Its time to dredge out this old baby again!!

Here is a challenge for Graham:
A while ago Arnie got into a heated argument against John Atkinson
and Glenn Z regarding aspects of the J Test about whether or
not the signal should be dithered.
I would like you to revisit this issue and tell us whether
you think Arnie was correct, or if John and Glenn were correct.
Since you are on a similar level of technical expertise as
Glenn, I think your input into this matter
would be very helpful and interesting.
And a new discussion on the technical merits should
prove interesting, as well.

Here are some excerpts:
Arny can provide his own, and I am sure he gladly wil do so.
And anyone else out there can see what history they
can come up with

HERE WE GO!

John Atkinson said 12-26-2003:

It always gratifying to learn, rather late of course, that I had
bested Arny Krueger in a technical discussion. My evidence for
this statement is his habit of waiting a month, a year, or even more
after he has ducked out of a discussion before raising the subject
again on Usenet as though his arguments had prevailed. Just as he has
done here. (This subject was discussed on r.a.o in May 2002, with Real
Audio Guys Paul Bamborough and Glenn Zelniker joining me in pointing
out the flaws in Mr. Krueger's argument.)

So let's examine what the Audio Engineering Society (of which I am
a long-term member and Mr. Krueger is not) says on the subject of
testing digital gear, in their standard AES17-1998 (revision of
AES17-1991):

Section 4.2.5.2: "For measurements where the stimulus is generated in
the digital domain, such as when testing Compact-Disc (CD) players,
the reproduce sections of record/replay devices, and digital-to-analog
converters, the test signals shall be dithered.

I imagine this is what Mr. Krueger means when wrote "The AES says don't
do it." But unfortunately for Mr. Krueger, the very same AES standard
goes on to say in the very next section (4.2.5.3):

"The dither may be omitted in special cases for investigative purposes.
One example of when this is desirable is when viewing bit weights on an
oscilloscope with ramp signals. In these circumstances the dither signal
can obscure the bit variations being viewed."

As the first specific test I use an undithered signal for is indeed for
investigative purposes -- looking at how the error in a DAC's MSBs
compare to the LSB, in other words, the "bit weights" -- it looks as if
Mr. Krueger's "The AES says don't do it" is just plain wrong.

Mr. Krueger is also incorrect about the second undithered test signal
I use, which is to examine a DAC's or CD player's rejection of
word-clock jitter, to which he refers in his next paragraph:

He does other tests, relating to jitter, for which there is no
independent confirmation of reliable relevance to audibility. I hear
that this is not because nobody has tried to find correlation. It's
just that the measurement methodology is flawed, or at best has no
practical advantages over simpler methodologies that correlate better
with actual use.

And once again, Arny Krueger's lack of comprehension of why the latter
test -- the "J-Test," invented by the late Julian Dunn and implemented as
a commercially available piece of test equipment by Paul Miller -- needs
to use an undithered signal reveals that he still does not grasp the
significance of the J-Test or perhaps even the philosophy of measurement
in general. To perform a measurement to examine a specific aspect of
component behavior, you need to use a diagnostic signal. The J-Test
signal is diagnostic for the assessment of word-clock jitter because:

1) As both the components of the J-Test signal are exact integer
fractions of the sample frequency, there is _no_ quantization error.
Even without dither. Any spuriae that appear in the spectra of the
device under test's analog output are _not_ due to quantization.
Instead, they are _entirely_ due to the DUT's departure from
theoretically perfect behavior.

2) The J-Test signal has a specific sequence of 1s and 0s that
maximally stresses the DUT and this sequence has a low-enough frequency
that it will be below the DUT's jitter-cutoff frequency.

Adding dither to this signal will interfere with these characteristics,
rendering it no longer diagnostic in nature. As an example of a
_non-diagnostic terst signal, see Arny Krueger's use of a dithered
11.025kHz tone in his website tests of soundcards at a 96kHz sample
rate. This meets none of the criteria I have just outlined.

He does other tests, relating to jitter, for which there is no
independent confirmation of reliable relevance to audibility.

One can argue about the audibility of jitter, but the J-Test's lack of
dither does not render it "really weird," merely consistent and
repeatable. These, of course, are desirable in a measurement technique.
And perhaps it worth noting that, as I have pointed out before,
consistency is something lacking from Mr. Krueger's own published
measurements of digital components on his website, with different
measurement bandwidths, word lengths, and FFT sizes making comparisons
very difficult, if not impossible.

I have snipped the rest of Mr. Krueger's comments as they reveal merely
that he doesn't actually read the magazine he so loves to criticize. :-)

John Atkinson
Editor, Stereophile
-----------------------------------------------------

Glenn Zelniker said 12-27-2001
(Not in response to JA's specific post at the top of this post):

Perfectly done, John. There are no holes in your logic; you have
connected the start and end points with all the points in
between. Good enough for me, good enough for any reasonable
person. I have come to the unfortunate conclusions that Arny is
not only constitutionally unable to admit error, but that he is
also fundamentally dishonest and pathologically unable to feel
shame about his disgusting behavior. It is therefore improbable
that he will ever acknowledge what is so painfully clear to any
sane person (namely, *most* of RAO): that he was 100% in the
wrong about the J-test, that his attacking Stereophile on the
basis of the J-test was a big mistake, and that he repeatedly
attempted to shift the context of the discussion to something
where he stood at least a small chance of being correct.

I address the following remarks to the remaining few RAO holdouts
who maintain that Arny is helpful, that Arny is useful, and that
we should be patient with Arny. I maintain that there is no
reason to engage in traditional human-style discourse with Arny
if you are on his list of enemies -- that is, you are a person
who has shown him to be wrong on a technical issue and who
refuses to back away from his snot and from his bullying. Arny is
very often very demonstrably wrong on any number of things, yet
debate with him is impossible because he refuses to admit defeat
even in the face of the most irrefutable evidence. To him,
iron-clad facts are malleable, the rules of logic are fluid, and
no amount of lying, squirming, context shifting, slander, and
deceit on his part are off-base. And he'll stoop to all of these
tactics in order to avoid being wrong about even the simplest of
matters. In short, *reality itself* must be transmogrified in
order to conform to Arny's overarching compulsion to be correct.

Arny's defenders and apologists need to recognize this sad state
of affairs and realize that doing so is not tantamount to
endorsing The Other Side. Rather, they should realize that there
are a number of good scientists and engineers who will not
entertain serious debate with Arny. It's exhausting and entirely
unrewarding. All I can hope to do is point out his serious
screw-ups and hope you (the rest of RAO, that is) will realize I
am correct. I simply won't weigh in definitively on a subject
with which I'm not confident of my own abilities. As for dealing
with Arny, the most expedient thing for me will be to treat him
with derision and dismissal unless he immediately capitulates.
Past and present history prove definitively that there's no other
way.

-------------------------------------------------------------