which contains a 4 bit preamble,
24 bits for a sample, and 4 control bits, one of which is a parity bit.
This parity bit is used mainly within a processing system;
it certainly doesn't intrude on the space for data samples.
.PP
The second matter is the section on Time Domain Errors which deals with jitter.
Or at least, it confuses together a number of issues under
the single label of jitter.
The most common use of the term jitter refers to the fact that
a CD player has at least two distinct places which either need or generate
a clock signal (the output bitstream of the CD reader mechanism and the clock
input for the DAC).
The clocks at these two parts of the circuit may vary with respect to each other;
the jitter is the average variance between the two clocks and is
typically measured in time (for example, 500 picoseconds RMS),
although its spectrum is at least as important as its magnitude.
Although I read a fair bit about jitter and other ``digital woes'',
I cannot recall anyone mentioning or worrying about jitter between the
clock of the source and the recorded signal.
.PP
Jitter is straightforward to solve, although it comes at a cost.
Good digital designers know about distributing clocks although
it can be tricky to get a really good clock signal out of the PLL
in the CD reader while maintaining an adequate capture range for the PLL.
In any case, the problem can be avoided completely by buffering
the bitstream between the CD reader and the DAC,
much as Sony's MD player does.
In this case, the DAC can have its own high quality independent clock and grab
the bits out of the buffer when it needs them.
.PP
This section also contains a digression on correction circuits introducing noise
into the analogue circuitry.
Yes, it happens, although it is not confined to the error correction
circuits and is often called
.I "logic induced noise"
because it can come from any digital IC.
And yes, it is straightforward to avoid (again, at a cost);
you put the DAC in another box and
connect them optically.
.PP
The final matter is the section on Dynamic Range Problems.
It is such a stunning confusion of concepts that Rejskind's comment
``... mathematical blunder, coupled with what may be outright fraud''
looks like a petard with its own hoist.
First, he starts off talking about dynamic range.
Using his parity bit mistake, he reaches a figure of 90.3dB.
He then subtracts 9.03dB because we measure noise by its RMS value.
Noise? Where did that come from?
Dynamic range is the ratio between the softest and loudest sounds recordable.
Apparently, we are now talking about a signal to noise ratio.
And although the 9.03dB (or dividing by $2 sqrt 2$) is plausible,
it would be more convincing to give a realistic level and spectrum of the noise.
Proceeding on (we are now down to about 81dB
of something),
we now mix in another concept, that of needing to allow for 8dB
headroom because digital has a ``hard ceiling''.
Apparently, we are now talking about the problem of live recording
and because of that, a CD is only 73dB which is worse than
analogue gear with Dolby SR.
.PP
Sifting through this incoherent rubbish, I can see a few concerns.
However, just because the CD has 16 bit samples, it doesn't mean we
have to record music with 16 bit samples.
It is clear that in most recording situations, the headroom problems
can be solved by using 18, 20 or 24 bit recorders.
It is also true that digitising low-level signals introduces sampling artifacts.
But proper dithering ameliorates these artifacts greatly.
And it is disingenuous to talk about the problems CDs have in reproducing
a sine wave recorded at -60dB when the best possible LP playback systems
have less than 70dB dynamic range and typically have a
signal/(noise+distortion) ratio of only 40dB.
.PP
Rejskind's article is an anti-digital rant, pure and simple.
I have no problem with that (although I don't agree with him).
However, what does publishing rants replete with errors accomplish?
The parity-bit error and the dynamic range fiasco are so egregious,
it makes me wonder how it got through
.I TAS 's
editorial process.
Either no one was knowledgeable enough, or no one cared enough.
As neither of these choices is consistent with the obvious pride HP has in
.TAS ,
I am just plain puzzled.
Can someone please explain what happened?
.in \n(qqu
.sp
Yours faithfully,
.sp .75i
Andrew Hume